Risk assessment model potency to detect patients most likely to benefit from thromboprophylaxis: An application of the TARGET-TP score.

Abstract

12116 Background: Interventional trials applying risk models for targeted-thromboprophylaxis (TP) for ambulatory cancer patients have previously excluded low risk patients, preventing quantification of residual risk and unmet need. We compare potency and pragmatic application of risk models, to guide routine clinical utilisation. Methods: TARGET-TP, a three arm phase 3 randomized trial of TP, classified ambulatory lung and gastrointestinal cancer patients into high or low thromboembolism (TE) risk groups using an algorithm derived from fibrinogen and d-dimer levels. High risk patients (randomized arms) received enoxaparin or no TP. Low risk patients were enrolled as an observation arm. Risk model potency was assessed by comparing cumulative TE incidence at 180 days between the two arms not receiving enoxaparin. In this analysis, we also compared other risk models using published risk thresholds (Khorana Score (KS), PROTECHT, CONKO, CATS/MICA) using associations of predicted TE risk with observed TE events (cause specific Cox proportional hazards regression), sensitivity and specificity. Results: Among 328 patients, 200 (61%) were classified high TE risk using the TARGET-TP algorithm. Without TP, TE incidence was 23% among high risk and 8% low risk patients – compared to 8% in high risk enoxaparin treated patients. There was notable cohort migration, with individual patients reclassified between high- and low-risk across other risk. Up to 75% of TARGET-TP high risk patients were classified low risk by other models, and would not be considered for TP, potentially exposing substantive residual TE risk (75% low risk by CATS/MICA, 61% KS, 60% CONKO, 32% PROTECHT). Up to 57% of low risk patients were high risk by other models, potentially exposing unnecessarily to TP (57% high risk by PROTECHT, 27% KS, 26% CONKO, 5% CATS/MICA). Among 228 patients in TARGET-TP trial non-intervention arms: TE incidence and comparative risk (hazard ratio, HR) for high versus low TE risk were: TARGET-TP (23% high vs. 8% low, HR 3.33 [95%CI 1.58-6.99]), KS (17% vs. 13%, HR 1. 50 [95%CI 0.74-3.02]), PROTECHT (16% vs. 12%, HR 1.50 [95%CI 0.69-3.05]), CONKO (18% vs. 13%, HR 1.54 [95%CI 0.76-3.09]), CATS/MICA (26% vs. 12%, HR 2.72 [95%CI 1.26-5.86]). Sensitivity and specificity respectively: TARGET-TP 70%/61%, KS 39%/68%, PROTECHT 70%/37%, CONKO 39%/69%, CATS/MICA 27%/87%. Conclusions: Application of TE risk models demonstrated some ineffectual and if utilised to define TP eligibility, 4/5 would exclude patient cohorts with TE rates exceeding 10%. TARGET-TP was the only model to achieve both high sensitivity and specificity. This simple pragmatic model considers only d-dimer and fibrinogen, can be applied without complex calculations or nomograms, in real-time for any patient. Clinical trial information: ACTRN12618000811202.