Abstract WP246: Probabilistic Linkage of De-identified Prehospital and Hospital Data for Stroke Patients Achieves Similar Match Rates to Manual Linkage

    loading  Checking for direct PDF access through Ovid

Abstract

Background: Acute ischemic stroke patients that are recognized by Emergency medical services (EMS) receive expedited care are more likely to receive t-PA. In order to study the continuum of acute stroke care and ensure accurate case ascertainment, prehospital and inpatient medical records must be linked. We tested the feasibility of probabilistic matching to link prehospital EMS data and in-hospital records using data from existing de-identified registries, and compared results to manual linkage.

Methods: A registry containing all EMS-transported suspect or confirmed stroke cases in Kent County, Michigan was used to identify subjects discharged from either of two primary stroke centers with a diagnosis of stroke or TIA. The registry contains EMS data obtained through direct query of EMS electronic medical record systems which is linked manually to hospital data using patient identifiers (name, date of birth). For the probabilistic match, de-identified data for EMS suspected stroke transports from the EMS information system (EMSIS) was provided by the county medical control authority. Probabilistic linkage between hospital stroke discharges and EMSIS data was done using LinkPlus software. Age, sex, transport agency, and date of service were used as matching variables. Probabilistic match rates were compared to manual registry match rates among EMS suspect stroke cases using a chi square test.

Results: Over a 6-month period in 2015, 172 hospital-confirmed stroke and TIA patients arrived by EMS at the target hospitals. Over the same period, 214 EMS suspect stroke cases were identified from the registry and 199 EMS suspect stroke cases were identified in EMSIS. Probabilistic linkage correctly matched 64.5% (111/172) of confirmed strokes to prehospital records in EMSIS, which was not significantly different from the manual registry match rate of 73.8% (127/172, p=0.24). There were 16 cases in the manual match that were missed by the probabilistic match: 5 (2.9%) due to age mismatch and 11 (6.4%) were missing from the EMSIS data. The remaining non-matched cases represented EMS missed stroke cases.

Conclusion: Probabilistic matching of de-identified pre-hospital and hospital data is a feasible approach for the evaluation of prehospital stroke care.

Related Topics

    loading  Loading Related Articles