Biomarker-Defined Phenotyping Reveals Patient Heterogeneity Not Captured by a Single Physiological Reserve Score

The study suggests that a two-class biomarker-defined phenotype structure may better reflect patient heterogeneity than a single physiological reserve score.

Summary and findings

This study examined patient heterogeneity in routine biomarkers among cohorts treated with dapagliflozin and/or semaglutide. It found that a one-dimensional physiological reserve score was not supported, leading to a two-class solution based on biomarkers. The findings suggest that these biomarkers may better capture patient heterogeneity than a single score.

Abstract

Background: Patient heterogeneity in routine biomarkers is often collapsed into a single physiological reserve score, but whether that one-dimensional representation holds across acute and chronic settings — or whether it obscures biomarker-defined subgroups relevant to individualized care — is unclear. Methods: We harmonized routine biomarkers (BMI, blood pressure, eGFR, LDL and total cholesterol, HbA1c) across two cohorts from the same health system: an acute cardiorenal syndrome cohort and a chronic outpatient cohort treated with dapagliflozin and/or semaglutide, excluding 15 patients common to both. We tested whether these biomarkers supported a one-factor structure via exploratory factor analysis, then a categorical alternative via Gaussian mixture modeling, assessed six-month class persistence in the chronic cohort, and applied the derived classes to the acute cohort for descriptive external validation. Results: A one-factor representation was not supported (KMO = 0.485; covariance concentrated in two definitionally related variable pairs). Gaussian mixture modeling favored a two-class solution, distinguished mainly by LDL, total cholesterol, HbA1c, and diastolic blood pressure. Class assignment showed moderate six-month agreement (κ = 0.46) in a subgroup enriched for closer monitoring. In the acute cohort, class structure was not clearly associated with age, BNP, ejection fraction, length of stay, CKD stage, HF phenotype, NYHA class, or comorbidities; associations with outcomes were descriptive, not causal. Conclusions: Rather than reducing to a single reserve score, these biomarkers were better described by an exploratory two-class, biomarker-defined phenotype structure that captured meaningful patient heterogeneity, showed moderate short-term persistence, and was not clearly related to acute illness severity. These findings are specific to the biomarkers and cohorts studied and should not be generalized without independent replication.

Limitations

• Not reported in abstract.

Elsewhere in the Semaglutide corpus