A partial signal, not a complete match
Electrocardiographic patterns associated with obesity and self-reported heart failure shared part of their structure, but they were far from identical, according to a cross-sectional analysis of US adults. The authors describe the result as statistically nonrandom alignment, not evidence that obesity reproduces the complete electrical signature associated with the heart-failure label.
How the comparison was built
To make that comparison, the researchers treated each condition as a pattern spread across 47 electrocardiographic features. They adjusted the feature space for age, sex and race or ethnicity, standardized the measurements and accounted for how the features varied together, using nonobese participants without heart failure as the reference. Each condition was then represented by a two-dimensional subspace, meaning the pair of leading mathematical directions used to summarize it, and the researchers compared the angles between those spaces.
The data came from the Third National Health and Nutrition Examination Survey, covering 1988-1994 and adults aged 40 years or older. Of 8,561 electrocardiographic records, 8,324 remained after the researchers excluded 17 records with unknown heart-failure status, 14 with unknown obesity status and 206 with incomplete core measurements.
Obesity meant a measured body mass index of at least 30 kg/m². Heart failure was based on a participant's report of a physician diagnosis, without independent medical-record adjudication, imaging or biomarker confirmation. The obesity subspace was derived from people with obesity but no heart failure, while the heart-failure subspace came from nonobese people with heart failure.
The cohort included 5,691 nonobese participants without heart failure in the reference group, 2,216 obese participants without heart failure, 287 nonobese participants with heart failure and 130 participants with both conditions.
A statistically unusual overlap
The main score for shared directions, mean squared canonical alignment, was 0.20485. That was higher than the average of 0.08756 in the permutation benchmark and above the benchmark's 99th percentile of 0.12036. In 5,000 shuffled analyses, none produced a value equal to or greater than the observed one, giving an empirical p value of 0.0002, which indicates the result was rare under that comparison.
The benchmark was built to respect the survey design. The researchers shuffled heart-failure labels within obesity-by-pseudo-stratum cells while keeping weights, strata, primary sampling units and split membership fixed, then repeated the full preprocessing and subspace calculation across 30 fixed half-sample assignments.
The finding was not equally strong in every split. Across 30 design splits, alignment ranged from 0.031 to 0.380. The mean minimum principal angle was 53.0 degrees, while the mean of the two angles was 67.5 degrees, figures that point to incomplete overlap between the two spaces.
Several sensitivity checks produced similar, though not identical, values. Alignment was 0.216 after clinical covariate adjustment and 0.219 when analyses were restricted to QRS duration below 120 milliseconds. It fell to 0.132 after excluding prior myocardial infarction and major conduction abnormalities, and to 0.149 after those strict exclusions were combined with clinical adjustment. The sensitivity results were treated as stability summaries, not independent hypothesis tests.
Why the result is only partial
A separate full-sample, descriptive analysis showed why the headline result is only partial. Its two principal angles were 31.96 and 81.81 degrees, with canonical cosines of 0.848 and 0.142 and mean squared alignment of 0.370. In ordinary terms, one direction was relatively close while the other was nearly at a right angle. The leading shared direction was dominated by T-wave amplitudes and the cosine of the frontal T-axis. Because this was a descriptive full-sample analysis, these figures were not an independent inferential test.
That distinction matters. The authors characterize the result as mathematical alignment between surface ECG signatures of survey-defined phenotypes. Because the analysis was cross-sectional, it cannot establish that obesity causes heart failure or causes the observed alignment, nor that obesity reproduces the complete electrocardiographic signature associated with the self-reported heart-failure label.
The study's heart-failure measure also limits what can be inferred. The label was self-reported and not independently adjudicated, while imaging, biomarkers, clinical records and heart-failure subtype information were unavailable. The result therefore remains a statistical comparison of surface ECG signatures in the survey-defined groups.
Paper data and sources
Original title: Partial Alignment of Electrocardiographic Subspaces Associated With Obesity and Heart Failure in the Third National Health and Nutrition Examination Survey
Authors: Dumitrescu P, Malik H, Cobas Hernandez J et al.
Journal/Repository: Cureus
Status: Peer-reviewed
First online: 2026-10-09
DOI: Not available
Original paper · Full text