Writing Committee

Jan–Paul Roozeman, Guido Mazzinari, Ary Serpa Neto, Markus W. Hollmann, Frederique Paulus, Marcus J. Schultz, and Luigi Pisani

STATISTICAL ANALYSIS PLAN, April 22, 2020

The primary aim of this analysis is to investigate the accuracy of SpO₂/FiO₂ at the first four days of invasive ventilation for prediction of 28–day mortality in invasively ventilated COVID–19 patients. A secondary aim is to determine if PaO₂/FiO₂ can be predicted from SpO₂/FiO₂ in these patients.

The primary endpoint of this analysis is 28–day mortality; secondary endpoints are duration of ventilation, ICU and hospital lengths of stay, ICU and hospital mortality, 90–day mortality and ventilator-free days at day 28.

Data will be expressed as mean ± standard deviation (SD), median with interquartile range (IQR) or number with percentage, where appropriate. Differences in baseline characteristics between survivors and non-survivors will be analyzed using the Pearson Chi-squared or Fisher exact tests for categorical variables and with a one-way ANOVA or Kruskal–Wallis test for continuous variables.

For this analysis, the first and second calendar day that a patient received invasive ventilation are merged and named ‘day 1’––this day, in theory, could last from a minimum of 24 hours to a maximum of 47 hours and 59 minutes. The next days are named ‘day 2’ and ‘day 3’.

The lowest SpO₂/FiO₂ on day 1, 2, and 3 with the corresponding PaO₂/FiO₂ are used to perform the analyses. SpO₂/FiO₂ and PaO₂/FiO₂ in the first hour after the start of invasive ventilation are ignored. A multivariable logistic regression model will be used to analyze the prognostic capacity of SpO₂/FiO₂ on 28–day mortality. For this, SpO₂/FiO₂ is assessed separately as independent variables, assessing the difference in sum of squared errors (partial sum of squares). Statistical significance is analyzed using a joint analysis of variance (ANOVA) and F–test. The model is then fitted, introducing clinically relevant confounders including age, PEEP, duration of prone positioning, arterial lactate level, arterial pH, vasopressor use and the presence of acute kidney injury at admission. The accuracy of the prediction will also be determined by constructing a receiver operator characteristics (ROC) curve; the area under the ROC is calculated and the optimal cut–off value for prediction of 28–day mortality is determined. An area under the ROC of ≥ 0.90 is considered excellent, 0.80 to 0.89 is considered good, 0.70 to 0.79 is considered fair, 0.60 to 0.69 is considered poor, and < 0.60 is considered a fail.¹ A calibration analysis will be used to assess the accuracy of the ROC.

To be able to compare SpO₂/FiO₂ and PaO₂/FiO₂, first the PaO₂ is estimated from SpO₂ using the non–linear formula by Severinghaus–Ellis² and the lowest SpO₂ on day 1. SpO₂ values of ≥ 98% are excluded from this analyses, as the oxyhemoglobin dissociation curve flattens above this level and large changes in the PaO₂may only result in marginal changes in SpO₂.^3-5 Subsequently, the correlation between SpO₂/FiO₂ and PaO₂/FiO₂ is analyzed by comparing the PaO₂/FiO₂ that was estimated from SpO₂ to the measured PaO₂/FiO₂ using a two–way scatterplot and Spearman correlation analysis_. Accuracy is assessed using Bland–Altman plots and a model II regression analysis (Deming regression). Additionally, to assess the predictive value of SpO₂ on PaO₂, an area under the receiver operator characteristics (ROC) curve is calculated at two diagnostic thresholds for PaO₂/FiO₂ (150 and 300 mm Hg).

All analyses are performed in R (version 4.0.3), in the R studio environment [www.rstudio.com]). A P value of < 0.05 is considered statistically significant.

References

1 Safari, S., Baratloo, A., Elfil, M. & Negida, A. Evidence Based Emergency Medicine; Part 5 Receiver Operating Curve and Area under the Curve. Emerg (Tehran) 4, 111-113 (2016).

2 Ellis, R. Determination of PO2 from saturation. J Appl Physiol 67, 902, doi:10.1152/jappl.1989.67.2.902 (1989).

3 Pisani, L. et al. Risk stratification using SpO2/FiO2 and PEEP at initial ARDS diagnosis and after 24 h in patients with moderate or severe ARDS. Ann Intensive Care 7, 108, doi:10.1186/s13613-017-0327-9 (2017).

4 Rice, T. W. et al. Comparison of the SpO2/FIO2 ratio and the PaO2/FIO2 ratio in patients with acute lung injury or ARDS. Chest 132, 410-417, doi:10.1378/chest.07-0617 (2007).

5 Pandharipande, P. P. et al. Derivation and validation of Spo2/Fio2 ratio to impute for Pao2/Fio2 ratio in the respiratory component of the Sequential Organ Failure Assessment score. Crit Care Med 37, 1317-1321, doi:10.1097/CCM.0b013e31819cefa9 (2009).