Faculty Bibliography

Objective/UNASSIGNED:The widespread deployment of electronic health records (EHRs) has introduced new sources of error and inefficiencies to the process of ordering medications in the hospital setting. Existing work identifies orders that require pharmacy intervention by comparing them to a patient's medical records. In this work, we develop a machine learning model for identifying medication orders requiring intervention using only provider behavior and other contextual features that may reflect these new sources of inefficiencies. Materials and Methods/UNASSIGNED:Data on providers' actions in the EHR system and pharmacy orders were collected over a 2-week period in a major metropolitan hospital system. A classification model was then built to identify orders requiring pharmacist intervention. We tune the model to the context in which it would be deployed and evaluate global and local feature importance. Results/UNASSIGNED:The resultant model had an area under the receiver-operator characteristic curve of 0.91 and an area under the precision-recall curve of 0.44. Conclusions/UNASSIGNED:Providers' actions can serve as useful predictors in identifying medication orders that require pharmacy intervention. Careful model tuning for the clinical context in which the model is deployed can help to create an effective tool for improving health outcomes without using sensitive patient data.

Importance/UNASSIGNED:Black and Hispanic populations have higher rates of coronavirus disease 2019 (COVID-19) hospitalization and mortality than White populations but lower in-hospital case-fatality rates. The extent to which neighborhood characteristics and comorbidity explain these disparities is unclear. Outcomes in Asian American populations have not been explored. Objective/UNASSIGNED:To compare COVID-19 outcomes based on race and ethnicity and assess the association of any disparities with comorbidity and neighborhood characteristics. Design, Setting, and Participants/UNASSIGNED:This retrospective cohort study was conducted within the New York University Langone Health system, which includes over 260 outpatient practices and 4 acute care hospitals. All patients within the system's integrated health record who were tested for severe acute respiratory syndrome coronavirus 2 between March 1, 2020, and April 8, 2020, were identified and followed up through May 13, 2020. Data were analyzed in June 2020. Among 11 547 patients tested, outcomes were compared by race and ethnicity and examined against differences by age, sex, body mass index, comorbidity, insurance type, and neighborhood socioeconomic status. Exposures/UNASSIGNED:Race and ethnicity categorized using self-reported electronic health record data (ie, non-Hispanic White, non-Hispanic Black, Hispanic, Asian, and multiracial/other patients). Main Outcomes and Measures/UNASSIGNED:The likelihood of receiving a positive test, hospitalization, and critical illness (defined as a composite of care in the intensive care unit, use of mechanical ventilation, discharge to hospice, or death). Results/UNASSIGNED:Among 9722 patients (mean [SD] age, 50.7 [17.5] years; 58.8% women), 4843 (49.8%) were positive for COVID-19; 2623 (54.2%) of those were admitted for hospitalization (1047 [39.9%] White, 375 [14.3%] Black, 715 [27.3%] Hispanic, 180 [6.9%] Asian, 207 [7.9%] multiracial/other). In fully adjusted models, Black patients (odds ratio [OR], 1.3; 95% CI, 1.2-1.6) and Hispanic patients (OR, 1.5; 95% CI, 1.3-1.7) were more likely than White patients to test positive. Among those who tested positive, odds of hospitalization were similar among White, Hispanic, and Black patients, but higher among Asian (OR, 1.6, 95% CI, 1.1-2.3) and multiracial patients (OR, 1.4; 95% CI, 1.0-1.9) compared with White patients. Among those hospitalized, Black patients were less likely than White patients to have severe illness (OR, 0.6; 95% CI, 0.4-0.8) and to die or be discharged to hospice (hazard ratio, 0.7; 95% CI, 0.6-0.9). Conclusions and Relevance/UNASSIGNED:In this cohort study of patients in a large health system in New York City, Black and Hispanic patients were more likely, and Asian patients less likely, than White patients to test positive; once hospitalized, Black patients were less likely than White patients to have critical illness or die after adjustment for comorbidity and neighborhood characteristics. This supports the assertion that existing structural determinants pervasive in Black and Hispanic communities may explain the disproportionately higher out-of-hospital deaths due to COVID-19 infections in these populations.

BACKGROUND:There is concern about the potential of an increased risk related to medications that act on the renin-angiotensin-aldosterone system in patients exposed to coronavirus disease 2019 (Covid-19), because the viral receptor is angiotensin-converting enzyme 2 (ACE2). METHODS:We assessed the relation between previous treatment with ACE inhibitors, angiotensin-receptor blockers, beta-blockers, calcium-channel blockers, or thiazide diuretics and the likelihood of a positive or negative result on Covid-19 testing as well as the likelihood of severe illness (defined as intensive care, mechanical ventilation, or death) among patients who tested positive. Using Bayesian methods, we compared outcomes in patients who had been treated with these medications and in untreated patients, overall and in those with hypertension, after propensity-score matching for receipt of each medication class. A difference of at least 10 percentage points was prespecified as a substantial difference. RESULTS:Among 12,594 patients who were tested for Covid-19, a total of 5894 (46.8%) were positive; 1002 of these patients (17.0%) had severe illness. A history of hypertension was present in 4357 patients (34.6%), among whom 2573 (59.1%) had a positive test; 634 of these patients (24.6%) had severe illness. There was no association between any single medication class and an increased likelihood of a positive test. None of the medications examined was associated with a substantial increase in the risk of severe illness among patients who tested positive. CONCLUSIONS:We found no substantial increase in the likelihood of a positive test for Covid-19 or in the risk of severe Covid-19 among patients who tested positive in association with five common classes of antihypertensive medications.

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.

PURPOSE/OBJECTIVE:To develop natural language processing (NLP) to identify incidental lung nodules (ILNs) in radiology reports for assessment of management recommendations. METHOD AND MATERIALS/METHODS:We searched the electronic health records for patients who underwent chest CT during 2014 and 2017, before and after implementation of a department-wide dictation macro of the Fleischner Society recommendations. We randomly selected 950 unstructured chest CT reports and reviewed manually for ILNs. An NLP tool was trained and validated against the manually reviewed set, for the task of automated detection of ILNs with exclusion of previously known or definitively benign nodules. For ILNs found in the training and validation sets, we assessed whether reported management recommendations agreed with Fleischner Society guidelines. The guideline concordance of management recommendations was compared between 2014 and 2017. RESULTS:The NLP tool identified ILNs with sensitivity and specificity of 91.1% and 82.2%, respectively, in the validation set. Positive and negative predictive values were 59.7% and 97.0%. In reports of ILNs in the training and validation sets before versus after introduction of a Fleischner reporting macro, there was no difference in the proportion of reports with ILNs (108 of 500 [21.6%] versus 101 of 450 [22.4%]; P = .8), or in the proportion of reports with ILNs containing follow-up recommendations (75 of 108 [69.4%] versus 80 of 101 [79.2%]; P = .2]. Rates of recommendation guideline concordance were not significantly different before and after implementation of the standardized macro (52 of 75 [69.3%] versus 60 of 80 [75.0%]; P = .43). CONCLUSION/CONCLUSIONS:NLP reliably automates identification of ILNs in unstructured reports, pertinent to quality improvement efforts for ILN management.

Background. Pneumonia (PNU) is the second most common nosocomial infection in the United States and is associated with substantial morbidity and mortality. While definitions from CDC were developed to increase the reliability of surveillance data, reduce the burden of surveillance in healthcare facilities, and enhance the utility of surveillance data for improving patient safety - the algorithm is still laborious. We propose an implementation of a refined algorithm script which combines two CDC definitions with the use of natural language processing (NLP), a tool which relies on pattern matching to determine whether a condition of interest is reported as present or absent in a report, to automate PNU surveillance. Methods. Using SAS v9.4 to write a query, we used a combination of National Healthcare Safety Network's (NHSN) PNU and ventilator-associated event (VAE) definitions that use discrete fields found in electronic medical records (EMR) and trained an NLP tool to determine whether chest x-ray report was indicative of PNU (Fig1). To validate, we assessed sensitivity/specificity of NLP tool results compared with clinicians' interpretations. Results. The NLP tool was highly accurate in classifying the presence of PNU in chest x-rays. After training the NLP tool, there were only 4% discrepancies between NLP tool and clinicians interpretations of 223 x-ray reports - sensitivity 92.2% (81.1-97.8), specificity 97.1% (93.4-99.1), PPV 90.4% (79.0-96.8), NPV 97.7% (94.1- 99.4). Combining the automated use of discrete EMR fields with NLP tool significantly reduces the time spent manually reviewing EMRs. A manual review for PNU without automation requires approximately 10 minutes each day per admission. With a monthly average of 2,350 adult admissions at our hospital and 16,170 patient-days for admissions with at least 2 days, the algorithm saves approximately 2,695 review hours. Conclusion. The use of discrete EMR fields with an NLP tool proves to be a timelier, cost-effective yet accurate alternative to manual PNU surveillance review. By allowing an automated algorithm to review PNU, timely reports can be sent to units about individual cases. Compared with traditional CDC surveillance definitions, an automated tool allows real-time critical review for infection and prevention activities