Login to Access Video or Poster Abstract: MP76-07
Sources of Funding: None

Introduction

Assessment of surgical risk is integral to patient counseling and shared clinical decision-making. The American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) surgical risk calculator is an easily accessible, online tool for predicting surgical outcomes after a variety of procedures. Little is known of the tool's applicability to urologic surgery. We sought to evaluate the predictive value of the calculator in a tertiary referral cohort of patients undergoing robot-assisted partial nephrectomy (RAPN).

Methods

We queried our prospectively maintained multi-institutional database of RAPN (n=1260) from 2008 to 2016. Preoperative details of 300 randomly selected patients were entered into the calculator. The predicted rates of complications were compared with the actual rates of observed complications. Validation of the calculator was assessed by receiver-operator area under the curve (AUC) for discrimination and Brier score (BS) for calibration. Calculated BS was also compared to a null model (null-BS); a BS lower than the null model indicates stronger predictive power for that individual outcome where a BS of zero indicates perfect prediction.

Results

The observed rate of any complication in our cohort was 14%, comparable with that reported in the literature, while the mean predicted rate of any complication was 5.42%. The calculated AUC for any complications was 0.51. Our cohort demonstrated a serious complication (Clavien Score ≥ 3) rate of 3.67%, lower than the predicted rate of 4.89% (AUC 0.55). The majority of the captured complications had a low BS, indicative of good calibration. However, the calculated AUC was low for all outcomes, indicating poor discrimination ability. Venous thromboembolism (VTE) and readmission had the highest AUCs - 0.67 and 0.69, respectively.

Conclusions

The ACS-NSQIP risk calculator poorly predicted most complications after RAPN. The model had marginal accuracy for predicting VTE and readmissions, and good accuracy for predicting the rate of serious complications, but it lacked the power to discriminate which patients were at risk to have such outcomes. These findings suggest the need for a more tailored outcome prediction model to accurately assist surgeons in counseling patients undergoing RAPN.

Funding

None