Can A.I. help us with hypoxemia?
What’s My Beef?
I have a beef with the Pao2/Fio2 ratio (P/F). The Berlin definition for Acute Respiratory Distress Syndrome (ARDS) includes the P/F ratio as a diagnostic criterion. Most cutoffs for ARDS interventions are based on the P/F ratio. PaO2 requires an arterial blood gas analysis (ABG), an invasive and potentially cost-prohibitive procedure in clinical settings with limited resources. ABG measurement overuse has been recognized for 20 years now, but the problem remains. PaO2 values can vary significantly from one blood-gas draw to another, and given the relative infrequency of checks, this can lead to erroneous conclusions and interventions. Finally, considering the current COVID-19 pandemic, frequent blood-gas checks may increase the risk of infection transmission. Many of these dogma-based processes in the ICU warrant a renewed risk-and-benefit analysis in the post-pandemic scenario.
SaO2 is a continuously available parameter. P/F ratio provides information about the pulmonary gas exchange adjusted for the quantity of oxygen delivered. SaO2/FiO2 (S/F) ratio can be calculated easily and can be considered a non-invasive alternative to P/F. A strong correlation between S/F and P/F has been reported in the literature. Brown et el. found that PaO2/FiO2 ratios could accurately be imputed with SaO2/FiO2 (S/F) ratios through non-linear equations. S/F correlates with P/F for diagnosing ARDS in medical and surgical patients. S/F ratio has the advantage of being easy to calculate, non-invasive, continuous, cost-effective, reliable, potentially lowering the risk of infection exposure, and avoids iatrogenic anemia. If there are so many potential benefits for a transition from S/F to P/F to assess hypoxia, why don’t we use it? Ok, so S/F quantitatively correlates with P/F, but how well does it identify ICU patients with the highest risk of mortality? Let’s ask the Machine!
What Did I Do?
Using a Machine-learning approach, we aimed to estimate the relative predictive capacity of S/F and P/F in measuring ICU mortality.
I used the eICU Collaborative Research Database. The features Age, Gender, SaO2, PaO2, FIO2, Admission Diagnosis, Apache IV, Mechanical Ventilation (MV), and ICU mortality were extracted. Mortality was the dependent variable for our prediction models. Missing data were imputed with Sklearn Iterative Imputer. Random assignment of all the encounters, 80% to the training (n=26690), and 20% to testing (n=6741). We used a gradient boosting decision tree algorithm variant called XGBoost as our model. We used AUC as our primary metric. Feature importance assessed using Permutation Importance. Permutation importance assesses the importance of a feature by randomly shuffling the values and calculating the predictive capacity reduction.
The XGboost model had an AUC of .85. Permutation importance demonstrates that S/F ratio outperforms P/F for mortality predictive ability. The Partial dependence plots illustrate that mortality rises significantly above S/F values of 200.
S/F was a stronger predictor of mortality than P/F based upon our data’s feature importance evaluation. S/F should be considered as the primary method for hypoxia assessment in the ICU.
- Ranieri VM, Rubenfeld GD, Thompson BT, et al. Acute respiratory distress syndrome: the Berlin Definition. Jama. Jun 20 2012;307(23):2526-33. doi:10.1001/jama.2012.5669
- Papazian L, Aubron C, Brochard L, et al. Formal guidelines: management of acute respiratory distress syndrome. Ann Intensive Care. Jun 13 2019;9(1):69. doi:10.1186/s13613-019-0540-9
- Howell MD, Davis AM. Management of ARDS in Adults. Jama. Feb 20 2018;319(7):711-712. doi:10.1001/jama.2018.0307
- Nikolac N, Supak-Smolcić V, Simundić AM, Celap I. Croatian Society of Medical Biochemistry and Laboratory Medicine: national recommendations for venous blood sampling. Biochem Med (Zagreb). 2013;23(3):242-54. doi:10.11613/bm.2013.031
- Merlani P, Garnerin P, Diby M, Ferring M, Ricou B. Quality improvement report: Linking guideline to regular feedback to increase appropriate requests for clinical tests: blood gas analysis in intensive care. BMJ. Sep 2001;323(7313):620-4. doi:10.1136/bmj.323.7313.620
- Walker HK, Hall WD, Hurst JW. Clinical methods : the history, physical, and laboratory examinations. Butterworths; 1990.
- Mallat J, Lazkani A, Lemyze M, et al. Repeatability of blood gas parameters, PCO2 gap, and PCO2 gap to arterial-to-venous oxygen content difference in critically ill adult patients. Medicine (Baltimore). Jan 2015;94(3):e415. doi:10.1097/md.0000000000000415
- Brown SM, Grissom CK, Moss M, et al. Non-linear Imputation of Pao2/Fio2 From Spo2/Fio2 Among Patients With Acute Respiratory Distress Syndrome. Chest. Aug 2016;150(2):307-13. doi:10.1016/j.chest.2016.01.003
- Bashar FR, Vahedian-Azimi A, Farzanegan B, et al. Comparison of non-invasive to invasive oxygenation ratios for diagnosing acute respiratory distress syndrome following coronary artery bypass graft surgery: a prospective derivation-validation cohort study. J Cardiothorac Surg. Nov 27 2018;13(1):123. doi:10.1186/s13019-018-0804-8
- Chen W, Janz DR, Shaver CM, Bernard GR, Bastarache JA, Ware LB. Clinical Characteristics and Outcomes Are Similar in ARDS Diagnosed by Oxygen Saturation/Fio2 Ratio Compared With Pao2/Fio2 Ratio. Chest. Dec 2015;148(6):1477-1483. doi:10.1378/chest.15-0169