A novel artificial intelligence-based methodology to predict non-specific response to treatment.

Area of Science:

• Psychiatry and Behavioral Sciences
• Computational Neuroscience
• Biostatistics

Background:

• Non-specific response to treatment (NSRT) significantly hinders the success of randomized clinical trials (RCTs) for major depressive disorder (MDD).
• Accurate prediction of individual NSRT is crucial for improving the reliability of treatment effect (TE) estimates in clinical trials.

Purpose of the Study:

• To develop artificial neural network (ANN) models for predicting individual probabilities of NSRT.
• To evaluate the utility of these ANN-derived NSRT probabilities in enhancing TE estimation within MDD RCTs.

Main Methods:

• Utilized pre-randomization data from a completed MDD trial to train ANN models predicting NSRT probability (prob-NSRT) in placebo-assigned subjects.
• Calculated NSRT propensity scores as the inverse of individual prob-NSRT.
• Applied mixed-effects models with propensity score weighting (PSW) to assess TE, comparing results with conventional analyses.

Main Results:

• ANN models successfully predicted individual NSRT probabilities.
• PSW analysis yielded a significantly larger estimate of TE compared to conventional analyses.
• The PSW methodology demonstrated enhanced signal detection of TE by adjusting for inter-individual variability in NSRT.

Conclusions:

• Propensity score weighting, informed by ANN-predicted NSRT probabilities, offers a promising approach to improve TE detection in MDD RCTs.
• This methodology can enhance the analysis of clinical trials by accounting for patient-specific non-specific treatment responses.
• External validation of the developed ANN models is necessary before widespread clinical or regulatory adoption.