Letter to the Editor: Regarding the use of decision curve analysis in predicting long-term complications of liver cirrhosis

To the editor, Predicting long-term complications of liver cirrhosis is crucial for facilitating early interventions and improving patient outcomes. With great interest, we read the recent study by Guo et al,1 which reported a noninvasive metabolomic state prediction model based on a UK Biobank cohort of 64,005 participants with extensive baseline data and metabolomic profiles. The metabolomic state model exhibited better performance than routine risk scores, such as the aspartate aminotransferase to platelet ratio index (APRI) and fibrosis-4 index (FIB-4), in both the training and validation cohorts. In addition, the authors developed and validated a nomogram integrating the metabolic state model with other predictors, which demonstrated better performance compared to the independent metabolic state model, APRI, and FIB-4. Here, we express a concern regarding the use of decision curve analysis (DCA) in assessing the clinical usefulness of the nomogram. In Figure 3 of the original paper, the authors exhibited DCA results for nomogram, metabolomic state model, APRI, and FIB-4 in both training and validation cohorts. As depicted in Figure 1, DCA demonstrates the range of threshold probabilities within which clinical decisions based on the model yield greater net benefits.2,3 According to the DCA from the training cohort, although not as effective as the metabolomic state model, decision based on APRI and FIB-4 still provide higher net benefits than "intervention for all" or "intervention for none" within a certain threshold probability ranges. However, in the validation cohort randomly grouped together with the training cohort, neither APRI nor FIB-4 demonstrated improved net benefits and their decision curves nearly overlapped with those of "intervention for all" and "intervention for none." This inconsistency appears unreasonable for such a large cohort, especially since APRI and FIB-4 are recognized predictive biomarkers,4,5 and in Figure 2B of the original paper. APRI and FIB-4 demonstrate modest predictive accuracy in the validation cohort. This observation prompts us to inquire about the data analysis process, seeking to understand the underlying factors better.FIGURE 1: Schematic diagram of the decision curve of a predictive model. Abbreviation: DCA, decision curve analysis.This work represents a significant step toward leveraging metabolic biomarkers for predicting long-term outcomes in patients with liver cirrhosis. Therefore, we encourage the authors to further elucidate the application of DCA to enhance the strength of the evidence, facilitating the broader adoption of the predictive model.