By 2030, the healthcare artificial intelligence market is expected to be worth almost $188 billion.
The Institute of Electrical and Electronic Engineers, the world’s largest nonprofit technical organization dedicated to the advancement of technology for the benefit of humanity, is keeping a sharp eye on AI – both the benefits and challenges of the technology that has been exploding in healthcare.
That’s why Healthcare IT News recently sat down with IEEE Fellow Chenyang Lu. We asked him how AI is being used by healthcare professionals to assist with improving patient outcomes, the challenges with implementing AI within healthcare and how to overcome those challenges, and what he thinks the future of AI in healthcare looks like. He offered some very insightful answers.
Q. What is your view on how AI can be used by healthcare professionals to assist with improving patient outcomes?
A. AI will effectively become copilots for our physicians and enable timely, precise and efficient treatment for each individual patient. AI models can make personalized predictions of a patient’s clinical outcomes, risk factors and response to different treatments. Here are three examples of AI in healthcare with great promise to improve patient outcomes.
First, depression screening. According to the WHO, more than 280 million people suffer from depression. Among them, more than 50% are not diagnosed or treated. The underdiagnosis problem stems from the substantial time and cost incurred to get diagnosed by psychiatrists. A recent study showed that deep learning models can detect depression and anxiety disorders using data collected with wearable devices, opening a new pathway to screen for depression unobtrusively.
This AI-based screening tool will enable clinicians to deliver selective prevention programs to individuals in a targeted and timely manner, addressing a critical evidence gap in depression prevention identified by the United States Preventive Services Task Force.
Second, cancer care. Cancer patients are at high risk for clinical deterioration: 6.4% of oncology inpatients have at least one ICU transfer, and 2.7% of them die on the hospital wards, according to a recent study. Machine learning models can generate early warnings for clinical deterioration of oncology inpatients by integrating heterogeneous data in the electronic health records.
AI-generated early warnings, alongside risk factors associated with the predictions, allow clinicians to identify patients at risks in advance and provide early interventions to prevent deterioration. Clinicians also face challenges in decisions about discharging patients from oncology wards. Prolonged stay diminishes availability of hospital access for patients with cancer. Machine learning models can be employed to determine when a patient hospitalized with cancer is clinically stable for hospital discharge, thereby improving cancer care efficiency while ensuring patient safety.
And third, perioperative care. Surgery incurs significant risks and cost to patients. Early identification of risk factors can be crucial to early intervention and improved outcomes. For example, pancreatic resection is the only cure for pancreatic cancer but is commonly associated with a high rate of severe complications. Using data collected with certain fitness wristbands, machine learning models can predict a patient’s risk for severe complications before surgery.
If a patient’s risk is high, they may be enrolled in prehabilitation programs to enhance their readiness for surgery. Using EHR data, machine learning models have also been developed to identify risks during surgery and to predict complications after surgery, for improving the safety and outcomes of patients in perioperative care.
Q. What do you see as the major challenges with implementing AI within healthcare and how can hospitals and health systems overcome these challenges?
A. Integration of AI models with the EHR and clinical workflow is essential for implementing AI in healthcare. However, there are significant challenges in implementing AI models on current EHR platforms, in contrast to commercial cloud platforms that have made it much easier to build and deploy AI.
Currently, we have numerous AI models in the pilot stages, but few have been deployed in EHRs. We are still at the early stage of AI in healthcare. Looking forward, it is imperative to lower the hurdles for implementing AI models in our infrastructure.
Furthermore, we need to retool our workflows and protocols so clinicians and AI can work together effectively. Experience in recent years has shown AI and clinicians provide complementary capabilities. AI will be copilots that work with clinicians to produce the best decisions and treatments collaboratively. Significant research is needed to develop effective human-in-the-loop AI in clinical settings.
Q. What do you see as the future of AI in healthcare? What’s next, and where is it anticipated to head in the coming years?
A. We are seeing early adoption of generative AI to improve operational efficiency by automating clinical documentation and patient communication. Despite the challenges in implementation, we will see increasing adoption of AI-based clinical decision support, driven by the great potential to improve patient outcomes and healthcare efficiency.
Importantly, we need to generate evidence for the efficacy and benefits of AI in healthcare in terms of patient outcomes and cost-effectiveness so we can incrementally build up AI capabilities in our health systems. In the meantime, we need to ensure fairness, safety, security, privacy and access to AI in healthcare through both policies and technologies.
This is another area where significant research is needed to enable sustained growth of AI in healthcare.
Follow Bill’s HIT coverage on LinkedIn: Bill Siwicki
Email him: [email protected]
Healthcare IT News is a HIMSS Media publication.
Source : Healthcare IT News