Mental health conditions such as bipolar disorder or depression can be difficult to definitively diagnose, but algorithms based on EHR data of confirmed diagnoses can predict individuals with 79 to 85 percent accuracy.
An article by University of Iowa psychiatry professor James B. Potash, MD, published in the American Journal of Psychiatry, discussed the prediction model, which aggregated 50,000 potential bipolar disorder cases and established an algorithm based on manual review, text features and coded data. Approximately 63 percent of the cases could be classified as bipolar disorder based on the DSM-IV definition, and of those, between 79 and 85 percent proved to be correct based on later direct interviews, Dr. Potash wrote.
The root causes of bipolar disorder have been hard to track over time. Repeated studies have shown that multiple small genetic predispositions lead to bipolar disorder, and a sample size of tens of thousands of patients will yield the most accurate trial result. However, it is hard to enroll so many patients in a single trial for comparison, Dr. Potash wrote.
EHR data provides that sample size. Big data, especially the proliferation of genetic analysis, allows researchers to more easily study the effects of genetic mutations on mental conditions, he wrote. This method can also be less expensive than a traditional trial, he wrote.