Treating autism with Bumetanide: Identification of responders using Q-Finder machine learning algorithm

Area of Science:

• Neuroscience
• Pharmacology
• Computational Biology

Background:

• Bumetanide, an NKCC1 inhibitor, shows promise for Autism Spectrum Disorders (ASD) by restoring GABAergic inhibition.
• Phase 2 trials demonstrated Bumetanide's efficacy in improving ASD symptoms.
• Large phase 3 trials failed to show overall efficacy, possibly due to ASD's heterogeneity.

Purpose of the Study:

• To reanalyze phase 3 clinical trial data using machine learning to identify responder subgroups in ASD patients treated with Bumetanide.
• To investigate if a precision medicine approach can uncover treatment benefits missed in large, heterogeneous trials.

Main Methods:

• Utilized Q-Finder, a supervised machine learning algorithm, on baseline clinical data from phase 3 trials.
• Applied the same standard endpoints and success criteria as the original phase 3 protocol.
• Cross-validated identified responder subgroups between two distinct study populations.

Main Results:

• Identified statistically significant responder subgroups within the phase 3 data, showing a positive response to Bumetanide.
• These responder subgroups represented up to 40% of the total participants.
• Findings were consistent across both phase 3 study populations.

Conclusions:

• Machine learning can identify meaningful treatment responses in heterogeneous conditions like ASD, even within negative large-scale trials.
• A precision medicine strategy, aided by machine learning, is crucial for uncovering subgroup-specific efficacy.
• This approach highlights the limitations of a one-size-fits-all treatment model for ASD.