Personalized medicine is based on a simple concept: tailoring healthcare to a specific individual. At its best, personalized medicine should be both predictive and preventive, promising fewer medical complications, less guesswork by providers, better outcomes and, ultimately, a more efficient system of care. However, this promise has been difficult to realize. Only in the last decade or so have we had access to the data, tools and technology to make it possible.
When most people think of personalized medicine, they think of genetics and genomics – using a patient’s molecular biomarkers to determine his or her risk for certain diseases or to decide which therapies to employ. But these alone are not enough: additional critical influences, including family history, socio-economic status, and environmental and behavioral factors are needed to develop a comprehensive individual profile that can support better predictions about treatment outcomes. And, all of the needed information must be made available and usable to clinicians rapidly-via clinical decision support tools-so that it can actually make a difference in the delivery of treatment.
CRI is now piloting a predictive model for personalized medicine that is designed to help clinicians more accurately plan and deliver individualized courses of treatment. This approach promises to improve patient outcomes and lower treatment costs by rapidly applying the best available, scientifically gathered evidence to treatment decisions. To date, intelligence derived by this approach has helped clinicians to accurately choose the most appropriate treatment option for a given patient over 70 percent of the time, a rate that compares favorably to the 25 percent rate of accurate first-time diagnosis and treatment using traditional methods.
EHRs: The starting point
CRI's new predictive model works with the data from electronic health records (EHRs), which are proliferating throughout healthcare. Normally, of course, EHR data can tell us only what happened in the past. But predictive modeling technology makes it possible to analyze past data and transform it into useful, actionable information--treatment recommendations that are individualized to the patient--as a support to clinical decision making.
As any clinician knows, there's an overwhelming supply of information available about possible treatment approaches for a given patient. For example, more than 20 different medications are available to treat depression, yet it is very difficult to know which one might work best. As a result, clinicians often are forced to make educated guesses. They sometimes may prescribe the right antidepressant, but other times not. Such trial and error is more expensive, more arduous for the patient, and perhaps even dangerous.
CRI's predictive model offers one way to improve the probability of providing the best treatment. It consists of a set of data-mining algorithms that:
- analyze multivariable patient treatment information,
- provide individualized predictions of patient outcomes, at the time of intake, and,
- help clinicians optimize treatment selection for mental health disorders.
The model also collects and incorporates direct patient feedback (patient-reported outcomes, or PROs) that better elucidate aspects of treatment that, due to the difficulty of measurement or collection, may not be fully considered (e.g. social functioning, pain). Understanding CRI’s predictive model
The key concept behind CRI's predictive model is that it combines knowledge of existing, evidence based practices with knowledge gained from real-time, real-life "practice-based evidence." This combination, and the algorithms that drive it, promise to reshape clinical opinion about the value and content of clinical decision support systems.