Summary
Acute respiratory failure is a feared complication of Myasthenia Gravis (MG). Many criteria for intubation and mechanical ventilation (MV) exist including clinical judgement and respiratory function parameters, but none are based on high-quality evidence. Thus, it is unclear if there are other predictors to consider. In this study, we aim to identify predictors of MV in patients admitted with MG crisis.
Original Article
Identifying Predictors of Acute Respiratory Failure in Myasthenia Gravis Using Machine Learning
Critical Care Medicine
Prasuna Kamireddi; Nanda Siva, Faraze Niazi, Swarna Rajagopalan
Introduction: Acute respiratory failure is a feared complication of Myasthenia Gravis (MG). Many criteria for intubation and mechanical ventilation (MV) exist including clinical judgement and respiratory function parameters, but none are based on high-quality evidence. Thus, it is unclear if there are other predictors to consider. In this study, we aim to identify predictors of MV in patients admitted with MG crisis.
Methods: We included all MG patients admitted to our hospital from January 1, 2009 to January 1, 2019 that were intubated requiring MV for acute respiratory failure from MG crisis. We also included a consecutive sample of MG patients from the same period that did not require MV. We collected patient data in these two groups, which included, age, sex, race, BMI, age at onset of MG, duration of disease, subtype of MG, Osserman’s grade, presence of bulbar or neck weakness including dysphagia, lowest and mean force vital capacity (FVC) and Negative Inspiratory Force (NIF), acute and chronic treatments, preceding infection, history of respiratory failure and presence of significant cardiac or respiratory comorbidities. These were used as features in a random forest model to predict whether the patient required intubation. Subsequently, we used average decrease in out of bag error and average decrease in Gini index as measures of relative importance for each predictor.
Results: We included 50 MG patients in this study, including 22 who required intubation and MV, and 28 that did not. FVC and NIF, both lowest and mean values had the highest predictive value for patients in MG crisis requiring MV for acute respiratory failure, followed by presence of dysphagia, preceding infection and BMI. The other variables had lower predictive value per our model. The sensitivity and specificity of our predictive model was 81.8% and 89.3% respectively using the above variables
Conclusions: Using machine learning, our model identified that FVC, NIF and bulbar weakness including dysphagia are the strongest predictors for requiring mechanical ventilation during MG crisis. This is a preliminary exploratory study that needs to be validated prospectively with a large sample size.
