Summary perioperative risk metrics within the electronic medical record predict patient-level cost variation in...

Automated data extraction from the electronic medical record is fast, scalable, and inexpensive compared with manual abstraction. However, concerns regarding data quality and control for underlying patient variation when performing retrospective analyses exist. This study assesses the ability of summary electronic medical record metrics to control for patient-level variation in cost outcomes in pancreaticoduodenectomy.

Patients that underwent pancreaticoduodenectomy from 2014 to 2018 at a single institution were identified within the electronic medical record and linked with the National Surgical Quality Improvement Program. Variables in both data sets were compared using interrater reliability. Logistic and linear regression modelling of complications and costs were performed using combinations of comorbidities/summary metrics. Models were compared using the adjusted R² and Akaike information criterion.

A total of 117 patients populated the final data set. A total of 31 (26.5%) patients experienced a complication identified by the National Surgical Quality Improvement Program. The median direct variable cost for the encounter was US$14,314. Agreement between variables present in the electronic medical record and the National Surgical Quality Improvement Program was excellent. Stepwise linear regression models of costs, using only electronic medical record-extractable variables, were non-inferior to those created with manually abstracted individual comorbidities (R² = 0.67 vs 0.30, Akaike information criterion 2,095 vs 2,216). Model performance statistics were minimally impacted by the addition of comorbidities to models containing electronic medical record summary metrics (R² = 0.67 vs 0.70, Akaike information criterion 2,095 vs 2,088).

Summary electronic medical record perioperative risk metrics predict patient-level cost variation as effectively as individual comorbidities in the pancreaticoduodenectomy population. Automated electronic medical record data extraction can expand the patient population available for retrospective analysis without the associated increase in human and fiscal resources that manual data abstraction requires.