Abstract

Background: High-sensitivity troponin, NT-proBNP and C-reactive protein are robustly and independently associated with cardiovascular events. Their addition to established risk models improves discrimination minimally, a result usually reported as disappointing and attributed to the limitations of the markers. Objectives: To collate reported associations alongside reported gains in discrimination for the same markers; to derive analytically what gain in the C-statistic a given hazard ratio can produce; and to determine whether the observed gains are smaller than the associations imply or exactly as large. Methods: Prospective cohort studies and syntheses reporting both an adjusted association and an incremental discrimination measure for a circulating cardiovascular biomarker were eligible. For a normally distributed marker added independently to an existing model, the resulting C-statistic was derived analytically. Predicted increments were computed from each reported hazard ratio and compared with the increments actually reported. Analyses were performed in Python 3. Results: Reported associations were strong: hazard ratios of 2.57 (95% CI 1.47-4.49) for high-sensitivity troponin I, 3.01 (1.66-5.48) for troponin T and 3.38 (2.04-5.60) for NT-proBNP comparing top with bottom quintile, and 1.39 to 1.64 per standard deviation in coronary artery disease. Reported discrimination gains were negligible: 0.002 for troponin I and 0.006 for C-reactive protein, with net reclassification indices of 0.011 (–0.064 to 0.086) and 0.024 (–0.042 to 0.091) respectively. Adding all three to SCORE2 moved the C-statistic from 0.81 to 0.82. Analytically, a marker added independently to a model with a C-statistic of 0.81 requires a hazard ratio per standard deviation of about 1.44 to raise it by 0.01, about 2.18 to raise it by 0.04 and about 3.74 to raise it by 0.09. The markers examined lie between 1.19 and 1.64, for which the predicted gains are 0.002 to 0.018—closely matching the 0.002 to 0.03 actually reported. Conclusions: The biomarkers are not underperforming. They are producing precisely the discrimination gain their effect sizes permit, and the apparent disappointment reflects an expectation that a strong association should translate into a large gain in the C-statistic. It cannot. A marker capable of moving the C-statistic by a clinically visible amount would need an independent hazard ratio far larger than any circulating cardiovascular biomarker has ever shown, and the relationship is calculable in advance of measuring anything.

Keywords: Cardiovascular risk prediction, Biomarkers, C-statistic, Discrimination, Net reclassification improvement, NT-proBNP