Excess Risk of Mortality and Complications Associated with Newly Diagnosed Cases of Diabetes in Ontario, Canada

Impact of diabetes on healthcare costs in a population-based cohort: a cost analysis

AIMS

To estimate the healthcare costs attributable to diabetes in Ontario, Canada using a propensity-matched control design and health administrative data from the perspective of a single-payer healthcare system.

METHODS

Incident diabetes cases among adults in Ontario were identified from the Ontario Diabetes Database between 2004 and 2012 and matched 1:3 to control subjects without diabetes identified in health administrative databases on the basis of sociodemographics and propensity score. Using a comprehensive source of administrative databases, direct per-person costs (Canadian dollars 2012) were calculated. A cost analysis was performed to calculate the attributable costs of diabetes; i.e. the difference of costs between patients with diabetes and control subjects without diabetes.

RESULTS

The study sample included 699 042 incident diabetes cases. The costs attributable to diabetes were greatest in the year after diagnosis [C$3,785 (95% CI 3708, 3862) per person for women and C$3,826 (95% CI 3751, 3901) for men], increasing substantially for older age groups and patients who died during follow-up. After accounting for baseline comorbidities, attributable costs were primarily incurred through inpatient acute hospitalizations, physician visits and prescription medications and assistive devices.

CONCLUSIONS

The excess healthcare costs attributable to diabetes are substantial and pose a significant clinical and public health challenge. This burden is an important consideration for decision-makers, particularly given increasing concern over the sustainability of the healthcare system, aging population structure and increasing prevalence of diabetic risk factors, such as obesity.

General population versus disease-specific event rate and cost estimates: potential bias for economic appraisals

Economic appraisals are increasingly being used for reimbursement decision making. Differences exist in the population data sources used in different studies and these differences may result in errors or biased estimates. A review of the literature suggests that very little has been written on this topic and guidelines and good practice documents are silent on the issue. Using illustrative examples, it was found that the population chosen for event/complication costing did not have a large impact on a cost-effectiveness analysis; however, the choice of population did have a large impact for cost-of-illness (COI) estimation. It was found that not controlling for event/complication rates in a nondiseased population resulted in a 15% inflated COI estimate and using event costs from the general population underestimated COI by 20-32%. Our analysis suggests that using event costs from the general population instead of a diseased population may not have a significant impact on cost-effectiveness estimates; however, COI studies should only use excess event/complication rates and should also only use event costs from populations with the disease.

A review of methods used in long-term cost-effectiveness models of diabetes mellitus treatment

Diabetes mellitus is a major healthcare concern from both a treatment and a funding perspective. Although decision makers frequently rely on models to evaluate the long-term costs and consequences associated with diabetes interventions, no recent article has reviewed the methods used in long-term cost-effectiveness models of diabetes treatment. The following databases were searched up to April 2008 to identify published economic models evaluating treatments for diabetes mellitus: OVID MEDLINE, EMBASE and the Thomson's Biosis Previews, NHS EED via Wiley's Cochrane Library, and Wiley's HEED database. Identified articles were reviewed and grouped according to unique models. When a model was applied in different settings (e.g. country) or compared different treatment alternatives, only the original publication describing the model was included. In some cases, subsequent articles were included if they provided methodological advances from the original model. The following data were captured for each study: (i) study characteristics; (ii) model structure; (iii) long-term complications, data sources, methods reporting and model validity; (iv) utilities, data sources and methods reporting; (v) costs, data sources and methods reporting; (vi) model data requirements; and (vii) economic results including methods to deal with uncertainty. A total of 17 studies were identified, 12 of which allowed for the conduct of a cost-effectiveness analysis and a cost-utility analysis. Although most models were Markov-based microsimulations, models differed with respect to the number of diabetes-related complications included. The majority of the studies used a lifetime time horizon and a payer perspective. The DCCT for type 1 diabetes and the UKPDS for type 2 diabetes were the trial data sources most commonly cited for the efficacy data, although several non-randomized data sources were used. While the methods used to derive the efficacy data were commonly reported, less information was given regarding the derivation of the utilities or the costs. New interventions were generally deemed cost effective based on ten studies presenting results. Authors relied mostly on univariate sensitivity analyses to test the robustness of their models. Although diabetes is a complex disease, several models have been developed to predict the long-term costs and consequences associated with diabetes treatment. Combined with previous findings, this review supports the development of a reference case that could be used for any new diabetes models. Models could be enhanced if they had the capacity to deal with both first- and second-order uncertainty. Future research should continue to compare economic models for diabetes treatment in terms of clinical outcomes, costs and QALYs when applicable.