Nonlinear Imputation of PaO2/FIO2 From SpO2/FIO2 Among Patients With Acute Respiratory Distress Syndrome

BACKGROUND: ARDS is an important clinical problem. The definition of ARDS requires testing of arterial blood gas to define the ratio of PaO2 to FIO2 (PaO2/FIO2 ratio). However, many patients with ARDS do not undergo blood gas measurement, which may result in underdiagnosis of the condition. As a consequence, a method for estimating PaO2 on the basis of noninvasive measurements is desirable. METHODS: Using data from three ARDS Network studies, we analyzed the enrollment arterial blood gas measurements to compare nonlinear with linear and log-linear imputation methods of estimating PaO2 from percent saturation of hemoglobin with oxygen as measured by pulse oximetry (SpO(2)). We compared mortality on the basis of various measured and imputed PaO2/FIO2 ratio cutoffs to ensure clinical equivalence. RESULTS: We studied 1,184 patients, in 707 of whom the SpO(2) <= 96%. Nonlinear imputation from the SpO(2)/FIO2 ratio resulted in lower error than linear or log-linear imputation (P < .001) for patients with SpO(2) <= 96% but was equivalent to log-linear imputation in all patients. Ninety-day hospital mortality was 26% to 30%, depending on the PaO2/FIO2 ratio, whether nonlinearly imputed or measured. On multivariate regression, the association between imputed and measured PaO2 varied by use of vasopressors and SpO(2). CONCLUSIONS: A nonlinear equation more accurately imputes PaO2/FIO2 from SpO(2)/FIO2 than linear or log-linear equations, with similar observed hospital mortality depending on SpO(2)/FIO2 ratio vs measured PaO2/FIO2 ratios. While further refinement through prospective validation is indicated, a nonlinear imputation appears superior to prior approaches to imputation.