Cardiomyopathy is a leading cause of maternal mortality, particularly prevalent in Nigeria with a reported incidence significantly higher than global averages. Diagnosis is challenging due to overlapping symptoms with physiological pregnancy changes, leading to delayed diagnosis and adverse outcomes. AI-enabled ECGs have shown promise in detecting various cardiovascular pathologies, including low LVEF. Prior studies demonstrated the effectiveness of AI-based screening in identifying pregnancy-related LVSD; however, its impact in improving cardiomyopathy detection beyond standard care remained unclear. This trial 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.

Existing literature highlights the high incidence of peripartum cardiomyopathy in Nigeria and the challenges in its timely diagnosis. Studies have shown the effectiveness of AI-enabled ECGs in identifying various cardiovascular conditions, including low LVEF. Previous retrospective and prospective studies, primarily conducted in the US and South Korea, demonstrated promising results for AI-based screening of pregnancy-related LVSD using 12-lead ECGs and digital stethoscopes. However, a large-scale pragmatic trial in a diverse, low-resource setting like Nigeria was needed to assess the real-world impact of AI-guided screening.

The SPEC-AI Nigeria trial was an open-label, pragmatic, multicenter, randomized clinical trial conducted across six hospitals in Nigeria. 1232 pregnant and postpartum women were randomized (1:1) to either the AI-guided screening arm or the usual care arm. The intervention arm received digital stethoscope recordings and 12-lead ECGs with AI-based predictions for LVSD, along with a confirmatory echocardiogram at baseline. The control arm received standard 12-lead ECGs and usual care. The primary outcome was the identification of LVSD (LVEF < 50%) during the study period, confirmed by echocardiography. Secondary outcomes included AI model performance across subgroups, effectiveness at different LVEF thresholds, and composite adverse events. Modified intention-to-treat (mITT) and full intention-to-treat (ITT) analyses were conducted. Two AI algorithms for 12-lead ECGs (one US FDA-cleared, one original Mayo Clinic model) and one for the digital stethoscope were used.

A total of 1195 participants completed the baseline visit. Using the AI-enabled digital stethoscope, the intervention arm identified significantly more cases of LVSD (4.1%) than the control arm (2.0%) (odds ratio 2.12, 95% CI 1.05–4.27, P=0.032). The 12-lead AI-ECG (US FDA-cleared) showed a similar trend (3.4% vs 2.0%), but the difference was not statistically significant (P=0.125). Subgroup analyses generally showed consistent results, with the exception of age group for the 12-lead AI-ECG. The digital stethoscope demonstrated high AUC values (0.976 for LVEF <50%, 0.985 for LVEF <40%) for LVSD detection. There was no statistically significant difference in composite adverse outcomes or cardiovascular events between the two groups. All-cause mortality was numerically higher in the intervention arm, but this was not statistically significant after adjusting for site.

The study demonstrates that AI-guided screening with a digital stethoscope significantly improves the diagnosis of pregnancy-related cardiomyopathy in a resource-limited setting. The high prevalence of LVSD detected highlights the need for widespread screening. The AI-stethoscope's portability, ease of use, and real-time results offer significant advantages in low-resource settings. Although the 12-lead AI-ECG did not reach statistical significance, it still showed a positive trend. The higher all-cause mortality in the intervention arm warrants further investigation but might be related to increased healthcare contact and potential biases in mortality ascertainment.

AI-guided screening with a digital stethoscope significantly enhances the diagnosis of peripartum cardiomyopathy in Nigeria. This low-cost, readily deployable technology holds the potential to reduce maternal mortality. Further research should focus on optimizing AI algorithms, evaluating long-term outcomes, and exploring the cost-effectiveness of AI-guided screening in diverse settings.

The study's pragmatic design and recruitment from tertiary care centers may limit the generalizability of findings. The high attrition rate, especially after the baseline visit, impacts the follow-up data. The definition of LVSD (LVEF < 50%) did not perfectly align with the training thresholds of all AI models, potentially affecting their performance. Mortality ascertainment may have been prone to error due to the lack of a national death registry in Nigeria.