OBJECTIVES: Obesity is frequently undercoded in health care or claims data sets. The aim of this project was to develop a predictive model based on comorbidities, identifying patients at high risk of obesity, as a solution to overcome the limited documentation in claims data sets.

METHODS: In a comprehensive literature search, comorbidities associated with obesity were identified. Additionally, associated comorbidities were identified in claims data (n=3,244,611) from the statutory health insurance system in Germany. From these data, a predictive model was developed, and then validated by data from the UK biobank (n=502,384). In contrast to German claims data, where obesity is documented only using the ICD-10 diagnosis code E66, in the UK biobank diagnosis could be validated in addition by documented body mass index (BMI) of >30 kg/m².

RESULTS: Using multiple comorbidities, the model showed good predictive performance in identifying patients living with obesity in German claims data, with an area under curve (AUC) of 83%. The validation based on UK biobank data showed a slightly decreasing predictive power (AUC: 76%), however, it confirmed that the model can be transferred to other databases. An enhanced validation using BMI to identify obesity revealed a further decrease in predictive performance (AUC: 69%).

CONCLUSIONS: The results indicate the robust predictive power of identifying obesity in claims data sets using this model. This validated model may be used to analyze obesity-related healthcare efforts in different secondary data sets. Furthermore, the lower predictive power of detecting patients with a BMI >30 kg/m², suggests that patients with obesity who have not yet been diagnosed as such by physicians, suffer from fewer comorbidities and are therefore more difficult to identify using such models.