Exploring Large Language Models for Acronym, Symbol Sense Disambiguation, and Semantic Similarity and Relatedness Assessment.

Acronyms, abbreviations, and symbols play a significant role in clinical notes. Acronym and symbol sense disambiguation are crucial natural language processing (NLP) tasks that ensure the clarity and consistency of clinical notes and downstream NLP processing. Previous studies using traditional machine learning methods have been relatively successful in tackling this issue. In our research, we conducted an evaluation of large language models (LLMs), including ChatGPT 3.5 and 4, as well as other open LLMs, and BERT-based models, across three NLP tasks: acronym and symbol sense disambiguation, semantic similarity, and relatedness. Our findings emphasize ChatGPT's remarkable ability to distinguish between senses with minimal or zero-shot training. Additionally, open source LLM Mixtrial-8x7B exhibited high accuracy for acronyms with fewer senses, and moderate accuracy for symbol sense accuracy. BERT-based models outperformed previous machine learning approaches, achieving an impressive accuracy rate of over 95%, showcasing their effectiveness in addressing the challenge of acronym and symbol sense disambiguation. Furthermore, ChatGPT exhibited a strong correlation, surpassing 70%, with human gold standards when evaluating similarity and relatedness.

Decreasing Opioid Addiction and Diversion Using Behavioral Economics Applied Through a Digital Engagement Solution: Protocol for a Randomized Controlled Trial.

BACKGROUND: Despite strong and growing interest in ending the ongoing opioid health crisis, there has been limited success in reducing the prevalence of opioid addiction and the number of deaths associated with opioid overdoses. Further, 1 explanation for this is that existing interventions target those who are opiate-dependent but do not prevent opioid-naïve patients from becoming addicted. OBJECTIVE: Leveraging behavioral economics at the patient level could help patients successfully use, discontinue, and dispose of their opioid medications in an acute pain setting. The primary goal of this project is to evaluate the effect of the 3 versions of the Opioid Management for You (OPY) tool on measures of opioid use relative to the standard of care by leveraging a pragmatic randomized controlled trial (RCT). METHODS: A team of researchers from the Center for Learning Health System Sciences (CLHSS) at the University of Minnesota partnered with M Health Fairview to design, build, and test the 3 versions of the OPY tool: social influence, precommitment, and testimonial version. The tool is being built using the Epic Care Companion (Epic Inc) platform and interacts with the patient through their existing MyChart (Epic Systems Corporation) personal health record account, and Epic patient portal, accessed through a phone app or the MyChart website. We have demonstrated feasibility with pilot data of the social influence version of the OPY app by targeting our pilot to a specific cohort of patients undergoing upper-extremity procedures. This study will use a group sequential RCT design to test the impact of this important health system initiative. Patients who meet OPY inclusion criteria will be stratified into low, intermediate, and high risk of opiate use based on their type of surgery. RESULTS: This study is being funded and supported by the CLHSS Rapid Prospective Evaluation and Digital Technology Innovation Programs, and M Health Fairview. Support and coordination provided by CLHSS include the structure of engagement, survey development, data collection, statistical analysis, and dissemination. The project was initially started in August 2022. The pilot was launched in February 2023 and is still running, with the data last counted in August 2023. The actual RCT is planned to start by early 2024. CONCLUSIONS: Through this RCT, we will test our hypothesis that patient opioid use and diverted prescription opioid availability can both be improved by information delivery applied through a behavioral economics lens via sending nudges directly to the opioid users through their personal health record. TRIAL REGISTRATION: ClinicalTrials.gov NCT06124079; https://clinicaltrials.gov/study/NCT06124079. INTERNATIONAL REGISTERED REPORT IDENTIFIER (IRRID): PRR1-10.2196/52882.

A Symptom-Based Natural Language Processing Surveillance Pipeline for Post-COVID-19 Patients.

Post-acute sequelae of SARS CoV-2 (PASC) are a group of conditions in which patients previously infected with COVID-19 experience symptoms weeks/months post-infection. PASC has substantial societal burden, including increased healthcare costs and disabilities. This study presents a natural language processing (NLP) based pipeline for identification of PASC symptoms and demonstrates its ability to estimate the proportion of suspected PASC cases. A manual case review to obtain this estimate indicated our sample incidence of PASC (13%) was representative of the estimated population proportion (95% CI: 19±6.22%). However, the high number of cases classified as indeterminate demonstrates the challenges in classifying PASC even among experienced clinicians. Lastly, this study developed a dashboard to display views of aggregated PASC symptoms and measured its utility using the System Usability Scale. Overall comments related to the dashboard's potential were positive. This pipeline is crucial for monitoring post-COVID-19 patients with potential for use in clinical settings.

Re-Aiming Equity Evaluation in Clinical Decision Support: A Scoping Review of Equity Assessments in Surgical Decision Support Systems.

OBJECTIVE: We critically evaluated the surgical literature to explore the prevalence and describe how equity assessments occur when using clinical decision support systems. BACKGROUND: Clinical decision support (CDS) systems are increasingly used to facilitate surgical care delivery. Despite formal recommendations to do so, equity evaluations are not routinely performed on CDS systems and underrepresented populations are at risk of harm and further health disparities. We explored surgical literature to determine frequency and rigor of CDS equity assessments and offer recommendations to improve CDS equity by appending existing frameworks. METHODS: We performed a scoping review up to Augus 25, 2021 using PubMed and Google Scholar for the following search terms: clinical decision support, implementation, RE-AIM, Proctor, Proctor's framework, equity, trauma, surgery, surgical. We identified 1415 citations and 229 abstracts met criteria for review. A total of 84 underwent full review after 145 were excluded if they did not assess outcomes of an electronic CDS tool or have a surgical use case. RESULTS: Only 6% (5/84) of surgical CDS systems reported equity analyses, suggesting that current methods for optimizing equity in surgical CDS are inadequate. We propose revising the RE-AIM framework to include an Equity element (RE 2 -AIM) specifying that CDS foundational analyses and algorithms are performed or trained on balanced datasets with sociodemographic characteristics that accurately represent the CDS target population and are assessed by sensitivity analyses focused on vulnerable subpopulations. CONCLUSION: Current surgical CDS literature reports little with respect to equity. Revising the RE-AIM framework to include an Equity element (RE 2 -AIM) promotes the development and implementation of CDS systems that, at minimum, do not worsen healthcare disparities and possibly improve their generalizability.

Effect of Integrating Access to a Prescription Drug Monitoring Program Within the Electronic Health Record on the Frequency of Queries by Primary Care Clinicians: A Cluster Randomized Clinical Trial.

Importance: Tools that are directly integrated with the electronic health record (EHR) workflow can reduce the hassle cost of certain guideline-concordant practices, such as querying a prescription drug monitoring program (PDMP) before prescribing opioids. Objective: To investigate the effect of integrating access to a PDMP within the EHR on the frequency of program queries by primary care clinicians. Design Settings and Participants: The PRINCE (Prescribing Interventions for Chronic Pain Using the Electronic Health Record) randomized trial used a factorial cluster design at the clinic level in 43 primary care clinics in Minnesota. In all, 309 clinicians participated; 161 clinicians were given EHR-integrated access to PDMP at the intervention clinics, and 148 clinicians had the usual access at the control clinics. The intervention went live on August 27, 2020, and data were collected through March 3, 2021. Intervention: Single sign-on access to the Minnesota PDMP was integrated into the EHR, allowing clinicians to query a patient's controlled substance prescription and dispensing history as recorded in the Minnesota PDMP directly from the patient's EHR record without logging into a separate web portal. Additionally, the integration tool alerted clinicians and reminded them to review the PDMP if a patient had 3 or more opioid prescriptions in the past year and 1 or more in the past 6 months. Clinics in the control group did not receive access to the EHR-integrated PDMP tool; instead, these participants logged into the PDMP web portal separately. Main Outcomes and Measures: Monthly PDMP query counts for primary care clinicians, overall and by modality (EHR-based, web-based, via a clinical delegate), adjusted for clinician characteristics, including type (physician, nurse practitioner, physician assistant), sex, and years in practice. Data were analyzed from August 2021 to May 2022. Results: Of the 43 participating clinics with 309 clinicians, 21 clinics with 161 clinicians (102 [63.4%] women; 114 [70.8%] physicians; tenure, 10.6 [4.4] years) received the PDMP integration intervention. Baseline unadjusted monthly PDMP query rates for the average clinician were 6.6 (95% CI, 4.4-9.9) vs 8.8 (95% CI, 6.0-13.1) queries in the control vs the PDMP integration group, respectively. During the intervention, PDMP query rates for the average clinician were 6.9 (95% CI, 4.7-10.3) vs 14.8 (95% CI, 10.0-22.0) queries among the control vs the PDMP integration group, respectively. Compared with the control group, the EHR-integrated PDMP tool produced a 60% greater increase in the relative change in monthly PDMP queries (95% CI, 51%-70%). An increase in PDMP queries via the EHR-integrated PDMP tool drove this increase, while web-based and delegate queries declined by 39% more among the intervention compared with the control group (95% CI, 34%-43%). Conclusions and Relevance: This cluster randomized clinical trial found that integrating access to the PDMP in the EHR increased PDMP-querying rates, suggesting that direct access reduced hassle costs and can dramatically improve adherence to guideline-concordant care practices among primary care clinicians. Trial Registration: ClinicalTrials.gov Identifier: NCT04601506.

Combined Expert and User-Driven Usability Assessment of Trauma Decision Support Systems Improves User-Centered Design.

BACKGROUND: Trauma clinical decision support systems improve adherence with evidence-based practice but suffer from poor usability and the lack of a user-centered design. The objective of this study was to compare the effectiveness of user and expert-driven usability testing methods to detect usability issues in a rib fracture clinical decision support system and identify guiding principles for trauma clinical decision support systems. METHODS: A user-driven and expert-driven usability investigation was conducted using a clinical decision support system developed for patients with rib fractures. The user-driven usability evaluation was as follows: 10 clinicians were selected for simulation-based usability testing using snowball sampling, and each clinician completed 3 simulations using a video-conferencing platform. End-users participated in a novel team-based approach that simulated realistic clinical workflows. The expert-driven heuristic evaluation was as follows: 2 usability experts conducted a heuristic evaluation of the clinical decision support system using 10 common usability heuristics. Usability issues were identified, cataloged, and ranked for severity using a 4-level ordinal scale. Thematic analysis was utilized to categorize the identified usability issues. RESULTS: Seventy-nine usability issues were identified; 63% were identified by experts and 48% by end-users. Notably, 58% of severe usability issues were identified by experts alone. Only 11% of issues were identified by both methods. Five themes were identified that could guide the design of clinical decision support systems-transparency, functionality and integration into workflow, automated and noninterruptive, flexibility, and layout and appearance. Themes were preferentially identified by different methods. CONCLUSION: We found that a dual-method usability evaluation involving usability experts and end-users drastically improved detection of usability issues over single-method alone. We identified 5 themes to guide trauma clinical decision support system design. Performing usability testing via a remote video-conferencing platform facilitated multi-site involvement despite a global pandemic.

Racial/Ethnic Disparities in Hospital Admissions from COVID-19 and Determining the Impact of Neighborhood Deprivation and Primary Language.

Background Despite past and ongoing efforts to achieve health equity in the United States, persistent disparities in socioeconomic status along with multilevel racism maintain disparate outcomes and appear to be amplified by COVID-19. Objective Measure socioeconomic factors and primary language effects on the risk of COVID-19 severity across and within racial/ethnic groups. Design Retrospective cohort study. Setting Health records of 12 Midwest hospitals and 60 clinics in the U.S. between March 4, 2020 to August 19, 2020. Patients PCR+ COVID-19 patients. Exposures Main exposures included race/ethnicity, area deprivation index (ADI), and primary language. Main Outcomes and Measures The primary outcome was COVID-19 severity using hospitalization within 45 days of diagnosis. Logistic and competing-risk regression models (censored at 45 days and accounting for the competing risk of death prior to hospitalization) assessed the effects of neighborhood-level deprivation (using the ADI) and primary language. Within race effects of ADI and primary language were measured using logistic regression. Results 5,577 COVID-19 patients were included, 866 (n=15.5%) were hospitalized within 45 days of diagnosis. Hospitalized patients were older (60.9 vs. 40.4 years, p<0.001) and more likely to be male (n=425 [49.1%] vs. 2,049 [43.5%], p=0.002). Of those requiring hospitalization, 43.9% (n=381), 19.9% (n=172), 18.6% (n=161), and 11.8% (n=102) were White, Black, Asian, and Hispanic, respectively. Independent of ADI, minority race/ethnicity was associated with COVID-19 severity; Hispanic patients (OR 3.8, 95% CI 2.72-5.30), Asians (OR 2.39, 95% CI 1.74-3.29), and Blacks (OR 1.50, 95% CI 1.15-1.94). ADI was not associated with hospitalization. Non-English speaking (OR 1.91, 95% CI 1.51-2.43) significantly increased odds of hospital admission across and within minority groups. Conclusions Minority populations have increased odds of severe COVID-19 independent of neighborhood deprivation, a commonly suspected driver of disparate outcomes. Non-English-speaking accounts for differences across and within minority populations. These results support the continued concern that racism contributes to disparities during COVID-19 while also highlighting the underappreciated role primary language plays in COVID-19 severity across and within minority groups.

System-Wide Improvement for Transitions After Ileostomy Surgery: Can Intensive Monitoring of Protocol Compliance Decrease Readmissions? A Randomized Trial.

BACKGROUND: Hospital readmission is common after ileostomy formation and frequently associated with dehydration. OBJECTIVE: This study was conducted to evaluate a previously published intervention to prevent dehydration and readmission. DESIGN: This is a randomized controlled trial. SETTING: This study was conducted in 3 hospitals within a single health care system. PATIENTS: Patients undergoing elective or nonelective ileostomy as part of their operative procedure were selected. INTERVENTION: Surgeons, advanced practice providers, inpatient and outpatient nurses, and wound ostomy continence nurses participated in a robust ileostomy education and monitoring program (Education Program for Prevention of Ileostomy Complications) based on the published intervention. After informed consent, patients were randomly assigned to a postoperative compliance surveillance and prompting strategy that was directed toward the care team, versus usual care. OUTCOME MEASURES: Unplanned hospital readmission within 30 days of discharge, readmission for dehydration, acute renal failure, estimated direct costs, and patient satisfaction were the primary outcomes measured. RESULTS: One hundred patients with an ileostomy were randomly assigned. The most common indications were rectal cancer (n = 26) and ulcerative colitis (n = 21), and 12 were emergency procedures. Although intervention patients had better postdischarge phone follow-up (90% vs 72%; p = 0.025) and were more likely to receive outpatient intravenous fluids (25% vs 6%; p = 0.008), they had similar overall hospital readmissions (20.4% vs 19.6%; p = 1.0), readmissions for dehydration (8.2% vs 5.9%; p = 0.71), and acute renal failure events (10.2% vs 3.9%; p = 0.26). Multivariable analysis found that weekend discharges to home were significantly associated with readmission (OR, 4.5 (95% CI, 1.2-16.9); p = 0.03). Direct costs and patient satisfaction were similar. LIMITATIONS: This study was limited by the heterogeneous patient population and by the potential effect of the intervention on providers taking care of patients randomly assigned to usual care. CONCLUSIONS: A surveillance strategy to ensure compliance with an ileostomy education program tracked patients more closely and was cost neutral, but did not result in decreased hospital readmissions compared with usual care. See Video Abstract at http://links.lww.com/DCR/A812.

Evaluation of Flowsheet Documentation in the Electronic Health Record for Residence, Living Situation, and Living Conditions.

Social determinants of health (SDOH) are important considerations in diagnosis, prevention, and health outcomes. However, they are often not well documented in the EHR and found primarily in unstructured or semi-structured text. Building upon previous work, we analyzed all flowsheet data in 2013 for information related to the SDOH topic areas of Residence, Living Situation, and Living Conditions. Overall, 91 rows were identified as being related to the topics areas resulting in 604,616 unique observations. Individual rows contained SDOH data often covered multiple concepts especially free-text entries. These data included most often references to the residence, residence details, and with whom the patient lives. Very few contained living condition references. Additionally, there was significant duplication and inconsistency of row labels, as well as variation in value list content for rows collecting the same concepts. Our findings demonstrate significant opportunities to improve and achieve better standardization in documentation around these SDOH.

Diagnostic Discordance, Health Information Exchange, and Inter-Hospital Transfer Outcomes: a Population Study.

BACKGROUND: Studying diagnostic error at the population level requires an understanding of how diagnoses change over time. OBJECTIVE: To use inter-hospital transfers to examine the frequency and impact of changes in diagnosis on patient risk, and whether health information exchange can improve patient safety by enhancing diagnostic accuracy. DESIGN: Diagnosis coding before and after hospital transfer was merged with responses from the American Hospital Association Annual Survey for a cohort of patients transferred between hospitals to identify predictors of mortality. PARTICIPANTS: Patients (180,337) 18 years or older transferred between 473 acute care hospitals from NY, FL, IA, UT, and VT from 2011 to 2013. MAIN MEASURES: We identified discordant Elixhauser comorbidities before and after transfer to determine the frequency and developed a weighted score of diagnostic discordance to predict mortality. This was included in a multivariate model with inpatient mortality as the dependent variable. We investigated whether health information exchange (HIE) functionality adoption as reported by hospitals improved diagnostic discordance and inpatient mortality. KEY RESULTS: Discordance in diagnoses occurred in 85.5% of all patients. Seventy-three percent of patients gained a new diagnosis following transfer while 47% of patients lost a diagnosis. Diagnostic discordance was associated with increased adjusted inpatient mortality (OR 1.11 95% CI 1.10-1.11, p < 0.001) and allowed for improved mortality prediction. Bilateral hospital HIE participation was associated with reduced diagnostic discordance index (3.69 vs. 1.87%, p < 0.001) and decreased inpatient mortality (OR 0.88, 95% CI 0.89-0.99, p < 0.001). CONCLUSIONS: Diagnostic discordance commonly occurred during inter-hospital transfers and was associated with increased inpatient mortality. Health information exchange adoption was associated with decreased discordance and improved patient outcomes.

African Americans and Short-Term Outcomes after Surgery for Crohn's Disease: An ACS-NSQIP Analysis.

BACKGROUND: Previous reports on racial disparities in the treatment of Crohn's disease [CD] in African American [AA] patients have shown differences in both medical and surgical treatments in this population. No study thus far has examined the effect of AA race on outcomes after surgery for CD. METHODS: Utilizing the National Surgical Quality Improvement Program [NSQIP] Participant User File [PUF] for the years 2005-2013, we examined the effect of AA race on postoperative complications in patients with CD undergoing intestinal surgery. RESULTS: AA patients had a significantly higher rate of complications overall compared to non-AA patients [23.5% vs 18.9%, p = 0.002]. Postoperative sepsis [10.9% vs 6.6%, p < 0.001] and surgical site infection [17.6% vs 14.8%, p = 0.037] were most significant. After adjustment for age, sex, preoperative disease severity and lifestyle factors [smoking], race remained a statistically significant factor in postoperative complication rate. Only after additional adjustment was made for comorbidities and American Society of Anesthesiologists class did race lose significance within our model. CONCLUSION: African Americans experience a greater amount of postoperative complications following surgery for Crohn's disease. Preoperative disease management, addressing smoking status and control of comorbid disease are important factors in addressing the racial disparities in the surgical treatment of Crohn's disease.

Residence, Living Situation, and Living Conditions Information Documentation in Clinical Practice.

Social determinants of health (SDOH) have an important role in diagnosis, prevention, health outcomes, and quality of life. Currently, SDOH information in electronic health record (EHR) systems is often contained in unstructured text. The objective of this study is to examine an important subset of SDOH documentation for Residence, Living Situation and Living Conditions in an enterprise EHR informed by previous model representations. In addition to two publically available clinical note sources, notes created by Social Work, Physical Therapy, and Occupational Therapy, along with free text Social Documentation entries were reviewed. Sentences were classified, annotated, and evaluated once mapped to element entities and attributes. Overall, 2,491 total notes yielded 616, 813, and 30 sentences related to Residence, Living Situation, and Living Conditions. This study demonstrated the need for additional elements in the model representation, more representative values and content culminating in a more comprehensive model representation for these key SDOH.

Assessing the Representation of Occupation Information in Free-Text Clinical Documents Across Multiple Sources.

There has been increasing recognition of the key role of social determinants like occupation on health. Given the relatively poor understanding of occupation information in electronic health records (EHRs), we sought to characterize occupation information within free-text clinical document sources. From six distinct clinical sources, 868 total occupation-related sentences were identified for the study corpus. Building off approaches from previous studies, refined annotation guidelines were created using the National Institute for Occupational Safety and Health Occupational Data for Health data model with elements added to increase granularity. Our corpus generated 2,005 total annotations representing 39 of 41 entity types from the enhanced data model. Highest frequency entities were: Occupation Description (17.7%); Employment Status - Not Specified (12.5%); Employer Name (11.0%); Subject (9.8%); Industry Description (6.2%). Our findings support the value of standardizing entry of EHR occupation information to improve data quality for improved patient care and secondary uses of this information.

Representation of Social History Factors Across Age Groups: A Topic Analysis of Free-Text Social Documentation.

As individuals age, there is potential for dramatic changes in the social and behavioral determinants that affect health status and outcomes. The importance of these determinants has been increasingly recognized in clinical decision-making. We sought to characterize how social and behavioral health determinants vary in different demographic groups using a previously established schema of 28 social history types through both manual analysis and automated topic analysis of social documentation in the electronic health record across the population of an entire integrated healthcare system. Our manual analysis generated 8,335 annotations over 1,400 documents, representing 24 (86%) social history types. In contrast, automated topic analysis generated 22 (79%) social history types. A comparative evaluation demonstrated both similarities and differences in coverage between the manual and topic analyses. Our findings validate the widespread nature of social and behavioral determinants that affect health status over populations of individuals over their lifespan.

Representing Residence, Living Situation, and Living Conditions: An Evaluation of Terminologies, Standards, Guidelines, and Measures/Surveys.

Social determinants of health play an important role in diagnosis, prevention, health outcomes, and quality of life. The objective of this study was to examine existing standards, vocabularies, and terminologies for items related to Residence, Living Situation, and Living Conditions and to synthesize them into model representations. Sources were identified through literature and keyword searches, and an examination of commonly used resources. Each source was systematically analyzed by two reviewers, mapped to topic area(s), and further mapped to a model representation. A total of 27 sources were identified and reviewed. Seven of the sources had no items, i.e. concepts, elements, or values, related to the three topic areas while SNOMED-CT had the most items at 436 followed by the US Census at 174. While none of the identified sources encompassed a complete representation for documenting the three topic areas, their synthesis together results overall in more comprehensive representations.

Patient-specific surgical outcomes assessment using population-based data analysis for risk model development.

Optimal surgical planning and decision making surrounding surgical interventions requires patient-specific risk assessment which incorporates patient pre-operative clinical assessment and clinical literature. In this paper, we utilized population-based data analysis to construct surgical outcome predictive models for spinal fusion surgery using hospital, patient and admission characteristics. We analyzed population data from the Nationwide Inpatient Sample (NIS) -a nationally representative database- to identify data elements affecting inpatient mortality, length of stay, and disposition status for patients receiving spinal fusion surgery in the years 2004-2008. In addition to outcomes assessment, we want to make the analytic model results available to clinicians and researchers for pre-operative surgical risk assessment, hospital resource allocation, and hypothesis generation for future research without an individual patient data management burden. Spinal fusion was the selected prototype procedure due to it being a high volume and typically inpatient procedure where patient risk factors will likely affect clinical outcomes.

Automated assessment of medical training evaluation text.

Medical post-graduate residency training and medical student training increasingly utilize electronic systems to evaluate trainee performance based on defined training competencies with quantitative and qualitative data, the later of which typically consists of text comments. Medical education is concomitantly becoming a growing area of clinical research. While electronic systems have proliferated in number, little work has been done to help manage and analyze qualitative data from these evaluations. We explored the use of text-mining techniques to assist medical education researchers in sentiment analysis and topic analysis of residency evaluations with a sample of 812 evaluation statements. While comments were predominantly positive, sentiment analysis improved the ability to discriminate statements with 93% accuracy. Similar to other domains, Latent Dirichlet Analysis and Information Gain revealed groups of core subjects and appear to be useful for identifying topics from this data.

Deidentification of facial images using composites.

PURPOSE: Maxillofacial surgeons rely on photography for education and documentation. Photographs of the face, unlike those of other body regions, are readily identifiable. Traditional methods of facial image deidentification decrease educational quality or fail to adequately conceal identity. In the present study, a method that uses blended facial composites to deidentify original facial images was developed. This method allows significant components of the original face to be visualized while concealing its identity. MATERIALS AND METHODS: The method was used to develop 20 different composite facial images that were viewed by student subjects. Ten of these images contained at least one third of a face that was familiar to the subjects. Subjects viewed the composite faces twice--first unaware that the faces were composites, and then primed to the presence of composites. Subjects later rated the efficacy of this method for image deidentification. RESULTS: When unaware that they were viewing composite images, no subjects recognized the familiar faces within the composites or rated them as familiar (0/120 total views, 0%). When later primed to the potential presence of familiar faces within composites, the identification rate increased significantly (74/120, 62%; P < .001). Results were similar no matter which portion of the familiar face (upper, 67%; mid, 54%; lower, 67%) was present. Subjects rated all composites as clinically realistic patient images. They also rated composites as more effective at deidentification than traditional methods. CONCLUSION: The use of composites appears to be a promising concept for facial image deidentification. Further larger-scale studies are needed to validate these findings.