ABSTRACT
In recent years, there has been a huge amount of data collected in hospitals in terms of electronic medical records (EMR), continuity of care documents (CCD), electronic healthcare records (EHR), and many more. These records contain information about patients such as their location, symptoms, diagnosed disease, blood test results, scan images and reports, recommended medications, and doctor’s notes. The aforementioned details can assist physicians and care providers in providing better healthcare solutions for the end-user. On the other hand, the amount of data available has increased tremendously, necessitating the use of analytical, clinical, and business intelligence tools to transform it into useful information. Researchers are attempting to integrate the latest breakthroughs in deep learning technologies to extract clinical information from data traditionally used for local billing and archiving purposes, which is advantageous to people. Because these records were created for the sole purpose of local administration and hospital management, they were unstructured, necessitating deep learning research, which has been shown effective in a number of other domains because of its 134efficacy in capturing data dependencies. Clinical risk prediction based on such big data analysis of healthcare records will undoubtedly aid in the prediction of undesirable occurrences such as cardiac arrest, lowering patient mortality rates. Several researchers are working on this topic in order to provide better patient care and gain more insights into how to enhance healthcare in the future. The most difficult aspect of dealing with these records is the disparity in how patient data is kept, such as weight as numbers, admission data as dates and times, categorical values for diagnosis, natural language for doctor’s remarks, and so on. Other issues, such as sparsity, missing data, and large data dimensionality, require specific research attention. As a result, more research is needed regarding mining these massive amounts of data; at present, research has shortcomings in model design, deep learning technique deployment, and the lack of globally acknowledged assessment standards that are highly needed. The present work examines the current state of the art, existing deep learning approaches used, the research gap, and recommendations for improved deep-learning deployment in EHR research.
