MI assist in predicting respiratory failure in patients in a pandemic

Acute respiratory failure (ARF) is a frequent medical condition that consumes many healthcare resources and is linked with high morbidity and death. Acute respiratory failure is difficult to classify, and it is frequently defined by the degree of mechanical assistance required or the difference between oxygen supply and absorption. Because of these characteristics, acute respiratory failure represents a spectrum of symptoms rather than a single illness mechanism. Early identification of risk factors for new or worsening acute respiratory failure may help to avoid this process.

Machine Learning

Predictive ML for ARF is defined as any conceivable regression and classification approach learnt from data of any modality (e.g., EMR vitals, laboratory data) that is also automatically used on such data for predicting ARF. As a result, we did not investigate semi-supervised ARF phenotyping clustering.

Requirements for Prediction Modeling

Numerous factors have been created to create and report prediction models, with 31 respiratory, sleep, and critical care journal editors referencing TRIPOD standards as a framework for guidance. The concept of prediction was reinforced, and three essential requirements for successful prediction models were underlined. Some prediction variables may contain causative elements, but not all variables must. Guidelines for ARF and ARDS prediction using EMR data:

1. To assess the “value” of the event of interest, and effective prediction model must employ known variables (predictors). If the outcome is binary, the model must include a classifier function. In different words, there must be a mechanism for labelling the data properly and consistently.

a. The existence or absence of a respiratory support level might be a binary result in the context of ARF (e.g., IMV or no IMV). ARF labels may have documentation issues (respiratory support may be developed first and then retroactively charted). This might result in ARF labelling being delayed.

References

Wong, An-Kwok Ian, et al. “Machine learning methods to predict acute respiratory failure and acute respiratory distress syndrome.” Frontiers in Big Data 3 (2020): 39.