Key result
Machine learning reclassifies ~56% of unexplained out-of-hospital cardiac arrest causes using longitudinal health data.
Why the study?
Does a semi-supervised machine learning framework using longitudinal health data improve the classification of unknown etiologies in adult out-of-hospital cardiac arrest?
Observational (n=22,170)
Yes
Does a semi-supervised machine learning framework using longitudinal health data improve the classification of unknown etiologies in adult out-of-hospital cardiac arrest?
Answers are grounded in this paper and cite their sources.
A semi-supervised machine learning model using longitudinal healthcare data can probabilistically reclassify over half of out-of-hospital cardiac arrests initially labeled as having an unknown etiology.
Background: Out-of-hospital cardiac arrest (OHCA) remains a major public health issue. However, most cases are classified as having “unknown” etiologies, because the majority of patients die on scene after unsuccessful or not initiated resuscitation. This limitation is compounded by the low autopsy rate in France, can make it difficult both family screening and epidemiological understanding. We developed a semi-supervised machine learning framework to retrospectively infer the most likely cause of OHCA from patients’ longitudinal healthcare histories. Methods: We included 22170 adults OHCA cases recorded in the Sudden Death Expertise Center registry between 2011 and 2020, of whom 18817 (84.9 %) had an unknown etiology at the time of inclusion. Using the French National Health Insurance Database (SNDS), we extracted all hospital discharge diagnoses, outpatient drug dispensations, and reimbursed diagnostic or therapeutic procedures over the five years preceding the OHCA. Leveraging patients with known causes, we trained two models within an expectation-maximization (EM) semi-supervised framework: (i) an observability model predicting whether a case was likely to have a known etiology, and (ii) a cause model assigning one of several medical categories (cardiac ischemic, cardiac rhythmic, cardiac other, pulmonary, neurologic, embolic, other). Models were validated using a temporal split. Results: With a posterior probability threshold of 0.5, overall accuracy reached 64.0%. Applying the algorithm to previously unknown cases enabled the reclassification of 10497 patients (55.8% of those initially unknown). After this “numerical autopsy,” the distribution of etiologies shifted from: cardiac ischemic (7.5% to 45.1%), pulmonary (2.6% to 10.0%), cardiac other (1.8% to 3.3%), neurologic (1.1% to 1.5%), other (1.0% to 1.1%), embolic (0.7% to 1.1%), and cardiac rhythmic (0.4% to 0.4%), while unknown cases decreased from 84.9% to 37.5%. Model performance was moderate yet clinically informative (AUC = 0.70, log-loss = 1.20, F1-score = 0.23), reflecting meaningful discrimination despite limited number of cases with known etiology. Conclusions: Nearly half of OHCA cases initially labeled as “unknown” can be probabilistically reclassified using routinely collected longitudinal health data. These “numerical autopsies” could inform family counseling and screening strategies, and strengthen population-level surveillance.
No takes yet. Share an insight, caveat, or question.
Cezard et al. (2026) conducted an observational in Out-of-hospital cardiac arrest (OHCA) (n=22,170). Semi-supervised machine learning framework was evaluated on Overall accuracy of the cause model. A semi-supervised machine learning framework reclassified 55.8% of out-of-hospital cardiac arrest cases with initially unknown etiologies, achieving an overall accuracy of 64.0%.
Synapse has enriched 10 related papers. Consider them for comparative context: