These and other improvements are briefly explained below.

Layout

All responses to questions in Version 2.0 are printed in a left-to-right (also referred to as “horizontal”) format, rather than with the mixture of horizontal and vertical listings of response choices that were printed below questions in the MOS and in the original SF-36. Mixed formats of response choices confuse respondents and cause missing and inconsistent responses, particularly among the elderly. Other improvements include more consistent use of indenting, numbering of instructions, deletion of useless item labels, and a simpler formatting of boxes that are checked by respondents.

Type-size and Bolding

A larger type size has been adopted throughout. Only instructions, as opposed to response choices, are bolded to simplify the “look and feel” of Version 2.0. These and other refinements were adopted on the basis of lessons learned in health care and from surveys in other fields.

Wording Changes

Evidence from numerous focus group studies, formal cognitive tests, and from empirical studies in more than a dozen countries support the improvements in item wording and the changes in some terms used to identify health concepts adopted in Version 2.0. These improvements make the English-language SF-36 easier to understand and administer as well as making it more objective. Version 2.0 is also more comparable with translations of the SF-36. Because most of the improvements in item wording were developed during the process of translating and adapting the SF-36 for use in other countries during the International Quality of Life Assessment (IQOLA) Project, Version 2.0 is sometimes referred to as the “international version”.

Five-Choice Response Scales

There is considerable empirical evidence that the Version 2.0 five-level response scales substantially improve the two SF-36 role functioning scales. Version 2.0 response scales extend the range measured and greatly increase score precision without increasing respondent burden. Specifically, Version 2.0 achieves a four-fold increase in the number of levels defined by both role scales, a substantially smaller standard deviation, and substantially reduces the percentage of respondents who score at both the ceiling and floor for both role scales. The elimination of one of the six response choices (“a good bit of the time”) from the MH and VT items was based on the finding that this response choice is not consistently ordered between adjacent categories in studies of item responses in Version 1.0 or in translations of the SF-36. Eliminating this choice simplified the format of the form with little or no loss of information.

Scoring and Norms

With the release of SF-36v2, norms were updated using data from the 1998 National Survey of Functional Health Status (NSFHS) and norm-based scoring (NBS) algorithms were introduced for all eight scales (Ware et al., 2000). NBS, which employs a linear T-score transformation with mean = 50 and standard deviation = 10, makes it possible to meaningfully compare scores for the eight-scale profile and the physical and mental summary measures in the same graph. SF-36v2 scoring software also yields less biased estimates of missing responses and makes it possible to estimate scores for more respondents with incomplete data (Kosinski, Bayliss, Bjorner, & Ware, 2000).

Comparability of Results

To make Version 1.0 easier to interpret and directly comparable to published results based on Version 2.0, cross-sectional and longitudinal norms for general and specific populations were re-estimated for Version 1.0 using NBS for all eight scales and for the two summary measures. Further, national calibration studies were fielded in the U.S. in 1998 and 1999 to evaluate the effect of all improvements and to assure the comparability of average scores across Versions 1.0 and 2.0 (Ware et al., 2000).

Acute (1-week recall) Form

The SF-36 is now available in both standard (4-week) and acute (1-week) recall versions. The more recently developed acute form was designed for applications in which health status would be measured weekly or biweekly. It was created by changing the recall period for six of the eight scales [Role-Physical (RP), Bodily Pain (BP), VT, Social Functioning (SF), Role-Emotional (RE) and MH] from “the past four weeks” to “the past week”. Two scales, Physical Functioning (PF) and General Health (GH) do not have a recall period; the items and instructions for these scales are identical across acute and standard forms.

The rationale behind a form with a 1-week recall period was that it would be more sensitive to recent changes in health status. This hypothesis was tested by comparing results for both the 1-week and original 4-week recall forms administered three times during a clinical trial of treatments for asthma (Keller, Bayliss, Ware, Hsu, Damiano, & Goss, 1997). As hypothesized, answers to SF-36 questions with a 1-week recall period tended to be more responsive to recent changes in disease state as estimated using several clinical criteria defining the severity of asthma. For example, changes in acute (1-week recall) SF-36 scale scores were generally more highly related to 1-week changes in asthma severity. Of some concern, from a normative perspective, the study also revealed higher mean scores for the acute version scales in comparison with the standard form scales. One explanation offered was a lower prevalence of negative events during the shorter recall period defined by the acute form. If so, this potential difference in mean scores would have implications for the norm-based interpretation of acute form scores. However, the findings from this one clinical trial were not replicated during the 1998 norming of the acute and standard forms in the general U.S. population (Ware et al., 2000).

Psychometric Considerations

SF-36 Measurement Model

Figure 1 illustrates the taxonomy of items and concepts underlying the construction of the SF-36 scales and summary measures. The taxonomy has three levels: (1) items; (2) eight scales that aggregate 2-10 items each; and, (3) two summary measures that aggregate scales. All but one of the 36 items (self-reported health transition) are used to score the eight SF-36 scales. Each item is used in scoring only one scale.



The eight scales are hypothesized to form two distinct higher-ordered clusters due to the physical and mental health variance that they have in common. Factor analytic studies have confirmed physical and mental health factors that account for 80-85% of the reliable variance in the eight scales in the U.S. general population (Ware et al., 1994), among MOS patients (McHorney et al., 1993; Ware et al., 1994), and in general populations in Sweden (Sullivan, Karlsson, & Ware, 1995) and the UK (Ware et al., 1994). As of 1998, these studies had been replicated in more than a dozen countries (Ware, Kosinski, Gandek, Aaronson, Alonso, Apolone, Bech, Brazier, Bullinger, Kaasa, Leplege, Prieto, & Sullivan, 1998; Fukuhara, Ware, Kosinski, Wada, & Gandek, 1998).

Three scales (PF, RP, BP) correlate most highly with the physical component and contribute most to the scoring of the Physical Component Summary (PCS) measure (Ware et al., 1994). The mental component correlates most highly with the MH, RE, and SF scales, which also contribute most to the scoring of the Mental Component Summary (MCS) measure. Three of the scales (VT, GH, and SF) have noteworthy correlations with both components.

The importance of these findings is illustrated below in the discussion of empirical validity. Specifically, scales that load highest on the physical component are most responsive to treatments that change physical morbidity, whereas scales loading highest on the mental component respond most to drugs and therapies that target mental health.

Scaling and Scoring Assumptions

A major objective in constructing the SF-36 was achievement of high psychometric standards. Guidelines for testing were derived from those recommended for use in validating psychological and educational measures by the American Psychological Association, the American Education Research Association, and the National Council on Measurement in Education (APA, 1974). Extensive psychometric testing has been conducted on the SF-36 in the United States (McHorney, Ware, Lu, & Sherbourne, 1994; Garratt, Ruta, Abdalla, Buckingham, & Russell, 1993; Jenkinson, Coulter, & Wright, 1993; Wagner, Keller, Kosinski, Baker, Jacoby, Hsu, Chadwick, & Ware, 1995), other countries (Sullivan, Karlsson, & Ware, 1994; Rampal, Martin, Marquis, Ware, & Bonfils, 1994; Sullivan et al., 1995; Bullinger, 1995; McCallum, 1995). Using the same tests of scaling and scoring assumptions that were used in developing the SF-36, results have been compared across general population studies in 10 countries (Gandek & Ware, 1998).

On the strength of favorable results from tests to date, nearly all studies have used the method of summated ratings and standardized SF-36 scoring algorithms documented elsewhere (MOT, 1991; Ware et al., 1993). This method assumes that items shown in the same scale in Figure 1 can be aggregated without score standardization or item weighing. Standardization of items within a scale was avoided by selecting or constructing items with roughly equivalent means and standard deviations. Weighting was avoided by using equally representative items (that is, items with roughly equivalent relationships to the underlying scale dimension). All items have been shown to correlate substantially (greater than 0.40, corrected for overlap) with their hypothesized scales with rare exceptions (McHorney et al., 1994; Ware et al., 1993).

More recent studies using item response theory (IRT) have shown strong linear associations between the original SF-36 simple summated ratings scores and those derived from IRT models, except at the extremes, as would be expected (Haley, McHorney, & Ware, 1994; McHorney, Haley, & Ware, 1997; Raczek, Ware, Bjorner, Gandek, Haley, Aaronson, Apolone, Bech, Brazier, Bullinger, & Sullivan, 1998). Results from these IRT studies have also suggested that improvements in scales and scoring algorithms are possible, especially for the PF scale. These models have also revealed substantial increases in the range of scale levels measured by both of the SF-36v2 role functioning scales in comparison with the original versions of those scales (Ware et al., 2000). Among the practical implications are greater score precision and reduced concentrations of scores at the “ceiling” and “floor”.

Reliability and Confidence Intervals

The reliability of the eight scales and two summary measures has been estimated using both internal consistency and test-retest methods. With rare exceptions, published reliability statistics have exceeded the minimum standard of 0.70 recommended for measures used in group comparisons in more than 25 studies (Tsai, Bayliss, & Ware, 1997); most have exceeded 0.80 (McHorney et al., 1994; Ware et al., 1993). Reliability estimates for physical and mental summary scores usually exceed 0.90 (Ware et al., 1994). A review of the first15 published studies revealed that the median reliability coefficients for each of the eight scales was equal or greater than 0.80 except for SF, which had a median reliability across studies of 0.76 (Ware et al., 1993). In addition, a reliability of 0.93 has been reported for the MH scale using the alternate forms method, suggesting that the internal-consistency method underestimated the reliability of that scale by about three percent (McHorney & Ware, 1995).

The trends in reliability coefficients for the SF-36 scales and summary measures summarized above have also been replicated across 24 patient groups differing in socio-demographic characteristics and diagnoses (Ware et al., 1993; Ware et al., 1994); McHorney et al., 1994). While studies of subgroups indicate slight declines in reliability for more disadvantaged respondents, reliability coefficients consistently exceeded recommended standards for group level analysis. Reliability estimates consistent with these trends have been published in more than 200 studies, results from more than 30 test-retest studies have also been summarized (Turner-Bowker et al., 2002).

Standard errors of measurement, 95% confidence intervals for individual scores, and distributions of change scores from test-retest and one-year stability studies have been published for the eight SF-36 scales and for the two summary scores (Brazier, Harper, Jones, O’Cathain, Thomas, Usherwood, & Westlake, 1992; Ware et al., 1993; Ware et al., 1994). Confidence intervals around individual scores are much smaller for the two summary measures than for the eight scales (+/- 6-7 points versus +/- 13-32 points, respectively) (Ware et al., 1994). For purposes of the Medicare Health Outcomes Survey – a very large federal effort to monitor health outcomes across health care plans serving the Medicare population – psychometrically-based standards have been established for SF-36 scores used to classify changes (better, same or worse) in physical (PCS) and mental (MCS) component summary scores (NCQA, 2002). Estimates of sample sizes required to detect differences in average scores of various magnitudes have been documented for five different study designs for each of the eight scales and for the two summary measures (Ware et al., 1993; Ware & Kosinski, 2001; Ware et al., 1994).

Validity

Studies of validity generally support the intended meaning of high and low SF-36 scores as documented in the original user’s manuals (Ware et al., 1993; Ware et al., 1994). Because of the widespread use of the SF-36 across a variety of applications, evidence from many types of validity research is relevant to these interpretations. Studies to date have yielded content, concurrent, criterion, construct, and predictive evidence of validity.

The content validity of the SF-36 has been compared to that of other widely used generic health surveys (Ware et al., 1993; Ware, 1995). Systematic comparisons indicate that the SF-36 includes eight of the most frequently measured health concepts. Among the content areas included in widely-used surveys, but not included in the SF-36, are; sleep adequacy, cognitive functioning, sexual functioning, health distress, family functioning, self-esteem, eating, recreation/hobbies, communication, and symptoms/problems that are specific to one condition. Symptoms and problems that are specific to a particular condition are not included in the SF-36 because the SF-36 is a generic measure.

To facilitate the evaluation of concepts not included, the SF-36 users" manuals include tables of correlations between the eight scales and the two summary measures and 32 measures of other general concepts (Ware et al., 1993; Ware et al., 1994), as well as 19 specific symptoms. SF-36 scales correlate substantially (r=0.40 or greater) with most of the omitted general health concepts and with the frequency and severity of many specific symptoms and problems. A noteworthy exception is sexual functioning, which correlates relatively weakly with SF-36 scales and is a good candidate for inclusion in questionnaires that supplement the SF-36.

Because most SF-36 scales were constructed to reproduce longer scales, attention was initially given to how well the short-form versions perform in empirical tests relative to the full-length versions. Relative to the longer MOS measures they were constructed to reproduce, SF-36 scales have been shown to achieve about 80-90% of their empirical validity in studies involving physical and mental health "criteria" (McHorney et al., 1993).

The validity, and therefore the interpretation, of each of the eight scales and the two summary measures has been shown to differ markedly, as would be expected from factor analytic studies of their construct validity (see Figure 2) (McHorney et al., 1993; Ware et al., 1994; Ware, Kosinski, Bayliss, McHorney, Rogers, & Raczek, 1995). Specifically, the MH, RE, and SF scales and the MCS summary measure have been shown to be the most valid of the SF-36 scales as mental health measures. This pattern of results has been replicated in both cross-cultural and longitudinal tests using the method of known-groups validity. The PF, RP, and BP scales and the PCS summary have been shown to be the most valid SF-36 scales for measuring physical health. Criteria used in the known-groups validation of the SF-36, which include accepted clinical indicators of diagnosis and severity of depression, heart disease, and other conditions, are well documented in peer-reviewed publications and in the two users" manuals (Kravitz, Greenfield, Rogers, Manning, Zubkoff, Nelson, Tarlov, & Ware, 1992; McHorney et al., 1993; Ware et al., 1993; Ware et al., 1994; Ware et al., 1995).



The MH scale has been shown to be useful in screening for psychiatric disorders (Berwick, 1991; Ware et al., 1994), as has the MCS summary measure (Ware et al., 1994). For example, using a cutoff score of 42, the MCS had a sensitivity of 74% and a specificity of 81% in detecting patients diagnosed with depressive disorder (Ware et al., 1994).

Relative to other published measures, SF-36 scales have performed well in most tests published to date (Weinberger, Samsa, Hanlon, Schmader, Doyle, Cowper, Uttech, Cohen, Feussner, 1991; Brazier et al., 1992; Kantz, Harris, Levitsky, Ware, & Davies, 1992; Krousel-Wood & Re, 1994; Krousel-Wood, McCune, Abdoh, & Re, 1994). As cited in the SF-36 bibliography (Turner-Bowker et al., 2002), studies have compared the SF-36 with 225 other measures. Predictive studies of validity have linked SF-36 scales and summary measures to utilization of health care services (Ware et al., 1994), the clinical course of depression (Wells, Burnam, Rogers, Hays, & Camp, 1992; Beusterien, Steinwald, & Ware, 1996), loss of job within one year (Ware et al., 1994), 180-day survival (Rumsfield, MaWhinney, McCarthy, Shroyer, Villa Nueva, O’Brien, Moritz, Henderson, Grover, Sethi, & Hammerstein, 1999) and five-year survival (Ware et al., 1994).

Results from clinical studies comparing scores for patients before and after treatment have largely supported hypotheses about the validity of SF-36 scales based on psychometric studies. For example, clinical studies have shown that three of the scales (PF, RP, and BP) with the most physical factor content (Figure 2) tend to be most responsive to the benefits of knee replacement (Kantz et al., 1992), hip replacement (Kantz et al., 1992; Lansky, Butler, & Waller, 1992), and heart valve surgery (Phillips & Lanky, 1992). In contrast, the three scales with the most mental factor content (MH, RE, and SF) in factor analytic studies have been shown to be most responsive in comparisons of patients before and after recovery from depression (Ware et al., 1995); change in the severity of depression (Beusterien et al., 1996); as well as drug treatment and interpersonal therapy for depression (Coulehan, Schulberg, Block, Madonia, & Rodrigues, 1997).

The discovery that 80-85% of the reliable variance in the eight SF-36 scales led to the construction of psychometrically-based physical and mental health summary measures. It was hoped that they would make it possible to reduce the number of statistical comparisons involved in analyzing the SF-36 (from eight to two) without substantial loss of information. In both cross-sectional and longitudinal studies reported to date, this appears to be the case (Ware et al., 1994; Ware et al., 1995). The advantages and disadvantages of analyzing the eight-scale SF-36 profile versus the two summary measures are illustrated and discussed elsewhere (Ware et al., 1994; Ware et al., 1995).

Finally, the SF-36 self-evaluated health transition item (five response categories ranging from "much better" to "much worse"), which is not used in scoring the scales or summary measures, has been shown to be useful in estimating average changes in health status during the year prior to its administration. In the MOS, measured changes in health status during a one-year follow-up period corresponded substantially, on average, to self-evaluated transitions at the end of the year. Using the 0-100 General Health Rating Index (GHRI) scale (Davies & Ware, 1981) as a "criterion", those who evaluated their health as "much better" improved an average of 13.2 points. The average change was 5.8 points for those who reported that they were "somewhat better". An average decline of -10.8 was observed for those who reported that their health was "somewhat worse" and 34.4 for those reporting "much worse". (It should be noted that the latter category had only 29 patients.) Change scores for those choosing the "about the same" category averaged 1.6 points. These results are encouraging with regard to the use and interpretation of self-evaluated transitions at the group level. Pending results from ongoing studies of the reliability of responses to the SF-36 self-evaluated transition item, it should be interpreted with caution at the individual level. Additional results and their implications are discussed elsewhere (Ware et al., 1993; Ware et al., 1994).

Administration Methods and Scoring

The SF-36 is suitable for self-administration, computerized administration, or administration by a trained interviewer in person or by telephone, to persons age 14 and older. The SF-36 has been administered successfully in general population surveys in the U.S. and other countries (Ware, Keller, Gandek, Brazier, & Sullivan, 1995), as well as to young and old adult patients with specific diseases (Ware et al., 1993; McHorney et al., 1994). It can be administered in 5-10 minutes with a high degree of acceptability and data quality (Ware et al., 1993). Indicators of data quality that have yielded satisfactory results in studies to date include very high item completion rates and favorable results for a response consistency index based on 15 pairs of SF-36 items, which is scored at the individual level (Ware et al., 1993). Computer administered and telephone voice recognition interactive systems of administration are currently being evaluated. Online administrations and scoring of SF-36 forms are demonstrated on the Internet.

Summary Measures

Table 1 summarizes information about the eight SF-36 scales and two summary measures that is important in their use and nterpretation. The eight scales are ordered in Table 1 in terms of their factor content (i.e., construct validity) as they are in the SF-36 profile to facilitate interpretation. The first scale is PF, which has been shown to be the best all around measure of physical health; the last scale, MH is the most valid measure of mental health in studies to date (McHorney et al., 1993; Ware et al., 1993; Ware et al., 1994). Interestingly, MH and PF are the poorest measures of the physical and mental components, respectively. Scales in between are ordered according to their validity in measuring physical and mental health. The VT and GH scales have substantial or moderate validity for both components of health status and should be interpreted accordingly.

Table 1: Summary of Information about SF-36® Scales and Physical and Mental Component    Summary Measures 

Summary of Information about SF-36 Scales and Physical and Mental Component Summary Measures

Acknowledgements and References