Effect of patient location on the performance of clinical models to predict pulmonary embolism.

Introduction: Current clinical likelihood models for predicting pulmonary embolism (PE) are used to categorize outpatients into low, intermediate and high clinical pre-test likelihood of PE. Since these clinical prediction rules were developed using outpatients it is not known if they can be applied universally to both inpatients and outpatients with suspected PE. Thus, the purpose of this study was to determine the effect of patient location on the performance of clinical models to predict PE.

Materials And Methods: Two clinical models (Wells and Wicki) were applied to data from the multi-centered PIOPED study. The Wells score was applied to 1359 patients and the Wicki score was applied to 998 patients. 361 patients (27%) from the PIOPED study did not have arterial gas measurement and were excluded from the Wicki score patient group. Patients were stratified by their location at the time of entry into the PIOPED study as follows: outpatient/emergency, surgical ward, medicine/coronary care unit or intensive care unit. The diagnostic performance of the two clinical models was applied to the various patient locations and the performance was evaluated using the area under a fitted receiver operating characteristic curve (AUC).

Results: The prevalence of PE in the three clinical probability categories were similar for the two scoring methods. Both clinical models yielded the lowest diagnostic performance in patients referred from surgical wards. The AUC for both clinical prediction rules decreased significantly when applied to inpatients in comparison to outpatients.

Conclusions: Current clinical prediction rules for determining the pre-test likelihood of PE yielded different diagnostic performances depending upon patient location. The performance of the clinical prediction rules decreased significantly when applied to inpatients. In particular, the rules performed least well when applied to patients referred from surgical wards suggesting these rules should not be used in this patient group. As expected the clinical prediction rules performed best in outpatients with the optimum diagnostic performance in patients referred from emergency and outpatient wards.