Computer-assisted Oxygen Therapy Weaning in Critically Ill Children

Investigators will perform a feasibility study followed by a randomized, controlled study on critically ill neonatal and pediatric patients on oxygen therapy. Investigators will use a model-based adaptive control architecture to regulate FiO2 settings. A major advantage of using an "adaptive" control system is that the response of the patient to previous FiO2 settings are used to recompute the most appropriate FiO2 setting in fixed time intervals. As a result of this iterative and self-correcting process, the FiO2 value converges to the most appropriate value over a period of time. Study Population: Children \<18 years of age admitted to the pediatric intensive care unit requiring oxygen therapy. Study Design: Fifteen patients will be recruited for the feasibility study and 50 patients for the efficacy trial. Sample size was calculated to provide 85% power with a type I error rate of 0.05 to detect a 10% decrease in duration of time in hyperoxic state. Definitions: Hypoxia: oxygen saturation by pulse oximetry (SpO2) \<91% \>1 min Prolonged hypoxia: hypoxia for \>2 min Severe hypoxia: SpO2 \<85% Prolonged severe hypoxia: SpO2 \<85% for \>1 min Hyperoxia: SpO2 \>96% on FiO2 \>0.21 \>1 min Target saturation range: SpO2 91-96% Increased FiO2 requirement: FiO2 \>0.60 Patients will be screened daily for eligibility. Informed consent will be obtained from parents of eligible patients. A randomization table created by the statistician will be used to randomized patients in a 1:1 ratio to either the manual or computer-assisted weaning group. The study will be carried out after the patient has been admitted to the unit for greater than 24 hours.Two hours of respiratory stability without the need for a sustained increase in FiO2 or change in ventilator settings will be required prior to initiation of the study. Target SpO2 range will be 91-96% for pediatric intensive care unit patients. If part of their usual care, patients will receive their scheduled respiratory treatments (e.g. suctioning, chest physical therapy) before the study period starts. During the 4-hour study period, patients in the manual weaning group will be cared for routinely with FiO2 changes made by the nurse or respiratory therapist according to the order placed by the clinician. A second pulse oximeter will be used for monitoring. In the computer-assisted group, a dedicated investigator will be at the bedside during the 4-hour study. Oxygen saturation will be monitored using a second pulse oximeter connected to a programmed computer. The computer will provide clinical decision support by making an FiO2 weaning recommendation every 5-10 min. If more frequent changes occur in the oxygen saturation above or below the target, the computer will make more frequent recommendations (every 30-60 sec) until the saturation is in target range. The investigator will wean the FiO2 based on the computer recommendation, if the suggested change is determined to be safe. If the recommendation is rejected, the investigator will document the reason. No other changes will be made during this time unless patient is hypoxic or hyperoxic. Respiratory support settings (flow, inspiratory pressure, tidal volume, positive end expiratory pressure (PEEP), pressure support) will remain unchanged during the study period, unless escalation of settings is necessary due to changes in clinical status. In case of hypoxia or hyperoxia, a manual adjustment in FiO2 may be made and any clinically necessary interventions (e.g. suctioning, nebulized treatments) will be performed if a recommendation is not made by the computer by 1 min. If an intervention other than an increase in FiO2 is necessary, the investigator may suspend further weaning recommendations until the intervention is complete. Data acquisition may be suspended if artifact is detected. Criteria for suspension of weaning by computer-assistance: * Acute respiratory deterioration \>2 min: oxygen desaturation requiring suctioning or nebulized treatment. During this time, the FiO2 may be adjusted manually. * Acute hemodynamic deterioration \>2 min: hypotension, tachycardia or bradycardia (defined by age). * Acute neurologic deterioration \> 2 min: agitation that interferes with respiratory support and affects oxygen saturation. * Artifact in oxygen saturation reading. Total time study is suspended will be recorded for each patient. Criteria for termination of the study for an individual subject: * Respiratory deterioration persists \>5 min and higher level of support is necessary. * Hemodynamic deterioration persists \>5 min and inotropes or vasopressors are initiated. * Patient develops altered mental status relative to pre-study condition or seizure activity \>2 min. * Parental request for withdrawal from the study. Data Collection: Patients will be assigned a unique subject identification (ID) number. Identifiable data (e.g. medical record number, date of birth) with the corresponding ID will be listed separately from other data and saved on a secure password-protected database (Box.com through Stanford University). During the study data will be collected under the subject ID number (de-identified). All electronic data will be stored in Box and collection sheets will be maintained in a locked locker in the pediatric intensive care unit until information is transferred to electronic form. Demographic data: age, gender, race, height, weight, acute diagnosis, chronic diagnoses, baseline oxygen/respiratory support requirement, date of admission, hours on respiratory support, type of respiratory support, blood gas results if obtained, hospital day. Blood will not be collected specifically for this study, but if a blood gas has been obtained on the patient, it will be recorded. Imaging results will be recorded. During study period: SpO2, respiratory rate, heart rate, blood pressure and temperature every 1 min. Medication administered during and within 1 hour prior to study period. FiO2 changes (computer recommendations and otherwise), respiratory support settings, and acute interventions (e.g. suctioning, breathing treatment, sedation) and reason. Blood gas result if ordered by clinician. Statistical Analysis The primary outcome will be duration of exposure to hyperoxia. Secondary outcomes will be incidence and duration of hypoxia, duration of oxygen saturation in target range, incidence and duration of increased oxygen requirement, and number of manual FiO2 adjustments. The durations of time will be calculated as a proportion of total time recorded. Oxygen saturation and FiO2 will be averaged for patients in the manual group and compared with that of patients in the computer-assisted group using a two-sample, one-sided t-test with margin to test whether the computer-assisted group performs at least 10% better than the manual group. The average number of manual changes to the FiO2 for patients in the manual group and the average number of manual changes in the computer-assisted group will also be compared using a t-test. Based on prior measurements in neonatal patients and after adjusting for potential data dropouts, investigators estimate that a population of 25 patients each for the manual and computer-assisted groups would be sufficient to test the hypothesis. While parametric two-sample, one-sided t-tests were used for this calculation, the results of using non-parametric statistics (based on Wilcoxon rank tests) are not expected to be significantly different. In the investigators' calculations, they used a Type-I error estimate of 0.05 (false positive) and a power of 0.85 (Type-II). In order to monitor the efficacy of the experiment, investigators will also be evaluating the experiment's performance at the half way stage after examining 25 patients. In addition, the principal investigator will also be closely monitoring the results of each experiment to watch for any significant deviations in performance. As this study involves minimal risk, site monitoring or auditing are not part of the plan. Data collection and entry into the registry and monitoring will be done by 2 investigators. If data are missing, the patient may be excluded from statistical analysis if missing data is significant enough to lead to false conclusions.