The Healthy Ageing Questionnaire Index: Validation in the Singapore Longitudinal Ageing Study

In the face of a global challenge of population ageing, there is wide consensus that ageing societies should strive to enable older people to remain in good health and maintain their personal wellbeing and social involvement in the community [1, 2]. Thus, a cornerstone of ageing policy actions is the promotion and support of healthy ageing (aka successful ageing). There is general agreement that healthy ageing should be viewed holistically, encompassing physical, psychological, and social dimensions [3]. However, a consensual definition remains lacking [4, 5].

The Rowe and Kahn model of successful ageing has for some time epitomised this multidimensional construct by specifically encompassing the avoidance of disease and disability, the maintenance of high physical and cognitive functions, and sustained engagement in social and productive activities [3]. However, the requisite for the absence of disease has been questioned, since many older adults rate themselves to be ageing well even though they have one or more chronic diseases [6]. A recent change in conceptual focus based on the WHO policy action framework for healthy ageing thus emphasizes that being free of disease or infirmity is not a requirement for healthy ageing [1]. The WHO defined healthy ageing as “the process of developing and maintaining the functional ability that enables wellbeing in older age.” Functional ability consists of the intrinsic capacity of the individual (mental and physical capacities) interacting with factors in the environment that enable them to meet their basic needs, be mobile, build and maintain relationships, and contribute to society.

Accordingly, there has been a recent increase in research on developing new indices of healthy ageing based on this conceptual definition. Generically, they are based on measures such as mobility, sensory skills, cognition, vitality, psychological symptoms, and functional ability for performing activities of daily living [7‒11]. Of note, the indices are developed for different purposes and use, and measurements are made using elaborate questionnaire scales, and/or clinical, performance-based, or blood testing which are lengthy, time-consuming, costly, or cumbersome to perform. To our knowledge, there is no index based on a brief questionnaire of healthy ageing that has hitherto been developed and validated for ease and accuracy of use in research and healthcare.

We report here the development of a simple and brief Healthy Ageing Questionnaire (HAQ) index for assessing functional ability as defined above by the WHO and examined its psychometric properties by assessing its internal reliability, construct validity in relation to known determinants and correlates of healthy ageing, and concurrent validity with Rowe and Kahn’s construct of successful ageing (RK-SA). We investigated its predictive validity for survival, and compared its performance with that of the RK-SA.

The study was conducted among participants in a sub-cohort study, the Singapore Study of Successful Ageing (SSOSA) of the Singapore Longitudinal Ageing Study second wave recruitment cohort (SLAS-2). As described in previous publications [12], the SLAS-2 recruited over 3,200 middle-aged and older adults aged 55 years and above (response rate: 72%) in 2009–2011 for participation in a study of ageing and health transition. Exclusion criteria were limited to being intellectually and physically unable to participate in questionnaire interviews, clinical, physical, and cognitive performance testing, due to mainly terminal chronic diseases or severe cognitive impairment. An extensive range of medical, physiological, social, lifestyle, behaviour, psychological, physical, and functional performance data were collected for each participant. The SSOSA was conducted on a sub-cohort of 500 SLAS participants aged 60 and above who were residents in two localities and consented to an additional questionnaire interview for an in-depth study of successful ageing [13].

The reference framework for the development of the HAQ was the WHO conceptualization of healthy ageing and the proposed operationalization by a recent Newcastle workshop of recommendations of indicators of physiological and metabolic health, physical capability, cognitive function, social wellbeing, and psychological wellbeing that characterise the Healthy Ageing Phenotype [14]. The SSOSA used some 280 individual questions assessing functional ability for successful ageing pertaining to domains of social network and support, financial and food security, physical performance and functional mobility, sensory ability, cognitive ability, vitality, psychological attributes, subjective wellbeing, life satisfaction, functional ability in performing physical and socio-occupational activities of daily living, healthcare and social services, and others. For the purpose of deriving the HAQ index, questions on the presence of chronic diseases were excluded, as were performance-based test measurements (such as Mini-Mental State Examination [MMSE] or gait speed). Interviews were conducted face-to-face with English, Chinese, and Malay translated versions of the questionnaire, and in the spoken language or dialect in which the respondents are most familiar with: English, Mandarin, Hokkien, Cantonese, or Malay; respondents of Indian ethnicity are literate or conversant in English or Malay, and interviewed accordingly. The brief HAQ was created by statistically deriving a parsimonious subset of strongly representative question items measuring healthy ageing from among the original 280 items in the SSOSA questionnaire.

The original SSOSA questions were all single-item questions with binary or Likert response scoring. The prior step in the statistical selection process was to re-order the ordinal response scales of question items such that they uniformly conform to a positive valence measure of healthy ageing. For example, with the original question item “Do you forget appointments?” we re-ordered the original 5-point Likert scoring so that “never” was given the highest positive score of 4 and “very often” was given the lowest score of zero. The Likert scores for each individual item range from 0 to 4 (maximum). Missing data were present for less than 5% of the observations, and were imputed using overall group median or mode values of item variables.

The iterative selection process initially involved exploratory regression analyses of individual item variables which were singly evaluated for its ability to predict time to death. In preliminary rounds of univariate analyses, item variables that failed to predict survival with a statistical probability for inclusion of p < 0.10 were rejected. This resulted in the shortlisting of 30 item variables (p < 0.10) that were entered together as candidate variables for selection into the final multivariate model predicting longer survival. We used iterative stepwise (“step-in,” step-out”) procedures to select significant variables that were all significant at p < 0.05 for both entry and retention in the final multivariate prediction model. We also used formal conditional forward and backward selection procedures for model building. All approaches consistently produced the same final model consisting of 15 significant variables independently predicting survival.

These 15 question items constitute the HAQ scale (see Table 1 and online suppl. Material S1; for all online suppl. material, see https://doi.org/10.1159/000533635). We evaluated its factor structure and internal consistency using exploratory factor analysis and reliability analysis. The final HAQ scale based on the model has a potential summed score range of 0–47. For ease of interpretation, we rescaled the summed score so that it ranges from 0 to 100 by dividing the observed value by 47 and multiplying by 100 (100*observed score/47).

We validated the measurement performance of the 15-item HAQ among 2,161 individuals in the parent SLAS-2 cohort who did not participate in the SSOSA sub-cohort study and were aged 60 and above at initial interview. Of the 15 HAQ question items, six of them were identical in both the SSOSA and SLAS. We substituted the non-identical questions with closely similar questions. These are shown in the online supplementary Material S2. For example, the SSOSA question, “As I get older, I am less useful” (agree/disagree) was substituted with the SLAS-2 question, “Do you feel worthless the way you are” (yes/no). The question “How would you rate the overall quality of the healthcare you receive?” (rated 0 = poor, 1 = fair, 2 = good, 3 = very good, 4 = excellent) was not available in the SLAS-2 cohort study, and was given a default score of 2 = good.

In the SLAS-2 validation cohort, we assessed concurrent validity with respect to RK-SA, and construct validity by examining HAQ associations with known determinants and correlates of healthy ageing. These included demographics, chronic diseases, multi-morbidity, fall, hospitalization, life satisfaction, frailty, IADL-ADL disability, logMAR vision, MMSE global cognition, and lower limb strength and functional mobility. The predictive validity of the HAQ was evaluated with respect to its prognostic value for future survival probabilities.

Self-reports of medical diagnoses and treatments, inspection of medication packages, blood pressure measurement, and fasting blood glucose were used to ascertain the presence of common chronic diseases and health conditions including self-report of fall(s) and hospitalization(s) in the previous year. Multi-morbidity was assessed by dichotomising the number of chronic diseases (0–4 and ≥5). Functional dependency was assessed by the basic and instrumental activity of living (ADL) scales. The Frailty index was calculated from the number of health deficits present among 98 evaluable health deficits and expressed as a fractional count from 0 to 1 [15]. Pre-frailty and frailty were defined by FI of 0.10–0.20, and >0.20, respectively [16‒18]. Global cognition was assessed by a validated translated version of the MMSE [19]. Visual acuity was assessed by logMAR testing, physical performance on the “5 times sit-to-stand” test [20, 21], and “the timed-up-and-go” (TUG) test [22, 23]. Data were complete for all co-variables except for TUG (available in 2,037 participants), LogMAR (1,957 participants), and MMSE (2,131 participants). Data were deleted pairwise for these variables in analyses of construct validity based on different numbers of observations, assuming that the data were missing completely at random.

RK-SA phenotype was determined using binary scores (0 or 1) for the absence of basic and instrumental ADL dependency and major chronic diseases, high levels of MMSE cognitive performance (≥27) [24], social engagement (participated once a month or more in at least one of 8 social activities and 6 productive activities), and physical functioning (not limited at all in lifting or carrying groceries, climbing several flights of stairs, bending, kneeling, or stooping, walking more than a mile or several blocks). The total summed score ranged potentially from 0 to 5, and was analysed both as a continuum and as a binary score (5 = successful ageing) [25].

Time to death was determined by follow-up of vital status in the cohort using computerized record matching with the National Death Registry at the National Disease Register Office. Time-to-event was estimated from date of initial interview at recruitment to date of death or censored on December 31, 2016.

Factor analysis was performed with principal component analysis for initial factor extraction and Varimax rotation with Kaiser normalization. Internal consistency of the factor scores were evaluated using Cronbach’s alpha (values above 0.70 indicate good internal consistency). The associations of HAQ index with known determinants and correlates were evaluated using analysis of variance (ANOVA) in linear regression models adjusting for age, sex, Chinese ethnicity, and education (no education, 1–6 years, and >6 years). Analysis of cross-tabulations of the HAQ index with RK-SA scores were performed with Somers’ d measure of strength and direction of correlation and Kappa measure of concordance. The associations of HAQ and RK-SA scores with survival probabilities were evaluated in Cox proportional hazard regression models with hazard ratio (HR) estimates adjusted for sex, age, Chinese ethnicity, educational level, living alone, smoking, alcohol, multi-morbidity. The predictive accuracy of the HAQ and RK-SA for survival were compared using area under the curve (AUC) from receiver operating characteristics analyses.

The mean age of the SSOSA cohort was 72.1 (SD = 5.7) years; 40.6% were men; 93% had at least one chronic disease, and 12.4% had 5 or more chronic diseases. The HAQ comprises 15 question items (Table 1 and online suppl. Material). The scores of the participants were clustered towards the favourable extremes (“ceiling” effect) for three questions: How many relatives do you feel at ease with, talk to about private matters, and/or call on for help? Physical health limited the kind of work or other activities you could undertake? During the past month, how much of the time has your physical health or emotional problems interfered with your social activities? The distribution of individual summed scores of participants ranged in values from 26 to 91, mean = 64.0 (SD = 11.8), which is skewed to the left, and fits a negative lognormal distribution (Fig. 1). The estimated time to complete an interview on the 15 items in the HAQ was 5–7 min.

Factor analysis supports three components: (1) cognitive and psychological wellbeing and support; (2) physical, socio-emotional and occupational functioning; (3) physical, social, and occupational activity participation, explaining 41.7% of total variance. The Cronbach’s alpha of 0.735 indicates good internal consistency of items in the scale.

In the SLAS-2 validation cohort, the mean age of the participants was 68.8 years, and 37.7% were men; 92% had at least one chronic diseases, and 10.3% has 5 or more medical conditions. The mean HAQ score was 64.5 (SD = 7.2; range: 13–85).

Table 2 shows the association of the HAQ index with known determinants and correlates in multiple linear regression models adjusted for age, sex, ethnicity, and education. Participants who were female, more educated, not living alone, non-smoking, never or infrequently drank alcohol, not hospitalized, without a fall history in the previous year, not disabled, had no or fewer than 5 medical conditions, and robust or pre-frail showed higher HAQ scores. Participants who were free of stroke, cardiac diseases, obstructive lung diseases (asthma/COPD), gastrointestinal disorders, and depression showed significantly higher HAQ scores. There were significant positive correlations of MMSE with HAQ index, and negative correlations of age, logMAR vision, timed 5XSTS, and TUG with HAQ index.

Table 3 shows the cross-tabulation of ordinal scores on the HAQ index and the RK-SA index, which were significantly correlated. The Somers’ d and the kappa statistics, however, showed low concordance. Of note, only 4.6% (99/2,161) of the participants were classified as “successful agers” by the Rowe and Kahn criteria, whereas 13.9% (310/2,161) were classified as “healthy ageing” by the top quintile of the HAQ.

Overall, there were 383 deaths observed from 19,619.7 person-years (p-y) of follow-up, giving a mortality rate of 19.5/1,000 p-y. Table 4 shows that increasing quintiles of the HAQ index were associated with decreasing mortality rates from 40.6/1,000 p-y to 9.7/1,000 p-y. The HR of association, adjusted for sex, age, Chinese ethnicity, educational level, living alone, smoking, alcohol, and multi-morbidity (model 2) decreased across quintiles of HAQ index, with the lowest HR of 0.44 (95% CI = 0.28–0.67) associated with the highest Q5 levels (HAQ score >70) (Fig. 2). The RK-SA index also significantly predicted survival outcome, with increasing score associated with decreased mortality rates from 68.4/1,000 p-y to 5.2/1,000 p-y, and with the highest score of 5 (“successful ager”) associated with adjusted HR of 0.32 (95% CI = 0.12–0.85) in model 2. Using receiver operating characteristics analysis of the predictive accuracy for survival, the AUC of HAQ (0.675, 95% CI = 0.644, 0.706) was marginally higher than that of RKSA (AUC = 0.661, 95% CI = 0.628, 0.691), p = 0.361.

A handful of measurement indices based on the WHO functional ability model have been developed and validated in recent years. The Successful Aging Index (SAI) was developed in the UK for use by general practitioners with items including modified Katz ADLs (5 items) and Lawton IADLs (7 items), MMSE (30 items), and 4 lay perspective items of optimism, interest, self-rated health, loneliness [7]. It showed good predictive validity for various health and informal care service use (AUC from 0.65 to 0.86). The Healthy Ageing Index (HAI) developed in the ATHLOS project [10] comprises a comprehensive list of 41 items, covering cognitive performance, sleep, energy, pain, urinary incontinence, vision, locomotion/mobility, Katz ADL, and Lawton IADL. It showed concurrent validity with sociodemographic, life and health factors, and predicted life expectancy (AUC measure of predictive accuracy was not reported). The scale is intended to be useful in research and policy-making for valid comparison of healthy ageing across international cohorts. A modified Physiological Index using blood, clinical, and cognitive testing measurements (systolic blood pressure, forced vital capacity, Digit Symbol Substitution Test, MMSE, serum cystatin-C or creatinine, and fasting blood glucose) predicted mortality with c-statistic of 0.656 [8, 9]. A HAI developed in six Latin American population cohorts [11] comprises 26 indicators including the WHO Disability Schedule-II, cognitive difficulties; sleep; not coping; getting worn out; MMSE and other cognitive performance, time in seconds taken to walk 10 m; hearing and vision problem. It predicted mortality with c-statistic of 0.74.

Our HAQ Index differs substantially from the above-mentioned indexes as it is based on a small number of single-question items. Also, the HAQ assesses cognitive or physical functioning without using performance, clinical measurement, or blood test. The questionnaire is accurate and takes only a few minutes to administer. As such, it may be recommended for use at the individual and population levels for research, healthcare, and policy-making. With a continuous measurement scale from 0 to 100, the HAQ index is easily interpreted when used for assessing how well an individual is ageing. Stratification by appropriate cut-offs allows the identification of population groups for targeted interventions. Population surveys using the HAQ can assess and monitor the level and prevalence of healthy ageing for policy planning and evaluation.

Response scores across HAQ question items were correlated with each other, indicating that all the items reflect the same underlying construct. The content validity of the HAQ was supported by the factor structure comprising three dimensions (cognitive and psychological wellbeing and support; physical, socio-emotional and occupational functioning; physical, social, and occupational activity participation) in congruence with the WHO concept and definition.

The construct validity of the HAQ index is shown by its good convergent and divergent associations with known demographic, behavioural, health, and clinical risk or protective factors. As such, the HAQ index shows that higher levels of healthy ageing were associated with favourable demographic, socio-economic, lifestyle behaviour, health functioning, and disease characteristics. Pertinently, the HAQ index is consistently associated with objective measures of physical and cognitive functioning (logMAR vision, MMSE, TUG, 5 times sit-and-stand time).

Not surprisingly, the Rowe-Kahn SA index and the HAQ index were correlated with each other, but show low concordance, given their construct differences (the presence of chronic disease being included in Rowe and Kahn’s model). Both of them predicted life expectancy in the cohort, but the predictive accuracy appeared to be marginally higher for the HAQ index but not statistically significant. It is important, however, to note from the utility perspective, that Rowe and Kahn’s criteria identifies far fewer healthy ageing individuals. The predictive accuracy of HAQ for survival is also comparable or better than other healthy ageing indices mentioned above. Hence, healthy ageing can be measured briefly and simply with the HAQ without compromising on measurement accuracy.

Participants in the top 5th quintile of HAQ index scores (>70) show the most favourable survival prospects. Hence, approximately 20% of individuals in this population cohort may be regarded as having a high level of healthy ageing. The 2nd to 4th quintiles with cut-offs at 60, 65, and 68 stratify approximately 60% with medium-level healthy ageing, whereas the bottom 20% with HAQ score <60 may be considered to have a low level of healthy ageing. In the public health setting, a cut-off of >70 on the HAQ represents a reasonable target for population-based mass interventions through public health education promoting healthy ageing. In integrated clinic- and community services-based settings, a cut-off of <60 on the HAQ represents a suitable screening threshold for identifying individuals with a low level of healthy ageing for personal medical and lifestyle interventions to enhance their mental and physical capacity (intrinsic capacity) and functional ability, and targeted at achieving a HAQ score of 70. This could potentially result in raising HAQ scores of individuals and increasing the number of those with higher levels of healthy ageing in the population. The mean HAQ score in this study population was 64.5 and could be increased with such population health interventions. Population-level monitoring of the mean HAQ score could help evaluate the overall impacts of public education and clinic- and community-based interventions for healthy ageing.

As it is, the HAQ has only been validated for the first time in a single population sample among middle-aged and older Asian adults in Singapore. Its generalizability and comparability across different populations remains uncertain. Given the possible cross-cultural differences in the conceptualization of HAQ questions, more studies are required to investigate question item response bias (“differential item functioning”) and to ascertain that they are invariant across different cultural groups. It is interesting to note that the HAQ index is demonstrably accurate in subjects drawn from a different but related population and using data collected from questions that were modified from those originally used in its development. This suggests that the HAQ is able to robustly represent all relevant aspects of the healthy ageing construct, and is both reproducible and transportable [26]. Further studies should be conducted using the HAQ in diverse populations and settings, and using varying operational equivalents.

We thank the following voluntary welfare organizations for their support: Geylang East Home for the Aged, Presbyterian Community Services, Thye Hua Kwan Moral Society (Moral Neighbourhood Links), Yuhua Neighbourhood Link, Henderson Senior Citizen’s Home, National Trade Union Congress Eldercare Co-op Ltd, Thong Kheng Senior Activity Centre (Queenstown), and Redhill Moral Seniors Activity Centre.

The study was conducted according to the guidelines laid down in the Declaration of Helsinki and all procedures involving human subjects were approved by the National University of Singapore Institutional Review Board (NUS-IRB; Reference Code: 04–140). Written informed consent was obtained from all subjects.

The authors have no conflicts of interest to declare.

This work was supported by research grants from the Agency for Science Technology and Research (A*STAR) Biomedical Research Council (BMRC/08/1/21/19/567) and the National Medical Research Council (NMRC/1108/2007; NMRC/CIRG/1409/2014).

T.P.N., X.Y.G., and S.L.W. reviewed the literature. T.P.N. analysed the data, drafted, and revised the manuscript; T.P.N., X.Y.G., D.Q.L.C., C.Y.C., P.Y., and K.B.Y. contributed to the study design and data collection; and all authors critically reviewed the results and drafts and approved the final manuscript for submission.

All data generated or analysed during this study are included in this article and its supplementary material files. Further enquiries can be directed to the corresponding author.