Heart Failure Protein Panels

Evidence

Two independent studies examined whether measuring many plasma proteins can sharpen specific heart-failure questions. One tested whether previously developed protein risk scores travel across study designs when predicting mortality in people with established heart failure. The other searched for a panel that could distinguish symptomatic heart failure with preserved ejection fraction (HFpEF) from people without heart failure. Their shared signal is incremental information beyond standard clinical measures, not a single universal heart-failure signature.

The mortality study evaluated three published scores built with the SomaScan aptamer platform. The scores had originally come from a community cohort, a clinical trial, and a registry, and contained between 8 and 57 available protein targets. Investigators recalculated all three in a community-based cohort from southeastern Minnesota. The analysis included 1,351 adults with heart failure, whose diagnoses had been validated with Framingham criteria. It tested associations with all-cause mortality and compared five-year prediction with established measures, including the MAGGIC clinical score and NT-proBNP.

Applied to the same cohort, the three protein scores were moderately correlated with one another, with Pearson correlations from 0.59 to 0.76. Each one-standard-deviation increase was associated with higher all-cause mortality: the unadjusted hazard ratios were 2.70 for the community-derived score, 1.76 for the trial-derived score, and 1.70 for the registry-derived score. After adjustment for MAGGIC and NT-proBNP, the respective hazard ratios were 2.40, 1.40, and 1.46. Results were similar across reduced- and preserved-ejection-fraction groups. Seven protein targets appeared in at least two scores, but renin was the only target in all three. Thus, prediction transferred more consistently than the exact protein lists.

The HFpEF study addressed diagnosis rather than prognosis. Researchers performed liquid-chromatography mass-spectrometry plasma proteomics in 249 people with symptomatic HFpEF from BIOSTAT-CHF and 99 controls without heart failure from PREVEND. HFpEF required a left-ventricular ejection fraction of at least 50 percent. Machine learning ranked differentially abundant proteins, and candidate measurements were technically checked with the SomaScan 7K assay. The study found 211 differentially abundant proteins and highlighted two downregulated candidates: tropomyosin-4, or TMP4, and the 14-3-3 zeta protein YWHAZ.

In the reported combined biomarker model, specificity increased from 73 percent for NT-proBNP to 91 percent while sensitivity was 97 percent. The area under the receiver-operating-characteristic curve increased from 0.96 to 0.99, with a reported p value of 0.036. These estimates show separation within the analyzed case-control samples. They do not establish performance in an unselected diagnostic population, and the authors explicitly stated that external validation is required before clinical implementation.

Analysis — Clinical Task Shapes the Panel

The cross-study pattern is that protein panels appear useful only after the clinical task is defined. This is analysis, not a demonstrated universal rule. In the mortality study, three substantially different protein lists retained prognostic information when moved into one community cohort; their moderate correlations and limited target overlap suggest that several molecular combinations can track the same broad outcome. In the HFpEF study, the target was not future mortality but present-day separation of symptomatic cases from non-heart-failure controls, so a different assay workflow and different proteins emerged. These findings complement rather than replicate each other. They support an emerging model in which proteomics adds a task-specific layer to conventional measures, while the endpoint and sampling frame determine what the layer contains. The important convergence is performance beyond MAGGIC or NT-proBNP within each study, not agreement on one biological signature. A decisive next step would lock each panel and threshold before testing it prospectively in external, clinically representative populations, with attention to calibration, subgroup performance, and whether added classification changes outcomes rather than only model statistics.

Limitations

Both studies were observational prediction analyses, so associations between protein scores and outcomes do not show that the measured proteins cause heart failure, death, or diagnostic differences. Neither study tested whether using a panel to guide care improves patient outcomes. In the mortality study, 855 of 1,351 participants overlapped the development sample for the community-derived score, creating possible optimism even though a validation-subset analysis showed only minor attenuation. Two source populations were predominantly White, and the cohorts predated widespread use of newer heart-failure therapies, limiting transportability to current, more diverse populations.

The HFpEF evidence available for this briefing was an abstract, which constrains interpretation to reported methods and results. The case and control groups came from different cohorts, so differences in recruitment, sample handling, comorbidity, or care setting could contribute to apparent protein separation. Technical checking on another assay is not the same as external clinical validation. The abstract does not provide enough detail to assess all missing-data procedures, model tuning, or subgroup calibration. Finally, high sensitivity, specificity, or area under the curve in selected samples does not by itself establish usefulness in routine practice, where disease prevalence and competing diagnoses differ.