Making healthcare cognitive

Most professions rely on precise data to make decisions. Can you imagine erecting a building without numeric measurements, angles and detailed plans? Healthcare data is not as precise. In fact, the most valuable data is often unstructured notes recorded from patient and doctor visits or encounters. Correctly interpreting that type of data can be literally a matter of life or death. The ability to accurately interpret unstructured data is essential, especially when the situation is often ambiguous and shifting.

Yet, when I think of how most healthcare organizations are analyzing their clinical data, I get a mental picture of the old shell game—lots of guessing, hand waving and looking under the wrong shell for a missing pea.

Many healthcare analytics systems and projects are like that. The missing component (the pea) in healthcare data analysis is unstructured data. Electronic medical records (EMR) systems today are optimized for structured data (20 percent of the available data). Yet, in healthcare, critical clinical information is hidden in the unstructured data, free text, images, recordings and other forms of content. Nurses’ notes, lab results and discharge summaries are just a few examples of unstructured information that should be analyzed, but in most cases … is not. Would analyzing only one-fifth of the available information yield solid conclusions?

When the University of North Carolina Health Care evaluated the accuracy of data mining in its cancer database, they found mammogram values in structured data 52 percent of the time, and in unstructured data 48 percent of the time. They also found that CRC screening (colon cancer) values are present in structured data just 17 percent of the time … and present in unstructured data 83 percent of the time. As a man of a certain age, this scares me in words that can’t be published.

Cognitive systems are designed to unlock all kinds of information and synthesize it into one picture—of a patient, a trend, a customer. They apply multiple techniques for merging and analyzing all types of data to reveal the truth or detect previously unknown patterns in the data. Those systems may bubble up new findings that will directly affect how we prevent and treat disease.

Project findings

Seton Healthcare Family project analyzed both structured and unstructured clinical (and operational) data using cognitive computing methods to help prevent hospital readmissions of congestive heart failure (CHF) patients. The following is a snapshot of some of their findings.

1. The data thought to be useful … wasn’t. In some cases, the unstructured data is more valuable and more trustworthy than the structured data:

• Left ventricle ejection fraction (volume of blood pumped by the heart) values were found in just 2 percent of the structured data from patient encounters, but in 74 percent of the unstructured data.
• Smoking status indicators were also found in both places. But here’s the kicker … The structured data values were only 65 percent accurate, and the unstructured data values were 95 percent accurate.
• Type of living arrangement can predict readmissions, but living arrangement was found in less than 1 percent of structured data. It was the unstructured data that revealed those insights (in 81 percent of patient encounters).

Without including the unstructured data in the analysis, the ability to make accurate predictions about readmissions is highly compromised—it significantly undermines (or even prevents) the identification of the patients who are most at risk of readmission … and the most in need of care.

2. New unexpected indicators emerged. Seton’s analysis identified 18 top CHF readmission predictors from hundreds of potential factors. The number one indicator came out of left field … jugular venous distention indicator. This was not one of the original candidate indicators and only surfaced through the analysis of both structured and unstructured data (or finding the pea).

The old shell game should be a thing of the past. Cognitive computing enables the “peas” to be found, trusted and used in the same way we trust and use structured data. If an ounce of prevention is worth a pound of cure, an ounce of perspective extracted from a ton of data is priceless in potential savings. In better diagnoses, more efficient healthcare systems, and more discovery of drug and disease causes and effects, cognitive computing becomes a powerful ally for healthcare professionals.