(019) A new tool to estimate comorbidities at the population-level in the United States

Publication Number: 1015

Background: No comprehensive database exists to provide readily accessible information about patterns of comorbidities needed by pharmacoepidemiologists and pharmacovigilance teams. Secondary analysis of Global Burden of Disease (GBD) estimates, which are widely used to understand distributions of population health measures over time, space, and by age and sex, can fill this evidence gap.

Objectives: To define a methodology to evaluate patterns of disease occurrence among subpopulations with key underlying conditions that are consistent with the existing GBD framework.

Methods: We leveraged patient-level data on disease comorbidity in combination with existing population-level GBD outputs and analytic architecture. We developed estimates of disease prevalence among specific patient populations in the US using GBD estimates to inform the prevalence “envelope” and patient-level survey data to inform the “overlap” between paired conditions. Briefly, nationally-representative patient-level data from 170,000+ adults aged 20 and older were compiled from multiple years of the Medical Expenditure Panel Survey (MEPS; 2012-2020). We mapped MEPS conditions (coded in ICD-9 and ICD-10) to the list of conditions modeled as part of the GBD. For each condition pair, we fit a logistic regression to determine odds of a comorbid condition associated with having an underlying condition, adjusting for age and sex, using the survey package in R to account for the survey design and population weights. The models were run twice for each condition pair, swapping the underlying and comorbid condition indicators as the independent and dependent variables, respectively. The mean of the two coefficients was used to transform the existing GBD estimates to arrive at sex-specific comorbid prevalence rates for each 5-year age band for all condition pairs included in the analysis. To readily query and visualize results from this analysis, we developed a web-based tool.

Results: The approach we developed allowed us to generate age- and sex-specific comorbidity estimates for 762 pairwise combinations of conditions. As an example, among an estimated 36m adults with type 2 diabetes in the US, we estimate that 8% (N=3m) have anxiety disorders as a comorbidity. Prevalence rates of anxiety are approximately 44% higher among females than males with diabetes, and peak for the 30-34 year old age group. The age band estimated to have the largest number of comorbid anxiety and type 2 diabetes cases is 60-64 (N=480k).

Conclusions: This analysis provides a standardized approach for generating estimates of prevalence of commonly occurring conditions across age and sex which are calibrated to the GBD’s population-based estimates of disease prevalence.