|| Checking for direct PDF access through Ovid
To derive a relationship between the SpO2/FIO2 ratio and PaO2/FIO2 ratio across the entire range of SpO2 values (0–100%) and to evaluate whether mortality prediction using the Pediatric Index of Mortality-3 can be improved by the use of PaO2/FIO2 values derived from SpO2/FIO2.Retrospective analysis of prospectively collected data.A regional PICU transport service.Children transported to a PICU.None.The relationship between SpO2/FIO2 and PaO2/FIO2 across the entire range of SpO2 values was first studied using several mathematical models in a derivation cohort (n = 1,235) and then validated in a separate cohort (n = 306). The best SpO2/FIO2-PaO2/FIO2 relationship was chosen according to the ability to detect respiratory failure (PaO2/FIO2 ≤ 200). The discrimination of the original Pediatric Index of Mortality-3 score and a derived Pediatric Index of Mortality-3 score (where SpO2/FIO2-derived PaO2/FIO2 values were used in place of missing PaO2/FIO2 values) were compared in a different cohort (n = 1,205). The best SpO2/FIO2-PaO2/FIO2 relationship in 1,703 SpO2/FIO2-to-PaO2/FIO2 data pairs was a linear regression equation of ln[PF] regressed on ln[SF]. This equation identified children with a PaO2/FIO2 less than or equal to 200 with a specificity of 73% and sensitivity of 61% in children with SpO2 less than 97% (92% and 33%, respectively, when SpO2 ≥ 97%) in the validation cohort. PaO2/FIO2 derived from SpO2/FIO2 (derived PaO2/FIO2) was better at predicting PICU mortality (area under receiver operating characteristic curve, 0.64; 95% CI, 0.55–0.73) compared with the original PaO2/FIO2 (area under receiver operating characteristic curve, 0.54; 95% CI, 0.49–0.59; p = 0.02). However, there was no difference in the original and derived Pediatric Index of Mortality-3 scores and their discriminatory ability for mortality.SpO2-based metrics perform no worse than arterial blood gas–based metrics in mortality prediction models. Future Pediatric Index of Mortality score versions may be improved by the inclusion of risk factors based on oxygen saturation values, especially in settings where PaO2 values are missing in a significant proportion of cases.