Age Specific Analysis on Multiclass Sequential Curated-Electronic Health Records (MSC-EHR) for CAD Survival Prediction using Deep Learning Techniques

Early risk assessment is essential for addressing cardiovascular disease, a major healthcare issue. Accurate diagnosis is essential for prompt medical care and medication. Deep learning (DL) approaches have yielded promising outcomes in the detection of coronary artery disease (CAD). Previous work [Waqar et al. Sci Programm. 2021;2021:1–12, Muntasir Nishat et al. Sci Programm. 2022;2022:1–17, Krishnan et al. Int J Electr Comput Eng. 2021;11(6):2088–8708] is mostly based on open repository data, it is necessary to use real-world data to observe the DL-based models’ performance. For this cutting-edge effort, real-world EHR is collected from Excelcare Hospital in Guwahati, Assam, India with three timelines spaced at six-month intervals concatenated into multiclass sequential curated-electronic health records (MSC-EHR). The FRS risk estimating method is used to convert multiclass categorization with four risk labels on the curated dataset. Hence, this work aims at harnessing the benefits of MSC-EHR data with age-specific cluster (ASC) over age-agnostic cluster (AAC) for improved CAD prediction. This work proposes two hybrid models based on deep learning methodology. The study has been divided into two phases to analyze AAC and ASC datasets independently. The purpose of this research endeavor is to analyze the performance of hybrid models using deep learning techniques on curated dataset and investigate the impact of data pre-processing and balancing techniques on the model performance. For the phase 1 experimentation, Hybrid-Model1 (RNN+GRU) achieved 93.27%, and Hybrid-Model2 (LSTM+GRU) achieved 94.01% accuracy, while in phase 2, Hybrid-Model1 attained 96.93% (ASC2) accuracy, and Hybrid-Model2 attained 97.28% (ASC2) accuracy, highlighting the importance of ASC data over AAC in CAD prediction.