Dataset Information

Learning from past respiratory failure patients to triage COVID-19 patient ventilator needs: A multi-institutional study.

ABSTRACT:

Background

Unlike well-established diseases that base clinical care on randomized trials, past experiences, and training, prognosis in COVID19 relies on a weaker foundation. Knowledge from other respiratory failure diseases may inform clinical decisions in this novel disease. The objective was to predict 48-hour invasive mechanical ventilation (IMV) within 48 h in patients hospitalized with COVID-19 using COVID-like diseases (CLD).

Methods

This retrospective multicenter study trained machine learning (ML) models on patients hospitalized with CLD to predict IMV within 48 h in COVID-19 patients. CLD patients were identified using diagnosis codes for bacterial pneumonia, viral pneumonia, influenza, unspecified pneumonia and acute respiratory distress syndrome (ARDS), 2008-2019. A total of 16 cohorts were constructed, including any combinations of the four diseases plus an exploratory ARDS cohort, to determine the most appropriate cohort to use. Candidate predictors included demographic and clinical parameters that were previously associated with poor COVID-19 outcomes. Model development included the implementation of logistic regression and three ensemble tree-based algorithms: decision tree, AdaBoost, and XGBoost. Models were validated in hospitalized COVID-19 patients at two healthcare systems, March 2020-July 2020. ML models were trained on CLD patients at Stanford Hospital Alliance (SHA). Models were validated on hospitalized COVID-19 patients at both SHA and Intermountain Healthcare.

Results

CLD training data were obtained from SHA (n = 14,030), and validation data included 444 adult COVID-19 hospitalized patients from SHA (n = 185) and Intermountain (n = 259). XGBoost was the top-performing ML model, and among the 16 CLD training cohorts, the best model achieved an area under curve (AUC) of 0.883 in the validation set. In COVID-19 patients, the prediction models exhibited moderate discrimination performance, with the best models achieving an AUC of 0.77 at SHA and 0.65 at Intermountain. The model trained on all pneumonia and influenza cohorts had the best overall performance (SHA: positive predictive value (PPV) 0.29, negative predictive value (NPV) 0.97, positive likelihood ratio (PLR) 10.7; Intermountain: PPV, 0.23, NPV 0.97, PLR 10.3). We identified important factors associated with IMV that are not traditionally considered for respiratory diseases.

Conclusions

The performance of prediction models derived from CLD for 48-hour IMV in patients hospitalized with COVID-19 demonstrate high specificity and can be used as a triage tool at point of care. Novel predictors of IMV identified in COVID-19 are often overlooked in clinical practice. Lessons learned from our approach may assist other research institutes seeking to build artificial intelligence technologies for novel or rare diseases with limited data for training and validation.

SUBMITTER: Carmichael H

PROVIDER: S-EPMC8159260 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:RationalePatients with ventilator-dependent respiratory failure have improved outcomes at centers with greater expertise; yet, most patients are not treated in such facilities. Efforts to align care for respiratory failure and hospital capability would necessarily require interhospital transfer.ObjectivesTo characterize the prevalence and the patient and hospital factors associated with interhospital transfer of adults residing in Florida with ventilator-dependent respiratory failure.MethodsWe performed a retrospective, observational study using Florida Healthcare Cost and Utilization Project data. We selected patients 18 years of age and older with International Classification of Diseases, Ninth Revision, codes of respiratory failure and mechanical ventilation during 2012 and 2013, and we identified cohorts of patients that did and did not undergo interhospital transfer. We obtained patient sociodemographic and clinical variables and categorized hospitals into subtypes on the basis of patient volume and services provided: large, medium (nonprofit or for-profit), and small.ResultsInterhospital transfer was our primary outcome measure. Patient sociodemographics, clinical variables, and hospital types were used as covariates. We identified 2,580 patients with ventilator-dependent respiratory failure who underwent interhospital transfer. Overall, transfer was uncommon, with only 2.9% of patients being transferred. In a hierarchical model, age less than 65 years (odds ratio [OR], 2.09; 95% confidence interval [CI], 1.77-2.45) and tracheostomy (OR, 3.19; 95% CI, 2.80-3.65) were associated with higher odds of transfer, whereas having Medicaid was associated with lower odds of transfer than having commercial insurance (OR, 0.65; 95% CI, 0.56-0.75). Additionally, care in medium-sized for-profit hospitals was associated with lower odds of transfer (OR, 1.37 vs. 2.70) than care in medium nonprofit hospitals, despite similar hospital characteristics.ConclusionsIn Florida, interhospital transfer of patients with ventilator-dependent respiratory failure is uncommon and more likely among younger, commercially insured, medically resource-intensive patients. For-profit hospitals are less likely to transfer than nonprofit hospitals. In future studies, researchers should test for geographic variations and examine the clinical implications of selectivity in interhospital transfer of patients with ventilator-dependent respiratory failure.

Project description:IntroductionWe previously showed in animals that the ratio of inspired tidal volume (Vtinsp) to inspiratory peak electrical activity of the diaphragm (EAdipk) can be used to quantify the respective patient and ventilator breath contributions (PVBCs) during neurally adjusted ventilatory assist (NAVA). The PVBC index has not been tested clinically.MethodsWe studied 12 intubated and mechanically ventilated patients with acute respiratory failure and measured EAdipk, airway (Paw) and inspiratory esophageal pressure (Pes) and Vtinsp. We applied 11 different NAVA levels, increasing them every 3 minutes in steps of 0.3 cm H₂O/μV from 0 to 3.0 cmH₂O/μV. At each NAVA level, one breath was non-assisted (NAVA level 0). PVBC indices were calculated by relating Vtinsp/EAdipk of the non-assisted breath to Vtinsp/EAdipk of the assisted breath(s) using one ((N1)PVBC) or the mean value of five preceding assisted breaths ((X5)PVBC). During assisted breaths, inspiratory changes in Pes (∆Pes) and transpulmonary (ΔPtp) pressures were used to calculate the relative contribution of patient to total inspiratory lung-distending pressures (ΔPes/ΔPtp). Matching of respiratory drive indices and squaring of the PVBC was evaluated for their effect on the correlation between PVBC and ΔPes/ΔPtp. Linear regression analysis and Bland-Altman analysis were applied to compare indices.ResultsUsing an average of five assisted breaths prior to the non-assisted breath and squaring the PVBC ((X5)PVBC(2)) improved determination coefficients (P <0.05), adjusted the regression slope and intercept between PVBC and ΔPes/ΔPtp toward identity (P <0.05) and reduced bias (P <0.05). Matching EAdipk between non-assisted and assisted breaths within the range of 0.77 to 1.30 improved the relationship between (X5)PVBC(2) and ΔPes/ΔPtp (P <0.05) and abolished the need for EAdi normalization in the PVBC calculation (R(2) = 0.96; bias = 0.16 ± 0.06; precision = 0.33 ± 0.08 (mean and 95% confidence interval)).ConclusionsThis clinical study confirms previous experimental results showing that the PVBC(2) predicts the contribution of the inspiratory muscles versus that of the ventilator during NAVA, when differences in effort (EAdi) between non-assisted and assisted breaths are limited. PVBC could help to quantify and standardize the adjustment of the level of assist, and hence reduce the risks of excessive ventilatory assist in patients.Trial registrationClinicalTrials.gov NCT01663480. Registered 9 August 2012.

Project description:Plans for allocating scarce healthcare resources during the COVID-19 pandemic commonly involve the activation of institutional triage teams. These teams would be responsible for selecting patients who are most likely to survive to be prioritized to receive scarce resources. However, there is little empirical support for this approach.DesignHigh-fidelity triage-team simulation study.SettingHealthcare institutions in Washington state.SubjectsTriage teams, consisting of at least two senior clinicians and a bioethicist.InterventionsParticipants reviewed a limited amount of deidentified information for a diverse sample of critically ill patients. Teams then assigned each patient to one of five prioritization categories defined by likelihood of survival to hospital discharge. The process was refined based on observation and participant feedback after which a second phase of simulations was conducted.Measurements and main resultsFeasibility was assessed by the time required for teams to perform their task. Prognostic accuracy was assessed by comparing teams' prediction about likelihood of survival to hospital discharge with real-world discharge outcomes. Agreement between the teams on prognostic categorization was evaluated using kappa statistics. Eleven triage team simulations (eight in phase 1 and three in phase 2) were conducted from December 2020 to February 2021. Overall, teams reviewed a median of 23 patient cases in each session (interquartile range [IQR], 17-29) and spent a median of 102 seconds (IQR, 50-268) per case. The concordance between expected survival and real-world survival to discharge was 71% (IQR, 64-76%). The overall agreement between teams for placement of patients into prognostic categories was moderate (weighted kappa = 0.53).ConclusionsThese findings support the potential feasibility, accuracy, and effectiveness of institutional triage teams informed by a limited set of patient information items as part of a strategy for allocating scarce resources in healthcare emergencies. Additional work is needed to refine the process and adapt it to local contexts.

Project description:PURPOSE:To assess the optimal structure of an introductory curriculum (IC) for radiation oncology residents, including the perceived utility of a 2-day off-site "boot camp," and evaluate the success of a pilot introductory radiation oncology curriculum (IROC) based on these initial data. METHODS AND MATERIALS:In the first phase, anonymous, web-based surveys were sent to US radiation oncology program directors and residents. Likert-type scores (1, not at all; 5, extremely) are reported as the median and interquartile range. Using the phase 1 results, IROC was developed, piloted, and evaluated. RESULTS:Of the 89 program directors and 697 residents, 47 (53%) and 165 (24%) responded, respectively. Of the 89 program directors, 37 (79%) reported offering a formal IC. However, only 83 residents (50%) reported having a formal IC. Program directors reported resident preparation for clinical training as "moderate" (median 3, interquartile range 2-3) on entering residency and "moderate" (median 3, interquartile range 3-4) after IC completion (P = .03). However, residents only believed they were "slightly" prepared (median 2, interquartile range 1-2) on entering residency and "moderately" (median 3, interquartile range 2-3) prepared after IC completion (P < .01). Program directors believed an off-site boot camp would be of "moderate" utility (median 3, interquartile range 3-4) with participation limited by funding (57%). Residents without an IC reported that having an IC would be "quite" beneficial (median 4, interquartile range 3-5). Residents preferred instruction before the clinical training (49%) and over 1 week (40%). Both program directors and residents rated lectures on radiation emergencies and simulation highly. Using these data, IROC was developed and piloted with incoming residents at 4 institutions. After IROC, residents reported improvement in overall preparedness for clinical training (before: median 1, interquartile range 1-2; vs after: median 3, interquartile range 2-3; P < .01) and among specific practice domains. CONCLUSIONS:Beginning radiation oncology residents frequently lack structured introductory curricula but desire instruction before the clinical training with a focus on practical aspects (emergency management, contouring). Program directors recognize the value of both off-site and on-site boot camps. An on-site IC could mitigate funding barriers. A standardized IC, IROC, piloted at 4 programs, showed promising outcomes.