Standard versus individualised positive end-expiratory pressure (PEEP) compared by electrical impedance tomography in neurocritical care: a pilot prospective single centre study.

BACKGROUND: Individualised bedside adjustment of mechanical ventilation is a standard strategy in acute coma neurocritical care patients. This involves customising positive end-expiratory pressure (PEEP), which could improve ventilation homogeneity and arterial oxygenation. This study aimed to determine whether PEEP titrated by electrical impedance tomography (EIT) results in different lung ventilation homogeneity when compared to standard PEEP of 5 cmH2O in mechanically ventilated patients with healthy lungs. METHODS: In this prospective single-centre study, we evaluated 55 acute adult neurocritical care patients starting controlled ventilation with PEEPs close to 5 cmH2O. Next, the optimal PEEP was identified by EIT-guided decremental PEEP titration, probing PEEP levels between 9 and 2 cmH2O and finding the minimal amount of collapse and overdistension. EIT-derived parameters of ventilation homogeneity were evaluated before and after the PEEP titration and after the adjustment of PEEP to its optimal value. Non-EIT-based parameters, such as peripheral capillary Hb saturation (SpO2) and end-tidal pressure of CO2, were recorded hourly and analysed before PEEP titration and after PEEP adjustment. RESULTS: The mean PEEP value before titration was 4.75 ± 0.94 cmH2O (ranging from 3 to max 8 cmH2O), 4.29 ± 1.24 cmH2O after titration and before PEEP adjustment, and 4.26 ± 1.5 cmH2O after PEEP adjustment. No statistically significant differences in ventilation homogeneity were observed due to the adjustment of PEEP found by PEEP titration. We also found non-significant changes in non-EIT-based parameters following the PEEP titration and subsequent PEEP adjustment, except for the mean arterial pressure, which dropped statistically significantly (with a mean difference of 3.2 mmHg, 95% CI 0.45 to 6.0 cmH2O, p < 0.001). CONCLUSION: Adjusting PEEP to values derived from PEEP titration guided by EIT does not provide any significant changes in ventilation homogeneity as assessed by EIT to ventilated patients with healthy lungs, provided the change in PEEP does not exceed three cmH2O. Thus, a reduction in PEEP determined through PEEP titration that is not greater than 3 cmH2O from an initial value of 5 cmH2O is unlikely to affect ventilation homogeneity significantly, which could benefit mechanically ventilated neurocritical care patients.

Recruitment-to-inflation ratio to assess response to PEEP during laparoscopic surgery: A physiologic study.

STUDY OBJECTIVE: During laparoscopic surgery, the role of PEEP to improve outcome is controversial. Mechanistically, PEEP benefits depend on the extent of alveolar recruitment, which prevents ventilator-induced lung injury by reducing lung dynamic strain. The hypotheses of this study were that pneumoperitoneum-induced aeration loss and PEEP-induced recruitment are inter-individually variable, and that the recruitment-to-inflation ratio (R/I) can identify patients who benefit from PEEP in terms of strain reduction. DESIGN: Sequential study. SETTING: Operating room. PATIENTS: Seventeen ASA I-III patients receiving robot-assisted prostatectomy during Trendelenburg pneumoperitoneum. INTERVENTIONS AND MEASUREMENTS: Patients underwent end-expiratory lung volume (EELV) and respiratory/lung/chest wall mechanics (esophageal manometry and inspiratory/expiratory occlusions) assessment at PEEP = 0 cmH2O before and after pneumoperitoneum, at PEEP = 4 and 12 cmH2O during pneumoperitoneum. Pneumoperitoneum-induced derecruitment and PEEP-induced recruitment were assessed through a simplified method based on multiple pressure-volume curve. Dynamic and static strain changes were evaluated. R/I between 12 and 4 cmH2O was assessed from EELV. Inter-individual variability was rated with the ratio of standard deviation to mean (CoV). MAIN RESULTS: Pneumoperitoneum reduced EELV by (median [IqR]) 410 mL [80-770] (p < 0.001) and increased dynamic strain by 0.04 [0.01-0.07] (p < 0.001), with high inter-individual variability (CoV = 70% and 88%, respectively). Compared to PEEP = 4 cmH2O, PEEP = 12 cmH2O yielded variable amount of recruitment (139 mL [96-366] CoV = 101%), causing different extent of dynamic strain reduction (median decrease 0.02 [0.01-0.04], p = 0.002; CoV = 86%) and static strain increases (median increase 0.05 [0.04-0.07], p = 0.01, CoV = 33%). R/I (1.73 [0.58-3.35]) estimated the decrease in dynamic strain (p ≤0.001, r = -0.90) and the increase in static strain (p = 0.009, r = -0.73) induced by PEEP, while PEEP-induced changes in respiratory and lung mechanics did not. CONCLUSIONS: Trendelenburg pneumoperitoneum yields variable derecruitment: PEEP capability to revert these phenomena varies significantly among individuals. High R/I identifies patients in whom higher PEEP mostly reduces dynamic strain with limited static strain increases, potentially allowing individualized settings.

Ventilation during cardiopulmonary resuscitation: A narrative review.

Ventilation during cardiopulmonary resuscitation is vital to achieve optimal oxygenation but continues to be a subject of ongoing debate. This narrative review aims to provide an overview of various components and challenges of ventilation during cardiopulmonary resuscitation, highlighting key areas of uncertainty in the current understanding of ventilation management. It addresses the pulmonary pathophysiology during cardiac arrest, the importance of adequate alveolar ventilation, recommendations concerning the maintenance of airway patency, tidal volumes and ventilation rates in both synchronous and asynchronous ventilation. Additionally, it discusses ventilation adjuncts such as the impedance threshold device, the role of positive end-expiratory pressure ventilation, and passive oxygenation. Finally, this review offers directions for future research.

Recruitment-Potential-Oriented Mechanical Ventilation Protocol and Narrative Review for Patients with Acute Respiratory Distress Syndrome.

Even though much progress has been made to improve clinical outcomes, acute respiratory distress syndrome (ARDS) remains a significant cause of acute respiratory failure. Protective mechanical ventilation is the backbone of supportive care for these patients; however, there are still many unresolved issues in its setting. The primary goal of mechanical ventilation is to improve oxygenation and ventilation. The use of positive pressure, especially positive end-expiratory pressure (PEEP), is mandatory in this approach. However, PEEP is a double-edged sword. How to safely set positive end-inspiratory pressure has long been elusive to clinicians. We hereby propose a pressure-volume curve measurement-based method to assess whether injured lungs are recruitable in order to set an appropriate PEEP. For the most severe form of ARDS, extracorporeal membrane oxygenation (ECMO) is considered as the salvage therapy. However, the high level of medical resources required and associated complications make its use in patients with severe ARDS controversial. Our proposed protocol also attempts to propose how to improve patient outcomes by balancing the possible overuse of resources with minimizing patient harm due to dangerous ventilator settings. A recruitment-potential-oriented evaluation-based protocol can effectively stabilize hypoxemic conditions quickly and screen out truly serious patients.

Comparing physiological impacts of positive pressure ventilation versus self-breathing via a versatile cardiopulmonary model incorporating a novel alveoli opening mechanism.

Mathematical models can be used to generate high-fidelity simulations of the cardiopulmonary system. Such models, when applied to real patients, can provide valuable insights into underlying physiological processes that are hard for clinicians to observe directly. In this work, we propose a novel modelling strategy capable of generating scenario-specific cardiopulmonary simulations to replicate the vital physiological signals clinicians use to determine the state of a patient. This model is composed of a tree-like pulmonary system that features a novel, non-linear alveoli opening strategy, based on the dynamics of balloon inflation, that interacts with the cardiovascular system via the thorax. A baseline simulation of the model is performed to measure the response of the system during spontaneous breathing which is subsequently compared to the same system under mechanical ventilation. To test the new lung opening mechanics and systematic recruitment of alveolar units, a positive end-expiratory pressure (PEEP) test is performed and its results are then compared to simulations of a deep spontaneous breath. The system displays a marked decrease in tidal volume as PEEP increases, replicating a sigmoidal curve relationship between volume and pressure. At high PEEP, cardiovascular function is shown to be visibly impaired, in contrast to the deep breath test where normal function is maintained.

Effects of starting one lung ventilation and applying individualized PEEP right after patients are placed in lateral decubitus position on intraoperative oxygenation for patients undergoing thoracoscopic pulmonary lobectomy: study protocol for a randomized controlled trial.

BACKGROUND: For patients receiving one lung ventilation in thoracic surgery, numerous studies have proved the superiority of lung protective ventilation of low tidal volume combined with recruitment maneuvers (RM) and individualized PEEP. However, RM may lead to overinflation which aggravates lung injury and intrapulmonary shunt. According to CT results, atelectasis usually forms in gravity dependent lung regions, regardless of body position. So, during anesthesia induction in supine position, atelectasis usually forms in the dorsal parts of lungs, however, when patients are turned into lateral decubitus position, collapsed lung tissue in the dorsal parts would reexpand, while atelectasis would slowly reappear in the lower flank of the lung. We hypothesize that applying sufficient PEEP without RM before the formation of atelectasis in the lower flank of the lung may beas effective to prevent atelectasis and thus improve oxygenation as applying PEEP with RM. METHODS: A total of 84 patients scheduled for elective pulmonary lobe resection necessitating one lung ventilation will be recruited and randomized totwo parallel groups. For all patients, one lung ventilation is initiated the right after patients are turned into lateral decubitus position. For patients in the study group, individualized PEEP titration is started the moment one lung ventilation is started, while patients in the control group will receive a recruitment maneuver followed by individualized PEEP titration after initiation of one lung ventilation. The primary endpoint will be oxygenation index measured at T4. Secondary endpoints will include intrapulmonary shunt, respiratory mechanics, PPCs, and hemodynamic indicators. DISCUSSION: Numerous previous studies compared the effects of individualized PEEP applied alone with that applied in combination with RM on oxygenation index, PPCs, intrapulmonary shunt and respiratory mechanics after atelectasis was formed in patients receiving one lung ventilation during thoracoscopic surgery. In this study, we will apply individualized PEEP before the formation of atelectasis while not performing RM in patients allocated to the study group, and then we're going to observe its effects on the aspects mentioned above. The results of this trial will provide a ventilation strategy that may be conductive to improving intraoperative oxygenation and avoiding the detrimental effects of RM for patients receiving one lung ventilation. TRIAL REGISTRATION: www.Chictr.org.cn ChiCTR2400080682. Registered on February 5, 2024.

Effectiveness of High-Flow Nasal Cannula (RAM Cannula) With T-piece Resuscitator to Deliver Continuous Positive Airway Pressure (CPAP) During Neonatal Transport.

Background Newborns frequently experience respiratory distress (RD), necessitating preventive management during transportation. The use of Continuous Positive Airway Pressure (CPAP) is crucial in mitigating RD in neonates, particularly during transit. This study aims to assess the feasibility and efficacy of utilizing a RAM cannula (Neotech Products, Valencia, USA) with a T-piece resuscitator to deliver CPAP during neonatal transport. The objective is to evaluate the response of transported neonates to this intervention, including improvements in distress, surfactant requirements, ventilator dependency, and complications. Method and material Neonates with RD qualifying for CPAP support at birth and requiring transport to Neonatal Intensive Care Unit (NICU) care were included. The average duration of transport was 38 minutes (range 12 minutes to 2 hours). RAM cannula with a T-piece resuscitator was used for CPAP delivery during transportation. Vital parameters and interventions were monitored during transit, and outcomes were compared with inborn neonates receiving standard CPAP in the labor room. Results Out of 48 babies, nine babies required surfactant, and four needed invasive ventilation, with three developing a nasal injury. Compared to in-house preterm babies, these babies had more Positive End Expiratory Pressure (PEEP) knob adjustment, desaturation episodes, late surfactant administration, and intubation needs. Conclusion A high-flow nasal cannula combined with a T-piece resuscitator emerges as a promising modality for CPAP delivery during neonatal transportation, demonstrating efficacy with minimal complications.

In Vivo Measurement of Tidal Volume During Non-invasive Respiratory Support by Continuous-Flow Helmet CPAP.

Recently, the interest in the Helmet interface during non-invasive respiratory support (NIRS) has increased due to the COVID-19 pandemic. During NIRS, positive end-expiratory pressure (PEEP) can be given as continuous positive airway pressure (CPAP), which maintains a positive airway pressure throughout the whole respiratory cycle with Helmet as an interface (H-CPAP). The main disadvantage of the H-CPAP is the inability to measure tidal volume (VT). Opto-electronic plethysmography (OEP) is a non-invasive technique that is not sensitive to gas compression/expansion inside the helmet. OEP acquisitions were performed on 28 healthy volunteers (14 females and 14 males) at baseline and during Helmet CPAP. The effect of posture (semi-recumbent vs. prone), flow (50 vs. 60 L/min), and PEEP (0 vs. 5 vs. 10 cmH2O) on the ventilatory and thoracic-abdominal pattern and the operational volumes were investigated. Prone position limited vital capacity, abdominal expansion and chest wall recruitment. A constant flow of 60 L/min reduced the need for the subject to ventilate while having a slight recruitment effect (100 mL) in the semi-recumbent position. A progressive increasing recruitment was found with higher PEEP but limited by the prone position. It is possible to accurately measure tidal volume during H-CPAP to deliver non-invasive ventilatory support using opto-electronic plethysmography during different clinical settings.

Efficacy and Safety of High PEEP NIV in COVID-19 Patients.

OBJECTIVE: To investigate the efficacy and safety of non-invasive ventilation (NIV) with high PEEP levels application in patients with COVID-19-related acute respiratory distress syndrome (ARDS). METHODS: This is a retrospective cohort study with data collected from 95 patients who were administered NIV as part of their treatment in the COVID-19 intensive care unit (ICU) at University Hospital Centre Zagreb between October 2021 and February 2022. The definite outcome was NIV failure. RESULTS: High PEEP NIV was applied in all 95 patients; 54 (56.84%) patients could be kept solely on NIV, while 41 (43.16%) patients required intubation. ICU mortality of patients solely on NIV was 3.70%, while total ICU mortality was 35.79%. The most significant difference in the dynamic of respiratory parameters between 2 patient groups was visible on Day 3 of ICU stay: By that day, patients kept solely on NIV required significantly lower PEEP levels and had better improvement in PaO2, P/F ratio, and HACOR score. CONCLUSION: High PEEP applied by NIV was a safe option for the initial respiratory treatment of all patients, despite the severity of ARDS. For some patients, it was also shown to be the only necessary form of oxygen supplementation.

The incidence of hypoxemia in dogs recovering from general anesthesia detected with pulse-oximetry and related risk factors.

The postoperative period is critical for the development of complications, including hypoxemia. To detect hypoxemia early and provide appropriate care, continuous monitoring of saturation is necessary: pulse oximetry is an easily accessible and simple method for this purpose. However, a SpO2 cut-off value to detect hypoxemia in dogs recovering from general anesthesia is lacking in the veterinary literature. The objectives of this clinical study are to validate the room air SpO2 test (SpAT), to identify a cut-off value to discriminate hypoxemia (Phase 1), and to apply the SpAT to study the incidence of transient postoperative hypoxemia (TPH) (Phase 2) in dogs with healthy lungs recovering from general anesthesia. Phase 1: 87 dogs recovering from general anesthesia with an arterial line were included. After extubation, SpAT was performed simultaneously with arterial blood sampling. A PaO2 < 80 mmHg was considered hypoxemia. Phase 2: 654 dogs were enrolled. They underwent general anesthesia with different ventilation settings for different procedures. After extubation, dogs were classified as hypoxemic if the SpO2 was lower than the cut-off obtained in phase 1. Phase 1 showed that the SpO2 cut-off is < 95% (sensitivity 100%, specificity 97.4%; area under the curve, AUC = 0.996; 95% Confidence Interval = 0.944-1; P<0.0001). In Phase 2, 169 dogs were hypoxemic. Body Condition Score (BCS) > 3/5, dorsal recumbency, FiO2 1, absence of Positive End-Expiratory Pressure (PEEP) had a significant odds ratio to induce TPH (5.8, 1.9, 3.7, 1.7, respectively). These results showed that SpO2 < 95% indicates PaO2 < 80 mmHg in dogs and TPH occurs in up to 28% of cases. Identification of associated risks could be useful to prevent and to increase awareness for monitoring and treatment.

Is EIT-guided positive end-expiratory pressure titration for optimizing PEEP in ARDS the white elephant in the room? A systematic review with meta-analysis and trial sequential analysis.

Electrical Impedance Tomography (EIT) is a novel real-time lung imaging technology for personalized ventilation adjustments, indicating promising results in animals and humans. The present study aimed to assess its clinical utility for improved ventilation and oxygenation compared to traditional protocols. Comprehensive electronic database screening was done until 30th November, 2023. Randomized controlled trials, controlled clinical trials, comparative cohort studies, and assessments of EIT-guided PEEP titration and conventional methods in adult ARDS patients regarding outcome, ventilatory parameters, and P/F ratio were included. Our search retrieved five controlled cohort studies and two RCTs with 515 patients and overall reduced risk of mortality [RR = 0.68; 95% CI: 0.49 to 0.95; I2 = 0%], better dynamic compliance [MD = 3.46; 95% CI: 1.59 to 5.34; I2 = 0%] with no significant difference in PaO2/FiO2 ratio [MD = 6.5; 95%CI -13.86 to 26.76; I2 = 74%]. The required information size except PaO2/FiO2 was achieved for a power of 95% based on the 50% reduction in risk of mortality, 10% improved compliance as the cumulative Z-score of the said outcomes crossed the alpha spending boundary and did not dip below the inner wedge of futility. EIT-guided individualized PEEP titration is a novel modality; further well-designed studies are needed to substantiate its utility.

Adjustment of positive end-expiratory pressure to body mass index during mechanical ventilation in general anesthesia: BodyVent, a randomized controlled trial.

BACKGROUND: In patients requiring general anesthesia, lung-protective ventilation can prevent postoperative pulmonary complications, which are associated with higher morbidity, mortality, and prolonged hospital stay. Application of positive end-expiratory pressure (PEEP) is one component of lung-protective ventilation. The correct strategy for setting adequate PEEP, however, remains controversial. PEEP settings that lead to a lower pressure difference between end-inspiratory plateau pressure and end-expiratory pressure ("driving pressure," ΔP) may reduce the risk of postoperative pulmonary complications. Preliminary data suggests that the PEEP required to prevent both end-inspiratory overdistension and end-expiratory alveolar collapse, thereby reducing ΔP, correlates positively with the body mass index (BMI) of patients, with PEEP values corresponding to approximately 1/3 of patient's respective BMI. Thus, we hypothesize that adjusting PEEP according to patient BMI reduces ΔP and may result in less postoperative pulmonary complications. METHODS: Patients undergoing general anesthesia and endotracheal intubation with volume-controlled ventilation with a tidal volume of 7 ml per kg predicted body weight will be randomized and assigned to either an intervention group with PEEP adjusted according to BMI or a control group with a standardized PEEP of 5 mbar. Pre- and postoperatively, lung ultrasound will be performed to determine the lung aeration score, and hemodynamic and respiratory vital signs will be recorded for subsequent evaluation. The primary outcome is the difference in ΔP as a surrogate parameter for lung-protective ventilation. Secondary outcomes include change in lung aeration score, intraoperative occurrence of hemodynamic and respiratory events, oxygen requirements and postoperative pulmonary complications. DISCUSSION: The study results will show whether an intraoperative ventilation strategy with PEEP adjustment based on BMI has the potential of reducing the risk for postoperative pulmonary complications as an easy-to-implement intervention that does not require lengthy ventilator maneuvers nor additional equipment. TRIAL REGISTRATION: German Clinical Trials Register (DRKS), DRKS00031336. Registered 21st February 2023. TRIAL STATUS: The study protocol was approved by the ethics committee of the Christian-Albrechts-Universität Kiel, Germany, on 1st February 2023. Recruitment began in March 2023 and is expected to end in September 2023.

Mean airway pressure as a parameter of lung-protective and heart-protective ventilation.

Mean airway pressure (MAP) is the mean pressure generated in the airway during a single breath (inspiration + expiration), and is displayed on most anaesthesia and intensive care ventilators. This parameter, however, is not usually monitored during mechanical ventilation because it is poorly understood and usually only used in research. One of the main determinants of MAP is PEEP. This is because in respiratory cycles with an I:E ratio of 1:2, expiration is twice as long as inspiration. Although MAP can be used as a surrogate for mean alveolar pressure, these parameters differ considerably in some situations. Recently, MAP has been shown to be a useful prognostic factor for respiratory morbidity and mortality in mechanically ventilated patients of various ages. Low MAP has been associated with a lower incidence of 90-day mortality, shorter ICU stay, and shorter mechanical ventilation time. MAP also affects haemodynamics: there is evidence of a causal relationship between high MAP and low perfusion index, both of which are associated with poor prognosis in mechanically ventilated patients. Elevated MAP values have also been associated with high central venous pressure and lactate, which are indicative of ventilator-associated right ventricular failure and tissue hypoperfusion, respectively. MAP, therefore, is an important parameter to measure in clinical practice. The aim of this review has been to identify the determinants of MAP, the pros and cons of using MAP instead of traditional protective ventilation parameters, and the evidence that supports the use of MAP in clinical practice.

The Respiratory Mechanics of COVID-19 Acute Respiratory Distress Syndrome-Lessons Learned?

Acute respiratory distress syndrome (ARDS) is a well-defined clinical entity characterized by the acute onset of diffuse pulmonary injury and hypoxemia not explained by fluid overload. The COVID-19 pandemic brought about an unprecedented volume of patients with ARDS and challenged our understanding and clinical approach to treatment of this clinical syndrome. Unique to COVID-19 ARDS is the disruption and dysregulation of the pulmonary vascular compartment caused by the SARS-CoV-2 virus, which is a significant cause of hypoxemia in these patients. As a result, gas exchange does not necessarily correlate with respiratory system compliance and mechanics in COVID-19 ARDS as it does with other etiologies. The purpose of this review is to relate the mechanics of COVID-19 ARDS to its underlying pathophysiologic mechanisms and outline the lessons we have learned in the management of this clinic syndrome.

Incidence of Weaning Failure in Obese Patients in Intensive Care Unit.

Background The escalating prevalence of obesity worldwide presents unique challenges in critical care management, especially in the context of mechanical ventilation and weaning processes in intensive care units (ICUs). The present study aimed to determine the incidence of weaning failure in obese patients in an ICU. Methods A prospective observational study was carried out to gather data on patients in the ICU of Shifa International Hospital located in Islamabad, Pakistan. The target population consisted of adult patients who were both male and female, ages 18 years and above. These individuals required intubation procedures as well as mechanical ventilation during their hospitalization. The researchers followed these patients prospectively and observed their medical conditions closely to gather data about how obesity might impact critical care interventions and outcomes. Results The sample size was 288 bearing a median age of 61.0 with an interquartile range of 19 years. Older age manifested a significantly higher frequency of failed extubation (p=0.065). Higher body mass index (BMI) was significantly associated with failed extubation among the study population. It was found that a higher significant difference was associated with BMI > 30 kg/m2 (obese) in failed and successful extubation. One-half of the patients with failed extubation and only 16 (5.9%) patients with successful extubation had end-stage renal disease (p<0.001). It was found that patients who underwent failed extubation had notably increased ICU mortality (p=0.108), 28-day mortality (p=0.067), as well as mean ICU (p<0.001) and hospital stay (p=0.007). Conclusion Our study revealed some insightful correlations between obesity, age, comorbidities, length of hospitalization, ICU stay, and mortality rate in terms of weaning failure among the study population.

Effectiveness of Helmet-CPAP in mild to moderate coronavirus type 2 hypoxemia: An observational study.

OBJECTIVE: To determine the relative effectiveness of Helmet-CPAP (H_CPAP) with respect to high-flow nasal cannula oxygen therapy (HFNO) in avoiding greater need for intubation or mortality in a medium complexity hospital in Chile during the year 2021. DESIGN: Cohort analytical study, single center. SETTING: Units other than intensive care units. PATIENTS: Records of adults with mild to moderate hypoxemia due to coronavirus type 2. INTERVENTIONS: None. MAIN VARIABLES OF INTEREST: Need for intubation or mortality. RESULTS: 159 patients were included in the study, with a ratio by support of 2:10 (H_CPAP:HFNO). The 46.5% were women, with no significant differences by sex according to support (p = 0.99, Fisher test). The APACHE II score, for HFNO, had a median of 10.5, 3.5 units higher than H_CPAP (p < 0.01, Wilcoxon rank sum). The risk of intubation in HFNO was 42.1% and in H_CPAP 3.8%, with a significant risk reduction of 91% (95% CI: 36.9%-98.7%; p < 0.01). APACHE II does not modify or confound the support and intubation relationship (p > 0.2, binomial regression); however, it does confound the support and mortality relationship (p = 0.82, RR homogeneity test). Despite a 79.1% reduction in mortality risk with H_CPAP, this reduction was not statistically significant (p = 0.11, binomial regression). CONCLUSIONS: The use of Helmet CPAP, when compared to HFNO, was an effective therapeutic ventilatory support strategy to reduce the risk of intubation in patients with mild to moderate hypoxemia caused by coronavirus type 2 in inpatient units other than intensive care. The limitations associated with the difference in size, age and severity between the arms could generate bias.

Reinforcement learning for intensive care medicine: actionable clinical insights from novel approaches to reward shaping and off-policy model evaluation.

BACKGROUND: Reinforcement learning (RL) holds great promise for intensive care medicine given the abundant availability of data and frequent sequential decision-making. But despite the emergence of promising algorithms, RL driven bedside clinical decision support is still far from reality. Major challenges include trust and safety. To help address these issues, we introduce cross off-policy evaluation and policy restriction and show how detailed policy analysis may increase clinical interpretability. As an example, we apply these in the setting of RL to optimise ventilator settings in intubated covid-19 patients. METHODS: With data from the Dutch ICU Data Warehouse and using an exhaustive hyperparameter grid search, we identified an optimal set of Dueling Double-Deep Q Network RL models. The state space comprised ventilator, medication, and clinical data. The action space focused on positive end-expiratory pressure (peep) and fraction of inspired oxygen (FiO2) concentration. We used gas exchange indices as interim rewards, and mortality and state duration as final rewards. We designed a novel evaluation method called cross off-policy evaluation (OPE) to assess the efficacy of models under varying weightings between the interim and terminal reward components. In addition, we implemented policy restriction to prevent potentially hazardous model actions. We introduce delta-Q to compare physician versus policy action quality and in-depth policy inspection using visualisations. RESULTS: We created trajectories for 1118 intensive care unit (ICU) admissions and trained 69,120 models using 8 model architectures with 128 hyperparameter combinations. For each model, policy restrictions were applied. In the first evaluation step, 17,182/138,240 policies had good performance, but cross-OPE revealed suboptimal performance for 44% of those by varying the reward function used for evaluation. Clinical policy inspection facilitated assessment of action decisions for individual patients, including identification of action space regions that may benefit most from optimisation. CONCLUSION: Cross-OPE can serve as a robust evaluation framework for safe RL model implementation by identifying policies with good generalisability. Policy restriction helps prevent potentially unsafe model recommendations. Finally, the novel delta-Q metric can be used to operationalise RL models in clinical practice. Our findings offer a promising pathway towards application of RL in intensive care medicine and beyond.