Bilateral Choroidal Detachments Associated with Perinuclear Anti-Neutrophil Cytoplasmic Antibodies (P-ANCA) - Associated Vasculitis.

PURPOSE: To report an uncommon, isolated presentation of bilateral choroidal detachments in a patient diagnosed with P-ANCA-associated vasculitis and to highlight the importance of an inflammatory work-up in such cases. METHODS: Case report. RESULTS: A 70-year-old male with a past medical history of autoimmune hepatitis presented with a sudden decrease in vision in both eyes. Over the course of the previous decade, he had experienced recurrent attacks of episcleritis, which were successfully managed with topical corticosteroid eye drops. The patient was diagnosed with bilateral detachments without accompanying scleritis or intraocular inflammation. Inflammatory markers revealed high P-ANCA and anti-MPO levels, confirming the diagnosis of P-ANCA-associated vasculitis. Treatment with systemic rituximab and corticosteroids led to the resolution of the choroidal detachment in both eyes. A 40-month follow-up confirmed the sustained resolution of the detachments. CONCLUSION: Choroidal detachment without other extraocular/intraocular inflammation can be associated with P-ANCA-associated vasculitis, a previously under-reported link. It is important to consider an inflammatory work-up for patients presenting with choroidal detachment to rule out conditions like P-ANCA-associated vasculitis.

Association of Gender and Operative Feedback Quality in Surgical Residents.

OBJECTIVE: Feedback is critical for learning, however, gender differences exist in the quality of feedback that trainees receive. For example, narrative feedback on surgical trainees' end-of-block rotations differs based on trainee-faculty gender dyads, with female faculty giving higher quality feedback and male trainees receiving higher quality feedback. Though this represents evidence of gender bias in global evaluations, there is limited understanding of how much bias might be present in operative workplace-based assessments (WBAs). In this study, we explore the quality of narrative feedback among trainee-faculty gender dyads in an operative WBA. DESIGN: A previously validated natural language processing model was used to examine instances of narrative feedback and assign a probability of being characterized as high quality feedback (defined as feedback which was relevant as well as corrective and/or specific). A linear mixed model was performed, with probability of high quality feedback as the outcome, and resident gender, faculty gender, PGY, case complexity, autonomy rating, and operative performance rating as explanatory variables. PARTICIPANTS: Analyses included 67,434 SIMPL operative performance evaluations (2,319 general surgery residents, 70 institutions) collected from September 2015 through September 2021. RESULTS: Of 36.3% evaluations included narrative feedback. Male faculty were more likely to provide narrative feedback compared to female faculty. Mean probabilities of receiving high quality feedback ranged from 81.6 (female faculty-male resident) to 84.7 (male faculty-female resident). Model-based results demonstrated that female residents were more likely to receive high quality feedback (p < 0.01), however, there was no significant difference in probability of high quality narrative feedback based on faculty-resident gender dyad (p = 0.77). CONCLUSIONS: Our study revealed resident gender differences in the probability of receiving high-quality narrative feedback following a general surgery operation. However, we found no significant differences based on faculty-resident gender dyad. Male faculty were more likely to provide narrative feedback compared to their female colleagues. Further research using general surgery resident-specific feedback quality models may be warranted.

Clostridioides difficile infection surveillance in intensive care units and oncology wards using machine learning.

OBJECTIVE: Screening individuals admitted to the hospital for Clostridioides difficile presents opportunities to limit transmission and hospital-onset C. difficile infection (HO-CDI). However, detection from rectal swabs is resource intensive. In contrast, machine learning (ML) models may accurately assess patient risk without significant resource usage. In this study, we compared the effectiveness of swab surveillance to daily risk estimates produced by an ML model to identify patients who will likely develop HO-CDI in the intensive care unit (ICU) setting. DESIGN: A prospective cohort study was conducted with patient carriage of toxigenic C. difficile identified by rectal swabs analyzed by anaerobic culture and polymerase chain reaction (PCR). A previously validated ML model using electronic health record data generated daily risk of HO-CDI for every patient. Swab results and risk predictions were compared to the eventual HO-CDI status. PATIENTS: Adult inpatient admissions taking place in University of Michigan Hospitals' medical and surgical intensive care units and oncology wards between June 6th and October 8th, 2020. RESULTS: In total, 2,979 admissions, representing 2,044 patients, were observed over the course of the study period, with 39 admissions developing HO-CDIs. Swab surveillance identified 9 true-positive and 87 false-positive HO-CDIs. The ML model identified 9 true-positive and 226 false-positive HO-CDIs; 8 of the true-positives identified by the model differed from those identified by the swab surveillance. CONCLUSION: With limited resources, an ML model identified the same number of HO-CDI admissions as swab-based surveillance, though it generated more false-positives. The patients identified by the ML model were not yet colonized with C. difficile. Additionally, the ML model identifies at-risk admissions before disease onset, providing opportunities for prevention.

Prospective evaluation of data-driven models to predict daily risk of Clostridioides difficile infection at 2 large academic health centers.

Many data-driven patient risk stratification models have not been evaluated prospectively. We performed and compared the prospective and retrospective evaluations of 2 Clostridioides difficile infection (CDI) risk-prediction models at 2 large academic health centers, and we discuss the models' robustness to data-set shifts.

Teaching artificial intelligence as a fundamental toolset of medicine.

Artificial intelligence (AI) is transforming the practice of medicine. Systems assessing chest radiographs, pathology slides, and early warning systems embedded in electronic health records (EHRs) are becoming ubiquitous in medical practice. Despite this, medical students have minimal exposure to the concepts necessary to utilize and evaluate AI systems, leaving them under prepared for future clinical practice. We must work quickly to bolster undergraduate medical education around AI to remedy this. In this commentary, we propose that medical educators treat AI as a critical component of medical practice that is introduced early and integrated with the other core components of medical school curricula. Equipping graduating medical students with this knowledge will ensure they have the skills to solve challenges arising at the confluence of AI and medicine.

Dynamic prediction of work status for workers with occupational injuries: assessing the value of longitudinal observations.

OBJECTIVE: Occupational injuries (OIs) cause an immense burden on the US population. Prediction models help focus resources on those at greatest risk of a delayed return to work (RTW). RTW depends on factors that develop over time; however, existing methods only utilize information collected at the time of injury. We investigate the performance benefits of dynamically estimating RTW, using longitudinal observations of diagnoses and treatments collected beyond the time of initial injury. MATERIALS AND METHODS: We characterize the difference in predictive performance between an approach that uses information collected at the time of initial injury (baseline model) and a proposed approach that uses longitudinal information collected over the course of the patient's recovery period (proposed model). To control the comparison, both models use the same deep learning architecture and differ only in the information used. We utilize a large longitudinal observation dataset of OI claims and compare the performance of the two approaches in terms of daily prediction of future work state (working vs not working). The performance of these two approaches was assessed in terms of the area under the receiver operator characteristic curve (AUROC) and expected calibration error (ECE). RESULTS: After subsampling and applying inclusion criteria, our final dataset covered 294 103 OIs, which were split evenly between train, development, and test datasets (1/3, 1/3, 1/3). In terms of discriminative performance on the test dataset, the proposed model had an AUROC of 0.728 (90% confidence interval: 0.723, 0.734) versus the baseline's 0.591 (0.585, 0.598). The proposed model had an ECE of 0.004 (0.003, 0.005) versus the baseline's 0.016 (0.009, 0.018). CONCLUSION: The longitudinal approach outperforms current practice and shows potential for leveraging observational data to dynamically update predictions of RTW in the setting of OI. This approach may enable physicians and workers' compensation programs to manage large populations of injured workers more effectively.

Early identification of patients admitted to hospital for covid-19 at risk of clinical deterioration: model development and multisite external validation study.

OBJECTIVE: To create and validate a simple and transferable machine learning model from electronic health record data to accurately predict clinical deterioration in patients with covid-19 across institutions, through use of a novel paradigm for model development and code sharing. DESIGN: Retrospective cohort study. SETTING: One US hospital during 2015-21 was used for model training and internal validation. External validation was conducted on patients admitted to hospital with covid-19 at 12 other US medical centers during 2020-21. PARTICIPANTS: 33 119 adults (≥18 years) admitted to hospital with respiratory distress or covid-19. MAIN OUTCOME MEASURES: An ensemble of linear models was trained on the development cohort to predict a composite outcome of clinical deterioration within the first five days of hospital admission, defined as in-hospital mortality or any of three treatments indicating severe illness: mechanical ventilation, heated high flow nasal cannula, or intravenous vasopressors. The model was based on nine clinical and personal characteristic variables selected from 2686 variables available in the electronic health record. Internal and external validation performance was measured using the area under the receiver operating characteristic curve (AUROC) and the expected calibration error-the difference between predicted risk and actual risk. Potential bed day savings were estimated by calculating how many bed days hospitals could save per patient if low risk patients identified by the model were discharged early. RESULTS: 9291 covid-19 related hospital admissions at 13 medical centers were used for model validation, of which 1510 (16.3%) were related to the primary outcome. When the model was applied to the internal validation cohort, it achieved an AUROC of 0.80 (95% confidence interval 0.77 to 0.84) and an expected calibration error of 0.01 (95% confidence interval 0.00 to 0.02). Performance was consistent when validated in the 12 external medical centers (AUROC range 0.77-0.84), across subgroups of sex, age, race, and ethnicity (AUROC range 0.78-0.84), and across quarters (AUROC range 0.73-0.83). Using the model to triage low risk patients could potentially save up to 7.8 bed days per patient resulting from early discharge. CONCLUSION: A model to predict clinical deterioration was developed rapidly in response to the covid-19 pandemic at a single hospital, was applied externally without the sharing of data, and performed well across multiple medical centers, patient subgroups, and time periods, showing its potential as a tool for use in optimizing healthcare resources.

Development and Validation of Models to Predict Pathological Outcomes of Radical Prostatectomy in Regional and National Cohorts.

PURPOSE: Prediction models are recommended by national guidelines to support clinical decision making in prostate cancer. Existing models to predict pathological outcomes of radical prostatectomy (RP)-the Memorial Sloan Kettering (MSK) models, Partin tables, and the Briganti nomogram-have been developed using data from tertiary care centers and may not generalize well to other settings. MATERIALS AND METHODS: Data from a regional cohort (Michigan Urological Surgery Improvement Collaborative [MUSIC]) were used to develop models to predict extraprostatic extension (EPE), seminal vesicle invasion (SVI), lymph node invasion (LNI), and nonorgan-confined disease (NOCD) in patients undergoing RP. The MUSIC models were compared against the MSK models, Partin tables, and Briganti nomogram (for LNI) using data from a national cohort (Surveillance, Epidemiology, and End Results [SEER] registry). RESULTS: We identified 7,491 eligible patients in the SEER registry. The MUSIC model had good discrimination (SEER AUC EPE: 0.77; SVI: 0.80; LNI: 0.83; NOCD: 0.77) and was well calibrated. While the MSK models had similar discrimination to the MUSIC models (SEER AUC EPE: 0.76; SVI: 0.80; LNI: 0.84; NOCD: 0.76), they overestimated the risk of EPE, LNI, and NOCD. The Partin tables had inferior discrimination (SEER AUC EPE: 0.67; SVI: 0.76; LNI: 0.69; NOCD: 0.72) as compared to other models. The Briganti LNI nomogram had an AUC of 0.81 in SEER but overestimated the risk. CONCLUSIONS: New models developed using the MUSIC registry outperformed existing models and should be considered as potential replacements for the prediction of pathological outcomes in prostate cancer.

Natural Language Processing to Estimate Clinical Competency Committee Ratings.

OBJECTIVE: Residency program faculty participate in clinical competency committee (CCC) meetings, which are designed to evaluate residents' performance and aid in the development of individualized learning plans. In preparation for the CCC meetings, faculty members synthesize performance information from a variety of sources. Natural language processing (NLP), a form of artificial intelligence, might facilitate these complex holistic reviews. However, there is little research involving the application of this technology to resident performance assessments. With this study, we examine whether NLP can be used to estimate CCC ratings. DESIGN: We analyzed end-of-rotation assessments and CCC assessments for all surgical residents who trained at one institution between 2014 and 2018. We created models of end-of-rotation assessment ratings and text to predict dichotomized CCC assessment ratings for 16 Accreditation Council for Graduate Medical Education (ACGME) Milestones. We compared the performance of models with and without predictors derived from NLP of end-of-rotation assessment text. RESULTS: We analyzed 594 end-of-rotation assessments and 97 CCC assessments for 24 general surgery residents. The mean (standard deviation) for area under the receiver operating characteristic curve (AUC) was 0.84 (0.05) for models with only non-NLP predictors, 0.83 (0.06) for models with only NLP predictors, and 0.87 (0.05) for models with both NLP and non-NLP predictors. CONCLUSIONS: NLP can identify language correlated with specific ACGME Milestone ratings. In preparation for CCC meetings, faculty could use information automatically extracted from text to focus attention on residents who might benefit from additional support and guide the development of educational interventions.

External Validation of a Widely Implemented Proprietary Sepsis Prediction Model in Hospitalized Patients.

Importance: The Epic Sepsis Model (ESM), a proprietary sepsis prediction model, is implemented at hundreds of US hospitals. The ESM's ability to identify patients with sepsis has not been adequately evaluated despite widespread use. Objective: To externally validate the ESM in the prediction of sepsis and evaluate its potential clinical value compared with usual care. Design, Setting, and Participants: This retrospective cohort study was conducted among 27 697 patients aged 18 years or older admitted to Michigan Medicine, the academic health system of the University of Michigan, Ann Arbor, with 38 455 hospitalizations between December 6, 2018, and October 20, 2019. Exposure: The ESM score, calculated every 15 minutes. Main Outcomes and Measures: Sepsis, as defined by a composite of (1) the Centers for Disease Control and Prevention surveillance criteria and (2) International Statistical Classification of Diseases and Related Health Problems, Tenth Revision diagnostic codes accompanied by 2 systemic inflammatory response syndrome criteria and 1 organ dysfunction criterion within 6 hours of one another. Model discrimination was assessed using the area under the receiver operating characteristic curve at the hospitalization level and with prediction horizons of 4, 8, 12, and 24 hours. Model calibration was evaluated with calibration plots. The potential clinical benefit associated with the ESM was assessed by evaluating the added benefit of the ESM score compared with contemporary clinical practice (based on timely administration of antibiotics). Alert fatigue was evaluated by comparing the clinical value of different alerting strategies. Results: We identified 27 697 patients who had 38 455 hospitalizations (21 904 women [57%]; median age, 56 years [interquartile range, 35-69 years]) meeting inclusion criteria, of whom sepsis occurred in 2552 (7%). The ESM had a hospitalization-level area under the receiver operating characteristic curve of 0.63 (95% CI, 0.62-0.64). The ESM identified 183 of 2552 patients with sepsis (7%) who did not receive timely administration of antibiotics, highlighting the low sensitivity of the ESM in comparison with contemporary clinical practice. The ESM also did not identify 1709 patients with sepsis (67%) despite generating alerts for an ESM score of 6 or higher for 6971 of all 38 455 hospitalized patients (18%), thus creating a large burden of alert fatigue. Conclusions and Relevance: This external validation cohort study suggests that the ESM has poor discrimination and calibration in predicting the onset of sepsis. The widespread adoption of the ESM despite its poor performance raises fundamental concerns about sepsis management on a national level.

Natural Language Processing and Assessment of Resident Feedback Quality.

OBJECTIVE: To validate the performance of a natural language processing (NLP) model in characterizing the quality of feedback provided to surgical trainees. DESIGN: Narrative surgical resident feedback transcripts were collected from a large academic institution and classified for quality by trained coders. 75% of classified transcripts were used to train a logistic regression NLP model and 25% were used for testing the model. The NLP model was trained by uploading classified transcripts and tested using unclassified transcripts. The model then classified those transcripts into dichotomized high- and low- quality ratings. Model performance was primarily assessed in terms of accuracy and secondary performance measures including sensitivity, specificity, and area under the receiver operating characteristic curve (AUROC). SETTING: A surgical residency program based in a large academic medical center. PARTICIPANTS: All surgical residents who received feedback via the Society for Improving Medical Professional Learning smartphone application (SIMPL, Boston, MA) in August 2019. RESULTS: The model classified the quality (high vs. low) of 2,416 narrative feedback transcripts with an accuracy of 0.83 (95% confidence interval: 0.80, 0.86), sensitivity of 0.37 (0.33, 0.45), specificity of 0.97 (0.96, 0.98), and an area under the receiver operating characteristic curve of 0.86 (0.83, 0.87). CONCLUSIONS: The NLP model classified the quality of operative performance feedback with high accuracy and specificity. NLP offers residency programs the opportunity to efficiently measure feedback quality. This information can be used for feedback improvement efforts and ultimately, the education of surgical trainees.

Using Natural Language Processing to Automatically Assess Feedback Quality: Findings From 3 Surgical Residencies.

PURPOSE: Learning is markedly improved with high-quality feedback, yet assuring the quality of feedback is difficult to achieve at scale. Natural language processing (NLP) algorithms may be useful in this context as they can automatically classify large volumes of narrative data. However, it is unknown if NLP models can accurately evaluate surgical trainee feedback. This study evaluated which NLP techniques best classify the quality of surgical trainee formative feedback recorded as part of a workplace assessment. METHOD: During the 2016-2017 academic year, the SIMPL (Society for Improving Medical Professional Learning) app was used to record operative performance narrative feedback for residents at 3 university-based general surgery residency training programs. Feedback comments were collected for a sample of residents representing all 5 postgraduate year levels and coded for quality. In May 2019, the coded comments were then used to train NLP models to automatically classify the quality of feedback across 4 categories (effective, mediocre, ineffective, or other). Models included support vector machines (SVM), logistic regression, gradient boosted trees, naive Bayes, and random forests. The primary outcome was mean classification accuracy. RESULTS: The authors manually coded the quality of 600 recorded feedback comments. Those data were used to train NLP models to automatically classify the quality of feedback across 4 categories. The NLP model using an SVM algorithm yielded a maximum mean accuracy of 0.64 (standard deviation, 0.01). When the classification task was modified to distinguish only high-quality vs low-quality feedback, maximum mean accuracy was 0.83, again with SVM. CONCLUSIONS: To the authors' knowledge, this is the first study to examine the use of NLP for classifying feedback quality. SVM NLP models demonstrated the ability to automatically classify the quality of surgical trainee evaluations. Larger training datasets would likely further increase accuracy.

Evaluating a Widely Implemented Proprietary Deterioration Index Model among Hospitalized Patients with COVID-19.

Rationale: The Epic Deterioration Index (EDI) is a proprietary prediction model implemented in over 100 U.S. hospitals that was widely used to support medical decision-making during the coronavirus disease (COVID-19) pandemic. The EDI has not been independently evaluated, and other proprietary models have been shown to be biased against vulnerable populations. Objectives: To independently evaluate the EDI in hospitalized patients with COVID-19 overall and in disproportionately affected subgroups. Methods: We studied adult patients admitted with COVID-19 to units other than the intensive care unit at a large academic medical center from March 9 through May 20, 2020. We used the EDI, calculated at 15-minute intervals, to predict a composite outcome of intensive care unit-level care, mechanical ventilation, or in-hospital death. In a subset of patients hospitalized for at least 48 hours, we also evaluated the ability of the EDI to identify patients at low risk of experiencing this composite outcome during their remaining hospitalization. Results: Among 392 COVID-19 hospitalizations meeting inclusion criteria, 103 (26%) met the composite outcome. The median age of the cohort was 64 (interquartile range, 53-75) with 168 (43%) Black patients and 169 (43%) women. The area under the receiver-operating characteristic curve of the EDI was 0.79 (95% confidence interval, 0.74-0.84). EDI predictions did not differ by race or sex. When exploring clinically relevant thresholds of the EDI, we found patients who met or exceeded an EDI of 68.8 made up 14% of the study cohort and had a 74% probability of experiencing the composite outcome during their hospitalization with a sensitivity of 39% and a median lead time of 24 hours from when this threshold was first exceeded. Among the 286 patients hospitalized for at least 48 hours who had not experienced the composite outcome, 14 (13%) never exceeded an EDI of 37.9, with a negative predictive value of 90% and a sensitivity above this threshold of 91%. Conclusions: We found the EDI identifies small subsets of high-risk and low-risk patients with COVID-19 with good discrimination, although its clinical use as an early warning system is limited by low sensitivity. These findings highlight the importance of independent evaluation of proprietary models before widespread operational use among patients with COVID-19.

Evaluating a Widely Implemented Proprietary Deterioration Index Model Among Hospitalized COVID-19 Patients.

INTRODUCTION: The Epic Deterioration Index (EDI) is a proprietary prediction model implemented in over 100 U.S. hospitals that was widely used to support medical decision-making during the COVID-19 pandemic. The EDI has not been independently evaluated, and other proprietary models have been shown to be biased against vulnerable populations. METHODS: We studied adult patients admitted with COVID-19 to non-ICU care at a large academic medical center from March 9 through May 20, 2020. We used the EDI, calculated at 15-minute intervals, to predict a composite outcome of ICU-level care, mechanical ventilation, or in-hospital death. In a subset of patients hospitalized for at least 48 hours, we also evaluated the ability of the EDI to identify patients at low risk of experiencing this composite outcome during their remaining hospitalization. RESULTS: Among 392 COVID-19 hospitalizations meeting inclusion criteria, 103 (26%) met the composite outcome. Median age of the cohort was 64 (IQR 53-75) with 168 (43%) African Americans and 169 (43%) women. Area under the receiver-operating-characteristic curve (AUC) of the EDI was 0.79 (95% CI 0.74-0.84). EDI predictions did not differ by race or sex. When exploring clinically-relevant thresholds of the EDI, we found patients who met or exceeded an EDI of 68.8 made up 14% of the study cohort and had a 74% probability of experiencing the composite outcome during their hospitalization with a median lead time of 24 hours from when this threshold was first exceeded. Among the 286 patients hospitalized for at least 48 hours who had not experienced the composite outcome, 14 (13%) never exceeded an EDI of 37.9, with a negative predictive value of 90% and a sensitivity above this threshold of 91%. CONCLUSION: We found the EDI identifies small subsets of high- and low-risk COVID-19 patients with fair discrimination. We did not find evidence of bias by race or sex. These findings highlight the importance of independent evaluation of proprietary models before widespread operational use among COVID-19 patients.

It's time to bring human factors to primary care policy and practice.

Primary health care is a complex, highly personal, and non-linear process. Care is often sub-optimal and professional burnout is high. Interventions intended to improve the situation have largely failed. This is due to a lack of a deep understanding of primary health care. Human Factors approaches and methods will aid in understanding the cognitive, social and technical needs of these specialties, and in designing and testing proposed innovations. In 2012, Ben-Tzion Karsh, Ph.D., conceived a transdisciplinary conference to frame the opportunities for research human factors and industrial engineering in primary care. In 2013, this conference brought together experts in primary care and human factors to outline areas where human factors methods can be applied. The results of this expert consensus panel highlighted four major research areas: Cognitive and social needs, patient engagement, care of community, and integration of care. Work in these areas can inform the design, implementation, and evaluation of innovations in Primary Care. We provide descriptions of these research areas, highlight examples and give suggestions for future research.

The Impact of Emergency Department Census on the Decision to Admit.

OBJECTIVE: We evaluated the effect of emergency department (ED) census on disposition decisions made by ED physicians. METHODS: We performed a retrospective analysis using 18 months of all adult patient encounters seen in the main ED at an academic tertiary care center. Patient census information was calculated at the time of physician assignment for each individual patient and included the number of patients in the waiting room (waiting room census) and number of patients being managed by the patient's attending (physician load census). A multiple logistic regression model was created to assess the association between these census variables and the disposition decision, controlling for potential confounders including Emergency Severity Index acuity, patient demographics, arrival hour, arrival mode, and chief complaint. RESULTS: A total of 49,487 patient visits were included in this analysis, of whom 37% were admitted to the hospital. Both census measures were significantly associated with increased chance of admission; the odds ratio (OR) per patient increase for waiting room census was 1.011 (95% confidence interval [CI] = 1.001 to 1.020), and the OR for physician load census was 1.010 (95% CI = 1.002 to 1.019). To put this in practical terms, this translated to a modeled rise from 35.3% to 40.1% when shifting from an empty waiting room and zero patient load to a 12-patient wait and 16-patient load for a given physician. CONCLUSION: Waiting room census and physician load census at time of physician assignment were positively associated with the likelihood that a patient would be admitted, controlling for potential confounders. Our data suggest that disposition decisions in the ED are influenced not only by objective measures of a patient's disease state, but also by workflow-related concerns.

Cherry Picking Patients: Examining the Interval Between Patient Rooming and Resident Self-assignment.

OBJECTIVE: We aimed to evaluate the association between patient chief complaint and the time interval between patient rooming and resident physician self-assignment ("pickup time"). We hypothesized that significant variation in pickup time would exist based on chief complaint, thereby uncovering resident preferences in patient presentations. METHODS: A retrospective medical record review was performed on consecutive patients at a single, academic, university-based emergency department with over 50,000 visits per year. All patients who presented from August 1, 2012, to July 31, 2013, and were initially seen by a resident were included in the analysis. Patients were excluded if not seen primarily by a resident or if registered with a chief complaint associated with trauma team activation. Data were abstracted from the electronic health record (EHR). The outcome measured was "pickup time," defined as the time interval between room assignment and resident self-assignment. We examined all complaints with >100 visits, with the remaining complaints included in the model in an "other" category. A proportional hazards model was created to control for the following prespecified demographic and clinical factors: age, race, sex, arrival mode, admission vital signs, Emergency Severity Index code, waiting room time before rooming, and waiting room census at time of rooming. RESULTS: Of the 30,382 patients eligible for the study, the median time to pickup was 6 minutes (interquartile range = 2-15 minutes). After controlling for the above factors, we found systematic and significant variation in the pickup time by chief complaint, with the longest times for patients with complaints of abdominal problems, numbness/tingling, and vaginal bleeding and shortest times for patients with ankle injury, allergic reaction, and wrist injury. CONCLUSIONS: A consistent variation in resident pickup time exists for common chief complaints. We suspect that this reflects residents preferentially choosing patients with simpler workups and less perceived diagnostic ambiguity. This work introduces pickup time as a metric that may be useful in the future to uncover and address potential physician bias. Further work is necessary to establish whether practice patterns in this study are carried beyond residency and persist among attendings in the community and how these patterns are shaped by the information presented via the EHR.