Multivariate modeling to identify patterns in clinical data: the example of chest pain

Area of Science:

• Cardiology
• Primary Care Medicine
• Diagnostic Studies

Background:

• Chest pain diagnosis presents challenges due to symptom interrelationships and diagnostic category ambiguity.
• Physicians need to differentiate between various causes of chest pain, impacting patient classification and management.

Purpose of the Study:

• To identify natural patient groupings based on risk factors, history, and clinical examination for chest pain.
• To validate these identified groups against final patient diagnoses.

Main Methods:

• A cross-sectional diagnostic study involving 1199 patients aged over 35 presenting with chest pain across 74 primary care practices.
• Utilized multiple correspondence analysis (MCA) for variable-level associations and multidimensional scaling (MDS), k-means, and fuzzy cluster analyses for patient-level subgroup discovery.
• Employed heatmaps for visual representation of analysis results.

Main Results:

• MCA identified six key factors: "chest wall syndrome", "vital threat", "stomach and bowel pain", "angina pectoris", "chest infection syndrome", and "self-limiting chest pain".
• Patient-level cluster analyses (MDS, k-means, fuzzy clustering) failed to identify distinct, interpretable patient subgroups.
• The statistical quality criteria for the resulting cluster solutions were insufficient.

Conclusions:

• Chest pain represents a clinically heterogeneous category.
• There is a lack of coherent associations between signs and symptoms at the patient level, hindering clear subgroup identification.