More than 90 percent of hospitals in the United States now rely on electronic medical records. Digitization offers huge benefits: different care providers within a system can access the same patient record, creating opportunities to streamline care, and patients can review their medications and care plan.
Indeed, electronic medical records have been shown to reduce costs and save lives. The downside, however, for both doctor and patient alike, is that care providers increasingly find themselves spending more time facing their computer screen than their patients.
Leora Horwitz, MD, and her colleagues in the predictive analytics unit at the Center for Healthcare Innovation and Delivery Science see artificial intelligence, or AI, as the perfect tool to help alleviate that burden. At the same time, it can perform tasks that are beyond the reach of human cognition, like finding hidden patterns in a sea of clinical records. “Patients can have problems, whether diagnosed or undiagnosed, that may seem small but aren’t,” says Dr. Horwitz. “Our AI system can help clinicians prioritize those problems.”
The predictive analytics unit has already rolled out systems designed to comb through common medical records and flag patients who are at higher risk for a range of conditions. One system looks for the telltale signs of potential heart failure, such as rapid heart rate and weight gain from accumulating fluid. Then, it analyzes a list of the patient’s health challenges. If heart failure isn’t on the list, the software flags that risk on the patient’s electronic medical record.
“Normally, heart failure is on the problem list for patients who are at high risk for the disease,” says Dr. Horwitz. “But if the patient is new to our system, the records may not yet have transferred in, and clinicians may be focusing on other, more immediate medical issues.” Almost all patients at risk for heart failure are eventually identified by clinicians, she notes, but the software provides a second line of defense.
Another system scrutinizes the medications, lab results, and clinical history of hospitalized patients with kidney disease to predict which ones are most likely to suffer kidney failure and require immediate dialysis. Under ordinary circumstances, dialysis patients undergo a surgical procedure to interconnect an artery and vein in their arm, forming a strong section of blood vessel, called a fistula, that can safely connect to a dialysis machine.
Without a fistula, they are at higher risk of infection and clotting. With the AI system, an alert could help clinicians decide to schedule a patient for dialysis before the need becomes urgent, allowing enough time to place a fistula.
Dr. Horwitz emphasizes that the unit’s efforts are not intended to offload any responsibilities from clinicians, but rather to augment their decision-making. “In the torrent of data that clinicians are exposed to today, it’s so easy for something to be overlooked,” she explains. “We’re building systems that bring some of that data to the top, so that clinicians can do what they think is appropriate with it.”