Abstract Body (Do not enter title and authors here): Introduction: More than 10 million 911 calls are made for non-traumatic CP each year, yet only 15% lead to a confirmed cardiac diagnosis. Identifying cases of cardiac chest pain (CP) during a 911 call remains a challenge. Recent advances in acoustic engineering make it possible to analyze voice features of the caller, which may reveal signs of underlying cardiac problems. This study explored whether voice-based acoustic biomarkers differ between 911 callers who did and did not have a confirmed cardiac diagnosis.

Methods: This retrospective analysis used 911 calls collected from Orange County, NC paired with clinical and outcome data. Calls was manually reviewed to determine the speaker, transcribed, pre-processed, and segmented. Predefined static acoustic features, including words per minute (WPM) and harmonics-to-noise ratio (HNR). The endpoint was cardiac diagnosis. We used non-patient callers (e.g. family) as a healthy control. Group comparisons were conducted using the Kruskal-Wallis test with post hoc pairwise Mann–Whitney U tests, and subgroup analyses were stratified by gender and age. Logistic regression was used to examine associations between acoustic biomarkers and cardiac diagnosis, adjusting for gender and age; all analyses were performed in Python 3.11 with p<0.1 considered significant.

Results: Of 47 911 callers in the database, 16 were patients themselves (mean age 61.3 years; 37.5% male) of whom 31.3% (n=5) reported chest pain and 31.3% (n=5) had a cardiac diagnosis. WPM and HNR differed between patients with and without a cardiac diagnosis and healthy controls (WPM: H=17.14, p < 0.01; HNR: H = 6.45, p = 0.04). Pairwise comparisons revealed that WPM was lower in cardiac patients (median 41.3) compared to non-patient callers (108.4, p < 0.01) and to patients without a cardiac diagnosis (67.5, p = 0.05). HNR was reduced in cardiac patients compared to those without a diagnosis (8.1 vs. 11.2, p=0.01), and significantly different from cardiac patients and non-patient callers (8.1 vs. 9.9, p=0.09). Female patients spoke significantly faster (73.94 vs. 39.25; p=0.01). HNR was an independent predictor of cardiac diagnosis (OR = 0.67, 95% CI: 0.45–0.96, p = 0.03).

Conclusion: Acoustic biomarkers differed significantly between non-traumatic CP callers with and without confirmed cardiac diagnoses. These results support the potential of integrating voice-based analytics into 911 triage workflows for emergency cardiac care.