Meeting Brochure and registration form      SMDM Homepage

Monday, 24 October 2005
4

VALIDATION AND UPDATING OF A PREDICTION RULE FOR SEVERE POSTOPERATIVE PAIN

Kristel J.M. Janssen, MSc1, C.J. Kalkman, PhD, MD1, K.G.M. Moons, PhD1, Diederick E. Grobbee, PhD, MD1, Gouke J. Bonsel, MD, PhD2, and Y. Vergouwe, PhD1. (1) University Medical Center Utrecht, Utrecht, Netherlands, (2) Academic Medical Center, Amsterdam, Netherlands

Purpose: To validate and update a multivariable prediction rule for the risk of severe postoperative pain in the first hour after surgery in outpatients. Methods: We used data from adult outpatients (n=549) participating in a randomized trial, which was designed to investigate the difference in incidence of postoperative nausea and vomiting between intravenous anaesthesia and inhalation anaesthesia. We used this data to study the validity of a prediction rule for severe postoperative pain within the first hour after surgery, defined as a numerical rating scale score Ан 8. The rule was developed in 1416 inpatients, participating in the same trial, and contained seven predictors: age, gender, preoperative pain, type of surgery, incision size, preoperative anxiety and need of information. Calibration (the agreement between predicted risks and observed frequencies) was studied with calibration plots and the Hosmer-Lemeshow test. Discrimination was studied with the area under the ROC curve (AUC). The rule was subsequently updated by combining the patient data of the validation set (outpatients) and the derivation set (inpatients). Interaction between surgical setting (inpatients/outpatients) and the other predictors was tested with the likelihood ratio test (significance level p < 0.30). Overoptimism was corrected with bootstrapping techniques. Results: The calibration of the rule was poor (p-value < 0.001 Hosmer-Lemeshow test) and the discriminative ability very low (AUC= 0.59). Therefore, we combined the datasets and included surgical setting in the updated rule. The likelihood ratio test showed statistically significant interaction. The AUC of the updated rule was 0.72 after correction for overoptimism. Discrimination was adequate for both inpatients and outpatients (AUC of 0.73 and 0.69 respectively). Conclusions: The prediction rule for severe postoperative pain showed poor predictive accuracy in outpatients. Interaction terms for surgical setting were required to allow for differences in strengths of predictors between inpatients and outpatients. If the updated rule proves to be robust in other hospitals, the rule may be helpful in identifying patients who will benefit from preemptive treatment.

See more of Poster Session III
See more of The 27th Annual Meeting of the Society for Medical Decision Making (October 21-24, 2005)