This study constructs a calibrated computational model that explicitly couples three elements implicated in obsessive-compulsive disorder (OCD) and its pharmacologic treatment: SSRI pharmacokinetics, a basal-ganglia–thalamocortical circuit, and corticostriatal synaptic plasticity. The author frames OCD as an elevated fixed point in the circuit's state space and treats SSRI administration as a perturbation that can move the system across a geometric boundary — specifically, a saddle-node fold (a type of tipping point). The model thus casts treatment response and remission in geometric dynamical terms rather than only as gradual changes in symptom means.
The calibrated model sets approximately 35 constants from prior literature or data and allows two shared parameters to vary across datasets. The circuit and plasticity components are intended to be physiologically informed; the pharmacokinetic element supplies a drug-specific driving input that differs across SSRI drugs and dosing arms in the analyzed trials.
Using published population response data from prior clinical trials, the model was fit to outcomes for six different SSRIs. Across 84 arm–timepoint observations, the model achieved an overall root-mean-square (RMS) error of 0.83 points on the Yale–Brown Obsessive Compulsive Scale (Y-BOCS). In an out-of-sample evaluation, the model predicted a held-out trial with an RMS error of 0.58 Y-BOCS points. These fits were produced with the stated ~35 fixed constants and only two shared free parameters, indicating the model can replicate mean trajectory data across multiple SSRI agents with parsimonious parameterization.
The study used previously published trial results rather than primary patient-level datasets; the author reports that all data used are summarized in supplemental information and that code/data repositories are publicly available (links provided in the manuscript). The work is presented as a preprint and the author emphasizes it has not undergone peer review and should not be used to guide clinical practice.
The author also evaluated a fold-free variant of the model (i.e., a model without the saddle-node bistability). That variant fits the group mean data about as well as the fold model, demonstrating that population mean trajectories alone cannot distinguish whether the underlying system is bistable. The crucial distinction between the two model geometries emerges not in means but in specific individual-level dynamical signatures that are possible only in the bistable (fold) geometry.
This finding emphasizes a general methodological point: reproducing means across trials is necessary but not sufficient evidence for a particular underlying dynamical mechanism. Mechanistic hypotheses involving state transitions and bistability require attention to individual-level temporal signatures and the presence of qualitatively discrete outcomes (e.g., durable off-drug remission) rather than only average symptom change.
The bistable (fold) model predicts two main individual-level signatures that the fold-free model cannot produce:
A discrete, durable off-drug remission subgroup: under the bistable geometry, some individuals cross the fold into a low-symptom attractor and remain in that state after drug withdrawal, producing a qualitatively distinct remission subgroup rather than a continuous distribution of partial improvements.
Critical slowing: as the system approaches the saddle-node fold, its return to baseline after perturbations slows; this dynamical slowing is a hallmark of approaching a tipping point and would be observable in individual temporal response patterns.
The paper notes there is qualitative support for these phenomena in OCD and adjacent disorders, but it also states explicitly that these specific signatures have not yet been tested as formal evidence of bistability in clinical data. Therefore, their presence remains a model-predicted hypothesis pending empirical verification.
Beyond explaining SSRI response and remission, the author proposes that the same bistable geometric framing can reinterpret augmentation strategies such as adding memantine or antipsychotics. In a bistable architecture, augmentation may act by shifting circuit state or plasticity parameters so as to facilitate crossing the fold into remission, or to stabilize the low-symptom attractor. The manuscript outlines this as a conceptual reframing rather than as empirically validated recommendations: the claims are derived from model geometry and fit to aggregated trial data, not from new randomized augmentation trials.
This reframing foregrounds which patients are likely to remit or relapse based on individual trajectories relative to the fold, rather than inference from group mean responses alone.
Key limitations reported in the manuscript include reliance on published aggregate trial outcomes rather than primary patient-level longitudinal data, and the fact that fold-specific individual-level signatures remain untested in clinical datasets. The author reports that all necessary metadata were obtained from published trials and that supplemental information summarizes the data used. Code and data links are provided in the manuscript's supplemental materials and an associated GitHub repository.
Finally, the paper is a medRxiv preprint and the author reiterates that it has not been peer reviewed; the manuscript explicitly states it should not be used to guide clinical practice. Competing interests were declared as none.