External validation of the intermountain risk scores for post-discharge (PD) thrombosis and bleeding in an independent population

Abstract

Abstract Introduction: Extended pharmacologic thromboprophylaxis in medical patients has largely failed to demonstrate clinical benefit in most randomized controlled trials, primarily due to an underestimated risk of bleeding. Accurate prediction of both post-discharge (PD) bleeding and VTE is therefore critical to prevent thrombotic events while minimizing harm. The Intermountain risk scores (IMRSs) were developed to estimate 90-day risk of PD bleeding and thrombosis. Objective: To externally validate the IMRSs for PD VTE and bleeding in an independent population. Methods: The study population (validation cohort) consisted of adult patients ≥ 18 years old admitted for a medical hospitalization and discharged alive from October 2010 to December 2019. A single patient could contribute to multiple hospitalizations. Similar to the Intermountain derivation study, we excluded hospitalizations with present on admission or hospital-acquired VTE. We also excluded people with less than one year observation period leading up to their hospitalization or who resided outside the catchment area of the corresponding medical center. PD VTE and bleeding requiring re-hospitalization within 90 days were identified using previously validated computable phenotypes. Variables included in the VTE model were age, white blood cell count, red blood cell count (RBC), platelet count, red blood cell distribution width (RDW), mean platelet volume, serum glucose, serum sodium (Na) and blood urea nitrogen. In the bleeding model the variables included were age, RBC, RDW, Na and serum creatinine. The laboratory data were collected within 24 hours of discharge. We calculated the IMRSs in the validation cohort and assessed calibration by comparing the observed incidence of events in the validation and Intermountain populations across the published risk categories. Cox proportional hazard models were used to estimate hazard ratios (HRs) for risk factors in the validation cohort, and model discrimination was assessed via the area under the receiver operating characteristic curve (AUC). Multiple imputation was used to account for missing data. Results: Among 14,556 medical hospitalizations, we identified 108 PD VTE events and 373 PD bleeding events. The mean age of the cohort was 68 (SD=16) years, 55% were female and 95% were White race. Notably, the incidence of PD VTE in the derivation cohort was higher than in the validation cohort (19.51 vs 7.42 per 1000 hospitalization), while the incidence of PD bleeding was lower in the derivation cohort compared to the validation cohort (13.3 vs 25.6 per 1000 hospitalization). In the validation cohort, the observed risk of PD-VTE was 1.05% in the high-risk category (IMH-VTE ≥ 7), compared to 0.46% for the low risk, compared with the observed risks in the derivation cohort of 2.6% and 0.46% for high-risk and low risk scores, respectively. For high-risk bleeding (IMH-bleeding score of ≥8), we found an observed risk of bleeding of 4.36%, compared to 1.84% for low risk. This compares to the derivation cohort, in which an observed risk of bleeding of 1.9% in high risk versus 0.8% for low risk. Except for hyponatremia (defined as Na<135 mmol/l), none of the variables included in the IMH risk score were associated with a statistically significant increase in the risk of bleeding or VTE in the validation cohort (aHR (95% CI) for Na<135 mmol/l was 2.13 (1.14,3.97) for VTE and 2.29 (1.24,4.23) for bleeding). Finally, the AUC for discrimination was 0.55 for the VTE model and 0.60 for the bleeding model compared to 0.65 and 0.69 in the derivation cohort, respectively. Conclusion: External validation revealed poor discrimination and calibration of the IMRSs for predicting 90-day PD VTE and bleeding leading to rehospitalization in an independent cohort. Alternative strategies are needed to enhance risk assessment following medical hospitalization.