Design and participants
This study used data from the mid-age cohort (born in 1946 to 1951) of the Australian Longitudinal Study on Women’s Health (ALSWH), an ongoing study of Australian women. Baseline surveys were mailed in 1996 when women were 45-50 years old (N=13,714), with follow up surveys from 1998 to 2016 at three years intervals. At baseline, the sample was largely representative of Australian women in this age group, but with a somewhat higher representation of partnered women and women with post-high school education (25). Ethics approval for the study was gained from the relevant ethics committees at the University of Newcastle (No. H-2010_0031) and the University of Queensland (No. 2010000411). Participants gave their informed consent to participate. Further details of the recruitment methods, response rates and data collection have been described elsewhere (25) and can be found at www.alswh.org.au.
For this study, we analysed data from women who responded to surveys in 2001 (50-55y), 2007 (56-61y) and 2013 (62-67y), and who consented to linkage with healthcare administrative data (N=7,792). Women who had missing data for PA in one or more surveys were excluded from the analyses (N=839). The analytical sample included 6,953 women (Supplementary Figure 1). We analysed the data in 2019.
Physical Activity
Physical activity was assessed in 2001, 2007 and 2013 using a modified version of the Active Australia questionnaire, which has acceptable reliability and validity for use in cohort studies (6). Participants were asked about frequency and duration of walking briskly (for recreation or exercise or to get to or from places), moderate-intensity leisure- activities (like social tennis, moderate exercise classes, recreational swimming, dancing), and vigorous-intensity leisure-time activities (that make you breathe harder or puff and pant) in the last week. They were asked to only report PA that lasted 10 min or more. Minutes per week spent in each activity were multiplied by a metabolic equivalent (MET) score: 3.33 for walking and moderate intensity leisure activities and 6.66 for vigorous leisure-time activity. The amount of PA was calculated as the sum of MET-minutes/week from each of the domains and categorised as: none (0 − <33.3; reference category); low (33.3 − <500); moderate (500 − <1000); or high (≥1000)(6).
Costs of health services
Health costs data from the Australian Medicare Benefits Scheme (MBS) and the Pharmaceutical Benefits Scheme (PBS) were analysed. Total health costs, which include the government benefit paid and the out-of-pocket cost paid by the patient for each service (MBS) or prescription (PBS), were calculated for each participant as the sum of total costs from 2013 (62-67y) to 2015 (64-69y). MBS is the Australian government’s system for subsidising general practitioner and some out-of-hospital specialist, pathology, radiology, dental and allied health services, and limited additional primary healthcare services, for all Australian citizens and permanent residents (MBS, 2019)(26) ; PBS is the system that subsidises the cost of approved prescribed medications.
Covariables
In 2013, participants provided information on age, area of residence, education, marital status, smoking, alcohol, and perceived health, categorised as shown in Table 1. They were also asked whether they had a health care card (which allows disadvantaged patients lower out-of-pocket costs). Body Mass Index (BMI) was categorized according to the World Health Organization classification: underweight (BMI<18.5 kg/m2); normal weight (BMI 18.5-24.9 kg/m2), overweight (BMI 25.0-29.9 kg/m2) or obese (BMI≥30 kg/m2). Number of chronic conditions was indicated using a list of nine conditions which are common in mid-age (diabetes, osteoarthritis, cardiovascular, cognitive, respiratory, cancer, mental health, ophthalmology and sexual).
Data analysis
Sociodemographic and behavioural characteristics of the analytical sample in 2013 were summarized using descriptive statistics (eg proportions, medians and Interquartile ranges -IQRs). Median and IQRs were calculated for MBS and PBS health costs during 2013-2015 according to sociodemographic characteristics in 2013.
For the trajectory analyses women were categorised as either 'inactive' (none or low) or 'active' (moderate or high) at each survey. The proportion of active women at each survey was calculated and changes in activity status (active/inactive) between 2001 and 2007 and between 2007 and 2013 were computed and illustrated using a lasagne plot. To elucidate the associations between physical activity and health costs, the analyses were conducted in four steps. First, median and interquartile ranges for total MBS and PBS costs in 2013-2015 were calculated for each physical activity category in 2001, 2007 and 2013. Second, costs were calculated for five groups according to PA trajectories: a) always inactive (inactive in all three surveys); b) always active (active in all three surveys); c) increasers (changed from inactive in 2001 to active in 2013); d) decreasers (changed from active in 2001 to inactive in 2013); e) 'fluctuaters' (classified in the same PA category in 2001 and 2013, but in the opposite category in 2007). Third, a cumulative PA score was created by summing the number of times women were categorized as active (0, 1, 2 or 3) and costs were calculated for women in each of these categories. Fourth, crude and adjusted quantile regression models were used to estimate the differences in median costs for each of the PA variables described in steps one to three. Analyses were adjusted for age, education, marital status, area of residence, having a health care card, smoking, alcohol and BMI. These potential confounders have been shown in previous ALSWH studies to be associated with both PA and health costs (16, 24, 27). All statistical analyses were performed using Stata 16.1.