In a recent survey, health care executives predicted that the secondary use of data which is captured during the care process will soon prove useful in improving the quality of care and public health, reducing costs, and getting drugs to market faster.
Now, Boston-based scientists have published a piece in the BMJ that supports this contention, and shows that potential insights from such data are limited only by the creativity of those who generate researchable questions.
In a truly novel use of secondary data, Ben Reis and colleagues from Children’s Hospital built a tool that enabled them to identify people at risk for domestic abuse fully 2 years before physicians actually established the diagnosis.
Domestic abuse is the most common cause of nonfatal injury among US women. It accounts for more than half of all murders of women, and involves as many as 16% of all married couples each year. Men can also be victims of domestic abuse.
The scientists reviewed 6 years worth of claims data from more than 500,000 hospital admissions and ER visits for patients who were at least 18 years of age. The patients had been assigned more than 16 million diagnoses in total. Cases of abuse were identified using established coding norms.
The scientists built a Bayesian model and inserted various elements of the past medical history to see which combination of elements could best predict cases in which a diagnosis of domestic abuse was eventually made.
They then used the final Bayesian model on a fresh set of patient data and found that it predicted a future diagnosis of abuse 10-30 months in advance of the time the actual diagnosis was made. The model achieved high degree of sensitivity and specificity (area under the ROC curve of 0.88).
For women, the factors that raised red flags were hospital visits for injuries, poisoning, and alcoholism. For men, psychosis and depression were associated with the highest risk.
The scientists don’t believe their work is quite ready for widespread use, but their hope is that it can serve as the basis for an early warning system that could help physicians decide which patients should be subjected to further screening or possibly intervention.
“This is not a diagnosis but a screening support system,” Reis told the Boston Globe. His group now plans to further refine the model, and use a similar approach to see if they can develop models that predict diabetes and depression.
In retrospect, the warning signs identified by the Boston scientists seem obvious, but harried primary care and ER physicians rarely have the opportunity to access, absorb and review the larger contextual issues surrounding any particular encounter when they’re trying to cram every patient visit into a 15 minute time slot.
Physicians “may not be taking full advantage of the growing amounts of longitudinal data stored in electronic health information systems,” Reis concluded. “This work has the potential to bring closer the vision of predictive medicine, where vast quantities of information are used to predict individuals’ future medical risks in order to improve medical care and diagnosis.”
Glenn Laffel, MD, PhD
Sr. VP, Clinical Affairs, Practice Fusion
