1050 — Validation of Stroke Meaningful Use Measures in a National Electronic Health Record System

Phipps MS, Stroke QUERI; Baltimore VA; University of Maryland School of Medicine; Fahner J, Stroke QUERI; Sager D, Stroke QUERI; VA Center for Health Information and Communication; Coffing J, Stroke QUERI; VA Center for Health Information and Communication; Maryfield B, Stroke QUERI; VA Center for Health Information and Communication; Williams LS, Stroke QUERI; VA Center for Health Information and Communication;

Objectives:
The Affordable Care Act included eight stroke indicators in its Meaningful Use (MU) program. This project evaluated stroke MU measures and one VHA stroke quality indicator (QI) in national VHA data and determined sources of error in using centralized electronic health record (EHR) data.

Methods:
We developed local SQL queries to generate the QIs, then modified them to run on VA Central Data Warehouse (CDW) data, mapping each local data element to the corresponding CDW data element and table. Numerator and denominator results were generated from CDW data in 2130 ischemic stroke admissions in 11 VA hospitals. Local and CDW results were compared to chart review. Mismatch reports were examined to identify, categorize, and correct sources of error. We calculated the raw proportion of matching cases, sensitivity/specificity, positive/negative predictive values (PPV/NPV) for the numerators and denominators, and an overall kappa statistic for each QI.

Results:
Four MU measures (venous thromboembolism (VTE) prophylaxis, antithrombotic (AT) by day 2, consideration for rehabilitation, and AT at discharge) and the VHA QI (documentation of NIH Stroke Scale score) are reported here. The proportion of matched cases between CDW and chart denominators ranged from 95.4%-99.7% and 87.7%-97.9% for the numerators. PPVs tended to be higher (range 96.8%-100% in CDW) and more similar between local and CDW data, with NPVs less stable and generally lower for both denominators and numerators (range 8.6%-98.8% in CDW). Common errors included difficulty in identifying: 1) comfort care orders, 2) mechanical VTE prophylaxis devices, 3) medications not recorded in VA bar code medication administration or pharmacy data (e.g. given in ED or purchased over the counter), and 4) provider documentation of contraindications.

Implications:
Stroke MU indicators can be relatively accurately generated from existing EHR systems but accuracy decreases slightly in central compared to local data sources. To improve stroke MU measure accuracy, EHRs should include standardized data elements for devices, Comfort Care status, recording contraindications, and electronic documentation for all medications.

Impacts:
EHR-based indicators match chart review in nearly 90% of cases, may be cost-effective, and would allow frontline providers timely access to estimates of hospital-level quality of stroke care.