We have developed and successfully verified a straightforward and expeditious prognostic model for the prompt onset of DVT in individuals suffering from ICH. This model is applicable to patients with ICH without history of DVT and involves 5 patient-related factors, including age, consciousness status, rehabilitation, total muscle strength of the weaker side, and infection. Our prediction model showed better discrimination in the validation cohorts than previous models. Therefore, it has practical implications and can be utilized in regular clinical practice to identify high-risk DVT patients timely and accurately amongst those with ICH.
Similar to previous studies, we found weak limb muscle strength to be a risk factor for development of DVT. It is consistent with the current commonly used medical PADUA score, which includes absolutely bedridden for more than three days as a risk factor for VTE [11]. A similar conclusion is supported by a comprehensive case-control study conducted in the United States which established leg paralysis as an autonomous risk factor for venous thrombosis with an odds ratio of 6.10 [17]. No matter muscle weakness, paralysis, or absolutely bedridden, all can lead to prolonged leg inactivity, thus resulting in venous stasis and thrombosis. It is noted that, compared to immobilization bed rest and paralysis, we used ‘total muscle strength of the worse side’ to further quantify the muscle weakness. As the different scores of the ‘total muscle strength of the worse side’ corresponded to different probability of DVT risk, we can precisely assess the individual risk of DVT in the ICH patients. Besides, we also quantified the status of consciousness according to the extent of consciousness disturbance as it is associated with the degree of immobility. As expected, more severe disturbance of consciousness corresponds to a higher risk of developing DVT based on the nomogram model. Therefore, we can better recognize high-risk individuals and make closer clinical follow-ups and examinations and take preventive measures with priority for them, especially when medical resources are limited.
In addition, we found that patients who received early rehabilitation training had a reduced risk of developing DVT. Our result are consistent with a study which found postoperative rehabilitation training is associated with significantly decreased risk of developing DVT in patients who underwent oblique lumbar inter body fusion operation [18][19]. In our stroke unit, paralysis patients with a grade of muscle strength < 4 of any limb usually received early rehabilitation training. Before that, all of the ICH patients underwent assessment of DVT risk using the Padua prediction score and received basic prevention nursing because all the patients got Padua scores of over 3 due to strict bed rest. For patients with a score ≥ 4, ultrasonic examination for DVT will be initiated as soon as possible and those excluding DVT may receive DVT pump as a mechanic prophylaxis, if possible. The patients receiving DVT pump have been excluded in our study because this treatment has been well recognized as an effective prophylaxis for VTE. As the Padua prediction score is not specifically designed for ICH patients, it cannot distinguish the risk of developing DVT in patients with different degree of neurological deficits. For example, a patient with complete paralysis usually had an increased risk of developing DVT, compared to a patient with no paralysis, yet early rehabilitation training at bedside could significantly decrease the risk. This is because early rehabilitation, including proper limb positioning, bed mobility techniques, and joint range of motion exercises, could improve the blood flow around the affected limbs and reduce the complications associated with immobility. However, using the Padua prediction score, both of them will get a score of 3 because all the patients with ICH should stay in bed strictly for more than three days to avoid the expansion of hemorrhage. As such, neither the extent of paralysis (immobility) nor the effect of clinical intervention will be reflected in assessing the risk of developing DVT. Our model exactly compensates for this defect and the result was more consistent with our clinical practice as we fully considered the effect of treatment.
Therefore, together with old age and infection, two well-recognized risk factors for DVT[20–23], we constructed a specific prediction model for the development of DVT following ICH with significantly improved model performance compared to existing models. Regarding the calibration, the IIDVT model was not well calibrated with significant overestimation, while our model and the LIU model were well calibrated without significant underestimation or overestimation. As for discrimination, our nomogram model performed significantly better than the LIU and the IIDVT models, with an AUROC of 0.794 versus 0.691 and 0.617 in our validation sample. As a complement measure to test the discrimination ability [24, 25], our prediction model exhibited better reclassification ability than the LIU model and IIDVT model, with an IDI of 0.1146 (P < 0.05) and 0.159 (P < 0.05), respectively. Accept for the calibration and discrimination, we also conducted DCA to assess the clinical decision-making ability of the models [26]. As a result, our model displayed a favorable overall advantage for projected probability thresholds spanning from 10–75%.
Undoubtedly, our study has some limitations. Firstly, we only included patients from the neurology department, excluding severe cases who were admitted to intensive care units or the neurosurgery department for surgical treatment. This might lead to an underestimation of outcome events (DVT). Secondly, as a retrospective study, 448 ICH patients were excluded due to the lack of ultrasound examinations, which will inevitably increase the selection bias. Lastly, this is a single-center study, and its conformity with external patients is yet to be validated.