Many studies have shown the utility and promise of artificial intelligence (AI), for the diagnosis of left ventricular hypertrophy (LVH). The aim of the present study was to conduct a meta-analysis to compare the accuracy of AI tools to electrocardiographic criteria, including Sokolow–Lyon and the Cornell, most commonly used for the detection of LVH in clinical practice.
Nine studies meeting the inclusion criteria were selected, comprising a sample size of 31 657 patients in the testing and 100 271 in the training datasets. Meta-analysis was performed using a hierarchal model, calculating the pooled sensitivity, specificity, accuracy, along with the 95% confidence intervals (95% CIs). To ensure that the results were not skewed by one particular study, a sensitivity analysis using the ‘leave-out-one approach’ was adopted for all three outcomes.
AI was associated with greater pooled estimates; accuracy, 80.50 (95% CI: 80.4–80.60), sensitivity, 89.29 (95% CI: 89.25–89.33) and specificity, 93.32 (95% CI: 93.26–93.38). Adjusting for weightage of individual studies on the outcomes, the results showed that while accuracy and specificity were unchanged, the adjusted pooled sensitivity was 53.16 (95% CI: 52.92–53.40).
AI demonstrates higher diagnostic accuracy and sensitivity compared with conventional ECG criteria for LVH detection. AI holds promise as a reliable and efficient tool for the accurate detection of LVH in diverse populations. Further studies are needed to test AI models in hypertensive populations, particularly in low resource settings.