Source/disclosure information
Development of a machine learning model for estimating PHQ-9 scores using clinical records from Mercy CD and other real-world data sources. Place of publication: American Society of Clinical Psychopharmacology. May 29-June 2, 2023. miami beach.
Disclosure:
Marci has not reported any additional relevant financial disclosures.
Important points:
- Machine learning models processed data from over 32,000 records containing 96,000 encounters of MDD patients.
- Application of the model produced an exponential amount of PHQ-9 scores and patient records.
MIAMI BEACH, Fla. — A new machine learning model accurately estimated depression questionnaire scores from complete and partial clinical notes, according to a poster at the American Society of Clinical Psychopharmacology Annual Meeting.
“In the real world, there is a lot of missing data, [Patient Health Questionnaire-9] Scores are commonly used and patients record them, but not everyone uses them every time. ” curl D. Dr. Mercy, Chief Psychiatrist and Managing Director of Mental Health and Neuroscience at OM1 Inc., a Boston-based healthcare technology company, told Healio. “We took a data science approach to see if we could take the psychiatrist’s notes and have a computer analyze them to generate a score.”
Marci et al. applied a new machine learning method to provide accurate estimated patient health questionnaire-9 scores from a set of unsorted and partially sorted clinical records for depressed patients.
They utilized data extracted from the OM1 PremiOM Major Depressive Disorder dataset. This dataset contains information on more than 490,000 MDD patients undergoing treatment nationwide.
Patients with MDD who recorded both the PHQ-9 score and some type of clinical record (32,802 records, 96,891 patient encounters) were included in the training cohort or validation cohort (15,792 records, 46,333 encounters) to generate the model. patient encounters). ).
The researchers employed the area under the receiver operating characteristic curve (AUC) to assess the performance of each model, a common Spearman and We used the continuous electronic PHQ-9 score, which was assessed using the Pearson R-value.
According to the results, applying the model across the original dataset generated ePHQ-9 scores for over 2.2 million patient encounters. This represented 2.7 times his PHQ-9 score of his 814,166 recorded, also generated for 208,692 patient encounters. This represents a 1.2-fold increase from her original 174,897 with a PHQ-9 score.
“We’re doing studies with real-world datasets like this, and once we start defining cohorts, we’re much more likely to hit our clinical endpoints,” said Marci.
