The Population Health Model (POHEM): an overview of rationale, methods and applications

To describe POHEM, it is first necessary to define dynamic microsimulation. Dynamic microsimulation, in the context of social science and population health, is the simulation of large samples of individuals (micro) and their behaviours, states and actions, over time (dynamic), in order to project the socioeconomic and demographic developments of society [7]. By necessity, individuals’ behaviours, states and actions are modeled using multiple sources of empirical data, including cross-sectional surveys, longitudinal surveys, administrative databases, vital statistics, and Census data. See Fig. 1 for an illustration of the data sources used in the POHEM cardiovascular disease model and the Appendix for a more comprehensive description of each data source. The explicit microanalytic foundation of these models is important in representing realistic population heterogeneities and is conducive to simulating the effects of policy interventions.

POHEM is one of a few population-based health dynamic microsimulation models worldwide; Zucchelli et al. and others have reviewed the field [6, 8‐10]. The model dynamically simulates individuals’ disease states, risk factors, and health determinants, in order to describe and project health outcomes, including disease incidence, prevalence, life expectancy, health-adjusted life expectancy, health-related quality of life, and healthcare costs. POHEM simulates discrete events, such as changes in disease states, on a case-by-case basis in continuous time. Events occur in chronological sequence at specific moments in time and the probability of an event occurring, or more precisely, the time to an event, is determined by a random process that draws from empirically observed and estimated waiting time distributions for each event. The occurrence of diverse events is embedded in a competing risk framework, that is, events compete with each other to be the next to occur. As events occur they influence subsequent draws from the relevant waiting time distributions, and hence the life trajectory of the individual. These many random components allow individuals’ life course trajectories to vary in ways that realistically reproduce observed patterns of behavior, including variation that is not “explained” by conventional epidemiological risk functions. The resulting variability in simulated individuals’ life course trajectories builds on the variations in the initial values assigned to the simulated individual. The resulting life trajectory of each simulated individual is influenced by exposure to simulated real world events, such as smoking initiation, changes in body mass index (BMI), and incidence of disease.

Producing estimates from POHEM involves six steps, described below and in Fig. 2.

Many of the examples discussed in this paper come from the POHEM model of cardiovascular disease (see Fig. 1 and Table 1 for covariates and risk factors modeled). This model was developed in collaboration with academic research teams and has been used to investigate the projected prevalence of risk factors for heart disease in Canada [33]. POHEM was well-suited for this study because cardiovascular disease is a complex disease process, with many contributing risk factors leading to multiple outcomes. Further, the incidence of cardiovascular disease has declined in the past 50 years, with the prevalence of risk factors changing significantly. For instance, smoking is in decline while obesity is on the rise, and the population is aging. The POHEM cardiovascular disease model has projected decreased smoking rates but increased obesity prevalence, with obesity projected to overtake smoking as the most prevalent risk factor by 2017. These projections can inform policymakers interested in decreasing cardiovascular disease risk and AMI hospitalizations in the population [33].

POHEM has been used extensively to study major cancer sites. Models have been developed to investigate cancer screening (colorectal and lung) [19, 23] cancer treatment (lung and breast) [15, 23] and costs of cancer (breast, colorectal, and lung) [18, 23, 34]. Flanagan et al. demonstrated that screening for colorectal cancer with fecal occult blood test and follow-up by colonoscopy could be cost-effective for the Canadian population but highlighted that the effectiveness of screening would greatly depend on reaching high screening participation rates [19]. These results informed the National Committee on Colorectal Cancer Screening which made updated colorectal cancer screening guidelines in Canada. Will et al. used the POHEM breast cancer model to demonstrate that while preventive tamoxifen treatment had a substantial benefit in reducing breast cancer incidence and mortality, the detrimental effects of tamoxifen on endometrial cancer, coronary heart disease, stroke, and deep vein thrombosis could actually outweigh its protective effects and even shorten women’s life expectancies overall [15].

The Cancer Risk Management Model (CRMM) is a spin-off model from POHEM developed by Statistics Canada in collaboration with CPAC to assess cancer control strategies in the areas of prevention, screening, and treatment for four major cancer sites: lung, colorectal, cervical and breast [23, 24]. They feature sophisticated natural history models of tumor onset and progression. Evans et al. projected estimates of future lung cancer incidence with current smoking rates holding steady (approximately 22 %) and counterfactually with smoking rates decreasing by 50 % over 3 years [23]. The CRMM lung module can also be used to compare the cost-effectiveness of similar smoking cessation initiatives to alternative interventions, including introduction of new screening programs, for instance, low dose CT scanning to detect early lung cancer [23, 24, 35]. The CRMM colorectal model has been used to evaluate different screening modalities (fecal occult blood, fecal immunological test, sigmoidoscopy and colonoscopy) under a range of program characteristics (age range, frequency, participation rates, costs) at a provincial level, as well as to assess the potential cost-effectiveness and budgetary impacts of implementation [36]. The CRMM cervical model can be used to evaluate the potential reorganization of cervical cancer screening in the context of vaccination for human papillomavirus (HPV) and HPV DNA testing. A wide range of scenarios have been evaluated to inform the Canadian Cervical Cancer Screening Network. This example illustrates the power of microsimulation to evaluate multiple intervention strategies including prevention, screening, and treatment, and provides common metrics with which to compare them. CRMM has also been made available on the web for easy use by cancer control and health policy analysts, who can use it to evaluate different cancer control strategies [23, 24].

Osteoarthritis (OA) is the most common form of arthritis and a leading cause of disability in Canada. OA is a chronic condition of aging and as our population ages, increasing numbers of people are living with this condition and suffering reductions in their health-related quality of life. In addition, OA is costly to the healthcare system, entailing drug and surgical treatment. The POHEM model provides a tool for researchers and policymakers to examine important outcomes of OA simultaneously. Kopec et al. developed POHEM–OA to quantify the future health and economic burden of OA under a range of scenarios incorporating changes in a key OA risk factor (BMI) and treatment levels (medications and surgery). Simulation modeling allowed these researchers to assess the future impact of potential changes in BMI on the prevalence of OA across population groups (see Fig. 3) [20]. In addition, this model also allowed researchers to examine the effects of BMI reduction on OA incidence and health-related quality of life [37], as well as projecting the direct [38] and indirect costs of OA in the next 20 years.

POHEM-Physical Activity (PA) was developed in collaboration with PHAC to project different types of physical activity and assess the potential impact of physical activity on health outcomes, such as number of cases of disease (diabetes, hypertension, heart disease, and cancer) and health-adjusted life expectancy. The Conference Board of Canada used POHEM-PA to do counterfactual analysis of increasing physical activity among 10 % of Canadians and examined the impacts on chronic disease incidence and ultimately the Canadian economy [39]. The POHEM-PA model methodology has been formally published [22].

POHEM-Neurological was developed in collaboration with PHAC to project the incidence and prevalence of neurological conditions as well as their impact on health-related quality of life, healthcare costs and caregivers [25]. The most prevalent neurological conditions were modeled: Alzheimer’s disease and other dementias, cerebral palsy, epilepsy, multiple sclerosis, Parkinson’s disease/parkinsonism, hospitalized traumatic brain injury, and hospitalized traumatic spinal cord injury. Each model accounted for changes in the Canadian population from births, immigration, emigration, and aging; however, risk factors were not included in this version of the model [25].