Doctors have long viewed artificial intelligence with skepticism. Most of them believe that AI and machine learning techniques are exaggerated, and cannot solve clinical problems in real life.
Doctors tend not to like the machines that dictate their decisions. They prefer to rely on their clinical acumen and judgment for diagnosis and clinical decision making.
But in today’s changing care delivery landscape and consumers demanding better engagement and a better care experience, clinicians are rethinking how to improve care delivery.
Enabling decision-making, improving health outcomes
The question has never been about artificial intelligence as opposed to clinician decision-making. As Atul Gawande says in his bestselling book, Complications, “No matter what measures are taken, doctors will sometimes falter, and it is unreasonable to demand that perfection be achieved. What is reasonable is to demand that we never stop aiming for it.”
In this book, Gawande provides real-life tales of mistakes made by surgeons and doctors.
When backed by a layer of an AI- and machine-learning-enabled assistant that navigates through historical data and draws similarities and relevant insights about a situation, the decision-making process can go a long way toward speeding up the diagnosis and decision-making process.
Consider a diagnosis of sepsis. AI algorithms are widely used in critical care units to diagnose sepsis. The sepsis sniffing algorithm alerts the clinician at least 3 to 4 hours before an escalating event leading to severe sepsis occurs. This could lead to lower death rates.
Early indications are given by an algorithm that runs in the background and collects all data generated from the patient’s bed and labs, producing intermittent results to alert the clinician to an impending crisis.
Hospitals experienced Rate 39.5% reduction in in-hospital mortality, 32.3% reduction in length of hospital stay, and 22.7% reduction in 30-day readmission rate for sepsis-related patient stays.
AI is one lever that can be used as a second opinion to make a diagnosis in complex cases and strive for perfection.
Patient participation and care experience
In today’s scenario of virtual care that integrates with personalized care, an AI-enabled clinician can delegate routine and mundane tasks such as sending educational materials, ordering prescription refills, and responding to patient inquiries with the support of AI algorithms.
In larger facilities, by using AI-powered tools such as symptom screening tools to triage patients, a clinician can improve the functionality of their clinic or department. Using AI-powered chatbots to answer routine questions, and book appointments are other uses of AI that help improve patient experience.
AI algorithms help identify chronically ill patients, send them medication reminders and educational materials, and alert clinicians of any changes in their vitals or labs when connected devices are used. In general, leading to increased patient participation and responsibility for their own health.
Choosing the use cases in which AI can be successfully implemented
It is important to identify use cases where AI algorithms can make a tangible impact in clinical areas. Some of the areas in which AI has been successfully implemented are radiology, internal medicine, neuroscience, and cardiology.
In all of these areas, algorithms run quietly in the background and help clinicians make a difference, sometimes by offering a second opinion or simply alerting any impending crisis. Nowhere has AI overshadowed the presence of a doctor.
Patients always prefer to hear their diagnosis from their doctors. In imaging, today, AI models are helping to automate the identification of healthy tissues and organs from tumors, the development of adaptive doses and treatment plans for radiotherapy, the diagnosis of early-stage cancers, the diagnosis of large vessel occlusions in stroke, and the identification of disease patterns for images. This is subsequently reviewed by the physician and radiologist, who is familiar with the patient’s general clinical, social and psychological profile.
Machine learning contains algorithmic bias and will always be accompanied by a slogan or disclaimer: “Clinical correlation essential.” However, artificial intelligence superseded by clinical intervention from a specialist who is aware of the human aspects of the patient is a good solution for an integrated machine and human model of care.
Several other use cases have been implemented or are in development to aid bedside diagnosis. Recently, techniques such as natural language programming to read unstructured information in physician notes and voice-enabled assistants to predict emotional and behavioral traits are subject to research.
AI has made significant advances in the administrative and operational areas of healthcare, and is setting a measurable mark in increasing revenue for large health systems.
But AI is also suffering from a series of failures in clinical areas, resulting in a lack of real-world deployments of machine learning algorithms in mainstream clinical practices. Recent examples include IBM Watson’s failure to diagnose and treat cancer, and Google’s failure to detect diabetic retinopathy using deep learning models from images of patients’ eyes.
The potential of AI in healthcare has not yet been realized. There are a limited number of reports available on the clinical benefits and cost arising from the actual use of AI algorithms in clinical practice.
Although slow, AI in clinical areas is steadily rebounding but it needs to deliver on its promise to make a difference at the point of care.
As health systems and hospitals are digitizing to improve care delivery and patient experience, doctors cannot be left behind. They must also change and contribute to making this shift into a more positive experience for themselves and their patients.
Dr. Jyoti Goswami is a Principal Consultant at Damo Consulting.