Challenges in Venous Thromboembolism Risk Stratification After Pancreatic and Liver Cancer Surgery Using Claims Data

Postoperative venous thromboembolism (VTE) is a major source of morbidity and mortality. Real-time risk stratification for VTE is important to guide measures for thromboprophylaxis. We sought to compare the discrimination of three VTE predictive systems in classifying 90-day postoperative VTE among cancer patients undergoing pancreas and liver surgery using administrative data.

We conducted a retrospective analysis of the 2001-2015 SEER-Medicare data for patients with newly diagnosed primary pancreatic or liver malignancies who underwent surgery. The primary outcome was 90-day postoperative VTE (including deep venous thrombosis and pulmonary embolism [PE]). We used the 2001-2010 dataset to create a new VTE predictive model using multivariable logistic regression. We then compared the discrimination of the Caprini score (adapted for claims data), the Elixhauser comorbidity score, and the new predictive model using the 2011-2015 dataset.

Our sample included 15 479 patients (12 005 pancreas and 3474 liver surgery). Median age was 74 years (interquartile range: 69-78) and 7619 were female (49%). The overall incidence of 90-day VTE was 10.0% (90-day PE: 4.5%). The new VTE model included age, race, history of VTE, cancer stage, preoperative chemotherapy, and Elixhauser score. The discrimination of all predictive models was modest for both VTE (c-statistic 63.0%, 59.4%, and 67.8%) and PE (61.5%, 59.7%, 65.6% for Caprini, Elixhauser, and new score respectively).

The existing tools for predicting postoperative VTE show poor performance using claims data. Future research should supplement administrative data with additional data sources to better predict patients at high risk for postoperative VTE.