Localizing at-risk and early psychosis populations along data-driven clinical, cognitive, and neuroanatomical spectra

Article activity feed

1. 

   PREreview

   Aug 7, 2026

   This Zenodo record is a permanently preserved version of a PREreview. You can view the complete PREreview at https://prereview.org/reviews/21837640.

   This study profiles 194 participants (i.e., 70 with first episode psychosis (FEP), 40 at clinical high risk (CHR), 43 at familial high risk (FHR), and 41 healthy controls) recruited from a single clinical infrastructure in Montreal and assessed with a dense clinical, cognitive, functional, and structural neuroimaging battery. Rather than relying on group contrasts alone, the authors derive data-driven axes of variability: an exploratory factor analysis yields three symptom dimensions (depression-anxiety; negative cognitive functioning; positive functioning), and a partial least squares correlation identifies the strongest axis of covariance between cortical thickness and symptoms. Group …

   This Zenodo record is a permanently preserved version of a PREreview. You can view the complete PREreview at https://prereview.org/reviews/21837640.

   This study profiles 194 participants (i.e., 70 with first episode psychosis (FEP), 40 at clinical high risk (CHR), 43 at familial high risk (FHR), and 41 healthy controls) recruited from a single clinical infrastructure in Montreal and assessed with a dense clinical, cognitive, functional, and structural neuroimaging battery. Rather than relying on group contrasts alone, the authors derive data-driven axes of variability: an exploratory factor analysis yields three symptom dimensions (depression-anxiety; negative cognitive functioning; positive functioning), and a partial least squares correlation identifies the strongest axis of covariance between cortical thickness and symptoms. Group means differ along these dimensions, but individual level separability between CHR and FEP is modest (AUC 0.57–0.69). Cortical thickness reductions relative to controls were evident in FEP but not in either high risk group, and the CHR anatomical profile did not replicate in two independent international cohorts.

   The main contribution is the direct, within-setting comparison of all three psychosis risk stages under a common assessment protocol. Also, the individual level framing is a strength: reporting AUC alongside mean differences makes explicit something that group contrast papers routinely obscure, namely that a significant mean difference can coexist with near-total distributional overlap. The sensitivity analyses (covarying for age and sex; repeating analyses in complete-case subsets; testing the CHR anatomical profile in held-out cohorts) are thorough, and the negative findings are reported rather than buried. The work makes a useful empirical case that the CHR–FEP boundary, defined largely by positive symptom severity, does not mark a clean breakpoint in the broader clinical or neuroanatomical landscape.

   Major issues

   • The specificity of the "schizophrenia-like" cortical thickness pattern in FEP is overstated. The abstract and discussion foreground the resemblance to schizophrenia (r = 0.51), but the correlation with the bipolar disorder map was numerically higher (r = 0.55) and OCD was also significant (r = 0.41). The FEP group by design includes affective psychosis (bipolar, depressive), which makes the bipolar result substantively interesting rather than incidental. More importantly, the ENIGMA case-control maps are themselves strongly intercorrelated, so a map that correlates with three of them is not evidence of a psychosis-specific profile — it is compatible with a general transdiagnostic thinning pattern. I would suggest either (a) reporting the inter-map correlations among the ENIGMA references and testing whether the schizophrenia correlation exceeds what would be expected given that shared structure, or (b) rewording the abstract and discussion to describe a pattern resembling several adult psychiatric conditions, of which schizophrenia is one. As written, the abstract sentence omits the bipolar and OCD results entirely.

   • The CHR null is treated as evidence of absence without inferential support for that interpretation. The conclusion that "no consistent anatomical pattern could be identified in CHR" rests on a non-significant result in a group of roughly 36 participants after quality control. The two replication cohorts strengthen the claim, but their sample sizes and the power available in each are not reported in the main text. Some form of positive evidence for the null — equivalence testing (e.g. TOST against a smallest effect size of interest derived from the ENIGMA CHR mega-analysis), a Bayes factor, or at minimum a power calculation against the effect magnitude reported by Jalbrzikowski et al. — would considerably strengthen this. There is also mild tension between the flat null in the abstract and Section 3.5, which reports a nominally significant CHR–22q11.2 correlation (r = 0.28, p = 0.045).

   • The extent and pattern of missing data warrant fuller treatment in the main text. Of 153 participants in the factor analysis, only 104 had complete summary-scale data, meaning roughly a third required imputation at the scale level rather than the item level — a substantially stronger assumption. The degrees of freedom in Table 2 show that missingness is concentrated in exactly the measures that drive the headline CHR finding: the QIDS ANOVA is based on 122 of 194 participants, with the BAI, SPIN, and DASS-21 also incomplete. Since the depression-anxiety factor is built almost entirely from these self-report scales, the finding that CHR expresses the most depression-anxiety symptoms is the one most dependent on imputation. Please report whether missingness was related to group, age, sex, or symptom severity (i.e. whether MAR is plausible), and consider bringing the key numbers from Supplementary Tables 4–5 into the main text.

   • The antipsychotic dosage covariate is close to collinear with group status. Chlorpromazine-equivalent dose is zero for every control and every FHR participant, non-zero for essentially all FEP participants, and non-zero for only six CHR participants. In a model comparing FEP to controls, dose is therefore very nearly a proxy for group membership, and the resulting coefficient is difficult to interpret as a medication effect. The claim that the adjusted map "could then be interpreted as approximating differences from controls for an unmedicated patient" also requires linear extrapolation to a dose of zero within a group where almost no patient sits at or near zero. I would suggest reporting the number of unmedicated FEP participants, and, if that subgroup is large enough, showing the unadjusted map for unmedicated FEP as a more direct check.

   • The age imbalance is substantial and may not be fully addressed by a linear covariate. The CHR group is roughly 4.5 years younger than both the FHR and FEP groups (20.75 vs 25.47 and 25.10). Cortical thickness changes non-linearly across the 14–35 range, so a linear age term may leave residual confounding precisely in the group where the anatomical null is reported. It would help to show either a non-linear age model, an age-matched subsample analysis, or the CHR–control comparison restricted to an overlapping age band. Years of education also differs markedly (10.94 in CHR vs 13.51 in controls) and is likely partly age-driven; its role, if any, in the functioning and cognition comparisons is not discussed.

   • No inferential test is reported for the PLS latent variable. The manuscript states that only LV1 was interpreted because it captures the strongest source of covariance, and bootstrap resampling is used for the contribution of individual variables. But bootstrapping the loadings does not establish that the latent variable itself exceeds what would arise from noise in a sample of 140 with 68 brain regions and 15 clinical variables. Given that the authors have written specifically on permutation-based inference for PLS (ref. 62), a sentence or two explaining the rationale for omitting an omnibus test — and what readers should use instead to gauge whether LV1 is above chance — would be valuable.

   • Distributional assumptions for the ANOVAs are not addressed, and floor effects are visible in the data. Several scales show standard deviations exceeding their means in the less-affected groups (SAPS in FHR: 0.88 ± 1.50; SANS in FHR: 1.29 ± 2.05), indicating strong floor effects and clear departures from normality. Since the FHR-versus-others comparisons are load-bearing for the "stepwise increase" framing, please report assumption checks or confirm the results with non-parametric or robust alternatives (e.g. Kruskal-Wallis with Dunn post-hoc, or permutation tests).

   • The service-delivery conclusion goes beyond what the design can support. The recommendation to incorporate early psychosis services into general transdiagnostic mental health infrastructure is a policy claim, while the data are cross-sectional, single-site, and observational. Phenotypic overlap between CHR and FEP is consistent with transdiagnostic service models but does not test whether such models improve outcomes, and the study includes no comparison of care configurations. I would suggest softening this to say the results are consistent with, or supportive of the rationale for, transdiagnostic frameworks, rather than that they favour a particular service arrangement.

   • No data availability, code availability, or author contributions statement is included. Given the reliance on specific software (FreeSurfer 7.1.0, the ENIGMA FreeSurfer protocol, the mifa R package, the ENIGMA toolbox, neuromaps) and a multi-step analytic pipeline, sharing analysis code would substantially aid reproducibility. If the clinical data cannot be shared, a statement setting out the access conditions would still be appropriate.

   Minor issues

   • The number of participants included in the final neuroimaging analyses is never stated directly. The reader must subtract seven quality-control failures and one processing failure from 190 to arrive at 182, then use the pass rates in Table 1 to recover the per-group numbers. Please state the final analysed N overall and by group.

   • Relatedly, the varying analytic subsamples (194 recruited; 153 factor analysis; 140 PLS; 105 longitudinal; 104 and 99 complete-case sensitivity analyses) are difficult to track. A participant flow diagram, or a single table listing each analysis with its N and inclusion criteria, would help considerably.

   • Table 1 reports chlorpromazine-equivalent dose for CHR as 61.72 (25.01), but it is unclear whether this is the mean across all 40 CHR participants or across the six taking antipsychotics. Please specify the denominator.

   • Table 1 gives no test statistic for the CHR-versus-FEP dose comparison, although one is provided for days since program entry. Empty cells for HC and FHR would be clearer marked with an em dash or "not applicable" rather than left blank.

   • The denominators in the Table 1 statistics vary (N = 194, 193, 190, 189) without explanation, implying missing values on race, education, and imaging. A footnote noting where data were missing would help.

   • Section 3.5 refers to "left pars orbitalis, rostral middle frontal, and supramarginal cortex." It is ambiguous whether "left" applies to all three regions or only the first. Please specify hemisphere for each.

   • Section 3.2 states that cognitive performance was lower in FEP relative to CHR and HC, but Table 2 shows this pattern for the Cogstate only; the WMS-IV comparison was significant for FEP versus HC alone. Please name the measure to which each claim applies.

   • The asterisk conventions differ between figures. Figure 1 uses ** for p(Tukey) < 0.05, whereas Figure 2 uses for uncorrected p < 0.05 and * for p(Holm) < 0.05. Using ** for a 0.05 threshold in one figure and a corrected threshold in another invites misreading; a single convention across all figures would be clearer.

   • Abbreviations differ between text and figures: PANSS-6 / PANSS6, DASS-21 / DASS21, WMS-IV / WMS. Please harmonise.

   • The correction threshold varies across analyses (Holm-Bonferroni for the scale comparisons and spatial correlations; a lenient 10% FDR for the vertex-wise maps) and the rationale for the lenient threshold is given only as "exploratory inference." A brief justification, and consistent labelling of which findings are exploratory, would help readers calibrate.

   • For the CHR spatial correlations, the 22q11.2 result is reported in the text but the 22q11.2-psychosis-positive result is not, despite the latter arguably being the more relevant comparison for a psychosis-risk sample. Figure 2B suggests it was negative. Please report it in text.

   • Section 2.3 states that BCIS subscales were "subtracted from each other" without specifying the direction. Please state explicitly (presumably self-reflectiveness minus self-certainty).

   • The SCID is described as the "Structured Clinical Interview for the Diagnostic and Statistical Manual of Mental Disorders (DSM) IV"; conventionally this is written DSM-IV, and the abbreviation SCID is never introduced. A brief note on why DSM-IV criteria were used rather than DSM-5 would also be helpful for readers assessing comparability with more recent cohorts.

   • The IQ < 70 exclusion criterion does not specify the instrument used to establish IQ.

   • Section 2.6 lists "motor movement disorders" among the MRI exclusion criteria; "movement disorders" alone would suffice.

   • In the abstract, "less afflicted stages" attributes affliction to stages rather than to individuals; something like "earlier or less severe stages" would read more naturally.

   • Several references to instruments cite database records rather than the original publications — for example ref. 39 (Beck Anxiety Inventory) and ref. 44 (Young Mania Rating Scale) are cited as "Published online 1988/1978" with PsycTests DOIs, rather than Beck et al. (1988), J Consult Clin Psychol 56(6):893–897 and Young et al. (1978), Br J Psychiatry 133:429–435. Refs 29, 31, and 32 are similarly formatted. Citing the primary sources would be preferable.

   • Several advance-access citations may now have final volume and page details, given the elapsed time since posting: refs 8, 10, 65, 84, and 85 are all listed as "Published online" with article-ID pagination. Worth a final check before journal submission.

   • Reference 35 contains a stray full stop in the author list ("Morosini PL., Magliano L").

   • Reference styling is inconsistent between sentence case and title case for article titles (compare refs 31 and 32). Ref. 3 carries an access date where other references do not.

   • The PDF contains a number of words broken across line breaks without hyphens — "negativecognitive-functioning," "casecontrol," "metaanalysis," "HolmBonferroni," "FirstEpisode." Some of these may be artefacts of PDF rendering rather than errors in the source file, but they are worth checking in the typeset version, particularly "negativecognitive-functioning," which appears in both the abstract and the Figure 1 legend.

   Competing interests

   The author declares that they have no competing interests.

   Use of Artificial Intelligence (AI)

   The author declares that they used generative AI to come up with new ideas for their review.

   Read the original source
2. 

   Version published to 10.64898/2026.03.04.26347618 on medRxiv

   Mar 5, 2026