144 Chapter 7 In order to focus our assessment on the subset of patients with hypoxaemic respiratory failure that is potentially modifiable by anti-inflammatory treatment, we repeated all analyses in subjects aged 20-75 who required supplementary oxygen therapy within 3 days of hospital admission, subjects aged 20-75 that were oxygen dependent on the day of admission, and subjects aged 20-75 without criteria for oxygen dependency. All included patients had SpO2 and FIO2 data available. While SpO2 is typically represented as a percentage, for S/F94 it is used as a fraction, with values ranging from 0-1. Estimation of S/F94 in observational data The S/F ratio was calculated by dividing SpO2 by FIO2 (with both as fractions, taking values between 0 and 1). For this evaluation, S/F94 was defined as an opportunistic measurement in which SpO2≤0.94, or the patient was receiving no supplementary oxygen (FIO2=0.21). Importantly, the retrospectively-defined subgroup of patients meeting the S/F94 criteria is not representative of all patients since there was an excess of patients who were not receiving respiratory support, with slight excess mortality, in the S/F94 group (Supplementary Table 1). This indicates at least two mechanisms of selection bias, acting in opposite directions, and precluding a direct comparison. Firstly, patients who have high blood oxygen levels on relatively little supplementary oxygen are excluded from the S/F94 group; by definition these patients have relatively mild disease. Secondly, the group in whom S/F94 could be measured includes patients who receive supplemental oxygen, and fail to reach adequate SpO2 values, but are not escalated to a higher level of respiratory support; this is a frail and multimorbid population with very severe disease. S/F94 was calculated at baseline (day 0) and on day 5 and day 8 from study enrolment. There is expected to be differential missingness between S/F94 and mortality: SpO2 and FIO2 data are only available for a proportion of cases, whereas outcome data is well-recorded. Patients who died or were discharged on given day and had a missing value for S/F94 were assigned values 0.5 (severe oxygenation defect) and 4.76 (perfect oxygenation), respectively. However, death/discharge was more likely to be recorded than S/F94, and this could introduce bias into our analysis. We addressed this by estimating the proportion of patients for whom S/F94 measurements were available among those who had not died or been sent home by a given day. We then resampled those who died/discharged according to these proportions. For example, if on day 5, 20% of those who had not died or discharged had S/F94 measurements available, we randomly resampled 20% of those who died/ had been discharged by then, assigning S/F94=0.5 to those who died, and S/F94=4.76 to those who were discharged.
RkJQdWJsaXNoZXIy MTk4NDMw