AI.
Machine learning.
Hospitals and healthcare.
The word “limitless” gets bandied about a bit when those ideas began to roil and coalesce, with ideas floated that relate to everything from predictive analytics and diagnoses to monitoring devices and health apps.
Turns out it’s kind of a big—or at least growing—thing.
Hospitals have a head start because they already have tons of big data—medical records, decades of procedures and more patients through the door every day.
But acquiring data is only the beginning; it needs to be accessible and actionable.
Efforts are in the early innings or, in hospital parlance, a preoperative stage.
Central Plan
Hospital systems have the data but it’s seriously siloed, said Michael Marino, senior vice president and chief of medical information at Providence St. Joseph Health. The Renton, Wash.-based health system and parent of a regional health system here, Providence St. Joseph Health Southern California, began two years ago to organize the data across its 51 hospitals and 829 clinics in seven states.
“It’s millions of encounters” a year, Marino said, and it grows “almost exponentially every day.”
Large databases of information—health records, recovery outcomes, financial performance—will help Providence “see patterns on things we didn’t understand before” and gain insight on treatment: how “complications, with these interventions, lead to these outcomes.”
A key issue, though, is how to store the material, which is by nature sensitive and often legally private: hospitals can have their own data storage centers, use cloud-based repositories—“data lakes”—from companies like Amazon and IBM, or develop a hybrid approach. Processing, analytic, and visualization tools bring more options to decide about.
MemorialCare Medical Foundation—which includes about 2,500 affiliated physicians—uses the Health Catalyst platform to warehouse and analyze data it collects, which, across metrics and treatment or prevention measures, lets “us quickly and simply sort, refine, call out, collate and drill down,” Chief Medical Officer Adam Solomon said.
Narrative Medicine
Once data is stored it needs to be used. Hospital staffs seek information giving value in real clinical settings, meaning they understand—and to some extent can tell—the story that letters and numbers represent.
It’s not about “a tool to remind them to check [for] fever,” Marino said, but one that connects symptoms in people to actions. “Tools that can [help] … not to annoy physicians but make their lives better.”
Children’s Hospital of Orange County Chief Health Information Officer Dr. William Feaster agrees.
On medical records, for instance: “I’m interested in how we can take all that information and enter it into the record without [creating] more work for physicians and causing burnout.”
Varieties of data from different sources—medical records, monitoring devices, wearables, scientific publications—is part of the challenge as well; materials need to be not just collected but collated and integrated. Feaster said a doctor, for instance, could spend 1,200 hours a year simply reading. Part of his work, then, is to get useful data from sources and translate that into daily work.
Solomon said via email that data can be the basis of “concierge outreach” to patients for scheduling procedures and “patients with ongoing health conditions” including diabetes, high blood pressure and other chronic ailments, “may derive even greater benefits.”
This is where AI, machine learning and predictive analytics can help.
Predictive Analytics
Kaiser Permanente and University of California-Irvine are aggressive in these areas. Kaiser runs an innovation center in Tustin and UCI opened its Center for Artificial Intelligence and Diagnostics Medicine in July.
Kaiser’s Dale Shim, managing director of healthcare innovation, said the health system’s data is “longitudinal: we stated collecting data [early] and tend to have longer history with our patients,” making for a rich storehouse of data.
It’s working on ways to use the material to help make better predictions and decisions, and delving more into virtual medicine, and data collection from wearables and at-home tools and devices.
UCI’s center is led by Drs. Peter Chang and Daniel Chow, neuroradiologists at the School of Medicine, and is focused on applying deep learning neural networks to areas including diagnostics, disease prediction and therapy planning.
AI can, for instance, contribute to 3D visualization and speeding up other work that ordinarily takes hours.
“There’s a lot of opportunity right now for AI to [make] the process infinitely faster,” Chang said.
Chow said detection and measurement might sound simple but have big effects.
“The standard of care literally is using a ruler to measure tumor size year-over-year,” he said. Accurate measurement affects diagnoses and drug dosage, among other issues.
Chang said implementation is less about the technology itself and more about “the dialogue between physicians and engineers making these algorithms—a lot is lost in these translations.”
Chow is “a physician interested in AI,” he said, “but I would be unable to initiate a program.”
Chang is “a software developer and physician” who can combine code and clinical need, Chow said.
The pair said this integrated expertise of doctors and computer scientists will help identify clinically relevant applications.
Precision Medicine
The goal is to use the tools in hospitals, helping patients and physicians.
CHOC, for instance, started looking at data from 40,000 admissions and came up with nearly a hundred variables that can be fed into healthcare’s developing big data machine, “using machine learning prediction [to] rate patients as low-, mid- or higher-risk for readmission,” Feaster said.
Such use could help predict, for instance, patients who “are likely to be readmitted within seven days of discharge.” Hospital caregivers could intervene with higher-risk patients—identified in electronic medical records systems—“prior to discharging the patient to reduce our readmission rate.”
CHOC has already seen a downward trend for readmission, Feaster said.
The healthcare provider also plans tools that can track common pediatric diseases and recommend care. In asthma patients, for instance, it has developed an alert system to remind patients to take pulmonary function tests—part of how the malady is managed.
MemorialCare’s hospital system has used data to improve on “a key safety focus across California”—lowering the rate of birth by caesarean section to 22.5%, below state goals of 23.9%, according to Chief Transformation Officer Helen Macfie and Chief Medical Officer James Leo.
Providence has created a sepsis identification and treatment protocol, first established at Hoag Memorial Hospital Presbyterian in Newport Beach, and applied it throughout its system, a key reason it wants to develop the clinical best practices in the first place. Sepsis—the body’s response to infection—can be an early-stage condition, and life-threatening if not identified and treated quickly.
“Population health depends on where you start on the continuum,” Marino said. “We focus on early intervention” with tools built from data that helps screen patients for potential problems.
