# Prediction and early detection of agitation and aggression in dementia using physiological data and artificial intelligence models - URL: https://alisanburak.com/en/wcp-2026-posterler/so177 - Language: en - Last updated: not specified - Page title: Prediction and early detection of agitation and aggression in dementia using physiological data and artificial intelligence models | WCP 2026 Summary > Multimodal physiological data obtained from a wearable device can predict episodes of agitation and aggression in dementia 10 minutes before onset with clinica… This page was translated automatically from Turkish. [Read the original](https://alisanburak.com/wcp-2026-posterler/so177) This page is an AI-assisted Turkish summary of the study presented at the 26th World Congress of Psychiatry. The study belongs to the presenting authors; please refer to the original source when citing it. SO177 Brief Oral Poster PresentationDigital Psychiatry and Artificial IntelligenceModel development using observational sensor data # Prediction and early detection of agitation and aggression in dementia using physiological data and artificial intelligence models PREDICTION AND EARLY IDENTIFICATION OF AGITATION AND AGGRESSION IN DEMENTIA USING PHYSIOLOGICAL DATA AND ARTIFICIAL INTELLIGENCE MODELS Presenter: Alexander Merkin Country: Germany Institution: Institute for Mental Health and Translational Neurosciences Session: 3030 - PARALLEL SHORT ORAL POSTER SESSIONS: GERIATRIC PSYCHIATRY 01 Date: Thu, 24.09.2026 Multimodal physiological data obtained from a wearable device can predict episodes of agitation and aggression in dementia 10 minutes before onset with clinically meaningful accuracy. The study aims to develop and comparatively evaluate deep learning models capable of identifying episodes of psychomotor agitation and aggression in dementia (PAAD) 10 minutes before onset. The data were drawn from previously published studies and include accelerometer, blood volume pulse, electrodermal activity, and skin temperature signals collected from 17 people with dementia over 140 monitoring days using the Empatica E4 wearable device; the dataset contains 182 clinically confirmed agitation events. Sixteen clinically meaningful features were extracted; three transformer-based models (Multimodal Clinical Transformer, Sensitivity Optimised Clinical Transformer, and Adaptive Ensemble) were compared with the Linear Complexity Physiological State Space Model (LC-PSSM), accounting for class imbalance, temporal shifts, and clinically relevant performance metrics. Despite the highly imbalanced dataset, LC-PSSM achieved the best performance across several dimensions, with an area under the ROC curve (AUC-ROC) of 0.7558, an area under the precision–recall curve (AUC-PR) of 0.5312, and a clinical score of 0.7062, while remaining computationally efficient. The abstract reports a sensitivity of 65.7% and a specificity of 70.6% for this model, while the sensitivity-optimised transformer achieved a sensitivity of 90.4% with moderate specificity; ensemble modelling yielded balanced performance. The poster figures show an ROC AUC of 0.724 and a PR AUC of 0.471 (baseline 0.250) for the transformer model. In conclusion, continuous physiological data enable the advance detection of PAAD episodes and lay the groundwork for a shift from reactive management to preventive, real-time intervention in dementia care. The authors note that validation in larger, multicentre datasets and integration into clinical workflows are required before clinical deployment. [View the original poster](https://cslide.ctimeetingtech.com/wcp26/attendee/eposter/file/2086) How to cite this study APA 7: Merkin, A., et al. (2026, September 24). Prediction and early identification of agitation and aggression in dementia using physiological data and artificial intelligence models [Poster presentation]. 26th WPA World Congress of Psychiatry (WCP 2026), Stockholm, Sweden. https://cslide.ctimeetingtech.com/wcp26/attendee/eposter/file/2086 Vancouver: Merkin A, et al. Prediction and early identification of agitation and aggression in dementia using physiological data and artificial intelligence models [poster]. 26th WPA World Congress of Psychiatry (WCP 2026); 2026 Sep 23-26; Stockholm, Sweden. Available from: https://cslide.ctimeetingtech.com/wcp26/attendee/eposter/file/2086 Author names and title formatting were generated automatically from the congress system; please check them against the original poster before use. ## On the same topic - [Siber tanıkların davranışını belirleyen etkenler](https://alisanburak.com/en/wcp-2026-posterler/ev0041) - [Z kuşağında sosyal medyanın "lunapark etkisi"nin modellenmesi: örtük değişken yaklaşımı](https://alisanburak.com/en/wcp-2026-posterler/ev0046) - [DEHB'de dijital girişimler: etkililik ve çok modlu tedavi yaklaşımlarına entegrasyon](https://alisanburak.com/en/wcp-2026-posterler/ev0072) - [Algoritmaların inşa ettiği beyinler: dijital çağda DEHB, patolojiyi mi tanılıyoruz yoksa yeni bir normali mi patolojikleştiriyoruz?](https://alisanburak.com/en/wcp-2026-posterler/ev0077) - [Genç erişkin bir örneklemde nomofobi ile yaygın anksiyete bozukluğu arasındaki ilişki](https://alisanburak.com/en/wcp-2026-posterler/ev0111) [All WCP 2026 summaries](https://alisanburak.com/en/wcp-2026-posterler) --- Kaynak / Source: https://alisanburak.com/en/wcp-2026-posterler/so177 İletişim / Contact: Bağdat Caddesi, Veli Güneysu Apt. No:453 K:4 D:9, Suadiye / Kadıköy, İstanbul · +90 532 642 95 25 · info@alisanburak.com · https://alisanburak.com/bagdat-caddesi-psikoterapi-iletisim