Antiretroviral therapy improves health, increases patients’ survival and reduces the risk of HIV transmission.1 Health-related quality of life (HRQoL) has then become an area of interest in HIV treatment and follow-up. Measuring the HRQoL may be helpful not only to evaluate treatment response, effectiveness of treatment, patients’ decline or improvement, but also adherence.2

HRQoL is a multidimensional concept related to the impact of health on the individual's self-perception of wellbeing, considering sex, lifestyle, physical, social and mental aspects.3 Measuring HRQoL in HIV should meet a set of standard psychometric properties, produce accurate results, and be sensitive to treatment adverse effects.4

The 36-Item Short Form Health Survey questionnaire (SF-36) is a widely used, standard instrument for evaluating HRQoL.5 The institution that hold all SF-36 copyrights and trademarks have developed licensing programs for academic research and business applications that evaluate internal consistency, accuracy of data punctuation and adequacy of interpretation.5 The SF-36 was originally derived from the most frequent measured concepts included in the Medical Outcomes Study (MOS).5 In 1997, the SF-36 version 1 was translated into Brazilian-Portuguese and cross-culturally adapted.6

Although SF-36 normative data for the population in the south of Brazil have been proposed,7 the SF-36 version 2.0 (SF-36v2) in HIV-infected patients has not yet been validated in Brazil. The SF-36v2 was introduced in 1996 to improve the original instrument on its psychometrics proprieties when considering the two SF-36 role functioning scales.8 The second version was developed to improve the response options for the items concerning role physical and role emotional limitations, reducing the ceiling effects of these scales.8,9 We aimed to assess data quality, reliability and validity of the SF-36v2 as a measure of health-related quality of life among individuals living with HIV, in Brazil.

The study protocol was approved by Ethical Review Board (protocol number 1.035.826) and is in accordance with the Declaration of Helsinki 2013 and the Brazilian National Health Council Resolution 499/12. All volunteers signed an informed consent form.

The study enrolled 392 HIV-infected patients (219 males; 173 females), mean age of 44.2±11.5 years, CD4 T lymphocytes count of 701.7±1273.6 cells/mm3 and CD8 count of 1100.9±601.0 cells/mm3. Most patients were racially mixed (54.6%) followed by black (31.1%), did not have a stable relationship (70.7%), had a family income (69.4%) higher than the Brazilian Minimal Wage (284.6 USD) and nine or more schooling years (69.4%). One-hundred-sixty patients (40.8%) had one or more comorbidities and one-hundred-three (26.3) diagnosis of depression (Table 2).

Confirmatory factor analysis (CFA) was performed to evaluate the factor structure and dimensionality of SF-36v2. The hypothetical measurement model composed of 35 items distributed by eight factors, as theoretically predicted, was tested. The adjustment indexes obtained for this model were: χ2 (gl): 1225.80 (532); CFI: 0.923; GFI: 0.841; TLI: 0.913; RMSEA: 0.058 (90% IC: 0.054-0.062). To improve the adherence of data to the model, five parameters were included (Table 3).

After inclusion of the five new parameters, we obtained the following adjusted indices: χ2 (gl): 980.7 (527); CFI: 0.949; GFI: 0.873; TLI: 0.943; RMSEA: 0.047 (90% IC: 0.042-0.051). Figure 1 showed the measurement model.

Figure 1.

The measurement model for CFA combining all eight domains of SF-36v2 with standardized factor loads and correlation coefficients between factors.

(0.43MB).

Based on the measurement model, a structural model was tested, grouping the eight factors into two components summaries, as described in the method section. The adjustment indexes obtained for this model were: χ2 (gl): 1702.2 (543); CFI: 0.871; GFI: 0.826; TLI: 0.858; RMSEA: 0.074 (90% IC: 0.070-0.078). The final structural model is described in Figure 2.

Figure 2.

SF-36v2 final structural model, with standardized factor loads and correlation coefficients between Summaries Components.

(0.35MB).

Once the factorial structure of the SF-36v2 was confirmed, we calculated the internal reliability indices of each of the eight factors (Table 4). No floor effects were detected in all SF-36v2 domains, while ceiling effects were observed for the PF, RP, SF and RE.

All SF-36 normalized mean scores were systematically lower in female patients. However, only RP (0.030), VT (0.029), RE (0.009), MH (0.001) and PCS (0.002) were statistically significant (Table 5).

The association of the predictive variables with the PCS and MCS scores was explored using multiple regression analyses (Table 6). Depression (p=0.001), family income (p=0.007), sex (p=0.001) and age (p <0.021) were strongly associated with the variation in MCS; Comorbidities (p=0.001), health self-perception (p=0.001), age (p=0.001) and sex (p=0.025) were good predictors of the PCS.

The SF-36v2 is a widely used instrument to assess Health-Related Quality Of Life. This is the first paper to report the psychometric qualities, reliability and validity of the SF-36v2 in Brazil. A previous study reported the translation and validation process of the SF-36v16 and a second one suggested normative data in south Brazil.7 We analyzed the psychometric quality of the SF-36v2 instrument and evaluated the HRQOL in patients living with HIV.

We did not test an alternative model, such as analyzing a structural model composed of a single factor, since SF-36 is a bidimensional instrument. We opted for performing the CFA using the previously established theoretical model of the SF-36. 8,9 The SF-36v2 Manual9 provides two additional forms of calculation of the instrument: the first step considers the eight domains (PF, RP, BP, GH, VT, SF, RE and MH) and the second step generates two summary measures, one physical (PCS) and the other mental (PCS). We tested both models and the results evidenced a good psychometric quality indicating that they generate reliable indicators.

The hypothetical measurement model for the eight factors presented satisfactory factorial loads11 ranging from 0.38 to 0.94 and adherence rates close to satisfactory following the recommended criteria.12,14 The modification indexes indicated a high potential for improvement for the model fit. In total, five parameters were included and all of them theoretically justified by the proximity of content. It is important to notice that the entered parameters were related to PF domain items (3h Walk Several Hundred Yards, 3i Walk One Hundred Yards and 3g Walk Mile). However, these items have high semantic proximity suggesting overlapping and redundancy. Maybe withdrawing some of them would raise the model parsimony without damaging the measurement quality. However, since it was not the objective of the study to reduce the SF-36v2 instrument, we decided to maintain all of them. After model adjustment, all indicators presented satisfactory values, except GFI, which was slightly below the threshold (0.873, below the 0.9 parameter) but still meeting the requirement suggested by specialized literature.14,15

Scoring the eight scales using the original non-normalized 0 to 100 algorithms (raw scores) is commonly reported in the literature, but normalization allows meaningful comparisons among domains and unbiased identification and quantification of the SF-36 domains.5,9 The normalized scores and the two summary components are performed by specific Scoring Software9 based on norms with a mean of 50 and a standard deviation of 10. Normalized scores below 50 are interpreted as below the US general population t-score and otherwise above.

Considering the CFA indicators, a previous study16 described lower values (RMSEA 0.14, CFI, 0.83 and TLI, 0.75) than the present study. We tested the psychometric characteristics of the two summaries components model. The structural model was tested, which already took into consideration the parameters previously included in the measurement model. The RMSEA was satisfactory and the other indicators (CFI, GFI and TLI) were close to desirable values (>0.9).11 Once the structural model is more complex than the measurement model, it is difficult to achieve a perfect fit.

To estimate the scale scores of reliability, we used the internal consistency method.13 We opted to use both the Cronbach's alpha and composite reliability coefficients. Cronbach's alpha coefficient is the most widely adopted index in literature and therefore enables comparison with a larger number of studies. However, recent reviews have pointed this measure as a faulty indicator as it underestimate scale/factor consistency. Thus, the composite reliability is a more robust and accurate measure.13

Based on the adopted criteria, the factors PF, RP, RE and MH presented high reliability and the others, apart from SF, provided acceptable reliability indicators.11 The SF has only two items, which may affect the internal consistency. As scales presenting many items tend to have greater internal consistency,11 our SF coefficient was higher than those found by other authors.16–19 Although we did not observe floor effects in any SF-36v2 factors, four domains (PF, RP, SF and RE) presented ceiling effects. All these ceiling effects were much lower than those found in general North American population.20 The application of questionnaires was conducted during face-to-face interviews. This fact may have contributed to a lower ceiling effect compared to the US population.

Our results showed systematically lower scores in females for all SF-36 normalized mean scores. Multiple regression analysis evidenced the association of depression, family income, sex and age with Mental Component Summary. Comorbidities, health self-perception, age and sex were associated with the Physical Component Summary. These results are similar to our previous report that observed lower HRQoL scores in females2,21 and the association of comorbities3,21 with lower HRQoL physical components and depression with lower HRQoL mental components.2,3

Our study has some limitations. Among them, the cross-sectional design and lack of cross validation14 with independent samples, since the SF-36 is widely used with acceptable and higher psychometric proprieties.8,9 However, we were able to provide a consistent validation of the tool in Brazilian HIV-infected patients. The lack of information on SF-36 psychometric proprieties in Brazil, including in HIV patients also reinforces the need of studies in this field. To our knowledge, this is the first study to validate the SF-36v2 in Brazil. Further studies with SF-36v2 psychometric analyses would be required in other populations to establish Brazilian normative data.

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

The authors declare no conflict of interest.