Beth Israel Deaconess Medical Center Research Articles

COVID-19 CXR Classification: Applying Domain Extension Transfer Learning and Deep Learning

The infectious coronavirus disease-19 (COVID-19) is a viral disease that affects the lungs, which caused great havoc when the epidemic rapidly spread around the world. Polymerase chain reaction (PCR) tests are conducted to screen for COVID-19 and respond to quarantine measures. However, PCR tests take a considerable amount of time to confirm the test results. Therefore, to supplement the accuracy and quickness of a COVID-19 diagnosis, we proposed an effective deep learning methodology as a quarantine response through COVID-19 chest X-ray image classification based on domain extension transfer learning. As part of the data preprocessing, contrast limited adaptive histogram equalization was applied to chest X-ray images using Medical Information Mart for Intensive Care (MIMIC)-IV obtained from the Beth Israel Deaconess Medical Center. The classification of the COVID-19 X-ray images was conducted using a pretrained ResNet-50. We also visualized and interpreted the classification performance of the model through explainable artificial intelligence and performed statistical tests to validate the reliability of the model. The proposed method correctly classified images with 96.7% accuracy, an improvement of about 9.9% over the reference model. This study is expected to help medical staff make an integrated decision in selecting the first confirmed case and contribute to suppressing the spread of the virus in the community.

The Effect of Obstructive Sleep Apnea on Sleep-dependent Emotional Memory Consolidation.

Rationale: A growing body of evidence suggests that sleep is critical for the adaptive processing and consolidation of emotional information into long-term memory. Previous research has indicated that emotional components of scenes particularly benefit from sleep in healthy groups, yet sleep-dependent emotional memory processes remain unexplored in clinical cohorts, including those with obstructive sleep apnea (OSA). This line of research is important as it will add to the understanding of how disrupted sleep in OSA contributes to both impaired cognition and emotion dysregulation. Objectives: To test the hypothesis that individuals with OSA will have impaired sleep-dependent memory consolidation, with the greatest impact being on memory for emotional content. Methods: In this study, a group of newly diagnosed patients with OSA (n = 26; 10 female; average age, 42.5 years) and a matched group of healthy control subjects (n = 24; 13 female; average age, 37 years) were enrolled in the study at Beth Israel Deaconess Medical Center. Participants encoded scenes with negative or neutral foreground objects placed on neutral backgrounds before a night of polysomnographically recorded sleep. In the morning, they completed a recognition test in which old and new scene objects and backgrounds, presented separately and one at a time, were judged as old, new, or similar compared with what had been previously viewed. Results: Patients with OSA had a deficit in recognition memory for the scenes. Overall recognition (the ability to recognize old items as either old or similar) was impaired across all scene elements, both negative and neutral objects and backgrounds, whereas specific recognition (correctly identifying old items as old) was impaired only for negative objects. Across all participants, successful overall recognition correlated positively with sleep efficiency and rapid eye movement (REM) sleep, whereas successful specific memory recognition correlated only with REM sleep. Conclusions: Our findings indicate that fragmented sleep and reduced REM sleep, both hallmarks of OSA, are associated with disruptions in general memory impairment and veridical memory for emotional content, which could alter emotional regulation and contribute to comorbid emotional distress in OSA.

Geographic Reach of Surgery Residency Interviews During In-Person and Virtual Interviews

INTRODUCTION: Virtual recruitment for Surgery residency may improve interview opportunities for applicants from URM and lower socioeconomic backgrounds. We aim to estimate the impact of virtual interviews for Surgery residency on applicants’ geographic reach. METHODS: Retrospective cohort of applications for surgery residency for the 2019 (in-person) and 2020 (virtual interviews) in an academic surgery residency. Geographic reach (GR) was the distance between applicants’ current location and Beth Israel Deaconess Medical Center where interviews were conducted. Demographics, USMLE scores and geographic distance to target program were collected. Federal Financial Institutions Examination Council’s (FFIEC) website was used to obtain socioeconomic data. Multivariable analyses with interaction terms were conducted. RESULTS: Six hundred sixty-seven and 698 applications were reviewed in the 2019 and 2020 match, respectively. Overall, there was no difference in GR for in-person and virtual interviews (p = 0.826). For non-URM applicants, virtual interviews had no impact on geographic reach (-52.57 miles, 95% CI -150.60-45.47; Table 4) while for URM applicants, virtual interviews increased GR by 240.14 miles (95% CI 7.55-472.72, p = 0.043). For applicants from communities with low poverty levels, virtual interviews decreased GR by 195.33 miles (95% CI 372.53-18.12, p = 0.0316) while for applicants from communities with high poverty levels, virtual interviews increased GR by 238.91 miles (95% CI 34.08-443.73, p = 0.022). Analyses controlled for USMLE Step 1 score (p > 0.05). CONCLUSION: Virtual recruiting might increase interview opportunities for surgery residency for applicants of URM and lower socioeconomic backgrounds. Educational leadership should optimize virtual recruitment opportunities as part of the strategies to promote diversity in surgical training.

INITIAL EXPERIENCE WITH INTRAPROCEDURAL ELECTRICAL IMPEDANCE TOMOGRAPHY FOR EVALUATING REGIONAL VENTILATION CHANGES DURING BRONCHOSCOPIC LUNG VOLUME REDUCTION

INITIAL EXPERIENCE WITH INTRAPROCEDURAL ELECTRICAL IMPEDANCE TOMOGRAPHY FOR EVALUATING REGIONAL VENTILATION CHANGES DURING BRONCHOSCOPIC LUNG VOLUME REDUCTION

INDWELLING PLEURAL CATHETERS FOR TREATMENT OF RECURRENT PLEURAL EFFUSION DUE TO CONGESTIVE HEART FAILURE: A SINGLE CENTER EXPERIENCE

INDWELLING PLEURAL CATHETERS FOR TREATMENT OF RECURRENT PLEURAL EFFUSION DUE TO CONGESTIVE HEART FAILURE: A SINGLE CENTER EXPERIENCE

EpiPen STAT: A quality improvement project for anaphylaxis management.

319 Background: The medical oncology infusion unit at Beth Israel Deaconess Medical Center (BIDMC) experienced six anaphylactic reactions to chemotherapy between June 2000 and April 2021. EpiPen was delayed in all cases. In one instance, a patient experienced an emergency where she could not breathe, and her blood pressure continued to decrease. The covering nurse hesitated before administering an EpiPen. In follow-up, nurses reported a lack of confidence in administering an EpiPen. No licensed independent provider (LIP) was stationed in the infusion unit to guide nurses. The project's goal was to facilitate timely EpiPen administration, guided by the Translating Evidence into Practice (TRIP) framework. Methods: The project lead and the medical director revised the hospital’s chemotherapy reaction guidelines to highlight anaphylaxis management. They established an interdisciplinary project team with medical, nursing, pharmacy, and informatics representatives. Interventions, which started in November 2021, included a reaction management algorithm and PRN-reaction orders allowing nurses to initiate interventions while waiting for a LIP arrival. The unit educator provided EpiPen training, and the pharmacy leaders created an adverse drug reactions (ADR) kit—a locked box with necessary medications that followed BIDMC’s institutional algorithm to manage infusion reactions. Results: The project team collected post-intervention data between December 2021 and April 2022. Nurses reported twenty mild-to-severe chemotherapy reactions, including four anaphylactic episodes. Nineteen patients were discharged home on the day of reactions, and one patient required hospital admission for persistent symptoms. The management algorithm was followed in ten reactions (50%). Deviations from the algorithm were noted in eight cases, with the most frequent reason being smaller doses of antihistamine or corticosteroid. The algorithm was not applicable in two cases with atypical patient symptoms. In seventeen of the twenty reactions (85%), PRN-reaction orders were available in the electronic ordering system. The ADR kit was used in all reactions. In two of the four anaphylactic reactions, EpiPen was administered immediately following hypotension or hypoxia. In the remaining two cases, it was administered for persistent symptoms. Conclusions: The project facilitated EpiPen administration. The critical steps were systemic analysis of barriers, problem-solving with the TRIP framework, and engagement of frontline nurses. In our next step, we will consider narrowing the scope to managing acute hypotension or hypoxia. The narrower scope will enable targeted education about emergency management, potentially at the hospital’s simulation center. To ensure every patient has an order, the department could consider a standing EpiPen order for hemodynamically unstable patients.

Invited Commentary: Importance of a Multidisciplinary Approach to Pulmonary Embolism in the Era of Catheter-directed Therapies.

HomeRadioGraphicsVol. 42, No. 6 PreviousNext Pulmonary Embolism and Deep Vein ThrombosisInvited Commentary: Importance of a Multidisciplinary Approach to Pulmonary Embolism in the Era of Catheter-directed TherapiesJulie C. Bulman , Jeffrey L. WeinsteinJulie C. Bulman , Jeffrey L. WeinsteinAuthor AffiliationsFrom the Department of Radiology, Beth Israel Deaconess Medical Center, Deaconess Road, Boston, MA 02215.Address correspondence to J.C.B. (email: [email protected]).Julie C. Bulman Jeffrey L. WeinsteinPublished Online:Oct 3 2022https://doi.org/10.1148/rg.220174MoreSectionsFull textPDF ToolsImage ViewerAdd to favoritesCiteTrack CitationsPermissionsReprints ShareShare onFacebookTwitterLinked In

On the growth of young neurologists—an interview with Prof. Louis R. Caplan

The article is an interview with Prof. Louis R. Caplan of the Department of Neurology, Beth Israel Deaconess Medical Center at the Harvard Medical School, conducted by Qi Qin from the National Clinical Research Center for Geriatric Diseases at the Capital Medical University, on behalf of Aging Pathobiology and Therapeutics.

Latest in Resuscitation Research: Highlights From the 2021 American Heart Association's Resuscitation Science Symposium.

Latest in Resuscitation Research: Highlights From the 2021 American Heart Association's Resuscitation Science Symposium.

Increased variability of mean arterial pressure is associated with increased risk of short-term mortality in intensive care unit: A retrospective study.

BackgroundIn intensive care unit (ICU), what thresholds of MAP variability are effective in distinguishing low- and high-risk patients for short-term mortality (in-hospital and 28-day) remains unclear.MethodsFifteen thousand five hundred sixty adult subjects admitted to ICU at Beth Israel Deaconess Medical Center (Boston, USA) between 2001 and 2012 were included in this retrospective study from MIMIC-III database. MAP within the first 24 h after admission were collected. Quantiles of MAP variability from 10% to 90% with 10% increasement each were considered to divide study participants into two groups, either having coefficients of variation of MAP greater or less than the given threshold. The threshold of MAP variability was identified by maximizing the odds ratio associated with increased risk of short-term mortality (in-hospital and 28-day). Logistic regression and Cox regression models were further applied to evaluate the association between increased variability of MAP and short-term mortality (in-hospital and 28-day).Results90% quantile of MAP variability was determined as the threshold generating the largest odds ratio associated with the increased risk of short-term mortality. Increased MAP variability, especially over 90% of MAP variability, was associated with increased risk of in-hospital mortality (odds ratio: 2.351, 95% CI: 2.064–2.673), and 28-day mortality (hazard ratio: 2.064, 95% CI: 1.820–2.337).ConclusionIncreased MAP variability, especially over 90% of MAP variability, is associated with short-term mortality. Our proposed threshold of MAP variability may aid in the early identification of critically ill patients with a high risk of mortality.

An efficient honey badger based Faster region CNN for chronc heart Failure prediction

If the blood circulation of the heart is not adequate then it causes arrhythmias and Congestive Heart Failure (CHF) which requires immediate medical attention or else it leads to the loss of one's life. An Electrocardiogram (ECG) is a golden standard to diagnose the fatal complications in the heart caused by arrhythmias and it comprises a massive information related to the heartbeat rhythm. The main challenge focused in this paper is to extract the crucial information present in the ECG signal by visual analysis and classify the different abnormalities exhibited in the ECG signal. This paper presents a Honey Badger Algorithm optimized Faster Region-based Convolutional Neural Network (HBA-FRCNN) for CHF prediction with higher diagnostic accuracy. The noisy input ECG signals such as muscle contraction, electrode touch noise, and different noise artifacts are preprocessed using the Delayed Normalized Least Mean Square (DNLMS). The electrocardiographic complex (QRS complex) consisting of the Q, R, and S waves are extracted using the Discrete Cosine Transform (DCT) and fast Fourier transforms(FFT). The target detection box and the anchor parameter for the FRCNN model are tuned using the HBA algorithm to overcome the missed target detection, overfitting, and computational cost. The ECG signals for this study were obtained from Beth Israel Deaconess Medical Center (BIDMC) Congestive Heart Failure Database and the Massachusetts Institute of Technology-Beth Israel Hospital (MIT-BIH) Normal Sinus Rhythm Database. The proposed methodology offers an accuracy, positive predictive value, sensitivity, and specificity score of 98.65%, 97.81%, 98.5%, and 98.2% respectively when evaluated with the ECG signals of the two datasets. For the Cardiac Arrhythmias (ARR), Congestive Heart Failure (CHF), and Normal Sinus Rhythm (NSR) classes present in the MIT-BIH dataset, the proposed model offers an accuracy of 99%, 100%, and 98% respectively and for the classes such as CHF severe and CHF normal in the BIDMC dataset, it offers an accuracy of 98% and 97%. The study mainly demonstrates the effectiveness of the FRCNN technique in predicting arrhythmias and CHF in patients by taking the increased number of features in the ECG signal. It also serves as a promising solution for physicians for long-time surveillance of patients prone to CHF with abnormal heartbeat rhythm.

The relation of residential greenness with fetal growth and the role of air pollution in Eastern Massachusetts, USA

Background and Aim: There is growing evidence that green spaces may be beneficial for health. However, the effects on pregnant individuals and fetuses, including the underlying mechanisms, have been underexplored. We examined the association between neighborhood greenness and fetal growth, and the role of air pollution in these relationships (interaction and mediation) in a Massachusetts-based cohort. Methods: Our study consisted of 9,446 pregnancies that delivered at Beth Israel Deaconess Medical Center, Massachusetts in 2011-2016. Neighborhood greenness was estimated using normalized difference vegetation index (NDVI) with 270 m resolution. Fetal growth parameters included birth weight (BW), and ultrasound measures of biparietal diameter (BPD), head circumference (HC), femur length, and abdominal circumference (AC). Ultrasound parameters were classified into three distinct gestational periods: 16-23 weeks, 24-31 weeks, and 32+ weeks. We used distributed lag models to estimate the time-varying association between monthly NDVI and fetal growth. All models were adjusted for sociodemographic characteristics, and time trends. For models in which we identified an association, we tested for potential interaction with particulate matter ≤2.5 µm (PM2.5) and nitrogen dioxide (NO2) —predicted from well-validated ensemble models— and assessed for mediation by these pollutants. Results: We found that higher NDVI was positively associated with BPD, HC, AC, and BW. For example, a 0.1 NDVI increase sustained up until week 32 was associated with a higher mean AC z-score of 0.03 (95% CI: 0.01, 0.06), and a higher mean BW z-score of 0.02 (95% CI: 0.01, 0.03). We found no evidence for interaction or mediation by air pollution. Conclusions: Residential greenness was associated with larger fetal growth measures. Our effect decomposition suggests that the association of greenness with fetal growth is independent from associations with local air pollution. Keywords: greenness, air pollution, fetal growth, ultrasound, pregnancy, interaction, mediation

Traffic-related air pollution and fetal growth in Eastern Massachusetts, USA

Background and Aim: Previous studies have examined the association between prenatal nitrogen dioxide (NO2) —a traffic emissions tracer— and fetal growth based on ultrasound measures during pregnancy. Yet, most have used exposure assessment methods with low temporal resolution, which limits the identification of critical exposure windows given that pregnancy occurs over a relatively short period. Here, we used NO2 data estimated from a high-resolution spatiotemporal model to examine the association between weekly NO2 exposure during pregnancy and fetal growth in a Massachusetts-based cohort. Methods: We used data from 9,446 singleton births at Beth Israel Deaconess Medical Center in 2011-2016. Fetal growth parameters included birth weight, and ultrasound measures of biparietal diameter (BPD), head circumference (HC), femur length, and abdominal circumference. Ultrasound parameters were classified into three distinct gestational periods: 16-23 weeks, 24-31 weeks, and 32+ weeks. We used distributed lag models to estimate the time-varying association between weekly NO2 exposure and fetal growth. All models were adjusted for sociodemographic characteristics, time trends, particulate matter less than 2.5 microns (PM2.5), and temperature. Results: We found that higher NO2 was negatively associated with all fetal growth parameters. The critical window differed depending on the fetal growth parameter and when it was assessed, with associations particularly strong for head parameters. For example, for BPD and HC measured after week 32, the critical exposure window appeared to be around weeks 15-25; a 10-ppb higher NO2 exposure sustained from conception to the time of measurement was associated with a lower mean z-score of -0.26 (95% CI: -0.39, -0.13) and -0.13 (95% CI: -0.25, 0.00) for BPD and HC, respectively. Conclusions: Higher NO2 was associated with smaller fetal growth measures. We identified several critical exposure windows that varied by parameter and timing of measurement. Keywords: air pollution, traffic, nitrogen dioxide, fetal growth, ultrasound, pregnancy, critical window

Mitral Regurgitation and Mortality Risk in Medicare Beneficiaries With Heart Failure and Preserved Ejection Fraction

The association of mitral regurgitation (MR) severity and mortality in heart failure with preserved ejection fraction (HFpEF) is uncertain. We sought to evaluate the relation between MR severity on transthoracic echocardiography (TTE) and subsequent all-cause mortality in Medicare beneficiaries with HFpEF. We linked 57,608 patients referred for TTE at Beth Israel Deaconess Medical Center to Medicare inpatient claims from 2003 to 2017. In those with a history of HF and a physician-reported left ventricular ejection fraction ≥50%, we evaluated the relation of MR severity and time to the primary end point of all-cause mortality using Kaplan-Meier methods. A total of 7,778 individuals (14.5%) met inclusion criteria (mean age 75.5 years ± 11.9, 55.9% female). Over a median follow-up of 8.1 years, 2,016 (25.9%) died at a median (interquartile range) of 1.7 (0.3 to 4.1) years. At 1 year, 15.8% with 3 to 4+ MR had died versus 10.5% with 0 to 2+ MR (hazard ratio 1.54, 95% confidence interval 1.22 to 1.95, p <0.001). After multivariable adjustment, 3 to 4+ MR continued to be associated with increased all-cause mortality (hazard ratio 1.48, 95% confidence interval 1.14 to 1.94, p=0.004) except in the subset with atrial fibrillation (interaction p=0.03) or recent (<3 months) HF hospitalization (p=0.54). In conclusion, in this large, single-institution retrospective study of Medicare beneficiaries with HFpEF who underwent TTE, moderate-to-severe and severe MR were significantly associated with an increased risk of all-cause mortality after multivariable adjustment, except in those with atrial fibrillation or recent HF. Prospective studies are needed to assess the role of MR reduction in mitigating this risk.

Modeling malnutrition with microfluidics

Environmental enteric dysfunction (EED) is a chronic intestinal syndrome that affects millions of children in low- and middle-income countries. It’s related to malnutrition, but some people don’t get better even with dietary supplements, so researchers want a better understanding of EED’s biological mechanisms. Researchers at Harvard University’s Wyss Institute have now turned their microfluidic organ-on-a-chip technology toward achieving that understanding. By making intestine chips with cells from people with EED, the group was able to simulate key genetic and structural features of the condition ( Nat. Biomed. Eng. 2022, DOI: 10.1038/s41551-022-00899-x ). Cicely W. Fadel, a neonatal and pediatric disease specialist at Beth Israel Deaconess Medical Center, who co-led the project as a clinical fellow at Wyss, presented the research at ACS Fall 2022. The researchers wanted to see how well their chip-based model could mimic features of EED in the lab. They found that chips with cells from EED

33892 Telehealth use at a specialty hidradenitis suppurativa clinic during the COVID-19 pandemic

COVID-19 has increased the patient burden on medical centers, resulting in a greater reliance on the use of telehealth. As there are few hidradenitis suppurativa (HS) specialty clinics in the U.S. (33 as of 2021), the need for telehealth is even greater in the HS patient population. This study analyzed trends in telehealth usage at a specialty HS clinic during the COVID-19 pandemic. Patients with a diagnosis of HS seen by 2 providers who manage predominantly HS clinics at Beth Israel Deaconess Medical Center (BIDMC) between May 2020-March 2021 were included. Correlation between the number of total telehealth visits per month and the 7-day average of confirmed cases in Massachusetts at the first or second day of every month was calculated. Limitations include being a single-site study and a small sample size. Of 541 visits between May 2020-March 2021, 354 (65.4%) were in-person visits and 187 (34.6%) were telehealth. There was a weak negative correlation between the number of telehealth visits and the number of COVID-19 cases in Massachusetts, (Pearson’s correlation coefficient r = -0.214, P = .528) which suggests a minimal relationship between the prevalence of COVID-19 and the use of telehealth. While COVID-19 may have been the impetus for the increase in telehealth use, patients and physicians continue to conduct a high number of telehealth visits despite COVID numbers. This is likely due to the decreased travel burden and the ability to address flares more urgently.

Machine learning nonresponse adjustment of patient-reported opioid consumption data to enable consumption-informed postoperative opioid prescribing guidelines.

Post-discharge opioid consumption is a crucial patient-reported outcome informing opioid prescribing guidelines, but its collection is resource-intensive and vulnerable to inaccuracy due to nonresponse bias. We developed a post-discharge text message-to-web survey system for efficient collection of patient-reported pain outcomes. We prospectively recruited surgical patients at Beth Israel Deaconess Medical Center in Boston, Massachusetts from March 2019 through October 2020, sending an SMS link to a secure web survey to quantify opioids consumed after discharge from hospitalization. Patient factors extracted from the electronic health record were tested for nonresponse bias and observable confounding. Following targeted learning-based nonresponse adjustment, procedure-specific opioid consumption quantiles (medians and 75th percentiles) were estimated and compared to a previous telephone-based reference survey. 6553 patients were included. Opioid consumption was measured in 44% of patients (2868), including 21% (1342) through survey response. Characteristics associated with inability to measure opioid consumption included age, tobacco use, and prescribed opioid dose. Among the 10 most common procedures, median consumption was only 36% of the median prescription size; 64% of prescribed opioids were not consumed. Among those procedures, nonresponse adjustment corrected the median opioid consumption by an average of 37% (IQR: 7, 65%) compared to unadjusted estimates, and corrected the 75th percentile by an average of 5% (IQR: 0, 12%). This brought median estimates for 5/10 procedures closer to telephone survey-based consumption estimates, and 75th percentile estimates for 2/10 procedures closer to telephone survey-based estimates. SMS-recruited online surveying can generate reliable opioid consumption estimates after nonresponse adjustment using patient factors recorded in the electronic health record, protecting patients from the risk of inaccurate prescription guidelines.

Abstract P015: Characteristics And Care Of Inpatients With Orthostatic Hypotension

Background: Orthostatic hypotension (OH) contributes to over 80,000 hospitalizations annually, yet most studies of OH are conducted in outpatient settings. We sought to better characterize the causes, care, and course of OH in hospitalized patients in a large academic center. Methods: In this observational study, we reviewed 101 electronic medical records of adults with OH during admission to Beth Israel Deaconess Medical Center in Boston between January 1, 2017 and June 30, 2021, extracting details related to symptoms, etiology, interventions, blood pressure (BP), and post-hospital follow-up. OH was defined as a drop in systolic or diastolic BP of at least 20 or 10 mm Hg, respectively. Results: Of these patients (39.6% female, 9.9% Black, mean length of stay of 6.8 days), 30.7% had evidence of OH on initial presentation (Table) . While dizziness (32.7%) and lightheadedness (29.7%) were common, 19.8% of patients were asymptomatic. Volume resuscitation (64.4%), medication changes (53.5%), and physical therapy (24.8%) were the most frequent interventions. Among medication changes, reductions in beta blockers (37.0%) and loop diuretics (27.8%) were most common. On average, orthostatic systolic and diastolic BP improved by 7.9 and 3.0 mm Hg, respectively, and OH improved or resolved in 64.3% of patients. Within six months of discharge, 24.8% of patients had a subsequent encounter for OH. Conclusion: This study identifies important foci for further study of inpatient OH, including detection among asymptomatic patients, clinical practices focused on BP augmentation strategies (volume resuscitation, down-titration of beta blockers and loop diuretics), and high recurrence rates.

Association of Artificial Intelligence–Aided Chest Radiograph Interpretation With Reader Performance and Efficiency

The efficient and accurate interpretation of radiologic images is paramount. To evaluate whether a deep learning-based artificial intelligence (AI) engine used concurrently can improve reader performance and efficiency in interpreting chest radiograph abnormalities. This multicenter cohort study was conducted from April to November 2021 and involved radiologists, including attending radiologists, thoracic radiology fellows, and residents, who independently participated in 2 observer performance test sessions. The sessions included a reading session with AI and a session without AI, in a randomized crossover manner with a 4-week washout period in between. The AI produced a heat map and the image-level probability of the presence of the referrable lesion. The data used were collected at 2 quaternary academic hospitals in Boston, Massachusetts: Beth Israel Deaconess Medical Center (The Medical Information Mart for Intensive Care Chest X-Ray [MIMIC-CXR]) and Massachusetts General Hospital (MGH). The ground truths for the labels were created via consensual reading by 2 thoracic radiologists. Each reader documented their findings in a customized report template, in which the 4 target chest radiograph findings and the reader confidence of the presence of each finding was recorded. The time taken for reporting each chest radiograph was also recorded. Sensitivity, specificity, and area under the receiver operating characteristic curve (AUROC) were calculated for each target finding. A total of 6 radiologists (2 attending radiologists, 2 thoracic radiology fellows, and 2 residents) participated in the study. The study involved a total of 497 frontal chest radiographs-247 from the MIMIC-CXR data set (demographic data for patients were not available) and 250 chest radiographs from MGH (mean [SD] age, 63 [16] years; 133 men [53.2%])-from adult patients with and without 4 target findings (pneumonia, nodule, pneumothorax, and pleural effusion). The target findings were found in 351 of 497 chest radiographs. The AI was associated with higher sensitivity for all findings compared with the readers (nodule, 0.816 [95% CI, 0.732-0.882] vs 0.567 [95% CI, 0.524-0.611]; pneumonia, 0.887 [95% CI, 0.834-0.928] vs 0.673 [95% CI, 0.632-0.714]; pleural effusion, 0.872 [95% CI, 0.808-0.921] vs 0.889 [95% CI, 0.862-0.917]; pneumothorax, 0.988 [95% CI, 0.932-1.000] vs 0.792 [95% CI, 0.756-0.827]). AI-aided interpretation was associated with significantly improved reader sensitivities for all target findings, without negative impacts on the specificity. Overall, the AUROCs of readers improved for all 4 target findings, with significant improvements in detection of pneumothorax and nodule. The reporting time with AI was 10% lower than without AI (40.8 vs 36.9 seconds; difference, 3.9 seconds; 95% CI, 2.9-5.2 seconds; P < .001). These findings suggest that AI-aided interpretation was associated with improved reader performance and efficiency for identifying major thoracic findings on a chest radiograph.

Adaptive Support Ventilation and Lung-Protective Ventilation in ARDS.

Adaptive support ventilation (ASV) is a partially closed-loop ventilation mode that adjusts tidal volume (VT) and breathing frequency (f) to minimize mechanical work and driving pressure. ASV is routinely used but has not been widely studied in ARDS. The study was a crossover study with randomization to intervention comparing a pressure-regulated, volume-targeted ventilation mode (adaptive pressure ventilation [APV], standard of care at Beth Israel Deaconess Medical Center) set to VT 6 mL/kg in comparison with ASV mode where VT adjustment is automated. Subjects received standard of care (APV) or ASV and then crossed over to the alternate mode, maintaining consistent minute ventilation with 1-2 h in each mode. The primary outcome was VT corrected for ideal body weight (IBW) before and after crossover. Secondary outcomes included driving pressure, mechanics, gas exchange, mechanical power, and other parameters measured after crossover and longitudinally. Twenty subjects with ARDS were consented, with 17 randomized and completing the study (median PaO2 /FIO2 146.6 [128.3-204.8] mm Hg) and were mostly passive without spontaneous breathing. ASV mode produced marginally larger VT corrected for IBW (6.3 [5.9-7.0] mL/kg IBW vs 6.04 [6.0-6.1] mL/kg IBW, P = .035). Frequency was lower with patients in ASV mode (25 [22-26] breaths/min vs 27 [22-30)] breaths/min, P = .01). In ASV, lower respiratory-system compliance correlated with smaller delivered VT/IBW (R2 = 0.4936, P = .002). Plateau (24.7 [22.6-27.6] cm H2O vs 25.3 [23.5-26.8] cm H2O, P = .14) and driving pressures (12.8 [9.0-15.8] cm H2O vs 11.7 [10.7-15.1] cm H2O, P = .29) were comparable between conventional ventilation and ASV. No adverse events were noted in either ASV or conventional group related to mode of ventilation. ASV targeted similar settings as standard of care consistent with lung-protective ventilation strategies in mostly passive subjects with ARDS. ASV delivered VT based upon respiratory mechanics, with lower VT and mechanical power in subjects with stiffer lungs.