Estimating Under-diagnosis of Patients in Chronically-ill Populations

Diagnosis coding in administrative data is known to be inconsistent and incomplete, introducing inaccurate assessment of patients health outcomes. The under-diagnosis of members with a target chronic condition reduces the correlation of that chronic condition with associated events. Yet, only a few studies have evaluated the extent of under-reporting of chronic conditions in administrative data. In this study, we developed a novel framework to identify latent members, or those who have not yet been identified with a target chronic condition through claims-based diagnosis but are likely to have some degree of the condition. We applied our innovative approach to estimate the prevalence of a chronic-related event, based on the population of observed and latent members. We provided a detailed illustration that treats incontinence as our target chronic condition while examining four types of incontinence-related events: urinary-tract infections (UTIs), slips and falls, dermatitis, and behavioral disruptions. All analysis relied on the 5% Medicare sample for a continuously enrolled cohort between 2014-2018. Using our novel approach, we were able to increase our identification of incontinence from the 11.2% diagnosed in 2018 to an estimated prevalence in 2018 of 34.7% among fee-for-service Medicare beneficiaries over the age of 65. Similarly, our estimation of UTIs associated with those with incontinence increased from 38% to 68%, from 20% to 41% for IAD, 22% to 54% for slips and falls, and from 26% to 57% for behavior disruptions.