The discriminative power of the EuroQol visual analog scale is sensitive to survey language in Singapore

The visual analog scale (VAS) in the EQ-5D-3L self-report questionnaire [1] is a single-item measure of global health that has demonstrated satisfactory psychometric properties in many populations [2‐5]. However, the Chinese version of the EQ-VAS exhibited weak construct validity in Singapore, a multi-ethnic urban country in South-East Asia. In two previous studies in Singapore [6, 7], expected associations between the EQ-VAS and other health or clinical measures were not observed among patients with rheumatic or Parkinson' disease who completed the Chinese EQ-5D-3L questionnaire; in contrast, the English version of the EQ-VAS showed good construct validity in the same studies [7, 8]. Hence, validity of the EQ-VAS among Chinese-speaking Singaporeans warrants further investigation.

The purpose of the present study was to investigate the construct validity of the EQ-VAS in Chinese-speaking patients with type 2 diabetes using data collected in a health survey of such patients in Singapore. Performance of the English version of the EQ-VAS was also assessed and served as a reference in this study.

A total of 335 and 298 participants completed the survey in Chinese and English, respectively. Demographic and health characteristics of the study sample are displayed in Table 1. Compared to participants who completed the survey in Chinese, participants completing the survey in English were younger, better educated, and more likely to be females and employed. Accordingly, more Chinese- than English-speaking patients reported one or more comorbidities (80.9% versus 70.8%, p = 0.003) and rated their health as 'fair' or 'poor' (38.8% versus 27.6%, p = 0.003). The majority of patients chose to complete the survey through an interviewer, although a larger proportion of English-speaking patients than Chinese-speaking patients completed the survey by themselves (29.2% versus 5.1%, p < 0.001).

For Chinese-speaking patients, the EQ-VAS was correlated with the EQ-5D-3L index (Spearman's correlation coefficient: 0.27) and self-assessed global health (Spearman's correlation coefficient: -0.51), suggesting convergent validity. However, there was no statistical difference in EQ-VAS scores between subgroups of patients known to differ in BMI, duration of diabetes, complication status, or comorbidity status in both univariate and multivariate analysis, suggesting poor known-groups validity (Table 2). For example, the multiple regression analysis showed that the difference in EQ-VAS score between patients with and without any comorbidity was 1.7 (p > 0.05, t-test) after adjusting for socio-demographic status. Subgroup analyses suggested that known-groups validity was not better among patients with formal education than those without formal education (Table 3), or among younger patients than older patients (Table 4).

In contrast, the EQ-VAS demonstrated both convergent and known-groups validity among patients who elected to complete the survey in English. Spearman's correlation coefficient was 0.31 between the EQ-VAS and EQ-5D-3L index and -0.56 between the EQ-VAS and self-assessed global health. Patients with 1 or more diabetes-related complications or comorbidities had lower EQ-VAS scores than those without such conditions, and patients who had diabetes for < 10 years had higher EQ-VAS scores than those who had diabetes for 10 or more years. Those differences were statistically significant even after controlling for the effect of socio-demographic status in the multiple regression models (Table 2). It was also as hypothesized that non-obese patients had higher EQ-VAS scores than obese patients, although the difference was not statistically significant. It should be noted that the magnitude of the mean differences between the comparison groups was not larger (range: 3.0 to 7.8)

In the present study, the EQ-VAS exhibited poor known-groups validity among Chinese-speaking patients with diabetes, although convergent validity was demonstrated by correlations between the EQ-VAS and two other measures of overall health. In contrast, both convergent and known-groups validity were observed for the English EQ-VAS. Similar results were also observed for the EQ-VAS in patients with rheumatic diseases [6], Parkinson's disease [7], and breast cancer (Yin-Bun Cheung, personal communication). Therefore, it appears that the Chinese EQ-VAS is not a sensitive measure for self-assessment of overall health in Singaporean patient populations. To the best of our knowledge, no previous studies questioned the sensitivity of the EQ-VAS in specific or the visual analog scale in general.

Our finding from the present study has some important implications. First, our study highlighted the importance of psychometric testing for health-status instruments. Good measurement properties of an instrument in one population may not necessarily be generalized to other populations especially those multi-cultural populations. This is true even for widely used simple instruments such as the VAS. Herdman et al. pointed out that measurement equivalence across language versions should be examined in cross-cultural application of health-related quality of life instruments [11]. Second, our study suggested that the EQ-VAS is not a sensitive measure for Chinese-speaking patients in Singapore. Although being a valid measure, the EQ-VAS may not be able to detect true differences between groups when such differences are small. When a measure for overall health is needed for this population, the EQ-5D-3L index may a better choices as it demonstrated better known-groups validity in the present study (see Table 5). Third, we can reasonably suspect that other variants of the VAS used in clinical research or practice in Singapore may suffer from similar problems when they are applied to Chinese-speaking patients. Since no previous studies have looked into the psychometric properties of other VAS variants in Singapore, investigators should be cautious when interpreting data collected from Chinese-speaking patients using such scales.

It is intriguing why the EQ-VAS performed differently among Chinese- and English-speaking Singaporeans. We thought older age and poor education might be the reasons as those were the main differences between Chinese- and English-speaking patients. We speculated that some older patients or poorly educated patients might not know how to use the EQ-VAS for self-rating because of age-related cognitive impairment or poor numeracy, respective. However, our results did not support this hypothesis; the insensitivity of EQ-VAS to different health status was not associated with education or age (Tables 3 and 4). Although determining the real reason for the observed results is beyond the scope of the present study, the possible reasons should be related to different response styles of the respondents. It may be that Chinese speakers in Singapore have some idiosyncratic response style such that relatively healthy Chinese-speaking patients score their own health lower than their English counterparts on the VAS. Chinese philosophies such as Middle Way [12] may make practitioners avoid using high or low VAS scores to describe their own health. Chinese people might be reluctant to say their health is very good because they are afraid that God may punish them for not being humble [13]. However, we cannot rule out the possibility that the poor performance of the Chinese EQ-VAS was due to the suboptimal translation of the instructions and anchor labels of the instrument. Future studies using qualitative research methods such as focus group discussion should be conducted to elicit the causes for the poor sensitivity to difference of the Chinese version of the EQ-VAS.

There were some limitations in our study. First, all data used in our study were self-reported data. Chinese-speaking patients might have reported less accurate information on complication or comorbidity profiles than English-speaking patients because they were older. If this was the case, the known-groups validity of the Chinese EQ-VAS would have been underestimated in this study. Second, the internal validity of our results might have been affected by respondents' self-selection of survey languages. It is possible that those bilingual respondents who choose the Chinese questionnaires happened to have different response style. Ideally, bilingual respondents were identified and randomized to complete the survey in English or Chinese. Third, the finding of our study only has limited external validity. Since our finding is purely based on patients with diabetes, it may not be generalized to all patient populations in Singapore. Nevertheless, similar results were also observed for rheumatic diseases, Parkinson's disease, and breast cancer. Additionally, our study may not be generalized to other Chinese-speaking populations such as Chinese in mainland China. A recent cross-sectional study of patients with diabetes in China found that the EQ-VAS score was associated with duration of diabetes and microvascular complications but not with BMI or macrovascular complications [14]. In spite of limited generalizability, our study demonstrated the necessity of cross-cultural validation of even simple health-status measures such as the VAS.

In conclusion, compared to its English counterpart, the EQ-VAS appears less sensitive to different health status in Chinese-speaking patients with type 2 diabetes in Singapore. Future studies using qualitative research methods are needed to ascertain the underlying reasons.