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

Pediatric sepsis imposes a significant burden of morbidity and mortality among children. While the speedy application of existing supportive care measures can substantially improve outcomes, further improvements in delivering that care require tools that go beyond recognizing sepsis and towards predicting its development. Machine learning techniques have great potential as predictive tools, but their application to pediatric sepsis has been stymied by several factors, particularly the relative rarity of its occurrence. We propose an alternate approach which focuses on predicting the provision of resuscitative care, rather than sepsis diagnoses or criteria themselves. Using three years of Emergency Department data from a large academic medical center, we developed a boosted tree model that predicts resuscitation within 6 hours of triage, and significantly outperforms existing rule-based sepsis alerts.

During the COVID-19 pandemic, rapid and accurate triage of patients at the emergency department is critical to inform decision-making. We propose a data-driven approach for automatic prediction of deterioration risk using a deep neural network that learns from chest X-ray images, and a gradient boosting model that learns from routine clinical variables. Our AI prognosis system, trained using data from 3,661 patients, achieves an AUC of 0.786 (95% CI: 0.742-0.827) when predicting deterioration within 96 hours. The deep neural network extracts informative areas of chest X-ray images to assist clinicians in interpreting the predictions, and performs comparably to two radiologists in a reader study. In order to verify performance in a real clinical setting, we silently deployed a preliminary version of the deep neural network at NYU Langone Health during the first wave of the pandemic, which produced accurate predictions in real-time. In summary, our findings demonstrate the potential of the proposed system for assisting front-line physicians in the triage of COVID-19 patients.

BACKGROUND:Automated systems that use machine learning to estimate a patient's risk of death are being developed to influence care. There remains sparse transparent reporting of model generalizability in different subpopulations especially for implemented systems. METHODS:A prognostic study included adult admissions at a multi-site, academic medical center between 2015 and 2017. A predictive model for all-cause mortality (including initiation of hospice care) within 60 days of admission was developed. Model generalizability is assessed in temporal validation in the context of potential demographic bias. A subsequent prospective cohort study was conducted at the same sites between October 2018 and June 2019. Model performance during prospective validation was quantified with areas under the receiver operating characteristic and precision recall curves stratified by site. Prospective results include timeliness, positive predictive value, and the number of actionable predictions. RESULTS:Three years of development data included 128,941 inpatient admissions (94,733 unique patients) across sites where patients are mostly white (61%) and female (60%) and 4.2% led to death within 60 days. A random forest model incorporating 9614 predictors produced areas under the receiver operating characteristic and precision recall curves of 87.2 (95% CI, 86.1-88.2) and 28.0 (95% CI, 25.0-31.0) in temporal validation. Performance marginally diverges within sites as the patient mix shifts from development to validation (patients of one site increases from 10 to 38%). Applied prospectively for nine months, 41,728 predictions were generated in real-time (median [IQR], 1.3 [0.9, 32] minutes). An operating criterion of 75% positive predictive value identified 104 predictions at very high risk (0.25%) where 65% (50 from 77 well-timed predictions) led to death within 60 days. CONCLUSION/CONCLUSIONS:Temporal validation demonstrates good model discrimination for 60-day mortality. Slight performance variations are observed across demographic subpopulations. The model was implemented prospectively and successfully produced meaningful estimates of risk within minutes of admission.

OBJECTIVE:<0.001). Rates of adverse events increased with the magnitude of D-dimer elevation; individuals with presenting D-dimer >2000 ng/mL had the highest risk of critical illness (66%), thrombotic event (37.8%), acute kidney injury (58.3%), and death (47%). CONCLUSIONS:Abnormal D-dimer was frequently observed at admission with COVID-19 and was associated with higher incidence of critical illness, thrombotic events, acute kidney injury, and death. The optimal management of patients with elevated D-dimer in COVID-19 requires further study.

Objective/UNASSIGNED:One primary consideration when developing predictive models is downstream effects on future model performance. We conduct experiments to quantify the effects of experimental design choices, namely cohort selection and internal validation methods, on (estimated) real-world model performance. Materials and Methods/UNASSIGNED:Four years of hospitalizations are used to develop a 1-year mortality prediction model (composite of death or initiation of hospice care). Two common methods to select appropriate patient visits from their encounter history (backwards-from-outcome and forwards-from-admission) are combined with 2 testing cohorts (random and temporal validation). Two models are trained under otherwise identical conditions, and their performances compared. Operating thresholds are selected in each test set and applied to a "real-world" cohort of labeled admissions from another, unused year. Results/UNASSIGNED: = 92 148). Both selection methods produce similar performances when applied to a random test set. However, when applied to the temporally defined "real-world" set, forwards-from-admission yields higher areas under the ROC and precision recall curves (88.3% and 56.5% vs. 83.2% and 41.6%). Discussion/UNASSIGNED:A backwards-from-outcome experiment manipulates raw training data, simplifying the experiment. This manipulated data no longer resembles real-world data, resulting in optimistic estimates of test set performance, especially at high precision. In contrast, a forwards-from-admission experiment with a temporally separated test set consistently and conservatively estimates real-world performance. Conclusion/UNASSIGNED:Experimental design choices impose bias upon selected cohorts. A forwards-from-admission experiment, validated temporally, can conservatively estimate real-world performance. LAY SUMMARY/UNASSIGNED:The routine care of patients stands to benefit greatly from assistive technologies, including data-driven risk assessment. Already, many different machine learning and artificial intelligence applications are being developed from complex electronic health record data. To overcome challenges that arise from such data, researchers often start with simple experimental approaches to test their work. One key component is how patients (and their healthcare visits) are selected for the study from the pool of all patients seen. Another is how the group of patients used to create the risk estimator differs from the group used to evaluate how well it works. These choices complicate how the experimental setting compares to the real-world application to patients. For example, different selection approaches that depend on each patient's future outcome can simplify the experiment but are impractical upon implementation as these data are unavailable. We show that this kind of "backwards" experiment optimistically estimates how well the model performs. Instead, our results advocate for experiments that select patients in a "forwards" manner and "temporal" validation that approximates training on past data and implementing on future data. More robust results help gauge the clinical utility of recent works and aid decision-making before implementation into practice.

BACKGROUND:Although clinical decision support (CDS) alerts are effective reminders of best practices, their effectiveness is blunted by clinicians who fail to respond to an overabundance of inappropriate alerts. An electronic health record (EHR)-integrated machine learning (ML) algorithm is a potentially powerful tool to increase the signal-to-noise ratio of CDS alerts and positively impact the clinician's interaction with these alerts in general. OBJECTIVE:This study aimed to describe the development and implementation of an ML-based signal-to-noise optimization system (SmartCDS) to increase the signal of alerts by decreasing the volume of low-value herpes zoster (shingles) vaccination alerts. METHODS:We built and deployed SmartCDS, which builds personalized user activity profiles to suppress shingles vaccination alerts unlikely to yield a clinician's interaction. We extracted all records of shingles alerts from January 2017 to March 2019 from our EHR system, including 327,737 encounters, 780 providers, and 144,438 patients. RESULTS:During the 6 weeks of pilot deployment, the SmartCDS system suppressed an average of 43.67% (15,425/35,315) potential shingles alerts (appointments) and maintained stable counts of weekly shingles vaccination orders (326.3 with system active vs 331.3 in the control group; P=.38) and weekly user-alert interactions (1118.3 with system active vs 1166.3 in the control group; P=.20). CONCLUSIONS:All key statistics remained stable while the system was turned on. Although the results are promising, the characteristics of the system can be subject to future data shifts, which require automated logging and monitoring. We demonstrated that an automated, ML-based method and data architecture to suppress alerts are feasible without detriment to overall order rates. This work is the first alert suppression ML-based model deployed in practice and serves as foundational work in encounter-level customization of alert display to maximize effectiveness.