This topic follows our gene-environment interactions theme. The “environmental signal” is the type of treatment, and the “response” (undoubtedly reflected in the expressivity of hundreds of genes) is “clinical outcome” in the patient. Research on “how to predict treatment outcome” for major depression disorder (MDD) patients has largely been carried out in randomized clinical trials involving strict standardization of treatments, stringent patient exclusion criteria, and careful selection and supervision of study clinicians. The extent — to which findings from such studies can be generalized to realistic psychiatric settings — is unclear. Authors [see attached article] sought to predict MDD outcomes for patients seeking treatment, within an intensive psychiatric-hospital setting, while comparing the performances of a range of 13 machine-learning approaches.
MDD patients (N = 484; ages 18–72; 89% Caucasian) — receiving treatment within a psychiatric-hospital program delivering pharmacotherapy and cognitive-behavioral therapy — were divided into a “training sample” and a “holdout sample”. Within the training sample, 51 pretreatment variables were submitted to 13 machine-learning algorithms to predict (via cross-validation) posttreatment (Patient Health Questionnaire–9) depression scores. The best performing modeling-approach (lowest mean-squared error; MSE) from the training sample was then selected to predict outcome in the holdout sample.
The best performing model in the training sample was discovered to be “elastic net regularization” (ENR; MSE = 20.5, R2 = 0.28) — which then was found to have a comparable performance in the holdout sample (MSE = 11.26; R2 = 0.38). Authors found 14 pretreatment variables that predicted outcome. To demonstrate “translation of the ENR model” to personalized prediction of treatment outcome, a patient-specific prognosis calculator is presented [see article for details]. Informed by pretreatment patient characteristics, such predictive models could be used to communicate prognosis to clinicians and to guide treatment planning. Identified predictors of poor prognosis might suggest important targets for drug intervention.
Knowledge of which patients are likely to exhibit a poor outcome may have important clinical implications regarding treatment recommendations (e.g. a more intensive, alternative or combination treatment) and can inform more careful symptom and treatment progress monitoring. In the present study, authors {see attached article] used machine-learning to develop predictions of treatment outcome for MDD patients seeking treatment in a “real-world” psychiatric hospital clinic. A prognosis calculator was shown to be successfully developed, which generates personalized predictions of treatment outcome for each individual MDD patient. 😊
DwN
J Consult Clin Psychol Jan 2020; 88: 25-38