Predictive analytics in healthcare pose unique challenges

Looking to take a page from Amazon, Netflix and Google, investors have poured $1.9 billion into companies pursing predictive analytics in healthcare, though predicting the best course of treatment is significantly different than recommending a book or movie, according to a new Rock Health report.

"Even if we had the technology to address interoperability issues, solve privacy concerns, and process unstructured data, hundreds of thousands of factors influence health--many of which medical science still can't explain. Additionally, health outcomes aren't instantaneous. Without an effective, closed-feedback loop, algorithms struggle to continue to learn and improve," Rock Health's Teresa Wang writes in a blog post.

Personalizing care through predictive analytics poses the opportunity to significantly reduce healthcare costs, the report states, including $192 billion in overtreatment, $128 billion in failures of care delivery and $35 billion in lack of care coordination, figures attributed to the Journal of the American Medical Association.

The report focuses on companies using predictive algorithms that directly affect patient care, such as clinical decision support, readmission prevention, adverse event avoidance, disease management and patient matching.

It identifies six components in reaching the goal of predicting which treatment will produce a favorable outcome for the patient: aggregating training data, searching for relationships in data, collecting data on the specific case, characterizing that individual case, contextualizing the recommendation, and capturing the performance of a recommendation.

The outcome of a treatment, however, can take years, making the time involved in creating a feedback loop a significant challenge, according to the report. The funded companies, it notes, are almost entirely focused on providers, practically ignoring patients. Another major challenge is presenting a clear path of action from the data.

While real-time data collection can reduce intervention response time, data can vary in accuracy and timeliness. Collecting data in real time requires new infrastructure and workflow, the report states, as provider and patient acceptance of algorithm-based treatments.

The health analytics market is poised for 25 percent annual growth over the next five years, IQ4I Research & Consultancy predicted recently.

Massachusetts General Hospital also uses a predictive analytics tool, including surgeon acceptance of its predictions on risk, according to David Ting, associate medical director for information systems at the hospital.

To learn more:
- check out the report
- read the blog post