Baseline medication exposure emerges as a prognostic variable in clinical high-risk psychosis
In clinical high-risk for psychosis populations, the first prescription on record may say as much about prognosis as the symptom checklist in the chart. Andrea Raballo, Michele Poletti and Antonio Preti argue that baseline pharmacological exposure should no longer be treated as a nuisance variable to be adjusted away and forgotten, but as a clinically meaningful marker embedded in the patient’s longitudinal story.
The premise is straightforward and unsettling at the same time. At the point of CHR-P identification, some patients are already taking antipsychotics, antidepressants, benzodiazepines, or combinations of these agents. Those exposures are rarely random. They may reflect prior symptom escalation, comorbid anxiety or depression, partial response to earlier treatment, clinician concern, help-seeking pathways, or a trajectory in which emerging psychosis has already altered behavior enough to trigger treatment before formal risk assessment.
One sentence captures the practical problem: medication history can confound causal inference and still carry prognostic signal.
Why baseline pharmacotherapy cannot be treated as statistical noise
The article frames three familiar biases that complicate interpretation of baseline medication use: confounding by indication, confounding by severity, and protopathic bias. Each can distort inference if the goal is to claim that a drug caused a downstream outcome. Yet those same forces may also encode the underlying clinical state. A patient who arrives at a CHR-P clinic already receiving an antipsychotic has not simply been exposed to a drug; that exposure may indicate a more advanced or more unstable clinical course, or a care pathway in which psychotic symptoms were already prominent enough to prompt treatment.
This is where prognostic modeling becomes more complicated than many current studies allow. Demographic variables, symptom scales, cognitive measures, and functional status remain essential, but they do not fully capture the treatment context in which CHR-P is identified. Baseline pharmacotherapy can add a layer of temporal information that is clinically interpretable and immediately available in routine care. Dose, duration, indication, adherence, and later treatment changes may further separate patients who look similar on paper but are not clinically equivalent.
The article’s central clinical message is not that medications should be rebranded as causal predictors. It is that they should be recognized as part of the risk phenotype.
What medication exposure may be signaling in early psychosis pathways
In CHR-P settings, medication use often reflects more than prescribing behavior. Antipsychotic exposure may indicate attenuated psychotic symptoms severe enough to merit intervention, a previous emergency evaluation, or a clinician’s concern that the trajectory is already drifting toward sustained psychosis. Antidepressant treatment may point to affective comorbidity, while benzodiazepine use may track anxiety, agitation, insomnia, or acute distress. Combined exposure suggests a more complex clinical picture, one in which overlapping syndromes and treatment attempts blur the line between prevention and active management.
From a pathophysiological standpoint, this matters because psychosis does not unfold in a vacuum. Neurodevelopmental vulnerability, stress reactivity, sleep disruption, affective dysregulation, substance use, and functional decline can converge long before a frank psychotic disorder is diagnosed. Pharmacotherapy may therefore act as a marker of disease burden that has already begun to organize the clinical presentation. If emerging psychosis has been partially suppressed, obscured, or modified by treatment, the apparent baseline state may understate risk rather than eliminate it.
Prognostic modeling needs treatment history, not just symptom snapshots
Current prognostic models often privilege static variables collected at one time point. That approach can be efficient, but it is not always faithful to the actual course of illness. The authors argue that incorporating medication exposure at CHR-P inception, and then tracking treatment changes over time, could reduce hidden heterogeneity and improve the identification of prognostically non-equivalent subgroups.
This is especially relevant when preventive interventions are being evaluated. If two CHR-P cohorts differ substantially in baseline pharmacotherapy, they may not share the same underlying risk architecture even if their demographic and symptom profiles look similar. A trial or observational study that ignores that difference risks blending together patients at very different stages of clinical evolution. The result is diluted signal, noisier effect estimates, and weaker inference about who benefits from what.
Medication history also has bedside implications. A clinician seeing baseline antipsychotic exposure in a young person labeled CHR-P should think carefully about whether the patient truly sits at the same stage of illness as an untreated peer with comparable attenuated symptoms. The distinction is not semantic. It changes the prior probability of transition, the intensity of follow-up, and the interpretation of apparent improvement or deterioration.
Causal humility, prognostic pragmatism
The article lands on a balanced position: baseline pharmacotherapy should be interpreted with causal humility but prognostic pragmatism. That phrasing fits the reality of early psychosis research, where the same variable can be both a biasing factor and an informative marker. Ignoring treatment history because it muddies causal analyses would discard clinically relevant data. Treating it as a clean causal exposure would go too far in the opposite direction.
A better approach is more disciplined. Medication exposure should be recorded alongside indication, dose, duration, adherence, and longitudinal changes, then incorporated into prognostic frameworks rather than relegated to a footnote. In CHR-P research, the patient’s prior treatment trajectory may be one of the clearest clues to where the illness has already been and where it may be heading.
The article does not provide a new trial, a diagnostic threshold, or a validated algorithm. Its force lies elsewhere: it reframes baseline pharmacotherapy as a clinically meaningful signal in a population already defined by uncertainty. The next step will be whether prognostic models that account for medication history can separate true risk from treatment-shaped appearance with enough precision to change care.