Hospitals and healthcare systems are introducing CDSS platforms to enable the use of machine learning to assist with diagnostic decisions and to predict treatment outcomes. The CDSS works by continuously monitoring information that clinicians enter into the EHR. As information is recorded, the CDSS can analyse the entries in real time along with other clinically relevant data that is linked to the EHR from other unrelated sources. These sources may include test results from pathology laboratories, radiological departments, genetics departments, and ambulatory settings, as well as research results stored in biobanks, clinical trials, and databanks of genome sequences. The CDSS can then make diagnostic recommendations based on algorithms that are typically programmed using rules informed by established clinical guidelines and published medical research reviews. Many diagnostic applications have been proposed for CDSS and some programs been approved for marketing as medical devices in several countries. This includes software with deep-learning algorithms that can analyse medical images to diagnose cardiovascular disease, wrist fractures, stroke, and diabetic retinopathy. Other applications undergoing clinical trials include programs use to diagnose breast and skin cancers, congenital cataract disease, Parkinson’s disease, and diabetes mellitus (Jiang et al. 2017). In many instances, the machine learning has been demonstrated to be at least as accurate as an experienced clinician with the added advantage of being able to recommend a diagnosis much faster. As the diseases being targeted for CDSS are typically chronic and/or degenerative, early diagnoses are critical for reducing complications, controlling symptoms, and improving outcomes. For example, CDSS can analyse the data stored in the EHR on patients already diagnosed with diabetes, such as symptoms, medical history, physical examinations, lab tests, treatments, and outcomes to make predictions and to formulate a possible diagnosis for a patient with similar traits and characteristics (El-Sappagh and Elmogy 2016). Similarly, the software is able to predict the likely outcomes of treatment options by accounting for information that a clinician may not readily have access to or be unaware of when recommending a management plan to their patient. The platforms can alert clinicians to potential problems and conflicts with the treatment plans, such as likely drug-efficacy probabilities, and allow the doctor to revise the management plan accordingly.