Combining administrative and clinical data to stratify surgical risk

Area of Science:

• Health Services Research
• Medical Informatics
• Surgical Outcomes Research

Background:

• Administrative claims data (ADM) from hospital discharges have limitations for risk stratification (RS) due to sparse clinical information and questionable predictive accuracy.
• Previous use of ADM for RS has been met with skepticism, hindering reliable assessment of surgical outcomes.

Purpose of the Study:

• To determine if administrative claims data (ADM) can be enhanced with present-on-admission (POA) codes and accessible clinical data.
• To create a refined database capable of supporting valid risk stratification (RS) for surgical outcomes.

Main Methods:

• Logistic regression analysis was employed to select predictor variables for RS of mortality and postoperative complications in specific surgical procedures.
• RS models were progressively developed, starting with age and ADM, and incrementally adding POA codes, laboratory data, vital signs, clinical findings, and composite scores.
• Model performance was evaluated using c-statistics, prediction errors, and measures of hospital-based systematic bias.

Main Results:

• The integration of POA codes and numerical laboratory results into ADM led to substantial improvements in all analytical performance measures.
• Conversely, incorporating difficult-to-obtain key clinical findings yielded only marginal improvements in prediction accuracy.

Conclusions:

• Enhancing administrative claims data (ADM) with POA codes and readily available laboratory data offers an efficient method for accurate risk stratification.
• This approach supports precise measurement of clinical outcomes for surgical patients.