QUANTIFYING EHR DISCONTINUITY IN PRAGMATIC TRIALS: INSIGHTS FROM THE NIDA CTN-074 PRIMARY CARE OPIOID USE DISORDERS TREATMENT (PROUD)STUDY
Author(s)
Ali Jalali, PhD1, Caroline Andy, MS1, Philip J. Jeng, BS, MS1, Catherine C. Rabin, BS1, Onchee Yu, MS2, Jennifer Bobb, PhD2, Paige Wartko, PhD, MPH2, Katharine Bradley, MD, MPH2, Sean M. Murphy, PhD1;
1Weill Cornell Medical College, New York, NY, USA, 2Kaiser Permanente Washington Health Research Institute, Seattle, WA, USA
1Weill Cornell Medical College, New York, NY, USA, 2Kaiser Permanente Washington Health Research Institute, Seattle, WA, USA
OBJECTIVES: Pragmatic trials and retrospective observational studies increasingly rely on electronic health records (EHRs) to evaluate patient‑level outcomes. To improve generalizability, these studies often pool data across multiple health systems. However, when systems differ in service integration or operate in overlapping catchment areas, health service utilization may be differentially captured, introducing systematic data discontinuity. The objective of this study was to quantify the extent to which variation in EHR discontinuity affects observed health service utilization among patients with opioid use disorder (OUD).
METHODS: We analyzed 24 months of pre‑intervention health service utilization data among 10,122 patients with OUD enrolled in the NIDA CTN‑074 PRimary Care Opioid Use Disorders Treatment (PROUD) study across 8 U.S. health systems. Each health system was classified a priori by expected level of data continuity (high, moderate, or low), separately for acute (emergency department admissions and inpatient days) and ambulatory care (outpatient and primary care visits). Differences in average per‑person utilization were estimated using random intercept linear mixed‑effects models adjusted for patient characteristics and comorbid conditions.
RESULTS: Relative to high-continuity systems, patients in low- and moderate-continuity systems had an expected 3.4 (95% CI: 2.4-4.5) and 2.5 (95% CI: 2.2-2.7) fewer captured ED visits, respectively, over the 24-month study period. Low- and moderate-continuity systems had an expected 3.3 (95% CI: 1.9-4.8) and 2.9 (95% CI: 2.1-3.9) fewer outpatient non-behavioral health visits relative to high-continuity systems over the same period. Inpatient days and primary care visits did not differ significantly.
CONCLUSIONS: Variation in health system EHR data continuity meaningfully affects the completeness of observed health service utilization data among patients with OUD. These findings demonstrate that EHR discontinuity is contributing source of measurement bias in multisystem studies and underscores the importance of incorporating health system-level stratification into comparative effectiveness and causal inference research using EHR data.
METHODS: We analyzed 24 months of pre‑intervention health service utilization data among 10,122 patients with OUD enrolled in the NIDA CTN‑074 PRimary Care Opioid Use Disorders Treatment (PROUD) study across 8 U.S. health systems. Each health system was classified a priori by expected level of data continuity (high, moderate, or low), separately for acute (emergency department admissions and inpatient days) and ambulatory care (outpatient and primary care visits). Differences in average per‑person utilization were estimated using random intercept linear mixed‑effects models adjusted for patient characteristics and comorbid conditions.
RESULTS: Relative to high-continuity systems, patients in low- and moderate-continuity systems had an expected 3.4 (95% CI: 2.4-4.5) and 2.5 (95% CI: 2.2-2.7) fewer captured ED visits, respectively, over the 24-month study period. Low- and moderate-continuity systems had an expected 3.3 (95% CI: 1.9-4.8) and 2.9 (95% CI: 2.1-3.9) fewer outpatient non-behavioral health visits relative to high-continuity systems over the same period. Inpatient days and primary care visits did not differ significantly.
CONCLUSIONS: Variation in health system EHR data continuity meaningfully affects the completeness of observed health service utilization data among patients with OUD. These findings demonstrate that EHR discontinuity is contributing source of measurement bias in multisystem studies and underscores the importance of incorporating health system-level stratification into comparative effectiveness and causal inference research using EHR data.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
MSR5
Topic
Methodological & Statistical Research
Topic Subcategory
Confounding, Selection Bias Correction, Causal Inference
Disease
No Additional Disease & Conditions/Specialized Treatment Areas