148 Chapter 7 Evaluation in ISARIC4C data 39,765 cases in the ISARIC4C study had SpO2, FIO2 and clinical data available for analysis and met the inclusion criteria (see Methods). Mortality in this population was 20.8% (Table 1). Since measurement of S/F94 was not protocolised in ISARIC4C, measurements were obtained for patients for whom SpO2 happened to be ≤0.94 or who were breathing room air (FIO2 = 0.21), therefore meeting the S/F94 definition. The conceptual advantage of S/F94 over S/F is that it offers a closer relationship to the pathophysiological process of interest. This is not expected to be apparent in the distribution of values observed, but rather in the sensitive detection of a real therapeutic effect. For this reason, and because of the risk of selection bias (see Methods), we did not undertake a direct comparison of patients meeting the criteria for S/F94 measurement, against patients who do not. Instead, we evaluated S/F94 against other commonly used outcome measures. Measure Distribution/ Event rate Estimated treatment effect Total n (β = 80% 2p = 0.05) Opportunistic S/F94 day 5 Mean = 2.39 SD = 1.29 ρ vs Day 0 = 0.31 ΔS/F94: 0.18 1,444 Protocolised S/F94 day 5 Mean = 2.39 SD = 1.25 ρ vs Day 0 = 0.57 ΔS/F94: 0.18 988 WHO day 5 (See Supplementary table 4) OR: 0.84 3,331 1-level sustained improvement 13,437/30,060 (44.7%) RR: 1.03 6,756 2-level sustained improvement 5,411/30,060 (18.0%) RR: 1.04 3,808 28-day mortality 8,262/39,765 RR: 0.85 5,143 Table 1. Comparison of outcome measures among 39,765 hospitalised patients aged 20-75, who required supplemental oxygen in the first 3 days in hospital. The estimated treatment effect is for a relative reduction in mortality. Sample size shows the total number of subjects needed in both arms to detect the estimated treatment effect shown, using a 1:1 allocation. Protocolised S/F94- hypothetical improvement in power using a protocolised measurement of S/F94. ΔS/F94- change in S/F94 associated with a 15% reduction in mortality. RR - risk ratio. OR - proportional odds ratio. In order to select the timepoint of S/F94, several aspects were taken into account. Firstly, we looked at data availability. Within the ISARIC4C dataset, S/F values were available for the largest numbers of patients on days 0, 2, 5 and 8 from study enrolment. Second, among patients who remained in hospital, the distribution of S/F94 values moves over the first few days from study enrolment towards a bimodal pattern
RkJQdWJsaXNoZXIy MTk4NDMw