A novel decision model to predict the impact of weight management interventions: The Core Obesity Model

Abstract Aims Models are needed to quantify the economic implications of obesity in relation to health outcomes and health‐related quality of life. This report presents the structure of the Core Obesity Model (COM) and compare its predictions with the UK clinical practice data. Materials and methods The COM is a Markov, closed‐cohort model, which expands on earlier obesity models by including prediabetes as a risk factor for type 2 diabetes (T2D), and sleep apnea and cancer as health outcomes. Selected outcomes predicted by the COM were compared with observed event rates from the Clinical Practice Research Datalink‐Hospital Episode Statistics (CPRD‐HES) study. The importance of baseline prediabetes prevalence, a factor not taken into account in previous economic models of obesity, was tested in a scenario analysis using data from the 2011 Health Survey of England. Results Cardiovascular (CV) event rates predicted by the COM were well matched with those in the CPRD‐HES study (7.8–8.5 per 1000 patient‐years across BMI groups) in both base case and scenario analyses (8.0–9.4 and 8.6–9.9, respectively). Rates of T2D were underpredicted in the base case (1.0–7.6 vs. 2.1–22.7) but increased to match those observed in CPRD‐HES for some BMI groups when a prospectively collected prediabetes prevalence was used (2.7–13.1). Mortality rates in the CPRD‐HES were consistently higher than the COM predictions, especially in higher BMI groups. Conclusions The COM predicts the occurrence of CV events and T2D with a good degree of accuracy, particularly when prediabetes is included in the model, indicating the importance of considering this risk factor in economic models of obesity.


| INTRODUCTION
The high prevalence and chronic nature of obesity are compounded by the large number of related complications. 1 There is extensive evidence of a link between body mass index (BMI) and type 2 diabetes (T2D), as well as cardiovascular disease (CVD), including both chronic complications, such as hypertension and coronary heart disease, and acute events such as myocardial infarction (MI) and stroke. 1,2 Furthermore, obesity is associated with other complications across multiple organ systems, including sleep apnea 3 and osteoarthritis, and is also implicated in the development of some types of cancer. 4 These complications incur a substantial proportion of obesity-related healthcare costs. [5][6][7] Health economic models of obesity are used to assess the costeffectiveness of weight management interventions, driving healthcare decision-making and allocation of resources. To do this, such models estimate the risk of BMI-related complications, the impact on health-related quality of life (HRQoL), and the associated economic costs. 8 Obesity models set in the UK healthcare system, assessing the long-term impact of T2D and CVD, and also incorporating mortality, have previously been developed, principally for use in economic predictions. These models have been used to assess the cost-utility of orlistat 9 and compare the cost-effectiveness of orlistat, sibutramine, and rimonabant, 8 and to assess the cost-effectiveness of the Light-erLife weight management program 10 and the Weight Action Program 11 However, these previous models can be refined and improved upon; given the multifactorial nature of overweight and obesity and the range of associated complications, the incorporation of additional comorbidities and risk factors offers the potential to improve the accuracy of predictions. Furthermore, models must be fit for purpose and interpretable by key stakeholders. This need for transparency and accuracy in model development has informed published best practice guidelines by the International Society for Pharmacoeconomics and Outcomes Research (ISPOR) and the Society for Medical Decision Making (SMDM). 12,13 This report presents the development and structure of the novel Core Obesity Model (COM; version 8.0). The COM was designed to encompass a broader range of complications than previous models and includes the impact of sleep apnea, knee replacement as a result of osteoarthritis, postmenopausal breast cancer, postmenopausal endometrial cancer, and colorectal cancer. The model also incorporates the impact of prediabetes, which is known to be associated with increased all-cause mortality, as well as a higher risk of T2D and CVD. 14 Furthermore, the results of an analysis comparing model predictions with observed rates of obesity-related complications in the UK clinical practice data are presented, to demonstrate the functionality of the COM and assess the impact of baseline glycemic status on its predictions while highlighting areas for further refinement of the model.

| Model structure
The COM is a Markov, closed-cohort model. In a Markov model, the disease being studied is divided into distinct and mutually exclusive states (health states) and transition probabilities are assigned to represent patients moving between these states over discrete time periods called "Markov cycles". By applying these transitions in the model and attaching estimates of resource use and health consequences to the states, followed by running the model over a large number of cycles, it is possible to estimate the longitudinal costs and outcomes associated with the disease.
The health states are chosen to represent clinically and economically important events in the disease process. The states are mutually exclusive: a patient can only be in one state at a time and cannot transition to a less severe health state. 15 The COM comprises 18 single or combined obesity-related health states ( Figure 1), intended to reflect the disease course and impact of effective weight management interventions for individuals who are currently living with BMI above 25 kg/m 2 . 16   Complications were selected for inclusion in the model because: (1) there is either strong or moderate evidence for their association with obesity (Table 3), based on a comprehensive report from the World Health Organization, 19 and also referenced in subsequent reports on the burden of obesity-related conditions; 20 Registry ( Figure 2D) may be used. 27,28 The risk for CVD as a recurrent event was based on estimates from the Framingham Recurring Coronary Heart Disease Study for individuals with normal glucose tolerance and for those with T2D; the UKPDS can be used as an alternative for individuals with T2D. 28,29 To reflect the increased risk of recurrent cardiovascular (CV) events in individuals with impaired glucose tolerance and a history of CV events, 14 the risk of CVD as a recurrent event for individuals with prediabetes was assumed to be the same as for those with T2D.
Risk estimates for other events and transitions to other health states are summarized in Table 4. These risk equations were selected following identification of relevant studies in the systematic review.
Appropriate studies for inclusion were those that focused on relevant populations; were relevant to the countries or regions of interest; reported on the association between BMI, other risk factors relevant to the model, where available, and the outcomes of interest; and were judged to be of high quality. High-quality studies were considered to be those with appropriate design and modeling, and use of large patient populations to develop and validate risk equations.

| Mortality
General population mortality (defined as age-and sex-specific allcause mortality) was included in the model based on country-specific life tables. Changes to the probability of mortality associated with MI, unstable angina, stroke, knee replacement, and certain cancers were made via adjustments to the general population mortality applied in the COM (Table 5).

| CPRD-HES study and Core Obesity Model comparative analysis
A recently published external validation of the COM showed that it reliably predicts the occurrence of obesity-related complications. 41 The aim of this analysis was to assess how baseline glycemic status Longitudinal data reflecting changes to BMI over time were not investigated in the CPRD-HES study; consequently, in this analysis, BMI was assumed to remain constant over time. No weight management intervention effects were considered in these analyses.

| Calculation of event rates
Cox-adjusted event rates for each BMI group were calculated by multiplying the crude event rates in the normal weight group by the To examine the impact of altering baseline prediabetes prevalence, a scenario analysis was conducted in which the rate of prediabetes for individuals of normal weight reported in the HSE was used for the reference group in the model. Baseline T2D prevalence was not altered in this analysis.   Table 8); however, event rates were still underpredicted in the other BMI groups (overall OLS LRL slope: 0.655; Table 9). The  Table 9); however, the negative R 2 value obtained from this analysis (−26.840; Table 9) limited the ability to fully interpret the result. All-cause mortality predictions were consistent in the scenario analysis ( Table 8), suggesting that prediabetes prevalence did not significantly affect mortality during the modeled time horizon.

| Comparison of predictions generated by different versions of the Core Obesity Model
As part of the development of the COM, a previous version (6.1) was subjected to an extensive validation process, according to best practice guidelines. 13 Results from a comparison between the current (8.0) and validated (6.1) versions of the COM (Table 8) showed that the validated version produced similar trends across BMI groups to those predicted by the current version, but that predicted event rates were generally slightly lower. Economic models can be used to extrapolate the long-term impacts of a disease and estimate the relative benefits of different treatment strategies. In conjunction with shorter-term data provided by clinical trials and observational studies, such projections are relevant to clinicians, payers, and policy-makers, particularly in the case of a common, chronic condition such as obesity. Best-practice guidance highlights the need for transparency and validation to ensure that the outputs of economic models can be interpreted with confidence by all stakeholders. 13 The aim of this study was to present the structure and components of the COM in a transparent manner and provide a single-study example of its predictive ability.
The COM incorporates a broad range of obesity-related health states, allowing for the presence of single and multiple comorbidities, and including complications both strongly and moderately related to obesity. When deriving data to develop risk equations for these complications, multiple relevant studies were considered, and those that were most appropriate based on study population and setting were selected. In their final appraisal determination for liraglutide, T A B L E 8 Observed event rates from the CPRD-HES study versus those predicted by the Core Obesity Model (versions 8.0 and 6.1)  It must also be noted that mortality rates in the CPRD-HES study (index period: January 2000-December 2010) are higher than those reported in several more recent studies. The 11.6%-14.0% mortality across BMI groups in this data set contrasts with rates of 7.1% in a study conducted by the Global BMI Mortality Collaboration, 48 8.0%

Incidence, crude event rates/1000 patient-years
in a 2018 study using CPRD data, 47 and 3.9% 51 and 4.0%, 50 respectively, in studies published in 2019 using data from the UK Biobank. This pattern is supported by the findings of a study that examined mortality in five survey periods from 1986 to 2009, which concluded that mortality is decreasing over time. 52 Such trends may be attributable, in part, to improvements in the management of obesity-related diseases during more recent decades. Therefore, the fact that mortality estimates generated by the COM are low Predictive models and the economic analyses performed by them are necessarily limited by the quality and scope of the data available.
For example, in the COM, some of the studies used to derive risk estimates did not include BMI as an independent risk factor 28 or did not estimate the impact of BMI above a certain threshold. 22,23,26 Therefore, the COM may underpredict disease risk for individuals with a BMI greater than 40 kg/m 2 ; this is reflected in the predicted T2D incidence in these analyses, which was lower than observed values in the highest BMI group. Furthermore, the COM is intended to reflect clinical practice as accurately as possible; however, epidemiological and database studies cannot capture all factors that affect obesity and disease risk. Adherence to and persistence with 278medication, as well as demographic characteristics and medical history, which may constitute important risk factors, are unlikely to be recorded fully in these databases. For example, socioeconomic status is implicated in a considerable proportion of obesity 53 but is not captured in CPRD or similar retrospective data sources. Finally, outcomes relating to CVD risk equations were subject to some assumptions as a result of the source material available: the risk of CVD as a first-time event was assumed to be the same for individuals with normal glucose tolerance and for those with prediabetes, and once an individual developed prediabetes, their risk of CVD as a recurrent event was the same as for those with T2D.
The COM improves on previous economic models of obesity [8][9][10][11] due to the inclusion of additional health states and baseline characteristics. The results of this study show that in the context of the UK clinical practice, the COM can predict rates of CV events across BMI groups and T2D in certain BMI groups, both of which are strongly linked to obesity. Further adjustment to the model prediction of mortality rates, especially at higher BMI levels, will improve and refine its overall ability to estimate the occurrence and health economic burden of obesity-related complications, providing a valuable tool to support healthcare decision-making.

ACKNOWLEDGMENT
The authors acknowledge the medical writing assistance of Phar-maGenesis Oxford Central, which was funded by Novo Nordisk A/S.

AUTHOR CONTRIBUTIONS
All authors contributed to the study design, data interpretation, and writing and critical review of manuscript content. Sandra Lopes, Mark Lamotte, and Anamaria-Vera Olivieri were involved in performing and reviewing the data analysis.