How Healthcare Companies Can Use Data More Effectively
The future of healthcare data is in electronic medical records and electronic health records (EMR/EHR).
13:23 05 September 2020
The US has already invested heavily in EMR/EHR, while medtech startups count on widespread adoption of digital health solutions by both providers and patients to drive a revolution in effective and affordable healthcare.
More patient data is available to practitioners and researchers than ever before. With clinical visits feeding into EMR/EHR databases, and with patients feeding the beast constantly, between clinic visits, thanks to telehealth, mail-in testing, and remote monitoring -- there’s a real risk of information overload. Once they have the data, how can healthcare companies put it to the best use?
Here are three ways healthcare companies can use data more effectively.
1. Use Predictive Analytics to Reduce Costs
One of the most powerful uses of large datasets is predictive analysis. Consumers experience predictive analytics every day, in the way that Amazon uses their past buying and searching history to recommend products; in the way Netflix uses our past viewing habits to recommend our next binge watch.
But according to Orthogonal, predictive analytics—that is, analyzing available data sets to make predictions about comparable circumstances—does more than drive buying decisions. Healthcare providers can use it to reduce costs in any number of ways. Examples include predicting needs of high-cost patients or predicting which patients might be at risk of readmission. 5% of all patients result in 50% of healthcare costs, and as many as one third of all readmissions are estimated to be preventable if the risk factors can be identified ahead of time.
One key function of healthcare providers is triage—evaluating the risk of complications of each patient and prioritizing treatment to ensure the best possible outcome for the patient population as a whole. Kaiser Permanente Northern California led the way by testing out two protocols for data-driven triage:
- A two-step protocol developed in partnership with Harvard, UC Santa Cruz, and UC San Francisco to identify newborns at highest risk of early-onset sepsis. The goal is to reduce the number of infants to whom antibiotics are administered.
- A protocol to integrate pre-existing “patient risk” rubrics into a unified risk rubric. The existing systems have not been widely integrated into triage, but if effective may represent a data-driven system for emergency rooms to quickly triage.
2. Leverage Artificial Intelligence to Optimize Care Plans
Again, a key challenge of massive data sets is information overload—there’s so much data, it would take several human lifetimes to extract anything of value from them, on top of which combing through all that data would turn a human mind to jelly. (Metaphor, notmedical opinion.)
That’s where Artificial Intelligence (AI) comes in. Less flashy and ponderous than the Steven Spielberg movie, AI is simply a class of computer programs that performs certain tasks better than any one human brain.
This is an important distinction. Despite the warnings of the sci fi films, AI isn’t about machines being smarter than us, and it certainly isn’t about machines becoming self-aware. A human mind could easily detect patterns in the vast datasets currently accumulating in EMR/EHR systems -- but that human mind probably has better things to do, like treating a patient or enjoying a glass of wine and a sunset.
AI computers are capable of taking in those massive datasets and running pattern-recognition algorithms on them to discover subtleties hard to spot in smaller datasets. For example, human eyes looking at three MRIs might not detect the shadow of the earliest stages of cancer, but a machine looking at one million MRIs with image-recognition software might be able to spot the subtle clues and learn to spot them in individual MRIs fed into the machine in the future.
A key application for AI pattern recognition in healthcare is in optimizing treatment plans for diseases that affect multiple organ systems. It might be easy to track the progress of or calibrate treatments for a disease that only affects the lungs, but if it involves the heart, lymphatic system, and pancreas simultaneously, picking the right treatment plan becomes much more of a shot in the dark.
AI algorithms can process data sets that are not only large, but also multivariate. It can be used to build treatment optimization models for data taken from multiple organ systems, assessing the risks and rewards of a single treatment plan for every vector of the disease. The powerful potential of this kind of treatment analysis is a hot topic in healthcare AI.
3. Analyze Simultaneous Data Streams in Remote Monitoring
Remote data monitoring is already widespread. Since the first Fitbit shipped, the Internet of Medical Things has evolved to include remote monitors for heart rate, breathing, ECG, sleep cycles, blood glucose, and much more.
But much of that remote data collection is geared toward patients who are healthy, or at least relatively stable. Patients recovering from a surgery or serious illness still regularly convalesce in ICUs, at great cost to the patient and to the system.
The rationale of the ICU is that any patient is likely to experience decompensation—a sudden worsening in condition—and should therefore be monitored in the ICU, often for several days.
Remote monitoring devices of the same vital statistics monitored in an ICU have leapt forward in sophistication in recent years. The problem is false positives. For remote cardiac telemetry monitors, for example, an alarm represents a legitimate decompensation only 5-10% of the time. That’s a lot of false positives. Deploying resources to help that many patients who might not even be seriously ill would be even more taxing on the system.
Medtech companies are solving this problem by developing simultaneous data stream remote monitoring. For example, some discharged cardiac patients go home with a small disk that gets inserted under the mattress. This one device tracks three vectors of the patient’s condition—the patient’s breathing, heart rate, and movement. When an alarm is triggered, it represents a legitimate decompensation almost 50% of the time—significantly better than the one-variable cardiac monitor.
The adoption of EMR/EHR systems means that healthcare data collection is here to stay. How well healthcare companies and providers use that data will determine whether we are healthier and richer for it.