Limited Validity in Clinical Prediction Tools for Major Depression Treatment

Despite the availability of 16 clinical prediction instruments for use in major depressive disorder (MDD), all identified studies exhibit a high risk for bias, particularly in outcome determination and data analysis, limiting their current clinical utility, according to a review published in the Journal of Affective Disorders.

Researchers conducted a systematic review to identify and evaluate diagnostic and prognostic clinical prediction instruments developed for the treatment of adults with MDD, with a focus on tools currently available for real-world clinical use. They also assessed the instruments’ validation status and potential biases to guide clinicians in informed decision-making. Following PRISMA guidelines, the researchers searched MEDLINE, PsycINFO, and Embase for studies published between January 2010 and March 2023. Eligible studies included development or validation of clinical prediction instruments (excluding symptom severity scales or individual predictors) for adult patients with MDD.

Out of 5879 unique records screened, 15 studies met eligibility criteria, reporting on 16 distinct clinical prediction instruments for MDD. Six instruments were diagnostic, 6 prognostic, and 4 combined both functions. Of the 11 development studies, only 3 underwent external validation. Additionally, 4 of the instruments were designed to assess relapse risk and 2 were designed to be completed by the patient.

Studies included both outpatient and mixed patient populations, with sample sizes ranging from 76 to 3,695.

Among diagnostic tools, the Decision Tool Unipolar Depression (DTUD) achieved a sensitivity of 0.67 and specificity of 0.83. Prognostic instruments such as the Multivariable Prediction Model reported area under the curves (AUCs) of 0.73–0.82 for predicting remission.

Combined tools like The Dutch Measure for quantification of Treatment Resistance in Depression (DM-TRD) showed significant associations between treatment resistance and clinical course (Akaike information criterion [AIC] 26,847.2 vs 26,970.9 without TRD). However, all studies demonstrated high bias risk, primarily due to flaws in outcome determination and data analysis. Applicability concerns were noted in 42% of assessments, mainly due to non-representative patient samples. Only 1 study used rigorous methods such as multiple imputation to address missing data, suggesting limited generalizability and reliability of most instruments for routine clinical use.

The study authors concluded, “Given the importance of clinical characteristics and information on previously unsuccessful treatments in decision-making and prognosis, we recommend instruments that assess both aspects – such as the MSM and the DM-TRD – for use in diagnostic assessments or at the initiation of new treatment.”

Study limitations include the heterogeneity among included prediction instruments limiting comparability, a high risk of bias in all studies affecting reliability, and the exclusion of pre-2010 studies potentially omitting relevant instruments.

This article originally appeared on Psychiatry Advisor