Tutorial on Multiple Mediation Analysis Using Causal Networks: Application to Diagnosing COVID-19 From Its Early and...

Area of Science:

• Causal inference
• Network analysis
• Machine learning

Background:

• Multiple mediation analysis is crucial for understanding complex relationships.
• Traditional methods involve analyzing coefficients in regression equations.
• Network-based approaches offer an alternative for studying mediation.

Purpose of the Study:

• To demonstrate a network-based method for multiple mediation analysis.
• To apply this method to diagnose COVID-19 using symptom data.
• To evaluate the utility of causal networks in medical diagnosis.

Main Methods:

• Constructed causal networks by sequentially regressing variables.
• Utilized LASSO regression to identify direct and indirect effects of symptoms on COVID-19 diagnosis.
• Employed twin modeling to calculate mediated effects by comparing real data to counterfactual models.

Main Results:

• The network model achieved a cross-validated area under the receiver curve of 0.82, indicating moderate to high accuracy.
• Identified that later symptoms, such as chills, mediate the effect of earlier symptoms, like fever.
• Integrated findings from 19 individual regressions into a cohesive network model.

Conclusions:

• Network-based multiple mediation analysis provides novel insights into complex symptom relationships.
• The developed procedure demonstrates the potential of artificial intelligence in triaging COVID-19 patients based on symptoms.
• This approach can aid in preliminary diagnosis before laboratory test results are available.