LONGITUDINAL PROFILING OF TREATMENT TRAJECTORIES IN SSRI-ANCHORED MDD USING AGENTIC AI

Background

The longitudinal characterization of treatment trajectories in Major Depressive Disorder (MDD) is critical for navigating the clinical heterogeneity of SSRIanchored populations1, yet conventional metrics often fail to capture the dynamics of patient response. This gap results in discordance, where residual deficits in sleep, energy, and executive function persist despite standard-of-care intervention2. Traditional episodic clinical assessments lack the temporal resolution to capture these dynamics. This longitudinal population study leverages Headlamp Health’s Lumos AITM agentic platform, to identify multi-factorial biomarker signatures of responders and non-responders.

Objective

To identify clinically relevant patient phenotypes in an SSRI-anchored MDD cohort using a Lumos AITM analysis of longitudinal patient-reported outcome and medication data.

Methods

This observational study utilized the Headlamp Health platform, Lumos AI™, to evaluate behavioral trajectories in patients with a SSRI backbone. From an initial cohort of 705 patients with verified medication continuity, 360 met the analytical threshold of ≥6 days of active behavioral data . The Lumos AI temporal correlation engine defined “treatment episodes” by identifying continuous usage, where gaps of ≥30 days were classified as treatment cessation. By analyzing these episodes, the engine identified seven discrete response patterns, which were aggregated into four clinical phenotypes: Energy and Mood Responders, Worsening from a Good Baseline, Stable Functioning with Emotional Dampening, and Cognition Rebound with Sleep Disruption. These phenotypes were then linked to more than 22,000 available patient biomarker data points.

Results

Lumos AI™ consolidated patient trajectories into four distinct phenotypes: Energy/Mood phenotype exhibited significant energy and mood gains with stable cognition; Worsening Baseline phenotypes showed a systemic decline in energy, mood, and cognitive efficiency; Stable/Dampened phenotype saw functional improvements in energy, sleep, and cognition paired with declining pleasantness; and Cognitive Rebound phenotype featured sharp cognitive restoration alongside acute sleep disruption with stagnant mood/energy. Analysis showed that outcomes were gated by the patient’s biosignature profile rather than diagnosis alone. In younger, low-inflammation cohorts the Aminoketone Advantage leverages a receptive neuro-receptor environment to drive therapeutic gains. Conversely, CNS Stimulant Failure in older cohorts hit a metabolic ceiling where elevated CRP suggested pro-inflammatory cytokines can desensitize receptors, blunting drug efficacy. These two phenotypes with similar diagnoses, baseline characteristics, and an SSRI backbone showed different outcomes, suggesting that direct catecholaminergic stimulation in this subgroup may contribute to diminishing benefit over time.

Conclusions

This population-scale analysis demonstrates that longitudinal patient phenotyping can resolve distinct treatment trajectories in MDD that are not apparent from diagnosis alone. Utilizing Lumos AI™ to synthesize real-world medication episodes, we identified divergent outcomes between CNS stimulant and aminoketone associated biotypes. This analysis suggests that pro-inflammatory baseline states may be associated with reduced treatment effects under direct stimulation, with tolerability varying by age and baseline characteristics. Lumos AI identified baseline cognition and inflammatory markers as high-fidelity biomarkers for characterizing response variability in SSRI-anchored regimens.