- Researchers previously demonstrated that applying natural language processing to the narrative section of electronic health records can capture information that isn't reflected in diagnostic and treatment codes
- In a retrospective analysis of health records from 535,814 patients, the team demonstrated that a cognitive symptom score derived from natural language processing of discharge summaries can stratify dementia risk up to eight years before diagnosis
- The higher the cognitive symptom score at discharge, the earlier the subsequent diagnosis of dementia
- Individuals in the greatest quartile of cognitive symptom score who survived at least five years after hospitalization were 7.4 times more likely to receive a diagnosis of dementia than those in the lowest quartile
Early detection of dementia symptoms is a cornerstone of disability reduction and a critical step in developing strategies for early intervention. Existing treatments are most useful when started early and efforts to develop new treatments have likewise focused on individuals early in the course of illness. However, existing techniques for risk stratification have not been widely used.
Subscribe to the latest updates from Psychiatry Advances in Motion
Researchers at Massachusetts General Hospital previously demonstrated that applying natural language processing to the narratives in electronic health records (EHRs) may capture useful elements of clinical presentation that aren't necessarily reflected in diagnostic and treatment codes. For example, in a prior paper in Depression & Anxiety, natural language processing of psychiatric symptoms that were captured in narrative notes improved the prediction of suicide risk.
Roy H. Perlis, MD, MSc, director of the Center for Quantitative Health in the Department of Psychiatry, Thomas McCoy, MD, the center's director of research, and colleagues recently extended this research to stratify the risk of dementia. In Alzheimer's & Dementia, they explain that a cognitive symptom score, derived from the narrative text in EHRs, was associated with subsequent dementia risk among hospitalized individuals without a prior diagnosis of dementia.
The researchers developed software to identify relevant symptoms reported in discharge notes of adults who were hospitalized at two large academic medical centers between January 1, 2005, and December 31, 2017. They included all individuals who did not have a dementia diagnosis recorded during the hospitalization or at any prior time.
The team used a validated method of estimating neuropsychiatric symptoms from narrative text in EHRs, previously described in another paper in Biological Psychiatry. Clinical experts developed lists of relevant terms (e.g., attention, memory, behavior), which were subsequently expanded through unsupervised machine learning. The final cognitive symptom score is simply the percentage of these terms that appear in any given note, as determined by a software program.
Cognitive Symptom Score Predicts Dementia Diagnosis
In the first academic medical center, there were 267,855 index hospitalizations, and 6,516 patients (2.4%) received a diagnosis of dementia over eight years of follow-up. The higher the cognitive symptom score at discharge, the earlier the subsequent diagnosis of dementia (HR, 1.63).
These results proved to be generalizable to the second hospital system. Following 267,959 hospitalizations, 3,902 patients received a dementia diagnosis, and the cognitive symptom score was again positively associated with dementia risk (HR, 1.37).
In a pooled analysis on the two hospital systems—535,814 patients representing 2,571,793 patient-years of follow-up—the cognitive symptom score remained significantly associated with the risk of dementia diagnosis (HR, 1.5).
The same pattern was observed in subgroup analyses of other groups, including:
- Patients ≥50 years old
- Patients under 50
- Patients who had a primary discharge diagnosis of stroke
The cognitive symptom score was able to identify a high-risk patient subset. Individuals in the highest quartile of cognitive symptom scores who survived at least five years after hospitalization were 7.4 times more likely to receive a diagnosis of dementia than those in the lowest quartile, and 1% incidence of dementia diagnosis was reached 1,361 days sooner.
Hope for the Future
It appears that up to eight years before a formal diagnosis of dementia, symptoms documented in discharge summaries can identify people who are at high risk. Further evaluation or application of biomarkers may have the greatest yield in such individuals.
Learn more about the Center for Quantitative Health
Explore research in the Mass General Department of Psychiatry