A Multimodality Machine Learning Approach to Differentiate Severe and Nonsevere COVID-19: Model Development and Validation

Abstract

Background: Effectively and efficiently diagnosing patients who have COVID-19 with the accurate clinical type of the disease is essential to achieve optimal outcomes for the patients as well as to reduce the risk of overloading the health care system. Currently, severe and nonsevere COVID-19 types are differentiated by only a few features, which do not comprehensively characterize the complicated pathological, physiological, and immunological responses to SARS-CoV-2 infection in the different disease types. In addition, these type-defining features may not be readily testable at the time of diagnosis.

Objective: In this study, we aimed to use a machine learning approach to understand COVID-19 more comprehensively, accurately differentiate severe and nonsevere COVID-19 clinical types based on multiple medical features, and provide reliable predictions of the clinical type of the disease.

Methods: For this study, we recruited 214 confirmed patients with nonsevere COVID-19 and 148 patients with severe COVID-19. The clinical characteristics (26 features) and laboratory test results (26 features) upon admission were acquired as two input modalities. Exploratory analyses demonstrated that these features differed substantially between two clinical types. Machine learning random forest models based on all the features in each modality as well as on the top 5 features in each modality combined were developed and validated to differentiate COVID-19 clinical types.

Results: Using clinical and laboratory results independently as input, the random forest models achieved >90% and >95% predictive accuracy, respectively. The importance scores of the input features were further evaluated, and the top 5 features from each modality were identified (age, hypertension, cardiovascular disease, gender, and diabetes for the clinical features modality, and dimerized plasmin fragment D, high sensitivity troponin I, absolute neutrophil count, interleukin 6, and lactate dehydrogenase for the laboratory testing modality, in descending order). Using these top 10 multimodal features as the only input instead of all 52 features combined, the random forest model was able to achieve 97% predictive accuracy.

Conclusions: Our findings shed light on how the human body reacts to SARS-CoV-2 infection as a unit and provide insights on effectively evaluating the disease severity of patients with COVID-19 based on more common medical features when gold standard features are not available. We suggest that clinical information can be used as an initial screening tool for self-evaluation and triage, while laboratory test results should be applied when accuracy is the priority.

Keywords: COVID-19; classification; clinical type; decision support; diagnosis; machine learning; multimodality; prediction; reliable.

Figure 1

Comparison of clinical features of patients with nonsevere and severe COVID-19. Note that because these features were binary, the y-axis indicates the prevalence of a positive result. CAR: cardiovascular disease; CHL: chills and shaking; CNC: cancer; CON: contact with patients with COVID-19; COU: coughing; CPD: chronic obstructive pulmonary disease; DIA: diabetes; DIR: diarrhea; FAM: family members with COVID-19; FEV: fever; FTG: fatigue; HED: headache; HIF: high fever; HYP: hypertension; KID: renal disease; LOF: low fever; MOF: medium fever; MSA: muscle ache; MUC: phlegm; NAP: loss of appetite; OLD: older age; PREV: prevalence; SEX: male sex; SHB: chest congestion; SMK: history of smoking; SOR: sore throat; VOM: vomiting.

Figure 2

Forest plot of the importance of clinical features of patients with nonsevere and severe clinical types of COVID-19. Chronic obstructive pulmonary disease is not included because only patients with severe COVID-19 showed it as a comorbidity. The threshold for a feature to be positively or negatively associated with severe COVID-19 was 1 (dashed line), not 0. CAR: cardiovascular disease; CHL: chills and shaking; CNC: cancer; CON: contact with patients with COVID-19; COU: coughing; CPD: chronic obstructive pulmonary disease; DIA: diabetes; DIR: diarrhea; FAM: family members with COVID-19; FEV: fever; FTG: fatigue; HED: headache; HIF: high fever; HYP: hypertension; KID: renal disease; LOF: low fever; MOF: medium fever; MSA: muscle ache; MUC: phlegm; NAP: loss of appetite; OLD: older age; SEX: male sex; SHB: chest congestion; SMK: history of smoking; SOR: sore throat; VOM: vomiting. *P<.05, **P<.01; ***P<.001 from the 2×2 contingency table for each feature.

Figure 3

Comparison of laboratory testing features of patients with nonsevere and severe COVID-19. Values shown on the y-axis were obtained after feature scaling and are between 0 and 1. The error bars represent the standard error of each laboratory testing feature. ALB: albumin; ALT: alanine transaminase; AST: aspartate aminotransferase; BNP: Brain natriuretic peptide; CK: creatine kinase; CREA: creatinine; CRP: C-reactive protein; DD: dimerized plasmin fragment D; ESR: erythrocyte sedimentation rate; FERR: ferritin; HGB: hemoglobin; hsCRP: high-sensitivity C-reactive protein; TNI: troponin I; IL6: interleukin 6; LDH: lactate dehydrogenase; LY: lymphocyte; LY%: percent of lymphocytes; NE: neutrophil; NE% percent of neutrophils; PCT: procalcitonin; PLT: platelet; TBIL: total bilirubin; WBC: white blood cell.

Figure 4

ROC curves from the random forest models based on clinical, laboratory testing, and multimodal features. Left: the symptom feature as the sole input; middle: the laboratory testing feature as the sole input; right: both features combined as the input. AUC: area under the curve; ROC: receiving operator characteristic.