Staying active, staying sharp: the relationship between physical activity and health-related quality of life for people living with cognitive impairment

Our empirical analyses utilized data from the Household, Income, and Labour Dynamics in Australia (HILDA) Survey. Since 2001, the survey has been collecting yearly data from a sample of Australians who are representative of the whole country. The study monitors a cohort of almost 17,000 people over their life course, gathering data on a range of topics including family and household dynamics, employment and earnings, educational attainment, and health outcomes. The HILDA Survey used a mix of face-to-face interviews and self-completion questionnaires to obtain this information [38]. A comprehensive overview of the HILDA dataset can be found elsewhere [39].

Our analysis used data from two time points of the HILDA Survey: the year 2012 (wave 12) and 2016 (wave 16). These specific waves were chosen for the study because they are the only waves within the dataset that include questions specifically designed to assess cognitive impairment. The HILDA survey assessed cognitive function of the survey respondents using validated instruments, including the Symbol Digit Modalities Test (SDMT) and the Backward Digit Span Test (BDS). While the scores obtained from the BDS and SDMT provide valuable insights into specific cognitive domains (working memory and processing speed, respectively), they may not fully capture the overall cognitive profile of an individual. A limitation of these tests is the lack of established cut-off scores, which may hinder accurate diagnosis of cognitive impairment. More comprehensive neuropsychological assessments, such as the Montreal Cognitive Assessment (MoCA) [40], Mini-Mental Status Examination (MMSE) [41], or Saint Louis University Mental Status (SLUMS) examination [42], offer a nuanced and reliable evaluation of cognitive function. Due to the unavailability of data measured through these scales, we utilized the BDS and SDMT to assess cognitive impairment for this study. Evidence suggests that the BDS and SDMT have been previously used to detect cognitive impairment, particularly in people with multiple sclerosis [43, 44] and hospitalized patients [45].

The BDS cognitive evaluation exam involves individuals reciting a sequence of numbers in reverse order [45]. The BDS evaluates the cognitive capacity of working memory on a scale of 0 to 8. The SDMT is a cognitive assessment tool that measures a person’s ability to process information quickly and accurately. The SDMT requires participants to match a list of numbers with corresponding geometric shapes as quickly and accurately as possible [46]. The SDMT evaluates the cognitive function of the central brain and provides scores that range from 0 to 110. The threshold for identifying cognitive impairment in this study was informed by an established criterion. Previous research has categorized cognitive impairment based on standardized thresholds: individuals scoring ≥ 1.0 standard deviation (SD) below the mean on either the BDS or SDMT (or both) are classified as having mild cognitive impairment, while those scoring ≥ 1.5 SD below the mean on both tests are identified as having severe cognitive impairment [47‐49]. This study consolidated mild and severe cognitive impairment into a single category, focusing on cognitive impairment as a unified construct. Cognitive impairment was therefore defined as scoring ≥ 1 SD below the mean on both the BDS and SDMT. This criterion aligns with empirical thresholds reported in prior studies and supports the objectives of this research. Accordingly, participants were considered cognitively impaired if they scored ≤ 3 on the BDS and ≤ 30 on the SDMT. Given that cognitive impairment primarily affects older people, our study focused on Australians aged 50 years and over. Therefore, the inclusion criteria for people in the sample were as follows: (i) being 50 years of age or older; (ii) identified as living with cognitive impairment; and (iii) having valid information on the outcome and key factors of interest. Applying these inclusion criteria resulted in an unbalanced panel of 1,168 yearly observations from 985 unique individuals. Figure 1 depicts the sample selection process and missing data analysis.

The primary outcome variable in our study was HRQoL, which we measured using the 36-item Short-Form Health Survey (SF-36). The SF-36 is a widely used and reliable instrument that assesses an individual’s physical and mental health using a standardized questionnaire [50]. The questionnaire comprises 36 items that evaluate eight specific health domains: physical functioning (PF), role physical functioning (RP), role emotional functioning (RE), social functioning (SF), mental health (MH), vitality (VT), bodily pain (BP), and general health (GH). Each dimension of the SF-36 has a theoretical range of 0 to 100, where 0 indicates the worst possible health and 100 represents the best possible health. The SF-36 data generally yields two summary measures: the physical-component summary (PCS) and the mental-component summary (MCS) [51]. The PCS and MCS were standardised by linear z-score transformations, resulting in a mean of 50 and a standard deviation of 10. The theoretical ranges of PCS and MCS scores are 4.54 to 76.09 and − 1.21 to 76.19, respectively, with higher scores indicating improved health [30].

In addition to the PCS and MCS, the SF-36 can also be used to generate the SF-6D, a health-state utility index, which is another internationally recognized measure of HRQoL [52]. The SF-6D utility index is derived from a subset of six subscales of the SF-36 (PF, RP, RE, SF, VT, and BP), with a theoretical range from 0.29 (indicating poor health) to 1 (representing optimal health) [52].

This study investigates the frequency of moderate-to-intense physical activity. To assess this, the research relies on a self-administrated question consistently employed across all waves of the HILDA Survey. This question inquires: ‘In general, how often do you participate in moderate or intensive physical activity for at least 30 minutes? Moderate physical activity will cause a slight increase in breathing and heart rate such as brisk walking.’ The participants’ replies were categorised into six pre-determined categories: ‘not at all’, ‘less than once per week’, ‘1 or 2 times per week’, ‘3 times per week’, ‘more than 3 times per week but not every day’, and ‘every day’. We simplified the physical activity categories into three groups: ‘not at all,’ ‘<1 to 3 times per week’ (by merging the groups ‘less than once per week’, ‘1 or 2 times per week’, and ‘3 times per week’), and ‘more than 3 times per week to everyday’ (by merging the groups ‘more than 3 times per week but not every day’, and ‘every day’). The categorization of physical activity in this study adheres to the established criteria of the Australian National Physical Activity Guidelines for Adults [53], demonstrating close concordance with the World Health Organization’s guidelines [54]. This methodological approach has been adopted in prior empirical research utilizing the HILDA survey data [31, 55].

We incorporated several individual-level socio-demographic, health-related behavioural characteristics, and health-related characteristics as covariates. The socio-economic and demographic characteristics analysed included: age (50–64 years, and ≥ 65 years), gender (male, and female), marital status (unpartnered and partnered), highest level of education (year 12 and below, professional qualifications, and university qualifications), household yearly disposable income (Quintile 1 [poorest], and Quintile 5 [richest]), participation in the labour force (employed, and unemployed or not in the labour force), Indigenous origin (non- Aboriginal or Torres Strait Islander, and Aboriginal or Torres Strait Islander), geographic residency (major city, and regional or remote area). Additionally, two health-related behavioural characteristics that can impact health outcomes were included: smoking habits (former smoker or never smoked, and currently smoking), and alcohol drinking (former drinker or never drunk, and active drinker). Furthermore, Body Mass Index (underweight, healthy weight, overweight, and obese) and disability status (no versus yes), were considered as the proxy of participants’ health-related characteristics.

We commence our analysis by calculating descriptive statistics for the study sample. For categorical variables, frequencies and percentages are calculated to describe their distribution. For continuous variables, means and standard deviations are calculated to summarize their central tendency and variability. These descriptive statistics are calculated separately for baseline, final wave, and pooled data to provide an overview of the data across different time points. We also present a summary of participants’ SF-36 component summary scores, and SF-6D utility values according to their physical activity levels.

The outcome variables (PCS, MCS, and SF-6D) used in our study were measured on a continuous scale. Therefore, we employed a longitudinal random-effects GLS regression model to examine the relationship between physical activity and HRQoL, allowing us to identify individual variations in this association. Additionally, the random-effects structure enables us to control for unobserved individual heterogeneity and potential confounders, thereby enhancing the reliability of our findings.

The random-effects GLS regression can be expressed in the following functional form:

HRQoL denotes the three outcome variables: PCS, MCS and SF-6D. The level of physical activity, PA, is an exposure variable. \(\:{Z}_{it}\) represents the vector consisting of time-varying and time-invariant control variables. The model parameter of interest to be estimated is denoted as \(\>{\beta _1}\), and \(\>{\beta _2}\) indicates the vector of coefficients while α is the model’s grand intercept. The analysis considers two parts of the error: individual-specific components, \(\:{\upmu\:}\), that stays the same over time, and time and person-specific error, \(\:{\text{€}}_{it}\), which is assumed to be uncorrelated with the independent variables.

Statistical significance was determined using a p-value threshold of 0.05. Lower p-values (< 0.01 and < 0.001) were reported to indicate stronger evidence of significance. Stata version 17.0 (StataCorp LLC, College Station, TX: USA) was employed for all statistical analyses.

We also fitted random-effects generalised estimating equation (GEE) regression model as part of the sensitivity analysis. The model initially reported in Table 3 was re-assessed and presented in Appendix Table A1 in the online supplemental material. The findings of the sensitivity analysis closely matched the baseline values for all the factors of interest. We found that older Australians with cognitive impairment who engaged in moderate or intense physical activity had higher PCS, MCS and SF-6D values compared to those who were physically inactive. For example, participants engaged in moderate or intense physical activity for more than 3 times per week to everyday had greater PCS (β = 7.46, SE = 0.76), MCS (β = 4.61, SE = 0.81) and SF-6D (β = 0.08, SE = 0.008) scores, in comparison to those who did not engage in physical activity.

We conducted a missing data analysis, as presented in Appendix Table A2. With the exception of Indigenous origin, SF-6D utility score, and BMI, the proportion of missing observations for most variables was less than 5%. To address the missing data, we applied the last value carry forward imputation technique. After imputation, we applied the random-effects GLS technique and compared the results with the estimates obtained from the complete case analysis (Table 3). The regression results from the imputed data were consistent in direction with the baseline regression findings. However, the magnitudes of the physical activity estimates varied slightly across the HRQoL measures. For example, participants engaged in moderate or intense physical activity for more than 3 times per week to everyday had greater PCS (β = 7.53, SE = 0.79), MCS (β = 3.97, SE = 0.83) and SF-6D (β = 0.07, SE = 0.009) scores, in comparison to those who did not engage in physical activity. The imputed regression analysis results are detailed in Appendix Table A3.

To further explore the relationship between physical activity and HRQoL, we ran a series of random-effects Generalized Least Squares (GLS) regression models. These models were used to investigate how the association between physical activity and HRQoL might vary across different age and gender subgroups within the study population. The results indicated that, across both age groups (50–64 years and 65 years and older), people who participated in any form of moderate or intense physical activity for at least 30 min had significantly higher PCS, MCS, and SF-6D values compared to those who were physically inactive (see Table 4). For instance, among participants aged 65 years and over, those engaging in moderate or intense physical activity more than 3 times per week to everyday had significantly higher scores in PCS (β = 7.41, SE = 0.91), MCS (β = 4.51, SE = 0.96), and SF-6D (β = 0.07, SE = 0.011) compared to those who did not engage in physical activity (Models 4, 5 and 6 in Table 4). Similarly, irrespective of gender, individuals participating in any moderate or intense physical activity showed higher PCS, MCS, and SF-6D scores than those who were inactive (see Table 5). For example, female participants who engaged in moderate or intense physical activity more than 3 times per week to daily had greater PCS (β = 7.16, SE = 1.21), MCS (β = 2.10, SE = 1.23), and SF-6D (β = 0.05, SE = 0.01) scores compared to their inactive counterparts (Models 4, 5 and 6 in Table 5).

Our study’s strengths include its use of a comprehensive longitudinal design. To the best of our knowledge, this is the first observational study to examine the association between physical activity and HRQoL among people with cognitive impairment. The use of a validated instrument to measure HRQoL and cognitive impairment ensures the reliability of our findings. However, our study has some limitations that need to be mentioned. Firstly, the HILDA Survey’s physical activity frequency measure may not have accurately captured the exact amount of time participants spent on physical activities, which could affect the precision of our results. Secondly, the reliance on self-reported data for physical activity and other covariates may introduce potential biases. Social desirability bias, characterised by the inclination of individuals to overstate their physical activity levels or underreport health limitations, is a significant concern. Recall bias, especially among older adults and those with cognitive impairments, can also distort the accuracy of reported physical activity engagement and HRQoL experiences. These biases may lead to either under- or overestimation of the true association between physical activity and HRQoL. Third, the absence of standardized cut-off scores for the SDMT and BDS limits their ability to accurately define and comprehensively assess the full range of cognitive impairment. Furthermore, while these tests provide valuable insights into specific cognitive domains, they may not adequately capture memory deficits, a core feature of cognitive impairment and dementia. This limitation could have potentially underestimated the prevalence of cognitive impairment in this study. The inclusion of memory-focused assessments, such as the Montreal Cognitive Assessment, Mini-Mental State Examination, or Saint Louis University Mental Status test, could have provided a more targeted approach and improved comparability with findings from other studies. Finally, the analysis did not account for potential confounding factors such as diet, stress levels, and comorbid conditions. These factors can significantly influence both physical activity levels and HRQoL. The unavailability of data on these factors in the HILDA Survey represents a major limitation. By not accounting for these factors, the study may not precisely estimated the relationship between physical activity and HRQoL among people with cognitive impairment. This may lead to confounding bias, distorting the observed relationship between physical activity and HRQoL due to the influence of these unobserved variables. This could result in spurious associations.

Future research could explore the economic implications of physical activity in people with cognitive impairment, including cost savings, work productivity, and reduction of overall health and economic burdens on the healthcare system. These research would provide valuable insights for policymakers and healthcare providers, guiding strategies and investments in initiatives concerning physical activity.

The findings of this research can be directly utilized by healthcare professionals who work with older adults with cognitive impairment to develop and implement more effective health promotion interventions that emphasize the importance of physical activity. In 2018, Australia became one of the pioneering nations in developing national physical activity guidelines tailored to older adults with cognitive impairment [66]. A narrative review further supported these guidelines, providing evidence that the recommendations for older adults could be adapted to meet the needs of individuals with cognitive impairment [67]. This conclusion was reached after comparing these guidelines with similar Canadian guidelines. Evidence suggests that older Australians today place a higher value on information and advice from healthcare professionals compared to other sources, highlighting their critical role in adopting recommendations [67]. By offering tailored guidance and support, healthcare professionals can assist individuals in achieving the recommended physical activity levels, which can lead to improved well-being. Our findings highlight the necessity of including physical activity in comprehensive care plans through bulk-billed GP services for older adults with cognitive impairment.