Derivation and validation of the In-hospital Mortality for PulmonAry embolism using Claims daTa (IMPACT) prediction rule

Abstract

Introduction:Existing prediction rules for prospectively prognosticating early mortality following pulmonary embolism (PE) require clinical and/or laboratory data, and are rarely suitable for claims database analyses. We sought to develop a claims-based prediction rule that retrospectively classifies PE patients into low- or higher-risk in-hospital mortality categories.Materials and methods:We randomly assigned MarketScan database patient admitted for PE between April 2010 and September 2013 into derivation (80%) and validation (20%) cohorts. A prediction rule (In-hospital Mortality for PulmonAry embolism using Claims daTa or IMPACT) was derived using multivariable logistic regression, with in-hospital mortality as the dependent variable and demographic/comorbidity data available in claims databases as independent variables. In-hospital mortality rates for low- and higher-risk patients were compared across the derivation and validation cohorts, and prediction rule performance was assessed by evaluating sensitivity and specificity estimates.Results:A total of 27,833 patients admitted for PE were included. The IMPACT rule consisted of 12 risk factors, and categorized 46% of patients as low-risk in both cohorts. Patients classified as low-risk by IMPACT (defined as an estimated in-hospital mortality risk ≤1.5%) had average in-hospital mortality rates of 0.81% (95% confidence interval [CI], 0.65–1.00) in the derivation and 0.77% (95% CI, 0.50–1.18) in the validation cohort. Higher-risk patients had average in-hospital mortality rates of 4.61% (95% CI, 4.25–5.01) and 5.02% (95% CI, 4.30–5.85), respectively. The IMPACT rule had high sensitivity for classifying in-hospital mortality risk (0.87 in both cohorts), but moderate specificity (0.47 for both cohorts).Limitations:We were unable to assess 30 day mortality as an endpoint. IMPACT was only validated in an internal sample.Conclusions:The IMPACT prediction rule may be able to retrospectively classify PE patients’ in-hospital mortality risk with high sensitivity and moderate specificity.