What This Means
This research suggests that blood clots in the deep veins of the legs (deep vein thrombosis, or DVT) are a significant concern for stroke patients undergoing rehabilitation, occurring in about 1 in 4 patients in this study. Researchers examined 240 stroke patients admitted to a rehabilitation unit and found that three factors were independently linked to DVT risk: lower scores on the Barthel Index (a measure of functional independence, where lower scores mean greater disability), having a malignant tumor (which roughly tripled the odds of DVT), and having diabetes (which was associated with lower DVT odds, an unexpected direction that warrants further investigation). These findings highlight that functional status and certain medical conditions may help identify which rehabilitation patients need closer monitoring for DVT.
The study also tested whether computer-based prediction models could improve DVT risk assessment. Among six machine learning approaches compared, a 'random forest' model performed best, correctly distinguishing DVT from non-DVT cases about 80% of the time (AUC of approximately 0.797). To understand what the model was 'thinking,' the researchers used a technique called SHAP analysis, which revealed that patient age, a blood clotting marker called D-dimer, functional independence (Barthel Index), a measure of clotting time (APTT), and how long since the stroke occurred were the most important factors driving predictions. Interestingly, these top machine learning predictors partially overlapped with but were not identical to the factors found by traditional statistical methods.
This research suggests that combining traditional clinical risk factors with machine learning tools could improve the identification of stroke rehabilitation patients at high risk for DVT, potentially enabling earlier and more targeted preventive measures. However, the study was conducted at a single center with a relatively small sample size, and the models would need validation in larger, more diverse populations before being used in routine clinical practice.