NICU Oxygen Control Study

We intend to enroll between 20-30 subjects at each of two separate institutions (University of Missouri Women's and Children's Hospital and Studer Family Children's Hospital in Pensacola, FL). The subjects will be premature infants \<34 weeks post conceptual age (PCA) requiring respiratory support. This randomized clinical trial will utilize a 24 period 2 treatment crossover design to show that the device performs no worse than (non-inferiority trial) an NICU nurse in keeping a premature neonate's SpO2 within the prescribed range while the infant is on respiratory support. Due to the nature of the study, any masking of the intervention is not possible. Upon enrollment, the subjects will be randomized into enrollment in two groups (A and B). The primary endpoint will be the mean time required to re-establish SpO2 within the prescribed range, as measured from the time an out-of-range alarm is triggered. A secondary endpoint will be proportion of time SpO2 is within the prescribed range, using an area-under-the-curve approach (with a discrete state) to account for varying time-on-test. These outcome measures are complementary because the former doesn't account for the number of alarms, while the latter does. This is important because the oxygen control device operates continuously in a proactive manner, rather than only reacting due alarms, so it is doing more than mimicking the nurse -- the second measure allows us to capture that.

Group A will initially have the automatic device interface with HFNC for 6 hours. The device will have the target SpO2 parameters ordered by the treating physician input into the device. A study laptop will interface with the device, cardiopulmonary monitor, and pulse oximeter to record the data for the study. Sensors will be used to record all adjustments to the device/respiratory support equipment (i.e. blend valve and flow valve used in HFNC). These sensors will continuously record the data for later analysis. The device will constantly evaluate data sent to it from the pulse oximeter and bedside monitor recording all of the data and alarms. In response to alarms, displayed data, doctor's orders, etc., nurses will continue to apply manual inputs to make adjustments to flow and provide tactile stimulus to the subject but not adjust FiO2 unless manual mode is selected. Recorded sensor measurements and manual inputs by the nurse will be used to refine the existing models as well as new models of response in HR, RR, and SpO2 to flow adjustments and tactile stimulus.

After the first 6 hours, the device will be switched to manual mode for the subject (nurse makes all adjustments for FiO2), but the laptop and sensor data logging system will continue to record data from the patient and the respiratory support equipment. This will record the information for the nurse intervention/baseline care part of the study, which continue for 6 hours. During the entire process, the bedside nurse will keep a diary of any events/interventions using the time prominently displayed on the monitoring laptop. This "time-stamped" diary system will allow for easier retrieval of and comparison to the data from the device and monitoring laptop. Also, the monitoring laptop will have a record of all the data, including alarms from the pulse oximeter as well as the bedside monitor to allow for easier retrieval of data related to alarm events and interventions. The laptop will also record any interventions made by the device to allow for easier retrieval of data related to device interventions. The treatment will then alternate periods of each treatment for a total of 6 days (24 6-hour periods).

Group B will have the exact opposite order as group A. Group B infants will initially have the laptop interface with all of the monitors and sensor measurements. However, the nurse intervention/baseline care stage of the study will take place for the first 6 hours. Next, group B will have the device interface with their respiratory equipment, and the data will be recorded as described above for the next 6 hours of the study. The the treatment will alternate every 6 hours for a total of 6 days.

This design was chosen because the premature infants should have fewer events as they grow older each day, and it will help take into account this potential bias. Also, the subjects will be randomized to group A or B in sets of 8 (i.e. in each group of 8 envelopes, 4 will be group A and 4 will be group B). During the entire study process the infants will receive normal NICU care and the parameters for the SpO2 range will be set by the physician caring for the infant. There are also built in manual overrides for the device which allow the NICU to make changes while the subject is on the device phase of the study. The device will be able to record these changes and the staff will record their manual interventions in the study diary.

We have planned our sample size using a non-inferiority test for a cross-over design, based on our primary endpoint, t_delta. For a given patient, define t_delta = (mean elapsed time needed for device to re-establish SpO2 after alarm) - (mean elapsed time needed for nurse to re-establish SpO2 after alarm). The margin of non- inferiority will be chosen as t_delta > -10 sec, so that a device which is no worse than 10 sec, on average, than a nurse will be considered non-inferior. Assuming the standard deviation of t_delta =12 and the true mean difference is zero under the alternate hypothesis, a sample size of 48 achieves 88% with alpha=0.05. If there is 16% patient drop-out before crossover, so that the final n=40, the power drops to 82%. In all analysis, a (paired) t-test will be used. Our secondary endpoint is secondary endpoint is the proportion of time SpO2 is within the prescribed range, using an area-under-the-curve approach (with a discrete state) to account for varying time-on-test. Our secondary endpoint will be analyzed in a similar manner.

We will plan for one interim analysis to determine if the trial should be stopped early due to futility (strong evidence of inferiority, where a confidence interval for t_delta lies entirely to the left of -10 and doesn't intersect -10) or for efficacy (strong evidence of superiority with margin > +20 sec). This will be carried out when n=32 (16 subjects per site) is attained and stopping decisions will be based on O'Brien-Fleming stopping principles. The interim analysis will be carried out by an independent statistician on the University of Missouri's Data Safety and Monitoring Committee, which is also available to monitor the study for adverse events if requested by the IRB. In the event that the patient drop-out is greater than 16% before crossover, then a more complicated estimation procedure will be employed using mixed effects models; otherwise, complete cases will be used.