Date of Award
5-2023
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Computer Science
Committee Chair/Advisor
Ron Gimbel
Committee Member
Brent Egan
Committee Member
Xia Jing
Committee Member
Jihad Obeid
Committee Member
Lior Rennert
Abstract
Background
Hypertension is the leading modifiable risk factor for cardiovascular disease and consequent mortality worldwide. In the U.S., more than half of hypertension cases remain uncontrolled, despite availability of effective pharmaceutical treatment options. Evidence suggests that therapeutic inertia, defined as clinician failure to initiate or increase therapy when treatment goals are unmet, is the most influential barrier to improving hypertension control. Substantial rates of therapeutic inertia have been reported in ambulatory primary care settings where hypertension is typically treated and managed. Understanding and overcoming the forces driving therapeutic inertia in hypertension management is a critical strategy to reach population health goals for blood pressure control and cardiovascular disease prevention.
Objectives
Three embedded studies within this dissertation that include: (1) descriptive and predictive modeling of antihypertensive therapeutic inertia, (2) a model of antihypertensive treatment selection, and (3) a propensity-score matched model of observed reductions in blood pressure after increasing dose or adding new classes of antihypertensive medication using electronic health record (EHR) data generated from real-world clinical practice.
Materials and Methods
Data for defining and modeling antihypertensive therapeutic inertia comes from five health care organizations; four located in the Southeast and one in the Midwest U.S. EHR data extracted from each system used in these analyses include patient demographic information, diagnoses, procedures, medications, vital signs, and laboratory measurements. Mixed-effects regression, classification trees, and ensemble learning, and propensity-score matching are applied to produce descriptive and predictive models of antihypertensive therapeutic inertia and intensification, treatment selection, and treatment effectiveness.
Results
For 120,755 patients with hypertension, therapeutic inertia was indicated at 84.1% of 168,222 visits where BP was uncontrolled (>140/>90mmHg). Therapeutic intensification occurred via dose increase of existing medication at 6.6% of visits, and addition of a new medication class at 9.2% of visits with uncontrolled BP. Mixed-effects modeling of patient and clinical variables extracted from the electronic health record accounted for 13.2% of the variance in therapeutic inertia vs. intensification among visits with uncontrolled BP. Gradient boosted classification trees produced the strongest predictive model of therapeutic inertia (test AUC: 0.748). Mixed-effects modeling explained 38.5% of the variance between treatment selection options. Propensity-score matched cases of treatment selection groups found a 1.31 mmHg greater reduction in SBP when a new class of medication was added.
Discussion
Patient, clinical, and encounter related variables extracted from the EHR did not account for a significant proportion of the observed variance in antihypertensive therapeutic inertia vs. intensification and increasing dose vs. adding a new medication. Consequently, predictive modeling using these variables was limited in performance. However, modeling of the relationship between EHR derived variables and therapeutic inertia/intensification and treatment selection was sufficiently robust to determine the contribution of patient and visit related clinical factors to likelihood of antihypertensive treatment action, and to evaluate the best methods for prediction of hypertension treatment events.
Recommended Citation
Martin, Benjamin, "Modeling Antihypertensive Therapeutic Inertia And Intensification To Support Clinical Action Toward Hypertension Control" (2023). All Dissertations. 3278.
https://tigerprints.clemson.edu/all_dissertations/3278