Topics:

Mining clinical notes could detect harmful drug reactions earlier

Tools

A team of researchers at the Stanford University School of Medicine has developed a data mining method that could divulge important information about possible drug side effects using records of interactions between doctors and patients. The new system might be able to detect drug side effects years before the FDA issues an official alert, a Stanford study found.

The computerized method, developed by assistant professor of medicine Nigam Shah and his colleagues, is designed to sift through the contents of clinical notes in electronic medical records, as opposed to insurance claims, lab tests or medical codes, which researchers say reveal only part of a patient's record. Researchers decided to look into clinical notes because those include information about what a doctor advises or prescribes for the patient as well as what a caregiver dictates into a patient's record, such as the patient's symptoms or other medical issues. Using the data mining technology, researchers mined about 10 million notes for about 1.8 million patients over 15 years to examine how often specific drugs and diseases were mentioned.

The idea is that the technology could eventually complement the FDA's Adverse Event Reporting System, a voluntary system that has collected information from healthcare professionals and consumers about drugs on the market since 1968.

The new application allows researchers to process 11 million clinical notes in about 7 hours on hardware comparable to a laptop computer by grouping content into "ontologies"--information graphs organized by associative relationships instead of a rigid linear structure. Data from clinical notes could potentially be more valuable than information gleaned from FDA's AER system because it's generated from what is observed and recorded in the hospital or doctor's office and doesn't rely on voluntary reporting.

The study authors believe this new system for mining previously untapped data could be used by healthcare providers and clinical data warehouses alike to help improve patient care, but they point out that it does have limitations. First, it requires a big database to accurately recognize trends, and the sheer volume of information the system processes makes it better suited to pick out common events, such as heart attacks, whereas the FDA reporting system is likely still better for looking at rare diseases or medical events. The system is also incapable of evaluating adverse drug reactions that are dose dependent.

- here's the Stanford University School of Medicine article
- read the study

Related Articles:
Mining data on FDA-OK'd drugs leads to potential cancer treatment
Selventa and Linguamatics team on mining details in journals
Algorithms mine articles to predict bad drug reactions