142 Chapter 7 oxygen would have S/F=0.99. In order to improve the accuracy of measurement of lung oxygenation, we propose limiting the ceiling effect in prospective data by protocolising measurement of SpO2 to control high values or in retrospective (opportunistic) analyses by excluding values recorded with SpO2 above a given threshold value. We first evaluated an optimal threshold using both synthetic and real data from arterial blood gas (ABG) samples, predicting that SpO2≤0.94 provides optimal predictive validity, at a level of induced hypoxia that is broadly acceptable to clinicians. We defined the S/F94 measurement as S/F measured when SaO2≤0.94 or FIO2=0.21. In opportunistic data, S/F94 can be estimated by excluding SpO2 values above 0.94 unless FIO2=0.21. In prospective, protocolised measurements, SaO2≤0.94 can be achieved by reducing FIO2 to a minimum of =0.21 (the fraction of oxygen in ambient air). Since many patients receive oxygen through devices for which FIO2 is not accurately quantified (e.g. Hudson mask, nasal cannulae), prospective studies measuring S/F94 will require a modification of oxygen delivery devices which, in itself, is expected to improve accuracy of measurement (Appendix: Protocol). In order to assess S/F94 as an outcome measure, we first used physiological model to evaluate the relationship with a reference standard, the P/F ratio. Second, we compared the predictive validity of S/F94 with several other measures of pulmonary oxygenation function, including the S/F ratio and the alveolar-arterial difference (Aa). We then used the ISARIC4C dataset to train models for a range of intermediate outcomes, including the WHO ordinal scale and S/F94, as predictors of 28-day mortality. We used these models to estimate sample sizes that would be required to see a given treatment effect. Finally, using data from the RECOVERY trial we estimated the expected improvement in required sample size when using a protocolised, rather than opportunistic, S/F94 measurement. Methods Ethical approval All research described in this study complies with all relevant ethical regulations. Ethical approval was given by the South Central-Oxford C Research Ethics Committee in England (13/SC/0149), the Scotland A Research Ethics Committee (20/SS/0028), and the WHO Ethics Review Committee (RPC571 and RPC572, April 2013). In England and Wales, consent was not required for the collection of depersonalised routine healthcare research data. In Scotland, a waiver for consent was given by the Public Benefit and Privacy Panel.
RkJQdWJsaXNoZXIy MTk4NDMw