Muhammad F. Walji

Objective:Clinically meaningful information in dental progress notes remains largely embedded in unstructured free text, limiting its reuse for research and quality improvement. Because these notes contain protected health information, extraction must run locally rather than relying on external proprietary models. We developed and evaluated a self-contained framework enabling locally deployable small language models (SLMs) to perform clinical named-entity recognition (NER) in dental notes without transmitting clinical text outside the institution.Methods:Using 1,200 annotated progress notes (200 training, 1,000 held-out gold standard) spanning 19 entity types, each candidate SLM generated, verified, refined, and evaluated its own entity-specific prompts, with self-prompt performance serving as a label-free model-selection signal. Inference combined two-stage multi-prompt ensembling consisting of sentence-level Boolean screening followed by entity extraction with union-based merging. Selected models were further adapted using QLoRA-based supervised fine-tuning (SFT) followed by direct preference optimization (DPO) under local hardware constraints. Performance was assessed by entity-level precision, recall, and F1 using soft matching.Results:Self-prompt generation performance was broadly concordant with gold-standard inference, supporting its use as a task-specific screening signal, and predicted task performance more reliably than general-purpose leaderboard rankings. Prompts incorporating entity descriptions and examples improved extraction, while error feedback added only modest benefit. SFT increased recall but encouraged over-extraction, and DPO recalibrated this balance by improving precision while largely preserving recall gains. Qwen2.5-14B-Instruct with DPO achieved the best overall performance (0.864 micro F1, 0.837 macro F1), with gains across most models advanced to DPO.Conclusion:Self-generated prompt optimization combined with lightweight preference-based post-training supports accurate, privacy-preserving clinical information extraction from unstructured dental notes using locally deployed SLMs. Because the framework makes no dentistry-specific assumptions, it offers an extensible pathway for privacy-sensitive clinical NER more broadly, pending external validation.

Background: The OWLL (Open-Wide Learning Lab) intervention was developed using Human-Centered Design to improve the quality and safety of pediatric dental sedation. The intervention includes a patient-facing informational sedation brochure and video, and an enhanced set of clinical sedation records. Here, we report on the process evaluation 6 months after implementation in daily clinical practice. Methods: This study was conducted at the outpatient pediatric dental clinics of 2 large US academic dental institutions. We used a mixed-methods design to assess the fidelity, acceptability, appropriateness, and feasibility of the OWLL intervention. Quantitative data were gathered through chart reviews, while qualitative data were collected through surveys and interviews with providers and patients. Results: During implementation, the intervention was used for 81 pediatric sedation patients across sites. Overall, the intervention was well-received by both clinicians and patients. Chart reviews indicated that sedation records were consistently utilized, though occasional gaps in data completeness were observed. Providers welcomed the intervention, finding it compatible with their daily routines and easy to use. However, some challenges were reported, primarily related to time constraints and the need to adjust clinical encounters for effective intervention integration. Parents appreciated the sedation brochure and video, describing them as a valuable and positive addition to the care process. Conclusion: The findings suggest that the intervention was used and is a welcomed addition, with the potential for broader application in clinical settings to further improve patient care. Future efforts should examine the intervention’s impact on patient safety outcomes across diverse clinical settings.

Background Despite established evidence-based guidelines, the prevention and management of dental caries varies across clinical settings. This study aimed to identify and understand quality gaps in caries prevention and management across four dental institutions. Methods A mixed-methods study was conducted across four large dental institutions. Three data sources were integrated: (1) structured chart reviews (n = 2,000) using six validated dental quality measures; (2) semi-structured interviews (n = 102) with patients, caregivers, staff, and dental providers; and (3) ethnographic observations of clinical care (n = 64) using the AEIOU framework. Data were triangulated using thematic coding and root cause analysis. Quality gaps were categorized as quantitative and qualitative gaps, and system-level challenges. Results Three categories of gaps emerged. (1) Quantitative gaps included low sealant placement (33.7%–54.6% in ages 6–9; 17.5%–43.0% in ages 10–14), inconsistent caries risk documentation (50.4%–99.6%), and high rates of untreated (23.1%–56.4%) and no new decay rates (65.9%–80.6%). (2) Qualitative gaps included limited documentation of preventive therapies, oral hygiene instruction, and nutritional counseling. Providers cited time constraints, unclear CRA protocols, and EHR usability issues. Interviews revealed unclear team roles and limited patient awareness of preventive options. (3) System-level challenges included workflow inefficiencies, fragmented responsibilities, and poor integration of CRA and preventive codes into clinical routines. Observations confirmed misalignments between documented and delivered care and missed opportunities for risk communication and same-day preventive interventions. Conclusions Substantial quality gaps persist in caries prevention and management despite institutional protocols. These gaps result from a complex interplay of systems, providers, and patient-level factors. Findings emphasize the need for coordinated data-informed strategies to improve the consistency and effectiveness of preventive care. This study highlights the value of integrating multiple data sources and user-centered methods to inform quality improvement in dental settings.

Purpose/Objective(s) Osteoradionecrosis of the jaw (ORN) is a severe complication of head and neck cancer (HNC) radiotherapy (RT), significantly impacting patient quality of life. Current dose assessment relies on whole-mandible dosimetry, limiting personalized, tooth-specific region risk evaluation. The lack of standardized methods to document and analyze dose distributions to tooth-bearing regions further hampers dental and maxillofacial decision-making. This study develops and evaluates RADMAP, a (semi-)automated tool for segmenting tooth-based jaw regions, enabling patient- and tooth-specific radiation dose mapping to improve dental dose reporting and ORN risk assessment. Methods and Materials A total of 736 tooth locations from 23 HNC patients treated with definitive RT were analyzed, including 11 who developed ORN. The RADMAP tool applies an angular ray-based algorithm to automatically segment the mandible and maxilla into 32 tooth-specific jaw regions, with optional manual refinement (semi-automated), outputting radiation dose mapping in both tabular and odontogram formats. Mean dose values from manually contoured tooth roots were compared with RADMAP-segmented alveolar regions (fully and semi-automated). Interobserver agreement among six users was evaluated. Whole-mandible, manual tooth, and tooth-based jaw segment (alveolar and basal) doses were compared between regions with and without ORN development. Results Mean dose correlation between RADMAP-segmented alveolar jaw regions and manually segmented tooth roots was high for both fully and semi-automated approaches (R² = 0.98, pConclusion RADMAP enables accurate, tooth-specific dose mapping of the mandible and maxilla, and potential improved ORN risk differentiation beyond whole-mandible dosimetry. Tooth-based dose reporting can personalize treatment, enhance multidisciplinary communication, and support prevention of radiation-related orodental sequelae.

Background: There are gaps in understanding the experiences of children, parents, and providers during dental conscious sedation. This study aimed to capture and analyze these experiences to identify opportunities for improvement and enhance the quality and safety of pediatric dental conscious sedation. Methods: A human-centered design approach was used to examine the conscious sedation experience in 2 US advanced education pediatric dental clinics. Researchers conducted field observations, interviews with providers and parents, and providers focus groups to explore experiential factors. Data were transcribed and analyzed using content thematic analysis. Insights from the data were used to explore and generate new solutions to improve dental conscious sedation quality and safety. Results: A total of 25 observations, 18 interviews (9 providers, 9 parents), and 4 provider focus groups were conducted across both sites. The process identified 4 key improvement opportunities: helping providers navigate the ambiguity of patient behavior, facilitating rapport building between providers and patients/parents, aligning expectations and supporting sedation sensemaking, and making the sedation experience more patient-centered. A multicomponent intervention was developed to address these needs, including a parent-facing brochure, a patient educational video, and an enhanced set of sedation records for providers to document patient and sedation information. Conclusions: This study used human-centered design to identify key challenges in pediatric dental conscious sedation and develop a multicomponent intervention in collaboration with patients, parents, and providers. The research demonstrates the potential of this approach to enhance sedation quality and safety, with future studies needed to assess its impact.

Objective Osteoradionecrosis of the jaw (ORNJ) is a chronic radiation-associated toxicity that lacks standardized classification criteria and treatment guidelines. Understanding early signs of tissue injury could help us better predict, prevent, and conservatively manage ORN. Our primary aims were to identify initial clinically-detected signs of ORN, determine the frequency of imaging-detected ORNJ, and validate the ability to classify cases using the novel system, ClinRad. Study Design A retrospective electronic health record review of 91 patients treated for head and neck cancer at The University of Texas MD Anderson Cancer Center with suspected ORN was performed by an Oral Medicine specialist to identify initial signs of ORN. Patients who received reirradiation to the head and neck or did not have enough evidence of ORN were excluded. A descriptive analysis was performed. Results 51 patients met the inclusion criteria. Half (53%) presented with imaging findings and exposed bone. Imaging findings in the absence of bone exposure were identified in 37%, of which disease progression was observed in 26%. All cases were classifiable using ClinRad. Conclusion Subclinical signs of bony changes consistent with ORN may be evident on imaging without exposed bone, supporting the use of imaging surveillance. ClinRad provided a mechanism to classify all cases at early onset. Data availability statement Anonymized data for the reported analyses is made publicly available on figshare at 10.6084/m9.figshare.28292186. Reporting guideline compliance statement In accordance with the EQUATOR Network (Enhancing the QUAlity and Transparency Of health Research) guidance, we have utilized the RECORD checklist, a guideline for the “REporting of studies Conducted using Observational Routinely-collected health data” (Benchimool El at al., 2015) The RECORD checklist is provided as a Supplementary file and available via 10.6084/m9.figshare.28292219. Data was anonymized in accordance with the EQUATOR guideline “Preparing raw clinical data for publication: guidance for journal editors, authors, and peer reviewers” (Hrynaszkiewicz I et al., 2010).