Digital Treatment Signatures: Cardiovascular Risk Prediction Using Antihypertensive Medication Fill History
- 1Department of Clinical Pharmacy Translational Science, College of Pharmacy, University of Michigan
Abstract
ObjectiveTaking a blood pressure medication is not a single fact but a trajectory of fills, gaps, and regimen changes. Dispensing records capture that trajectory, yet risk calculators reduce it to a yes/no treatment indicator and quality programs to proportion of days covered (PDC). We asked how much of that discarded signal is recoverable.
Materials and MethodsWe studied 7,625 adults treated for hypertension using linked electronic health record and pharmacy dispensing data. From a 1-year medication-history window, we predicted 3-year myocardial infarction, stroke, or all-cause death. Holding cohort, covariates, horizon, and validation fixed, we compared PDC, engineered temporal and regimen features, and the ordered fill sequence modeled with an attention-based bidirectional LSTM survival network. Factorial analyses separated representation from model class.
ResultsC-index increased from 0.7113 with clinical factors alone to 0.7131 with PDC, 0.7273 with engineered summaries, and 0.7396 with the ordered sequence ({Delta}C=0.0283; 95% CI, 0.0189-0.0382). The sequence exceeded engineered summaries by 0.0123 (0.0040-0.0202). Richer representations improved discrimination by 0.0181-0.0191, approximately three times the 0.0059-0.0078 from changing model class. Among patients with PDC [≥]0.80, PDC ranked risk poorly (C=0.4628), whereas sequence-model strata had 3-year event rates of 3.6%-22.8%.
DiscussionA medication historys prognostic value lies in its temporal structure, which aggregate adherence measures discard, so the sequence advantage reflects preserved information rather than model complexity. External validation is needed.
ConclusionTraditional metrics of antihypertensive treatment conceal substantial risk heterogeneity, whereas a tokenized dispensing timeline recovers a prognostic treatment signature that improves cardiovascular risk prediction from existing records.
Related
- Patient preferences for antihypertensive medication dispensing trade-offs in cardiovascular disease prevention: national survey in Australia
- Medication history, not clinical profiles, predicts antihypertensive treatment persistence across health systems
- Antihypertensive Medication Class and Incident Alzheimers Disease and Related Dementias: External Replication Across Two Independent Healthcare Data Sources-the Vanderbilt EHR Synthetic Derivative and US Medicare Claims
- Knowledge and family support in relation to antihypertensive medication adherence: A cross-sectional study
- PATIENT PROFILING AND OUTCOME OF ANTIHYPERTENSIVE TREATMENT. An Exploratory Analysis of the VALUE Trial.
BibTeX
@misc{zheng2026digital,
title = {{Digital Treatment Signatures: Cardiovascular Risk Prediction Using Antihypertensive Medication Fill History}},
author = {Y. Zheng and K. B. Farris and A. B. Coe and M. P. Dorsch and K. Najarian and C. A. Lester},
howpublished = {medRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.03.26362175},
url = {https://www.medrxiv.org/content/10.64898/2026.09.03.26362175v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Digital Treatment Signatures: Cardiovascular Risk Prediction Using Antihypertensive Medication Fill History
AU - Y. Zheng
AU - K. B. Farris
AU - A. B. Coe
AU - M. P. Dorsch
AU - K. Najarian
AU - C. A. Lester
PY - 2026
DA - 2026/09/03
PB - medRxiv
DO - 10.64898/2026.09.03.26362175
UR - https://www.medrxiv.org/content/10.64898/2026.09.03.26362175v1
AB - ObjectiveTaking a blood pressure medication is not a single fact but a trajectory of fills, gaps, and regimen changes. Dispensing records capture that trajectory, yet risk calculators reduce it to a yes/no treatment indicator and quality programs to proportion of days covered (PDC). We asked how much of that discarded signal is recoverable. Materials and MethodsWe studied 7,625 adults treated for hypertension using linked electronic health record and pharmacy dispensing data. From a 1-year medication-history window, we predicted 3-year myocardial infarction, stroke, or all-cause death. Holding cohort, covariates, horizon, and validation fixed, we compared PDC, engineered temporal and regimen features, and the ordered fill sequence modeled with an attention-based bidirectional LSTM survival network. Factorial analyses separated representation from model class. ResultsC-index increased from 0.7113 with clinical factors alone to 0.7131 with PDC, 0.7273 with engineered summaries, and 0.7396 with the ordered sequence ({Delta}C=0.0283; 95% CI, 0.0189-0.0382). The sequence exceeded engineered summaries by 0.0123 (0.0040-0.0202). Richer representations improved discrimination by 0.0181-0.0191, approximately three times the 0.0059-0.0078 from changing model class. Among patients with PDC [≥]0.80, PDC ranked risk poorly (C=0.4628), whereas sequence-model strata had 3-year event rates of 3.6%-22.8%. DiscussionA medication historys prognostic value lies in its temporal structure, which aggregate adherence measures discard, so the sequence advantage reflects preserved information rather than model complexity. External validation is needed. ConclusionTraditional metrics of antihypertensive treatment conceal substantial risk heterogeneity, whereas a tokenized dispensing timeline recovers a prognostic treatment signature that improves cardiovascular risk prediction from existing records.
ER -