DERIVATION AND VALIDATION OF THE HA-VTE AND HA-MB INTERMOUNTAIN RISK SCORES FROM UBIQUITOUS CLINICAL BIOMARKERS TO PREDICT 90-DAY HOSPITAL-ASSOCIATED VENOUS THROMBOEMBOLISM AND MAJOR BLEEDING AMONG MEDICAL PATIENTS

Abstract

SESSION TITLE: Late-breaking Abstracts SESSION TYPE: Original Investigations PRESENTED ON: October 18-21, 2020 PURPOSE: Identify hospitalized medical patients who at discharge have significant risk for hospital-associated venous thromboembolism (HA-VTE) or major bleeding (HA-MB) to inform extended-duration thromboprophylaxis (EDT) decision-making. METHODS: The Intermountain risk score (IMRS) is a highly predictive mortality risk estimation tool derived from components of the complete blood count (CBC) and basic metabolic panel (BMP). While minimally predictive for the outcome of 90-day HA-VTE [Snyder,L 2020] we hypothesized that IMRS components from the CBC and BMP could be re-weighted to be more predictive of HA-VTE and major bleeding (“HA-VTE IMRS” & “HA-MB IMRS”). To do so, a dataset of 45,669 medical patients who survived hospitalization in which 1038 (1.6%) VTE and 611 (1.3%) major bleeding events were recorded (as determined using validated natural language processing in EMR interrogation [Evans,RS 2010]) was split into a derivation (2/3) and validation (1/3) cohort. In the derivation set the CBC and BMP candidate variables were split into quintiles, and the referent quintile was determined as that with the lowest event rate. Backward stepwise regression in 500 bootstrapped samples was performed to identify predictors from the BMP and CBC candidate variables. A Cox model was fit in the derivation dataset with the variables included in 80%, 70%, 60%, and 50% of bootstrapped samples; these models were compared using likelihood ratio tests to determine if the additional variables included at each threshold significantly improved model fit. Coefficients from the Cox model were used to assign weights for the quintiles to generate the HA-VTE IMRS and HA-MB IMRS cutoff thresholds. Covariates selected for the final VTE model were: RDW, BUN, age, glucose, WBC, platelet count, RBC and sodium. Covariates selected for the final major bleeding model were: RDW, age, creatinine, MPV, Sodium, and RBC. RESULTS: In the derivation set 563/30,445 patients experienced HA-VTE and 399/30,445 experienced HA-MB. The HA-VTE IMRS ≥ 7 generated an AUC=0.603 and the HA-MB IMRS ≥ 8 generated an AUC=0.624. In the validation set 137/5242 had VTE in the HA-VTE IMRS ≥ 7 cohort vs. 160/9982 with a HA-VTE IMRS < 7 (2.6% vs 1.6%; HR=1.69; 95% CI 1.35-2.13; AUC=0.60). 135/7101 had MB in the HA-MB IMRS ≥ 8 cohort vs. 67/8123 with a HA-MB IMRS < 8 (1.9% vs. 0.8%; HR=2.35; 95% CI 1.75-3.16; AUC=0.64). CONCLUSIONS: HA-VTE and HA-MB risk may be estimated electronically at the time of hospital discharge based on components of the CBC and BMP. This information may inform decision-making regarding the prescription of EDT. CLINICAL IMPLICATIONS: If validated, this electronic program may be deployed in an EMR to calculate 90-day HA-VTE risk at the time of discharge with no additional time or expense based on laboratory values that are ubiquitously available and could inform decision-making regarding extended duration thromboprophylaxis for medical patients. DISCLOSURES: Consultant relationship with BD Please note: $5001 - $20000 Added 07/17/2020 by Joseph Bledsoe, source=Web Response, value=Consulting fee No relevant relationships by masarret fazili, source=Web Response PI of research grant relationship with AstraZeneca Please note: $20001 - $100000 Added 07/17/2020 by Benjamin Horne, source=Web Response, value=Grant/Research Support PI of research grant relationship with GlaxoSmithKline Please note: $1-$1000 Added 07/17/2020 by Benjamin Horne, source=Web Response, value=Grant/Research Support PI of research grant relationship with CareCentra Please note: $1-$1000 Added 07/17/2020 by Benjamin Horne, source=Web Response, value=Grant/Research Support Owner of intellectual property relationship with Alluceo Please note: $1-$1000 Added 07/17/2020 by Benjamin Horne, source=Web Response, value=IP was licensed Owner of intellectual property relationship with CareCentra Please note: $1-$1000 Added 07/17/2020 by Benjamin Horne, source=Web Response, value=IP was licensed No relevant relationships by James Lloyd, source=Web Response No relevant relationships by Gregory Snow, source=Web Response Research contract (no longer active) relationship with BMS/Pfizer Please note: $1-$1000 Added 07/17/2020 by Scott Stevens, source=Web Response, value=Grant/Research Support No relevant relationships by Scott Woller, source=Web Response