From bias to signal: baseline medication carries prognostic meaning in clinical high-risk for psychosis

Researchers at the Università della Svizzera italiana in Lugano have proposed a new way of interpreting a variable that most studies of early psychosis try to eliminate rather than understand: the medications that young people are already taking when they first come to clinical attention. In a perspective article published in the journal Schizophrenia, psychiatrist Andrea Raballo and colleagues Antonio Preti and Michele Poletti argue that baseline pharmacotherapy in people at Clinical High Risk for Psychosis, often abbreviated CHR-P, should be treated as a meaningful clinical signal rather than dismissed as statistical noise. Their argument challenges a deeply rooted habit in psychiatric research, where medication exposure at study entry is usually handled as a confounder to be controlled, adjusted, or excluded.

The CHR-P state describes a clinical condition in which help-seeking young people show attenuated or brief psychotic-like symptoms, combined with a deterioration in functioning that suggests something serious may be developing. Not everyone in this state goes on to develop a full psychotic disorder, and the field has struggled for decades with the same problem: the transition rate from high risk to psychosis has fallen in more recent cohorts, and the heterogeneity of outcomes makes it difficult to know who genuinely needs intensive intervention. Current prediction models rely heavily on demographic details, cognitive performance, symptom severity, and social functioning, but they tend to ignore one piece of information that is available in almost every clinical encounter: what the person was already taking before the assessment began.

Raballo and colleagues argue that this omission reflects a misunderstanding of what baseline medication actually represents. When a young person arrives at a specialized early detection clinic already prescribed an antidepressant, an antipsychotic, a benzodiazepine, or some combination of these agents, that prescription history is not a random artifact. It encodes a sequence of clinical decisions made by earlier treating clinicians, reflecting the severity of previous symptoms, the presence of comorbid conditions such as anxiety or depression, the pathways through which the person entered care, and the degree to which previous treatments helped or failed. In other words, the medication record is a condensed biography of the clinical problem, written in the language of pharmacology.

The researchers acknowledge that this signal is entangled with a well-known methodological problem called confounding by indication. People are prescribed medication because they are unwell, so any apparent relationship between taking a drug and a later outcome could simply reflect the fact that sicker people receive more treatment. This is the standard argument for treating baseline medication as a nuisance variable, and it is statistically correct as far as it goes. But Raballo and colleagues push back against the conclusion that follows from it. If confounding by indication means that medication exposure is contaminated, it also means that medication exposure is informative, because it carries information about the very severity that drives the confounding in the first place.

The perspective article suggests that the field should stop asking whether baseline medication is a bias and start asking what it tells us about the clinical trajectory. A young person prescribed a low dose of an antipsychotic before referral to a high-risk service is not the same kind of clinical case as someone who arrives medication-free, even if both meet the same formal research criteria for CHR-P. The first has already been recognized as troubled enough to warrant pharmacological intervention, has already been through a treatment relationship, and may already have shown some degree of response or non-response. The second may be earlier in the course of the problem, may have entered care through a different route, or may have been less severely impaired in the period before assessment. Treating these two individuals as statistically equivalent discards information that could sharpen prediction.

The authors propose that medication exposure should be characterized in far more detail than current studies typically allow. Rather than recording a simple yes-or-no variable for whether someone was taking a drug at baseline, researchers could document the specific class of medication, the dose, the duration of exposure, the indication for which it was prescribed, the degree of adherence, the severity of the condition it was meant to treat, and the longitudinal pattern of changes in the prescription over time. Each of these dimensions carries distinct clinical information, and together they could form a much richer picture of where a particular individual sits on the spectrum of emerging vulnerability. This level of detail is already available in most clinical records; it simply has not been systematically extracted or analyzed.

The implications for prediction research are substantial. If baseline medication carries genuine prognostic information, incorporating it into prediction models could improve the accuracy with which researchers identify who is most likely to transition to psychosis. It could also help explain some of the puzzling heterogeneity in outcomes that has characterized the field, where individuals who appear similar on paper follow very different trajectories. The authors suggest that some of the apparent decline in transition rates across successive cohorts might be partly related to changes in prescribing practices, with more young people receiving pharmacological treatment earlier in the course of emerging problems, which could alter the composition of the groups recruited into research studies. This hypothesis cannot be confirmed without systematic data, but it illustrates the kind of insight that becomes possible when medication history is treated as a variable worth studying rather than one worth adjusting away.

The perspective also carries implications for clinical practice. Clinicians working in early detection services already know, in an informal way, that a young person with a history of prior treatment is different from one without such a history. The formal research criteria for CHR-P do not capture this difference, and prediction models built on those criteria inherit the blind spot. Raballo and colleagues argue that the field should develop a more nuanced clinical vocabulary for describing the treatment history of help-seeking young people, one that distinguishes between different classes of prior medication, different indications, different durations, and different patterns of response. Such a vocabulary would allow clinicians to communicate more precisely about the clinical state of the people they assess, and would give researchers a richer set of variables to work with when building the next generation of prediction tools.

The authors are careful to note that their argument does not imply that medication causes any particular outcome, nor that prescribing practices should change based on the current evidence. The relationship between baseline pharmacotherapy and later trajectory is observational, and disentangling the causal structure would require study designs that go beyond the observational data currently available. What the perspective does argue is that the field has been too quick to discard a variable that is available in nearly every clinical encounter, that the discarded information is clinically meaningful, and that recovering it could improve both prediction and clinical understanding. In a field where the stakes involve young people at a vulnerable moment in their lives, and where the difference between over-treatment and under-treatment can be consequential, the argument for paying closer attention to the medication record is one that researchers and clinicians alike may find difficult to ignore.

Subject of Research: Prognostic significance of baseline pharmacotherapy in clinical high-risk for psychosis

Article Title: From bias to signal: the prognostic meaning of baseline pharmacotherapy in clinical high-risk for psychosis

Article References: Raballo, A., Poletti, M., & Preti, A. (2026). From bias to signal: the prognostic meaning of baseline pharmacotherapy in clinical high-risk for psychosis. Schizophrenia. https://doi.org/10.1038/s41537-026-00819-x

Image Credits: AI Generated

DOI: 10.1038/s41537-026-00819-x

Keywords: bias, signal, prognostic, meaning, baseline, pharmacotherapy, clinical, high-risk, psychosis, scientific research