Back

Population level risk stratification of hypertensive patients by predictive modeling

Bandhakavi, S.; Liu, Z.; Karigowda, S.; McCammon, J.; Rahmanian, F.; Lavoie, H. M.

2021-04-30 health informatics
10.1101/2021.04.27.21256198 medRxiv
Show abstract

ObjectiveWe recently reported that hypertension (HTN) patients having at least three rounds of distinct treatment options (atl_three_roto) in a 12-month window have elevated risk of next-year complications. However, early identification of these "challenge to treat" patients is non-trivial and "drivers" of complications in these vs remaining HTN patients are not fully defined. To address these challenges/gaps, we present predictive models for preceding outcomes, delineate their "drivers", and highlight value of their integration for population level risk stratification/management of HTN patients. Materials and Methods2.47 million HTN patients enrolled through 2015-2016 were selected from a nation-wide commercial claims database. Features associated with their treatment patterns, comedications, and comorbidities were extracted for 2015 and used to model/predict 2016 outcomes of atl_three_roto status and/or HTN complications. Logistic regression-derived odds-ratios were used to delineate drivers of each outcome. ResultsPrior year treatment patterns, specific hypertension drugs (anti-hypertensives, calcium channel blockers, beta blockers), and congestive heart failure most increased future odds of atl_three_roto status. Regardless of prior year atl_three_roto status, specific comorbidities (renal disease, congestive heart failure, myocardial infarction, vascular disease, diabetes with chronic complications) and comedications (beta blockers, cardiac agents, anti-lipidemics) most increased future odds of HTN complications. Proof-of-concept analysis with an independent dataset demonstrated that integrating these model predictions/drivers thereof can be leveraged for risk stratification/management of HTN patients. DiscussionIntegrating predictions and their "drivers" from above models supports early identification and targeted management of "at-risk" HTN patients. ConclusionWe have developed a predictive modeling based approach for risk stratification and management of HTN patients.

Matching journals

The top 7 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.