Artificial intelligence guided screening for cardiomyopathies in an obstetric population: a pragmatic randomized clinical trial

Cardiomyopathy is a significant cause of maternal mortality, particularly in postpartum period. Nigeria has the highest reported incidence of peripartum cardiomyopathy globally. Diagnosing cardiomyopathy during pregnancy and postpartum is challenging due to overlapping symptoms with physiological pregnancy changes. AI-enabled ECGs have shown promise in identifying various cardiovascular pathologies, including low LVEF. Previous studies demonstrated the effectiveness of AI-based screening in identifying pregnancy-related LVSD; however, its impact on improving cardiomyopathy detection beyond standard care remained unknown. This study aimed to evaluate whether AI-guided screening, using a digital stethoscope and 12-lead ECG, improves the diagnosis of pregnancy-related LVSD in a Nigerian obstetric population compared to usual care.

Studies in the US and Republic of Korea have shown good performance of AI-ECG models for detecting perinatal LVSD. A retrospective study showed an AUC of 0.89 and a pilot prospective study showed AUC of 1.00 using a 12-lead ECG and 0.98 using a digital stethoscope in identifying pregnancy-related LVSD with LVEF < 45%. These studies suggested the potential of AI to enhance LVSD screening; however, a rigorous clinical trial was needed to confirm this in a real-world setting, particularly in a high-incidence region like Nigeria.

The SPEC-AI Nigeria trial was an open-label, pragmatic, multicenter, randomized clinical trial. 1232 pregnant and postpartum women were randomized (1:1) to either AI-guided screening (intervention) or usual care (control). The intervention involved digital stethoscope recordings with point-of-care AI predictions and a 12-lead ECG with asynchronous AI predictions for LVSD. Participants in the intervention arm underwent a confirmatory echocardiogram at baseline for AI model validation. Data was collected at six hospitals in Nigeria between August 2022 and September 2023, with follow-up through May 2024. The primary endpoint was identification of LVSD during the study period, defined as the number of participants with LVSD confirmed by echocardiography. A modified intention-to-treat (mITT) analysis was performed, excluding participants who did not complete baseline assessments, withdrew, or died before baseline testing. Secondary outcomes included AI model performance in subgroups and the effectiveness of AI in identifying various LVEF thresholds. Statistical analyses were performed using logistic regression and receiver operating characteristic (ROC) curves.

A total of 1195 participants completed baseline assessments (587 intervention, 608 control). Using the AI-enabled digital stethoscope, 24/587 (4.1%) participants in the intervention arm were identified with LVSD compared to 12/608 (2.0%) in the control arm (odds ratio 2.12, 95% CI 1.05–4.27, P=0.032). With the 12-lead AI-ECG, 20/587 (3.4%) in the intervention arm were identified versus 12/608 (2.0%) in the control arm (odds ratio 1.75, 95% CI 0.85–3.62, P=0.125). Subgroup analyses showed consistent results for the digital stethoscope, but the 12-lead ECG showed varied performance across age groups. The digital stethoscope had an AUC of 0.976 (95% CI 0.953–0.998) for detecting LVEF <50%. There were no serious adverse events related to study participation. A full intention-to-treat analysis yielded similar results.

This study demonstrates that AI-guided screening using a digital stethoscope significantly improves the diagnosis of pregnancy-related cardiomyopathy in a high-risk population in Nigeria. The improved detection rate, along with the high prevalence of LVSD found in this population, highlights the need for widespread screening. The study's findings support the use of AI-powered digital tools to enhance cardiovascular care in resource-limited settings and reduce health disparities. The relatively low number needed to screen (NNS = 47) emphasizes the cost-effectiveness of this intervention compared to other obstetric screening modalities.

AI-guided screening with a digital stethoscope significantly increased the diagnosis of cardiomyopathy associated with LVSD compared to usual care. This intervention holds the potential to improve cardio-obstetric care by reducing delays in diagnosing a life-threatening condition. Further research should explore the impact of AI screening on treatment interventions, healthcare costs, utilization, and long-term outcomes.

The study's pragmatic design and enrollment at teaching hospitals with readily available cardiology expertise may limit generalizability. The high attrition rate, largely due to the study design and participant entry points, could have affected the results. Echocardiograms were not consistently performed in the control group, limiting the ability to fully assess LVSD prevalence in this arm. The selected cutoff for LVSD (LVEF <50%) did not precisely align with the original model training thresholds. The potential impact of out-of-pocket costs on obtaining echocardiograms in the control group was not assessed.