Protein Model Shows Higher Cardiovascular Risk Discrimination

Evidence

A model combining two sugar-modified protein fragments with six clinical indicators showed higher measured ability to distinguish cardiovascular risk than clinical indicators alone in 102 patients receiving peritoneal dialysis, a form of kidney dialysis. Cardiovascular disease affects the heart and blood vessels; the protein fragments measured here are called glycated peptides.

In the peritoneal-dialysis prediction study, external-validation performance was 0.822 on the area under the receiver operating characteristic curve, versus 0.780 for the clinical-only model. This measure summarizes discrimination—how well a model separates people who develop an outcome from those who do not—with higher values indicating better separation. The respective 95% confidence intervals were 0.718–0.927 and 0.666–0.894, describing uncertainty around each estimate.

The observational study began with a matched discovery comparison of 48 patients with cardiovascular disease and 48 without it. Researchers analyzed stored serum collected before disease occurred using liquid chromatography–tandem mass spectrometry, a technique for separating and identifying molecules. They identified 257 glycated peptides with differing abundance, prioritized four candidates and retained two. This was followed by a prospective internal model-development cohort of 240 patients and a multicentre external validation cohort of 102.

At the training-derived risk-score cutoff of 0.373, external sensitivity—the proportion of cases correctly identified—was 70.6%, with a reported confidence interval of 44.0–89.7%. Specificity—the proportion of non-cases correctly identified—was 77.6%, with an interval of 67.3–86.0%. Patients classified as high risk had 5.50 times the one-year cardiovascular-disease risk of those classified as low risk in external validation (95% confidence interval 2.12–14.27). The corresponding internal-validation relative risk was 3.35 (1.19–9.42).

The separate veterinary prescribing study found topical-antiseptic use of 55.0% in dermatology consultations versus 45.5% in the comparison category. Topical antiseptics are substances applied to body surfaces to reduce microbes. The reported odds ratio was 1.47 (95% confidence interval 1.06–2.02), indicating higher prescribing odds in dermatology consultations—not treatment effectiveness.

That study examined retrospective 2022 records for dogs, cats and other small animals at one Colombian university veterinary hospital. Systematically selecting the first four prescriptions each day yielded 1,460 prescriptions, of which 731 involved nonantibiotic antimicrobials. Its cross-sectional design characterized prescribing patterns rather than testing treatment effects.

Analysis

Interpretation: The dialysis study offers stronger direct quantitative support for prediction because it tested performance in a multicentre external cohort and reported uncertainty. The veterinary result is not corroboration of the protein model. Their connection is methodological: both quantify observations in defined populations, without establishing causal treatment effects or benefit from acting on the findings.

External validation tests a model beyond its development cohort, but it does not establish broad applicability or show that model-guided care improves patient outcomes.

Limitations

The dialysis model’s higher discrimination estimate does not, by itself, establish a statistically reliable improvement over clinical indicators alone. The wide sensitivity and relative-risk intervals leave substantial uncertainty. Predictive separation also does not demonstrate benefit from using the score in practice.

The veterinary findings reflect one hospital and a sampling method limited to the first four daily prescriptions. No clinical-outcome assessment establishes safety, effectiveness or benefit from antiseptic use. The findings cannot directly establish implications for human skin disease.

Both studies were published on November 1, 2026. The veterinary records date to 2022; that collection period describes the data’s age, not the publication’s age.

What to watch

Further dialysis-model testing could measure both discrimination and whether predicted risks match observed one-year disease rates. Veterinary research would need broader sampling and animal outcomes to move beyond prescribing patterns.

Confidence that the protein measurements add predictive value would increase if a larger external cohort showed better discrimination than the clinical-only model, with a confidence interval for that difference excluding zero.