Glycated Proteomics in Peritoneal Dialysis

Evidence

This single-study explainer examines a two-phase investigation of glycated proteomics for cardiovascular disease prediction among patients receiving peritoneal dialysis. The first phase was a matched case-control biomarker-discovery analysis within an established cohort at Zhongshan Hospital, Fudan University. It used stored serum collected before cardiovascular disease occurred; cases and controls were selected retrospectively according to subsequent cardiovascular disease outcomes. The second phase prospectively developed and validated a prediction model in independent cohorts: a single-centre internal derivation cohort and a multicentre external validation cohort. The discovery cohort included 48 peritoneal dialysis patients with cardiovascular disease and 48 without cardiovascular disease; the prospective internal derivation cohort included 240 participants and the external validation cohort included 102. (pmid:42947327)

The study’s protein-focused measurement strategy was liquid chromatography-tandem mass spectrometry, used to identify glycated peptides in the discovery phase. The investigators used glmnet to prioritize candidate biomarkers, then quantified candidates in the prospective cohorts. For model construction, they applied F-test feature selection and elastic-net logistic regression. The final combined model retained two glycated peptide biomarkers—GEAFTLK(g)ATVLNYLPK and K(g)QHLFVK—alongside phosphate, diabetes, total Kt/V, albumin, cardiac troponin T, and glycated albumin. This design therefore evaluated a multivariable prediction model rather than the performance of either glycated peptide as an isolated test. (pmid:42947327)

In discovery, 257 glycated peptides were reported as differentially abundant between the cardiovascular disease and non-cardiovascular-disease groups. Four candidates were prioritized, with the two named peptides retained in the final model. Both retained glycated peptide biomarkers were reported to be significantly higher among patients who developed cardiovascular disease than among those who did not. Because the source describes the discovery comparison as a matched case-control analysis in which groups were retrospectively selected by later outcome, these measurements describe an observed association between peptide levels and subsequent cardiovascular disease status in that selected peritoneal dialysis cohort. (pmid:42947327)

The primary reported external comparison was between the glycated-proteomics-plus-clinical-indicator combined model and a clinical-only model. In the multicentre external validation cohort, the combined model’s area under the receiver operating characteristic curve was 0.822 (95% confidence interval 0.718–0.927), compared with 0.780 (95% confidence interval 0.666–0.894) for the clinical-only model. The source also reports that the combined model had higher area under the receiver operating characteristic curves than three published cardiovascular risk models, without providing their numerical estimates in the provided excerpt. These discrimination results quantify separation between participants with and without the outcome in the evaluated cohort; they do not independently establish clinical usefulness. (pmid:42947327)

For risk stratification, the investigators used a Youden cutoff of 0.373 derived from training data. Participants classified as high risk had higher 1-year cardiovascular disease incidence than low-risk participants in both validation cohorts. In the internal validation cohort, the reported relative risk for high versus low risk was 3.35 (95% confidence interval 1.19–9.42; p = .021); in the external validation cohort, it was 5.50 (95% confidence interval 2.12–14.27; p < .001). The high- and low-risk groups were defined by combined-model predicted probability at or above 0.373 and below 0.373, respectively. (pmid:42947327)

Analysis — Prediction performance and scope

The reported pattern is consistent with a staged observational prediction study: retrospective outcome-based selection was used to screen a broad glycated-peptide signal, and prospectively assembled derivation and validation cohorts were used to assess a model incorporating selected peptides and clinical indicators. The external area-under-the-curve comparison indicates that, in the multicentre validation cohort, the combined model discriminated cardiovascular disease status more strongly than the clinical-only comparator. The high-versus-low-risk relative risks similarly describe outcome separation after application of a cutoff chosen in training data. Neither result identifies whether the peptide markers themselves participate in cardiovascular disease biology, because the model combines them with six clinical variables and the design does not manipulate peptide exposure or treatment. The observed pattern is also shaped by the study’s selection process: discovery cases and controls were retrospectively defined by subsequent cardiovascular disease outcomes, whereas validation evaluated prediction in the stated cohorts. Confidence intervals around the external area under the receiver operating characteristic curve and relative risks convey uncertainty in those estimates. Statistical associations and discrimination are not clinical significance or evidence of patient benefit. (pmid:42947327)

Limitations

The retrospective design cannot establish causation. In particular, the discovery component selected cases and controls according to subsequent cardiovascular disease outcomes, so its peptide-group differences are susceptible to selection bias and cannot show that higher glycated peptide levels caused cardiovascular disease. The study measured glycated peptides in serum, clinical indicators, combined-model predicted probabilities, and cardiovascular disease outcomes; it did not test a biological mechanism for the retained peptides. The final model’s comparison with a clinical-only model is informative about discrimination within the external cohort, but it does not isolate the contribution of either peptide from the other model variables. (pmid:42947327)

The prospective phase included internal derivation and multicentre external validation, which broadens evaluation beyond the single-centre derivation setting, but all described participants were receiving peritoneal dialysis. Accordingly, the reported model performance and high-versus-low-risk incidence pattern are bounded to the studied peritoneal dialysis cohorts. The provided evidence reports confidence intervals for selected performance and stratification estimates, including a relatively wide interval around the external-validation relative risk, which indicates imprecision in that estimate. The available source material does not provide additional uncertainty details beyond the stated confidence intervals and validation results. (pmid:42947327)