During the current study, Dr. Hurria's salary was supported by the American Society of Clinical Oncology Young Investigator Award and the American Federation for Aging Research.
Developing a cancer-specific geriatric assessment†
A feasibility study
Version of Record online: 4 OCT 2005
Copyright © 2005 American Cancer Society
Volume 104, Issue 9, pages 1998–2005, 1 November 2005
How to Cite
Hurria, A., Gupta, S., Zauderer, M., Zuckerman, E. L., Cohen, H. J., Muss, H., Rodin, M., Panageas, K. S., Holland, J. C., Saltz, L., Kris, M. G., Noy, A., Gomez, J., Jakubowski, A., Hudis, C. and Kornblith, A. B. (2005), Developing a cancer-specific geriatric assessment. Cancer, 104: 1998–2005. doi: 10.1002/cncr.21422
Presented at the 40th annual meeting of the American Society of Clinical Oncology, New Orleans, Louisiana, June 5–8, 2004.
- Issue online: 17 OCT 2005
- Version of Record online: 4 OCT 2005
- Manuscript Accepted: 16 MAY 2005
- Manuscript Revised: 23 MAR 2005
- Manuscript Received: 13 DEC 2004
- Memorial Sloan-Kettering Cancer Center (MSKCC) Patrick Byrne Research Fund
- MSKCC NIH P20
- geriatric assessment;
- older patient;
As the U.S. population ages, there is an emerging need to characterize the “functional age” of older patients with cancer to tailor treatment decisions and stratify outcomes based on factors other than chronologic age. The goals of the current study were to develop a brief, but comprehensive, primarily self-administered cancer-specific geriatric assessment measure and to determine its feasibility as measured by 1) the percentage of patients able to complete the measure on their own, 2) the length of time to complete, and 3) patient satisfaction with the measure.
The geriatric and oncology literature was reviewed to choose validated measures of geriatric assessment across the following domains: functional status, comorbidity, cognition, psychological status, social functioning and support, and nutritional status. Criteria applied to geriatric assessment measurements included reliability, validity, brevity, and ability to self-administer. The measure was administered to patients with breast carcinoma, lung carcinoma, colorectal carcinoma, or lymphoma who were fluent in English and receiving chemotherapy at Memorial Sloan-Kettering Cancer Center (New York, NY) or the University of Chicago (Chicago, IL).
The instrument was completed by 43 patients (mean age, 74 yrs; range, 65–87 yrs). The majority had AJCC Stage IV disease (68%). The mean time to completion of the assessment was 27 minutes (range, 8–45 mins). Most patients were able to complete the self-administered portion of the assessment without assistance (78%) and were satisfied with the questionnaire length (90%). There was no association noted between age (P = 0.56) or educational level (P = 0.99) and the ability to complete the assessment without assistance.
In this cohort, this brief but comprehensive geriatric assessment could be completed by the majority of patients without assistance. Prospective trials of its generalizability, reliability, and validity are justified. Cancer 2005. © 2005 American Cancer Society.
Cancer is a disease of aging, with approximately 60% of all cancers and 70% of cancer mortality occurring in people age ≥ 65 years. The absolute number of older patients with cancer is increasing as the population ages. Older patients of the same chronologic age differ widely in physical and psychological functioning. The individual's level of functioning is an important factor to consider when weighing the risks and benefits of a treatment that can be life sustaining yet potentially toxic. A comprehensive yet brief measure needs to be developed to characterize the “functional age” of an older patient to select the most appropriate treatment and stratify outcomes based on factors other than chronologic age. Such a measure would include the essential domains evaluated by geriatricians including functional status, comorbid medical conditions, cognition level, nutritional status, psychological status, and extent of social support. Each of these domains, collectively termed “geriatric assessment,” serves as an independent predictor of morbidity and mortality in the older patient.1–9
The rationale for incorporation of geriatric assessment in the care of the older oncology patient is to use this assessment to derive an estimation of life expectancy and tolerance to cancer therapy, allowing a common language beyond that provided by chronologic age, in the classification of older adults. The role of geriatric assessment in the care of the older patient was summarized in a metaanalysis of 28 controlled trials of geriatric assessment, which demonstrated that geriatric assessment, if linked to geriatric interventions, reduced early rehospitalization and mortality in older patients through early identification and treatment of problems.10 A randomized controlled trial in the setting of the Veterans Administration (VA) health system demonstrated that geriatric evaluation and management modestly reduce functional decline and improve mental health. However, in contrast to the result of the metaanalysis, in the current trial, geriatric assessment had no effect on survival.11
An outpatient multidisciplinary geriatric assessment can take up to 2 hours to complete, not including the time taken to review the data and to formulate a plan for further management. Although it provides useful information, it is time intensive and impractical to perform in a typical oncology practice. Furthermore, the existing tools have not been specifically validated for use in oncology patients. The goals of the current pilot study were to develop a brief yet comprehensive cancer-specific geriatric assessment measure (CSGA) that would be primarily self-administered and to establish its feasibility in an oncology practice.
MATERIALS AND METHODS
The study was approved by the institutional review board. Inclusion criteria for the current study were: 1) age ≥ 65 years; 2) a diagnosis of breast, lung, or colorectal carcinoma, or lymphoma; 3) receipt of standard chemotherapy for either adjuvant or metastatic treatment; and 4) fluency in English. Patients not fluent in English were excluded because not all instruments have been validated in other languages. Written informed consent was obtained from all patients.
This geriatric assessment was completed in the physician's office. Three items required assistance by a healthcare professional or member of the research team at the enrolling institution: 1) Blessed Orientation-Memory-Concentration (BOMC) test; 2) Timed Up and Go; and 3) Karnofsky performance status (KPS). The remainder of the geriatric assessment was either self-administered or administered by a research assistant if the patient required assistance. Participating institutions included Memorial Sloan-Kettering Cancer Center (New York, NY) and the University of Chicago (Chicago, IL).
The primary endpoint of the study was to determine the feasibility of administering the geriatric assessment measure, including: 1) the percentage of patients able to complete it on their own; 2) the length of time to complete; and 3) patient satisfaction with the questionnaire, identifying items that were distressing or difficult to comprehend. The secondary end points were to determine the percentage of patients who did not enter the study and to analyze the reasons why. This was accomplished by determining: 1) the number of patients asked to participate and 2) the number of patients who declined and the reasons why. Participants were asked for feedback regarding the questionnaire's length and questions that were either not easily understandable or that were upsetting, and for suggestions for items that were not covered but should be added.
The study population consisted of a convenience sample of patients referred to the study by their primary oncologists. Of the 48 patients approached for the study, 10% (5 patients) refused participation, mainly citing fatigue. Of the 43 patients who enrolled, 5% (2 patients) decided not to participate in the study after informed consent was obtained and 1 was ineligible based on a BOMC score > 11, suggesting cognitive impairment. Therefore, 40 patients participated, representing 83% of eligible participants.
The measures included in this brief geriatric assessment were chosen for their reliability, validity, brevity, and prognostic ability to determine risk for morbidity or mortality in an older patient. The measures selected are summarized in Table 1.
|Domain||Measures||No. of Items||References|
|Functional status||1) Activities of Daily Living (subscale of MOS Physical Health)||10||13|
|2) Instrumental Activities of Daily Living (subscale of the OARS)||7||12|
|3) Karnofsky physician-rated performance rating scale||1||14, 15|
|4) Karnofsky self-reported performance rating scale||1||16|
|5) Timed Up and Go||1||17|
|6) No. of falls in last 6 mos||1||19|
|Comorbidity||Physical Health Section (subscale of the OARS)||48||12|
|Cognition||Blessed Orientation-Memory-Concentration test||6||20, 21|
|Psychologic||Hospital Anxiety and Depression Scale||14||22–26|
|Social functioning||MOS Social Activity Limitations Measure||4||27|
|Social support||1) MOS Social Support Survey: Emotional/Information and Tangible subscales||13||27|
|2) Seeman and Berkman Social Ties||4||9|
|Nutrition||1) Body mass index||1||4|
|2) Percent unintentional weight loss in the last 6 mos||1||5, 6|
Instrumental Activities of Daily Living subscale
The Instrumental Activities of Daily Living (IADL) is a subscale of the Multidimensional Functional Assessment Questionnaire (MFAQ): Older American Resources and Services (OARS). The OARS MFAQ was developed to provide a profile of the level of functioning and need for services of older persons who live at home but may have some degree of impairment. The MFAQ has been tested in > 6000 older community residents. The IADL subscale consists of 7 questions rated on a 3-point Likert scale measuring the degree to which an activity can be performed independently. The 5-week test-retest correlation is 0.71 for the IADL subscale.12
Activities of Daily Living subscale
The Activities of Daily Living (ADL) is a subscale of the Medical Outcomes Study (MOS) Physical Health. The MOS Physical Health Scale measures a broad range of physical functioning, with questions ranging from “Can you bathe and dress yourself?” to “Can you perform vigorous activities, such as running or lifting heavy objects?” Items are rated on a 3-point Likert scale measuring independence in performing the activity. Internal consistency of the physical function score is 0.92.13
Karnofsky Physician-Rated Performance Rating scale
The KPS is a general measure of patient independence in carrying out normal activities and self-care needs. The scale, developed in 1948, has been widely used in the evaluation of patients with cancer. Patients are given a score on a numerical scale of 0–100 as a global indicator of functional status. There is a moderate degree of interrater reliability between nurse and social worker KPS ratings with a Pearson correlation of 0.69 (P < 0.001). In terms of validity, the KPS most strongly correlates with variables related to physical functioning (difficulty with stairs, difficulty with balance; Pearson correlation = 0.63 [P < 0.001] and 0.61 [P < 0.001] respectively).14, 15
Karnofsky Self-Reported Performance Rating scale
A self-reporting version of the KPS was developed to assess the patients' perception of their own performance status. With this scale, patients rate their own functional status and choose from a range of functioning from “able to carry out normal activities requiring no assistance” to “severely disabled, requiring continuous nursing care.” Among patients with cancer participating in clinical trials, the patient-rated KPS was significantly related to survival (P < 0.05) and provided information independent from that obtained by the physician-scored performance status.16
Timed Up and Go
The Timed Up and Go is a performance test of physical mobility. The test measures how many seconds it takes an individual to stand up from a standard armchair (approximate seat height of 46 cm), walk a distance of 3 m (10 ft), turn, walk back to the chair, and sit down again. The test was originally reported by Mattias et al., and subsequently modified by Podsiadlo et al.17 to be a timed test. In a population of frail, community-dwelling older adults, there was good interrater and intrarater reliability (intraclass correlation coefficient = 0.99 for both). The Timed Up and Go score correlated to the scores on the Berg Balance Scale (correlation coefficient [r] = −0.72), gait speed (r = −0.55), and Barthel Index of ADL (r = −0.51).17 Guralnik et al.18 reported gait speed as an important predictor of disability.
Number of falls in last 6 months
Older patients are at risk for falls because of limited mobility, gait, and balance impairments.19 Falls may place patients with cancer at greater than average risk for injury because bony metastases place them at risk for a pathologic fracture, and patients receiving chemotherapy may have a low platelet count, which puts them at greater risk of hemorrhage. Patients were asked to report their number of falls in the last 6 months.
Physical Health Section
The OARS Physical Health Section is a comorbidity scale that contains a list of current illnesses and conditions an individual might have, and the degree to which they impair daily activities, rated on a 3-point scale of “not at all” to “a great deal.” A list of current medications is also recorded. The test-retest reliability for the Physical Health subscale over 5 weeks was 0.66. In terms of validity, the Physical Health subscale correlated significantly with health professional ratings (Kendall tau coefficients = 0.75).12
Blessed Orientation-Memory-Concentration test
The BOMC consists of six questions designed to screen for cognitive impairment. A score > 11 signifies cognitive impairment. The test-retest reliability is high (Spearman rank correlation = 0.96; P < 0.001).20 The BOMC has excellent validity as a screening instrument, correlates highly with clinicians' ratings of dementia severity (r = 0.89), predicts results from a longer (26-item) mental status questionnaire, and discriminates between patients with mild, moderate, and severe cognitive deficits.21
Hospital Anxiety and Depression Scale
The Hospital Anxiety and Depression scale (HADS) is a 14-item self-administered measure that has been extensively tested in cancer populations. It has two, seven-item subscales that assess depression and anxiety. The scale is considered particularly appropriate for use with medically ill patients because of the absence of somatic items, which often confound the determination of psychiatric problems in a medically ill population. The findings on the HADS Anxiety and Depression subscales demonstrate medium to strong correlations with other questionnaires for anxiety and depression, such as the Beck Depression Inventory, the Spielberger State-Trait Anxiety Inventory, the Clinical Anxiety Scale, and the Symptom Checklist 90 Scale (correlation coefficients range from 0.60 to 0.80).22–26
Medical Outcomes Study Social Activity Limitations Measure
The impact of overall health on social functioning in the older patient will be assessed by the Social Activity Limitations subscale from the Medical Outcomes Study Social Activity Limitations Measure (MOS). The four-item subscale identifies the extent to which physical or emotional problems interfere with an individual's social activities. All items are rated on a 5-point Likert scale, with response categories varying in each item. The mean of the total score is transformed to a scale of 0–100, with a higher number indicating greater support. Internal consistency is good (alpha coefficient = 0.77). The results on this measure correlate with role limitations due to physical (r = 0.52) and emotional (r = 0.49) health, psychologic distress (r = 0.64), and pain (r = 0.55).27
Medical Outcomes Study Social Support Survey: Emotional/Information and Tangible subscales
This is a 20-item measure of perceived availability of social support, with 4 subscales: Emotional/Informational, Tangible, Affectionate, and Positive Social Interactions. In the current study, we use the Tangible (access to material aid or behavioral assistance) and Emotional/Information (the expression of positive affect and empathetic understanding; the offering of advice, information, guidance, or feedback) subscales. All but 1 item is rated on a 5-point Likert scale from “none of the time” to “all of the time.” These 2 subscales consist of 12 items. Internal consistency of the subscales and total score is excellent (alpha coefficient > 0.90). Convergent validity was demonstrated by significant correlations of the social support total score with measures of mental health (r = 0.45; P < 0.01).27, 28
Seeman and Berkman Social Ties
This measure includes 4 questions regarding the presence of social ties: 1) marital status, 2) close contact with ≥ 2 close friends/relatives, 3) regular church attendance, and 4) membership in other types of groups. In a study by Seeman et al.,9 the presence of social ties was inversely related to mortality, independent of age. In comparison to the MOS Social Support Survey, this measure quantifies the number of social ties, but does not describe whether the patient feels supported by these individuals.
Body mass index
In a prospective cohort study of 214 older community-dwelling adults, a low body mass index (BMI), defined as a BMI < 22 kg/m2, was associated with dependency in activities of daily living (odds ratio = 1.21; 95% confidence interval [CI], 1.01–1.45) and decreased 1-year survival (relative risk = 0.85; 95% CI, 0.74–0.97).4
Percent unintentional weight loss in the last 6 months
The prognostic effect of unintentional weight loss in patients with cancer was evaluated in a study of 3047 patients enrolled in Eastern Cooperative Oncology Group chemotherapy trials. Weight loss during the 6 months before chemotherapy was associated with poorer survival, lower chemotherapy response rates, and decreased performance status.5, 6
The mean age of patients included in the study was 74 years (range, 65–87 yrs). The sample consisted of patients receiving chemotherapy for breast carcinoma (25%), colorectal carcinoma (18%), lung carcinoma (38%), and lymphoma (20%). The majority of the participants were married (55%), retired (83%), and white (90%). Sixty-five percent had some college experience or beyond. Ninety-five percent of patients reported other comorbid medical conditions, with a mean of three nonindex cancer medical problems. The top 4 reported comorbid conditions were hypertension in 43% of patients (17 patients), arthritis in 38% (15 patients), heart disease in 28% (11 patients), and other cancers or leukemias in 25% (10 patients). Twenty-three percent of patients (9 patients) rated their hearing as fair and 8% (3 patients) rated their hearing as poor. Fifteen percent (6 patients) rated their eyesight (with glasses or contacts) as fair, 5% (2 patients) as poor, and 3% (1 patient) as totally blind. Patient characteristics are summarized in Table 2.
|Characteristic||Percent||No. of patients|
|Mean age in yrs (range)||74 (65–87)|
|Types of cancer|
|Less than high school||15||6|
|High school graduate||20||8|
|Some college/junior college||23||9|
The mean time to completion of the geriatric assessment was 27 minutes (standard deviation [SD] = 10; range, 8–45 mins). This is inclusive of patient time to complete the self-administered portion of the assessment and the interviewer time to administer the BOMC test and Timed Up and Go, and to rate the patient's KPS. Seventy-eight percent of the patients (31 patients) completed the self-administered portion of the assessment without assistance. There was no bivariate association between age (P = 0.56) or educational level (P = 0.99) and the ability to complete the assessment. Eighty-three percent of the patients (33 patients) stated that the assessment was easy to understand, and all stated that no items were upsetting. Ninety percent of the patients (36 patients) were satisfied with the length of the questionnaire. One patient was ineligible based on a BOMC score > 11, suggesting cognitive impairment.
Functional status was measured in the following ways. Thirty-eight percent were dependent in one or more instrumental activities of daily living. However, the mean score was 13 (SD = 2) on a scale of 0–14. The median score was 14 and 63% of patients (25 patients) achieved a perfect score of 14, demonstrating evidence of a ceiling effect. In contrast, the MOS Physical Health scale captured a wide variability in physical functioning, with a mean score of 63 (SD = 6), a median score of 70, and a range of scores of 5–100 on a scale of 0–100. Physicians reported a median KPS score of 80, and patients reported a median KPS score of 90, with a 0.72 correlation coefficient between physician and patient-reported KPS. There was a 58% (23 patients) discordance between physician and patient-rated KPS, with 83% (19 of 23 patients) of patients more likely to report a higher KPS score than the physician. However, 26% of patients (6 of 23 patients) had a > 10% discordance between physician and patient-rated KPS in either direction. The mean score on the Timed Up and Go was 13 (SD = 5) seconds. Twelve percent of patients (4 patients) scored above a cutoff score of 15 seconds. Twenty-eight percent of patients (11 patients) had ≥ 1 fall in the last 6 months. Correlation coefficients among the OARS IADL, MOS Physical Health scale, physician-rated KPS, and patient-rated KPS ranged from 0.52 to 0.67. There was a poor correlation noted between these measures of physical functioning and the Timed Up and Go or number of falls in the last 6 months (correlation coefficients < 0.07). The correlation between the Timed Up and Go and number of falls in last 6 months was 0.28.
Psychologic state was measured by the HADS. Eight percent of patients (3 patients) reported significant anxiety or depression, based on the total score ≥ 15. Five percent (2 patients) reported significant depression, and no patients reported significant anxiety based on a subscale score ≥ 11. Social functioning and support were measured by the MOS Social Activities Limitations Measure, MOS Social Support Survey, and the Seeman and Berkman Social Ties questionnaire. Forty-five percent of patients (18 patients) noted limitations in social activities because of their physical or emotional states, and 93% (37 patients) stated that they had ≥ 2 close relatives. There was a wide range of scores for emotional support (mean, 85 [SD = 18]; median, 89; and range, 44–100) and tangible support (mean, 85 [SD = 25]; median, 100; and range, 6–100), despite relatively high median scores.
Nutritional status was assessed by measuring BMI and recording the degree of weight loss in the past 6 months. The mean BMI was 28 (SD = 5; range, 22–46; median, 26). Only 5% of patients (2 patients) had a BMI < 22 kg/m2. In contrast, 23% (9 patients) were obese as defined by a BMI > 30 kg/m2. Forty-eight percent of patients (19 patients) reported that they had lost weight in the past 6 months, whereas 38% of patients (15 patients) reported a weight loss of > 5% of their body weight in the last 6 months.
In the current pilot study, we determined that a brief but comprehensive, mainly self-administered geriatric assessment questionnaire for older patients with cancer is feasible in the setting of an outpatient oncology clinic. Currently, there is no consensus within the geriatric or oncology community regarding a standard geriatric assessment measure for older patients with cancer. Furthermore, these two disciplines often disagree about what constitutes an appropriate assessment secondary to different goals, time constraints, and patient populations. The standard assessment performed by a geriatrician, requiring up to 2 hours, may not address the needs of the oncologist, who requires a brief assessment to specifically identify the seemingly fit-appearing older individual whose limited functional reserve places him/her at risk for toxicities and, perhaps, poorer outcomes overall. The goal of the current pilot study was to develop an assessment measure for older patients with cancer that meets the needs of the oncology community.
Certain common measures of assessment in the geriatric population demonstrated a ceiling effect in this ambulatory population of older patients with cancer receiving chemotherapy. That is, the likely selection bias for healthier elderly patients in referral for chemotherapy yielded relatively few with deficits severe enough to be detected by these tools. These included the IADL, Timed Up and Go, and the HADS. Others demonstrated more variance (MOS Physical Health, MOS Social Activities Limitations Measure, and risk of nutritional problems manifested as a greater than 5% weight loss in 6 mos). This suggests that chemotherapy is offered or given to a functional group of older patients. Despite this, greater than one-fourth of patients experienced a fall in the last 6 months, a common geriatric syndrome. Therefore, a geriatric assessment devised to predict functional outcomes in older patients receiving chemotherapy will need to include measures that differentiate among patients of higher functional status, but also address the common problems faced in a geriatric population.
A limited number of investigators have reported on the use of a geriatric assessment for older patients with cancer, each using different assessment tools and administration methods. Balducci29 studied geriatric assessment in the older patient with cancer, including an evaluation of functional status, serious comorbidity (by Charlson scale and Cumulative Illness Rating Scale-Geriatrics), memory impairment, poor nutrition, and polypharmacy. Ingram et al.30 reported on the feasibility of a self-reported assessment, including questions regarding demographics, comorbid conditions, functional status, pain, financial well-being, social support, emotional state, spiritual well-being, and quality of life.
In comparison to the geriatric assessment performed by these other investigators, the geriatric assessment proposed in our study includes a more comprehensive assessment of physical functioning (including a performance-based functional status measure and questions regarding higher-order physical functioning and the ability to perform more rigorous physical activities) and assessment of cognition (not included in the assessment by Ingram et al.). Each of these domains is critical in assessing older patients with cancer. Our goal for future studies is to determine whether these measures can be used to predict an older patient's ability to tolerate and comply with a rigorous treatment course.
There are limitations to the current study. The study sample consisted of a highly educated, primarily white group of patients who were receiving chemotherapy. The feasibility of this tool in a less educated, more ethnically diverse, or potentially sicker population not offered chemotherapy needs to be determined. In addition, because this is a self-report questionnaire, there is no validation of the patient's responses. Last, the majority of the sample was derived from a single tertiary care cancer center. The feasibility in a multicenter trial needs to be studied. A prospective feasibility study is planned by the Cancer and Leukemia Group B, via the Cancer in the Elderly Committee, to test the feasibility of this assessment measure in a broader group of patients. If this study demonstrates general feasibility, prospective trials will be developed to determine this assessment's ability to: 1) predict the risk of serious toxicity to chemotherapy, and 2) help physicians address geriatric issues not covered by a routine history and physical examination, thereby tailoring therapy and potentially improving overall outcomes. The goal is to develop a geriatric functional index, identifying the key questions that will identify an older patient's physical, cognitive, and psychosocial functioning to determine those who would benefit from referral to a geriatrician.
In summary, the information provided by a geriatric assessment would be potentially valuable for several reasons. First, it would help the clinician get a sense of the functional age of a patient. Such information can help a physician tailor therapy and perhaps address issues that might not be obvious from a routine physical and history. In this way, the patient's therapy tolerance could be improved as well as overall outcome. Second, it would identify patients at higher risk for functional decline or toxicity due to treatment, so that interventions to lessen these effects could be implemented. Third, it would provide valuable information regarding older patients in clinical trials, allowing a comparison of patient characteristics across studies and a control for possible confounding factors contributing to mortality.
The authors thank the patients who enrolled in the current study and the following individuals who also contributed: Christopher Azzoli, M.D.; George J. Bosl, M.D.; Gabriella D'Andrea, M.D.; Maura Dickler, M.D.; Jeremy Kortmansky, M.D.; Lee Krug, M.D.; Diana Lake, M.D.; Vincent Miller, M.D.; Craig Moskowitz, M.D.; Mary Ellen Moynahan, M.D.; Carol Portlock, M.D.; Naiyer Rizvi, M.D.; Nancy Sklarin, M.D.; and Carol Pearce, M.F.A.
- 13Physical functioning measures. In: StewartAL, WareJEJr., editors. Measuring functioning and well-being; the Medical Outcomes Study approach. Durham, NC: Duke University Press, 1992: 86–101., .
- 14The clinical evaluation of chemotherapeutic agents in cancer. In: MacleodCM, editor. Evaluation of chemotherapeutic agents. New York: Columbia University Press, 1948: 191–205., .
- 27Measuring functioning and well-being: the Medical Outcomes Study approach. Durham, NC: Duke University Press, 1992.,