(40) Bayesian Nonparametric Comorbidity Analysis of Type 2 Diabetes Mellitus (T2DM)

Sunday, August 28, 2022

8:00 AM – 1:45 PM CEST

Location: Hall C

Publication Number: 1085

Presenting Author(s)

Adrian Martinez-De la Torre
– Institute of Pharmaceutical Sciences, Department of Chemistry and Applied Biosciences, ETH Zurich, Zurich, Switzerland, Zurich, Z�rich, Switzerland

Co-Author(s)

Fernando Perez-Cruz
– Swiss Data Science Center, ETH Zurich and EPFL, Switzerland, Switzerland
Stefan Weiler
– Institute of Pharmaceutical Sciences, Department of Chemistry and Applied Biosciences, ETH Zurich, Zurich, Switzerland., Switzerland
Andrea Burden
– Professor, Institute of Pharmaceutical Sciences, Department of Chemistry and Applied Biosciences, Zurich, Switzerland

Background: Patients with Type 2 Diabetes Mellitus (T2DM) tend to develop a constellation of chronic comorbidities over time. These are associated with high drug utilization and adverse events.
There is an increasing need for understanding the frequently complex disease progression over time. Some methodological approaches have been applied, for instance using latent class growth analysis (LCGA) or principal component analysis (PCA). Nonetheless, Bayesian nonparametric methods have not yet been used for T2DM comorbidity progression.

Objectives: We aimed to find different profiles of patients and their disease progression after T2DM onset using Bayesian nonparametric models in primary care electronic medical records.

Methods: The IQVIA Medical Research Data incorporating data from THIN, a Cegedim database of anonymized electronic health records, was used to identify all patients with a first-ever prescription for a non-insulin antidiabetic drug (NIAD) between January 2006 and December 2019. In order to understand disease progression, we selected 50 chronic comorbidities of interest. We used Bayesian nonparametric latent models to model disease clusters and their progression over time and computed the Observed-Expected (O/E) ratio to identify comorbidities with a higher prevalence than expected.

Results: A total of 175,383 patients were included in the study (97,148 males, 78,235 females), who were on average 60.4 years old and had a BMI of 32.5 at index date (start of NIAD). Progression patterns could be explained by 14 latent features, each of them associated with a specific main disease. We found that cardiovascular diseases such as congestive heart failure or intermittent claudication were rapidly progressing in T2DM patients. Moreover, we observed that the onset of certain diseases led to the appearance of further complications. For instance, the onset of congestive heart failure was associated to an increase in the prevalence of chronic kidney disease and atrial fibrillation.

Conclusions: This was the first study that used a Bayesian nonparametric latent model to characterize T2DM comorbidities in an electronic health record database. We observed that the onset of certain comorbidities increases the probability of developing further complications. The identified comorbidity clusters may be used to improve treatment strategies for multimorbid T2DM patients.