Polypharmacy and trajectories of health-related quality of life in older adults: an Australian cohort study

This cohort study was based on data from the Household Income and Labour Dynamics in Australia (HILDA) Survey. In the present study, HRQoL was assessed annually from 2013 to 2017, and polypharmacy status was assessed in 2013 and 2017. Briefly, the HILDA survey is one of the largest longitudinal surveys in Australia that collects information from community-dwelling participants covering a wide range of dimensions from family relationships and finance to health and education. Further details about the survey and its methodology have been reported elsewhere [26]. The sample population included in the present study were older adults aged 65 years and over who participated in 2013, had medication history recorded, and complete information on covariates at baseline and at follow-up. Participants who do not satisfy any of these criteria were excluded.

A total of 17,501 participants completed wave 13. Of these, 3021 (17.3%) were aged 65 years and older and identified as potential participants. Of these, 2645 (87.6%) had complete information on covariates. Those who were excluded due to incomplete information at baseline (n = 376) were older, less educated, less likely to be married, and more socially active. Of the initial 2645, 464 were excluded due to missing covariates at follow-up. Overall, 2181 participants fulfilled the inclusion criteria and were included in the final analysis. Figure 1 shows a flow chart of the study.

In Australia, the Pharmaceutical Benefits Scheme (PBS) provides universal health care that allows all citizens and permanent residents to access a range of subsidised healthcare services including subsidised medicines [27]. In the present study, the use of regular prescription medications was measured in 2013 and 2017 only, and this was obtained from responses to the question: “All together, how many prescription medications do you take on a regular basis?”. In this study, polypharmacy was defined as the concurrent use of five or more regular prescription medications [3]. Participants with polypharmacy at neither baseline nor follow-up were classified as ‘no polypharmacy’; those with polypharmacy in 2013 only were classified as ‘baseline only polypharmacy’; those with polypharmacy in 2017 only were classified as ‘incident polypharmacy’; and those with polypharmacy at both time points were classified as ‘persistent polypharmacy’.

HRQoL was assessed in HILDA through the Medical Outcomes Study Short Form (SF-36). The SF-36 is one of the most widely used measures of HRQoL [28]. Detailed description of SF-36 has been explained elsewhere [29]. Briefly, it is a self-completed questionnaire comprising 36 items that cover eight domains including general health, pain, vitality, social, mental and physical function, and role limitation due to emotional or physical health. Using factor analysis of the eight scales, two summary scores of health known as the physical component summary (PCS) and the mental component summary (MCS) were generated. Traditionally, scores from all domains are combined and transformed to generate PCS and MCS. However, as explained by Dockery et al. this may lead to inaccurate results in older adults [30]; for instance, in the factor analysis, the physical function domains negatively correlate with mental health, and hence, MCS scores would increase, not necessarily due to improvement of mental health but rather because of deterioration of physical function [30]. Therefore, in the present study, PCS scores were calculated from the four subscales that were positively correlated with physical function in the factor analysis. Those subscales were role-physical, bodily pain, physical functioning and general health. Similarly, MCS scores were calculated from the four subscales that were positively correlated to mental health in the factor analysis. Those subscales included vitality, social functioning, role limitations and mental health. A higher score (ranging from 0 to 100) on a component reflects a better HRQoL [29]. The SF-36 scores reported in HILDA have been previously validated to support their use as general measures of physical and mental health status in the Australian population [28].

Covariates included age, gender, education level, marital status, comorbidities, socio-economic status, and lifestyle behaviours at baseline (2013). Education was dichotomised into high school (i.e. 12 years of education) or higher vs less than high school. Marital status was dichotomised into ‘married/defacto’ vs ‘single/separated/divorced/widowed’. Number of comorbidities was obtained from questions that assessed the presence of a diagnosis of arthritis, asthma, any type of cancer, chronic bronchitis and/or emphysema, depression and/or anxiety, type 1 diabetes, type 2 diabetes, hypertension, heart disease, or other serious circulatory condition, e.g. stroke. These comorbidities were included as they are likely to impact HRQoL [31]. Change in comorbidities was calculated as the difference in number of comorbidities between 2013 and 2017, and categorised as ‘stable/decrease’ and ‘increase’. Socio-economic Index For Areas (SEIFA) is a relative measure of socio-economic advantage and disadvantage of areas in Australia provided by the Australian Bureau of Statistics [32]. SEIFA has multiple indices; in HILDA, the Index of Relative Socio-Economic Advantage and Disadvantage (IRSAD) was used [26]. Detailed explanation of SEIFA and its indices is reported elsewhere [32]. Briefly, IRSAD provides a summary measure of relative socio-economic advantages and disadvantages experienced, on average, by individuals living in that area [32]. For example, an area with majority of the households with low income or unskilled occupations, and few households with high income or skilled occupations would generally have a low score indicating a relatively greater disadvantage and a lack of advantage [32]. We treated SEIFA-IRSAD deciles as a continuous variable ranking areas from 1 (most disadvantaged) to 10 (most advantaged). Physical activity was measured as the number of days participants engaged in moderate or intensive physical activity for at least 30 minutes. This was included in the analysis as a dichotomous variable: ‘3 days a week or more’ or ‘less than 3 days a week’. Social life was measured by asking participants how often they get together socially with friends or relatives not living with them, and was dichotomised into ‘once a month or less’ or ‘more than once a month’. Frequency of alcohol intake was dichotomised into: drinking on ‘less than 5 days a week’ or ‘5 or more days a week’. Lastly, smoking was dichotomised into ‘smoker’ or ‘non-smoker’.

Baseline characteristics were described according to polypharmacy status and reported as means and standard deviations (SDs), or frequencies and percentages, as appropriate. Linear mixed-effects models were used to model the change in PCS and MCS scores from 2013 to 2017. Models were stratified according to polypharmacy status (none/baseline only/incident/prevalent) and change in comorbidities (stable/decrease vs. increase), resulting in 8 strata. The final models included intercept and slope (time) as random effects. Variances of the residuals and random effects were assumed to be uncorrelated. Allowing the random intercept and random slope to be correlated resulted in non-convergence of the models. For the stratum ‘baseline only polypharmacy and increased comorbidities’, the model contained a random intercept only due to the small sample size of this stratum. Time was based on the exact date of interviews. A quadratic slope was tested but was not significant. Hence, HRQoL trajectories were modelled as a linear function of time in years with 2013 as time = 0. Assumptions of linearity, homoscedasticity and normality of residuals were confirmed by visual inspection of graphs. The main analysis was adjusted for baseline comorbidities, education, marital status, SEIFA decile, social life, physical activity, smoking and alcohol intake. All covariates were mean centred and included as intercept and slope predictors. The results of these models were reported as estimated fixed effects (β value) and 95% confidence intervals (CIs). To assess the impact of different definitions of polypharmacy on the observed results, we conducted a sensitivity analysis in the incident polypharmacy model using different cut offs for polypharmacy (i.e. defining polypharmacy as ≥ 3, ≥ 4, ≥ 6, and ≥ 7 medications). All analyses were conducted using the Statistical Package for the Social Sciences (SPSS) version 25 (IBM Corp Armonk, NY) and GraphPad Prism 9.2.0 for Windows, GraphPad Software, San Diego, California USA.