Comparing Psychometric Properties of the SF-36 and SF-12 in  Measuring Quality of Life among Adolescents in China: A Large Sample Cross-sectional Study


 Objective: By comparing the psychometric properties of the Short Form 36 (version 1, SF-36) and the Short Form 12 (version 1, SF-12), we supply evidence for the selection of instruments measuring the quality of life (QOL) and decision-making processes of adolescents in China. Methods: Stratified cluster random sampling was adopted, and the SF-36 was administered to assess QOL. The Pearson correlation coefficient was used to show correlation. Cronbach’s alpha and construct reliability (CR) were used to evaluate the reliability of SF-36 and SF-12, while criterion validity and average variance extracted (AVE, convergence validity) were used to evaluate validity. Confirmatory factor analysis was used to calculate the load factor for each item, then to obtain the CR and AVE. The Semejima grade response model (logistic two-parameter module) in item response theory was used to estimate item discrimination, item difficulty, and item average information for each item. Results: 19,428 samples were included in the study. The mean age of respondents was 14.78 years (SD = 1.77). High correlations between the corresponding domains and components of both scales were found. Reliability of each domain of the SF-36 was better than for the corresponding domain of SF-12. The domains of PF, RP, BP, and GH in SF-36 had good construct reliability (CR > 0.6). The criterion validities of SF-36 were a little higher in some corresponding dimensions, except for PCS. The convergence validities of SF-12 were higher than SF-36 in PF, RP, BP, and PCS. The items of BP, SF, RP, and VT in SF-12 had acceptable discrimination of items that were higher than in SF-36. The items’ average amounts of information of BP, VT, SF, RE, and MH in SF-36 and SF-12 were poor. Conclusion: Two component measurements of the SF-12 (PCS and MCS) appeared to perform at least as well as the SF-36 in cross-sectional settings in adolescence, but the reliability and validity of 8 domains of the SF-36 were better than those of SF-12. Some domains, for instance SF and BP, were not suitable for adolescents or need no further study.


Introduction
Youth involves identity-building. Experiences during this developmental period can shape long-term attributes and attitudes and may lead to the adoption of a lifetime of healthy or risky behaviors [1]. The determinants of current and future health and disease for adolescents span the social and psychological elds [2]. A deeper understanding of how adolescents view their lives allows a greater understanding of their present health. The health-related quality of life (HRQOL) of school adolescents has been the subject of international interest. The term refers to a comprehensive model of subjective health that covers physical, social, psy chological, and functional aspects of individual well-being as a multidimensional and subjec tive construct [3,4]. The point of all this interest is to guide the organization of resources and decision-making processes to promote adolescents' quality of life. To accomplish this, understanding the current quality of adolescents' life is essential [5,6].
The SF-36 was developed and validated as a generic short-form instrument for measuring HRQOL; it was widely applied to assess important QOL domains in the Medical Outcomes Study [7]. The SF-36 consists of eight QOL domains: PF, physical functioning; RP, role physical; BP, bodily pain; GH, general health; VT, vitality; SF, social functioning; RE, role emotional; MH, mental health; with two summary scales having been constructed to summarize the physical and mental components (PCS and MCS, respectively) [8]. The factor structures of SF-36 that have been identi ed in China suggest that PCS is primarily a comprehensive measure of PF, RP, BP, and GH and MCS is mainly encompassed in the VT, SF, RE, and MH subscales. However, the two summaries somewhat overlap, and especially the VT, GH, and SF scales have noteworthy correlations with both components [9].
One of the major advantages of using the SF-36 is that it allows for QOL scores to be compared to scores in different groups [10], However, because the SF-36 was not originally designed to measure important QOL domains speci c to adolescents, some studies have determined the scale, especially the mental component summary, to be relatively insensitive to variations in different populations over time [11][12][13].
A substantially shorter questionnaire, the SF-12 was developed by Ware and colleagues, reducing the number of items from 36 to 12 to create an abbreviated version of the SF-36. [14,15]. Most of the respondents testing the new scale completed the SF-12 in less than a third of the usual time needed to complete the SF-36 [8]. Ware showed that the two instruments are highly correlated, and about 90% of the variation in both the physical and the mental component summaries measured in the SF-36 was explained by the same summary measures of the SF-12 [16]. Subsequent studies comparing the two scales have suggested varying results based upon the disease or health condition of interest [17][18][19]. The SF-12 and SF-36 are available in many languages and have been applied to all kinds of groups, including adolescents [20]. Although studies have demonstrated that both scales are valid instruments for this age group, they have rarely been used to evaluate the QOL of adolescents in China. In fact, few studies have focused on the quality of life of healthy adolescents in China [2,21].
Most studies of adolescent QOL in China have surveyed the per ception of QOL among chronically ill adolescent patients and were conducted in hospital or outpatient settings [22,23]. Recently there has been a growing interest in the study of healthy groups of adolescents, leading to studies being performed in other contexts, such as in schools [24,25], because of a growing awareness of the need to recognize and monitor adolescents who are most vulnerable to a poor healthrelated quality of life [26,27]. In some studies, though the SF-12 and SF-36 were used to investigate per ceived adolescent QOL, it was unclear which of the two scales was more suitable to the age group [21]. Thus, our study aimed to evaluate the QOL of healthy adolescent students at schools in China by using the SF-36 and SF-12 and comparing the psychometric properties of each, supplying evidence for the selection of instruments measuring quality of life and decision-making processes and thereby promoting the quality of life of adolescents.

Study design and sample
Strati ed cluster random sampling was adopted [28], rst dividing regions by geographical location: Dongguang, Shanghai, Shenyang, Wuhan, Xi'an, and Kunming represented the south, east, north, central, northwest and southwest regions, respectively. These areas were chosen in order to ensure proper representation by including participants from geographically diverse areas. Second, middle schools were randomly selected and followed by grade ( rst grade of junior school to third grade of high school). The basic sampling frameworks were all middle schools, as reported by each city. In each city, middle schools were selected by simple random sampling according to a random number table. Finally, 17 middle schools were included (4 in Dongguan, 1 in Shanghai, 3 in Shenyang, 1 in Wuhan, 4 in Xi'an and 4 in Kunming). The number of schools in each city was limited by the research group's local investigative capacity.
All students enrolled from the rst grade of junior high school to the third grade of high school were included in the survey. The exclusion criteria were those with any physical or mental condition that made them unable to complete questionnaires or students and their parents who had not signed an informed consent form. The study was approved by the Institutional Review Board (IRB) at the A liated Hospital of Guangdong Medical University. Verbal informed consent for publication was obtained from the participants and/or their relatives, as approved by the IRB. The response rate was almost 80%. This present study included 19,428 adolescents with complete information on quality of life measures. The sample sizes for each region were Dongguan (4,490, 23.1%), Shanghai (1,039, 5.3%), Shenyang (3,539, 18.2%), Wuhan (1,371, 7.1%), Xi'an (4,197,21.6%) and Kunming (4,792, 24.7%)

Instruments and variables
SF-36 (version 1) was used to assess QOL. Based on the response to individual items comprising the 8 subscales and using a z-score transformation, the scores of each subscale were calculated. [29]. First, the subscale items were coded; second, the items were scored; and nally, the scores were converted as shown in Formula 1. (see Formula 1 in the Supplemental Files) Using Z-transform scores and factor score coe cients (based on American norms), we calculated PCS and MCS scores according to Formulas 2 and 3: (see Formulas 2 and 3 in the Supplemental Files) SF-12 component summary scores (eight subscales, PCS-12, and MCS-12) were calculated using the SF-12 items that are embedded in the SF-36 [30]. This method has been presented as being equivalent to calculating the SF-12 as a standalone questionnaire [17]. All summary scores range from 0-100, where higher scores indicate better QOL.

Statistical analysis
For descriptive analyses, we aimed to show overall demographics and QOL. We calculated average and standard deviations in QOL scores by SF-36 and SF-12. For testing their relevance, the Pearson correlation coe cient was used to show correlation between the subscales of SF-36 and SF-12.
Cronbach's alpha for domains composed of multiple items and construct reliability (CR) were used to evaluate the reliability of SF-36 and SF-12, and validity indicators were represented by criterion validity and convergence validity (average variance extracted, AVE). Criterion validity was expressed by the correlation between the response of each subscale and self-reported health status. The calculation of the formulas for CR and AVE are shown in Formulas 4 and 5. (see Formulas 4 and 5 in the Supplemental Files) The sample was randomly split into a training set (50%) and a validation set (50%) to examine the construct validity of the SF-36 and SF-12. Using the training set, an exploratory factor analysis (EFA) was performed to explore the latent structure based on correlations matrix, and factor loadings (λ) were estimated by maximum likelihood estimation and rotated by promax. Using the validation set, con rmatory factor analysis (CFA) was used to validate the identi ed two-factor structure in some Chinese populations. The weighted least-square method was used for the estimation of CFA parameters. Factor loadings (λ) were taken for standardized regression coe cients. The classic goodness-of-t χ 2 statistic and its degrees of freedom were reported. However, as the χ 2 statistic is highly sensitive in large samples, assessment of goodness-of-t was based on the t indices as recommended: the root mean square error of approximation (RMSEA, close to 0.06 or lower) and the comparative t index (CFI, close to 0.95) [31]. The basic two-factor CFA model (model I ) without correlated errors was rst assessed (PCS associated with PF, RP, BP, and GH, whereas MCS associated with VT, SF, RE, and MH). Subsequently, the factor structures PCS and MCS, associated with most of the 8 domains (model II ) as described above, were also incorporated [21]. Then, the EFA or CFA was repeated on another data set, and mean estimates were reported.
According to the evaluation results of the samples, and taking into account the characteristics of the ordered and multicategory forms of the scale items, the Semejima grade response model (logistic two-parameter module) in item response theory was used to estimate item discrimination, item di culty, and item average information of each item [32]. RStudio, Amos 20.0, and Multilog 7.03 were used to process data.

Sample characteristics
Of the 20,226 questionnaires received, 798 had no responses on some of the SF-36 items. In the end, 19 49.4% (9,595) were boys. Among the SF-36 and SF-12 sub-scales, the physical functioning (PF) mean score was the highest, and the role emotional (RE) mean score was the lowest. The biggest mean difference in scores between the two scales was in the social functioning domain (SF). Of the corresponding domains of the two scales, the RE dimensions were the most relevant (r = 0.923), while the smallest correlation coe cient was in the VT dimension (r = 0.670), which means domains of the SF-12 scale could re ect the information of 67.0% to 92.3% of the corresponding domains of the SF-36 scale (Table 1).  Table 2.
Similarly, the construct validity of the SF-12 was also good in adolescents; the Kaiser-Meyer-Olkin Measure of Sampling Adequacy was 0.732. Eight components were extracted and explained 63.50% of the total variance. Due to the strong correlations between MH and SF and between VT and MH, the domains of SF, VT, and MH were not divided into identi ed structures in SF-12 (Table 3). Table 3 Results of factors analysis of SF-12 among adolescents (n = 9,741)

Factor analysis by CFA
We con rmed two conceptual models. Conceptual Model I assumed that PCS was associated with PF, RP, BP, and GH, whereas MCS was associated with VT, SF, RE, and MH. Conceptual Model II assumed that PCS and MCS were associated with most of the 8 domains. Fit indices of the two models revealed that no matter whether SF-36 or SF-12 were used, Conceptual Model I was better than Conceptual Model II in the structures identi ed ( Table 4). The structure of this model has been used widely in studies in China. In our study, we selected the structures of Model I as the two summary scales (PCS and MCS) of SF-36 and SF-12. Standardized parameter estimates for CFA on each path are shown in Figure 1.

Validity and reliability of domains of SF-36 and SF-12
As mentioned above, standardized parameter estimates for CFA in Model I was selected as factor loading. CR and AVE were calculated according to Formulas 4 and 5.
Except for SF, scales composed of multiple items had generally acceptable internal reliability ( Table 2). The low internal reliability of SF was probably because of inconsistent understanding of the meaning of the only two items in that subscale, which may be biased or di cult to parse for adolescents ("To what extent has your physical health or emotional problems interfered with…" and "How much of the time has your physical health or emotional problems interfered with…"). Moreover, consistent with related studies, the internal reliability of MH in SF-12 was low (0.369). On the other hand, the internal reliability of SF-36 in each domain was better than that of the corresponding domains of SF-12, which was consistent with higher internal reliability due to more items. The domains of PF, RP, BP, GH, and PCS in SF-36 had good construct reliability (CR > 0.6). Except for RP and PCS, the domains in SF-12 were not good at construct reliability.
Criterion validity was calculated based on the item of self-reported health ("In general, would you say your health is…."). It is worth noting that criterion validities of all the domains of the two scales were low, but especially so for PF, RP, and SF, which suggested that the correlation between physical health and self-perceived health was weak. Moreover, in PCS, the criterion validity of SF-12 was much higher than the criterion validity of SF-36. Although the criterion validities of SF-36 were higher in other corresponding dimensions, the differences were small. PF, RP, BP, and PCS had generally acceptable convergence validity whether in SF-36 or SF-12. Moreover, in these domains, the convergence validities of SF-12 were higher than SF-36, while there was little difference in other domains (Table 5).

Psychometric properties in item response theory
The parameter values and information content of the items according to the Samezima grade response model are shown in Table 6. The discriminations of items were between 0.45-2.73, with a large gap. The di culty of the items ascended from the lowest level to the highest level unidirectionally, which meets the di culty assumptions estimated by the model. The average amount of information of each item was between 0.07-1.02.
In SF-36, the domains of PF, RP, GH, and RE had acceptable discriminations of items (> 1), but the remaining dimensions were less differentiated, especially BP and SF, probably because for teenagers, there was strong homogeneity between individuals in terms of physical pain and social function. On the other hand, in SF-12, BP, SF, RP, and VT had higher discriminations of items than in SF-36.
With reference to the relevant literature, the amount of information measured on the scales > 25 indicated that the quality of the evaluation items was good; the amount of information < 16 indicates that the evaluation items were poor [28].
Given the number of items on the scale for SF-36, we divided 25 and 16 by 36 to get the average information amount for each item, so as to obtain the determination criterion: the average information amount of excellent items was > 0.69 (25/36), while items < 0.44 (16/36) were judged to be poor. Similarly, for SF-12, the average information amount of the excellent items was > 2.08, while items < 1.33 were judged to be poor. Except for PF05 and PF09, the items of the PF domain in SF-36 were excellent, and the items of the GH domain in SF-36 were excellent too, though the items of BP, VT, SF, RE, and MH were poor. On the other hand, the average amounts of information in SF-12 items were poor.

Discussion
Psychometric standards were used to evaluate the reliability and validity of the standard Chinese SF-36 and SF-12 scales in a large sample of Chinese adolescents. Our study suggested that the SF-12 and the SF-36 correlated very highly in this population. Although the reliability and average amount of information of the SF-12 domains and items were lower than that of the SF-36, the convergence validity and item discrimination were somewhat better. No matter whether the SF-36 or the SF-12 is considered, high correlations existed between some domains, for example, between MH and VT, SF dimensions, etc. The psychometric properties of the two broader components (PCS and MCS) were better than the individual domains.
Studies showed the two scales discriminated between adolescents with physical and mental health problems and performed well in associating with other clinical criteria [20,33,34] . A study of 31,357 adolescents in Hong Kong showed the two components and a single general health component of the standard Chinese SF-12 were appropriate health indicators for Chinese adolescents [21]. Studies have also shown that the SF-12 correlated highly with SF-36 in obese and non-obese patients [3,4]. However, many problems with the two instruments still existed, such as a high correlation between two components, low internal reliability, and the ceiling effect within individual domains [35]. Comparing the SF-12 and the SF-36, previous studies in patients with speci c diseases or health conditions have generally found moderate to high correlations between corresponding domains and components of both scales [19,36]. Our study also demonstrated these correlations. Since the SF-12 is embedded in the SF-36, we expected reasonably high correlations.
Overall, the dimensions of the SF-12 scale could re ect 64.5% to 92.3% of the corresponding dimensions of the SF-36 scale.
A low reliability and validity of the social functioning domain was also noted. This might indicate questionable reliability and validity of the instruments or the lack of representation [3]. On the other hand, it may also be attributed to the presence of inconsistent responses, which might occur if respondents completed a questionnaire without comprehending the items, which might occur with adolescents [21]. Due to the brevity of the SF-12 instrument, related research has shown that it is not possible to get reliable information for each of the eight domains, so that one would not be able to draw conclusions about speci c domains [37]. Indeed, we found the SF-36 was better than the SF-12 in terms of reliability. At the same time, comparing the SF-12 and the SF-36 in terms of validity, no loss in effectiveness was shown, and there was even a slight improvement. But we also found that the criterion validities of PF, SF, and MH were low, and self-reported health was our criterion. Relevant research has found that for most adolescents performing moderately strenuous activities or climbing several ights of stairs would not present problems because this age group is typically physically t and active, but when combined with a limited social life and less satisfactory mental state, inconsistent responses would be possible [21].
Unlike previous research [35,[37][38][39], we found the domains of BP and SF in general had poor discrimination of items, while PF in general, as well as BP, SF, RP, and VT on the SF-12, had higher discriminations of items than in SF-36. We suggest that, compared with PF items, the items in these other domains were not easy for teenagers to understand, resulting in a lack of sensitivity in the measurement. Similarly, a loss of information had been found in SF-12 that would be provided by the eight dimensions of the SF-36, but utilization of the two summary dimensions of SF-12 had an advantage for adolescents, which was consistent with the results of other population studies [21].
Methodological limitations should be mentioned. The participants were strati ed regarding geographical areas for minimizing the risk of possible regional differences. However, the regions chosen were vast and included small towns and big cities as well as rural areas [40,41]. Differences due to these circumstances might exist but not have come to light in this design. Additionally, there was a difference in response consistency between the samples because of the characteristics of adolescence, leading to bias in the results [42].

Conclusion
In general, our study suggested that the SF-12 correlated highly with the SF-36 in adolescent groups in China. If focus is restricted to the two broad component measurements (