Yale Cancer Center and Smilow Cancer Hospital, Yale School of Management, and Yale New Haven Health, New Haven, CT; and Massachusetts General Hospital, Boston, MA Corresponding author: Kerin Adelson, MD, Yale Cancer Center and Smilow Cancer Hospital, 20 York St, North Pavilion 15, Suite 3006, New Haven, CT 06510; e-mail: kerin.adelson@yale.edu. Disclosures provided by the authors are available with this article at jop.ascopubs.org. DOI: https://doi.org/10.1200/JOP. 2017.023200; published online ahead of print at jop.ascopubs.org on December 5, 2017. Development of Imminent Mortality Predictor for Advanced Cancer (IMPAC), a Tool to Predict Short-Term Mortality in Hospitalized Patients With Advanced Cancer Kerin Adelson, Donald K.K. Lee, Salimah Velji, Junchao Ma, Susan K. Lipka, Joan Rimar, Peter Longley, Teresita Vega, Javier Perez-Irizarry, Edieal Pinker, and Rogerio Lilenbaum QUESTION ASKED: End-of-life care for patients with advanced cancer is aggressive and costly. Oncologists inconsistently esti- mate life expectancy and address goals of care. Current available prognostication tools are based on subjective clinical assessment. We asked whether we could use objective data from the Electronic Health Record to develop the Imminent Mortality Predictor in Advanced Cancer (IMPAC), a tool that could predict short-term mortality in hospitalized patients with advanced cancer. If so, such a tool could be used by oncologists to guide end-of-life conversations. SUMMARY ANSWER: For mortality within 90 days at a 40% sensitivity level, IMPAC has close to 60% positive predictive value. Patients estimated to have a greater than 50% chance of death within 90 days had a median survival time of 47 days. Patients estimated to have a less than 50% chance of death had a median survival of 290 days (Fig). METHODS: Statistical learning techniques were applied to data from electronic health records (EHRs) for 669 patients with advanced cancer discharged from Yale Cancer Center/ Smilow Cancer Hospital to develop a tool that could estimate survival probabilities. To char- acterize the pattern of end-of-life care among this cohort, we examined the use of aggressive interventions within the last 30 days of life. For every visit in which IMPAC correctly identified the patient as likely to die within 90 days of admission, we calculated the po- tentially avoidable cost had the patient instead been cared for in hospice. BIAS, CONFOUNDING FACTOR(S), DRAWBACKS: The data used to generate the model were based on patterns of care at a single academic research institution in which patients and physicians could self-select for more ag- gressive care. IMPAC uses data from the Roth- man Index, a proprietary commercial product, thus limiting applicability at hospitals that do not purchase it. Our cost avoidance model is built on the assumption that patients flagged as likely to die would all receive only hospice care from 48 hours into a hospitalization onward and does not incorporate actual cost data from patients transitioned to hospice. REAL-LIFE IMPLICATIONS: We have devel- oped a novel prognostic tool, IMPAC, which uses objective data to generate life expectancy probabilities automatically from EHR data in real time. If it is integrated into the standard clinical workflow, IMPAC will signal oncol- ogists that goals-of-care conversations are imperative and will help facilitate prognostic understanding and informed decisions re- garding downstream health care interventions. Potentially avoidable costs are significant. 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Positive Predictive Value Sensitivity 30-day 60-day 90-day 180-day Fig. Average positive predictive value versus sensitivity for 30-, 60-, 90-, and 180-day mortality horizons. The average value is taken across the 20 test set splits. ReCAPs (Research Contributions Abbreviated for Print) provide a structured, one-page summary of each paper highlighting the main findings and significance of the work. The full version of the article is available online at jop.ascopubs.org. Copyright © 2017 by American Society of Clinical Oncology jop.ascopubs.org 1 Original Contribution FOCUS ON QUALITY Original Contribution FOCUS ON QUALITY Downloaded from ascopubs.org by YALE MEDICAL LIBRARY on December 6, 2017 from 130.132.173.019 Copyright © 2017 American Society of Clinical Oncology. All rights reserved.