The purpose of this thesis was to develop an approach that could evaluate the burden of unhealthy behaviours over a lifetime through linked population-based health surveys. The Canadian Community Health Survey (CCHS) is one such cross-sectional survey that is routinely administered to the household population and has been linked to a multitude of administrative healthcare databases. Utilizing the linked CCHS to evaluate the burden of unhealthy behaviours over a lifetime is challenging. Health behaviours naturally change over a lifetime due to many factors, and the burden of unhealthy behaviours has many different dimensions (e.g., mortality, disability, and healthcare costs) that are interconnected with each other. The degree to which lifetime disability and healthcare costs vary in relation to differences in life expectancy remains an area of debate. It is unclear whether individuals with healthy behaviours actually experience less lifetime disability and healthcare costs than individuals with unhealthy behaviours since they typically live much longer. Through several studies, this thesis developed various components that can be potentially combined into a lifetime approach which incorporates multivariable transitions.
The first two studies assessed the burden of unhealthy behaviours on period life expectancy and period lifetime healthcare costs. In the first study, CCHS-based multivariable risk algorithms were constructed to provide estimates of the causal associations between each unhealthy behavior (smoking history, leisure physical inactivity, non-active transport, leisure sedentary activity, and poor diet) and mortality. The burden of unhealthy behaviours on period life expectancy was estimated to be 7.5 (6.5-8.3) life years in 2000-2004 and 6.7 (5.8-7.4) life years in 2010-2014. The largest burdens were attributed to non-active transport and smoking. In the second study, CCHS-based multivariable risk algorithms were constructed to provide estimates of the causal associations between each unhealthy behavior and healthcare costs within different phases of life (i.e., defined by proximity to death). Unhealthy behaviours were attributed with 10.2% (2.5%-17.7%) of the period lifetime healthcare costs in 2000-2004, and 12.9% (5.6%-19.8%) in 2010-2014. Leisure sedentary activity and non-active transport were responsible for almost this entire burden, while the other unhealthy behaviours appeared to actually reduce period lifetime healthcare costs. The degree to which these estimates are accurate is unclear given the limitations of period life tables and the potential for unhealthy behaviours relating to physical activity to be a product of aging and prior illness.
The third study focused on developing methods by which to derive CCHS-based multivariable transition risk algorithms, which would allow for the creation of cohort life tables rather than period life tables. Novel methods involving multiple imputation models were utilized to create quasi-longitudinal CCHS cohorts from multiple cycles of the CCHS. These quasi-longitudinal cohorts were leveraged to develop multivariable risk algorithms for transitions towards different levels of immobility, an exposure that had been included in the prior algorithms for mortality and healthcare costs. Transitions towards moderate immobility were predicted by all unhealthy behaviours except poor diet, and transitions towards severe immobility were predicted by all unhealthy behaviours except sedentary activity. This approach can also be utilized to develop multivariable transitions for the unhealthy behaviours, which were simultaneously allowed to transition in the quasi-longitudinal CCHS cohorts. Such multivariable transition algorithms could potentially be combined with the previously derived algorithms for mortality and healthcare costs to generate more realistic estimates of life expectancy and lifetime healthcare costs. Large variability in the imputed quasi-longitudinal CCHS cohorts requires further examination, and may be reduced by including comorbidities, healthcare costs, and other information from linked administrative healthcare databases.
The last two studies evaluated the representativeness of linked CCHS respondents for population-based studies. Response and consent (to linkage) rates in the CCHS have been declining since its introduction raising concerns surrounding the comparability of CCHS samples over time. Similar to other population-based surveys, survey weights are provided that are designed address biases that may arise from non-response and non-consent to linkage. Unfortunately, these survey weights are not necessarily appropriate for many linked health outcomes that are rare. As a result, CCHS-based multivariable health risk algorithms are frequently derived from pooled unweighted CCHS samples. Fortunately, relative to wider sampling frames, unweighted linked CCHS samples were observed to be comparable over time. Nevertheless, linked CCHS respondents were observed to be healthier than comparable individuals in the community-dwelling and general populations at older ages, where they demonstrated lower risks of mortality, long-term care admission, and healthcare costs. This was not unexpected given that important segments of the population (e.g., residents of retirement homes and long-term care care) are excluded from the CCHS sampling frame. These studies highlighted the difficulties of estimating life expectancy and corresponding lifetime healthcare costs from the household population, and the necessity to ensure that such estimates realistically incorporate the time individuals may live outside of the household population over a lifetime.
These series of studies therefore resulted in mortality, healthcare cost, and transition risk algorithms that could potentially be combined to generate lifetime estimates of life expectancy, disability, and healthcare costs for a CCHS respondent. The development of transition risk algorithms requires further research. Once these methods are optimized and transition risk algorithms for all exposures of interest are generated, all the components required for this framework will be complete. At that point, explicit methods by which to combine the algorithms and validate projections will be required. This framework will enable a cause-deleted approach to be applied that simultaneously considers the impact of unhealthy behaviours on mortality, disability, transitions, and healthcare costs. This thesis represents an initial first step towards creating a framework that has the potential to generate lifetime estimates, as well as counterfactual estimates, which better reflect the complex nature of lifetime trajectories.
Identifer | oai:union.ndltd.org:uottawa.ca/oai:ruor.uottawa.ca:10393/44246 |
Date | 10 November 2022 |
Creators | Perez, Richard |
Contributors | Manuel, Douglas |
Publisher | Université d'Ottawa / University of Ottawa |
Source Sets | Université d’Ottawa |
Language | English |
Detected Language | English |
Type | Thesis |
Format | application/pdf |
Rights | Attribution-NonCommercial-NoDerivatives 4.0 International, http://creativecommons.org/licenses/by-nc-nd/4.0/ |
Page generated in 0.0016 seconds