Researchers developed a model to predict one year risk of diagnosis conversion from major depressive disorder to bipolar disorder that was validated with data from multiple countries. Their research was published in Translational Psychiatry.
The investigators sought to hasten practitioners’ recognition of BD.
They collected data from the Observational Medical Outcomes Partnership (OMOP) and trained the model on five databases in the OMOP that included sociodemographic and clinical data of individuals who first had a diagnosis of MDD. The investigators identified whether the patients were diagnosed with BD within one year of their MDD diagnosis. The patients (n=2,687,578 patients) they included were at least 10 years old when they were diagnosed with MDD, which followed at least one year of observation without use of antipsychotic, antidepressant, lithium, or mood-stabilizing anticonvulsant (MSA).
The researchers analyzed data up to 10 years following a diagnosis of MDD.
The investigators externally validated the final prediction model with nine databases including Columbia University Irving Medical Center (CUIMC), Ajou University School of Medicine in South Korea (AUSOM), STAnford medicine Research data Repository (STARR), IQVIA (including France, Belgium, and US records), Japan Medical Data Center database (JMDC), and the US Veterans Health Administration EMR.
Area under the curve (AUC) varied 0.633 to 0.745 for the training model databases.
Performance exceeded random prediction in all validation databases (range 0.570-0.785).
The researchers found that the data suggested patients who were younger, had anxiety, had more severe initial depressive episodes, and had experienced psychotic features during the index depression tended to develop BD within one year. Pregnancy tended to predict a lower risk of transitioning to BD within one year.
Limitations of the study included those related to extracting data from electronic health records and administrative claims, exclusion of lab test results, possible overfitting, and possibility BD type II cases were missed.
“Despite moderate AUC performance, our model can identify patients spanning a 100-fold magnitude difference in risk of MDD to BD transition,” the researchers said.
“Accounting for interactions and nonlinear relationships between the variables using XGBoost [a decision-tree-based ensemble Machine Learning algorithm] did not result in higher AUC values, suggesting that substantial improvements in model performance are unlikely to be gained by more feature engineering of the training databases, though temporal relationships might be explored further. Because one can code and bill for MDD without specifying severity, we anticipate that models that go beyond claims records to incorporate physician notes and results of depression psychometric assessment may lead to more precise predictions of the risk of future conversion.”
Disclosure: Some study authors declared affiliations with biotech, pharmaceutical, and/or device companies. Please see the original reference for a full list of authors’ disclosures.
Nestsiarovich A, Reps JM, Matheny ME, et al. Predictors of diagnostic transition from major depressive disorder to bipolar disorder: a retrospective observational network study. Translational Psychiatry. Published online December 20, 2021. doi:10.1038/s41398-021-01760-6