Identifying Activity Patterns From Different Actigraphy Devices in People With Schizophrenia Spectrum Disorders: A Comparative Study

Identifying Activity Patterns From Different Actigraphy Devices in People With Schizophrenia Spectrum Disorders: A Comparative Study | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Identifying Activity Patterns From Different Actigraphy Devices in People With Schizophrenia Spectrum Disorders: A Comparative Study Damiano Archetti, Cesare Michele Baronio, Elisa Caselani, Marta Magno, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9493731/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 10 You are reading this latest preprint version Abstract Actigraphy is a reliable and non-invasive method for assessing motor activity and how disease affects the daily activity of patients. Machine learning models applied to actigraphy data have been used to identify activity patterns in patients suffering from either major depressive disorder or Schizophrenia Spectrum Disorders (SSDs). However, heterogeneous collection protocols across different datasets significantly hinder generalizability, making direct comparison difficult. In this work, we used actigraphy data to inform a machine learning model that distinguishes between the activity patterns of healthy controls (HC) and patients with SSDs. Actigraphy recordings from a total of 258 subjects (126 HC and 132 with SSDs) from the DiAPASon and PSYKOSE datasets, each spanning one week, were used. Actigraphy patterns were classified using principal component analysis (PCA) followed by a support vector machine (SVM) classifier trained on DiAPASon data. The model was validated using independent data from PSYKOSE, collected with a different actigraphy device, after a rescaling procedure to match DiAPASon. After rescaling, the concordance between DiAPASon and PSYKOSE was 99.9% (30.7% pre-rescaling). The best performing model achieved a classification accuracy of 0.86 for DiAPASon data and 0.80 for PSYKOSE, demonstrating the model’s generalizability. Using Shapley analysis, distinctive activity patterns driving classifications towards HC or SSDs classes were detected. Our results stand as a proof of concept that machine learning models, able to identify activity patterns, can be generalized to other actigraphy devices, but further testing on more diverse datasets and devices is required. Biological sciences/Computational biology and bioinformatics Health sciences/Diseases Health sciences/Health care Physical sciences/Mathematics and computing Biological sciences/Neuroscience Figures Figure 1 Figure 2 Figure 3 1. Introduction Schizophrenia spectrum disorders (SSDs) are severe and highly disabling mental disorders characterized by marked clinical heterogeneity, encompassing positive and negative symptoms, cognitive impairment, and profound disturbances in everyday functioning [ 1 ] . Among the less visible yet clinically meaningful dimensions of these disorders are abnormalities in motor activity patterns and circadian rhythms [ 2 , 3 , 4 ] , which reflect the integrated functioning of neurobiological systems, psychopathology, and real-world behavior. In recent years, actigraphy has emerged as a reliable, non-invasive, and ecologically valid method for the continuous assessment of motor activity in naturalistic settings [ 5 ] , overcoming several limitations of traditional clinical assessments based on episodic observation or self-report [ 6 ] . A growing body of evidence indicates that individuals with SSDs exhibit distinct actigraphic signatures compared with healthy controls, including reduced overall activity levels, increased fragmentation of motor behavior, circadian irregularity, and altered temporal distribution of movement across the 24-hour cycle [ 7 , 8 , 9 ] . These features have been linked to negative symptom severity, cognitive dysfunction, social withdrawal, and poorer functional outcomes [ 10 ] . Consequently, actigraphy has gained increasing relevance as an objective behavioral marker with potential applications in diagnosis, longitudinal monitoring, and outcome prediction. Parallel to these developments, the application of Machine Learning (ML) techniques to actigraphy data has opened new avenues for identifying complex activity patterns that may not be detectable through conventional statistical approaches [ 11 ] . Supervised and unsupervised models have been used to distinguish between diagnostic groups [ 12 ] , characterize symptom profiles [ 13 , 14 , 15 ] , and explore behavioral phenotypes across different mental disorders [ 16 , 17 ] . Within the broader framework of digital phenotyping, such approaches hold promise for advancing precision psychiatry by providing scalable and data-driven tools capable of capturing clinically meaningful behavioral information. Despite this progress, a major methodological challenge limits the translational potential of actigraphy-based machine learning models: the substantial heterogeneity in data acquisition protocols and devices. Actigraphy devices differ in sensor technology, sampling frequency, signal processing pipelines, and proprietary algorithms, leading to systematic discrepancies in recorded activity levels. As a result, models trained on data from a single device or cohort often show reduced performance when applied to external datasets, undermining their generalizability and clinical utility [ 18 ] . This issue is particularly critical for multi-center studies and for real-world implementation, where different devices are frequently used. Within this context, the present study aims to evaluate the generalizability of actigraphy-based ML models for distinguishing between healthy controls and individuals with SSDs using data derived from two independent datasets [ 19 , 20 ] collected with different actigraphy devices. By combining dimensionality reduction techniques with supervised classification and implementing a rescaling procedure to harmonize activity signals, this work provides a proof of concept (PoC) that meaningful motor activity patterns can be transferred across devices. Ultimately, these findings contribute to the growing evidence base supporting the feasibility of device-agnostic digital phenotyping approaches in SSDs and highlight both the potential and the remaining challenges for their integration into clinical research and practice. 2. Methods 2.1 Participants Data for this work were acquired from the DiAPASon [ 19 ] and PSYKOSE [ 20 , 21 ] datasets. Diapason is a multi-centric study carried out in northern Italy between 2020 and 2021. The study was carried out in accordance to national and institutional committees on human experimentation, as well as the Helsinki Declaration, and was approved by local Ethical Committees [ 19 ] . PSYKOSE is a fully anonymized publicly available dataset that comprises data from individuals enrolled at Haukeland University Hospital, Norway. The study was approved by the local Ethical Committee and carried out in accordance with local regulations [ 20 ] . All participants signed an informed consent form, please refer to the original works for detailed information about recruitment [ 19 , 20 ] . Only individuals with SSDs and healthy controls (HC) were considered. Additional inclusion criteria were the availability of actigraphy data spanning at least seven consecutive days. 2.2 Actigraphy data Actigraphy data were collected with the ActiGraph GT9X Link for a continuative period of one week for DiAPASon and with Actiwatch AW4 for a continuative period of two weeks for PSYKOSE. The sampling frequency was set to 30Hz for ActiGraph GT9X and 32 Hz for Actiwatch AW4, and actigraphy counts were computed separately with the proprietary software for each device in 60-second epochs. All actigraphy tracks were trimmed to the same length of 10,080 timepoints (one per minute), starting at 00.00 AM of Monday and ending at 11.59 PM of the Sunday. Actigraphy tracks with zero counts for extended periods were excluded, while sporadic missing timepoints were imputed as the average of that timepoint for the respective diagnostic class separately for each cohort. In the case of DiAPASon, where actigraphy recordings were available for one week only, the available days were reordered to match the Monday-Sunday ordering. In the case of PSYKOSE, where actigraphy recordings were available for two weeks, we picked the first available Monday-Sunday span. As the actigraphy data were collected with different devices, possible biases could arise [ 18 ] , therefore we tried several methods to rescale PSYKOSE data to match activity distributions of DiAPASon data. Specifically, we tried the following: Match punctual average : for each timepoint of the actigraphy track, PSYKOSE counts were multiplied by a factor that matched the average value observed in PSYKOSE at that timepoint to the average value observed in DiAPASon at that timepoint. Match local average : for each timepoint of the actigraphy track, PSYKOSE counts were multiplied by a factor that matched the average value observed in PSYKOSE in an interval centered at that timepoint to the average value observed in DiAPASon in an interval centered at that timepoint. Intervals with radii of 5, 10, 30 and 60 minutes timepoints were explored. Match global average : data of all timepoints of the PSYKOSE actigraphy tracks were multiplied by a factor that matched the average value observed globally in PSYKOSE to the average value observed globally in DiAPASon; Linear model : PSYKOSE data underwent a linear transformation so that the relation between timepoint-specific count averages in DiAPASon and PSYKOSE was an identity; Linear model + floor : same as linear model, but after the linear transformation all negative counts were set to 0. All cases were tested either using HCs from DiAPASon as a reference. The goodness of rescaling was assessed by measuring concordance, calculated as the fraction of statistical comparisons between distributions of actigraphy counts at the same time-point that were not significantly different between DiAPASon and PSYKOSE. Only PSYKOSE data scaled with the method that reached the best concordance were considered for subsequent analyses. 2.3 Activity pattern classification To avoid possible model overfitting caused by the high dimensionality of actigraphy data, and to identify a model capable of differentiating between healthy-like and SSD-like activity patterns, a grid-search over dimensionality reduction algorithms and over Support Vector Machine (SVM) classifiers was performed. Specifically, we cycled over principal component analysis (PCA), Uniform Manifold Approximation and Projection (UMAP) and T-distributed Stochastic Neighbor Embedding (TSNE) dimensionality reduction algorithms, and over different SVM kernels (linear, polynomial, RBM), all with multiple parameters, producing a total of 1,584 possible combinations. Data from DiAPASon were used as a training set, and only the model that achieved the best classification accuracy over repeated k-fold cross validation (50 repetitions, 10 folds for each repetition) was considered. Generalizability of the best performing model was tested by classifying individuals from PSYKOSE that served as independent test set. 2.4 Model explainability and Statistical Analysis To enhance the explainability of the model a SHAPLEY analysis [ 22 ] was performed to investigate which components of the embedded space were the most influential for the classification task. Population-level statistical comparisons were performed with chi-squared test for categorical variables and with t-test for continuous variables. Inter-cohort statistical comparisons between distributions of actigraphy counts for a specific timepoint was performed with the Kolmogorov-Smirnov test, and before computing concordance a correction against multiple tests comparison was performed via Benjamini-Hochberg false discovery rate. Significance level for each of the statistical tests was set at α = 0.05. Modelling and all statistical analyses were performed with python version 3.10.10, specifically, scikit-learn 1.4.0 was used for machine learning models and scipy 1.12.0 for statistical analyses. 3. Results A total of 213 individuals from DiAPASon and 45 individuals from PSYKOSE were considered, relevant demographics are available in Table 1 and detailed demographics of the DiAPASon sample are available in the Supplementary Table 1. Populations of HC and SSDs patients were mostly comparable inside either data cohort, with the only significant difference (p-value of chi squared test = 0.0024) being the sex of PSYKOSE individuals, where male population was more numerous for SSDs patients than for healthy controls. The total populations of DiAPASon and PSYKOSE showed no significant differences. The SSDs population form DiAPASon was composed of either inpatients (61) or outpatients (51), this information was not available for the PSYKOSE dataset. Table 1 Socio-demographic characteristics of patients and healthy controls from DiAPASon and PSYKOSE Variables DiAPASon PSYKOSE HC SSDs TOTAL HC SSDs TOTAL Sex, n (%) ° Female 42 (41.6%) 42 (37.5%) 84 (39.4%) 14 (58.3%) 3 (14.3%) 17 (37.8%) Male 59 (58.4%) 70 (62.5%) 129 (60.6%) 10 (41.7%) 18 (85.7%) 28 (62.2%) Age, n (%) 20–29 16 (15.8%) 21 (18.6%) 37 (17.4%) 7 (29.2%) 1 (4.8%) 8 (17.8%) 30–44 33 (32.7%) 40 (35.7%) 73 (34.3%) 8 (33.3%) 10 (47.6%) 18 (40.0%) ≥ 45 52 (51.5%) 51 (45.6%) 103 (48.3%) 9 (37.5%) 10 (47.6%) 19 (42.2%) BPRS Sum (average ± SD) - 46 ± 13 - - 49.4 ± 8.8 - Acronyms: BPRS= brief psychiatric rating scale; HC= Healthy Control; n= total number; SD= standard deviation; SSDs= Schizophrenia Spectrum Disorders patient; *= significant differences between HC and SSDs patients inside DiAPASon, °= significant differences between HC and SSDs patients inside PSYKOSE; §= global significant differences between DiAPASon and PSYKOSE populations. 3.1 Data Scaling Before scaling, DiAPASon and PSYKOSE data showed poor comparability, with an average concordance score between activity counts distributions equal to 30.7%. On average PSYKOSE data had lower activity counts than DiAPASon (1,600 ± 2,200 for DiAPASon, 240 ± 350 for PSYKOSE). The best performing actigraphy scaling method resulted to be the “ Match local average” case on a 60 minutes radius interval, which produced an average concordance of 99.9% (Table 2 ). Other scaling methods produced relevant improvements as well, except for the “linear model” method without flooring. For further analyses only PSYKOSE data scaled with the best performing method were considered. Table 2 performance of scaling methods. Scaling Concordance (HC) Concordance (SSDs) No scaling (reference) 28.4% 33.0% Match punctual average 99.8% 98.7% Match local average (5 minutes radius) 100% 99.5% Match local average (10 minutes radius) 100% 99.5% Match local average (30 minutes radius) 100% 99.6% Match local average (60 minutes radius) 100% 99.8% Match global average 99.0% 99.2% Linear model 41.0% 17.9% Linear model with flooring 97.0% 96.4% Acronyms: HC= Healthy Control; SSDs= Schizophrenia Spectrum Disorders patient 3.2 Classification Model Grid search over dimensionality reduction methods and SVM parameters returned as the best performing model a PCA with 9 components followed by a SVM with RBF kernel (gamma = 2.50e-11), which returned a classification accuracy of 0.860 (95% CI: [0.854–0.866]) on a repeated k-fold cross validation (10 folds, 50 repetitions) on the DiAPASon training set. Scaled PSYKOSE data were projected in the PCA space defined on the DiAPASon training set and classified according to the SVM informed on DiAPASon (Fig. 1 ). In this case classification accuracy was 0.80. Acronyms: HC= Healthy Control; SSDs= Schizophrenia Spectrum Disorders patient; Acc= classification accuracy. 3.3 Model Explainability Shapley analysis returned the first two PCA directions and the fifth PCA direction as the main driving factors for classification for both the training and test sets. High values for these features drove the classification towards the HC class, while low values drove the classification towards SSD class. The ranking of the other PCA features was almost the same for the two datasets except for the seventh and ninth directions, which however had only marginal contributions to classification (Fig. 2 ). Acronyms: HC= Healthy Control; SSDs= Schizophrenia Spectrum Disorder patient. To improve the explainability of the model, we computed the eigenvectors of the PCA directions, as shown in Fig. 3 . The orange colour represents eigenvector components aligned with activity (high values for the PCA feature indicate high activity), while blue eigenvector components had a verse opposite to the activity counts (low values for the PCA feature correspond to high activity at that time). Since the first two and fifth components were the main driving factors for classification and since they were the only ones with definite orientations in the latent time space, we will comment only those. High values of PCA feature 1, which drove the classification towards the HC class, corresponded to intense activity during daytime hours (between 7am and 9pm); conversely, sleep times were not relevant. For component 2, high values of the PCA corresponded to intense activity in the evening hours (from 6pm to 11 pm), while low values of PCA corresponded to intense activity in the early morning (from 6 am to 10 am). The fifth PCA component instead mirrored different weekly patterns between HCs and SSDs patients, as lower daytime activities for working weekdays (Monday to Friday) and higher daytime activities during weekend days (Saturday and Sunday) drove the classification toward the HC class. 4. Discussion In this work we used actigraphy data from DiAPASon and PSYKOSE datasets [ 19 , 20 ] , comprising more than 250 individuals, to inform and validate a classification model employing PCA and SVM to distinguish physical activity patterns of people with SSDs and healthy controls with a classification accuracy of 0.86 on the training data, which is slightly subpar compared with other classification models for SSD based on activity data [ 23 , 24 , 25 ] . However, most of the current works are affected by major methodological limitations influencing their scientific relevance. Chiefly, models are trained on small datasets (< 50 individuals) that lack variety [ 26 , 27 ] . Also, actigraphy-based ML models lack validation on an independent test, introducing possible biases and overfit that prevent model generalizability [ 12 , 28 , 29 ] . In our work we validated the classification model on independent data from the PSYKOSE initiative, achieving a classification accuracy of 0.80 that serves as a PoC that ML models are generalizable to unseen data. This result was possible due to the unprecedented sample size of DiAPASon for the training set compared to other similar works [ 12 , 23 , 25 ] , which provided sufficient variety for the model to learn daily movement dynamics. The lack of independent validation in previous works has largely stemmed from the poor comparability of different datasets, driven by discrepancies between actigraphy signals recorded with different device models [ 18 ] . In this study, we provide strong evidence that ML models, equipped with appropriate scaling methods, can be generalized to new actigraphy devices. In particular, the "match local average" rescaling method enabled unprecedented harmonization of actigraphy data from distinct devices by aligning the mean activity level at each corresponding timepoint, ensuring that the average value observed in PSYKOSE matched the average value in DiAPASon at every time point independently. This overcomes a long-standing barrier in actigraphy research. Unlike previous studies constrained to single-device datasets [ 30 ] , our approach supports meaningful comparison and merging of data from independent cohorts. This methodological advance opens the door to more reliable cross-device assessments of psychomotor activity and is poised to strengthen digital phenotyping efforts [ 23 , 31 ] in SSDs beyond current capabilities. Nevertheless, further validation on a wider array of devices is needed to fully establish the generalizability of this approach. By prying on explainable AI methods, namely SHAP analysis, we identified the time periods that drove classification. Healthy controls consistently showed higher activity volumes during the daytime and evening hours, while individuals with SSD were more active in the early morning. Also, we detected a distinctive weekly pattern characterized by an increased activity of HCs during the weekend. These results align with previous literature reporting reduced activity volume in individuals with SSDs due to psychomotor slowing [ 32 ] . However, the increased early morning activity observed in the SSDs group may be because some participants were living in residential facilities, which follow precise daily routines. Interestingly, no influence of sleep activity was relevant for the classification [ 23 ] , which could be due to the lower amplitude of nighttime activity compared to daytime activity or device limitations. Further research focusing specifically on sleep hours could be conducted. Compared to previous works that used actigraphy-derived metrics for classification [ 28 ] , our work employs the whole activity track for classification, allowing for a more fine-grained characterization of activity patterns. Collecting data passively with an actigraphy device, given further validation, could be extremely important for implementing classification models on wearable technologies capable of detecting disease-like activity patterns. A methodological limitation of this work might be the method of comparison of actigraphy counts between datasets, as our definition of inter-cohort concordance assumed independent measures while actigraphy is characterized by cyclic measures and pseudo-replication. However, our approximation of actigraphy counts as independent measures was justified by the high-dimensional nature of data, that prevented more refined comparison models from reaching convergence. The major limitation of this work is that the generalizability of the model has been tested on only a single independent actigraphy device, further validation on other devices must be performed. With the availability of open-access actigraphy datasets, classification models are beginning to attract interest for other psychiatric conditions [ 33 , 34 ] or in other fields [ 35 ] . It would also be valuable to investigate model generalizability for these cases. 5. Conclusions This work is one of the first examples of ML classification models solely based on passively-recorded actigraphy data capable of identifying healthy-like and SSDs-like activity patterns with proven generalizability to independent data collected using different actigraphy devices. Explainability analyses revealed the most influent daytimes that drove classification, returning differences in activity patterns between healthy control and individuals with SSDs that confirm clinical observations. Results are a necessary PoC towards employment of classification models based on passively collected data that may be implemented in mobile devices. Declarations Data Availability Statement: Data from the DiAPASon Initiative can be made available upon reasonable request, PSYKOSE data is publicly available at: https://datasets.simula.no/psykose/ Acknowledgements: The present work was partially supported by “Ministero della Salute”, IRCCS Research Program, Ricerca Corrente - Linea n. 1 “Utilizzo di strumenti di Intelligenza Artificiale (AI) per l’analisi dei disturbi psichici”. The authors thank also the DiAPASon consortium (DAily time use, Physical Activity, quality of care and interpersonal relationships in patients with Schizophrenia spectrum disorders) for supporting and enabling this study. Funding: The present work received no funding Conflicts of interest: The authors have no conflicts of interest to declare Author Contributions: DA: Conceptualization, Data curation, Formal Analysis, Methodology Software, Visualization, Writing – original draft, Writing – review and editing CMB : Conceptualization, Data curation, Writing – review and editing EC: Writing – review and editing MM: Writing – review and editing SC: Writing – review and editing GDG: Funding acquisition, Investigation, Methodology, Project administration, Writing – review and editing AR: Conceptualization, Supervision, Writing – review and editing References Dziwota, E., Stepulak, M. Z., Włoszczak-Szubzda, A. & Olajossy, M. Social functioning and the quality of life of patients diagnosed with schizophrenia. Ann. Agric. Environ. Med. 25 (1), 50–55. 10.5604/12321966.1233566 (2018). Freeman, D., Sheaves, B., Waite, F., Harvey, A. G. & Harrison, P. J. Sleep disturbance and psychiatric disorders. Lancet Psychiatry . 7 (7), 628–637. 10.1016/S2215-0366(20)30136-X (2020). Meyer, N. et al. Sleep and Circadian Rhythm Disturbance in Remitted Schizophrenia and Bipolar Disorder: A Systematic Review and Meta-analysis. Schizophr Bull. 46 (5), 1126–1143. 10.1093/schbul/sbaa024 (2020). Nadesalingam, N. et al. Motor abnormalities are associated with poor social and functional outcomes in schizophrenia. Compr. Psychiatr. 115 , 152307. 10.1016/j.comppsych.2022.152307 (2022). Liu, F., Wanigatunga, A. A. & Schrack, J. A. Assessment of Physical Activity in Adults Using Wrist Accelerometers. Epidemiol. Rev. 43 (1), 65–93. 10.1093/epirev/mxab004 (2021). Sallis, J. F. & Saelens, B. E. Assessment of physical activity by self-report: status, limitations, and future directions. Res. Q. Exerc. Sport . 71 (Suppl 2), 1–14. 10.1080/02701367.2000.11082780 (2000). Aronica, R. et al. Digital sleep phenotype and wrist actigraphy in individuals at clinical high risk for psychosis and people with schizophrenia spectrum disorders: a systematic review and meta-analysis. BMJ Ment Health . 28 (1), e301337. 10.1136/bmjment-2024-301337 (2025). Mayeli, A. et al. Sleep abnormalities in individuals at clinical high risk for psychosis. J. Psychiatr Res. 137 , 328–334. 10.1016/j.jpsychires.2021.03.006 (2021). Nadesalingam, N. et al. The Behavioral Mapping of Psychomotor Slowing in Psychosis Demonstrates Heterogeneity Among Patients Suggesting Distinct Pathobiology. Schizophr Bull. 49 (2), 507–517. 10.1093/schbul/sbac170 (2023). Wee, Z. Y. et al. Actigraphy studies and clinical and biobehavioural correlates in schizophrenia: a systematic review. J. Neural Transm (Vienna) . 126 (5), 531–558. 10.1007/s00702-019-01993-2 (2019). Benoit, J., Onyeaka, H., Keshavan, M. & Torous, J. Systematic Review of Digital Phenotyping and Machine Learning in Psychosis Spectrum Illnesses. Harv. Rev. Psychiatry . 28 (5), 296–304. 10.1097/HRP.0000000000000268 (2020). Price, G. D. et al. An unsupervised machine learning approach using passive movement data to understand depression and schizophrenia. J. Affect. Disord . 316 , 132–139. 10.1016/j.jad.2022.08.013 (2022). Jacobson, N. C., Weingarden, H. & Wilhelm, S. Digital biomarkers of mood disorders and symptom change. NPJ Digit. Med. 2 (3). 10.1038/s41746-019-0078-0 (2019). Jacobson, N. C., Lekkas, D., Huang, R. & Thomas, N. Deep learning paired with wearable passive sensing data predicts deterioration in anxiety disorder symptoms across 17–18 years. J. Affect. Disord . 282 , 104–111. 10.1016/j.jad.2020.12.086 (2021). Pedrelli, P. et al. Monitoring Changes in Depression Severity Using Wearable and Mobile Sensors. Front. Psychiatry . 11 , 584711. 10.3389/fpsyt.2020.584711 (2020). Maruani, J., Mauries, S., Zehani, F., Lejoyeux, M. & Geoffroy, P. A. Exploring actigraphy as a digital phenotyping measure: A study on differentiating psychomotor agitation and retardation in depression. Acta Psychiatr Scand. 151 (3), 401–411. 10.1111/acps.13739 (2025). Vecchio, I., Mifsud, L., Castro e Almeida, S. & Passecker, J. Diagnostic digital phenotyping in schizophrenia-spectrum disorders: a systematic review. npj Digit. Med. 9 10.1038/s41746-025-02194-w (2026). Liu, F. et al. Comparison of sleep parameters from wrist-worn ActiGraph and Actiwatch devices. Sleep 47 (2), zsad155. 10.1093/sleep/zsad155 (2024). de Girolamo, G. et al. DAily time use, Physical Activity, quality of care and interpersonal relationships in patients with Schizophrenia spectrum disorders (DiAPASon): an Italian multicentre study. BMC Psychiatry . 20 (1), 287. 10.1186/s12888-020-02588-y (2020). Jakobsen, P. et al. PSYKOSE: A Motor Activity Database of Patients with Schizophrenia. IEEE 33rd International Symposium on Computer-Based Medical Systems (CBMS) . 303–308. (2020). 10.1109/CBMS49503.2020.00064 (2020). Fasmer, O. B., Hauge, E., Berle, J. Ø., Dilsaver, S. & Oedegaard, K. J. Distribution of Active and Resting Periods in the Motor Activity of Patients with Depression and Schizophrenia. Psychiatry Investig . 13 (1), 112–120. 10.4306/pi.2016.13.1.112 (2016). Lundberg, S. M. & Lee, S. I. A unified approach to interpreting model predictions. Proceedings of the 31st International Conference on Neural Information Processing Systems . 4768–4777; (2017). 10.48550/arXiv.1705.07874 László, S. et al. The two ends of the spectrum: comparing chronic schizophrenia and premorbid latent schizotypy by actigraphy. BMC Psychiatry . 25 (1), 531. 10.1186/s12888-025-06971-5 (2025). Misgar, M. M. & Bhatia, M. P. S. Unveiling Psychotic Disorder Patterns: A Deep Learning Model Analysing Motor Activity Time-series Data With Explainable AI. Biomed. Signal Process. Control . 91 , 106000. 10.1016/j.bspc.2024.106000 (2024). Nagy, Á. et al. The Actigraphy-Based Identification of Premorbid Latent Liability of Schizophrenia and Bipolar Disorder. Sens. (Basel) . 23 (2), 958. 10.3390/s23020958 (2023). Foteinopoulou, N. M. & Patras, I. Machine learning approaches for fine-grained symptom estimation in schizophrenia: A comprehensive review. Artif. Intell. Med. 165 , 103129. 10.1016/j.artmed.2025.103129 (2025). Idowu, O., Aderinto, N., Olatunji, G. & Kokori, E. Machine Learning in Schizophrenia: A Systematic Review and Meta-Analysis of Diagnostic and Predictive Models. BJPsych Open. 11 (S1), S44. 10.1192/bjo.2025.10148 (2025). Krane-Gartiser, K., Henriksen, T. E. G., Morken, G., Vaaler, A. E. & Fasmer, O. B. Motor activity patterns in acute schizophrenia and other psychotic disorders can be differentiated from bipolar mania and unipolar depression. Psychiatry Res. 270 , 418–425. 10.1016/j.psychres.2018.10.004 (2018). Umbricht, D., Cheng, W. Y., Lipsmeier, F., Bamdadian, A. & Lindemann, M. Deep Learning-Based Human Activity Recognition for Continuous Activity and Gesture Monitoring for Schizophrenia Patients With Negative Symptoms. Front. Psychiatry . 11 , 574375. 10.3389/fpsyt.2020.574375 (2020). Tahmasian, M., Khazaie, H., Golshani, S. & Avis, K. T. Clinical application of actigraphy in psychotic disorders: a systematic review. Curr. Psychiatry Rep. 15 (6), 359. 10.1007/s11920-013-0359-2 (2013). Pieters, L. E., Deenik, J., de Vet, S., Delespaul, P. & van Harten, P. N. Combining actigraphy and experience sampling to assess physical activity and sleep in patients with psychosis: A feasibility study. Front. Psychiatry . 14 , 1107812. 10.3389/fpsyt.2023.1107812 (2023). Morrens, M., Hulstijn, W. & Sabbe, B. Psychomotor slowing in schizophrenia. Schizophr Bull . 33(4), 1038–1053; (2007). 10.1093/schbul/sbl051 Garcia-Ceja, E. et al. OBF-Psychiatric, a motor activity dataset of patients diagnosed with major depression, schizophrenia, and ADHD. Sci. Data . 12 (1), 32. 10.1038/s41597-025-04384-3 (2025). Panchal, P. et al. Toward a Digital Future in Bipolar Disorder Assessment: A Systematic Review of Disruptions in the Rest-Activity Cycle as Measured by Actigraphy. Front. Psychiatry . 13 , 780726. 10.3389/fpsyt.2022.780726 (2022). Gramkow, M. H. et al. Diagnostic performance of actigraphy in Alzheimer's disease using a machine learning classifier - a cross-sectional memory clinic study. Alzheimers Res. Ther. 17 (1), 111. 10.1186/s13195-025-01751-5 (2025). Additional Declarations No competing interests reported. Supplementary Files ActigraphySubmittedR1Supplementary.pdf Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 17 May, 2026 Reviewers agreed at journal 10 May, 2026 Reviewers agreed at journal 07 May, 2026 Reviewers agreed at journal 07 May, 2026 Reviewers agreed at journal 07 May, 2026 Reviewers invited by journal 05 May, 2026 Editor assigned by journal 05 May, 2026 Editor invited by journal 04 May, 2026 Submission checks completed at journal 30 Apr, 2026 First submitted to journal 30 Apr, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9493731","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":637693938,"identity":"e5704dc0-fb3e-46a9-aa4a-d70e7921f36e","order_by":0,"name":"Damiano Archetti","email":"data:image/png;base64,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","orcid":"","institution":"IRCCS Istituto Centro San Giovanni di Dio Fatebenefratelli","correspondingAuthor":true,"prefix":"","firstName":"Damiano","middleName":"","lastName":"Archetti","suffix":""},{"id":637693939,"identity":"c24e482d-ba1b-4e59-be3b-ee049694a801","order_by":1,"name":"Cesare Michele Baronio","email":"","orcid":"","institution":"IRCCS Istituto Centro San Giovanni di Dio Fatebenefratelli","correspondingAuthor":false,"prefix":"","firstName":"Cesare","middleName":"Michele","lastName":"Baronio","suffix":""},{"id":637693940,"identity":"9d926c97-5c1a-4961-8133-43dfde9de91e","order_by":2,"name":"Elisa Caselani","email":"","orcid":"","institution":"IRCCS Istituto Centro San Giovanni di Dio Fatebenefratelli","correspondingAuthor":false,"prefix":"","firstName":"Elisa","middleName":"","lastName":"Caselani","suffix":""},{"id":637693941,"identity":"1cfc4c08-bd58-40e9-aaec-a6fe0c4b43e4","order_by":3,"name":"Marta Magno","email":"","orcid":"","institution":"IRCCS Istituto Centro San Giovanni di Dio Fatebenefratelli","correspondingAuthor":false,"prefix":"","firstName":"Marta","middleName":"","lastName":"Magno","suffix":""},{"id":637693944,"identity":"0fb8aad9-82c5-4a09-accb-707dc6b3d5e6","order_by":4,"name":"Stefano Calza","email":"","orcid":"","institution":"University of Brescia","correspondingAuthor":false,"prefix":"","firstName":"Stefano","middleName":"","lastName":"Calza","suffix":""},{"id":637693947,"identity":"1b3a975f-3621-4878-a7ca-891b0d3891d6","order_by":5,"name":"Giovanni De Girolamo","email":"","orcid":"","institution":"IRCCS Istituto Centro San Giovanni di Dio Fatebenefratelli","correspondingAuthor":false,"prefix":"","firstName":"Giovanni","middleName":"","lastName":"De Girolamo","suffix":""},{"id":637693948,"identity":"1f5c63c6-5955-4897-bacd-777cc4b09bb9","order_by":6,"name":"Alberto Redolfi","email":"","orcid":"","institution":"IRCCS Istituto Centro San Giovanni di Dio Fatebenefratelli","correspondingAuthor":false,"prefix":"","firstName":"Alberto","middleName":"","lastName":"Redolfi","suffix":""}],"badges":[],"createdAt":"2026-04-22 09:12:25","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9493731/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9493731/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":109296260,"identity":"8515bc88-4c28-40bb-8275-ffb8f5fa1847","added_by":"auto","created_at":"2026-05-15 08:46:24","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":113464,"visible":true,"origin":"","legend":"\u003cp\u003eDistribution of individuals on the first and fifth components of the PCA space. (a) HC and SSDs individuals from DiAPASon; (b) predicted labels of individuals from DiAPASon; (a) HC and SSDs individuals from PSYKOSE; (b) predicted labels of individuals from PSYKOSE.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-9493731/v1/31565b5a9fdd2d93ec2e1de2.png"},{"id":109263834,"identity":"de2889e5-74fb-409d-87b6-2bcdb13e748c","added_by":"auto","created_at":"2026-05-14 12:08:02","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":86899,"visible":true,"origin":"","legend":"\u003cp\u003eBeeswarm plots of Shap analysis for DiAPASon (a) and PSYKOSE (b) individuals.\u003c/p\u003e\n\u003cp\u003eAcronyms: HC= Healthy Control; SSDs= Schizophrenia Spectrum Disorder patient.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-9493731/v1/710105b5c47b6dca1c412ef5.png"},{"id":109263835,"identity":"b57b66d9-afbb-4a13-9067-e5ffcf620ab1","added_by":"auto","created_at":"2026-05-14 12:08:02","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":193164,"visible":true,"origin":"","legend":"\u003cp\u003eeigenvectors of each PCA direction ordered by SHAP importance. Colors indicate the magnitude (transparency intensity) and verse (orange for positive and blue for negative) of each component of the eigenvector.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-9493731/v1/069f1e221619ff0bd8b12a25.png"},{"id":109297309,"identity":"69419cad-97f2-4aa2-91ef-71b76858dfb4","added_by":"auto","created_at":"2026-05-15 08:56:01","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":632508,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9493731/v1/9c380bd6-c2b3-44d5-88db-b4b3b27ad616.pdf"},{"id":109263832,"identity":"cf48c64d-035e-47e8-9f9a-df40ae734321","added_by":"auto","created_at":"2026-05-14 12:08:02","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":140725,"visible":true,"origin":"","legend":"","description":"","filename":"ActigraphySubmittedR1Supplementary.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9493731/v1/d011eedecc10dcc362e2b3b2.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"\u003cp\u003eIdentifying Activity Patterns From Different Actigraphy Devices in People With Schizophrenia Spectrum Disorders: A Comparative Study\u003c/p\u003e","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eSchizophrenia spectrum disorders (SSDs) are severe and highly disabling mental disorders characterized by marked clinical heterogeneity, encompassing positive and negative symptoms, cognitive impairment, and profound disturbances in everyday functioning\u003csup\u003e[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]\u003c/sup\u003e. Among the less visible yet clinically meaningful dimensions of these disorders are abnormalities in motor activity patterns and circadian rhythms\u003csup\u003e[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]\u003c/sup\u003e, which reflect the integrated functioning of neurobiological systems, psychopathology, and real-world behavior. In recent years, actigraphy has emerged as a reliable, non-invasive, and ecologically valid method for the continuous assessment of motor activity in naturalistic settings\u003csup\u003e[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]\u003c/sup\u003e, overcoming several limitations of traditional clinical assessments based on episodic observation or self-report\u003csup\u003e[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eA growing body of evidence indicates that individuals with SSDs exhibit distinct actigraphic signatures compared with healthy controls, including reduced overall activity levels, increased fragmentation of motor behavior, circadian irregularity, and altered temporal distribution of movement across the 24-hour cycle\u003csup\u003e[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]\u003c/sup\u003e. These features have been linked to negative symptom severity, cognitive dysfunction, social withdrawal, and poorer functional outcomes\u003csup\u003e[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]\u003c/sup\u003e. Consequently, actigraphy has gained increasing relevance as an objective behavioral marker with potential applications in diagnosis, longitudinal monitoring, and outcome prediction.\u003c/p\u003e \u003cp\u003eParallel to these developments, the application of Machine Learning (ML) techniques to actigraphy data has opened new avenues for identifying complex activity patterns that may not be detectable through conventional statistical approaches\u003csup\u003e[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]\u003c/sup\u003e. Supervised and unsupervised models have been used to distinguish between diagnostic groups\u003csup\u003e[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]\u003c/sup\u003e, characterize symptom profiles\u003csup\u003e[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003e, and explore behavioral phenotypes across different mental disorders\u003csup\u003e[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]\u003c/sup\u003e. Within the broader framework of digital phenotyping, such approaches hold promise for advancing precision psychiatry by providing scalable and data-driven tools capable of capturing clinically meaningful behavioral information.\u003c/p\u003e \u003cp\u003eDespite this progress, a major methodological challenge limits the translational potential of actigraphy-based machine learning models: the substantial heterogeneity in data acquisition protocols and devices. Actigraphy devices differ in sensor technology, sampling frequency, signal processing pipelines, and proprietary algorithms, leading to systematic discrepancies in recorded activity levels. As a result, models trained on data from a single device or cohort often show reduced performance when applied to external datasets, undermining their generalizability and clinical utility\u003csup\u003e[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]\u003c/sup\u003e. This issue is particularly critical for multi-center studies and for real-world implementation, where different devices are frequently used.\u003c/p\u003e \u003cp\u003eWithin this context, the present study aims to evaluate the generalizability of actigraphy-based ML models for distinguishing between healthy controls and individuals with SSDs using data derived from two independent datasets\u003csup\u003e[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/sup\u003e collected with different actigraphy devices. By combining dimensionality reduction techniques with supervised classification and implementing a rescaling procedure to harmonize activity signals, this work provides a proof of concept (PoC) that meaningful motor activity patterns can be transferred across devices. Ultimately, these findings contribute to the growing evidence base supporting the feasibility of device-agnostic digital phenotyping approaches in SSDs and highlight both the potential and the remaining challenges for their integration into clinical research and practice.\u003c/p\u003e"},{"header":"2. Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Participants\u003c/h2\u003e \u003cp\u003eData for this work were acquired from the DiAPASon\u003csup\u003e[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]\u003c/sup\u003e and PSYKOSE\u003csup\u003e[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]\u003c/sup\u003e datasets. Diapason is a multi-centric study carried out in northern Italy between 2020 and 2021. The study was carried out in accordance to national and institutional committees on human experimentation, as well as the Helsinki Declaration, and was approved by local Ethical Committees\u003csup\u003e[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]\u003c/sup\u003e. PSYKOSE is a fully anonymized publicly available dataset that comprises data from individuals enrolled at Haukeland University Hospital, Norway. The study was approved by the local Ethical Committee and carried out in accordance with local regulations\u003csup\u003e[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/sup\u003e. All participants signed an informed consent form, please refer to the original works for detailed information about recruitment\u003csup\u003e[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eOnly individuals with SSDs and healthy controls (HC) were considered. Additional inclusion criteria were the availability of actigraphy data spanning at least seven consecutive days.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Actigraphy data\u003c/h2\u003e \u003cp\u003eActigraphy data were collected with the ActiGraph GT9X Link for a continuative period of one week for DiAPASon and with Actiwatch AW4 for a continuative period of two weeks for PSYKOSE. The sampling frequency was set to 30Hz for ActiGraph GT9X and 32 Hz for Actiwatch AW4, and actigraphy counts were computed separately with the proprietary software for each device in 60-second epochs. All actigraphy tracks were trimmed to the same length of 10,080 timepoints (one per minute), starting at 00.00 AM of Monday and ending at 11.59 PM of the Sunday. Actigraphy tracks with zero counts for extended periods were excluded, while sporadic missing timepoints were imputed as the average of that timepoint for the respective diagnostic class separately for each cohort. In the case of DiAPASon, where actigraphy recordings were available for one week only, the available days were reordered to match the Monday-Sunday ordering. In the case of PSYKOSE, where actigraphy recordings were available for two weeks, we picked the first available Monday-Sunday span.\u003c/p\u003e \u003cp\u003eAs the actigraphy data were collected with different devices, possible biases could arise\u003csup\u003e[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]\u003c/sup\u003e, therefore we tried several methods to rescale PSYKOSE data to match activity distributions of DiAPASon data. Specifically, we tried the following:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003e \u003cem\u003eMatch punctual average\u003c/em\u003e: for each timepoint of the actigraphy track, PSYKOSE counts were multiplied by a factor that matched the average value observed in PSYKOSE at that timepoint to the average value observed in DiAPASon at that timepoint.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e \u003cem\u003eMatch local average\u003c/em\u003e: for each timepoint of the actigraphy track, PSYKOSE counts were multiplied by a factor that matched the average value observed in PSYKOSE in an interval centered at that timepoint to the average value observed in DiAPASon in an interval centered at that timepoint. Intervals with radii of 5, 10, 30 and 60 minutes timepoints were explored.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e \u003cem\u003eMatch global average\u003c/em\u003e: data of all timepoints of the PSYKOSE actigraphy tracks were multiplied by a factor that matched the average value observed globally in PSYKOSE to the average value observed globally in DiAPASon;\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e \u003cem\u003eLinear model\u003c/em\u003e: PSYKOSE data underwent a linear transformation so that the relation between timepoint-specific count averages in DiAPASon and PSYKOSE was an identity;\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e \u003cem\u003eLinear model\u0026thinsp;+\u0026thinsp;floor\u003c/em\u003e: same as linear model, but after the linear transformation all negative counts were set to 0.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eAll cases were tested either using HCs from DiAPASon as a reference. The goodness of rescaling was assessed by measuring concordance, calculated as the fraction of statistical comparisons between distributions of actigraphy counts at the same time-point that were not significantly different between DiAPASon and PSYKOSE. Only PSYKOSE data scaled with the method that reached the best concordance were considered for subsequent analyses.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Activity pattern classification\u003c/h2\u003e \u003cp\u003eTo avoid possible model overfitting caused by the high dimensionality of actigraphy data, and to identify a model capable of differentiating between healthy-like and SSD-like activity patterns, a grid-search over dimensionality reduction algorithms and over Support Vector Machine (SVM) classifiers was performed. Specifically, we cycled over principal component analysis (PCA), Uniform Manifold Approximation and Projection (UMAP) and T-distributed Stochastic Neighbor Embedding (TSNE) dimensionality reduction algorithms, and over different SVM kernels (linear, polynomial, RBM), all with multiple parameters, producing a total of 1,584 possible combinations. Data from DiAPASon were used as a training set, and only the model that achieved the best classification accuracy over repeated k-fold cross validation (50 repetitions, 10 folds for each repetition) was considered. Generalizability of the best performing model was tested by classifying individuals from PSYKOSE that served as independent test set.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Model explainability and Statistical Analysis\u003c/h2\u003e \u003cp\u003eTo enhance the explainability of the model a SHAPLEY analysis\u003csup\u003e[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/sup\u003e was performed to investigate which components of the embedded space were the most influential for the classification task.\u003c/p\u003e \u003cp\u003ePopulation-level statistical comparisons were performed with chi-squared test for categorical variables and with t-test for continuous variables. Inter-cohort statistical comparisons between distributions of actigraphy counts for a specific timepoint was performed with the Kolmogorov-Smirnov test, and before computing concordance a correction against multiple tests comparison was performed via Benjamini-Hochberg false discovery rate. Significance level for each of the statistical tests was set at α\u0026thinsp;=\u0026thinsp;0.05.\u003c/p\u003e \u003cp\u003eModelling and all statistical analyses were performed with python version 3.10.10, specifically, scikit-learn 1.4.0 was used for machine learning models and scipy 1.12.0 for statistical analyses.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Results","content":"\u003cp\u003eA total of 213 individuals from DiAPASon and 45 individuals from PSYKOSE were considered, relevant demographics are available in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e and detailed demographics of the DiAPASon sample are available in the Supplementary Table\u0026nbsp;1. Populations of HC and SSDs patients were mostly comparable inside either data cohort, with the only significant difference (p-value of chi squared test\u0026thinsp;=\u0026thinsp;0.0024) being the sex of PSYKOSE individuals, where male population was more numerous for SSDs patients than for healthy controls. The total populations of DiAPASon and PSYKOSE showed no significant differences. The SSDs population form DiAPASon was composed of either inpatients (61) or outpatients (51), this information was not available for the PSYKOSE dataset.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSocio-demographic characteristics of patients and healthy controls from DiAPASon and PSYKOSE\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eVariables\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e \u003cp\u003eDiAPASon\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003ePSYKOSE\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSSDs\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTOTAL\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eHC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSSDs\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eTOTAL\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSex, n (%)\u003c/b\u003e\u0026deg;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eFemale\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e42 (41.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e42 (37.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e84 (39.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e14 (58.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3 (14.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e17 (37.8%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eMale\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e59 (58.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e70 (62.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e129 (60.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e10 (41.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e18 (85.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e28 (62.2%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAge, n (%)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003e20\u0026ndash;29\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e16 (15.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e21 (18.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e37 (17.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e7 (29.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1 (4.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e8 (17.8%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003e30\u0026ndash;44\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e33 (32.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e40 (35.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e73 (34.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e8 (33.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e10 (47.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e18 (40.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003e\u0026ge;\u0026thinsp;45\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e52 (51.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e51 (45.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e103 (48.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e9 (37.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e10 (47.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e19 (42.2%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eBPRS Sum (average\u0026thinsp;\u0026plusmn;\u0026thinsp;SD)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e46\u0026thinsp;\u0026plusmn;\u0026thinsp;13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e49.4\u0026thinsp;\u0026plusmn;\u0026thinsp;8.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAcronyms: BPRS= brief psychiatric rating scale; HC= Healthy Control; n= total number; SD= standard deviation; SSDs= Schizophrenia Spectrum Disorders patient; *= significant differences between HC and SSDs patients inside DiAPASon, \u0026deg;= significant differences between HC and SSDs patients inside PSYKOSE; \u0026sect;= global significant differences between DiAPASon and PSYKOSE populations.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Data Scaling\u003c/h2\u003e \u003cp\u003eBefore scaling, DiAPASon and PSYKOSE data showed poor comparability, with an average concordance score between activity counts distributions equal to 30.7%. On average PSYKOSE data had lower activity counts than DiAPASon (1,600\u0026thinsp;\u0026plusmn;\u0026thinsp;2,200 for DiAPASon, 240\u0026thinsp;\u0026plusmn;\u0026thinsp;350 for PSYKOSE). The best performing actigraphy scaling method resulted to be the \u0026ldquo;\u003cem\u003eMatch local average\u0026rdquo;\u003c/em\u003e case on a 60 minutes radius interval, which produced an average concordance of 99.9% (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Other scaling methods produced relevant improvements as well, except for the \u003cem\u003e\u0026ldquo;linear model\u0026rdquo;\u003c/em\u003e method without flooring. For further analyses only PSYKOSE data scaled with the best performing method were considered.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eperformance of scaling methods.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eScaling\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eConcordance (HC)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eConcordance (SSDs)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo scaling (reference)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e28.4%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e33.0%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMatch punctual average\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e99.8%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e98.7%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMatch local average (5 minutes radius)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e100%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e99.5%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMatch local average (10 minutes radius)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e100%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e99.5%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMatch local average (30 minutes radius)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e100%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e99.6%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMatch local average (60 minutes radius)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e100%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e99.8%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMatch global average\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e99.0%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e99.2%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLinear model\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e41.0%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e17.9%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLinear model with flooring\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e97.0%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e96.4%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"3\"\u003eAcronyms: HC= Healthy Control; SSDs= Schizophrenia Spectrum Disorders patient\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Classification Model\u003c/h2\u003e \u003cp\u003eGrid search over dimensionality reduction methods and SVM parameters returned as the best performing model a PCA with 9 components followed by a SVM with RBF kernel (gamma\u0026thinsp;=\u0026thinsp;2.50e-11), which returned a classification accuracy of 0.860 (95% CI: [0.854\u0026ndash;0.866]) on a repeated k-fold cross validation (10 folds, 50 repetitions) on the DiAPASon training set.\u003c/p\u003e \u003cp\u003eScaled PSYKOSE data were projected in the PCA space defined on the DiAPASon training set and classified according to the SVM informed on DiAPASon (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). In this case classification accuracy was 0.80.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAcronyms: HC= Healthy Control; SSDs= Schizophrenia Spectrum Disorders patient; Acc= classification accuracy.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Model Explainability\u003c/h2\u003e \u003cp\u003eShapley analysis returned the first two PCA directions and the fifth PCA direction as the main driving factors for classification for both the training and test sets. High values for these features drove the classification towards the HC class, while low values drove the classification towards SSD class. The ranking of the other PCA features was almost the same for the two datasets except for the seventh and ninth directions, which however had only marginal contributions to classification (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAcronyms: HC= Healthy Control; SSDs= Schizophrenia Spectrum Disorder patient.\u003c/p\u003e \u003cp\u003eTo improve the explainability of the model, we computed the eigenvectors of the PCA directions, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe orange colour represents eigenvector components aligned with activity (high values for the PCA feature indicate high activity), while blue eigenvector components had a verse opposite to the activity counts (low values for the PCA feature correspond to high activity at that time). Since the first two and fifth components were the main driving factors for classification and since they were the only ones with definite orientations in the latent time space, we will comment only those. High values of PCA feature 1, which drove the classification towards the HC class, corresponded to intense activity during daytime hours (between 7am and 9pm); conversely, sleep times were not relevant. For component 2, high values of the PCA corresponded to intense activity in the evening hours (from 6pm to 11 pm), while low values of PCA corresponded to intense activity in the early morning (from 6 am to 10 am). The fifth PCA component instead mirrored different weekly patterns between HCs and SSDs patients, as lower daytime activities for working weekdays (Monday to Friday) and higher daytime activities during weekend days (Saturday and Sunday) drove the classification toward the HC class.\u003c/p\u003e \u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eIn this work we used actigraphy data from DiAPASon and PSYKOSE datasets\u003csup\u003e[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/sup\u003e, comprising more than 250 individuals, to inform and validate a classification model employing PCA and SVM to distinguish physical activity patterns of people with SSDs and healthy controls with a classification accuracy of 0.86 on the training data, which is slightly subpar compared with other classification models for SSD based on activity data\u003csup\u003e[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/sup\u003e. However, most of the current works are affected by major methodological limitations influencing their scientific relevance. Chiefly, models are trained on small datasets (\u0026lt;\u0026thinsp;50 individuals) that lack variety\u003csup\u003e[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]\u003c/sup\u003e. Also, actigraphy-based ML models lack validation on an independent test, introducing possible biases and overfit that prevent model generalizability\u003csup\u003e[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]\u003c/sup\u003e. In our work we validated the classification model on independent data from the PSYKOSE initiative, achieving a classification accuracy of 0.80 that serves as a PoC that ML models are generalizable to unseen data. This result was possible due to the unprecedented sample size of DiAPASon for the training set compared to other similar works\u003csup\u003e[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/sup\u003e, which provided sufficient variety for the model to learn daily movement dynamics.\u003c/p\u003e \u003cp\u003eThe lack of independent validation in previous works has largely stemmed from the poor comparability of different datasets, driven by discrepancies between actigraphy signals recorded with different device models\u003csup\u003e[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]\u003c/sup\u003e. In this study, we provide strong evidence that ML models, equipped with appropriate scaling methods, can be generalized to new actigraphy devices. In particular, the \"match local average\" rescaling method enabled unprecedented harmonization of actigraphy data from distinct devices by aligning the mean activity level at each corresponding timepoint, ensuring that the average value observed in PSYKOSE matched the average value in DiAPASon at every time point independently. This overcomes a long-standing barrier in actigraphy research. Unlike previous studies constrained to single-device datasets\u003csup\u003e[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]\u003c/sup\u003e, our approach supports meaningful comparison and merging of data from independent cohorts. This methodological advance opens the door to more reliable cross-device assessments of psychomotor activity and is poised to strengthen digital phenotyping efforts\u003csup\u003e[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]\u003c/sup\u003e in SSDs beyond current capabilities. Nevertheless, further validation on a wider array of devices is needed to fully establish the generalizability of this approach.\u003c/p\u003e \u003cp\u003eBy prying on explainable AI methods, namely SHAP analysis, we identified the time periods that drove classification. Healthy controls consistently showed higher activity volumes during the daytime and evening hours, while individuals with SSD were more active in the early morning. Also, we detected a distinctive weekly pattern characterized by an increased activity of HCs during the weekend. These results align with previous literature reporting reduced activity volume in individuals with SSDs due to psychomotor slowing\u003csup\u003e[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/sup\u003e. However, the increased early morning activity observed in the SSDs group may be because some participants were living in residential facilities, which follow precise daily routines. Interestingly, no influence of sleep activity was relevant for the classification\u003csup\u003e[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003c/sup\u003e, which could be due to the lower amplitude of nighttime activity compared to daytime activity or device limitations. Further research focusing specifically on sleep hours could be conducted.\u003c/p\u003e \u003cp\u003eCompared to previous works that used actigraphy-derived metrics for classification\u003csup\u003e[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]\u003c/sup\u003e, our work employs the whole activity track for classification, allowing for a more fine-grained characterization of activity patterns. Collecting data passively with an actigraphy device, given further validation, could be extremely important for implementing classification models on wearable technologies capable of detecting disease-like activity patterns.\u003c/p\u003e \u003cp\u003eA methodological limitation of this work might be the method of comparison of actigraphy counts between datasets, as our definition of inter-cohort concordance assumed independent measures while actigraphy is characterized by cyclic measures and pseudo-replication. However, our approximation of actigraphy counts as independent measures was justified by the high-dimensional nature of data, that prevented more refined comparison models from reaching convergence. The major limitation of this work is that the generalizability of the model has been tested on only a single independent actigraphy device, further validation on other devices must be performed. With the availability of open-access actigraphy datasets, classification models are beginning to attract interest for other psychiatric conditions\u003csup\u003e[\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]\u003c/sup\u003e or in other fields\u003csup\u003e[\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]\u003c/sup\u003e. It would also be valuable to investigate model generalizability for these cases.\u003c/p\u003e"},{"header":"5. Conclusions","content":"\u003cp\u003eThis work is one of the first examples of ML classification models solely based on passively-recorded actigraphy data capable of identifying healthy-like and SSDs-like activity patterns with proven generalizability to independent data collected using different actigraphy devices. Explainability analyses revealed the most influent daytimes that drove classification, returning differences in activity patterns between healthy control and individuals with SSDs that confirm clinical observations. Results are a necessary PoC towards employment of classification models based on passively collected data that may be implemented in mobile devices.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eData Availability Statement:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData from the DiAPASon Initiative can be made available upon reasonable request, PSYKOSE data is publicly available at: https://datasets.simula.no/psykose/\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe present work was partially supported by \u0026ldquo;Ministero della Salute\u0026rdquo;, IRCCS Research Program, Ricerca Corrente - Linea n. 1 \u0026ldquo;Utilizzo di strumenti di Intelligenza Artificiale (AI) per l\u0026rsquo;analisi dei disturbi psichici\u0026rdquo;. The authors thank also the DiAPASon consortium (DAily time use, Physical Activity, quality of care and interpersonal relationships in patients with Schizophrenia spectrum disorders) for supporting and enabling this study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe present work received no funding\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflicts of interest:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors have no conflicts of interest to declare\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDA:\u0026nbsp;\u003c/strong\u003eConceptualization, Data curation, Formal Analysis, Methodology Software, Visualization, Writing \u0026ndash; original draft, Writing \u0026ndash; review and editing\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCMB\u003c/strong\u003e\u003cstrong\u003e:\u0026nbsp;\u003c/strong\u003eConceptualization, Data curation, Writing \u0026ndash; review and editing\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEC:\u0026nbsp;\u003c/strong\u003eWriting \u0026ndash; review and editing\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMM:\u0026nbsp;\u003c/strong\u003eWriting \u0026ndash; review and editing\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSC:\u0026nbsp;\u003c/strong\u003eWriting \u0026ndash; review and editing\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eGDG:\u0026nbsp;\u003c/strong\u003eFunding acquisition, Investigation, Methodology, Project administration, Writing \u0026ndash; review and editing\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAR:\u0026nbsp;\u003c/strong\u003eConceptualization, Supervision, Writing \u0026ndash; review and editing\u003cstrong\u003e\u003cbr\u003e\u003c/strong\u003e\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eDziwota, E., Stepulak, M. Z., Włoszczak-Szubzda, A. \u0026amp; Olajossy, M. Social functioning and the quality of life of patients diagnosed with schizophrenia. \u003cem\u003eAnn. Agric. Environ. Med.\u003c/em\u003e \u003cb\u003e25\u003c/b\u003e (1), 50\u0026ndash;55. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.5604/12321966.1233566\u003c/span\u003e\u003cspan address=\"10.5604/12321966.1233566\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFreeman, D., Sheaves, B., Waite, F., Harvey, A. G. \u0026amp; Harrison, P. J. Sleep disturbance and psychiatric disorders. \u003cem\u003eLancet Psychiatry\u003c/em\u003e. \u003cb\u003e7\u003c/b\u003e (7), 628\u0026ndash;637. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/S2215-0366(20)30136-X\u003c/span\u003e\u003cspan address=\"10.1016/S2215-0366(20)30136-X\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMeyer, N. et al. Sleep and Circadian Rhythm Disturbance in Remitted Schizophrenia and Bipolar Disorder: A Systematic Review and Meta-analysis. \u003cem\u003eSchizophr Bull.\u003c/em\u003e \u003cb\u003e46\u003c/b\u003e (5), 1126\u0026ndash;1143. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/schbul/sbaa024\u003c/span\u003e\u003cspan address=\"10.1093/schbul/sbaa024\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNadesalingam, N. et al. Motor abnormalities are associated with poor social and functional outcomes in schizophrenia. \u003cem\u003eCompr. Psychiatr.\u003c/em\u003e \u003cb\u003e115\u003c/b\u003e, 152307. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.comppsych.2022.152307\u003c/span\u003e\u003cspan address=\"10.1016/j.comppsych.2022.152307\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, F., Wanigatunga, A. A. \u0026amp; Schrack, J. A. Assessment of Physical Activity in Adults Using Wrist Accelerometers. \u003cem\u003eEpidemiol. Rev.\u003c/em\u003e \u003cb\u003e43\u003c/b\u003e (1), 65\u0026ndash;93. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/epirev/mxab004\u003c/span\u003e\u003cspan address=\"10.1093/epirev/mxab004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSallis, J. F. \u0026amp; Saelens, B. E. Assessment of physical activity by self-report: status, limitations, and future directions. \u003cem\u003eRes. Q. Exerc. Sport\u003c/em\u003e. \u003cb\u003e71\u003c/b\u003e (Suppl 2), 1\u0026ndash;14. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1080/02701367.2000.11082780\u003c/span\u003e\u003cspan address=\"10.1080/02701367.2000.11082780\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2000).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAronica, R. et al. Digital sleep phenotype and wrist actigraphy in individuals at clinical high risk for psychosis and people with schizophrenia spectrum disorders: a systematic review and meta-analysis. \u003cem\u003eBMJ Ment Health\u003c/em\u003e. \u003cb\u003e28\u003c/b\u003e (1), e301337. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1136/bmjment-2024-301337\u003c/span\u003e\u003cspan address=\"10.1136/bmjment-2024-301337\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMayeli, A. et al. Sleep abnormalities in individuals at clinical high risk for psychosis. \u003cem\u003eJ. Psychiatr Res.\u003c/em\u003e \u003cb\u003e137\u003c/b\u003e, 328\u0026ndash;334. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jpsychires.2021.03.006\u003c/span\u003e\u003cspan address=\"10.1016/j.jpsychires.2021.03.006\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNadesalingam, N. et al. The Behavioral Mapping of Psychomotor Slowing in Psychosis Demonstrates Heterogeneity Among Patients Suggesting Distinct Pathobiology. \u003cem\u003eSchizophr Bull.\u003c/em\u003e \u003cb\u003e49\u003c/b\u003e (2), 507\u0026ndash;517. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/schbul/sbac170\u003c/span\u003e\u003cspan address=\"10.1093/schbul/sbac170\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWee, Z. Y. et al. Actigraphy studies and clinical and biobehavioural correlates in schizophrenia: a systematic review. \u003cem\u003eJ. Neural Transm (Vienna)\u003c/em\u003e. \u003cb\u003e126\u003c/b\u003e (5), 531\u0026ndash;558. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00702-019-01993-2\u003c/span\u003e\u003cspan address=\"10.1007/s00702-019-01993-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBenoit, J., Onyeaka, H., Keshavan, M. \u0026amp; Torous, J. Systematic Review of Digital Phenotyping and Machine Learning in Psychosis Spectrum Illnesses. \u003cem\u003eHarv. Rev. Psychiatry\u003c/em\u003e. \u003cb\u003e28\u003c/b\u003e (5), 296\u0026ndash;304. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/HRP.0000000000000268\u003c/span\u003e\u003cspan address=\"10.1097/HRP.0000000000000268\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePrice, G. D. et al. An unsupervised machine learning approach using passive movement data to understand depression and schizophrenia. \u003cem\u003eJ. Affect. Disord\u003c/em\u003e. \u003cb\u003e316\u003c/b\u003e, 132\u0026ndash;139. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jad.2022.08.013\u003c/span\u003e\u003cspan address=\"10.1016/j.jad.2022.08.013\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJacobson, N. C., Weingarden, H. \u0026amp; Wilhelm, S. Digital biomarkers of mood disorders and symptom change. \u003cem\u003eNPJ Digit. Med.\u003c/em\u003e \u003cb\u003e2\u003c/b\u003e (3). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41746-019-0078-0\u003c/span\u003e\u003cspan address=\"10.1038/s41746-019-0078-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJacobson, N. C., Lekkas, D., Huang, R. \u0026amp; Thomas, N. Deep learning paired with wearable passive sensing data predicts deterioration in anxiety disorder symptoms across 17\u0026ndash;18 years. \u003cem\u003eJ. Affect. Disord\u003c/em\u003e. \u003cb\u003e282\u003c/b\u003e, 104\u0026ndash;111. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jad.2020.12.086\u003c/span\u003e\u003cspan address=\"10.1016/j.jad.2020.12.086\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePedrelli, P. et al. Monitoring Changes in Depression Severity Using Wearable and Mobile Sensors. \u003cem\u003eFront. Psychiatry\u003c/em\u003e. \u003cb\u003e11\u003c/b\u003e, 584711. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fpsyt.2020.584711\u003c/span\u003e\u003cspan address=\"10.3389/fpsyt.2020.584711\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMaruani, J., Mauries, S., Zehani, F., Lejoyeux, M. \u0026amp; Geoffroy, P. A. Exploring actigraphy as a digital phenotyping measure: A study on differentiating psychomotor agitation and retardation in depression. \u003cem\u003eActa Psychiatr Scand.\u003c/em\u003e \u003cb\u003e151\u003c/b\u003e (3), 401\u0026ndash;411. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/acps.13739\u003c/span\u003e\u003cspan address=\"10.1111/acps.13739\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVecchio, I., Mifsud, L., Castro e Almeida, S. \u0026amp; Passecker, J. Diagnostic digital phenotyping in schizophrenia-spectrum disorders: a systematic review. \u003cem\u003enpj Digit. Med.\u003c/em\u003e \u003cb\u003e9\u003c/b\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41746-025-02194-w\u003c/span\u003e\u003cspan address=\"10.1038/s41746-025-02194-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2026).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, F. et al. Comparison of sleep parameters from wrist-worn ActiGraph and Actiwatch devices. \u003cem\u003eSleep\u003c/em\u003e \u003cb\u003e47\u003c/b\u003e (2), zsad155. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/sleep/zsad155\u003c/span\u003e\u003cspan address=\"10.1093/sleep/zsad155\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ede Girolamo, G. et al. DAily time use, Physical Activity, quality of care and interpersonal relationships in patients with Schizophrenia spectrum disorders (DiAPASon): an Italian multicentre study. \u003cem\u003eBMC Psychiatry\u003c/em\u003e. \u003cb\u003e20\u003c/b\u003e (1), 287. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12888-020-02588-y\u003c/span\u003e\u003cspan address=\"10.1186/s12888-020-02588-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJakobsen, P. et al. PSYKOSE: A Motor Activity Database of Patients with Schizophrenia. \u003cem\u003eIEEE 33rd International Symposium on Computer-Based Medical Systems (CBMS)\u003c/em\u003e. 303\u0026ndash;308. (2020). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1109/CBMS49503.2020.00064\u003c/span\u003e\u003cspan address=\"10.1109/CBMS49503.2020.00064\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFasmer, O. B., Hauge, E., Berle, J. \u0026Oslash;., Dilsaver, S. \u0026amp; Oedegaard, K. J. Distribution of Active and Resting Periods in the Motor Activity of Patients with Depression and Schizophrenia. \u003cem\u003ePsychiatry Investig\u003c/em\u003e. \u003cb\u003e13\u003c/b\u003e (1), 112\u0026ndash;120. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.4306/pi.2016.13.1.112\u003c/span\u003e\u003cspan address=\"10.4306/pi.2016.13.1.112\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLundberg, S. M. \u0026amp; Lee, S. I. A unified approach to interpreting model predictions. \u003cem\u003eProceedings of the 31st International Conference on Neural Information Processing Systems\u003c/em\u003e. 4768\u0026ndash;4777; (2017). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.48550/arXiv.1705.07874\u003c/span\u003e\u003cspan address=\"10.48550/arXiv.1705.07874\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eL\u0026aacute;szl\u0026oacute;, S. et al. The two ends of the spectrum: comparing chronic schizophrenia and premorbid latent schizotypy by actigraphy. \u003cem\u003eBMC Psychiatry\u003c/em\u003e. \u003cb\u003e25\u003c/b\u003e (1), 531. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12888-025-06971-5\u003c/span\u003e\u003cspan address=\"10.1186/s12888-025-06971-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMisgar, M. M. \u0026amp; Bhatia, M. P. S. Unveiling Psychotic Disorder Patterns: A Deep Learning Model Analysing Motor Activity Time-series Data With Explainable AI. \u003cem\u003eBiomed. Signal Process. Control\u003c/em\u003e. \u003cb\u003e91\u003c/b\u003e, 106000. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.bspc.2024.106000\u003c/span\u003e\u003cspan address=\"10.1016/j.bspc.2024.106000\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNagy, \u0026Aacute;. et al. The Actigraphy-Based Identification of Premorbid Latent Liability of Schizophrenia and Bipolar Disorder. \u003cem\u003eSens. (Basel)\u003c/em\u003e. \u003cb\u003e23\u003c/b\u003e (2), 958. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/s23020958\u003c/span\u003e\u003cspan address=\"10.3390/s23020958\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFoteinopoulou, N. M. \u0026amp; Patras, I. Machine learning approaches for fine-grained symptom estimation in schizophrenia: A comprehensive review. \u003cem\u003eArtif. Intell. Med.\u003c/em\u003e \u003cb\u003e165\u003c/b\u003e, 103129. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.artmed.2025.103129\u003c/span\u003e\u003cspan address=\"10.1016/j.artmed.2025.103129\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIdowu, O., Aderinto, N., Olatunji, G. \u0026amp; Kokori, E. Machine Learning in Schizophrenia: A Systematic Review and Meta-Analysis of Diagnostic and Predictive Models. \u003cem\u003eBJPsych Open.\u003c/em\u003e \u003cb\u003e11\u003c/b\u003e (S1), S44. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1192/bjo.2025.10148\u003c/span\u003e\u003cspan address=\"10.1192/bjo.2025.10148\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKrane-Gartiser, K., Henriksen, T. E. G., Morken, G., Vaaler, A. E. \u0026amp; Fasmer, O. B. Motor activity patterns in acute schizophrenia and other psychotic disorders can be differentiated from bipolar mania and unipolar depression. \u003cem\u003ePsychiatry Res.\u003c/em\u003e \u003cb\u003e270\u003c/b\u003e, 418\u0026ndash;425. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.psychres.2018.10.004\u003c/span\u003e\u003cspan address=\"10.1016/j.psychres.2018.10.004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUmbricht, D., Cheng, W. Y., Lipsmeier, F., Bamdadian, A. \u0026amp; Lindemann, M. Deep Learning-Based Human Activity Recognition for Continuous Activity and Gesture Monitoring for Schizophrenia Patients With Negative Symptoms. \u003cem\u003eFront. Psychiatry\u003c/em\u003e. \u003cb\u003e11\u003c/b\u003e, 574375. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fpsyt.2020.574375\u003c/span\u003e\u003cspan address=\"10.3389/fpsyt.2020.574375\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTahmasian, M., Khazaie, H., Golshani, S. \u0026amp; Avis, K. T. Clinical application of actigraphy in psychotic disorders: a systematic review. \u003cem\u003eCurr. Psychiatry Rep.\u003c/em\u003e \u003cb\u003e15\u003c/b\u003e (6), 359. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s11920-013-0359-2\u003c/span\u003e\u003cspan address=\"10.1007/s11920-013-0359-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePieters, L. E., Deenik, J., de Vet, S., Delespaul, P. \u0026amp; van Harten, P. N. Combining actigraphy and experience sampling to assess physical activity and sleep in patients with psychosis: A feasibility study. \u003cem\u003eFront. Psychiatry\u003c/em\u003e. \u003cb\u003e14\u003c/b\u003e, 1107812. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fpsyt.2023.1107812\u003c/span\u003e\u003cspan address=\"10.3389/fpsyt.2023.1107812\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMorrens, M., Hulstijn, W. \u0026amp; Sabbe, B. Psychomotor slowing in schizophrenia. \u003cem\u003eSchizophr Bull\u003c/em\u003e. 33(4), 1038\u0026ndash;1053; (2007). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/schbul/sbl051\u003c/span\u003e\u003cspan address=\"10.1093/schbul/sbl051\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGarcia-Ceja, E. et al. OBF-Psychiatric, a motor activity dataset of patients diagnosed with major depression, schizophrenia, and ADHD. \u003cem\u003eSci. Data\u003c/em\u003e. \u003cb\u003e12\u003c/b\u003e (1), 32. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41597-025-04384-3\u003c/span\u003e\u003cspan address=\"10.1038/s41597-025-04384-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePanchal, P. et al. Toward a Digital Future in Bipolar Disorder Assessment: A Systematic Review of Disruptions in the Rest-Activity Cycle as Measured by Actigraphy. \u003cem\u003eFront. Psychiatry\u003c/em\u003e. \u003cb\u003e13\u003c/b\u003e, 780726. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fpsyt.2022.780726\u003c/span\u003e\u003cspan address=\"10.3389/fpsyt.2022.780726\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGramkow, M. H. et al. Diagnostic performance of actigraphy in Alzheimer's disease using a machine learning classifier - a cross-sectional memory clinic study. \u003cem\u003eAlzheimers Res. Ther.\u003c/em\u003e \u003cb\u003e17\u003c/b\u003e (1), 111. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s13195-025-01751-5\u003c/span\u003e\u003cspan address=\"10.1186/s13195-025-01751-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-9493731/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9493731/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eActigraphy is a reliable and non-invasive method for assessing motor activity and how disease affects the daily activity of patients. Machine learning models applied to actigraphy data have been used to identify activity patterns in patients suffering from either major depressive disorder or Schizophrenia Spectrum Disorders (SSDs). However, heterogeneous collection protocols across different datasets significantly hinder generalizability, making direct comparison difficult.\u003c/p\u003e \u003cp\u003eIn this work, we used actigraphy data to inform a machine learning model that distinguishes between the activity patterns of healthy controls (HC) and patients with SSDs.\u003c/p\u003e \u003cp\u003eActigraphy recordings from a total of 258 subjects (126 HC and 132 with SSDs) from the DiAPASon and PSYKOSE datasets, each spanning one week, were used. Actigraphy patterns were classified using principal component analysis (PCA) followed by a support vector machine (SVM) classifier trained on DiAPASon data. The model was validated using independent data from PSYKOSE, collected with a different actigraphy device, after a rescaling procedure to match DiAPASon.\u003c/p\u003e \u003cp\u003eAfter rescaling, the concordance between DiAPASon and PSYKOSE was 99.9% (30.7% pre-rescaling). The best performing model achieved a classification accuracy of 0.86 for DiAPASon data and 0.80 for PSYKOSE, demonstrating the model\u0026rsquo;s generalizability. Using Shapley analysis, distinctive activity patterns driving classifications towards HC or SSDs classes were detected.\u003c/p\u003e \u003cp\u003eOur results stand as a proof of concept that machine learning models, able to identify activity patterns, can be generalized to other actigraphy devices, but further testing on more diverse datasets and devices is required.\u003c/p\u003e","manuscriptTitle":"Identifying Activity Patterns From Different Actigraphy Devices in People With Schizophrenia Spectrum Disorders: A Comparative Study","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-05-14 12:07:55","doi":"10.21203/rs.3.rs-9493731/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2026-05-17T13:18:54+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"106521130914288898463948119545953985536","date":"2026-05-10T15:19:25+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"201453480611002528479449350954612400834","date":"2026-05-07T16:23:54+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"66109078684833527074267346676032550793","date":"2026-05-07T15:32:05+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"291758507754959165627613569529128505412","date":"2026-05-07T14:51:44+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-05-05T14:36:15+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-05-05T14:25:16+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-05-04T10:48:10+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-04-30T08:32:27+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2026-04-30T07:36:35+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"59bb22a6-6464-4c8e-8736-e12882e57db9","owner":[],"postedDate":"May 14th, 2026","published":true,"recentEditorialEvents":[{"type":"editorInvitedReview","content":"","date":"2026-05-17T13:18:54+00:00","index":41,"fulltext":""},{"type":"reviewerAgreed","content":"106521130914288898463948119545953985536","date":"2026-05-10T15:19:25+00:00","index":40,"fulltext":""},{"type":"reviewerAgreed","content":"201453480611002528479449350954612400834","date":"2026-05-07T16:23:54+00:00","index":39,"fulltext":""},{"type":"reviewerAgreed","content":"66109078684833527074267346676032550793","date":"2026-05-07T15:32:05+00:00","index":38,"fulltext":""},{"type":"reviewerAgreed","content":"291758507754959165627613569529128505412","date":"2026-05-07T14:51:44+00:00","index":37,"fulltext":""},{"type":"reviewersInvited","content":"10","date":"2026-05-05T14:36:15+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-05-05T14:25:16+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-05-04T10:48:10+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-04-30T08:32:27+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2026-04-30T07:36:35+00:00","index":"","fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":68117869,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":68117870,"name":"Health sciences/Diseases"},{"id":68117871,"name":"Health sciences/Health care"},{"id":68117872,"name":"Physical sciences/Mathematics and computing"},{"id":68117873,"name":"Biological sciences/Neuroscience"}],"tags":[],"updatedAt":"2026-05-14T12:07:55+00:00","versionOfRecord":[],"versionCreatedAt":"2026-05-14 12:07:55","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9493731","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9493731","identity":"rs-9493731","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}