A Novel Computerized Clinical Decision Support System for Treating Thrombolysis in Patients with Acute Ischemic Stroke

Symptomatic Intracranial Hemorrhage After Endovascular Stroke Treatment: External Validation of Prediction Models

BACKGROUND

Symptomatic intracranial hemorrhage (sICH) is a severe complication of reperfusion therapy for ischemic stroke. Multiple models have been developed to predict sICH or intracranial hemorrhage (ICH) after reperfusion therapy. We provide an overview of published models and validate their ability to predict sICH in patients treated with endovascular treatment in daily clinical practice.

METHODS

We conducted a systematic search to identify models either developed or validated to predict sICH or ICH after reperfusion therapy (intravenous thrombolysis and/or endovascular treatment) for ischemic stroke. Models were externally validated in the MR CLEAN Registry (n=3180; Multicenter Randomized Clinical Trial of Endovascular Treatment for Acute Ischemic Stroke in the Netherlands). The primary outcome was sICH according to the Heidelberg Bleeding Classification. Model performance was evaluated with discrimination (c-statistic, ideally 1; a c-statistic below 0.7 is considered poor in discrimination) and calibration (slope, ideally 1, and intercept, ideally 0).

RESULTS

We included 39 studies describing 40 models. The most frequently used predictors were baseline National Institutes of Health Stroke Scale (NIHSS; n=35), age (n=22), and glucose level (n=22). In the MR CLEAN Registry, sICH occurred in 188/3180 (5.9%) patients. Discrimination ranged from 0.51 (SPAN-100 [Stroke Prognostication Using Age and National Institutes of Health Stroke Scale]) to 0.61 (SITS-SICH [Safe Implementation of Treatments in Stroke Symptomatic Intracerebral Hemorrhage] and STARTING-SICH [STARTING Symptomatic Intracerebral Hemorrhage]). Best calibrated models were IST-3 (intercept, -0.15 [95% CI, -0.01 to -0.31]; slope, 0.80 [95% CI, 0.50-1.09]), SITS-SICH (intercept, 0.15 [95% CI, -0.01 to 0.30]; slope, 0.62 [95% CI, 0.38-0.87]), and STARTING-SICH (intercept, -0.03 [95% CI, -0.19 to 0.12]; slope, 0.56 [95% CI, 0.35-0.76]).

CONCLUSIONS

The investigated models to predict sICH or ICH discriminate poorly between patients with a low and high risk of sICH after endovascular treatment in daily clinical practice and are, therefore, not clinically useful for this patient population.

A new machine learning algorithm with high interpretability for improving the safety and efficiency of thrombolysis for stroke patients: A hospital-based pilot study

Background

Thrombolysis is the first-line treatment for patients with acute ischemic stroke. Previous studies leveraged machine learning to assist neurologists in selecting patients who could benefit the most from thrombolysis. However, when designing the algorithm, most of the previous algorithms traded interpretability for predictive power, making the algorithms hard to be trusted by neurologists and be used in real clinical practice.

Methods

Our proposed algorithm is an advanced version of classical k-nearest neighbors classification algorithm (KNN). We achieved high interpretability by changing the isotropy in feature space of classical KNN. We leveraged a cohort of 189 patients to prove that our algorithm maintains the interpretability of previous models while in the meantime improving the predictive power when compared with the existing algorithms. The predictive powers of models were assessed by area under the receiver operating characteristic curve (AUC).

Results

In terms of interpretability, only onset time, diabetes, and baseline National Institutes of Health Stroke Scale (NIHSS) were statistically significant and their contributions to the final prediction were forced to be proportional to their feature importance values by the rescaling formula we defined. In terms of predictive power, our advanced KNN (AUC 0.88) outperformed the classical KNN (AUC 0.75, p = 0.0192 ).

Conclusions

Our preliminary results show that the advanced KNN achieved high AUC and identified consistent significant clinical features as previous clinical trials/observational studies did. This model shows the potential to assist in thrombolysis patient selection for improving the successful rate of thrombolysis.

The feasibility and accuracy of machine learning in improving safety and efficiency of thrombolysis for patients with stroke: Literature review and proposed improvements

In the treatment of ischemic stroke, timely and efficient recanalization of occluded brain arteries can successfully salvage the ischemic brain. Thrombolysis is the first-line treatment for ischemic stroke. Machine learning models have the potential to select patients who could benefit the most from thrombolysis. In this study, we identified 29 related previous machine learning models, reviewed the models on the accuracy and feasibility, and proposed corresponding improvements. Regarding accuracy, lack of long-term outcome, treatment option consideration, and advanced radiological features were found in many previous studies in terms of model conceptualization. Regarding interpretability, most of the previous models chose restrictive models for high interpretability and did not mention processing time consideration. In the future, model conceptualization could be improved based on comprehensive neurological domain knowledge and feasibility needs to be achieved by elaborate computer science algorithms to increase the interpretability of flexible algorithms and shorten the processing time of the pipeline interpreting medical images.

Cross-Validation of Machine Learning Models for the Functional Outcome Prediction after Post-Stroke Robot-Assisted Rehabilitation

The state of the art is still lacking an extensive analysis of which clinical characteristics are leading to better outcomes after robot-assisted rehabilitation on post-stroke patients. Prognostic machine learning-based models could promote the identification of predictive factors and be exploited as Clinical Decision Support Systems (CDSS). For this reason, the aim of this work was to set the first steps toward the development of a CDSS, by the development of machine learning models for the functional outcome prediction of post-stroke patients after upper-limb robotic rehabilitation. Four different regression algorithms were trained and cross-validated using a nested 5×10-fold cross-validation. The performances of each model on the test set were provided through the Median Average Error (MAE) and interquartile range. Additionally, interpretability analyses were performed, to evaluate the contribution of the features to the prediction. The results on the two best performing models showed a MAE of 13.6 [13.4] and 13.3 [14.8] on the Modified Barthel Index score (MBI). The interpretability analyses highlighted the Fugl-Meyer Assessment, MBI, and age as the most relevant features for the prediction of the outcome. This work showed promising results in terms of outcome prognosis after robot-assisted treatment. Further research should be planned for the development, validation and translation into clinical practice of CDSS in rehabilitation. Clinical relevance- This work establishes the premises for the development of data-driven tools able to support the clinical decision for the selection and optimisation of the robotic rehabilitation treatment.

A risk score for prediction of symptomatic intracerebral haemorrhage following thrombolysis

BACKGROUND AND PURPOSE

Intravenous recombinant tissue plasminogen activator (rt-PA) remains the only FDA approved pharmacological therapy for acute ischemic stroke (AIS), but this treatment is associated with symptomatic intracerebral haemorrhage (SICH). The aim of this study was to derive and validate an accurate measure of SICH risk in ischemic stroke patients treated with rt-PA using data readily available from patient clinical records.

METHODS

Demographics, physiological parameters, and clinical data were obtained from 1,270 ischemic stroke patients treated with thrombolysis at 20 hospitals. This included age, sex, weight, blood pressure, glucose levels, smoking preferences, and presence of previous clinical conditions. Using a bivariate analysis on a training dataset of 890 patients, SICH cases were compared against SICH-free patients and key risk factors associated with SICH were identified. Continuous variables were stratified using k-means clustering, and odds ratios computed for each of the categorical risk factors employed in the risk score. The SICH risk score, which was assessed using an independent validation dataset comprising 380 patients, was defined between 0 and 53, and stratified into 4 categories: very low risk (0-6), low risk (7-12), moderate risk (13-19), and high risk (>20).

RESULTS

Older age (age > 75 years), higher blood pressure, higher severity of stroke, pre-treatment antithrombotic and history of hypertension and hyperlipidaemia, were shown to be significant risk factors for SICH following rt-PA treatment (p < 0.05). A number of interaction effects with age produced greater overall SICH risk than that of individual variables alone, including age*weight, age*NIHSS, age*diastolic blood pressure, and age*hypertension. The SICH prediction tool demonstrated a C-statistic of 0.75 for continuous risk scoring (0-53) and 0.71 for stratified risk levels.

CONCLUSION

A novel, computationally efficient risk score utilising data readily available from patient clinical records was shown to predict SICH risk following thrombolysis treatment with high accuracy. This tool may be useful for pre-screening patients for SICH risk to reduce the morbidity and mortality associated with thrombolysis treatment.