Multiple treatments are effective for major depressive disorder (MDD), but the outcomes of each treatment vary broadly among individuals. Accurate prediction of outcomes is needed to help select a treatment that is likely to work for a given person. We aim to examine the performance of machine learning methods in delivering replicable predictions of treatment outcomes.

Of 7732 non-duplicate records identified through literature search, we retained 59 eligible reports and extracted data on sample, treatment, predictors, machine learning method, and treatment outcome prediction. A minimum sample size of 100 and an adequate validation method were used to identify adequate-quality studies. The effects of study features on prediction accuracy were tested with mixed-effects models. Fifty-four of the studies provided accuracy estimates or other estimates that allowed calculation of balanced accuracy of predicting outcomes of treatment.

Eight adequate-quality studies reported a mean accuracy of 0.63 [95% confidence interval (CI) 0.56–0.71], which was significantly lower than a mean accuracy of 0.75 (95% CI 0.72–0.78) in the other 46 studies. Among the adequate-quality studies, accuracies were higher when predicting treatment resistance (0.69) and lower when predicting remission (0.60) or response (0.56). The choice of machine learning method, feature selection, and the ratio of features to individuals were not associated with reported accuracy.

The negative relationship between study quality and prediction accuracy, combined with a lack of independent replication, invites caution when evaluating the potential of machine learning applications for personalizing the treatment of depression.

Pharmacogenomic testing has emerged to aid medication selection for patients with major depressive disorder (MDD) by identifying potential gene-drug interactions (GDI). Many pharmacogenomic tests are available with varying levels of supporting evidence, including direct-to-consumer and physician-ordered tests. We retrospectively evaluated the safety of using a physician-ordered combinatorial pharmacogenomic test (GeneSight) to guide medication selection for patients with MDD in a large, randomized, controlled trial (GUIDED).

Patients diagnosed with MDD who had an inadequate response to ≥1 psychotropic medication were randomized to treatment as usual (TAU) or combinatorial pharmacogenomic test-guided care (guided-care). All received combinatorial pharmacogenomic testing and medications were categorized by predicted GDI (no, moderate, or significant GDI). Patients and raters were blinded to study arm, and physicians were blinded to test results for patients in TAU, through week 8. Measures included adverse events (AEs, present/absent), worsening suicidal ideation (increase of ≥1 on the corresponding HAM-D17 question), or symptom worsening (HAM-D17 increase of ≥1). These measures were evaluated based on medication changes [add only, drop only, switch (add and drop), any, and none] and study arm, as well as baseline medication GDI.

Most patients had a medication change between baseline and week 8 (938/1,166; 80.5%), including 269 (23.1%) who added only, 80 (6.9%) who dropped only, and 589 (50.5%) who switched medications. In the full cohort, changing medications resulted in an increased relative risk (RR) of experiencing AEs at both week 4 and 8 [RR 2.00 (95% CI 1.41–2.83) and RR 2.25 (95% CI 1.39–3.65), respectively]. This was true regardless of arm, with no significant difference observed between guided-care and TAU, though the RRs for guided-care were lower than for TAU. Medication change was not associated with increased suicidal ideation or symptom worsening, regardless of study arm or type of medication change. Special attention was focused on patients who entered the study taking medications identified by pharmacogenomic testing as likely having significant GDI; those who were only taking medications subject to no or moderate GDI at week 8 were significantly less likely to experience AEs than those who were still taking at least one medication subject to significant GDI (RR 0.39, 95% CI 0.15–0.99, p=0.048). No other significant differences in risk were observed at week 8.

These data indicate that patient safety in the combinatorial pharmacogenomic test-guided care arm was no worse than TAU in the GUIDED trial. Moreover, combinatorial pharmacogenomic-guided medication selection may reduce some safety concerns. Collectively, these data demonstrate that combinatorial pharmacogenomic testing can be adopted safely into clinical practice without risking symptom degradation among patients.

Myriad Neuroscience/Assurex Health

Recently, artificial intelligence-powered devices have been put forward as potentially powerful tools for the improvement of mental healthcare. An important question is how these devices impact the physician-patient interaction.

Aifred is an artificial intelligence-powered clinical decision support system (CDSS) for the treatment of major depression. Here, we explore the use of a simulation centre environment in evaluating the usability of Aifred, particularly its impact on the physician–patient interaction.

Twenty psychiatry and family medicine attending staff and residents were recruited to complete a 2.5-h study at a clinical interaction simulation centre with standardised patients. Each physician had the option of using the CDSS to inform their treatment choice in three 10-min clinical scenarios with standardised patients portraying mild, moderate and severe episodes of major depression. Feasibility and acceptability data were collected through self-report questionnaires, scenario observations, interviews and standardised patient feedback.

All 20 participants completed the study. Initial results indicate that the tool was acceptable to clinicians and feasible for use during clinical encounters. Clinicians indicated a willingness to use the tool in real clinical practice, a significant degree of trust in the system's predictions to assist with treatment selection, and reported that the tool helped increase patient understanding of and trust in treatment. The simulation environment allowed for the evaluation of the tool's impact on the physician–patient interaction.

The simulation centre allowed for direct observations of clinician use and impact of the tool on the clinician–patient interaction before clinical studies. It may therefore offer a useful and important environment in the early testing of new technological tools. The present results will inform further tool development and clinician training materials.

The Genomics Used to Improve DEpresssion Decisions (GUIDED) trial assessed outcomes associated with combinatorial pharmacogenomic (PGx) testing in patients with major depressive disorder (MDD). Analyses used the 17-item Hamilton Depression (HAM-D17) rating scale; however, studies demonstrate that the abbreviated, core depression symptom-focused, HAM-D6 rating scale may have greater sensitivity toward detecting differences between treatment and placebo. However, the sensitivity of HAM-D6 has not been tested for two active treatment arms. Here, we evaluated the sensitivity of the HAM-D6 scale, relative to the HAM-D17 scale, when assessing outcomes for actively treated patients in the GUIDED trial.

Outpatients (N=1,298) diagnosed with MDD and an inadequate treatment response to >1 psychotropic medication were randomized into treatment as usual (TAU) or combinatorial PGx-guided (guided-care) arms. Combinatorial PGx testing was performed on all patients, though test reports were only available to the guided-care arm. All patients and raters were blinded to study arm until after week 8. Medications on the combinatorial PGx test report were categorized based on the level of predicted gene-drug interactions: ‘use as directed’, ‘moderate gene-drug interactions’, or ‘significant gene-drug interactions.’ Patient outcomes were assessed by arm at week 8 using HAM-D6 and HAM-D17 rating scales, including symptom improvement (percent change in scale), response (≥50% decrease in scale), and remission (HAM-D6 ≤4 and HAM-D17 ≤7).

At week 8, the guided-care arm demonstrated statistically significant symptom improvement over TAU using HAM-D6 scale (Δ=4.4%, p=0.023), but not using the HAM-D17 scale (Δ=3.2%, p=0.069). The response rate increased significantly for guided-care compared with TAU using both HAM-D6 (Δ=7.0%, p=0.004) and HAM-D17 (Δ=6.3%, p=0.007). Remission rates were also significantly greater for guided-care versus TAU using both scales (HAM-D6 Δ=4.6%, p=0.031; HAM-D17 Δ=5.5%, p=0.005). Patients taking medication(s) predicted to have gene-drug interactions at baseline showed further increased benefit over TAU at week 8 using HAM-D6 for symptom improvement (Δ=7.3%, p=0.004) response (Δ=10.0%, p=0.001) and remission (Δ=7.9%, p=0.005). Comparatively, the magnitude of the differences in outcomes between arms at week 8 was lower using HAM-D17 (symptom improvement Δ=5.0%, p=0.029; response Δ=8.0%, p=0.008; remission Δ=7.5%, p=0.003).

Combinatorial PGx-guided care achieved significantly better patient outcomes compared with TAU when assessed using the HAM-D6 scale. These findings suggest that the HAM-D6 scale is better suited than is the HAM-D17 for evaluating change in randomized, controlled trials comparing active treatment arms.

Assurex Health, Inc.

Major depressive disorder is associated with significant impairment in occupational functioning and reduced productivity, which represents a large part of the overall burden of depression.

To examine symptom-based and work functioning outcomes with combined pharmacotherapy and psychotherapy treatment of major depressive disorder.

Employed patients with a DSM-IV diagnosis of major depressive disorder were treated with escitalopram 10–20mg/day and randomised to: (a) telephone-administered cognitive-behavioural therapy (telephone CBT) (n = 48); or (b) adherence-reminder telephone calls (n = 51). Outcomes included the Montgomery-åsberg Depression Rating Scale (MADRS), administered by masked evaluators via telephone, and self-rated work functioning scales completed online. (Registered at clinicaltrials.gov: NCT00702598.)

After 12 weeks, there were no significant between-group differences in change in MADRS score or in response/remission rates. However, participants in the telephone-CBT group had significantly greater improvement on some measures of work functioning than the escitalopram-alone group.

Combined treatment with escitalopram and telephone- administered CBT significantly improved some self-reported work functioning outcomes, but not symptom-based outcomes, compared with escitalopram alone.