A misdiagnosis can impact us all — think about it, it not only changes the lives of patients, but also impacts their friends, their families and clinicians. After hearing countless stories of poor healthcare experiences stemming from misdiagnosis (not just on the news, but in my close circle of friends), I couldn’t help but wonder why are they so common, and most importantly, what can we do about it?
For many patients and doctors, getting to the bottom of a diagnosis can feel like you are going down a rabbit hole. Clinicians are often time poor and the stakes are very high. An accurate diagnosis is one of the first and most crucial steps when it comes to providing quality patient care and increasing the likelihood of positive health outcomes. A report by the Australian Institute of Health and Welfare found that for colorectal cancer, a person diagnosed at the early stage had a 99% chance of surviving 5 years and only a 13% chance when diagnosed at the later stage. This statistic alone shows how valuable it can be to detect diseases early on.
The process of reaching a medical diagnosis includes evaluating the patient’s symptoms, medical history and test results. The end goal is to determine the root cause of the problem and make an accurate diagnosis in order to provide an effective treatment.
The Medical Journal of Australia estimated that 140,000 diagnostic errors occur in Australia per annum. Of those cases, 21,000 are of serious harm, and between 2,000–4,000 result in deaths. More than 75% of diagnostic errors were a result of cognitive factors in decision-making (that is 105,000 cases!). Two of the main contributors to this were a failure to form an adequate diagnosis (i.e., using mental shortcuts, fatigue and distractions) and overconfidence in a misdiagnosis.
Human error is inevitable when it comes to diagnosing patients, but how can we reduce this? This is where AI can come in to help clinicians make more informed decisions when diagnosing patients.
Hospitals produce ~50 petabytes of data per annum. This data consists of clinical notes, medical images, lab tests, operational data and more. The World Economic Forum found that ~97% of data produced by hospitals goes unused. Among many transformative applications, AI enables healthcare professionals to tap into this unused data, structure it and leverage it to improve diagnostics accuracy and the quality of patient care.
One of AI’s most promising applications includes assisting healthcare professionals with diagnosis and decisions around suitable treatment plans. This not only saves precious clinician time and practice money by removing inefficiencies, but also improves patient outcomes.
AI can ingest and analyse large amounts of patient data from various sources (i.e., medical images) and be trained to identify patterns that are predictive of disease. As more data is input into the model, the accuracy of the ML algorithm improves as the model learns (i.e., clustering information into useful groupings). The model can generate insights, predictions and recommendations, which assist the clinician in making more informed decisions, ultimately reducing the chances of misdiagnosis.
There are an increasing number of examples of AI tools demonstrating improvements in medical diagnostics popping up, including Sybil. MIT researchers recently developed an AI model trained on CT scans, which assesses future lung cancer risk in patients. One study demonstrated that the model can accurately predict whether a patient will develop lung cancer in the next year, 86% — 94% of the time. Lung cancer is the fifth most commonly diagnosed cancer in Australia and the deadliest cancer globally. Given the difficulty in treatment (especially as the cancer progresses), it is promising that there are more models being developed that can assist doctors in early detection.
This all sounds promising — so why isn’t every medical provider using AI?
There are many complex factors that have resulted in some medical providers choosing not to adopt AI. Some of these factors include a lack of familiarity with AI and ML, gaps in regulatory guidance, costs, and the effectiveness not yet being proved in a wide variety of clinical settings.
With amazing companies building in the medical diagnostics space such as Freenome, Color and Paige (among many others!), I am optimistic that in due time, AI will be leveraged widely to enhance traditional healthcare practices, especially with regard to prevention, causation, prediction and treatment.