Predicting Acute Brain Injury in Venoarterial Extracorporeal Membrane Oxygenation Patients with Tree-Based Machine Learning: Analysis of the Extracorporeal Life Support Organization Registry.

Objective: To determine if machine learning (ML) can predict acute brain injury (ABI) and identify modifiable risk factors for ABI in venoarterial extracorporeal membrane oxygenation (VA-ECMO) patients. Design: Retrospective cohort study of the Extracorporeal Life Support Organization (ELSO) Registry (2009-2021). Setting: International, multicenter registry study of 676 ECMO centers. Patients: Adults (≥18 years) supported with VA-ECMO or extracorporeal cardiopulmonary resuscitation (ECPR). Interventions: None. Measurements and Main Results: Our primary outcome was ABI: central nervous system (CNS) ischemia, intracranial hemorrhage (ICH), brain death, and seizures. We utilized Random Forest, CatBoost, LightGBM and XGBoost ML algorithms (10-fold leave-one-out cross-validation) to predict and identify features most important for ABI. We extracted 65 total features: demographics, pre-ECMO/on-ECMO laboratory values, and pre-ECMO/on-ECMO settings.Of 35,855 VA-ECMO (non-ECPR) patients (median age=57.8 years, 66% male), 7.7% (n=2,769) experienced ABI. In VA-ECMO (non-ECPR), the area under the receiver-operator characteristics curves (AUC-ROC) to predict ABI, CNS ischemia, and ICH was 0.67, 0.67, and 0.62, respectively. The true positive, true negative, false positive, false negative, positive, and negative predictive values were 33%, 88%, 12%, 67%, 18%, and 94%, respectively for ABI. Longer ECMO duration, higher 24h ECMO pump flow, and higher on-ECMO PaO2 were associated with ABI.Of 10,775 ECPR patients (median age=57.1 years, 68% male), 16.5% (n=1,787) experienced ABI. The AUC-ROC for ABI, CNS ischemia, and ICH was 0.72, 0.73, and 0.69, respectively. The true positive, true negative, false positive, false negative, positive, and negative predictive values were 61%, 70%, 30%, 39%, 29% and 90%, respectively, for ABI. Longer ECMO duration, younger age, and higher 24h ECMO pump flow were associated with ABI. Conclusions: This is the largest study predicting neurological complications on sufficiently powered international ECMO cohorts. Longer ECMO duration and higher 24h pump flow were associated with ABI in both non-ECPR and ECPR VA-ECMO.

Coarse Race and Ethnicity Labels Mask Granular Underdiagnosis Disparities in Deep Learning Models for Chest Radiograph Diagnosis.

See also the editorial by Nikolic in this issue.

An Electronic Algorithm to Identify Vancomycin-Associated Acute Kidney Injury.

Background: The burden of vancomycin-associated acute kidney injury (V-AKI) is unclear because it is not systematically monitored. The objective of this study was to develop and validate an electronic algorithm to identify cases of V-AKI and to determine its incidence. Methods: Adults and children admitted to 1 of 5 health system hospitals from January 2018 to December 2019 who received at least 1 dose of intravenous (IV) vancomycin were included. A subset of charts was reviewed using a V-AKI assessment framework to classify cases as unlikely, possible, or probable events. Based on review, an electronic algorithm was developed and then validated using another subset of charts. Percentage agreement and kappa coefficients were calculated. Sensitivity and specificity were determined at various cutoffs, using chart review as the reference standard. For courses ≥48 hours, the incidence of possible or probable V-AKI events was assessed. Results: The algorithm was developed using 494 cases and validated using 200 cases. The percentage agreement between the electronic algorithm and chart review was 92.5% and the weighted kappa was 0.95. The electronic algorithm was 89.7% sensitive and 98.2% specific in detecting possible or probable V-AKI events. For the 11 073 courses of ≥48 hours of vancomycin among 8963 patients, the incidence of possible or probable V-AKI events was 14.0%; the V-AKI incidence rate was 22.8 per 1000 days of IV vancomycin therapy. Conclusions: An electronic algorithm demonstrated substantial agreement with chart review and had excellent sensitivity and specificity in detecting possible or probable V-AKI events. The electronic algorithm may be useful for informing future interventions to reduce V-AKI.

Utilizing Machine Learning to Predict Neurological Injury in Venovenous Extracorporeal Membrane Oxygenation Patients: An Extracorporeal Life Support Organization Registry Analysis.

Background: Venovenous extracorporeal membrane oxygenation (VV-ECMO) is associated with acute brain injury (ABI), including central nervous system (CNS) ischemia (defined as ischemic stroke or hypoxic-ischemic brain injury) and intracranial hemorrhage (ICH). There is limited data on prediction models for ABI and neurological outcomes in VV-ECMO. Research Question: Can machine learning (ML) accurately predict ABI and identify modifiable factors of ABI in VV-ECMO? Study Design and Methods: We analyzed adult (≥18 years) VV-ECMO patients in the Extracorporeal Life Support Organization Registry (2009-2021) from 676 centers. ABI was defined as CNS ischemia, ICH, brain death, and seizures. Overall, 65 total variables were extracted including clinical characteristics and pre-ECMO and on-ECMO variables. Random Forest, CatBoost, LightGBM, and XGBoost ML algorithms (10-fold leave-one-out cross-validation) were used to predict ABI. Feature Importance Scores were used to pinpoint variables most important for predicting ABI. Results: Of 37,473 VV-ECMO patients (median age=48.1 years, 63% male), 2,644 (7.1%) experienced ABI: 610 (2%) and 1,591 (4%) experienced CNS ischemia and ICH, respectively. The median ECMO duration was 10 days (interquartile range=5-20 days). The area under the receiver-operating characteristics curves to predict ABI, CNS ischemia, and ICH were 0.67, 0.63, and 0.70, respectively. The accuracy, positive predictive, and negative predictive values for ABI were 79%, 15%, and 95%, respectively. ML identified pre-ECMO cardiac arrest as the most important risk factor for ABI while ECMO duration and bridge to transplantation as an indication for ECMO were associated with lower risk of ABI. Interpretation: This is the first study to use machine learning to predict ABI in a large cohort of VV-ECMO patients. Performance was sub-optimal due to the low reported prevalence of ABI with lack of standardization of neuromonitoring/imaging protocols and data granularity in the ELSO Registry. Standardized neurological monitoring and imaging protocols may improve machine learning performance to predict ABI.

Programmatic Adoption and Implementation of Video-Observed Therapy in Minnesota: Prospective Observational Cohort Study.

BACKGROUND: In-person directly observed therapy (DOT) is standard of care for tuberculosis (TB) treatment adherence monitoring in the US, with increasing use of video-DOT (vDOT). In Minneapolis, vDOT became available in 2019. OBJECTIVE: In this paper, we aimed to evaluate the use and effectiveness of vDOT in a program setting, including comparison of verified adherence among those receiving vDOT and in-person DOT. We also sought to understand the impact of COVID-19 on TB treatment adherence and technology adoption. METHODS: We abstracted routinely collected data on individuals receiving therapy for TB in Minneapolis, MN, between September 2019 and June 2021. Our primary outcomes were to assess vDOT use and treatment adherence, defined as the proportion of prescribed doses (7 days per week) verified by observation (in person versus video-DOT), and to compare individuals receiving therapy in the pre-COVID-19 (before March 2020), and post-COVID-19 (after March 2020) periods; within the post-COVID-19 period, we evaluated early COVID-19 (March-August 2020), and intra-COVID-19 (after August 2020) periods. RESULTS: Among 49 patients with TB (mean age 41, SD 19; n=27, 55% female and n=47, 96% non-US born), 18 (36.7%) received treatment during the post-COVID-19 period. Overall, verified adherence (proportion of observed doses) was significantly higher when using vDOT (mean 81%, SD 17.4) compared to in-person DOT (mean 54.5%, SD 10.9; P=.001). The adoption of vDOT increased significantly from 35% (11/31) of patients with TB in the pre-COVID-19 period to 67% (12/18) in the post-COVID-19 period (P=.04). Consequently, overall verified (ie, observed) adherence among all patients with TB in the clinic improved across the study periods (56%, 67%, and 79%, P=.001 for the pre-, early, and intra-COVID-19 periods, respectively). CONCLUSIONS: vDOT use increased after the COVID-19 period, was more effective than in-person DOT at verifying ingestion of prescribed treatment, and led to overall increased verified adherence in the clinic despite the onset of the COVID-19 pandemic.

Structural basis of ion transport and inhibition in ferroportin.

Ferroportin is an iron exporter essential for releasing cellular iron into circulation. Ferroportin is inhibited by a peptide hormone, hepcidin. In humans, mutations in ferroportin lead to ferroportin diseases that are often associated with accumulation of iron in macrophages and symptoms of iron deficiency anemia. Here we present the structures of the ferroportin from the primate Philippine tarsier (TsFpn) in the presence and absence of hepcidin solved by cryo-electron microscopy. TsFpn is composed of two domains resembling a clamshell and the structure defines two metal ion binding sites, one in each domain. Both structures are in an outward-facing conformation, and hepcidin binds between the two domains and reaches one of the ion binding sites. Functional studies show that TsFpn is an electroneutral H+/Fe2+ antiporter so that transport of each Fe2+ is coupled to transport of two H+ in the opposite direction. Perturbing either of the ion binding sites compromises the coupled transport of H+ and Fe2+. These results establish the structural basis of metal ion binding, transport and inhibition in ferroportin and provide a blueprint for targeting ferroportin in pharmacological intervention of ferroportin diseases.