Prediction of individual treatment allocation between electroconvulsive therapy or ketamine using the Personalized Advantage Index.
Study Goal
The researchers aimed to develop a predictive model using EHR data to determine whether ECT or ketamine would be the more effective treatment for individual patients with depression.
Results Summary
The study found that patients who received their predicted optimal treatment (either ECT or ketamine) had significantly lower minimum QIDS scores compared to those who received the non-optimal treatment, suggesting personalized treatment selection improves outcomes. The model identified pretreatment factors to guide antidepressant treatment selection.
Population
Patients with depression (2506 ECT and 196 ketamine patients, with propensity score matching yielding 196 per treatment).
Effective Dosage
Not specified
Duration
Acute treatment period (exact duration not specified)
Interactions
None mentioned
| Intervention | Direction | Endpoint | Population | Dosage | Impact | Claim # |
|---|---|---|---|---|---|---|
Electroconvulsive therapy (ECT) | decrease | depression | - | - | effective treatments | #1 |
ketamine | decrease | depression | - | - | effective treatments | #2 |
Personalized Advantage Index (PAI) using machine learning | neutral | optimal treatment assignment to ECT or ketamine | 2506 ECT and 196 ketamine patients | - | predicted optimal treatment assignment | #3 |
predicted optimal treatment | decrease | minimum QIDS scores (min-QIDS) | Patients with large PAI scores | mean difference = 1.19 [95% CI: 0.32, ∞], t = 2.25, q < 0.05, d = 0.26 | significantly lower | #4 |
Electroconvulsive therapy (ECT) and ketamine are effective treatments for depression; however, evidence-based guidelines are needed to inform individual treatment selection. We adapted the Personalized Advantage Index (PAI) using machine learning to predict optimal treatment assignment to ECT or ketamine using EHR data on 2506 ECT and 196 ketamine patients. Depressive symptoms were evaluated using the Quick Inventory of Depressive Symptomatology (QIDS) before and during acute treatment. Propensity score matching across treatments was used to address confounding by indication, yielding a sample of 392 patients (n = 196 per treatment). Models predicted differential minimum QIDS scores (min-QIDS) over acute treatment using pretreatment EHR measures and SHAP values identified prescriptive predictors. Patients with large PAI scores who received a predicted optimal had significantly lower min-QIDS compared to the non-optimal treatment group (mean difference = 1.19 [95% CI: 0.32, ∞], t = 2.25, q < 0.05, d = 0.26). Our model identified candidate pretreatment factors to provide actionable, effective antidepressant treatment selection guidelines.