Zina Ben Miled

Objectives: To evaluate multimodal fusion strategies for postoperative delirium prediction using structured electronic health record (EHR) data and clinical notes. Materials and Methods: Using EHR data from patients undergoing major surgery (2013-2021), we compared unimodal and multimodal models. Structured data were modeled using XGBoost and Self-Attention and Intersample Attention Transformer (SAINT), and clinical notes using a fine-tuned Bioformer, a biomedical transformer language model. Multimodal approaches included feature-level, decision-level, gated, cross-attention, and representation-level fusion. Results: Representation-level fusion achieved the highest area under the receiver operating characteristic curve (AUC-ROC = 0.791), followed by decision-level fusion (0.780), gated fusion (0.774), feature-level fusion (0.772), and cross-attention fusion (0.765). Among unimodal models, Bioformer achieved an AUC-ROC of 0.761, while SAINT and XGBoost each achieved an AUC-ROC of 0.753. Discussion: Multimodal approaches consistently improved performance, suggesting complementary predictive information from structured variables and clinical narratives. Notably, representation-level fusion outperformed more complex fusion strategies despite its simpler architecture. Conclusion: Representation-level fusion achieved the best performance, supporting multimodal learning for postoperative delirium prediction.

UNSTRUCTURED Digital therapeutics are digital technology-based interventions designed to treat, manage, or prevent medical or behavioral conditions. Early products delivered fixed educational, behavioral, or monitoring content. Newer products increasingly resemble adaptive software ecosystems that incorporate large language models, multimodal real-time data, memory, retrieval, tool use, and multi-agent coordination. Such systems change in two distinct ways. The first is planned: a team updates the model, data, configuration, or permissions. Regulatory mechanisms such as predetermined change control plans govern these updates. The second is not planned. An agentic system can act differently at runtime with nothing newly deployed, because retrieval surfaces different context, memory accumulates, or agents select different tools. These changes cannot be enumerated in advance, and neither clinical stage models nor software governance frameworks specify how to develop, evaluate, or retire products that exhibit unplanned behavior. We propose an Agile Stage Model for minimally viable development and governance of digital therapeutics, including those built on agentic artificial intelligence. The model draws on the NIH Stage Model, software-as-a-medical device principles, Agile Science and its derivatives, the artificial intelligence agent literature, intelligent delegation, and recent work on agentic artificial intelligence for executing clinical trials. It extends the Agile Analytic Engine, an architecture that connects population evidence to a local treatment-to-target decision contract specifying target, baseline, action, reassessment interval, and response rule, and then to a governed chain of minimally viable experiments. The model comprises six stages: minimum viable specification (Stage 0), minimum viable product and internal validation (Stage 1), first-in-human sensor-action testing (Stage 2), local real-world minimally viable experiment (Stage 3), population effectiveness and cross-site learning (Stage 4), and lifecycle stewardship and deimplementation (Stage 5). The stages are learning gates rather than a fixed sequence, and teams may advance, repeat, revise, pause, or stop. At each stage the team asks a single question: what is the smallest valid experiment that can inform the next decision? For agentic systems, every stage adds explicit requirements for bounded authority, human oversight, permissioning, verification, and monitoring that separates degradation in performance from change in system behavior. One rule governs the model: standardize only what protects patient safety and the integrity of evidence and let everything else adapt to context. Oversight should scale with the severity, irreversibility, and reach of potential harm rather than being uniform across products. The model is conceptual and requires prospective evaluation.

Introduction: This study assesses the utility of integrating multimodal preoperative electronic health records (EHRs), including unstructured medical notes, for early postoperative delirium prediction. We aim to demonstrate the feasibility of this data integration to facilitate future expansion of the methodology towards risk stratification. Methods: Transformer-based models were trained and evaluated using EHR data (2013–2021) from the Indiana Network for Patient Care for patients undergoing major surgeries such as laparotomies, bowel resections and cardiac procedures. Delirium was assessed retrospectively using (1) International Classification of Diseases, 10th revision codes and (2) clinician/nurse assessments from both structured and unstructured documentation. Results: From an initial cohort of 35 128 control patients, postoperative delirium was identified in 3872 (assessment 1) and 15 351 (assessment 2) cases. Prevalent background comorbidities included pre-existing conditions such as hypertension, diabetes, hypothyroidism and chronic pulmonary disease. Following data balancing, our late fusion multimodal transformer model achieved an area under the receiver operating characteristic curve of 0.80. The model demonstrated increased sensitivity to patients with high disease burden, highlighting the potential of advanced architectures for using unstructured clinical data. Conclusions: Using up to 4 years of preoperative EHR data offers the significant advantage of earlier risk identification, facilitating enhanced preoperative surgical patient care. By successfully unlocking the nuanced clinical signals within preoperative unstructured data, this study provides a robust foundation for incorporating medical notes into early postoperative delirium prediction.

We present a scalable, AI-powered system that identifies and extracts evidence-based behavioral nudges from unstructured biomedical literature. Nudges are subtle, non-coercive interventions that influence behavior without limiting choice, showing strong impact on health outcomes like medication adherence. However, identifying these interventions from PubMed's 8 million+ articles is a bottleneck. Our system uses a novel multi-stage pipeline: first, hybrid filtering (keywords, TF-IDF, cosine similarity, and a "nudge-term bonus") reduces the corpus to about 81,000 candidates. Second, we use OpenScholar (quantized LLaMA 3.1 8B) to classify papers and extract structured fields like nudge type and target behavior in a single pass, validated against a JSON schema. We evaluated four configurations on a labeled test set (N=197). The best setup (Title/Abstract/Intro) achieved a 67.0% F1 score and 72.0% recall, ideal for discovery. A high-precision variant using self-consistency (7 randomized passes) achieved 100% precision with 12% recall, demonstrating a tunable trade-off for high-trust use cases. This system is being integrated into Agile Nudge+, a real-world platform, to ground LLM-generated interventions in peer-reviewed evidence. This work demonstrates interpretable, domain-specific retrieval pipelines for evidence synthesis and personalized healthcare.

Automated ultrasonic testing (AUT) serves as a vital method for evaluating critical infrastructure in industries such as oil and gas. However, a significant challenge in deploying artificial intelligence (AI)-based interpretation methods for AUT data lies in improving their reliability and effectiveness, particularly due to the inherent scarcity of real-world defective data. This study directly addresses data scarcity in a weld defect classification task, specifically concerning the detection of lack of fusion (LOF) defects in weld inspections using a proprietary industrial ultrasonic B-scan image dataset. This paper leverages state-of-the-art generative models, including Generative Adversarial Networks (GANs) and Denoising Diffusion Probabilistic Models (DDPM) (StyleGAN3, VQGAN with an unconditional transformer, and Stable Diffusion), to produce realistic B-scan images depicting LOF defects. The fine-tuned Transformer-based models, including ViT-Base, Swin-Tiny, and MobileViT-Small classifiers, on the regular B-scan image dataset are then applied to retain only high-confidence positive synthetic samples from each method. The impact of these synthetic images on the classification performance of a ResNet-50 model is evaluated, where it is fine-tuned with cumulative additions of synthetic images, ranging from 10 to 200 images. Its accuracy on the test set increases by 38.9% relative to the baseline with the addition of either 80 synthetic images using VQGAN with an unconditional transformer or 200 synthetic images by StyleGAN3 to the training set, and by 36.8% with the addition of 150 synthetic images by Stable Diffusion. This also outperforms Transformer-based vision models that are trained on regular training data. Concurrently, knowledge distillation experiments involve training ResNet-50 as a student model, leveraging the expertise of ViT-Base and Swin-Tiny as teacher models to demonstrate the effectiveness of adding the synthetic data to the training set, where the greatest enhancement is observed to be 34.7% relative to the baseline. This work contributes to advancing robust, AI-assisted tools for critical infrastructure inspection and offers practical pathways for enhancing available models in resource-constrained industrial environments.

Background: Most adults aged ≥ 65 years live with multiple chronic conditions (MCC), and nearly one in four have recognized or unrecognized Alzheimer’s disease and related dementias (ADRD), including an estimated 7.2 million Americans. Together, MCC and ADRD increase treatment complexity, medication burden, and the risk of adverse outcomes. Among patients who meet clinical criteria for mild cognitive impairment (MCI) or ADRD but lack a formal diagnosis, MCC burden remains unclear. This study examined the association between MCC burden and undiagnosed MCI and ADRD in a diverse cohort of older adults in primary care. Methods: We conducted a cross-sectional analysis of 324 adults aged ≥ 65 from primary care clinics in Indiana and South Florida (2021– 2023), as part of a larger ADRD detection study. Patients without documented MCI or ADRD completed standardized cognitive assessments. Cognitive status (normal, MCI, ADRD) was determined by interdisciplinary consensus. Chronic conditions and medications were extracted from electronic health records. Multinomial logistic regression was used to examine the association between MCC profiles and cognitive status. Results: Among 324 older adults, 51.9% were determined to have MCI and 8% ADRD. Patients with MCI and ADRD had more chronic conditions (mean = 5– 6) and medications (mean = 4– 5) than those with normal cognition (p < 0.001). Anticholinergic use was more common in the MCI (23.8%) and ADRD (23.1%) groups than in those with normal cognition (10.8%). In adjusted models, MCI and ADRD were associated with higher odds of having more chronic conditions. Cerebrovascular disease was associated with both MCI and ADRD; diabetes, sleep apnea, and insomnia with MCI; and ischemic heart disease and insomnia with ADRD. Conclusion: Older adults with unrecognized MCI and ADRD experience substantial MCC and medication burden. These findings highlight the need for targeted primary care interventions that integrate cognitive screening, support MCC management, optimize self-management capacity, and promote safer prescribing. Plain Language Summary: Many older adults live with multiple long-term health conditions and take several medications. Some also have memory or thinking problems, like mild cognitive impairment (MCI) or dementia, that have not yet been diagnosed. In this study, we looked at older adults receiving care at primary care clinics to understand how undiagnosed memory problems are related to other health issues. We found that people with unrecognized MCI or dementia had more chronic conditions and were taking more medications than those with normal memory. Certain health problems, like stroke, diabetes, and sleep disorders, were more common in people with memory issues. We also found that medications known to affect memory (called anticholinergics) were used more often in people with memory problems, especially in younger patients. These findings suggest that doctors and care teams should consider memory problems when treating older adults with complex health needs. By doing so, they can create safer care plans, choose better medications, and help patients manage their health more effectively.

Background: Electronic health records (EHR) are now widely available in healthcare institutions to document the medical history of patients as they interact with healthcare services. In particular, routine care EHR data are collected for a large number of patients.These data span multiple heterogeneous elements (i.e., demographics, diagnosis, medications, clinical notes, vital signs, and laboratory results) which contain semantic, concept, and temporal information. Recent advances in generative learning techniques were able to leverage the fusion of multiple routine care EHR data elements to enhance clinical decision support. Objective: A scoping review of the proposed techniques including fusion architectures, input data elements, and application areas is needed to synthesize variances and identify research gaps that can promote re-use of these techniques for new clinical outcomes. Design: A comprehensive literature search was conducted using Google Scholar to identify high impact fusion architectures over multi-modal routine care EHR data during the period 2018 to 2023. The guidelines from the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) extension for scoping review were followed. The findings were derived from the selected studies using a thematic and comparative analysis. Results: The scoping review revealed the lack of standard definition for EHR data elements as they are transformed into input modalities. These definitions ignore one or more key characteristics of the data including source, encoding scheme, and concept level. Moreover, in order to adapt to emergent generative learning techniques, the classification of fusion architectures should distinguish fusion from learning and take into consideration that learning can concurrently happen in all three layers of new fusion architectures (i.e., encoding, representation, and decision). These aspects constitute the first step towards a streamlined approach to the design of multi-modal fusion architectures for routine care EHR data. In addition, current pretrained encoding models are inconsistent in their handling of temporal and semantic information thereby hindering their re-use for different applications and clinical settings. Conclusions: Current routine care EHR fusion architectures mostly follow a design-by-example methodology. Guidelines are needed for the design of efficient multi-modal models for a broad range of healthcare applications. In addition to promoting re-use, these guidelines need to outline best practices for combining multiple modalities while leveraging transfer learning and co-learning as well as semantic and temporal encoding.

Background: Postoperative delirium (POD) is a common complication after major surgery and is associated with poor outcomes in older adults. Early identification of patients at high risk of POD can enable targeted prevention efforts. However, existing POD prediction models require inpatient data collected during the hospital stay, which delays predictions and limits scalability. Objective: This study aimed to develop and externally validate a machine learning-based prediction model for POD using routine electronic health record (EHR) data. Methods: We identified all surgical encounters from 2014 to 2021 for patients aged 50 years and older who underwent an operation requiring general anesthesia, with a length of stay of at least 1 day at 3 Indiana hospitals. Patients with preexisting dementia or mild cognitive impairment were excluded. POD was identified using Confusion Assessment Method records and delirium International Classification of Diseases (ICD) codes. Controls without delirium or nurse-documented confusion were matched to cases by age, sex, race, and year of admission. We trained logistic regression, random forest, extreme gradient boosting (XGB), and neural network models to predict POD using 143 features derived from routine EHR data available at the time of hospital admission. Separate models were developed for each hospital using surveillance periods of 3 months, 6 months, and 1 year before admission. Model performance was evaluated using the area under the receiver operating characteristic curve (AUROC). Each model was internally validated using holdout data and externally validated using data from the other 2 hospitals. Calibration was assessed using calibration curves. Results: The study cohort included 7167 delirium cases and 7167 matched controls. XGB outperformed all other classifiers. AUROCs were highest for XGB models trained on 12 months of preadmission data. The best-performing XGB model achieved a mean AUROC of 0.79 (SD 0.01) on the holdout set, which decreased to 0.69-0.74 (SD 0.02) when externally validated on data from other hospitals. Conclusions: Our routine EHR-based POD prediction models demonstrated good predictive ability using a limited set of preadmission and surgical variables, though their generalizability was limited. The proposed models could be used as a scalable, automated screening tool to identify patients at high risk of POD at the time of hospital admission.

BACKGROUND Postoperative delirium (POD) is a common complication after major surgery and is associated with poor outcomes in older adults. Early identification of patients at high risk of POD can enable targeted prevention efforts. However, existing POD prediction models require inpatient data collected during the hospital stay, which delays predictions and limits scalability. OBJECTIVE This study aimed to develop and externally validate a machine learning-based prediction model for POD using routine electronic health record (EHR) data. METHODS We identified all surgical encounters from 2014 to 2021 for patients aged 50 years and older who underwent an operation requiring general anesthesia, with a length of stay of at least 1 day at 3 Indiana hospitals. Patients with preexisting dementia or mild cognitive impairment were excluded. POD was identified using Confusion Assessment Method records and delirium International Classification of Diseases (ICD) codes. Controls without delirium or nurse-documented confusion were matched to cases by age, sex, race, and year of admission. We trained logistic regression, random forest, extreme gradient boosting (XGB), and neural network models to predict POD using 143 features derived from routine EHR data available at the time of hospital admission. Separate models were developed for each hospital using surveillance periods of 3 months, 6 months, and 1 year before admission. Model performance was evaluated using the area under the receiver operating characteristic curve (AUROC). Each model was internally validated using holdout data and externally validated using data from the other 2 hospitals. Calibration was assessed using calibration curves. RESULTS The study cohort included 7167 delirium cases and 7167 matched controls. XGB outperformed all other classifiers. AUROCs were highest for XGB models trained on 12 months of preadmission data. The best-performing XGB model achieved a mean AUROC of 0.79 (SD 0.01) on the holdout set, which decreased to 0.69-0.74 (SD 0.02) when externally validated on data from other hospitals. CONCLUSIONS Our routine EHR-based POD prediction models demonstrated good predictive ability using a limited set of preadmission and surgical variables, though their generalizability was limited. The proposed models could be used as a scalable, automated screening tool to identify patients at high risk of POD at the time of hospital admission.