I read the study by Huang et al (1) with interest; their study provides valuable insights into the association between remnant cholesterol and the development of nonalcoholic fatty liver disease (NAFLD) in a prognostic setting. However, I believe that their conclusion regarding the predictive value of remnant cholesterol is unwarranted.

Huang et al reported statistically significant associations between remnant cholesterol and NAFLD development after adjusting for confounding factors. However, it is crucial to distinguish between statistical associations and predictive value. It is important to recognize that association testing, as conducted by Huang et al, aims to provide group-level inference rather than individual-level prediction. Without robust predictive analysis, it is difficult to ascertain whether remnant cholesterol can truly provide meaningful information for predicting the development of NAFLD. To assess the predictive/discriminative value of biomarkers, it is essential to use appropriate metrics. Instead of using metrics of association, such as odds or hazards ratios (2), one can evaluate discrimination in a prognostic context by reporting confusion matrix-derived metrics (eg, sensitivities, specificities, false-positive and false-negative rates), or by reporting related summary measures like C-statistics (3). When examining whether remnant cholesterol enhances the predictive ability for NAFLD compared to models that include other biomarkers, the improvement in discriminative ability can be assessed (3).

Huang et al did not perform such statistical tests, even though it would have been useful to evaluate the predictive performance of remnant cholesterol in individual-level prediction or its ability to improve conventional models that are used in the clinical prediction of NAFLD, especially as they have the ability to assess this is a prognostic setting while the majority of the published literature appears to be focused on diagnostic models developed and tested in cross-sectional studies. Without such prognostic assessments of predictive ability, the claim that remnant cholesterol is a predictor of the outcome remains unsupported.

The confusion between association and prediction is a recurring issue in the scientific literature, and many of us argue for a more rigorous terminology used to differentiate between association and prediction (2‐7). Biomarkers that exhibit statistically significant associations may not possess adequate predictive ability for the same outcome (4). Hence, it is imperative to exercise caution when interpreting study findings and drawing conclusions about the predictive value of biomarkers, especially with clinically interesting and important questions, such as the one posed by Huang et al.

As pointed out in a previous letter (8), I suggest that journal editors and statistical reviewers play a significant role in addressing this issue. By emphasizing the importance of rigorous predictive analysis and distinguishing between association and prediction in study conclusions, we can promote more accurate and responsible reporting of research findings.

I urge the scientific community to consider the implications of using inappropriate terminology and drawing unwarranted conclusions based on statistical associations. By doing so, we can prevent the dissemination of misleading information, avoid unnecessary clinical practices (7), and allocate resources more effectively to benefit researchers and patients.

I appreciate your attention to these concerns and hope that my comments will contribute to promoting better reporting practices in cardiometabolic research.