Many sources of patient data are in an unstructured text format, in free text buried in electronic heath records (EHRs). For example, there is key information hidden in free-text Myocardial Perfusion Imaging (MPI) reports that can help isolate the patients at higher risk of acute myocardial infarction or death. However, manual review of these reports is laborious and time consuming. Natural language processing can help extract key concepts from these reports in a highly scalable and repeatable fashion.
Why use NLP to abstract results from MPI reports?
Approximately 15-20 million MPI scans are performed worldwide. It is known that abnormal MPI reports are associated with a higher likelihood of cardiac diseases. It therefore becomes important to be able to extract certain key concepts from these reports which can aid in isolating patients with abnormal reports. This can then allow for close monitoring of these patients. NLP is an artificial intelligence (AI) technology that can transform unstructured free-text into normalized structured data with a high degree of scalability and repeatability. It is therefore an excellent fit when a large number of reports need to be reviewed to extract the same set of concepts.
Abstraction of free-text MPI reports to inform patient care
A recent publication by Zheng et al discusses the use of NLP to automatically abstract free-text MPI reports. This study was performed at Kaiser Permanente Southern California (KPSC). Their main objective was to identify patients who are at high risk for Acute Coronary Syndrome (ACS). They found that the results obtained from the NLP algorithm achieved a higher sensitivity and specificity and were in good agreement with those derived from manual review by physicians.
Their study population included 16,957 high risk patients, who have more cardiovascular-related comorbidities and potentially a much higher rate of abnormal MPI findings. They built queries in the Linguamatics NLP platform to extract information such as test type, cardiovascular defects, artifacts, wall motion and ejection fraction (EF) from the MPI reports. The extracted information was then used to categorize patients into the following four groups: 1. Ischemia (ischemic or reversible defect identified), 2. Infarction (no definitive ischemic finding but a fixed or irreversible defect identified), 3. Non-diagnostic ischemia (ischemia or infarction cannot be ruled out due to the presence of artifacts or sub-optimal test quality) and 4. Normal. The authors assessed the NLP query results and found both high sensitivity (96.7%) and specificity (98.9%) for the MPI categorical results.
Using the NLP abstracted results, they were able to identify patients with short-term cardiac risk. These patients have a much higher 30-day cardiac event rates and their MPI abnormalities were worsening. The authors state: “NLP is an accurate and efficient strategy to abstract results from the free-text MPI reports”. The researchers have also used NLP to extract other cardiovascular variables such as Ejection Fraction, Aspirin and Warfarin usages as well as extracting clinical variables from the Electrocardiogram Treadmill Test (ETT) reports.
In addition, NLP has also been previously used to abstract ECG reports as well as patient stratification for heart risk. We look forward to seeing NLP being applied to this area in clinical operations as well as many other clinical areas to help advance patient care.