{"id":{"repo_id":"westminster","oai_identifier":"oai:westminsterresearch.westminster.ac.uk:x35wq"},"canonical_url":"https://search.dev.ndltd.org/etd/westminster/oai:westminsterresearch.westminster.ac.uk:x35wq","repository":{"repo_id":"westminster","name":"University of Westminster","base_url":"https://westminsterresearch.westminster.ac.uk/oai2"},"display":{"title":"Multimodal AI for Hospital Readmission Prediction Among Older Adults","abstract":"This thesis develops a multimodal artificial intelligence (AI) model to predict 30-day hospital readmission risk among older adults, including those receiving home care. The aim is to create a robust predictive framework that leverages comprehensive patient data collected during hospitalisation to enhance risk assessment and clinical decision-making. By integrating multiple data sources, this approach improves predictive accuracy and provides deeper insights into the key factors driving readmissions, making AI models more informed and interpretable by using the most available data in Electronic Health Records (EHR). This work uses various fusion strategies and integrates structured EHR data, temporal physiological measurements and unstructured clinical narratives in various fusion models to construct a holistic view of each patient. This thesis introduces several innovations to improve feature encoding, interpretability, and predictive performance. First, a score-based target encoder refines categorical feature representation, ensuring a more meaningful clinical data integration. Second, based on association rule mining, a rule-based drug-drug interaction (DDI) detection method identifies potential DDIs, offering additional insights into medication safety. And third, to better process unstructured clinical text, an enhanced ClinicalBERT model, trained on MIMIC-IV discharge summaries, extracts richer representations from narratives, improving the model’s ability to capture nuanced risk factors. Finally, several fusion strategies are evaluated to integrate diverse clinical data sources, and recommendations on fusion architectures are provided. These advancements significantly boost predictive performance, improving the detection of high-risk patients while reducing false negatives. A focused analysis of home care patients reveals distinct patterns of readmission risk, driven by higher comorbidity burdens, frequent transitions in care, and greater social support needs. These factors are often overlooked by conventional models, underscoring the importance of multimodal AI approaches that account for the interaction between medical, functional, and social health determinants. By integrating diverse sources of patient information, this research demonstrates the potential of multimodal AI models to enhance hospital readmission prediction. The findings contribute to the advancement of clinical decision support systems, enabling more proactive and personalised interventions to reduce avoidable readmissions and improve healthcare outcomes for elderly patients.","abstract_html":"This thesis develops a multimodal artificial intelligence (AI) model to predict 30-day hospital readmission risk among older adults, including those receiving home care. The aim is to create a robust predictive framework that leverages comprehensive patient data collected during hospitalisation to enhance risk assessment and clinical decision-making. By integrating multiple data sources, this approach improves predictive accuracy and provides deeper insights into the key factors driving readmissions, making AI models more informed and interpretable by using the most available data in Electronic Health Records (EHR). This work uses various fusion strategies and integrates structured EHR data, temporal physiological measurements and unstructured clinical narratives in various fusion models to construct a holistic view of each patient. This thesis introduces several innovations to improve feature encoding, interpretability, and predictive performance. First, a score-based target encoder refines categorical feature representation, ensuring a more meaningful clinical data integration. Second, based on association rule mining, a rule-based drug-drug interaction (DDI) detection method identifies potential DDIs, offering additional insights into medication safety. And third, to better process unstructured clinical text, an enhanced ClinicalBERT model, trained on MIMIC-IV discharge summaries, extracts richer representations from narratives, improving the model’s ability to capture nuanced risk factors. Finally, several fusion strategies are evaluated to integrate diverse clinical data sources, and recommendations on fusion architectures are provided. These advancements significantly boost predictive performance, improving the detection of high-risk patients while reducing false negatives. A focused analysis of home care patients reveals distinct patterns of readmission risk, driven by higher comorbidity burdens, frequent transitions in care, and greater social support needs. These factors are often overlooked by conventional models, underscoring the importance of multimodal AI approaches that account for the interaction between medical, functional, and social health determinants. By integrating diverse sources of patient information, this research demonstrates the potential of multimodal AI models to enhance hospital readmission prediction. The findings contribute to the advancement of clinical decision support systems, enabling more proactive and personalised interventions to reduce avoidable readmissions and improve healthcare outcomes for elderly patients.","abstract_has_math":false,"creators":["Nazyrova, Nodira"],"institution":"University of Westminster","degree_name":"Ph.D.","degree_level":"PhD thesis","degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":["Chahed, S.","Chaussalet, T.J.","Dwek, M."],"committee_chairs":[],"committee_members":[],"year":2025,"date_issued":"2025","date_published":"2025","updated_at":"2026-07-24T06:01:04Z","subjects":[],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier","label":"Identifier","values":["oai:westminsterresearch.westminster.ac.uk:x35wq"],"render_values":[{"text":"oai:westminsterresearch.westminster.ac.uk:x35wq","href":null,"code":true}]}]},"links":{"outbound_url":"https://doi.org/10.34737/x35wq","outbound_label":"DOI","outbound_source":"dc:identifier.doi"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Chahed, S.","Chaussalet, T.J.","Dwek, M."]},{"key":"dc:creator","label":"Author","values":["Nazyrova, Nodira"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2025"]},{"key":"dc:date.issued","label":"Date","values":["2025"]},{"key":"dc:publisher","label":"Institution","values":["University of Westminster"]},{"key":"dc:publisher.department","label":"Dc Publisher Department","values":["Computer Science and Engineering"]},{"key":"dc:publisher.institution","label":"Dc Publisher Institution","values":["University of Westminster"]},{"key":"dc:relation","label":"Dc Relation","values":["https://westminsterresearch.westminster.ac.uk/item/x35wq/multimodal-ai-for-hospital-readmission-prediction-among-older-adults"]},{"key":"dc:relation.isreferencedby","label":"Dc Relation Isreferencedby","values":["https://westminsterresearch.westminster.ac.uk/item/x35wq/multimodal-ai-for-hospital-readmission-prediction-among-older-adults"]},{"key":"dc:type","label":"Dc Type","values":["Thesis or dissertation"]},{"key":"dc:type.qualificationlevel","label":"Dc Type Qualificationlevel","values":["PhD thesis"]},{"key":"dc:type.qualificationname","label":"Dc Type Qualificationname","values":["Ph.D."]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["oai:westminsterresearch.westminster.ac.uk:x35wq"]},{"key":"dc:identifier.doi","label":"DOI","values":["https://doi.org/10.34737/x35wq"]},{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://westminsterresearch.westminster.ac.uk/download/0ef915b81185e49eb234d98236f98a76125bf04891e5d239af1c98839ca6c146/15536389/Nodira_Nazyrova_PhD_Thesis_Final.pdf"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["This thesis develops a multimodal artificial intelligence (AI) model to predict 30-day hospital readmission risk among older adults, including those receiving home care. The aim is to create a robust predictive framework that leverages comprehensive patient data collected during hospitalisation to enhance risk assessment and clinical decision-making. By integrating multiple data sources, this approach improves predictive accuracy and provides deeper insights into the key factors driving readmissions, making AI models more informed and interpretable by using the most available data in Electronic Health Records (EHR). This work uses various fusion strategies and integrates structured EHR data, temporal physiological measurements and unstructured clinical narratives in various fusion models to construct a holistic view of each patient. This thesis introduces several innovations to improve feature encoding, interpretability, and predictive performance. First, a score-based target encoder refines categorical feature representation, ensuring a more meaningful clinical data integration. Second, based on association rule mining, a rule-based drug-drug interaction (DDI) detection method identifies potential DDIs, offering additional insights into medication safety. And third, to better process unstructured clinical text, an enhanced ClinicalBERT model, trained on MIMIC-IV discharge summaries, extracts richer representations from narratives, improving the model’s ability to capture nuanced risk factors. Finally, several fusion strategies are evaluated to integrate diverse clinical data sources, and recommendations on fusion architectures are provided. These advancements significantly boost predictive performance, improving the detection of high-risk patients while reducing false negatives. A focused analysis of home care patients reveals distinct patterns of readmission risk, driven by higher comorbidity burdens, frequent transitions in care, and greater social support needs. These factors are often overlooked by conventional models, underscoring the importance of multimodal AI approaches that account for the interaction between medical, functional, and social health determinants. By integrating diverse sources of patient information, this research demonstrates the potential of multimodal AI models to enhance hospital readmission prediction. The findings contribute to the advancement of clinical decision support systems, enabling more proactive and personalised interventions to reduce avoidable readmissions and improve healthcare outcomes for elderly patients."]},{"key":"dc:description.abstract","label":"Abstract","values":["This thesis develops a multimodal artificial intelligence (AI) model to predict 30-day hospital readmission risk among older adults, including those receiving home care. The aim is to create a robust predictive framework that leverages comprehensive patient data collected during hospitalisation to enhance risk assessment and clinical decision-making. By integrating multiple data sources, this approach improves predictive accuracy and provides deeper insights into the key factors driving readmissions, making AI models more informed and interpretable by using the most available data in Electronic Health Records (EHR). This work uses various fusion strategies and integrates structured EHR data, temporal physiological measurements and unstructured clinical narratives in various fusion models to construct a holistic view of each patient. This thesis introduces several innovations to improve feature encoding, interpretability, and predictive performance. First, a score-based target encoder refines categorical feature representation, ensuring a more meaningful clinical data integration. Second, based on association rule mining, a rule-based drug-drug interaction (DDI) detection method identifies potential DDIs, offering additional insights into medication safety. And third, to better process unstructured clinical text, an enhanced ClinicalBERT model, trained on MIMIC-IV discharge summaries, extracts richer representations from narratives, improving the model’s ability to capture nuanced risk factors. Finally, several fusion strategies are evaluated to integrate diverse clinical data sources, and recommendations on fusion architectures are provided. These advancements significantly boost predictive performance, improving the detection of high-risk patients while reducing false negatives. A focused analysis of home care patients reveals distinct patterns of readmission risk, driven by higher comorbidity burdens, frequent transitions in care, and greater social support needs. These factors are often overlooked by conventional models, underscoring the importance of multimodal AI approaches that account for the interaction between medical, functional, and social health determinants. By integrating diverse sources of patient information, this research demonstrates the potential of multimodal AI models to enhance hospital readmission prediction. The findings contribute to the advancement of clinical decision support systems, enabling more proactive and personalised interventions to reduce avoidable readmissions and improve healthcare outcomes for elderly patients."]},{"key":"dc:title","label":"Title","values":["Multimodal AI for Hospital Readmission Prediction Among Older Adults"]}]}],"canonical_facts":{"dc:contributor.advisor":["Chahed, S.","Chaussalet, T.J.","Dwek, M."],"dc:creator":["Nazyrova, Nodira"],"dc:date":["2025"],"dc:date.issued":["2025"],"dc:description":["This thesis develops a multimodal artificial intelligence (AI) model to predict 30-day hospital readmission risk among older adults, including those receiving home care. The aim is to create a robust predictive framework that leverages comprehensive patient data collected during hospitalisation to enhance risk assessment and clinical decision-making. By integrating multiple data sources, this approach improves predictive accuracy and provides deeper insights into the key factors driving readmissions, making AI models more informed and interpretable by using the most available data in Electronic Health Records (EHR). This work uses various fusion strategies and integrates structured EHR data, temporal physiological measurements and unstructured clinical narratives in various fusion models to construct a holistic view of each patient. This thesis introduces several innovations to improve feature encoding, interpretability, and predictive performance. First, a score-based target encoder refines categorical feature representation, ensuring a more meaningful clinical data integration. Second, based on association rule mining, a rule-based drug-drug interaction (DDI) detection method identifies potential DDIs, offering additional insights into medication safety. And third, to better process unstructured clinical text, an enhanced ClinicalBERT model, trained on MIMIC-IV discharge summaries, extracts richer representations from narratives, improving the model’s ability to capture nuanced risk factors. Finally, several fusion strategies are evaluated to integrate diverse clinical data sources, and recommendations on fusion architectures are provided. These advancements significantly boost predictive performance, improving the detection of high-risk patients while reducing false negatives. A focused analysis of home care patients reveals distinct patterns of readmission risk, driven by higher comorbidity burdens, frequent transitions in care, and greater social support needs. These factors are often overlooked by conventional models, underscoring the importance of multimodal AI approaches that account for the interaction between medical, functional, and social health determinants. By integrating diverse sources of patient information, this research demonstrates the potential of multimodal AI models to enhance hospital readmission prediction. The findings contribute to the advancement of clinical decision support systems, enabling more proactive and personalised interventions to reduce avoidable readmissions and improve healthcare outcomes for elderly patients."],"dc:description.abstract":["This thesis develops a multimodal artificial intelligence (AI) model to predict 30-day hospital readmission risk among older adults, including those receiving home care. The aim is to create a robust predictive framework that leverages comprehensive patient data collected during hospitalisation to enhance risk assessment and clinical decision-making. By integrating multiple data sources, this approach improves predictive accuracy and provides deeper insights into the key factors driving readmissions, making AI models more informed and interpretable by using the most available data in Electronic Health Records (EHR). This work uses various fusion strategies and integrates structured EHR data, temporal physiological measurements and unstructured clinical narratives in various fusion models to construct a holistic view of each patient. This thesis introduces several innovations to improve feature encoding, interpretability, and predictive performance. First, a score-based target encoder refines categorical feature representation, ensuring a more meaningful clinical data integration. Second, based on association rule mining, a rule-based drug-drug interaction (DDI) detection method identifies potential DDIs, offering additional insights into medication safety. And third, to better process unstructured clinical text, an enhanced ClinicalBERT model, trained on MIMIC-IV discharge summaries, extracts richer representations from narratives, improving the model’s ability to capture nuanced risk factors. Finally, several fusion strategies are evaluated to integrate diverse clinical data sources, and recommendations on fusion architectures are provided. These advancements significantly boost predictive performance, improving the detection of high-risk patients while reducing false negatives. A focused analysis of home care patients reveals distinct patterns of readmission risk, driven by higher comorbidity burdens, frequent transitions in care, and greater social support needs. These factors are often overlooked by conventional models, underscoring the importance of multimodal AI approaches that account for the interaction between medical, functional, and social health determinants. By integrating diverse sources of patient information, this research demonstrates the potential of multimodal AI models to enhance hospital readmission prediction. The findings contribute to the advancement of clinical decision support systems, enabling more proactive and personalised interventions to reduce avoidable readmissions and improve healthcare outcomes for elderly patients."],"dc:identifier":["oai:westminsterresearch.westminster.ac.uk:x35wq"],"dc:identifier.doi":["https://doi.org/10.34737/x35wq"],"dc:identifier.uri":["https://westminsterresearch.westminster.ac.uk/download/0ef915b81185e49eb234d98236f98a76125bf04891e5d239af1c98839ca6c146/15536389/Nodira_Nazyrova_PhD_Thesis_Final.pdf"],"dc:publisher":["University of Westminster"],"dc:publisher.department":["Computer Science and Engineering"],"dc:publisher.institution":["University of Westminster"],"dc:relation":["https://westminsterresearch.westminster.ac.uk/item/x35wq/multimodal-ai-for-hospital-readmission-prediction-among-older-adults"],"dc:relation.isreferencedby":["https://westminsterresearch.westminster.ac.uk/item/x35wq/multimodal-ai-for-hospital-readmission-prediction-among-older-adults"],"dc:title":["Multimodal AI for Hospital Readmission Prediction Among Older Adults"],"dc:type":["Thesis or dissertation"],"dc:type.qualificationlevel":["PhD thesis"],"dc:type.qualificationname":["Ph.D."]},"updated_at":"2026-07-24T06:01:04Z"}