Predicting Outcome of Endovascular Treatment for Acute Ischemic Stroke: Potential Value of Machine Learning Algorithms

Introduction: Stroke is a common cause of physical disability. Women generally suffer from more severe strokes, have poorer stroke outcomes, and higher mortality than that of men. Cytokines play an important role in post-stroke inflammation. Prior studies have examined differences in individual cytokine levels in patients with acute ischemic stroke (AIS), but comprehensive cytokine expression profiling across different sex and clinical characteristics are lacking. Hypothesis: Stroke is a sexually dimorphic disease with well-known sex differences in immune cell prevalence, cytokine expression, and outcome. A comprehensive cytokine and immune cell network may help identify sex-specific immune response and further provide guidance for stroke research in females. Methods: Patients with AIS were recruited from 2011-2015 at a Comprehensive Stroke Center. Multiplex analysis (Luminex 200 IS) was used to measure serum levels of 30 common cytokines. Data were analyzed with SPSS 26.0 (IBM) and machine learning algorithms. Spearman’s correlation, Mann-Whitney U test, and two-way ANOVA analyses were used to determine the relationships among the variables. The network between cytokines and immune cell types was predicted by CIBERSORT and modified ssGSEA in R package. Results: We examined sex differences in serum cytokine profiles on stroke severity and immune cells profiles using 144 patients with AIS. Among 30 cytokines, IFN-A2, IFNγ, IL-1RA, IL-6, IL-8, IP-10, RANTES, TNFα, and VEGF were found to have statistically significant differences between male and female. Additionally, female survivors with higher admission NIHSS exhibited higher levels of IFN-A2, IFNγ, IL-6, and IL-8 (F=2.722, p=.011; F=2.245, p=.034; F=7.626, p<.001; F=4.599, p<.001, respectively). A cytokine-immune cell network was created using computer algorithms resulting in identification of an upregulation of Th22 in the female. Sex-specific expression of Th22 cells was then validated in human PBMC. Conclusion: Our study suggests sex is an important factor which determines clinical outcome. Reducing Th22 may improve stroke recovery in females. Analyzing clinical data using machine learning algorithms can identify prognostic indicators of stroke.

Download Full-text

Machine learning algorithms vs. thresholding to segment ischemic regions in patients with acute ischemic stroke

IEEE Journal of Biomedical and Health Informatics ◽

10.1109/jbhi.2021.3097591 ◽

2021 ◽

pp. 1-1

Author(s):

Luca Tomasetti ◽

Liv Jorunn Hllesli ◽

Kjersti Engan ◽

Kathinka Dhli Kurz ◽

Martin Wilhelm Kurz ◽

...

Keyword(s):

Machine Learning ◽

Ischemic Stroke ◽

Acute Ischemic Stroke ◽

Learning Algorithms ◽

Machine Learning Algorithms

Download Full-text

Comparative analysis, applications, and interpretation of electronic health record-based stroke phenotyping methods

10.21203/rs.3.rs-16812/v1 ◽

2020 ◽

Author(s):

Phyllis Thangaraj ◽

Benjamin R Kummer ◽

Tal Lorberbaum ◽

Mitchell S.V. Elkind ◽

Nicholas P Tatonetti

Keyword(s):

Machine Learning ◽

Ischemic Stroke ◽

Acute Ischemic Stroke ◽

Learning Algorithms ◽

Case Control ◽

Machine Learning Algorithms ◽

Care Hospital ◽

Diagnosis Codes ◽

Wide Range ◽

Electronic Health

Abstract Background: Accurate identification of acute ischemic stroke (AIS) patient cohorts is essential for a wide range of clinical investigations. Automated phenotyping methods that leverage electronic health records (EHRs) represent a fundamentally new approach cohort identification without current laborious and ungeneralizable generation of phenotyping algorithms. We systematically compared and evaluated the ability of machine learning algorithms and case-control combinations to phenotype acute ischemic stroke patients using data from an EHR.Materials and Methods: Using structured patient data from the EHR at a tertiary-care hospital system, we built and evaluated machine learning models to identify patients with AIS based on 75 different case-control and classifier combinations. We then estimated the prevalence of AIS patients across the EHR. Finally, we externally validated the ability of the models to detect AIS patients without AIS diagnosis codes using the UK Biobank.Results: Across all models, we found that the mean AUROC for detecting AIS was 0.963±0.0520 and average precision score 0.790±0.196 with minimal feature processing. Classifiers trained with cases with AIS diagnosis codes and controls with no cerebrovascular disease codes had the best average F1 score (0.832±0.0383). In the external validation, we found that the top probabilities from a model-predicted AIS cohort were significantly enriched for AIS patients without AIS diagnosis codes (60-150 fold over expected). Conclusions: Our findings support machine learning algorithms as a generalizable way to accurately identify AIS patients without using process-intensive manual feature curation. When a set of AIS patients is unavailable, diagnosis codes may be used to train classifier models.

Download Full-text

Comparative analysis, applications, and interpretation of electronic health record-based stroke phenotyping methods

10.1101/565671 ◽

2019 ◽

Cited By ~ 1

Author(s):

Phyllis M. Thangaraj ◽

Benjamin R. Kummer ◽

Tal Lorberbaum ◽

Mitchell V. S. Elkind ◽

Nicholas P. Tatonetti

Keyword(s):

Machine Learning ◽

Ischemic Stroke ◽

Acute Ischemic Stroke ◽

Learning Algorithms ◽

External Validation ◽

Case Control ◽

Machine Learning Algorithms ◽

Receiver Operating Curve ◽

Diagnosis Codes ◽

Electronic Health

Background and PurposeAccurate identification of acute ischemic stroke (AIS) patient cohorts is essential for a wide range of clinical investigations. Automated phenotyping methods that leverage electronic health records (EHRs) represent a fundamentally new approach cohort identification. Unfortunately, the current generation of these algorithms is laborious to develop, poorly generalize between institutions, and rely on incomplete information. We systematically compared and evaluated the ability of several machine learning algorithms and case-control combinations to phenotype acute ischemic stroke patients using data from an EHR.MethodsUsing structured patient data from the EHR at a tertiary-care hospital system, we built machine learning models to identify patients with AIS based on 75 different case-control and classifier combinations. We then determined the models’ classification ability for AIS on an internal validation set, and estimated the prevalence of AIS patients across the EHR. Finally, we externally validated the ability of the models to detect self-reported AIS patients without AIS diagnosis codes using the UK Biobank.ResultsAcross all models, we found that the mean area under the receiver operating curve for detecting AIS was 0.963±0.0520 and average precision score 0.790±0.196 with minimal feature processing. Logistic regression classifiers with L1 penalty gave the best performance. Classifiers trained with cases with AIS diagnosis codes and controls with no cerebrovascular disease diagnosis codes had the best average F1 score (0.832±0.0383). In the external validation, we found that the top probabilities from a model-predicted AIS cohort were significantly enriched for self-reported AIS patients without AIS diagnosis codes (65-250 fold over expected).ConclusionsOur findings support machine learning algorithms as a way to accurately identify AIS patients without relying on diagnosis codes or using process-intensive manual feature curation. When a set of AIS patients is unavailable, diagnosis codes may be used to train classifier models. Our approach is potentially generalizable to other academic institutions and further external validation is needed.

Download Full-text

Comparative analysis, applications, and interpretation of electronic health record-based stroke phenotyping methods

10.21203/rs.3.rs-16812/v3 ◽

2020 ◽

Author(s):

Phyllis Thangaraj ◽

Benjamin R Kummer ◽

Tal Lorberbaum ◽

Mitchell S.V. Elkind ◽

Nicholas P Tatonetti

Keyword(s):

Machine Learning ◽

Ischemic Stroke ◽

Acute Ischemic Stroke ◽

Learning Algorithms ◽

Case Control ◽

Machine Learning Algorithms ◽

Care Hospital ◽

Diagnosis Codes ◽

Wide Range ◽

Electronic Health

Abstract Background: Accurate identification of acute ischemic stroke (AIS) patient cohorts is essential for a wide range of clinical investigations. Automated phenotyping methods that leverage electronic health records (EHRs) represent a fundamentally new approach cohort identification without current laborious and ungeneralizable generation of phenotyping algorithms. We systematically compared and evaluated the ability of machine learning algorithms and case-control combinations to phenotype acute ischemic stroke patients using data from an EHR.Materials and Methods: Using structured patient data from the EHR at a tertiary-care hospital system, we built and evaluated machine learning models to identify patients with AIS based on 75 different case-control and classifier combinations. We then estimated the prevalence of AIS patients across the EHR. Finally, we externally validated the ability of the models to detect AIS patients without AIS diagnosis codes using the UK Biobank.Results: Across all models, we found that the mean AUROC for detecting AIS was 0.963±0.0520 and average precision score 0.790±0.196 with minimal feature processing. Classifiers trained with cases with AIS diagnosis codes and controls with no cerebrovascular disease codes had the best average F1 score (0.832±0.0383). In the external validation, we found that the top probabilities from a model-predicted AIS cohort were significantly enriched for AIS patients without AIS diagnosis codes (60-150 fold over expected). Conclusions: Our findings support machine learning algorithms as a generalizable way to accurately identify AIS patients without using process-intensive manual feature curation. When a set of AIS patients is unavailable, diagnosis codes may be used to train classifier models.

Download Full-text

Comparative analysis, applications, and interpretation of electronic health record-based stroke phenotyping methods

BioData Mining ◽

10.1186/s13040-020-00230-x ◽

2020 ◽

Vol 13 (1) ◽

Author(s):

Phyllis M. Thangaraj ◽

Benjamin R. Kummer ◽

Tal Lorberbaum ◽

Mitchell S. V. Elkind ◽

Nicholas P. Tatonetti

Keyword(s):

Machine Learning ◽

Ischemic Stroke ◽

Acute Ischemic Stroke ◽

Learning Algorithms ◽

Case Control ◽

Machine Learning Algorithms ◽

Care Hospital ◽

Diagnosis Codes ◽

Wide Range ◽

Electronic Health

Abstract Background Accurate identification of acute ischemic stroke (AIS) patient cohorts is essential for a wide range of clinical investigations. Automated phenotyping methods that leverage electronic health records (EHRs) represent a fundamentally new approach cohort identification without current laborious and ungeneralizable generation of phenotyping algorithms. We systematically compared and evaluated the ability of machine learning algorithms and case-control combinations to phenotype acute ischemic stroke patients using data from an EHR. Materials and methods Using structured patient data from the EHR at a tertiary-care hospital system, we built and evaluated machine learning models to identify patients with AIS based on 75 different case-control and classifier combinations. We then estimated the prevalence of AIS patients across the EHR. Finally, we externally validated the ability of the models to detect AIS patients without AIS diagnosis codes using the UK Biobank. Results Across all models, we found that the mean AUROC for detecting AIS was 0.963 ± 0.0520 and average precision score 0.790 ± 0.196 with minimal feature processing. Classifiers trained with cases with AIS diagnosis codes and controls with no cerebrovascular disease codes had the best average F1 score (0.832 ± 0.0383). In the external validation, we found that the top probabilities from a model-predicted AIS cohort were significantly enriched for AIS patients without AIS diagnosis codes (60–150 fold over expected). Conclusions Our findings support machine learning algorithms as a generalizable way to accurately identify AIS patients without using process-intensive manual feature curation. When a set of AIS patients is unavailable, diagnosis codes may be used to train classifier models.

Download Full-text

Abstract 1122‐000106: Comparing Machine Learning Algorithms and Regression Models for Predicting Functional Outcome in the STRATIS Registry

Stroke: Vascular and Interventional Neurology ◽

10.1161/svin.01.suppl_1.000106 ◽

2021 ◽

Vol 1 (S1) ◽

Author(s):

Mouhammad A Jumaa ◽

Zeinab Zoghi ◽

Syed Zaidi ◽

Nils Mueller‐Kronast ◽

Osama Zaidat ◽

...

Keyword(s):

Machine Learning ◽

Logistic Regression ◽

Ischemic Stroke ◽

Acute Ischemic Stroke ◽

Functional Outcome ◽

Learning Algorithms ◽

Learning Model ◽

Machine Learning Algorithms ◽

Systematic Evaluation ◽

Machine Learning Model

Introduction : Machine learning algorithms have emerged as powerful predictive tools in the field of acute ischemic stroke. Here, we examine the predictive performance of a machine algorithm compared to logistic regression for predicting functional outcomes in the prospective Systematic Evaluation of Patients Treated With Neurothrombectomy Devices for Acute Ischemic Stroke (STRATIS) Registry. Methods : The STRATIS Registry was a prospective, observational study of the use of the Solitaire device in acute ischemic stroke patients. Patients with posterior circulation stroke or missing 90‐day mRS were excluding from the analysis. A statistical algorithm (logistic regression) and a machine learning algorithm (decision tree) were implemented on the preprocessed dataset using 10‐fold cross‐validation method where 80% of the data were fed into the models to be trained and the remaining 20% were utilized in the test phase to evaluate the performance of the models for prediction of 90‐day mRS score as dichotomous output. Results : Of the 938 STRATIS patients, 702 with 90‐day mRS were included. The machine learning model outperformed the logistic regression model with a 0.92±0.026 Area Under Curve (AUC) score compared to a 0.88±0.028 AUC score obtained by implementing logistic regression. Conclusions : Our machine learning model delivered improved performance in comparison with the statistical model in predicting 90‐day functional outcome. More studies are needed to understand and externally validate the predictive capacity of our machine learning model.

Download Full-text

Comparative analysis, applications, and interpretation of electronic health record-based stroke phenotyping methods

10.21203/rs.3.rs-16812/v4 ◽

2020 ◽

Author(s):

Phyllis Thangaraj ◽

Benjamin R Kummer ◽

Tal Lorberbaum ◽

Mitchell S.V. Elkind ◽

Nicholas P Tatonetti

Keyword(s):

Machine Learning ◽

Ischemic Stroke ◽

Acute Ischemic Stroke ◽

Learning Algorithms ◽

Case Control ◽

Machine Learning Algorithms ◽

Care Hospital ◽

Diagnosis Codes ◽

Wide Range ◽

Electronic Health

Abstract Background: Accurate identification of acute ischemic stroke (AIS) patient cohorts is essential for a wide range of clinical investigations. Automated phenotyping methods that leverage electronic health records (EHRs) represent a fundamentally new approach cohort identification without current laborious and ungeneralizable generation of phenotyping algorithms. We systematically compared and evaluated the ability of machine learning algorithms and case-control combinations to phenotype acute ischemic stroke patients using data from an EHR.Materials and Methods: Using structured patient data from the EHR at a tertiary-care hospital system, we built and evaluated machine learning models to identify patients with AIS based on 75 different case-control and classifier combinations. We then estimated the prevalence of AIS patients across the EHR. Finally, we externally validated the ability of the models to detect AIS patients without AIS diagnosis codes using the UK Biobank.Results: Across all models, we found that the mean AUROC for detecting AIS was 0.963±0.0520 and average precision score 0.790±0.196 with minimal feature processing. Classifiers trained with cases with AIS diagnosis codes and controls with no cerebrovascular disease codes had the best average F1 score (0.832±0.0383). In the external validation, we found that the top probabilities from a model-predicted AIS cohort were significantly enriched for AIS patients without AIS diagnosis codes (60-150 fold over expected). Conclusions: Our findings support machine learning algorithms as a generalizable way to accurately identify AIS patients without using process-intensive manual feature curation. When a set of AIS patients is unavailable, diagnosis codes may be used to train classifier models.

Download Full-text

Comparative analysis, applications, and interpretation of electronic health record-based stroke phenotyping methods

10.21203/rs.3.rs-16812/v2 ◽

2020 ◽

Author(s):

Phyllis Thangaraj ◽

Benjamin R Kummer ◽

Tal Lorberbaum ◽

Mitchell S.V. Elkind ◽

Nicholas P Tatonetti

Keyword(s):

Machine Learning ◽

Ischemic Stroke ◽

Acute Ischemic Stroke ◽

Learning Algorithms ◽

Case Control ◽

Machine Learning Algorithms ◽

Care Hospital ◽

Diagnosis Codes ◽

Wide Range ◽

Electronic Health

Abstract Background: Accurate identification of acute ischemic stroke (AIS) patient cohorts is essential for a wide range of clinical investigations. Automated phenotyping methods that leverage electronic health records (EHRs) represent a fundamentally new approach cohort identification without current laborious and ungeneralizable generation of phenotyping algorithms. We systematically compared and evaluated the ability of machine learning algorithms and case-control combinations to phenotype acute ischemic stroke patients using data from an EHR.Materials and Methods: Using structured patient data from the EHR at a tertiary-care hospital system, we built and evaluated machine learning models to identify patients with AIS based on 75 different case-control and classifier combinations. We then estimated the prevalence of AIS patients across the EHR. Finally, we externally validated the ability of the models to detect AIS patients without AIS diagnosis codes using the UK Biobank.Results: Across all models, we found that the mean AUROC for detecting AIS was 0.963±0.0520 and average precision score 0.790±0.196 with minimal feature processing. Classifiers trained with cases with AIS diagnosis codes and controls with no cerebrovascular disease codes had the best average F1 score (0.832±0.0383). In the external validation, we found that the top probabilities from a model-predicted AIS cohort were significantly enriched for AIS patients without AIS diagnosis codes (60-150 fold over expected). Conclusions: Our findings support machine learning algorithms as a generalizable way to accurately identify AIS patients without using process-intensive manual feature curation. When a set of AIS patients is unavailable, diagnosis codes may be used to train classifier models.

Download Full-text