Supportive Care Needs Survey: A guide to administration, scoring and analysis

Tags: SCNS, visit, prostate cancer, cancer patients, domain, analysis of covariance, breast cancer, items, Health Research, Supportive Care Needs Survey, cancer, psychological domain, psychological needs, psychological need, score analysis, confounding factors, SUPPORTIVE CARE, variation, health information, example data sets, treatment centres, information needs, Needs Survey, Allison Boyes, cancer survivors, treatment centre, Analysing data, statistical methods, baseline level, population level, validation study, random intercept
Content: Supportive Care Needs Survey: A guide to administration, scoring and analysis Prepared by Patrick McElduff, Allison Boyes, Alison Zucca, Afaf Girgis January 2004
There is no charge for administering either the long form (SCNS-LF59) or the short form (SCNS-SF34) of the Supportive Care Needs Survey. Any publications which describe the use of the Supportive Care Needs Survey should cite the following publications as appropriate: SCNS-SF34 Boyes A, Girgis A, Lecathelinais C. Brief assessment of adult cancer patients' perceived needs: Development and validation of the 34-item Supportive Care Needs Survey (SCNS-SF34). Journal of Evaluation in Clinical Practice 2009; 15(4):602-606. SCNS-LF59 Bonevski B, Sanson-Fisher RW, Girgis A, Burton L, Cook P, Boyes A, et al (the Supportive Care Review Group). Evaluation of an instrument to assess the needs of patients with cancer. Cancer 2000; 88(1):217-25. All rights reserved. No part of this manual may be reproduced or transmitted without prior permission of the Centre for Health Research & Psycho-oncology. Requests for permission to use the SCNS, its modules or to reproduce material contained in this manual should be addressed to: The Director Centre for Health Research & Psycho-oncology (CHeRP) Cancer Council NSW & University of Newcastle Room 230A, Level 2, David Maddison Building Callaghan NSW 2308 Australia Tel: + 61 2 4913 8605 Fax: +61 2 4913 8601 Email: [email protected] http://www.newcastle.edu.au/research-centre/cherp/ This manual was prepared on behalf of the Centre for Health Research & Psycho-oncology by Patrick McElduff, Allison Boyes, Alison Zucca and Afaf Girgis. Suggested citation: McElduff P, Boyes A, Zucca A, Girgis A. The Supportive Care Needs Survey: A guide to administration, scoring and analysis. Newcastle: Centre for Health Research & PsychoOncology, 2004.
CONTENTS BACKGROUND....................................................................................................................... 2 PROFILE OF THE SCNS ....................................................................................................... 4 SCNS-LF59 ............................................................................................................................ 4 SCNS-SF31 ............................................................................................................................ 4 SCNS-SF34 ............................................................................................................................ 4 Response scale..................................................................................................................... 5 Completion time.................................................................................................................... 5 Readability ............................................................................................................................ 5 Psychometric properties ...................................................................................................... 5 DESCRIPTION OF THE SCNS DOMAINS .......................................................................... 7 Psychological needs ............................................................................................................ 7 Health system & information needs .................................................................................... 8 Physical & daily living needs ............................................................................................... 9 patient care & support needs .............................................................................................. 9 Sexuality needs................................................................................................................... 10 Additional items.................................................................................................................. 10 SUPPLEMENTARY MODULES .......................................................................................... 11 Access to Health Care and ancillary support services ..................................................... 11 breast cancer ..................................................................................................................... 11 Melanoma ............................................................................................................................ 11 Prostate cancer................................................................................................................... 12 Colostomy ........................................................................................................................... 12 UTILITY OF THE SCNS ....................................................................................................... 14 Identifying priorities for action .......................................................................................... 14 quality assurance tool ....................................................................................................... 15
Intervention tool.................................................................................................................. 15 SURVEY ADMINISTRATION............................................................................................... 16 Pen-and-paper..................................................................................................................... 16 Touchscreen computer ...................................................................................................... 16 Telephone............................................................................................................................ 17 SCORING ............................................................................................................................... 18 Standardising a Likert summated score ........................................................................... 18 Missing data ........................................................................................................................ 22 ANALYSIS ............................................................................................................................. 23 Data checking ..................................................................................................................... 23 Example datasets ............................................................................................................... 23 SAS code ............................................................................................................................. 23 Missing data ........................................................................................................................ 23 Potential confounding factors ........................................................................................... 24 Analysing data .................................................................................................................... 24 Continuous outcome .................................................................................................. 24 Dichotomous outcome ............................................................................................... 36 REFERENCES ......................................................................................................................40 APPENDIX 1: SUPPORTIVE CARE NEEDS SURVEY ­ LONG FORM (SCNS-LF59) ........ 42 APPENDIX 2: SUPPORTIVE CARE NEEDS SURVEY ­ SHORT FORM (SCNS-SF34)......47 APPENDIX 3: DESCRIPTION OF EXAMPLE DATA SETS..................................................51 APPENDIX 4: SAS CODE .....................................................................................................54
Supportive Care Needs Survey (SCNS): A guide to administration, scoring and analysis The Supportive Care Needs Survey (SCNS) is an instrument for assessing the perceived needs of people diagnosed with cancer. The SCNS is the product of more than a decade of research by the Centre for Health Research & Psychology (CHeRP). Researchers and clinical groups around Australia and internationally have used the SCNS for research and quality assurance purposes. This guide describes the recommended procedures for administering, scoring and analysing the Supportive Care Needs Survey. It also contains summary information about supplementary modules that can be used in conjunction with the core surveys. All publications relating to the SCNS should use the scoring and analysis procedures described in this guide This guide will be updated at regular intervals to include advances in the ongoing psychometric development of the SCNS and to incorporate new reference data. Acknowledgements We wish to thank Gita Mishra, Madeleine King, Robert Gibberd, Kerrie Mengersen, Sally Burrows, Christophe Lecanthelinais, Cate D'este, Dianne O'Connell and David Sibritt for contributing their knowledge and expertise to the development of this guide. We also wish to recognise the conceptual and early developmental work undertaken by Rob Sanson-Fisher, Glenda Foot and Billie Bonevski. 1
Background In the past, the effectiveness of cancer care has been assessed using biomedical endpoints such as tumour shrinkage, survival and length of remission. With the increase in rates of survival from cancer, assessment of the quality of cancer patients' survival has become more important. Outcomes such as psychological sequelae and quality of life are now recognised as valuable indicators of care and a number of different strategies have been developed to assess patients' well-being including quality of life questionnaires (Aaronson et al, 1993), symptom checklists (McCorkle & Young, 1978), pain scales (Cleeland & Ryan, 1994) and satisfaction with care surveys (Wiggers et al, 1990). These instruments generally take a problem or symptombased approach and ask patients to indicate the frequency of an issue (eg "during the past week, have you had pain?") or its severity (eg "during the past week, how severe was your pain?"). If the patient reports experiencing an issue at an elevated rate or level then it is inferred that the patient must want help with it. Needs assessment is an alternative to these approaches (Rainbird, 1999; Lattimore-Foot, 1996). Needs assessment is a direct measure of the gap between patients' experience and their expectations. Needs assessment not only directly measures patients' own perceptions of their need for help on given issues but also directly measures the magnitude of patients' desire for help in dealing with the unmet needs. Obtaining a direct index of patients' perceived needs means that assumptions do not have to be made about patients' care requirements. This ability to identify specific issues that patients need help with and to directly assess the perceived urgency of the need for help enables care to be focused on the issues patients themselves have identified as most needing help with. On a broader scale, it also enables service providers to pinpoint gaps in existing services and prioritise resource allocation to those aspects of care that need improving (Rainbird, 1999; Lattimore-Foot, 1996). A review of the needs assessment literature in the oncology field published between 1985 and 1995 revealed that no instrument met all of the following criteria for an acceptable needs assessment tool (Lattimore-Foot, 1996): measures multidimensional and comprehensive range of needs directly assesses patients' perceptions of their needs assesses whether issues of need have been experienced, which issues remain unmet needs and the magnitude of such needs on one response scale measures outcomes within a defined temporal context demonstrates acceptable reliability and validity is user friendly. 2
CHeRP embarked upon a program of work to develop and evaluate a new generic assessment tool for adult cancer patients that complied with all of the above criteria. The original core questionnaire, the Cancer Needs Questionnaire (CNQ), was developed in the early 1990s and consisted of a 52-item long-form survey and a 32-item short-form survey assessing cancer patients' perceived need for help across the following five factors: psychological, health information, physical & daily living, patient care & support and interpersonal communication (Lattimore-Foot,1996; Foot & Sanson-Fisher,1995). A second-generation core survey, the Supportive Care Needs Survey (SCNS) was developed in the late 1990s and consists of a 59-item long-form (SCNS-LF59), a 31-item short-form survey (SCNS-SF31) and the recently developed 34-item short-form survey (SCNS-SF34) (Boyes et al, 2009). Several cancer specific modules have been developed for use in conjunction with either the long or short form of the SCNS to provide more detailed information about the perceived needs of specific cancer populations. Modules are currently available for breast cancer, melanoma, prostate cancer, colostomy, and access to health care and ancillary support services. Modules for lymphoedema following breast cancer, long-term survivors of cancer and patients with advanced incurable cancer are in development. The SCNS-LF59 and SCNS-SF34 are the current core surveys and are recommended for all new activities. 3
Profile of the SCNS SCNS-LF59 The long-form of the Supportive Care Needs Survey (SCNS-LF59) consists of 59 items. It was developed in 1995 following a review of the original Cancer Needs Questionnaire by oncology specialists and further testing with cancer patients (Bonevski et al, 2000). A number of new items were created, non-informative items were removed and existing items were rephrased to increase clarity and relevance. exploratory factor analysis revealed five factors that correspond closely to those of the original CNQ. While four of the five factors are similar to those of the CNQ, sexuality appeared as a new construct and interpersonal communication failed to emerge. The 59 items of SCNS-LF59 map to the following five domains of need: psychological, health system & information, physical & daily living, patient care & support and sexuality. SCNS-SF31 To enhance the practical utility of the SCNS-LF59, the first short-form of the survey was developed in 1998. This was achieved by consecutively removing the item that correlated the least with the remaining items in the domain until all correlations resulted in coefficients of 0.65 or greater. Factor analysis was conducted with the remaining items and the factor structure interpreted. Factors with less than three items were discarded due to factor instability. This resulted in a 31-item short-form of the survey (SCNS-SF31) covering the following four domains of need: psychological, health system & information, physical & daily living and patient care & support. SCNS-SF34 Further psychometric development of the short-form survey was undertaken during 2002 (Boyes et al, 2009). In addition to consecutively removing the weakest correlating item with the remaining items within the domain until all correlations resulted in coefficients of 0.57 or greater, the frequency distribution and clinical significance of each item was considered using data from the validation study (refer to description of psychometric properties pages 8-9). A total of 34 items were selected and a factor analysis performed to determine their factor structure. The 34 items mapped to the following five domains, which are identical to those of the longer version: psychological, health system, physical & daily living, patient care & support, and sexuality. The 34-item short-form of the survey (SCNS-SF34) is recommended for all new activities. 4
Response scale For each item, respondents are asked to indicate their level of need for help over the last month as a result of having cancer, using the following response options:
No need
Not Applicable 1
Satisfied 2
Low need 3
Some need Moderate need 4
High need 5
The five response options are described as follows:
1 = No Need:
Not applicable This was not a problem for me as a result of having cancer.
2 = No Need:
Satisfied I did need help with this, but my need for help was satisfied at the time.
3 = Some Need: Low need for help This item caused me little concern or discomfort. I had little need for additional help.
4 = Some Need: Moderate need for help This item caused me some concern or discomfort. I had some need for additional help.
5 = Some Need: High Need for Help This item caused me a lot of concern or discomfort. I had a strong need for additional help.
Completion time The SCNS­LF59 takes approximately 15 to 20 minutes to complete. The SCNS-SF34 takes approximately 10 minutes to complete. Readability Reading level was calculated using the computerised Flesch-Kincaid Grade Level score. The SCNS-LF59 has a reading level of sixth to seventh grade (age 11 to 13 years). The SCNS-SF34 has a reading level of seventh to eighth grade (age 12 to 14 years). Psychometric properties A validation study was conducted to assess the construct validity and the internal reliability of the SCNS-LF59 (Bonevski et al, 2000). Consecutive patients from nine major public cancer treatment centres in New South Wales, Australia were asked to participate. Patients had to have been diagnosed with cancer at least 3 months prior to the study, were aged between 18 5
and 85 years, able to speak and write English and were physically able to complete the survey. Patients were given the SCNS-LF59 to complete at home and return by mail within seven days. There were 1350 eligible patients and 888 of these completed the survey.
The SCNS-SF31 and SCNS-SF34 (Boyes et al, 2009) were developed from a secondary analysis of this dataset. In addition, convergent validity of the SCNS-SF34 with other measures of psychosocial wellbeing was established through secondary analysis of the Cancer Survival Study dataset (Boyes et al, 2012).
Construct validity: Construct validity of the SCNS-LF59 and SCNS-SF34 was evaluated using principal components factor analysis with orthogonal rotation.
For the SCNS-LF59, five factors were identified which together accounted for 64% of the total variance. The five factors identified were labelled: psychological needs (22 items), health system & information needs (15 items), physical & daily living needs (7 items), patient care & support needs (8 items), and issues with sexuality (3 items). The four remaining items failed to map to a factor.
For the SCNS-SF34, five factors accounting for 72.1% of the total variance were identified. The five factors were identical to those of the SCNS-LF 59: psychological needs (10 items), health system & information needs (11 items), physical & daily living needs (5 items), patient care & support needs (5 items), and issues with sexuality (3 items).
Internal reliability: Internal reliability of items within each factor was assessed using Cronbach alpha with the coefficient criteria set at 0.7. As shown in Table 3, the reliability coefficients were found to be substantial and, for both surveys, exceeded 0.8 in all domains.
Table 1:
Cronbach alpha co-efficients for each factor
Domain Psychological Health system & information Physical & daily living Patient care & support Sexuality
Cronbach alpha co-efficient
SCNS-LF59
SCNS-SF34
0.97
0.95
0.96
0.96
0.90
0.87
0.87
0.90
0.87
0.88
6
Description of the SCNS domains This section provides a description of each of the five domains of need, the specific items within each domain and their primary factor loadings for both the SCNS-LF59 and SCNS-SF34.
Psychological needs The psychological domain assesses needs related to emotions and coping (see Table 2). This domain is represented by 22 items in the SCNS-LF59 and 10 items in the SCNS-SF34.
Table 2:
Psychological needs items and factor loadings
Item Fears about losing your independence Confusion about why this has happened to you Feeling bored and/or useless Anxiety Feeling down or depressed Feelings of sadness Fears about the cancer spreading Fears about the cancer returning Fears about pain Anxiety about having any treatment Fears about physical disability or deterioration Accepting changes in your appearance Worry that the results of treatment are beyond your control Uncertainty about the future Learning to feel in control of your situation Making the most of your time Keeping a positive outlook Finding meaning in this experience Feelings about death and dying Concerns about the worries of those close to you Changes to usual routine and lifestyle Concerns about the ability of those close to you to cope with caring for you
Factor Loading SCNS-LF59 SCNS-SF34
56
-
70
-
54
-
73
74
71
76
78
81
78
74
74
-
56
-
64
-
66
-
65
-
74
76
78
80
75
74
57
-
67
67
62
-
73
75
56
59
58
-
55
-
7
Health system & information needs The health system and information domain assesses needs related to the treatment centre and for information about the disease, diagnosis, treatment and follow-up (see Table 3). This domain is represented by 15 items in the SCNS-LF59 and 11 items in the SCNS-SF34.
Table 3:
Health system & information needs items and factor loadings
Item Hospital staff to convey a sense of hope to you and your family
Factor loading SCNS-LF59 SCNS-SF34
54
-
The opportunity to talk to someone who understands and has
63
-
been through a similar experience
To be given written information about the important aspects of
76
76
your care
To be given information (written, diagrams, drawings) about
74
76
aspects of managing your illness and side-effects at home
To be given explanations of those tests for which you would
81
83
like explanations
To be adequately informed about the benefits and side-effects
82
84
of treatments before you choose to have them
To be informed about your test results as soon as feasible
81
84
To be informed about cancer which is under control or diminishing (that is, remission)
78
82
To be informed about things you can do to help yourself get
78
78
well
To be informed about support groups in your area
61
-
To have access to professional counselling (eg, psychologist,
66
63
social worker, counsellor, nurse specialist) if you/family/friends
need it
To be treated like a person, not just another case
70
71
To be treated in a hospital or clinic that is as physically pleasant as possible
70
73
To be given choices about when you go in for tests or treatment
68
-
To have one member of hospital staff with whom you can talk
70
72
to about all aspects of your condition, treatment and follow-up
8
Physical & daily living needs The physical and daily living domain assesses needs related to coping with physical symptoms, side effects of treatment and performing usual tasks and activities (see Table 4). This domain is represented by 7 items in the SCNS-LF59 and 5 items in the SCNS-SF34.
Table 4:
Physical and daily living needs items and factor loadings
Pain
Item
Factor Loading SCNS-LF59 SCNS-SF34
64
70
Lack of energy/tiredness
73
75
Nausea/vomiting Feeling unwell
54
-
72
70
Not sleeping well
65
-
Work around the home
64
73
Not being able to do the things you used to do
65
72
Patient care & support needs The patient care and support domain assesses needs related to health care providers showing sensitivity to physical and emotional needs, privacy and choice (see Table 5). This domain is represented by 8 items in the SCNS-LF59 and 5 items in the SCNS-SF34..
Table 5:
Patient care and support needs items and factor loadings
Item Waiting a long time for clinic appointments
Factor Loading SCNS-LF59 SCNS-SF34
56
-
Family or friends to be allowed with you in hospital whenever you
68
-
want
More fully protected rights for privacy when you're at the hospital
67
-
More choice about which cancer specialist you see
65
69
More choice about which hospital you attend
71
74
Reassurance by medical staff that the way you feel is normal
56
62
Hospital staff to attend promptly to your physical needs Hospital staff to acknowledge, and show sensitivity to, your feelings and emotional needs
70
71
67
70
9
Sexuality needs The sexuality domain assesses needs related to sexual relationships (see Table 6). This domain is represented by three items in both the SCNS-LF59 and SCNS-SF34.
Table 6:
Sexuality needs items and factor loadings
Item Changes in sexual feelings
Factor Loading SCNS-LF59 SCNS-SF34
86
89
Changes in sexual relationships
86
90
To be given information about sexual relationships
67
73
Additional items Four items (see Table 7) are not incorporated within any of the identified domains of needs as they either failed to reach the cutpoint (0.5) for inclusion within a domain or their factor mapping was ambiguous. These items are included in the SCNS-LF59 because they are considered to be clinically important.
Table 7:
Items with no specific factor loading
Item Talking to other people about the cancer
Changes in other peoples attitudes and behaviour towards you
Concerns about your financial situation
Concerns about getting to and from the hospital
10
Supplementary modules A number of supplementary modules have been developed for use in conjunction with the SCNS-LF59 and SCNS-SF34. The modules provide detailed information about perceived needs specific to cancer site, stage of disease and type of treatment. The response options for each module are identical to the core survey. Modules are currently available for access to services, breast cancer, melanoma, prostate cancer and colostomy users (see Table 8). Modules are currently being developed for long-term survivors of cancer, patients with advanced incurable cancer and women who experience lymphoedema following breast cancer and will be published in the next edition of this guide. Following is a brief summary of existing modules. Access to health care and ancillary support services The access to services module is applicable to all cancer patients regardless of cancer site, disease stage, treatment modality or time since diagnosis. It consists of 16 items assessing need for help to access transport facilities, financial assistance, information resources, counselling and support services, home help (cleaning, gardening, nursing) and hospital facilities (child minding, food, drink). The module takes 8 minutes to complete and has a reading level of 9th to 10th grade (age 14-16 years) (Bonevski et al, 2000). Breast Cancer Two breast cancer modules have been developed; one for long-term survivors of breast cancer and another for recent survivors of breast cancer. Long-term survivors of breast cancer: This module consists of 8 items assessing needs related to self-image, interpersonal relationships, lymphoedema, prosthesis and genetic aspects of breast cancer. The module takes less than 5 minutes complete and has a reading level of 6th to 7th grade (age 11-14 years) (Girgis et al, 2000). Recent survivors of breast cancer: This module consists of 40 items assessing needs related to coping, fertility, impact of pain, access to services and resources, and information and medical communication. The module takes approximately 20 minutes to complete and has a reading level of 6th grade (age 11 -12 years). Preliminary evidence indicates that the module has demonstrated reliability and validity (Thewes et al, 2003). Melanoma The melanoma module is applicable to a wide range of patients with melanoma, varying in disease stage, treatment modality and time since diagnosis. It consists of 12 items assessing needs related to skin soreness, recurrence, and information about treatment and skin 11
protection. The module takes approximately 6 minutes to complete and has a reading level of 9th-10th grade (age 14-16 years) (Bonevski et al, 1999). Prostate cancer The prostate module is applicable to men with prostate cancer varying in disease stage, treatment modality and time since diagnosis. It consists of 7 items assessing needs related to urinary function, bowel function and masculine self-image. The module takes less than 5 minutes to complete and has a reading level of 6th to 7th grade (age 11-13 years). Preliminary evidence indicates that the module has demonstrated internal consistency and validity (Steginga et al, 2001). Colostomy The colostomy module is applicable to patients diagnosed with colon cancer who use a colostomy bag. It contains 10 items assessing needs related to emotional adjustment, lifestyle changes, interpersonal relations and management of the colostomy bag. The module takes 5 minutes to complete and has a reading level of 8th-9th grade (age 13-15 years). 12
Table 8: Summary of supplementary modules
Patient group No. of items Domains covered Test-retest reliability Internal consistency Content validity Construct validity Criterion validity Reading age Average completion time
Version Access to services
General cancer 16 Items concerning access to transport
--
--
patients
facilities, financial assistance, information
resources, counselling and support
services, home help and hospital
facilities.
4
--
--
14-16 years
8 mins
Breast cancer
Long term survivors of breast cancer
8 Items concerning, self-image, prosthesis, --
--
4
--
--
11-13 < 5 mins
interpersonal relationships, lymphoedema
years
and genetic aspects of breast cancer.
Recent survivors of breast cancer
40 Items concerning informational and communication, coping, services and resources, fertility and impact of pain.
4
4
4
4
4
11-12 20 mins
years
Melanoma
Patients with malignant melanoma
12 Items concerning soreness of skin, recurrence, lymphoedema and information about treatment and skin protection.
--
--
4
--
--
14-16 years
6 mins
Prostate
Patients with
7 Items concerning urinary function, bowel
--
4
4
4
--
11-13 < 5 mins
prostate cancer
function and masculine self-image.
years
Colostomy
Patients with colon cancer
10 Items concerning emotional adjustment,
--
--
lifestyle changes, interpersonal relations
and management of a colostomy bag.
4
--
--
13-15 years
5 mins
13
Utility of the SCNS The SCNS is applicable to both the research and clinical settings. It can be administered to cancer patients at a single point in the cancer journey or repeatedly over time. To date the SCNS has been used to identify priorities for action, to assess the adequacy of current practice in order to identify areas for improvement, and as an intervention tool to reduce patients' perceived needs (Girgis & Burton, 2001). Following is a brief summary of activities undertaken in relation to each of these applications. Identifying priorities for action Numerous descriptive studies have been conducted to identify cancer patients' broad domains of need as well as the specific issues challenging different groups of cancer patients. The original CNQ and associated modules have been administered to large cross-sectional samples of patients with melanoma (Bonevski et al, 1999), breast cancer (Girgis at al, 2000), colon cancer and patients undergoing chemotherapy (Newell et al, 1999). During 1996-1997, the SCNS-LF59 was administered in the largest cross-sectional assessment of cancer patients' needs in Australia and internationally to date (Sanson-Fisher et al, 2000). The SCNS-SF31 has been administered to cross-sectional samples of men with prostate cancer (Steginga et al, 2001), breast cancer survivors (Thewes et al, 2003) and patients undergoing radiotherapy. Currently, the SCNS-SF31 is one of a number of instruments being administered to a population-based cross-sectional sample of 850 long-term cancer survivors. Together these descriptive studies have demonstrated that cancer patients' perceived needs are highest in the psychological, health information, and physical and daily living domains and that individuals with active disease report more perceived needs than those in remission, as do younger patients compared to older patients (Steginga et al, 2001; Sanson-Fisher et al, 2000; Girgis et al, 2000; Bonevski et al, 1999). Currently, a large-scale population based longitudinal study is underway to determine how cancer survivors' perceived needs change over the course of the cancer journey. Newly diagnosed cancer patients are being administered a battery of surveys including the SCNSSF34 four times over the first five years after diagnosis. The prevalence and correlates of survivors' unmet need at six months post-diagnosis have been recently published (Boyes et al, 2012). When completed, this longitudinal study will provide critical information about the onset, duration and frequency of cancer patients' perceived needs and will pinpoint the type of assistance required by sub-groups of cancer survivors at various stages of recovery or disease progression. It will also enable reference data to be developed for gender, age, spread of disease and some cancer sites against which individual and group comparisons can be made. 14
Quality assurance tool One study has been undertaken to assess medical oncologists' ability to accurately detect their patients' levels of needs and other psychosocial outcomes. While waiting to see their oncologist, cancer patients completed a touchscreen computer survey which included the CNQ. Immediately after each consultation, oncologists completed a deskpad survey about the patient. The results showed that while oncologists accurately detected some of the more common physical symptoms, they tended to overestimate cancer patients' perceived needs and to underestimate their levels of anxiety and depression (Newell et al, 1998). This study demonstrated the importance of finding ways to provide oncologists with accurate information about their patients' needs and psychosocial problems. Intervention tool Several studies focusing on `feedback' have been undertaken in an attempt to identify effective ways of reducing the perceived needs experienced by cancer patients. A pilot-study was undertaken at one hospital to assess the impact of routinely providing patient-reported psychosocial information to oncologists. Prior to each consultation, patients completed a touchscreen computer survey which included the SCNS-SF31. A report summarising their responses was immediately printed and placed in the patient's file to facilitate discussion and referral during the consultation. Patients' psychological needs, levels of anxiety and physical symptoms were found to decrease (non-significant) over the course of their treatment period. These results were comparable to those obtained in a similar study where a summary of patients' self-reported psychosocial outcomes including perceived needs, were made available to a care coordinator and other members of the health care team at a single consultation (McLachlan et al, 2001). Building upon these data, CHeRP undertook a trial to assess the impact of two models of coordinated care on advanced cancer patients' perceived needs, quality of life and physical symptoms. Patients' responses to a telephone interview including the SCNS-SF34 were summarised and, depending on which group they were allocated to, forwarded to either their nominated GP and oncologist for discussion and action at their next appointment, or to a designated telephone caseworker who contacted them with a focus on linking them with local Community Services (Girgis et al, 2009). 15
Survey administration The survey was developed as a self-report tool for adults (18 years and older) diagnosed with cancer. Detailed instructions on how to complete the survey and a worked example are provided at the beginning of the instrument and are usually sufficient for respondents to complete the survey without any additional directives. This enables the survey to be distributed to respondents by mail or completed in a clinical setting without supervision. We do not recommend completion of the survey by a proxy (eg carer, health care provider) as patients' responses have been found to differ to those provided by proxies (Newell et al, 1998; Rainbird, 1999). No specific qualifications are required to administer the survey however, users should be mindful that respondents may be unwell and/or distressed as a result of cancer and/or its treatment. Respondents should be given adequate time to complete the survey and offered appropriate support if issues of concern arise during survey completion. Respondents who are well and perceive the survey as irrelevant to them should be provided with a brief explanation about the importance of gathering information from individuals who are coping well, in addition to those in need of assistance, to ensure that the full range of cancer experiences are represented. Pen-and-paper The SCNS-LF59 and SCNS-SF34 were developed and psychometrically tested as a pen-andpaper survey and are most commonly administered in this format. Several methods have been successfully used to distribute the surveys including: · Mail out of survey to respondents to complete at home and return by reply-paid mail; · Face to face distribution of survey to respondents in the clinic setting for immediate completion and return; · Face to face distribution of survey to respondents in the clinic setting to complete at home and return by reply-paid mail. Touchscreen computer The survey has been successfully adapted for electronic administration via touchscreen computer. Each item should be presented one at a time in exactly the same format as the penand-paper version to ensure the integrity of the data collected. Administration of the survey via touchscreen computer is highly acceptable to respondents (Newell et al, 1999) and the data collected are equivalent to that gathered via the pen-and-paper version (Boyes et al, 2002). 16
The main advantages of this form of survey administration are instant data entry, reduced occurrence of missing data, instant scoring of data and production of reports summarising respondents' answers in "real time". Telephone The survey has been successfully administered via telephone interview. Given the complexity of the response scale, we recommend sending the survey to participants in advance to refer to during the telephone interview. 17
Scoring We recommend calculating a Likert summated scale by summing the individual items within a domain. The summated scale can be analysed as the crude total of all items in the domain or it can be standardised as shown below.
Standardising a Likert summated score If m equals the number of questions in a scale and k is the value of the maximum response for each item, the standardised score is obtained by summing the individual items, subtracting m, and then multiplying the resulting value by 100/(mЧ(k-1)). A standardised Likert summated score has possible values ranging from zero to 100. For example, there are 3 items in the sexuality needs domain of the SCNS and each item has a possible response of 1 to 5. For an individual who responded with answers of 3, 2 and 4, their Likert summated scale would be 9 (i.e. 3+2+4), their adjusted value would be 6 (i.e. 9 ­ 3) and their standardised score would be 50 (i.e. 6Ч100/(3*(5-1))).
An alternative to using the Likert summated scale is to weight items according to their standardised factor scores derived from the factor analysis. However factor scores can be unstable and are subject to sample variation (Fayers & Machin, 2000). Furthermore, the factor scores for items within each domain of the SCNS appear to be similar, which means there would be very little difference to the summated value if the weights were used.
Data from the validation study: Using data from the validation study (Bonevski et al, 2000) some basic statistics for the Likert summated scales from the 5 domains in the SCNS-LF59 and SCNS-SF34 are given in Tables 9 and 10, respectively. The distributions of the summated scales for the SCNS-LF34 are shown in Figures 1 to 5.
Table 9:
Summary data for the Likert summated scale for the SCNS-LF59
Domain
Number of Minimum Maximum Mean (SD) Median (IQR) items
Psychological
22
22
110
54.7 (23.7) 51.0 (38.4)
Health system & information
15
15
75
37.6 (16.5) 32.0 (23.0)
Physical & daily living
7
7
35
16.6 (7.1) 15.4 (11.0)
Patient care & support
8
8
40
15.7 (7.1) 15.0 (8.0)
Sexuality
3
3
15
5.8 (3.3)
5.0 (5.0)
SD = standard deviation; IQR = interquartile range
18
Table 10: Summary data for the Likert summated scale for the SCNS-SF34
Domain
Number of Minimum Maximum Mean (SD) items
Median (IQR)
Psychological
10
10
50
25.7 (11.4) 24.0 (19)
Health system & information
11
11
55
28.5 (12.8) 24.0 (18)
Physical & daily living
5
5
25
12.3 (5.4) 11.0 (8.3)
Patient care & support
5
5
25
10.1 (4.8) 10.0 (6.0)
Sexuality
3
3
15
5.8 (3.3)
5.0 (5.0)
SD = standard deviation; IQR = interquartile range
Figure 1:
Distribution of the Likert summated scores for the psychological needs domain from the SCNS-SF34.
19
Figure 2:
Distribution of the Likert summated scores for the health system and information needs domain from the SCNS-SF34.
Figure 3:
Distribution of the Likert summated scores for the physical and daily living needs domain from the SCNS-SF34.
20
Figure 4:
Distribution of the Likert summated scores for the patient care and support needs domain from the SCNS-SF34.
Figure 5:
Distribution of the Likert summated scores for the sexuality needs domain from the SCNS-SF34
21
Missing data There were between 7% and 12% missing data for each question in the SCNS-LF59 among the 888 subjects who participated in the validation study. Approximately 53% of subjects answered all questions, 17% missed one question and 7% missed 2 questions. Only 2% of subjects missed all questions. When a subject completes only part of the questionnaire, data will be missing for specific items or for covariates necessary for regression analysis. So as not to lose the available information recorded for these subjects, we recommend imputing the missing values. In the case where less than half of the items within a domain are missing, one option is to impute a value that is equal to the mean for the individual of the other items in that domain (Fayers & Machin, 2000). An alternative method, which is statistically more robust and can be used more generally, is to use multiple imputation techniques to impute the missing values (Rubin, 1987). Multiple imputation can now be in implemented in SAS using proc MI (see http://support.sas.com/rnd/app/papers/multipleimputation.pdf). 22
Analysis Data checking We strongly suggest that time is spent checking the data. Firstly the responses should be checked using frequency plots to ensure that all recorded answers are within the limits of the possible responses. Secondly, for a randomly selected group of individuals the original data should be cross-checked with the computer records to check the quality of data entry. Example datasets Two datasets will be used to illustrate the methods in this Section. The first dataset is a random sample of 200 subjects from the 888 subjects who participated in the validation study and the second dataset is a subset of the data for subjects who participated in a trial of printed feedback to oncologists. A description of each dataset is given in Appendix 3 and the data are available on request. SAS code A copy of the SAS code used to the analysis is included as Appendix 4A for continuous outcomes and Appendix 4B for dichotomous outcomes. A detailed description of each procedure is given in the SAS manuals, which are freely available at http://jeff-lab.queensu.ca/stat/sas/sasman/sashtml/stat/ Missing data In longitudinal studies, missing data will occur because subjects fail to attend interviews or because they are lost to follow-up. Multiple imputation could be used to generate these missing values. We recommend using an analytical technique that deals with the missing data in a sensible way and that is what we present here. Rubin et al (1976) categorised missing data as · missing completely at random (MCAR): data missing independently of the individual's previous values and independently of the value on the day when the data are missing but they may be correlated with one or more covariates · missing at random (MAR): data missing independently of the value at the time when they are missing but are correlated with the previous values and may be correlated with one or more covariates 23
· missing not at random (MNAR): assumes missing data are dependent on their value at the time they are missing. Statistical methods for analysing longitudinal data will either assume that the data are MAR or MCAR. There is some development on methods that assume the data are MNAR but these methods are not currently available. Potential confounding factors The SCNS will be used in many situations where it is necessary to adjust for potential confounding factors. As a guide to researchers who have yet to begin collecting data, we recommend collecting information on variables that were found to be predictors of reporting "some need" in at least one of the domains in the validation study (Bonevski et al., 2000). Variables that were predictors of reporting some needed included remission status, age, gender, centre, treatment received, type of cancer, site of cancer and duration since diagnosis (Sanson-Fisher et al., 2000). Analysing data The analysis is divided into two broad sections: A. methods required for analysing data when the outcome of interest is continuous B. methods for analysing data when the outcome is dichotomous A. Continuous outcome (See Appendix 4A, p55) This section considers appropriate methods of analysis when the outcome of interest is the level of need of the subject as measured by a total domain score. The section is divided according to the type of study from which the data are obtained. The first part considers data from the SCNS when it is used for a clinical purpose or when the main aim is to characterise the level of need of an individual. The second, third and fourth parts consider the analyses required for the comparison of two randomised groups, the comparison of two non-randomised groups, and factors influencing population variation, respectively. 1. Outcome is the score for an individual subject The SCNS may be used for clinical purposes or for comparing the score of an individual to scores for a population, for example comparing an individual score with the average score for a treatment centre, a population of patients with cancer or a general population. Tables 9 and 10 24
(see page 18-19) provide average scores for a population of patients with cancer. The patients represent consecutive patients attending a surgical oncology department or a medical or radiation oncology outpatient clinics at one of the nine participating treatment centres in New South Wales, Australia (Bonevski et al, 2000). 2. Comparison of needs between two randomised groups Prospective users of the SCNS should consider randomisation based on stratification, i.e. stratifying study subjects according to their baseline level of need and any other potential confounding factors (Kernan et al., 1999). In many situations there will be a limited number of subjects available for randomisation and stratifying the study subjects may increase the efficiency of the study. Data from the feedback trial are used to illustrate the statistical methods in this section and the groups being compared are subjects allocated to active treatment and those allocated to the control group (a description is given in Appendix 3). The outcome of interest is the total psychological domain score. The first analysis considered is the effect of treatment on total psychological domain score from baseline to one of the follow-up visits. The second analysis is the effect of treatment on trends in total psychological domain score from baseline to the end of the study period using data from all follow-up visits. There are several ways of testing the effect of treatment between baseline and a single followup visit. In a randomised trial it is perfectly valid to test the effect of treatment by doing an unpaired t-test on the psychological needs score from the follow visit. However, the efficiency of the analysis could be improved by taking into account the baseline level of psychological need. This can be achieved either by doing a change score analysis or by including baseline level of psychological need as a covariate in a regression model, a technique known as analysis of covariance. The analysis could also be undertaken by fitting a random effects model. For the change score analysis a new variable is created by subtracting each individual's value in the psychological needs domain at baseline from their score at the follow-up visit. Then an unpaired t-test is performed using the new variable to compare its mean value between the two groups. Analysis of covariance is performed by fitting a linear regression model with the psychological domain score at the second time period as the dependent variable, treatment group as the independent variable and score at baseline included as a covariate. Both of these methods take into account baseline scores but they are not modelling the same thing. The change score analysis assumes that the entire baseline score for each individual is carried through to his or her outcome score, an assumption that is unlikely to be true in the presence of measurement error or where there are other factors not related to the treatment that cause 25
variation in a person's score. The analysis of covariance method only adjusts for the proportion of the baseline score that is carried through to the outcome score and therefore analysis of covariance is often described as taking account of `regression to the mean'. 2.1 Comparison of psychological domain scores at follow-up 2.1.1 Comparison of domain score at follow-up (See Appendix 4A, p56) The independent t-test has four assumptions and they are similar to the assumptions of analysis of variance and linear regression: the groups are independent; observations within a group are independent; within each group the outcome variable is normally distributed; and the variance of the two groups is similar. The first two of these assumptions relate to the study design and are not considered further. Creating a histogram of the psychological domains scores and comparing the histogram with a normal distribution can be used to check the assumption of normality. The t-test is fairly robust to the assumption of normality and it is not necessary to be too critical when doing this visual check. However, if it is consider that the data are too skewed to assume normality, a non-parametric test should be used. In this case, the appropriate nonparametric test is the Wilcoxon Rank Sum Test otherwise known as the Mann-Whitney U test. The fourth assumption can be tested using an equality of variance test which is likely to be provided with the statistical output for the t-test. Under the heading of equality of variances we are told that the method used is Folded's F statistic and the p-value associated with the test is 0.5665, which is considered not to be statistically significant at the 5% level (see output below). Therefore, we can conclude that the variation in psychosocial needs scores is not statistically significantly different between groups and we can use the results from the method that pools the variances, which gives a p-value of 0.5487. Since the p-value is greater than 0.05 we can conclude there is no statistically significant difference in needs in the psychological domain between the treatment and control group.
T-Tests
Variable Method
Variances DF t Value Pr > |t|
totalpsy Pooled
Equal
64 -0.60 0.5487
totalpsy Satterthwaite Unequal 63.9 -0.60 0.5474
Equality of Variances
F Variable Method Num DF Den DF Value Pr > F
totalpsy Folded F
33
31
1.23 0.5665
26
2.1.2 Comparison of change in domain score between follow-up and baseline (See Appendix 4A, p57) The change score analysis is done by creating a new variable equal to the difference in each individual's score at baseline and their score at follow-up. A t-test is conducted on the new variable (called "diff" in this analysis). All the assumptions of the t-test still apply but now the outcome of interest is the difference in total psychological needs between the two periods and not the total psychological need at the follow-up visit. The SAS output indicates that it is reasonable to assume the variances are equal and therefore the p-value of 0.1552 is appropriate (see output below). Again the p-value is greater than 0.05 so we conclude that there is no statistically significant difference in change in psychological needs between the two groups.
T-Tests
Variable Method
Variances DF t Value Pr > |t|
diff
Pooled
Equal
64 -1.44 0.1552
diff
Satterthwaite Unequal 63.1 -1.45 0.1531
Equality of Variances
F Variable Method Num DF Den DF Value Pr > F
diff
Folded
F
33
31
1.44 0.3117
2.1.3 Analysis of Covariance (See Appendix 4A, p57) There are several procedures that will fit a linear regression model in SAS but the natural choice is the Proc Reg procedure. In the Analysis of Variance Table from the SAS output the degrees of freedom (DF) of the total sum of squares is 65 indicating that 66 observations were used in the analysis (see output below). The 14 missing observations are from the subjects who did not complete the questionnaire or did not attend the second visit.
Source Model Error Corrected Total
Analysis of Variance Sum of Mean DF Squares Square 2 1669.39420 834.69710 63 3205.04520 50.87373 65 4874.43939
F Value Pr > F 16.41 <.0001
27
The p-value for the treatment variable is for the null hypothesis that the treatment effect is equal to zero. In this example the p-value is 0.2205, so again we conclude that there is no statistically significant difference in psychological domains scores at follow-up after adjusting for baseline level of need.
parameter estimates
Parameter Standard Variable DF Estimate Error t Value Pr > |t|
Intercept 1
6.33852 2.19293
2.89 0.0053
treatment 1
2.18255 1.76370
1.24 0.2205
totalpsy0 1
0.52020 0.09157
5.68 <.0001
2.1.4 Using mixed models (See Appendix 4A, p58) To illustrate the use of mixed models we now consider the analysis of change from baseline to follow-up. The mixed model is a repeated measurement analysis where psychological needs scores from the two visits are repeat observations on individuals. For a good tutorial in fitting mixed models in SAS see the paper by Judith Singer (Singer, 1998). For this analysis the data need to be in long format, i.e. each observation on an individual should have a separate line of data. To test if the change in psychological needs is different between the treatment and control group we fit a linear regression model with psychological need as the outcome variable, visit as one dependent variable, treatment as a second dependent variable and an interaction term of treatment by visit. The p-value of the interaction term is the p-value associated with the null hypothesis that the treatment effect is zero, i.e. the mean change in psychological needs scores is not different between the treatment and control group. However, an ordinary linear regression model assumes the subjects at the follow-up visit are independent of the subjects at baseline. This is clearly not the case here and one way to take account of the dependence is to fit a categorical variable for subject to the linear regression model. This is known as a fixed effect model, i.e. fit the term for the individual as a `fixed' effect. In this context the term fixed means that we allow each subject to have a separate intercept in the model and therefore the model estimates a separate curve for each individual. The model allows the intercept to vary, but in this case it assumes all the curves have the same slope. It is important to note that the intercept for each individual is a fixed distance from the intercept of the subject deemed to be the reference category. The fixed effect model is not very efficient because it requires a substantial number of additional parameters to be estimated. An alternative to the fixed effects model is to allow each individual to have a separate intercept term, but instead of estimating a fixed effect for the individual, you 28
assume that the intercepts have a normal distribution with some mean and variance. Therefore there is no need to estimate a separate parameter for each individual; all that is needed is to estimate the mean and variance of the distribution of intercept terms. This type of model is known as a `random effects' model or more specifically in this case a random intercept model. The random intercept model will be particularly useful in the analysis of trends over more than two time periods. A select portion of the output from a random effects model is shown below. Under the heading of Dimensions there is information relating to the number of subjects used in the analysis. The output indicates that there are 80 subjects, a maximum of 2 observations per subject and 146 observations all together. This is the same as we saw previously, 80 subjects were included at baseline and 14 of them did not have a repeat measurement. In the change score analysis and the analysis of covariance, the baseline information for these 14 subjects would have been dropped from the analysis. In the random effects model all the data are used because a random effects model pools cross-sectional and longitudinal effects. The p-value for the treatment by visit interaction term is 0.2266, again indicating that the effect of treatment is not statistically significantly different between groups.
Dimensions
Covariance Parameters
2
Columns in X
4
Columns in Z Per
1
Subject
Subjects
80
Max Obs Per Subject
2
Observations Used
146
Observations Not Used
0
Total Observations
146
Covariance Parameter Estimates
Cov Parm
Standard
Z
Subject Estimate Error Value Pr Z
Intercept STUDYNO 45.9749 10.9335 4.20 <.0001
Residual
35.3919
6.1499 5.75 <.0001
29
Fit Statistics
-2 Res Log Likelihood 1017.4
AIC (smaller is better) 1021.4
AICC (smaller is better)
1021.5
BIC (smaller is better) 1026.2
Solution for Fixed Effects
Effect
Standard
Pr >
Estimate Error DF t Value
|t|
Intercept
18.8684
1.4633 78 12.89 <.0001
VISIT
-2.7470
1.4615 64 -1.88 0.0647
treatment
-0.5113
2.0195 78 -0.25 0.8008
VISIT*treatment
2.4817
2.0327 64
1.22 0.2266
There are several other statistics of interest in the statistical output from the random intercept model. The table headed, `Covariance Parameter Estimates' provides estimates and standard errors of the random effects. For example, in the row where `Intercept' is the first entry, the value of 45.9749 is an estimate of the variance of the random intercept term. The standard error of the estimate of the variance is 10.9355 giving a Wald Statistic of 4.20 and a p-value of <0.001. The Wald statistic is testing the null hypothesis that the variance of the random effect is zero. The p-value is less than 0.05 so we reject the null hypothesis and conclude there is a statistically significant difference in the value of the intercepts for individuals. The intercept represents the predicted value for the individual at baseline in this particular model because visit is coded 0 and 1. Therefore the significant p-value infers there is a significant difference in total psychological need between individuals at baseline. The other random effect is the residual error term (2), which is the variance of the difference between the observed and the predicted value for each individual. The fit statistics provide information for comparing models when one of the models is a subset of the other. Care must be taken when using these statistics to ensure that the same number of observations is used in each model, which may not occur if there is missing data for variables that have been added to the model. For more details on the fit statistics see the SAS manual (http://jeff-lab.queensu.ca/stat/sas/sasman/sashtml/stat/chap41/sect23.htm).
30
2.2 Comparison of trends in psychological domain scores A number of methods were outlined above for comparing differences in the level of psychological domain scores at follow-up between the two groups. Not all the above methods are generalisable for comparing trends in psychological domains scores when there are more than two time points. Although both the fixed effect and the random effects models are suitable, only the random effects models are considered in this workbook because they are far more efficient than other methods and they can be extended to more flexible and informative models. We also include the generalised estimating equation (GEE) approach to this analysis and comment on why it is not the recommended method when the outcome of interest is normally distributed but is recommended when the output of interest is dichotomous.
2.2.1 The random intercept model (See Appendix 4A, p58) The random intercept model is fitted the same way as in the two time-point situation but now the data include observations in total psychological needs scores for four occasions. The only additional assumption we make is that the change in total psychological needs scores is constant over time. The dimensions table indicates there are 80 subjects with a maximum of 4 observations per subject. There are 251 observations in total clearly indicating that every individual is not measured on all 4 occasions. If a person had their psychological need measured at least once their data for that visit will be included in the analysis. For any visit in which no data are available it is assumed the data is MAR.
Dimensions
Covariance Parameters
2
Columns in X
4
Columns in Z Per Subject
1
Subjects
80
Max Obs Per Subject
4
Observations Used
251
Observations Not Used
0
Total Observations
251
31
Covariance Parameter Estimates
Cov Parm
Standard Z Subject Estimate Error Value Pr Z
Intercept STUDYNO 40.9720 8.4423 4.85 <.0001
Residual
32.5485 3.5055 9.28 <.0001
The variance of the random intercept term is 40.9720 with a p-value of < 0.001 indicating that there is statistically significant variation in total psychological needs at baseline among subjects.
The random intercept model that has been fitted is: yij = 0i + 1, treatmentij + 2 visitij + 3 treatmentij visitij + ij where, yij = the total psychological needs for individual i at time j, 0i = the random intercept term for individual i, 1, 2, and 3 are the values of the parameters for treatment, visit and the interaction term of treatment by visit, respectively, and ij = the error term for individual i at time j The interpretation of the parameters in the model depends on how the variables are coded. In this analysis, the visit variable is coded as 0, 1, 2, and 3 and the treatment variable is coded as zero for the control group and 1 for the treatment group.
Effect Intercept VISIT treatment VISIT*treatment
Solution for Fixed Effects
Standard
Estimate
Error
DF
18.0820
1.3140
78
-0.5020
0.4893
169
0.4182
1.8151
78
-0.2003
0.6859
169
t Value 13.76 -1.03 0.23 -0.29
Pr > |t| <.0001 0.3064 0.8184 0.7706
The estimate of the parameter (1) for treatment is not statistically significant but is equal to 0.4182 indicating that the treatment group had a 0.4182 lower mean level of total psychological needs at baseline that the control group. The value of the parameter (2) for visit is -0.5020 indicating that there was an average reduction between visits of 0.5020 in the total psychological needs scores in the control group. The value of the parameter (3) for the 32
interaction term of visitЧtreatment is -0.2003, which represents the difference in the slopes between the treatment and control group. This means the average change between visits in the treatment group is -0.5020 - 0.2003 or -0.7023 per visit. The interaction parameter is not statistically significant indicating that there is no statistically significant difference in the trends between groups.
Fit Statistics
-2 Res Log Likelihood 1700.5
AIC (smaller is better) 1704.5
AICC (smaller is better)
1704.6
BIC (smaller is better) 1709.3
2.2.2 Generalised estimating equations (GEE) (See Appendix 4A, p59) There is an alternative to fitting a random effects model. GEEs were introduced to adjust for correlation in data resulting from repeated measurements on the same observations or through sampling (Liang & Zeger, 1986). For linear models in which the error term is assumed to be normally distributed AND there are no missing data, the GEE will give the same marginal effects (i.e., average effect of a group) as the random effects models. In the presence of missing data the random effects model will give more accurate estimates of the true effect as they assume the data are MAR rather than MCAR. Therefore it is recommended that random effects models be used in these situations. Nonetheless, the GEE analysis is also presented because we recommend the use of GEE when the data are not normally distributed as will be the case when the outcome of interest is dichotomous, such as whether or not a subject has a need.
The results of fitting the GEE are similar to the results from the random effects models and the interpretation given above can be adopted. The comparative estimates and p-values are not too dissimilar.
Model Information
Data Set
WORK.FEEDBACK
Distribution
Normal
Link Function
Identity
Dependent Variable
totalpsy
Observations
251
Used
33
Analysis Of GEE Parameter Estimates
Empirical Standard Error Estimates
Parameter
95% Standard Confidence Estimate Error Limits
Z Pr > |Z|
Intercept
18.0798
1.4169 15.3028 20.8568 12.76 <.0001
VISIT
-0.5032
0.6125 -1.7037 0.6973 -0.82 0.4114
treatment
0.4205
1.9650 -3.4308 4.2718 0.21 0.8306
VISIT*treatment -0.1947 0.8036 -1.7698 1.3804 -0.24 0.8085
3. Comparison of needs between two observed groups (See Appendix 4A, p59) This part of the analysis section is considered separately to part 2 because the comparison of the two groups may need to be done while controlling for potential confounding factors. Some researchers may want to adjust for confounding factors when the groups have been randomly allocated, especially in the presence of missing data, therefore they should consider using the methods outlined in this section rather than the methods outlined in part 2. However, it is more likely that these methods will be used when the data are obtained from observational studies and confounding is likely to be a concern. The methods are very similar to those outlined in section 2 except only the regression models are possible, i.e. analysis of covariance, random effects models or GEEs.
4. Factors influencing variation at a population level (See Appendix 4A, p60-61) In this part we consider methods that are appropriate for examining factors that influence variation at the population level. Population in this sense could be any factor that identifies groups of subjects, for example doctor, hospital or area health service. The methods could also be used for examining factors that influence variation between individuals. Data from the validation study are used to illustrate this method. In the validation study patients were only measured once but they were selected from eight treatment centres. I have chosen to fit a mixed model to this data where the random term is the intercept for centre. The outcome of interest in the model is the total health information score and a selection of the output from the model is shown below. The random intercept term is statistically significantly different from zero indicating that average total health information scores are different between centres. Age and gender were included in the model as covariates so the difference between centres is not due to any difference in the age range or sex distribution of the patients.
34
Covariance Parameter Estimates
Cov Parm
Standard Z Subject Estimate Error Value Pr Z
Intercept centre
0.09417 0.05318 1.77 0.0383
Residual
0.08747 0.009545 9.16 <.0001
Solution for Fixed Effects
Standard
Pr >
Effect Estimate Error DF t Value
|t|
Intercept 3.6827 0.1636 7 22.51 <.0001
sex
-0.02726 0.04717 168 -0.58 0.5641
age
-0.03206 0.01867 168 -1.72 0.0878
The data also include a variable for the number of clinicians (GPs) working at each centre. The number of GPs is a centre specific variable because all individuals within a centre have the same level of the variable. To test if the number of clinicians at a centre influences the variation between centres we fit the GPs variable to the model.
There are two things to be considered. Firstly, the estimate of the coefficient for GPs is ­0.2731 and is statistically significant which indicates that the average level of health information needs decreases as the number of clinicians increase. Secondly, estimate of the variation in the intercept terms has fallen from 0.09417 to 0.003196, a reduction of approximately 97%, strongly indicating that the number of clinicians at a centre explains most of the variation between centres.
Covariance Parameter Estimates
Cov Parm
Standard Z Subject Estimate Error Value Pr Z
Intercept centre
0.003196 0.004620 0.69 0.2445
Residual
0.08764 0.009575 9.15 <.0001
Solution for Fixed Effects
Standard
Pr >
Effect Estimate Error DF t Value
|t|
Intercept 4.4175 0.1440 6 30.68 <.0001
sex
-0.03328 0.04672 168 -0.71 0.4773
age
-0.02698 0.01832 168 -1.47 0.1426
gps
-0.2731 0.03041 6 -8.98 0.0001
35
B. Dichotomous outcome (See Appendix 4B, p62) This section considers appropriate methods of analysis when the outcome of interest is dichotomous, in particular when the outcome of interest is having a need for any one of the items in a domain. Having a need may be defined as having `some' level of need or it may be defined as having a `moderate to high' level of need. A subject will be deemed to have `some' level of need for an item if their response was a 3, 4 or 5 to that item in the questionnaire and they will be deemed to have a `moderate to high' level of need if their response was a 4 or 5. This section considers the analyses required for the comparison of differences between two randomised groups and the comparison of trends between two non-randomised groups. I. Comparison of needs between two randomised groups Data from the feedback trial are used to illustrate the statistical methods in this section and the two groups being compared are those allocated to active treatment and those allocated to the control group. The outcome of interest is `some' need in the psychological domain. 1.1 Comparison of the effect of treatment from baseline to follow-up between two groups 1.1.1. Comparison of level of need at follow-up A total score is calculated for each individual at follow-up by summing the number of questions in which they indicate they have a need. If the total scores are normally distributed then an independent t-test should be used to compare the mean of the total scores between the two groups. However, if the total scores are not normally distributed then it would be more appropriate to use non-parametric tests such as the Wilcoxon Rank Sum Test. 1.1.2. Comparison of change in level of need between baseline and follow-up Testing for a difference in change between groups in the total level of need between baseline and follow-up is the same as the change score analysis for continuous data. A new variable is created by calculating the difference in each individual's total score at follow-up and their total score at baseline. Then a t-test or a non-parametric test comparing the average level of the differences between the two groups is performed (see section A on continuous outcomes). 1.1.3. Compare trends in the probability of having a need One option is to calculate a total score for each individual at each time point by summing the number of questions in which they indicate they have a need and then treat the total score as a continuous variable. 36
An alternative method is to keep all responses to items within a domain as dichotomous and include them separately in a statistical model that will adjust for the correlation of the repeat measurements (questions and times) on an individual. There were 8 items in the psychological domain of the questionnaire in the feedback trial therefore each subject will have eight outcomes for each time point. To conduct this analysis the data must be put in long format with a separate observation for each question (see the SAS code in Appendix 4B, p62). Once the data are in long format it can be analysed either as a random effects model or using a GEE. The choice in this case is not as straightforward as the choice when the data are normally distributed. When the data are normally distributed and there are no missing data, the two models will give approximately the same answer and we recommended the random effects model because it has a more realistic assumption about the missing data and therefore it will provide a more accurate estimate of the true effect.
When the data are dichotomous the choice of model is more complex. Consider the following simple example where the outcome of interest is the presence of some need for the question "In the last month what was your level of need with learning to feel in control of your situation." Only subjects who were measured at baseline and the first follow-up visit are included in the analysis (see Table 11).
Table 11: Distribution of responses to a question in the psychological domain at baseline and follow-up
First follow-up visit
Some need
No need
Baseline visit
Some need No need
33 13 46 (69.7%)
4 16 20 (30.3%)
37 (56.1%) 29 (43.9%) 66
It is possible to fit a logistic model to these data either as a random intercepts model or using a GEE but the interpretation of the coefficients from the two models will be different. In the random intercept model you get a coefficient from which you can estimate the usual odds ratio (OR) for dependent samples. The OR is 0.31 because it is equivalent to the ratio of the discordant pairs 4/13. 37
The exponential of the coefficient from a GEE with a logit link function (and binomial error term) is the odds of having a need at the follow-up visit relative to the odds of having a need at baseline, and in the given example is equal to: An alternative to the logistic model is a GEE with a different link function. For example the exponential of the coefficient of a GEE with a log link function is an estimate of the relative change (RR) in the level of need from baseline to follow-up and in the example given above would equal: representing a 31% relative reduction in the level of need between baseline and the follow-up visit. Another GEE with an easily interpretable coefficient is a GEE with an identity link function. In this case the coefficient can be interpreted as the absolute change (AR) in the level of need from baseline to follow-up: representing a 14% absolute reduction in the level of need. 1.2. Comparison of the effect of treatment on trends in needs between groups (See Appendix 4B, p63) We recommend the GEE approach for comparing the effect of treatment between two groups. In this example I have fit a GEE with a log link and a binomial error term. The outcome is some need for questions in the psychological domain and the independent variables are treatment, visit, a treatment by visit interaction term, and a term for question. The output below shows the interaction term is not statistically significant (p-value = 0.0985) suggesting there is no difference in the change in psychological need between groups. Analysis Of GEE Parameter Estimates 38
Empirical Standard Error Estimates
Parameter
95% Standard Confidence Estimate Error Limits
Z Pr > |Z|
Intercept
-1.0879
0.1716 -1.4243 -0.7515 -6.34 <.0001
treatment
-0.0612
0.2046 -0.4622 0.3399 -0.30 0.7650
VISIT
-0.0386
0.0772 -0.1898 0.1126 -0.50 0.6172
treatment*VISIT
-0.0318
0.1011 -0.2300 0.1664 -0.31 0.7534
question
d405
0.2243
0.0876 0.0526 0.3959 2.56 0.0105
question
d406
0.2793
0.0841 0.1145 0.4440 3.32 0.0009
question
d407
0.2979
0.0911 0.1194 0.4763 3.27 0.0011
question
d408
0.2157
0.0810 0.0570 0.3745 2.66 0.0077
question
d409
0.2235
0.0817 0.0634 0.3835 2.74 0.0062
question
d410
0.1683
0.0768 0.0178 0.3188 2.19 0.0284
Question
d411 -0.1113
0.0865 -0.2807 0.0582 -1.29 0.1982
Question
d412
0.0000
0.0000 0.0000 0.0000
.
.
2. Comparison of needs between two observed groups (See Appendix 4B, p63) This part of the analysis section is considered separately to part 1 because the comparison of the two groups may need to be done while controlling for potential confounding factors. Some researchers may want to adjust for confounding factors when the groups have been randomly allocated and they should also consider using the methods outlined here. The model is fit in the same way as in the previous section but covariates are added to the model. There may be some problems with convergence when the number of variables is large compared with the number of observations.
39
References Aaronson N, Ahmedzai S, Bergman B, Bullinger M, Cull A, Duez NJ, et al. The European Organisation for Research and Treatment of Cancer QLQ-C30: A quality-of-life instrument for use in international clinical trials in oncology. Journal of the National Cancer Institute 1993;85:365-376. Boyes AW, Girgis A, D'Este C, Zucca A. Prevalence and correlates of cancer survivors' supportive care needs 6 months after diagnosis: a population-based cross-sectional study. BMC Cancer 2012;12:150. Boyes A, Girgis A, Lecathelinais C. Brief assessment of adult cancer patients' perceived needs: Development and validation of the 34-item Supportive Care Needs Survey (SCNS-SF34). Journal of Evaluation in Clinical Practice 2009; 15(4):602-606 Boyes A, Newell S, Girgis A. Rapid assessment of psychosocial well-being: Are computers the way forward in a clinical setting? Quality of Life Research 2002;11:27-35 Bonevski B, Sanson-Fisher R, Girgis A, Burton L, Cook P, Boyes A, et al (the Supportive Care Review Group). Evaluation of an instrument to assess the needs of patients with cancer. Cancer 2000 88:217-25. Bonevski B, Sanson-Fisher R, Hersey P, Paul C, Foot G. Assessing the perceived needs of patients attending an outpatient melanoma clinic. Journal of Psychosocial Oncology 1999;17:101-18. Cleeland C & Ryan K. Pain assessment: global use of the Brief Pain Inventory. Annals of the Academy of Medicine 1994;23:129-38. Fayers PM, Machin D. Quality of Life: Assessment, analysis and interpretation. Chichester, England: John Wiley & Sons LTD, 2000. Foot G, Sanson-Fisher R. Measuring the unmet needs of people living with cancer. Cancer Forum 1995;19(2):131-35. Girgis A, Boyes A, Sanson-Fisher RW, Burrows S. Perceived needs of women diagnosed with breast cancer: A focus on rural versus urban location. Australian and New Zealand Journal of Public Health 2000;24:166-73. Girgis A, Burton L. Cancer patient's supportive care needs: Strategies for assessment and intervention. NSW Public Health Bulletin 2001;12:269-271. Girgis A, Breen S, Stacey F, Lecathelinais C. Impact of two supportive care interventions on anxiety, depression, quality of life, and unmet needs in patients with nonlocalized breast and colorectal cancers. Journal of Clinical Oncology 2009; 27:6180-6190. Kernan WN, Viscoli CM, Makuch RW, Brass LM & Horwitz RI. Stratified randomization for clinical trials. Journal of Clinical Epidemiology 1999;52:19-26. Lattimore-Foot GG. Needs assessment in tertiary and secondary oncology practice: A conceptual and methodological exposition [PhD thesis]. Newcastle: University of Newcastle, 1996. Liang K Zeger SL. Longitudinal data analysis using generalised linear models. Biometrika 1986;73:13-22. 40
McLachlan SA, Allenby A, Matthews J, Wirth A, Kissane D, Bishop M, et al. Randomised trial of coordinated psychosocial interventions based on patient self-assessments versus standard care to improve the psychosocial functioning of patients with cancer. Journal of Clinical Oncology 2001;19:4117-4125. McCorkle R & Young K. Development of a symptom distress scale. Cancer Nursing 1978;1:373378 Newell S, Sanson-Fisher RW, Girgis A, Ackland S. The physical and psycho-social experiences of patients attending an outpatient medical oncology department: A cross-sectional study. European Journal of Cancer Care 1999;8:73-82. Newell S, Sanson-Fisher RW, Girgis A, Bonaventura A. How well do medical oncologists' perceptions reflect their patients' reported physical and psychosocial problems? Data from a survey of five oncologists. Cancer 1998;8:1640-651. Rainbird K. Measuring the perceived needs of patients with advanced, incurable cancer: Towards evidence-based care of the dying [PhD thesis]. Newcastle: University of Newcastle, 1999. Rubin DB. Inference and missing data. Biometrika 1976;63:581-92. Rubin, B.D. Multiple Imputation for Nonresponse in Surveys. 1987 New York: John Wiley & Sons. Sanson-Fisher R, Girgis A, Boyes A, Bonevski B, Burton L, Cook P, et al (the Supportive Care Review Group). The unmet supportive care needs of patients with cancer. Cancer 2000,88(1):225-36. Steginga S, Occhipinti S, Dunn J, Gardiner RA, Heathcote P, Yaxley J. The supportive care needs of men with prostate cancer (2000). Psycho-Oncology 2001;10:66-75 Singer JD. Using Proc Mixed to fit multilevel models, hierarchical models, and individual growth models). Journal of Educational and Behavioral Statistics 1998;24:323-355. Thewes B; Meiser B; Rickard J; Friedlander M. The fertility and menopause-related information needs of younger women with a diagnosis of breast cancer: A qualitative study. Psycho-Oncology. 2003;12:500-511. Wiggers JH, O'Donovan K, Redman S, Sanson-Fisher R. Cancer patient satisfaction with care. Cancer 1990;66:610-616. 41
Appendix 1: Supportive Care Needs Survey ­ long form (SCNS-LF59) 42
SUPPORTIVE CARE NEEDS SURVEY ­ LONG FORM 59 (SCNS-LF59)
Centre for Health Research & Psycho-oncology (CHeRP)
INSTRUCTIONS
To help us plan better services for people diagnosed with cancer, we are interested in whether or not needs which you may have faced as a result of having cancer have been met. For every item on the following pages, indicate whether you have needed help with this issue within the last month as a result of having cancer. Put a circle around the number which best describes whether you have needed help with this in the last month. There are 5 possible answers to choose from:
NO NEED SOME NEED
1 Not applicable ­ This was not a problem for me as a result of having cancer. 2 Satisfied - I did need help with this, but my need for help was satisfied at the time. 3 Low need - This item caused me concern or discomfort. I had little need for additional help. 4 Moderate need ­ This item caused me concern or discomfort. I had some need for additional help. 5 High need - This item caused me concern or discomfort. I had a strong need for additional help.
For example In the last month, what was your level of need for help with: 1. Being informed about things you can do to help yourself to get well
No need
Some need
Not applicable Satisfied
1
2
Low need 3
Moderate need 4
High need 5
If you put the circle where we have, it means that you did not receive as much information as you wanted about things you could do to help yourself get well, and therefore needed some more information.
Now please complete the survey on the next 3 pages.
43
In the last month, what was your level of need for help with: 1. Pain
No need
Some need
Not applicable 1
Low Satisfied need
2
3
Moderate High need need
4
5
2. Lack of energy and tiredness
1
2
3
4
5
3. Nausea and/or vomiting
1
2
3
4
5
4. Feeling unwell a lot of the time
1
2
3
4
5
5. Not sleeping well
1
2
3
4
5
6. Work around the home
1
2
3
4
5
7. Not being able to do the things you used to
1
do
2
3
4
5
8. Fears about losing your independence
1
2
3
4
5
9. The confusion about why this has happened
1
to you
2
3
4
5
10. Feeling bored and/or useless
1
2
3
4
5
11. Anxiety
1
2
3
4
5
12. Feeling down or depressed
1
2
3
4
5
13. Feelings of sadness
1
2
3
4
5
14. Fears about the cancer spreading
1
2
3
4
5
15. Fears about the cancer returning
1
2
3
4
5
16. Fears about pain
1
2
3
4
5
17. Anxiety about having any treatment
1
2
3
4
5
18. Fears about physical disability or deterioration
1
2
3
4
5
19. Accepting changes in your appearance
1
2
3
4
5
20. Worry that the results of treatment are beyond your control
1
2
3
4
5
21. Uncertainty about the future
1
2
3
4
5
22. Learning to feel in control of your situation
1
2
3
4
5
23. Making the most of your time
1
2
3
4
5
24. Keeping a positive outlook
1
2
3
4
5
44
In the last month, what was your level of need for help with: 25. Finding meaning in this experience
No need
Some need
Not applicable 1
Low Satisfied need
2
3
Moderate High need need
4
5
26. Feelings about death and dying
1
2
3
4
5
27. Changes to your usual routine and lifestyle
1
2
3
4
5
28. Talking to other people about the cancer
1
2
3
4
5
29. Changes in other people' s attitudes and behaviour towards you
1
2
3
4
5
30. Changes in sexual feelings
1
2
3
4
5
31. Changes in your sexual relationships
1
2
3
4
5
32. Concerns about the worries of those close
1
to you
2
3
4
5
33. Concerns about the ability of those close to
1
you to cope with caring for you
2
3
4
5
34. Concerns about your financial situation
1
2
3
4
5
35. Concerns about getting to and from the hospital
1
2
3
4
5
36. Waiting a long time for clinic appointments
1
2
3
4
5
37. Family or friends being allowed with you in
1
hospital whenever you want?
2
3
4
5
38. More fully protected rights for privacy when
1
you're at the hospital
2
3
4
5
39. More choice about which cancer specialists
1
you see
2
3
4
5
40. More choice about which hospital you attend
1
2
3
4
5
41. Reassurance by medical staff that the way
1
you feel is normal
2
3
4
5
42. Hospital staff attending promptly to your physical needs
1
2
3
4
5
43. Hospital staff acknowledging, and showing
1
sensitivity to, your feelings and emotional
needs
2
3
4
5
44. Hospital staff conveying a sense of hope to
1
you and your family
2
3
4
5
45
In the last month, what was your level of need for help with: 45. The opportunity to talk to someone who understands and has been through a similar experience
No need
Some need
Not applicable 1
Low Satisfied need
2
3
Moderate High need need
4
5
46. Being given written information about the
1
2
3
4
5
important aspects of your care
47. Being given information (written, diagrams,
1
drawings) about aspects of managing your
illness and side-effects at home
2
3
4
5
48. Being given explanations of those tests for
1
which you would like explanations
2
3
4
5
49. Being adequately informed about the benefits and side-effects of treatments before you choose to have them
1
2
3
4
5
50. Being informed about your test results as
1
2
3
4
5
soon as feasible
51. Being informed about cancer which is under
1
control or diminishing (that is, remission)
2
3
4
5
52. Being informed about things you can do to
1
help yourself to get well
2
3
4
5
53. Being informed about support groups in your area
1
2
3
4
5
54. Having access to professional counselling
1
(eg, psychologist, social worker, counsellor,
nurse specialist) if you, family or friends
need it
2
3
4
5
55. Being given information about sexual relationships
1
2
3
4
5
56. Being treated like a person not just another
1
case
2
3
4
5
57. Being treated in a hospital or clinic that is as
1
physically pleasant as possible
2
3
4
5
58. Being given choices about when to go in for
1
tests or treatment
2
3
4
5
59. Having one member of hospital staff with
1
2
3
4
5
whom you can talk to about all aspects of
your condition, treatment and follow-up
46
Appendix 2: Supportive Care Needs Survey ­ short form (SCNS-SF34) 47
SUPPORTIVE CARE NEEDS SURVEY SHORT FORM 34 (SCNS-SF34)
Centre for Health Research & Psycho-oncology (CHeRP)
INSTRUCTIONS
To help us plan better services for people diagnosed with cancer, we are interested in whether or not needs which you may have faced as a result of having cancer have been met. For every item on the following pages, indicate whether you have needed help with this issue within the last month as a result of having cancer. Put a circle around the number which best describes whether you have needed help with this in the last month. There are 5 possible answers to choose from:
NO NEED SOME NEED
1 Not applicable ­ This was not a problem for me as a result of having cancer. 2 Satisfied - I did need help with this, but my need for help was satisfied at the time. 3 Low need - This item caused me concern or discomfort. I had little need for additional help. 4 Moderate need ­ This item caused me concern or discomfort. I had some need for additional help. 5 High need - This item caused me concern or discomfort. I had a strong need for additional help.
For example In the last month, what was your level of need for help with: 2. Being informed about things you can do to help yourself to get well
No need
Not applicable 1
Satisfied 2
Some need
Low need 3
Moderat High e need need
4
5
If you put the circle where we have, it means that you did not receive as much information as you wanted about things you could do to help yourself get well, and therefore needed some more information.
Now please complete the survey on the next 2 pages.
48
In the last month, what was your level of need for help with: 1. Pain
No need
Some need
Not applicable 1
Satisfied 2
Low Moderate High need need need
3
4
5
2. Lack of energy/tiredness
1
2
3
4
5
3. Feeling unwell a lot of the time
1
2
3
4
5
4. Work around the home
1
2
3
4
5
5. Not being able to do the things you used
1
to do
2
3
4
5
6. Anxiety
1
2
3
4
5
7. Feeling down or depressed
1
2
3
4
5
8. Feelings of sadness
1
2
3
4
5
9. Fears about the cancer spreading
1
2
3
4
5
10. Worry that the results of treatment are beyond your control
1
2
3
4
5
11. Uncertainty about the future
1
2
3
4
5
12. Learning to feel in control of your situation
1
2
3
4
5
13. Keeping a positive outlook
1
2
3
4
5
14. Feelings about death and dying
1
2
3
4
5
15. Changes in sexual feelings
1
2
3
4
5
16. Changes in your sexual relationships
1
2
3
4
5
17. Concerns about the worries of those close
1
to you
2
3
4
5
18. More choice about which cancer specialists you see
1
2
3
4
5
19. More choice about which hospital you attend
1
2
3
4
5
20. Reassurance by medical staff that the way
1
you feel is normal
2
3
4
5
21. Hospital staff attending promptly to your
1
physical needs
2
3
4
5
22. Hospital staff acknowledging, and showing
1
sensitivity to, your feelings and emotional
needs
2
3
4
5
49
In the last month, what was your level of need for help with:
No need
Some need
Not
Low Moderate High
applicable Satisfied need need need
23. Being given written information about the
1
important aspects of your care
2
3
4
5
24. Being given information (written, diagrams,
1
drawings) about aspects of managing your
illness and side-effects at home
2
3
4
5
25. Being given explanations of those tests for
1
which you would like explanations
2
3
4
5
26. Being adequately informed about the benefits and side-effects of treatments before you choose to have them
1
2
3
4
5
27. Being informed about your test results as
1
soon as feasible
2
3
4
5
28. Being informed about cancer which is under control or diminishing (that is, remission)
1
2
3
4
5
29. Being informed about things you can do to
1
help yourself to get well
2
3
4
5
30. Having access to professional counselling
1
(eg, psychologist, social worker,
counsellor, nurse specialist) if you, family
or friends need it
2
3
4
5
31. To be given information about sexual relationships
1
2
3
4
5
32. Being treated like a person not just another case
1
2
3
4
5
33. Being treated in a hospital or clinic that is
1
as physically pleasant as possible
2
3
4
5
34. Having one member of hospital staff with
1
whom you can talk to about all aspects of
your condition, treatment and follow-up
2
3
4
5
Thank you for completing this survey
50
Appendix 3: Description of example data sets 51
Example data set: Validation study
The data are a sample of 200 subjects randomly selected from participants in the validation study.
Variable name Variable description
Possible responses
Id
Unique identifying number for each
patient
Centre
Centre that the patient attended
Q44 to Q54, Q56 to Q59
Questions from the health system and information domain
1='Not applicable' 2='Satisfied' 3='Low need' 4='Moderate need' 5='High need'
Q61
First told had cancer
1='Within last month' 2='1-3 months ago' 3='3-6 months ago' 4='6-12 months ago' 5='1-2 years ago' 6='2-3 years ago' 7='3-5 years ago' 8='> 5 years ago' 9='Cant remember'
Age
Age group
1='18-30' 2='31-40' 3='41-50' 4='51-60' 5='61-70' 6='71-90'
Sex
Sex
1='Male' 2='Female'
GPs
The number of GPs at the centre the
patient attended.
52
Example dataset: Feedback trial
The data (feedbacksample) are a subset of the variables for all subjects who participated in the feedback trial. There were 80 subjects recruited and of these 66 attended the second visit, 57 the third visit and 48 the final visit.
Variable name Variable description
Possible responses
Id
Unique identifying number
Visit
Visit number
Flag
Number of visits that the person
attended
Age
Age in years
Anxiety
Anxiety score
Higher number represents a higher level of anxiety
Depress
Depression score
Higher number represents a higher level of depression
D405 to d412 Items in the psychological domain
1='Not applicable' 2='Satisfied' 3='Low need' 4='Moderate need' 5='High need'
Totalpsy
Total score in the psychological domain
Treatment
Treatment group
0='Control' 1='Treatment'
53
Appendix 4: SAS code 54
Appendix 4A: Continuous Outcome (refers to Section A, pp24-36)
*-------------------------------------------------------------------------------*
|
File:
SCNSWorkbookContinuous.sas
|
|
Date:
1 May 2003
|
|
Author:
Patrick McElduff
|
|
Purpose:
To do the analysis related to total domain scores
|
|
for the SCNS workbook
|
|
|
*-------------------------------------------------------------------------------*;
*Location of the data; libname dire 'C:\SCNS\data';
/********************************************************************************/
/*
Read in the sample of data from the feedback trial
*/
/********************************************************************************/
data feedback; set dire.feedbacksample; run;
/******************* ANALYSIS REQUIRED FOR SECTION A2 (p25-34) *******************/
/********************************************************************************/
/*
Put the data into wide format to do the change score analysis and the */
/* analysis of covariance
*/
/********************************************************************************/
data baseline(keep=studyno treatment age anxiety) long(keep=studyno treatment age anxiety totalpsy visit); set feedback; if visit in (0, 1) then output long; if visit = 0 then output baseline; run;
/********************************************************************************/
/*
Plot the change for each individual.
*/
/********************************************************************************/
goptions interpol=join;
axis1 color=blue width=2.0 order=(0 to 1 by 1) label = ('Visit number'); axis2 color=blue width=2.0 order=(0 to 40 by 10);
proc sort data=long; by treatment; run;
proc gplot data=long; plot totalpsy*visit=studyno/haxis=axis1 vaxis=axis2 nolegend; by treatment; title 'Change in total psychological need score'; title2 'between baseline and the first follow-up visit'; run;
/********************************************************************************/
/*
Creating a new dataset that will be based on 1 observation per study
*/
/*
number, instead of 1 observation per visit.
*/
/********************************************************************************/ proc sort data=long; by studyno; run;
55
proc transpose data=long out=wide prefix=totalpsy; by studyno; id visit; var totalpsy; run; proc sort data=wide; by studyno; run; proc sort data=baseline; by studyno; run; data diffdata(keep=studyno treatment age anxiety diff totalpsy0 totalpsy1); merge baseline wide; by studyno; diff = totalpsy1-totalpsy0; run;
** SECTION A2.1.1, p26 **;
/********************************************************************************/
/*
Check to see that the data are normally distributed.
*/
/********************************************************************************/
proc univariate data=feedback;
where visit = 1;
var totalpsy;
histogram / noframe normal(color=red) cfill=grey midpoints=5 to 30 by 5;
title 'Histogram of total psychological need at follow-up';
run;
/********************************************************************************/
/*
Compare total psychological needs scores at the first follow-up visit */
/********************************************************************************/
proc ttest data=feedback; where visit = 1; title 'T-test: comparison of total psychological domain scores'; title2 'between treatment groups at the first follow-up visit'; class treatment; var totalpsy; run;
proc npar1way data=feedback wilcoxon; where visit = 1; title 'Non-parametric test: comparison of total psychological domain scores'; title2 'between treatment groups at the first follow-up visit'; class treatment; var totalpsy; run;
56
** SECTION A2.1.2, p27 **;
/********************************************************************************/
/*
Conducting the change score analysis using data from baseline and the */
/*
first followup visit.
*/
/********************************************************************************/
proc ttest data=diffdata; title 'T-test: comparison of the change in total psychological domain scores'; title2 'from baseline (visit=0) to followup between treatment groups'; title3 'Parametric change score analysis'; class treatment; var diff; run;
proc npar1way data=diffdata wilcoxon; title1 'Non-parametric: comparison between treatment groups of the change in total'; title2 'pychological domain score from baseline (visit=0) to visit=1'; title3 'Non-parametric change score analysis'; class treatment; var diff; run;
** SECTION A2.1.3, pp27-28 **;
/********************************************************************************/
/*
Conducting the analysis of covariance using data from baseline and the */
/*
first followup visit.
*/
/********************************************************************************/
proc genmod data=diffdata; title 'GEE model comparing trends in the total psychological domain score over time'; model totalpsy1 = treatment totalpsy0 / dist=normal link=identity; run;
proc reg data=diffdata; title 'GEE model comparing trends in the total psychological domain score over time'; model totalpsy1 = treatment totalpsy0; run;
57
** SECTION A2.1.4, pp28-30 **;
/********************************************************************************/
/*
Fit the fixed effects model using data from baseline and the
*/
/*
first followup visit.
*/
/*
NOTE: the data being used is the data in long format (i.e. data=long). */
/********************************************************************************/
proc genmod data=long; title 'Fixed model comparing trends in the total psychological domain score from'; title2 'baseline to the first follow-up visit'; class studyno; model totalpsy = treatment visit treatment*visit studyno / dist=normal link=identity; run;
/********************************************************************************/
/*
Fit the random effects model using data from baseline and the
*/
/*
first followup visit.
*/
/*
NOTE: the data being used is the data in long format (i.e. data=long). */
/********************************************************************************/
proc mixed data=long noclprint covtest; title1 'Mixed model comparing trends in the total psychological domain score from'; title2 'baseline to the first follow-up visit'; class studyno; model totalpsy = visit treatment treatment*visit/solution ddfm=bw notest; random intercept / subject=studyno; run;
** SECTION A2.2.1, pp31-33 **;
/********************************************************************************/
/*
Fit the random effects model using data from all four visits
*/
/********************************************************************************/
/********************************************************************************/
/*
Plot the trend for each individual
*/
/********************************************************************************/
goptions interpol=join; axis1 color=blue width=2.0 order=(0 to 3 by 1) label=('Visit number'); axis2 color=blue width=2.0 order=(0 to 50 by 10) label=('Total score');
proc sort data=feedback; by treatment; run;
proc gplot data=feedback; plot totalpsy*visit=studyno/haxis=axis1 vaxis=axis2 nolegend; by treatment; title 'Trends in total psychological domain scores'; run;
proc mixed data=feedback noclprint covtest; title1 'Mixed model comparing trends in the total psychological domain score over time'; class studyno; model totalpsy = visit treatment treatment*visit/solution ddfm=bw notest; random intercept / subject=studyno; run;
58
** SECTION A2.2.2, pp33-34 **;
/************************************************************************ */
/*
Fit the generalised estimating equation (GEE) using data from all */
/*
four visits
*/
/*************************************************************************/
proc genmod data=feedback; title 'GEE model comparing trends in the total psychological domain score over time'; class studyno; model totalpsy = visit treatment treatment*visit / dist=normal link=identity; repeated subject=studyno/type=exch; run;
/****************** ANALYSIS REQUIRED FOR SECTION A3 (p34) *********************/
/********************************************************************************/
/*
Conducting the analysis of covariance using data from baseline and
*/
/*
the first followup visit - but include age and sex as potential
*/
/*
confounding factors.
*/
/********************************************************************************/
proc genmod data=diffdata; title 'GEE model comparing trends in the total psychological domain score over time'; model totalpsy1 = treatment totalpsy0 age anxiety / dist=normal link=identity; run;
/********************************************************************************/
/*
Fit the fixed effects model using data from baseline and the first
*/
/*
followup visit.
*/
/*
NOTE: the data being used is the data in long format (i.e. data=long). */
/********************************************************************************/
proc genmod data=long; title 'Fixed model comparing trends in the total psychological domain score over time'; class studyno; model totalpsy = treatment visit treatment*visit studyno age anxiety / dist=normal
link=identity; run;
/********************************************************************************/
/*
Fit the random effects model using data from baseline and the first
*/
/*
follow-up visit.
*/
/*
NOTE: the data being used is the data in long format (i.e. data=long). */
/********************************************************************************/
proc mixed data=long noclprint covtest; title 'Mixed model comparing trends in the total psychological domain score over time'; class studyno; model totalpsy = visit treatment treatment*visit age anxiety/solution ddfm=bw notest; random intercept / subject=studyno; run;
59
/******************** ANALYSIS REQUIRED FOR SECTION A4 (pp34-35) **************/
/********************************************************************************/
/*
Read in the sample data set from the validation study. The data requires */
/*
some manipulation before doing the analysis. Firstly create a new
*/
/*
variables for the total health domain score for each individual and
*/
/*
then taking the natural log of it.
*/
/********************************************************************************/
data validation;*(keep=id centre sex age totalhs lntotal q61 gps); set dire.sample;
*** generate a variable (countmiss) representing the number of items missing in the health domain for each individual;
countmiss=nmiss(of q44--q54 q56--q59);
*** temp is equal to the mean of the scores missing values are excluded from the calculation;
temp = mean(q44,q45,q46,q47,q48,q49,q50,q51,q52,q53,q54,q56,q57,q58,q59);
*** if less than half of the items are missing then replace the missing values with the mean for that person - same as multiplying the mean by the number of items;
if countmiss <= 8 then totalhs = temp*15; else totalhs = .;
*** Create a new variable equal to the natural log of total health domain score. The transformed variable is more normally distributed than the untransformed score although I consider both to be suitable for analysis;
lntotal = log(totalhs); run;
/********************************************************************************/
/*
Check to see if total score and log of total score are normally
*/
/*
distributed
*/
/********************************************************************************/
proc univariate data=validation; var totalhs; histogram / noframe normal(color=red) cfill=grey midpoints=15 to 85 by 10; title 'Histogram of the total health information needs score'; run;
proc univariate data=validation; var lntotal; histogram / noframe normal(color=red) cfill=grey midpoints=2.5 to 4.5 by 0.2; title 'Histogram of the natural log of total health information needs score'; run;
60
/********************************************************************************/
/*
The aim of this analysis is to examine factors that may be influential */
/*
at a population level.
*/
/********************************************************************************/
proc mixed data=validation noclprint covtest; title1 'Mixed model examining factors at the centre level that influence the total'; title2 'health domain score over time'; class centre; model lntotal = /solution; random intercept / subject=centre; run;
proc mixed data=validation noclprint covtest; title1 'Mixed model examining factors at the centre level that influence the total'; title2 'health domain score over time'; class centre; model lntotal = sex age /solution ddfm=bw notest; random intercept / subject=centre; run;
proc mixed data=validation noclprint covtest; title1 'Mixed model examining factors at the centre level that influence the total'; title2 'health domain score over time'; class centre; model lntotal = sex age gps /solution ddfm=bw notest; random intercept / subject=centre; run;
61
Appendix 4B: Dichotomous Outcome (refers to Section B, pp36-39)
*-------------------------------------------------------------------------------*
|
File:
SCNSWorkbookDichotomous.sas
|
|
Date:
1 May 2003
|
|
Author:
Patrick McElduff
|
|
Purpose:
To do the analysis related to total domain scores
|
|
for the SCNS workbook
|
|
|
*-------------------------------------------------------------------------------*;
*Location of the data; libname dire 'C:\SCNS\data';
/********************************************************************************/
/*
Read in the data from the sample of data from the feedback trial
*/
/********************************************************************************/
data feedback; set dire.feedbacksample; run;
proc format; value needfmt 1,2='No need' 3,4,5='Some need'; run;
data baseline(keep=studyno treatment age anxiety) wide(keep=studyno treatment visit d405--d412) wide1(keep=studyno treatment visit d405--d412 age anxiety); set feedback; format d405--d412 needfmt.; if visit = 0 then output baseline; if visit in (0,1) then output wide; output wide1; run;
62
** WHEN THERE ARE ONLY 2 TIME POINTS **; ** SECTION B1.2, pp38-39 **;
/********************************************************************************/
/*
Fit a GEE comparing differences in change from baseline and follow-up */
/*
NOTE: the data being used is the data in long format (i.e. data=long1). */
/********************************************************************************/
proc sort data=wide;
by studyno visit;
run;
proc transpose data=wide out=long prefix=need; by studyno visit; var d405-d412; run;
proc sort data=long; by studyno; run;
proc sort data=baseline; by studyno; run;
data long1; merge long baseline; by studyno; rename _name_=question ; run;
proc genmod data=long1 descending; title 'GEE comparing differences in change in the prevalence of some need within'; title2 'the health domain from baseline to the first follow-up visit between'; title3 'treatment groups'; class studyno question; model need1 = treatment visit treatment*visit question / dist=binomial link=log; repeated subject=studyno/ type=exch; run;
** Accounting for confounding factors**; ** SECTION B2, p39 **; proc genmod data=long1 descending; title 'GEE comparing differences in change in the prevalence of some need within'; title2 'the health domain from baseline to the first follow-up visit between'; title3 'treatment groups - with adjustment for baseline age'; class studyno question; model need1 = treatment visit treatment*visit question age / dist=binomial link=log; repeated subject=studyno/ type=exch corrw; run;
63
** WHEN THERE ARE MORE THAN 2 TIME POINTS ** SECTION B1.2, pp38-39 **;
/********************************************************************************/
/*
Fit a GEE comparing trends in prevalence of some need within the
*/
/* psychological domain between treatment groups.
*/
/********************************************************************************/
proc sort data=wide1; by studyno visit; run;
proc transpose data=wide1 out=long2 prefix=need; by studyno visit; var d405-d412; run;
proc sort data=long2; by studyno; run;
proc sort data=baseline; by studyno; run;
data long3; merge long2 baseline; by studyno; rename _name_ = question ; run;
proc genmod data=long3 descending; title 'GEE comparing trends in the prevalence of some need within the health domain'; title2 'between treatment groups'; class studyno question; model need1 = treatment visit treatment*visit question / dist=binomial link=log; repeated subject=studyno/ type=exch; run;
** Accounting for confounding factors**; ** SECTION B2, p39 **; proc genmod data=long3 descending; title 'GEE comparing trends in the prevalence of some need within the health domain'; title2 'between treatment groups - with adjustment for baseline age'; class studyno question; model need1 = treatment visit treatment*visit question age /dist=binomial link=log; repeated subject=studyno/ type=exch; run;
64

File: supportive-care-needs-survey-a-guide-to-administration-scoring.pdf
Title: Cancer Education Research Program (CERP),
Author: Allison Boyes
Published: Tue Feb 25 09:55:36 2014
Pages: 68
File size: 0.5 Mb


PACKET STATUS REGISTER, 40 pages, 2.1 Mb

Surpassing the love of men, 3 pages, 0.04 Mb
Copyright © 2018 doc.uments.com