The world’s older population grows at an unprecedented rate. In their old age, individuals may encounter health adversities including comorbidity and limitations in activities of daily living, and be disengaged from productive activities and social networks. Despite such challenges and adversities, they may still be satisfied with their conditions (1). Against this backdrop, assessment of quality of life (QoL) among older people has been regarded as an important subject in public health (2).

QoL is coined by the World Health Organization (WHO) as “individuals’ perception of their position in life in the context of the culture and value systems in which they live and in relation to their goals, expectations, standards and concerns” (3). The WHO has been advocating research on QoL, and initiatives include the 100-item WHOQOL-100 (3, 4) and its shorter version – the 26-item WHOQOL-BREF (5, 6). Subsequently, the EUROHIS-QOL 8-item index (7) has been derived based on the WHOQOL-100 and the WHOQOL-BREF. These instruments are generic measures of QoL and its applicability to older adults are questionable. For instance, they do not highlight main concerns in later life such as sensory abilities and autonomy. To address this problem, the 24-item WHOQOL-OLD has been developed with a focus on such concerns (8).

Nevertheless, the WHOQOL-OLD as a supplementary module needs to be administered together with the WHOQOL-BREF, and this combination seems too lengthy and overburdening to the more-impaired older respondents and imposes difficulties to assessment practices. Therefore, adapting the EUROHIS-QOL 8-item index (7) and the WHOQOL-OLD 6-item short form version 1 (9), researchers have advanced the 13-item WHOQOL-AGE as a standalone, time-saving measure of QoL of the older population (10).

To our knowledge, there exist three published studies of the factor structure of the WHOQOL-AGE. The first study was conducted on Europeans aged 18+ (10). Briefly, confirmatory factor analysis (CFA) identified a second-order model in which a second-order factor oversaw two first-order factors (Q1-Q8 on factor 1, items Q1 and Q9-Q13 on factor 2). Subsequent studies involved Europeans aged 18+ (11) and Chinese Taiwanese aged 70+ (2). Their CFA findings supported a bifactor model that incorporated two specific factors (Q1-Q8 on factor 1, items Q9-Q13 on factor 2) in addition to the general factor (that comprised all items). The methodological explanation is that specific factor 1 items adopt a bipolar rating format whereas factor 2 items adopt a unipolar format. Another explanation is that factor 1 items seem to capture satisfaction in personal asset while factor 2 items seem to capture self-efficacy in activities of daily living (2).

The knowledge of the factor structure of the WHOQOL-AGE remains problematic. Q1 (“How would you rate your quality of life?”), which is a generic item of QoL (11), should only load on the general factor but not on any specific factor with non-generic items (4). Moreover, the conceptual meanings of the specific factors are unclear. One factor assesses health (e.g., Q2) and external concerns including social relationships (Q6) and living conditions (Q7). Similarly, another factor assesses internal [e.g., autonomy (Q10)] as well as external [e.g., social relationships (Q13)] aspects. Of note, both factors cover social relationships. Perhaps due to these issues, these two factors were merely vaguely named as “factor 1” and “factor 2.” Furthermore, there is a lack of examination on the correlates of these two factors (10, 12–14).

The scale development literature dictates that the applicability of a measure is limited unless its factor structure is clear and its factors are interpretable and meaningful (15). In addition, the factors should show empirical specificity and are more or less differentially related to different correlates (16). With these principles in mind, here, we investigate the factor structure and validity of the WHOQOL-AGE. Compared with other model types (e.g., second-order models), bifactor models offer a number of advantages, one of which is recognizing multidimensionality while maintaining a unidimensional structure (17, 18). With the establishment of a bifactor model, researchers and practitioners can refer to the total score (general factor) and the scores of specific factors. Our first objective is thus, via factor analytic procedures, to establish a bifactor model of the WHOQOL-AGE, in which Q2 to Q13 load on the specific factors while all items including Q1 embed in the general factor (4). We study the measurement invariance of this model across gender and education level, two fundamental personal attributes that may shape interpretation of survey items (2).

Existing validation studies of the WHOQOL-Age have focused on its total score. For instance, concerning known-groups validity, individuals with chronic conditions were found to report a lower WHOQOL-AGE total score (10). Thus far, data on validity associated with the specific factors of the WHOQOL-AGE remains lacking. We are motivated to fill this gap because these data help ascertain factor structure and meaning (16). Our second objective is to revisit the known-groups validity of the WHOQOL-AGE. Following the usual practice in the QoL research (12, 19), we examine how demographic, social and health correlates relate to the identified factors of the WHOQOL-AGE. Pending better understanding of the specific factors, it is virtually illegitimate to detail formal hypotheses for the corresponding known-groups validity. Nevertheless, a reasonable speculation is, for example, that healthier (such as lower depressive symptomatology, fewer falls) individuals should score higher on the “health” factor (if found) than the less healthy counterparts. Moreover, health correlates should play a stronger role in the “health” factor than in other specific factors (19).

The two research objectives were addressed with a sample of Chinese aged ≥85 years in Singapore. The number of people aged 80+ is projected to increase threefold from 2019 to 2050 (20). South-East Asia is one of the regions where the oldest-old age group shows the largest percentage increases. It is meaningful to study QoL of the oldest-old (21), especially in South-East Asia. Thus far, research on the factor structure and validity WHOQOL-AGE exclusively among the oldest-old is lacking. Besides, this measure has rarely been scrutinized in Asian samples (2). Sociocultural and language issues may shape response patterns and affect scale development findings. It is desirable to evaluate factor structure and validity of a measure in samples with different backgrounds (11, 22). In this regard, the present examination of the WHOQOL-AGE among the oldest-old Chinese in Singapore is informative.

Among the 1,000 participants, the mean (standard deviation) age was 87.9 ± 2.4 years (between 85 and 97 years; Supplementary Table S1). Three hundred fifty-three individuals were men and 647 were women. Four hundred fifty-six individuals had received no formal education and 544 had received at least primary education. Supplementary Table S2 reports the correlation among the 13 WHOQOL-AGE items. All correlations were positive and significant (p < 0.01).

Against population aging, a brief and time-saving measure of QoL in old age is essential. The present study examined the factor structure (objective 1) and known-groups validity (objective 2) of the 13-item WHOQOL-AGE with a sample of the oldest-old Chinese in Singapore.

The WHOQOL-AGE exhibited a bifactor model that comprised three specific factors, namely “health,” “environment” and “mastery,” in addition to the general factor (“overall”). “Health” (about physical and psychological health) and “mastery” (autonomy, capacity, capital) focused on internal aspects, whereas “environment” focused on external issues (social relationships, activities, living conditions). The WHOQOL-AGE is partly rooted in the WHOQOL-100 (3, 4) which incorporates four domains: physical, psychological, social, environment. The resemblance between the observed specific factors of the WHOQOL-AGE and the conception of the WHOQOL-100 makes good sense.

The specific factors of the WHOQOL-AGE here are different from those reported previously (2, 10, 11). One possible account is that recognizing Q1 as a generic item (11), we did not put it under any specific factor (4). Besides, we applied the WHOQOL-AGE exclusively to the oldest-old. Mastery refers to capacity for participation and contribution, and perceived independence and usefulness (35). Perhaps because mastery is an increasing concern with age (36), it is more explicit and more readily identified in the oldest-old. In any event, compared with those reported in previous research, the factors identified in the present study are more meaningful and interpretable. Our findings better articulate what elements of QoL the WHOQOL-AGE captures, and this naturally enhances practicality and facilitates usage of this instrument.

The present bifactor model of the WHOQOL-AGE showed scalar invariance across education level. It also demonstrated partial scalar invariance across gender, with the intercept of Q13 (“How satisfied are you with your intimate relationships in your life?”) unconstrained. Women place a greater emphasis on intimacy than men (37), but they are less likely to remain married (12% women vs. 70% men in our dataset) because they have longer life expectancy and greater likelihood of being widowed (38). This backdrop may lead to a gendered interpretation and processing of Q13. Nevertheless, Q13 was the only item which intercept could not be constrained across gender. The general picture is thus that the present bifactor model holds across gender as well as education level.

Turning to the known-groups validity of the WHOQOL-AGE, BADL and IADL were positive correlates, and depressive symptomatology and falls were negative correlates of the “health” factor. Demographic and social correlates also played a role in “health.” Our measure of financial adequacy concerned the perception of having sufficient financial resources to pay for medical expenses. This should be why there was a positive association between financial adequacy and “health.” Living alone is a driver for lower health-related QoL (39). Indeed, we found living alone as a negative correlate of “health.”

Affective state, as depicted by depressive symptomatology, may be linked with diverse aspects of QoL (40). In line with the literature, our data showed a negative relationship of depressive symptomatology with not only “health” but also “environment” and “mastery.” Our data also revealed that apart from “health,” IADL was positively associated with “mastery” (36), which refers to autonomy and capacity (35).

Strength of social network and social engagement did not relate to “health.” On the other hand, strength of social network, which speaks to social connectedness in household and non-household settings, positively related to “environment” and “mastery.” Social engagement was a positive correlate of “mastery,” congruent with the wisdom that active participation in social activities may entail stronger perceptions of independence and usefulness (41).

Our multivariate findings showed that intact cognition was negatively associated with “health” as well as “overall,” but note that on a bivariate level, cognitive status was actually uncorrelated with these two factors (r = 0.01, p = 0.641; r = 0.05, p = 0.146). These data collectively indicate the emergence of the suppressor effect, a statistical phenomenon that is difficult to interpret (42). Some past studies have documented that cognitive function is largely unrelated to QoL in the oldest-old age on a multivariate level (21, 43). Taken together, the observed negative linkage of intact cognition with “health” and “overall” should not be overinterpreted.

In summary, “health,” “environment,” and “mastery” were accounted for by different arrays of correlates, with “health” being linked with more health correlates than the latter factors (19). At the same time, virtually all significant correlates of “health,” “environment,” and “mastery” were associated with “overall.” These data demonstrate the uniqueness of the specific factors and the known-groups validity of our bifactor model of the WHOQOL-AGE. That said, we call for more validation data of the WHOQOL-AGE – for example, its relationship with the WHOQOL-BREF and WHOQOL-OLD (convergent validity).

There are some caveats associated with the present study. It is not uncommon that published scales show low reliability (44), but one may be concerned that the Cronbach’s alpha of “mastery” was merely 0.62. Cronbach’s alpha is a function of number of items and average correlation between pairs of items (44). “Mastery” only showed marginally acceptable reliability likely because it only involved three items, which were, nevertheless, moderately positively intercorrelated (r > 0.3, p < 0.01; Supplementary Table S2). Given that the intercorrelations were moderate in size, and that CFA findings strongly supported the present bifactor model of the WHOQOL-AGE, the low reliability of “mastery” should not have created much noise and not be too problematic (45).

It has been recommended that the sample-size to parameters ratio for CFA should be at least 10:1 (16, 24). Our sample size just fulfilled this rule of thumb for the testing of measurement invariance in the context of CFA. Hence, we did not conduct EFA and CFA on subsamples (e.g., n = 300 for EFA; n = 700 for CFA) (15), but performed them on the whole sample (16, 46, 47). The latter practice is important in its own right–if congruence is not observed between EFA and CFA on the same data, it is unlikely that CFA will corroborate EFA findings in a different sample (46, 48, 49). In any case, we call for replication studies for the factor structure of the WHOQOL-AGE reported here.

Another issue is that the participants were survivors in a longitudinal cohort with follow-up of about 25 years. One may argue that our findings are limited to the relatively healthy oldest-old. Replication studies that adopt representative samples are needed. We also acknowledge that regression analysis was conducted with cross-sectional data. To develop better knowledge of known-groups validity, future research should analyze temporal causations with longitudinal data. In addition, scholars should study other forms of validity such as convergent validity and predictive validity.

This study examined the WHOQOL-AGE in Singapore, an Asian context. Our results suggested a good potential of the WHOQOL-AGE to be applied in other Asian societies, though accumulated investigations of this instrument in other regions and populations are surely needed. The cultural equivalence of the translated WHOQOL-AGE deserves more discussions. Singaporeans as Asians may be less likely to use the extreme values on scales than the Westerners (50). Besides, Singapore emphasizes family values. Our sample of the oldest-old Chinese in Singapore may consider the item on intimate relationships (Q13) of the WHOQOL-Age as sensitive information that they were unwilling to share with strangers. On the item about living conditions (Q7), all respondents were residents in public housing estates in Singapore at recruitment, and this may limit the heterogeneity in their responses to this item. Notably, nearly half of the studied oldest-old did not receive formal education. Hence, although the SG90 questionnaires were supposed to be administered in Mandarin during interviews, most of the interviews were conducted in the Hokkien or Cantonese dialect as a workaround, since the Chinese in this cohort were restricted to these two major dialect groups of Chinese in Singapore. Another language concern is that the Mandarin or dialects used in Singapore is not entirely the same as those used in Mainland China and Taiwan. Due to these sociocultural and language issues, our observations in the oldest-old Chinese in Singapore should not be mechanistically assumed to be generalizable to other samples.

The 13-item WHOQOL-AGE exhibits a bifactor structure and known-groups validity among the oldest-old Chinese in Singapore. Depending on their objectives, scholars and practitioners may use the whole WHOQOL-AGE scale (i.e., general factor), or focus on its subscales (specific factors). With good properties, flexibility and parsimony, the WHOQOL-AGE seems a promising measure of QoL in the concerned population.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving humans were approved by Institutional Review Board at the National University of Singapore (approval number H-17-027). The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.

GC designed the research, analyzed the data, and drafted the paper. QF provided feedback on the data analysis and result interpretation, and contributed to revising the paper. Y-SC and W-PK administered the data collection, and contributed to revising the paper. All authors contributed to the article and approved the submitted version.