Key result
Autonomous code-writing agents boost AI-ECG discrimination for structural heart disease by up to ~0.04 AUROC.
Why the study?
Clinical AI models reflect limited human search across many possible architectures and training variations, and autonomous agents may enable controlled, auditable model development without new data or human edits.
Does optimization by autonomous code-writing agents improve the performance (AUROC) of validated AI-ECG models for structural heart disease?
Does optimization by autonomous code-writing agents improve the performance (AUROC) of validated AI-ECG models for structural heart disease?
Autonomous code-writing agents can substantively rewrite and optimize previously validated AI-ECG models to improve discrimination for structural heart disease without requiring new data or human guidance.
Clinical artificial intelligence (AI) models are usually reported as finished artifacts, but each model reflects a limited human search across a much larger space of architectures, inputs, losses, optimizers, and training recipes. We tested whether autonomous code-writing agents could perform a controlled model-development experiment: proposing and evaluating code changes, and seeking performance gains without new data or human-guided edits. We built two such agents: an Iteration Agent that searches sequentially, keeping the best variant at each step, and an Evolution Agent that searches for variations in parallel using multiple large language models and prioritizes high-performing lineages across generations. In two architecturally distinct AI-enhanced electrocardiography (AI-ECG) models for structural heart disease, agent-optimized variants improved rank discrimination across held-out, external, and cross-institution evaluations, with area under the receiver operating characteristic curve gains of +0.006 to +0.039 (paired p < 0.05). At a fixed 90% sensitivity, specificity rose by up to 7.1 percentage points and positive predictive value by up to 4.8 percentage points. The selected code changes were substantive, spanning architecture, representation, and training recipe variations. These findings position autonomous agents as an auditable layer for clinical AI model improvement, provided that candidate selection, external validation, and post-update governance are explicit. We release these agents as an open, reusable toolkit.
No takes yet. Share an insight, caveat, or question.
Dhingra et al. (2026) studied Structural heart disease. Autonomous code-writing agents (Iteration and Evolution Agents) vs. Baseline (human-developed) AI-ECG models was evaluated on Rank discrimination (AUROC) for structural heart disease across held-out, external, and cross-institution datasets (AUROC gains of +0.006 to +0.039, p=<0.05). Autonomous code-writing agents improved the rank discrimination of two independently developed AI-ECG models for structural heart disease, yielding AUROC gains of +0.006 to +0.039 across held-out and external evaluations without additional data.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: