On the long list of “Things Physicians Hate,” one of the top items is shifting attention from the patient so they can enter EMR data using dropdown menus.
Since the days when Hippocrates used parchment, physicians have relied on their notes. That’s where all the relevant information and observations reside to inform clinical decision-making. But far too little of this information finds its way into the structured fields in an EMR.
For example, if a patient didn’t refill a prescription because it cost too much, that information is often found in the physician notes, not the EMR’s structured fields. The structured, EMR data may have a wealth of information on a patient’s health history, yet not include any mention of the patient’s recent job layoff (which can contribute significantly to comorbidities like depression). But all this highly relevant information can usually be found in the physician’s notes, where it can be searched and analyzed.
There’s a new generation of technology tools to do that very thing. They use artificial intelligence (AI) algorithms to comb through – and learn from – physician notes. For instance, without a programmer telling the software what to do, it can identify a potential asthmatic from words like “wheezing”, “albuterol”, and “nebulizer,” even if the word “asthma” never appears in the notes.
In addition to quickly and flexibly addressing questions about a population, AI technology also highlights subgroups of patients. With a chronic condition like CHF, it’s rarely just a group of patients who have CHF only. Rather, it’s subgroups of CHF patients who suffer from different mixes of comorbidities.
Improving Population Health
Using AI and predictive modeling to leverage the unstructured text in physician notes provides a strong foundation for population health management – much better than using claims data and ACO metrics to define best opportunities for intervention.
In a recent study, 19 percent of high frequency ED patients had a discharge diagnosis of schizophrenia. But AI-guided analysis of physician notes told a different story: 51 percent had schizophrenia.
With AI technology, healthcare clinicians and administrators get a much clearer picture of the trends underneath the metrics. For example, the structured data in an EMR could tell you that a large number of diabetic patients are out of target range for HbA1c. But an analysis of the physician notes can tell you why: poor eating habits, personal loss, no transportation, etc.
Predicting Future High Utilizers
Most CMOs and clinical administrators know the identity of their chronically high-cost patients. It’s the rising-risk patients who present the greatest opportunity for improving overall population health and reducing expense. AI technology can predict the Medicare patients who are most at risk of becoming high utilizers.
In one analysis of Medicare charts, AI technology identified a group of patients that had nearly three times the number of hospital admissions than would normally be expected for the population. Nearly a third of identified patients had no prior admissions whatsoever – and 65 percent were rising-risk patients. Guiding early interventions to rising-risk patients, AI technology can help save the healthcare system millions annually.
EMRs play an important role in healthcare, but they need to be augmented by AI tools and predictive modeling. This allows hospitals and practices to smartly deploy limited resources where they’ll have the greatest impact on both patient outcomes and the cost of care.