212 Chapter 9 be used, such as those discussed in the second part of this thesis, in other settings randomised controlled clinical trials are needed to inform medical decision making and public health policies. In the third part of this thesis, we aim to improve the efficiency of causal inference (and therefore, the efficiency of clinical trials) in infectious disease research, by improving the assessment of the outcome of interest, and by removing noise from heterogeneous diseases. In Chapter 7 an intermediate endpoint for clinical trials in COVID-19 was developed and evaluated: S/F94, a modification of the SpO2/FIO2 (S/F) ratio. SpO2 can be measured continuously and non-invasively using a pulse oximeter, and the ratio of SpO2 to the fraction of inspired oxygen, FIO2, (the S/F ratio) provides information on pulmonary oxygenation function. However, the S/F ratio ceases to reflect pulmonary oxygenation function at high levels of SpO2 because of the ceiling effect in the oxyhaemoglobin saturation curve49. The correlation of the S/F ratio with the reference standard (PaO2/FIO2, P/F ratio) improved when SpO2> 0.94 were excluded. S/F94 is defined as the S/F at which the SpO2 is 0.94 or less, or the FIO2 is = 0.21 (ambient air). First, the relationship of this new measure with the P/F ratio was evaluated. Second, the predictive validity of S/F94 was compared to other measures of pulmonary oxygenation function. Next, using data from over 39.000 hospitalised adult COVID-19 patients from the ISARIC4C data set, the sample size needed to see a specific treatment effect for S/F94 was compared to other commonly used outcome measures: the WHO ordinal scale, sustained improvement on the WHO ordinal scale and 28-day mortality. Finally, data from the RECOVERY trial was utilised to estimate the expected improvement when using a protocolised S/F94 measurement, rather than the opportunistic measurement from retrospective observational data. The expectation was that protocolised measurements would be less noisy due to standardised and steady state measurements, improving the relationship with mortality and decreasing sample size needed. When evaluating the relationship between the S/F ratio and the P/F ratio, excluding values with SpO2 over 94% improved the correlation with the P/F ratio, in both synthetic (Spearman r S/F: 0.40; S/F94: 0.85) and real (Spearman r S/F: 0.82; S/ F94: 0.97) data. A strong inverse relationship between S/F94 and 28-day mortality was found: for every 1 unit increase in S/F94 on day 5 after hospitalisation, the odds of 28-day mortality decrease by 75%. For the sample size calculation, when using a 15% relative reduction in mortality, a 1:1 allocation between groups and 80% power to detect the difference with p=0.05 (two-tailed test), the required sample size was smallest for S/F94 on day 5, needing 722 participants in each arm of the study to detect the predicted reduction in mortality. When using S/F94 on day 8 as the outcome measure, sample size would be 1342 participants in each arm and when using the WHO ordinal scale, each arm would need 1666 participants on day 5 or 1168 participants on day 8. If 1-level sustained improvement at day 28 on the WHO ordinal scale was used, 3378 participants were needed in each arm, and 1904 participants in each arm were necessary if 2-level sustained improvement was used. If 28-day mortality would be used as the outcome measure, 2572 participants would be needed in each arm. Finally, using data from the protocolised measurements that
RkJQdWJsaXNoZXIy MTk4NDMw