Hylke Salverda

92 Chapter 5 SLE6000 respirator (SLE Limited, South Croydon, UK) with OxyGenie automated oxygen titration to a MP70 bedside monitor (Philips, Eindhoven, the Netherlands) or, if no respiratory support was given, SpO2 was measured by a Masimo module on the Philips monitor. From the bedside monitor data is sent to two databases: a Philips Datawarehouse Connect feed to a database in which numerical data is stored once per second for 1 year; and a once per minute feed (HL7 data transfer protocol) which sends the exact value at the set interval time, which may be between 5-60 seconds (in our situation 1-per-minute). The HL7 message is picked up by our patient data management system (PDMS Metavision; IMDsoft, Tel Aviv, Israel). These data are stored for at least 15 years. No filtering, anti-aliasing, averaging or other processing is done on data prior to entry in the database. To prevent synchronization issues caused by systems running on different timeclocks we chose to process the one-per-second data into one-per-minute data: one value per minute was extracted from one-per-second data by taking the value at the change of the minute (i.e. at 0 seconds). For both the one-per-second and one-per-minute data for SpO2 we calculated the average, standard deviation, proportion of time within target range or hypoxia (<80). Within target range was defined as SpO2 between 91%-95% irrespective of FiO2, or 96%-100% when room air was being inspired. For the FiO2 average and oxygen days were calculated. An oxygen day was defined as at least half of the data FiO2 values of that day above 21%. Please note that this may not represent the true oxygen exposure, as the oxygen sensor can have a deviation of 1%. Finally, the number of data points and the difference between the first and last timepoint in each dataset were noted. Data are presented as mean (SD) and median [IQR] with standard tests for normality. Data processing and analyses were done by custom written software in MATLAB (Matlab R2020b; The MathWorks Inc., Natick, Massachusetts, USA). No statistical hypothesis testing was done as we were not testing for a differences between treatments, but examining for comparability. Results There was data available from 92 patients, with a median of 1151774 [577843 - 2586608] one-per-second data points per patient. An excerpt from a data recording is shown in Figure 1. When processed to one-per-minute data, there were 19462 [9129 – 43162] data points left. The time difference between the first and last entry in the