Full blood count data for clustering in clinical high risk (CHR) for psychosis individuals: validation of unsupervised and semi-supervised approaches

Background

Increased neutrophils activity relative to lymphocytes following a first episode of psychosis, reflective of the balance of innate-adaptive immunity (Bioque et al., 2024, Journal of Psychiatric Research 172, 334–339). Clinical high risk for psychosis (CHR) individuals who are at increased risk of developing psychosis have shown neurophysiological deficits measured using resting-state and task-based EEG associated with increased pro-inflammatory cytokines (Hamilton et al., 2025, Clin EEG Neurosci 56, 46–59). Leukocytes are the primary cells that produce cytokines; cytokine levels and cell counts are correlated with cytokine levels. Platelets play a significant role in immune modulation and chronic inflammation impacts red cells. There may be biological subtypes within the CHR population who differ in their full blood count (FBC) profile and linked neurophysiological and symptom outcomes. We have three main aims: Firstly, investigate the performance of two different cluster-based solutions against internal validation techniques to assess if identified clusters produced using FBC data resemble meaningful subtypes. Secondly, compare cluster solutions produced by different algorithms to establish degree of agreement between algorithms. Finally, to assess whether differing cluster allocations are associated with cross sectional differences in demographics, neurophysiological, and clinical symptoms

Methods

Two distinct algorithms were selected for investigation. Gaussian Mixture Modelling (GMM) was selected as the unsupervised clustering approach. Heterogeneity through Discriminative Analysis clustering (HYDRA) was selected as a semi-supervised clustering approach. These were applied to a population of 787 CHR participants (Wannan et al., 2024, Schizophr Bull 50, 496–512) based on baseline leukocyte data (Neutrophils, Monocytes, Lymphocytes). Algorithms were assessed using relevant internal validation metrics. GMM was assessed using allocation uncertainty and silhouette width. HYDRA was assessed using adjusted rand index (ARI) as a measure of solution stability and silhouette width. ARI was also used for between model agreement. Analysis of variance was used to assess correlation between cluster allocation and neurophysiology (mismatch negativity and P300 amplitude), and clinical symptoms (Brief Psychiatric Rating Scale subscales). Multinomial regression was used to determine relative importance of variables regarding association with cluster allocations. Analyses were controlled for demographic variation (age, sex, and body-mass index).

Results

There was low level of agreement between models (ARI = 0.194) and Silhouette Width indicated low cluster separation for both algorithms (GMM = 0.034, HYDRA = 0.19). GMM cluster allocation uncertainty was moderate (0.18). HYDRA showed good cluster stability (ARI = 0.65). There was significant association between GMM cluster allocation and P300 amplitude (p = 0.047), BPRS positive symptoms (p = 0.003), BPRS affect (p = 0.036), and BPRS activation (p = 0.001), after adjusting for sex, age, and BMI. Multinomial regression reinforced BPRS activation was the most relevant feature of interest. For HYDRA clusters only BPRS disorganisation reached significance (p = 0.010) after covariate adjustment, this association was reinforced by multinomial regression for cluster allocation.

Discussion

HYDRA and GMM showed overall disagreement and clusters did not show separation by distance in the feature space, however cluster allocation was significantly association with symptomatic and neurophysiological outcomes. This may suggest heuristic utility despite weak internal validity.

References

Bioque, M., Llorca-Bofí, V., Salmerón, S., García-Bueno, B., MacDowell, K.S., Moreno, C., Sáiz, P.A., González-Pinto, A., Hidalgo-Figueroa, M., Barcones, M.F., Rodriguez-Jimenez, R., Bernardo, M., Leza, J.C., 2024. Association between neutrophil to lymphocyte ratio and inflammatory biomarkers in patients with a first episode of psychosis. Journal of Psychiatric Research 172, 334–339. https://doi.org/10.1016/j.jpsychires.2024.02.044 Hamilton, H.K., Roach, B.J., Bachman, P.M., Belger, A., Carrion, R.E., Duncan, E., Johannesen, J.K., Light, G.A., Niznikiewicz, M.A., Addington, J., Bearden, C.E., Cadenhead, K.S., Cornblatt, B.A., Perkins, D.O., Tsuang, M.T., Walker, E.F., Woods, S.W., Cannon, T.D., Mathalon, D.H., 2025. Mismatch Negativity as an Index of Auditory Short-Term Plasticity: Associations with Cortisol, Inflammation, and Gray Matter Volume in Youth at Clinical High Risk for Psychosis. Clin EEG Neurosci 56, 46–59. https://doi.org/10.1177/15500594241294035 Wannan, C.M.J., Nelson, B., Addington, J., Allott, K., Anticevic, A., Arango, C., Baker, J.T., Bearden, C.E., Billah, T., Bouix, S., Broome, M.R., Buccilli, K., Cadenhead, K.S., Calkins, M.E., Cannon, T.D., Cecci, G., Chen, E.Y.H., Cho, K.I.K., Choi, J., Clark, S.R., Coleman, M.J., Conus, P., Corcoran, C.M., Cornblatt, B.A., Diaz-Caneja, C.M., Dwyer, D., Ebdrup, B.H., Ellman, L.M., Fusar-Poli, P., Galindo, L., Gaspar, P.A., Gerber, C., Glenthøj, L.B., Glynn, R., Harms, M.P., Horton, L.E., Kahn, R.S., Kambeitz, J., Kambeitz-Ilankovic, L., Kane, J.M., Kapur, T., Keshavan, M.S., Kim, S.-W., Koutsouleris, N., Kubicki, M., Kwon, J.S., Langbein, K., Lewandowski, K.E., Light, G.A., Mamah, D., Marcy, P.J., Mathalon, D.H., McGorry, P.D., Mittal, V.A., Nordentoft, M., Nunez, A., Pasternak, O., Pearlson, G.D., Perez, J., Perkins, D.O., Powers, A.R., Roalf, D.R., Sabb, F.W., Schiffman, J., Shah, J.L., Smesny, S., Spark, J., Stone, W.S., Strauss, G.P., Tamayo, Z., Torous, J., Upthegrove, R., Vangel, M., Verma, S., Wang, J., Rossum, I.W., Wolf, D.H., Wolff, P., Wood, S.J., Yung, A.R., Agurto, C., Alvarez-Jimenez, M., Amminger, P., Armando, M., Asgari-Targhi, A., Cahill, J., Carrión, R.E., Castro, E., Cetin-Karayumak, S., Mallar Chakravarty, M., Cho, Y.T., Cotter, D., D’Alfonso, S., Ennis, M., Fadnavis, S., Fonteneau, C., Gao, C., Gupta, T., Gur, R.E., Gur, R.C., Hamilton, H.K., Hoftman, G.D., Jacobs, G.R., Jarcho, J., Ji, J.L., Kohler, C.G., Lalousis, P.A., Lavoie, S., Lepage, M., Liebenthal, E., Mervis, J., Murty, V., Nicholas, S.C., Ning, L., Penzel, N., Poldrack, R., Polosecki, P., Pratt, D.N., Rabin, R., Rahimi Eichi, H., Rathi, Y., Reichenberg, A., Reinen, J., Rogers, J., Ruiz-Yu, B., Scott, I., Seitz-Holland, J., Srihari, V.H., Srivastava, A., Thompson, A., Turetsky, B.I., Walsh, B.C., Whitford, T., Wigman, J.T.W., Yao, B., Yuen, H.P., Ahmed, U., Byun, A. (Jin S., Chung, Y., Do, K., Hendricks, L., Huynh, K., Jeffries, C., Lane, E., Langholm, C., Lin, E., Mantua, V., Santorelli, G., Ruparel, K., Zoupou, E., Adasme, T., Addamo, L., Adery, L., Ali, M., Auther, A., Aversa, S., Baek, S.-H., Bates, K., Bathery, A., Bayer, J.M.M., Beedham, R., Bilgrami, Z., Birch, S., Bonoldi, I., Borders, O., Borgatti, R., Brown, L., Bruna, A., Carrington, H., Castillo-Passi, R.I., Chen, J., Cheng, N., Ching, A.E., Clifford, C., Colton, B.-L., Contreras, P., Corral, S., Damiani, S., Done, M., Estradé, A., Etuka, B.A., Formica, M., Furlan, R., Geljic, M., Germano, C., Getachew, R., Goncalves, M., Haidar, A., Hartmann, J., Jo, A., John, O., Kerins, S., Kerr, M., Kesselring, I., Kim, H., Kim, N., Kinney, K., Krcmar, M., Kotler, E., Lafanechere, M., Lee, C., Llerena, J., Markiewicz, C., Matnejl, P., Maturana, A., Mavambu, A., Mayol-Troncoso, R., McDonnell, A., McGowan, A., McLaughlin, D., McIlhenny, R., McQueen, B., Mebrahtu, Y., Mensi, M., Hui, C.L.M., Suen, Y.N., Wong, S.M.Y., Morrell, N., Omar, M., Partridge, A., Phassouliotis, C., Pichiecchio, A., Politi, P., Porter, C., Provenzani, U., Prunier, N., Raj, J., Ray, S., Rayner, V., Reyes, M., Reynolds, K., Rush, S., Salinas, C., Shetty, J., Snowball, C., Tod, S., Turra-Fariña, G., Valle, D., Veale, S., Whitson, S., Wickham, A., Youn, S., Zamorano, F., Zavaglia, E., Zinberg, J., Woods, S.W., Shenton, M.E., 2024. Accelerating Medicines Partnership® Schizophrenia (AMP® SCZ): Rationale and Study Design of the Largest Global Prospective Cohort Study of Clinical High Risk for Psychosis. Schizophr Bull 50, 496–512. https://doi.org/10.1093/schbul/sbae011