Faculty Bibliography

BACKGROUND: Clinical reasoning (CR) is a core component of medical training, yet residents often receive little feedback on their CR documentation. Here we describe the process of developing a machine learning (ML) algorithm for feedback on CR documentation to increase the frequency and quality of feedback in this domain.
METHOD(S): To create this algorithm, note quality first had to be rated by gold standard human rating. We selected the IDEA Assessment Tool-a note rating instrument across four domains (I=Interpretive summary, D=Differential diagnosis, E=Explanation of reasoning, A=Alternative diagnoses explained) that uses a 3-point Likert scale without descriptive anchors. To develop descriptive anchors we conducted an iterative process reviewing notes from the EHR written by medicine residents and validated the Revised-IDEA Assessment Tool using Messick's framework- content validity, response process, relation to other variables, internal structure, and consequences. Using the Hofstee standard setting method, cutoffs for high quality clinical reasoning for the IDEA and DEA scores were set. We then created a dataset of expertrated notes to create the ML algorithm. First, a natural language processing software was applied to the set of notes that enabled recognition and automatic encoding of clinical information as a diagnosis or disease (D's), a sign or symptom (E or A), or semantic qualifier (e.g. most likely). Input variables to the ML algorithm included counts of D's, E/A's, semantic qualifiers, and proximity of semantic qualifiers to disease/ diagnosis. ML output focused on DEA quality and was binarized to low or high quality CR. Finally, 200 notes were randomly selected for human validation review comparing ML output to human rated DEA score.
RESULT(S): The IDEA and DEA scores ranged from 0-10 and 0-6, respectively. IDEA score of >= 6.5 and a DEA score of >= 3 was deemed high quality. 252 notes were rated to create the dataset and 20% were rated by 3 raters with high intraclass correlation 0.84 (95% CI 0.74-0.90). 120 of these notes comprised the testing set for ML model development. The logistic regression model was the best performing model with an AUC 0.87 and a positive predictive value (PPV) of 0.65. 48 (40%) of the notes were high quality. There was substantial interrater reliability between ML output and human rating on the 200 note validation set with a Cohen's Kappa 0.64.
CONCLUSION(S): We have developed a ML algorithm for feedback on CR documentation that we hypothesize will increase the frequency and quality of feedback in this domain. We have subsequently developed a dashboard that will display the output of the ML model. Next steps will be to provide internal medicine residents' feedback on their CR documentation using this dashboard and assess the impact this has on their documentation quality. LEARNING OBJECTIVE #1: Describe the importance of high quality documentation of clinical reasoning. LEARNING OBJECTIVE #2: Identify machine learning as a novel assessment tool for feedback on clinical reasoning documentation

Background: Heart failure (HF) is an important contributor to the increased cardiovascular (CV) mortality incidence in ESKD. Therapies targeting HF's unique pathophysiology in ESKD are lacking. Hydralazine-isosorbide dinitrate (H-ISDN) targets reduced nitric oxide bioavailability and could improve CV mortality in ESKD Methods: Adult patients with HF on maintenance dialysis between January 2011 and December 31, 2016 were identified using the United States Renal Data System. There were 6306 patients with at least one prescription for H-ISDN and 75,851 non-users. The primary outcome was death from any cause. Secondary outcomes included cardiovascular death and sudden death. Treatment effects were estimated using stabilized inverse probability weights in Cox proportional hazards regression. Because H-ISDN has been shown to improve outcomes in Black HF patients, we investigated effect modification by race Results: Age was similar in H-ISDN users (66 +/- 13 years) and non-users (69 +/- 13 years) with 50% and 51% men, respectively. H-ISDN (51%) users were more likely to be of Black race than non-users (27%). Dialysis vintage was longer in H-ISDN (25 months) users compared with non-users (15 months). All characteristics were well balanced in weighted models. Risks of all-cause mortality, cardiovascular death, and sudden death were significantly reduced in H-ISDN users compared to non-users (Table). We did not identify significant effect modification by race (Figure)
Conclusion(s): To our knowledge, this is the first analysis of the impact of H-ISDN on mortality in ESKD. Our results suggest that combination H-ISDN improves survival in dialysis patients with HF

BACKGROUND: On the road to becoming competent, compassionate and ethical physicians, trainees need to reflect on their experiences, understand the clinical and social contexts, and integrate cognitive and affective reactions in ways that build resilience and a coherent professional identity. Using a qualitative approach, this study seeks to identify medical residents' stressors and challenges, and to understand their experience of the internal and external factors of such situations. Such information can guide educators to develop curricula that better meet residents' needs.
METHOD(S): Primary Care residents at NYU School of Med have ongoing Psychosocial Rounds (PSR) throughout their 3 years, facilitated by a faculty member and Chief Resident, where residents present challenging cases or situations, framed by a specific question. Semi-structured notes taken by facilitators, including question, case description, process of discussion and teaching points were compiled into a deidentified database of 119 cases spanning 2010-2019. These notes were coded by three coders using iterative thematic analysis.
RESULT(S): Seventy four of the 119 cases have been coded to date. Four general themes emerged, with each comprised of 2 to 4 main codes. These themes were 1) Self (S): including management of medical uncertainty, emotional reactions, roles and responsibilities, self-care; 2) Teams (T): including relationship with peers, supervisors, other health professionals; 3) Understanding Patient and Families (PF): including social and cultural context, mental health issues, patient/ family and provider disagreements; and 4) Hospital, Healthcare and Societal issues (HHS). There was a high co- occurrence of themes within cases, 60% had 2 themes present, 24% had 3, and only 16% had one theme. Cases with 3 themes most often included S, T and PF.
CONCLUSION(S): This analysis of PSR cases identifies issues for which residents seek help and support in a safe, case-oriented problem-solving discussion group, and allows for in-depth reflection and exploration. The cooccurrences of themes indicate the complexity of issues faced, and the importance of integrating multiple domains when beginning to understand these issues. LEARNING OBJECTIVE #1: Professionalism: Coping with challenges of becoming resilient physician with emotional and cognitive capacity to deal with complex situations LEARNING OBJECTIVE #2: Interpersonal and Communication Skills: Develop awareness and skills to negotiate interpersonal situations

LEARNINGOBJECTIVES 1: Interpersonal /Communication Skill: 1) Identify communication skills needed to recruit older adults LEARNING OBJECTIVES 2: 2) Assess feasibility of GOSCEs to enhance recruitment skills in RAs. SETTING AND PARTICIPANTS: Convenience sample of 18 (5 male, 13 female) Research Assistants (RAs) at an urban hospital who recruit older adults for clinical trials. DESCRIPTION: Increasing older adults' participation in clinical trials is urgently needed. We developed a remote, three station simulation (Group Objective Structured Clinical Exam - GOSCE) to teach RAs communication skills. This 2-hour course included a discussion of challenges in recruiting older adults; skills practice with Standardized Participants (SPs); and a debrief to review experiences, highlight best practices. After discussion, RAs rotated (3 per group) through the stations, each with SP and faculty observer who provided immediate feedback. Thus, learners had opportunities for active and observational learning.Scenarios were: 1) an older white woman with hearing impairment; 2) an older white woman and family member together; and 3) an older Black man mistrustful due to history of racism in medical research. SPs completed behaviorally anchored checklists (11 communication skills across all cases, and 5-7 case-specific questions). Learners completed a 36- item survey of self-assessed change in skill after the workshop; insights on recruitment practice; and educational value. EVALUATION: The communication checklist across all cases included: relationship development (5 items, mean of 58% well done (range: 50-75%), patient education (3 items, 44% (42-58%)), patient satisfaction (2 items, 54% (50-58%)), and information gathering (1 item, 92%). Seventeen RAs completed the survey, 100% felt the workshop provided valuable feedback and taught relevant material, 88% would participate again and 52%reported that the workshop improved their recruitment skills. All RAs reported encountering situations similar to hearing impairment and family member cases, and the majority rated the cases as high in educational value. Just 45% reported experiencing a case similar to the Black male case, and 100% rate it as high in educational value. Key points identified by RAs included the value of building a trusting relationship with potential subjects, recognizing possible barriers to communication early on and addressing these directly in a supportive and respectful style. DISCUSSION / REFLECTION / LESSONS LEARNED: Remote GOSCEs are a feasible mechanism for training RAs in subject recruitment focused on the unique needs of older adults. Responses to the RA survey suggest that GOSCEs are feasible for training RAs in simulated clinical scenarios with which participants are familiar and unfamiliar. SP assessment of RAs identified areas for further reinforcement to improve recruitment skills. This innovation is a feasible, high yield strategy for training research staff. It is highly adaptable to the specific recruitment needs and skills of a clinical trials and will add to the literature on educating RAs

A low serum bicarbonate (SB) level is predictive of adverse outcomes in kidney injury, infection, and aging. Because the liver plays an important role in acid-base homeostasis and lactic acid metabolism, we speculated that such a relationship would exist for patients with cirrhosis. To assess the prognostic value of admission SB on adverse hospital outcomes, clinical characteristics were extracted and analyzed from a large electronic health record system. Patients were categorized based on admission SB (mEq/L) into 7 groups based on the reference range (22-25) into mildly (18-21), moderately (14-17), and severely (<14) decreased groups and mildly (26-29), moderately (30-33), and severely (>30) increased groups, and the relationship of SB category with the frequency of complications (acute kidney injury/hepatorenal syndrome, portosystemic encephalopathy, gastrointestinal bleeding, ascites, and spontaneous bacterial peritonitis) and hospital metrics (length of stay [LOS], admission to an intensive care unit [ICU], and mortality) was assessed. A total of 2,693 patients were analyzed. Mean SB was 22.9 ± 4.5 mEq/L. SB was within the normal range (22-25 mEq/L) in 1,072 (39.8%) patients, and 955 patients (36%) had a low SB. As the SB category decreased, the incidence of complications progressively increased (p < 0.001). Increased MELD-Na score and low serum albumin also correlated with frequency of complications (p < 0.001). As the SB category decreased, LOS, ICU admission, and mortality progressively increased (p < 0.001). On multivariate analysis, the association of decreased SB with higher odds of complications, LOS, ICU admission, and mortality persisted. Conclusion. Low admission SB in patients with cirrhosis is associated with cirrhotic complications, longer LOS, increased ICU admissions, and increased hospital mortality.

Background Prior studies of SLE clusters based on autoantibodies have utilized cross-sectional data from single centers. We applied clustering techniques to longitudinal and comprehensive autoantibody data from a large multinational, multiethnic inception cohort of well characterized SLE patients to identify clusters associated with disease outcomes. Methods We used demographic, clinical, and serological data at enrolment and follow-up visits years 3 and 5 from 805 patients who fulfilled the 1997 Updated ACR SLE criteria and were enrolled within 15 months of diagnosis. For each visit, ANA, dsDNA, Sm, U1-RNP, SSA/Ro60, SSB/La, Ro52/ TRIM21, histones, ribosomal P, Jo-1, centromere B, PCNA, anti-DFS70, lupus anticoagulant (LAC), IgG and IgM for anticardiolipin, anti-b2GP1, and aPS/PT, and IgG anti-b2GP1 D1 were performed at a single lab (except LAC). K-means clustering algorithm on principal component analysis (10 dimensions) transformed longitudinal ANA/autoantibody profiles was used. We compared cluster demographic/clinical outcomes, including longitudinal disease activity (total and adjusted mean SLEDAI- 2K), SLICC/ACR damage index and organ-specific domains, SLE therapies, and survival, using one-way ANOVA test and a Benjamini-Hochberg correction with false discovery rate alpha=0.05. Results were visualized using t-distributed stochastic neighbor embedding. Results Four unique patient clusters were identified (table 1). Cluster 1, characterized by high frequency of anti-Sm and anti-RNP over time, was the youngest group at disease onset with a high proportion of subjects of Asian and African ancestry. At year 5, they had the highest disease activity, were more likely to have active hematologic and mucocutaneous involvement, and to be on/exposed to immunosuppressants/ biologics. Cluster 2, the largest cluster, had low frequency of anti-dsDNA, were oldest at disease onset, and at year 5, had the lowest disease activity, and were least likely to have nephritis and be on/exposed to immunosuppressants/biologics. Cluster 3 had the highest frequency of antiphospholipid antibodies over time, were more likely to be of European ancestry, have an elevated BMI, be former smokers, and by year 5, to have nephritis, neuropsychiatric involvement, including strokes and seizures (SLICC/ACR damage index). Cluster 4 was characterized by anti-SSA/Ro60, SSB/La, Ro52/TRIM21, histone antibodies, and low complements at year 5. Overall, survival of the 805 subjects was 94% at 5 years, and none of the clusters predicted survival. Conclusions Four SLE patient clusters associated with disease activity, organ involvement, and treatment were identified in this analysis of longitudinal ANA/autoantibody profiles in relation to SLE outcomes, suggesting these subsets might be identifiable based on extended autoantibody profiles early in disease and carry prognostic information

Bacterial infections represent a major cause of mortality and morbidity in patients with cirrhosis that can alter the clinical course of compensated cirrhosis. The most common infections are spontaneous bacterial peritonitis by gram-negative organisms, urinary-tract infection, and pneumonia. In this case report, we raise the question of considering infections in the prognosis scoring in this patient group.