Predicting surgical risk: exclusion of laboratory data set maintains predictive accuracy

Am J Med Qual. 2013 Jul-Aug;28(4):352-6. doi: 10.1177/1062860612474063. Epub 2013 Feb 11.

Abstract

Studies have demonstrated that predictive accuracy of acuity models flattens when more than 10 variables are used. The authors hypothesized that the exclusion of laboratory data would produce a reliable predictive model. The American College of Surgeons National Surgical Quality Improvement Program data set from 2005 to 2008 was reviewed. Logistic regression was used to develop models from 86 preoperative variables to predict 30-day morbidity and mortality. Models were compared by measuring area under the receiver operating characteristic (AUROC) values that then were analyzed with unpaired t test. As the number of variables decreased, the change in the AUROC for mortality and morbidity were not statistically significant for 10- and 5-variable models. Although the AUROC for acuity models decreased slightly for morbidity and mortality when laboratory values were excluded, these changes were not statistically significant. This study shows that models developed to predict surgical outcomes can achieve similar predictive accuracy without laboratory data.

Keywords: laboratory data; risk models; surgical outcomes.

MeSH terms

  • Diagnostic Tests, Routine*
  • Forecasting / methods*
  • Humans
  • Models, Statistical*
  • Postoperative Complications / epidemiology*
  • Quality of Health Care
  • ROC Curve
  • Risk Assessment
  • Surgical Procedures, Operative / adverse effects*
  • Surgical Procedures, Operative / standards
  • United States / epidemiology