Mortality Risk Variance Explained by High Sensitivity Troponin I Exceeds Putative COVID-19 Severity Biomarkers in Heart Failure Patients Hospitalized with Acute SARS-CoV-2 Infection

Abstract

Background: Accumulating evidence suggests hospitalized patients with heart failure (HF) are especially susceptible to COVID-19-related complications and mortality approaches 30%. We hypothesized that Boosted Tree (BT) mortality risk variance explained (VE%) by cardiac injury and putative COVID-19 thrombo-inflammatory biomarkers would be different between survivors and non-survivors in HF patients undergoing index hospitalization for SARS-CoV-2. Methods: Presentation demographics, anthropometrics, laboratory results (within 8h), and ICD-10 codes were collected. Discrete data summarized as proportions were compared with chi-squared test. Continuous data summarized with median [IQR] were compared using Kruskal-Wallis test. COVID-19 therapy across 6-pandemic surges, demographics, and comorbidities were statistically balanced. Univariate logistic regression tested association (p<.05) of COVID-19 clinical traits putatively linked with mortality risk and provided a receiver operating characteristic (ROC) curve while computing Youden’s cut point. BT algorithm optimized area under ROC curve (AUROC) representing prediction accuracy while providing VE% contributed by each cut point variable. Results: Among 5,841 consecutive patients discharged from index COVID-19 hospitalization between March 18, 2020, and April 30, 2022, 1039 harbored extant HF. Those with at least 1 high sensitivity troponin (hs-cTnI) assay, 483, were studied. Hospital mortality (n=62, 13%) and survival (n=421, 87%) patient demographics were statically similar with 63% males and 37% females aged 75[64-83] years distributed across Whites (82%), Blacks (11%), and other races (7%). Subsequent results are similarly sequenced. Most prevalent comorbidities were hypertension (81% vs. 92%, p=.007), diabetes (48% vs. 42%, p=.41), obesity (50% vs. 34%, p<.02), renal failure (34% vs. 38%, p=.58), iron deficiency anemia (35% vs. 35%, p=1.0), and chronic coagulation disorder (39% vs 17%, p=.0002). Biomarker cut points were hs-cTnI ≥26 ng/L (71% vs. 53%, p=.006), CRP ≥6.0 mg/dL (63% vs. 22%, p<.0001), d-dimer ≥2.98 ug/mL (45% vs. 26%, p=.015), LDH ≥444 U/L (61% vs. 20%, p<.0001), and ferritin ≥668 ng/mL (68% vs. 26%, p<.0001). BT model demonstrated 38% R2, AUROC=0.86 with VE% contributed by hs-cTnI (41%), CRP (27%), d-dimer (14%), LDH (10%), and ferritin (8%). Conclusion: Our training model demonstrates a machine learning strategy to integrate hs-cTnI with putative biomarkers of COVID-19 severity to enhance understanding of host response contributing to mortality risk at presentation in hospitalized HF patients.