Untangling profiles of postthrombotic syndrome using unsupervised machine learning.

Area of Science:

• Vascular Medicine
• Medical Informatics
• Biostatistics

Background:

• Post-thrombotic syndrome (PTS) is a chronic condition following deep vein thrombosis (DVT), characterized by heterogeneity.
• The Villalta scale is used for PTS diagnosis, but its ability to capture patient heterogeneity is unclear.

Purpose of the Study:

• To apply unsupervised machine learning to identify distinct PTS patient profiles.
• To investigate the heterogeneity within the Villalta scale by distinguishing between signs and symptoms.

Main Methods:

• Unsupervised clustering was applied to 818 patients from the IDEAL-DVT study.
• Villalta scale items were clustered to differentiate between signs and symptoms.
• Associations with clinical factors, residual venous obstruction, quality of life, and longitudinal outcomes were analyzed.

Main Results:

• Four distinct patient profiles emerged: younger patients with provoked DVT, women with joint pain, men with isolated popliteal DVT, and older men with diabetes and femoral vein involvement.
• Villalta scale clustering clearly separated signs and symptoms.
• Sign scores correlated with older age, male sex, higher BMI, and DVT extent, while symptom scores correlated with younger age, female sex, higher BMI, and provoked DVT.
• Residual venous obstruction was linked to sign scores, whereas quality of life was more strongly related to symptom scores.
• Sign and symptom scores varied significantly across profiles and over two years, with distinct trajectories for symptom-dominant (profile 1) and sign-dominant (profile 4) groups.

Conclusions:

• PTS exhibits significant patient heterogeneity that can be identified using machine learning.
• The Villalta scale's sign and symptom dimensions are distinct and differentially associated with clinical factors, outcomes, and quality of life.
• Revising the PTS scoring system to differentiate signs and symptoms may enable more personalized risk prediction and prevention strategies for patients with DVT.