Please upgrade your browser

We built this website using the latest browser technologies to deliver the very best experience.

This makes the site run faster and easier to use. Unfortunately, your browser is out of date and will not support some of these technologies.

We recommend that you use a modern browser such as Google Chrome or Microsoft Edge to view this website.

Download Microsoft EdgeDownload Microsoft Edge
Blog

Prediction of acute cardiovascular events from electronic health records

July 15, 2020

Sensyne Health, using anonymised patient data from Oxford University Hospitals NHS Foundation Trust, developed an algorithm to predict the probability of a patient having an acute cardiovascular event in the near future (for example 1, 3 or 12 months).

Cardiovascular diseases are associated with high mortality rates as well as long-lasting morbidities, accounting for 167,000 deaths in the UK each year. They represent a significant burden to the NHS and continued efforts to improve risk prediction are of great importance as they allow preventive interventions.

Existing clinical tools such as the QRISK2 score are often used in primary care settings to assess the 10-year cardiovascular risk of a patient. These tools are derived from population studies and consider well-established risk factors such as age, BMI, smoking, cholesterol levels, hypertension and diabetes. However, short-term predictions can be made following hospital encounter using secondary care electronic health records (EHR).

Fig. 1 – Illustration of acute cardiovascular event prediction. Previous patient observations are used as inputs for a deep-learning model which predicts future risk of ischaemic stroke or heart attack (myocardial infarction – MI) .

The new algorithm predicts the likelihood of a patient developing an ischaemic stroke or heart attack based on the EHR data recorded during their previous in-hospital observations.  Sensyne Health developed a deep learning-based approach which makes use of a recurrent neural network, with an attention mechanism, for multi-task prediction (prediction over several time horizons, from 1 month to 12 months after the last observation). The new algorithm leverages the full breadth of EHR data for prediction, including information on diagnoses, procedures, medications, encounters, demographics, laboratory values and vital signs.

The algorithm demonstrates advantages compared to traditional methods, when assessing risks over various prediction horizons. Through deep learning, the algorithm can analyze and integrate large amounts of data to predict one of two acute cardiovascular events and the attention mechanism provides clinicians with a better understanding of which specific data points at which time in the patient’s history contributed the most to the risk of a cardiovascular event.

Fig. 2 – Performance of different machine learning approaches when predicting the risk of MI within the next month. The rate of true positive predictions is shown with respect to the rate of false positives. Poor models form a straight line on the diagonal (shown dotted). The proposed method outperforms state-of-the-art techniques.
Fig.3 - Importance of clinical features for predicting the risk of MI. The contribution of each clinical variable to the overall prediction of a heart attack within 1 month is shown. Patient age, underlying comorbidities and selected laboratory results signal increase risk of heart attack.

This approach could be used as a risk stratification tool and has the potential to facilitate the targeting of preventive strategies to those most at-risk. The Sensyne Health team will be presenting the findings at the Healthcare Systems, Population Health, and the Role of Health-Tech workshop at the International Conference on Machine Learning (HSYS-ICML 2020).

Blog

Prediction of acute cardiovascular events from electronic health records

July 15, 2020

Sensyne Health, using anonymised patient data from Oxford University Hospitals NHS Foundation Trust, developed an algorithm to predict the probability of a patient having an acute cardiovascular event in the near future (for example 1, 3 or 12 months).

Cardiovascular diseases are associated with high mortality rates as well as long-lasting morbidities, accounting for 167,000 deaths in the UK each year. They represent a significant burden to the NHS and continued efforts to improve risk prediction are of great importance as they allow preventive interventions.

Existing clinical tools such as the QRISK2 score are often used in primary care settings to assess the 10-year cardiovascular risk of a patient. These tools are derived from population studies and consider well-established risk factors such as age, BMI, smoking, cholesterol levels, hypertension and diabetes. However, short-term predictions can be made following hospital encounter using secondary care electronic health records (EHR).

Fig. 1 – Illustration of acute cardiovascular event prediction. Previous patient observations are used as inputs for a deep-learning model which predicts future risk of ischaemic stroke or heart attack (myocardial infarction – MI) .

The new algorithm predicts the likelihood of a patient developing an ischaemic stroke or heart attack based on the EHR data recorded during their previous in-hospital observations.  Sensyne Health developed a deep learning-based approach which makes use of a recurrent neural network, with an attention mechanism, for multi-task prediction (prediction over several time horizons, from 1 month to 12 months after the last observation). The new algorithm leverages the full breadth of EHR data for prediction, including information on diagnoses, procedures, medications, encounters, demographics, laboratory values and vital signs.

The algorithm demonstrates advantages compared to traditional methods, when assessing risks over various prediction horizons. Through deep learning, the algorithm can analyze and integrate large amounts of data to predict one of two acute cardiovascular events and the attention mechanism provides clinicians with a better understanding of which specific data points at which time in the patient’s history contributed the most to the risk of a cardiovascular event.

Fig. 2 – Performance of different machine learning approaches when predicting the risk of MI within the next month. The rate of true positive predictions is shown with respect to the rate of false positives. Poor models form a straight line on the diagonal (shown dotted). The proposed method outperforms state-of-the-art techniques.
Fig.3 - Importance of clinical features for predicting the risk of MI. The contribution of each clinical variable to the overall prediction of a heart attack within 1 month is shown. Patient age, underlying comorbidities and selected laboratory results signal increase risk of heart attack.

This approach could be used as a risk stratification tool and has the potential to facilitate the targeting of preventive strategies to those most at-risk. The Sensyne Health team will be presenting the findings at the Healthcare Systems, Population Health, and the Role of Health-Tech workshop at the International Conference on Machine Learning (HSYS-ICML 2020).

Blog

Prediction of acute cardiovascular events from electronic health records

Prediction of acute cardiovascular events from electronic health records

July 15, 2020

Sensyne Health, using anonymised patient data from Oxford University Hospitals NHS Foundation Trust, developed an algorithm to predict the probability of a patient having an acute cardiovascular event in the near future (for example 1, 3 or 12 months).

Cardiovascular diseases are associated with high mortality rates as well as long-lasting morbidities, accounting for 167,000 deaths in the UK each year. They represent a significant burden to the NHS and continued efforts to improve risk prediction are of great importance as they allow preventive interventions.

Existing clinical tools such as the QRISK2 score are often used in primary care settings to assess the 10-year cardiovascular risk of a patient. These tools are derived from population studies and consider well-established risk factors such as age, BMI, smoking, cholesterol levels, hypertension and diabetes. However, short-term predictions can be made following hospital encounter using secondary care electronic health records (EHR).

Fig. 1 – Illustration of acute cardiovascular event prediction. Previous patient observations are used as inputs for a deep-learning model which predicts future risk of ischaemic stroke or heart attack (myocardial infarction – MI) .

The new algorithm predicts the likelihood of a patient developing an ischaemic stroke or heart attack based on the EHR data recorded during their previous in-hospital observations.  Sensyne Health developed a deep learning-based approach which makes use of a recurrent neural network, with an attention mechanism, for multi-task prediction (prediction over several time horizons, from 1 month to 12 months after the last observation). The new algorithm leverages the full breadth of EHR data for prediction, including information on diagnoses, procedures, medications, encounters, demographics, laboratory values and vital signs.

The algorithm demonstrates advantages compared to traditional methods, when assessing risks over various prediction horizons. Through deep learning, the algorithm can analyze and integrate large amounts of data to predict one of two acute cardiovascular events and the attention mechanism provides clinicians with a better understanding of which specific data points at which time in the patient’s history contributed the most to the risk of a cardiovascular event.

Fig. 2 – Performance of different machine learning approaches when predicting the risk of MI within the next month. The rate of true positive predictions is shown with respect to the rate of false positives. Poor models form a straight line on the diagonal (shown dotted). The proposed method outperforms state-of-the-art techniques.
Fig.3 - Importance of clinical features for predicting the risk of MI. The contribution of each clinical variable to the overall prediction of a heart attack within 1 month is shown. Patient age, underlying comorbidities and selected laboratory results signal increase risk of heart attack.

This approach could be used as a risk stratification tool and has the potential to facilitate the targeting of preventive strategies to those most at-risk. The Sensyne Health team will be presenting the findings at the Healthcare Systems, Population Health, and the Role of Health-Tech workshop at the International Conference on Machine Learning (HSYS-ICML 2020).

Blog

Prediction of acute cardiovascular events from electronic health records

Prediction of acute cardiovascular events from electronic health records

Sensyne Health, using anonymised patient data from Oxford University Hospitals NHS Foundation Trust, developed an algorithm to predict the probability of a patient having an acute cardiovascular event in the near future (for example 1, 3 or 12 months).

Cardiovascular diseases are associated with high mortality rates as well as long-lasting morbidities, accounting for 167,000 deaths in the UK each year. They represent a significant burden to the NHS and continued efforts to improve risk prediction are of great importance as they allow preventive interventions.

Existing clinical tools such as the QRISK2 score are often used in primary care settings to assess the 10-year cardiovascular risk of a patient. These tools are derived from population studies and consider well-established risk factors such as age, BMI, smoking, cholesterol levels, hypertension and diabetes. However, short-term predictions can be made following hospital encounter using secondary care electronic health records (EHR).

Fig. 1 – Illustration of acute cardiovascular event prediction. Previous patient observations are used as inputs for a deep-learning model which predicts future risk of ischaemic stroke or heart attack (myocardial infarction – MI) .

The new algorithm predicts the likelihood of a patient developing an ischaemic stroke or heart attack based on the EHR data recorded during their previous in-hospital observations.  Sensyne Health developed a deep learning-based approach which makes use of a recurrent neural network, with an attention mechanism, for multi-task prediction (prediction over several time horizons, from 1 month to 12 months after the last observation). The new algorithm leverages the full breadth of EHR data for prediction, including information on diagnoses, procedures, medications, encounters, demographics, laboratory values and vital signs.

The algorithm demonstrates advantages compared to traditional methods, when assessing risks over various prediction horizons. Through deep learning, the algorithm can analyze and integrate large amounts of data to predict one of two acute cardiovascular events and the attention mechanism provides clinicians with a better understanding of which specific data points at which time in the patient’s history contributed the most to the risk of a cardiovascular event.

Fig. 2 – Performance of different machine learning approaches when predicting the risk of MI within the next month. The rate of true positive predictions is shown with respect to the rate of false positives. Poor models form a straight line on the diagonal (shown dotted). The proposed method outperforms state-of-the-art techniques.
Fig.3 - Importance of clinical features for predicting the risk of MI. The contribution of each clinical variable to the overall prediction of a heart attack within 1 month is shown. Patient age, underlying comorbidities and selected laboratory results signal increase risk of heart attack.

This approach could be used as a risk stratification tool and has the potential to facilitate the targeting of preventive strategies to those most at-risk. The Sensyne Health team will be presenting the findings at the Healthcare Systems, Population Health, and the Role of Health-Tech workshop at the International Conference on Machine Learning (HSYS-ICML 2020).

Arrange to meet us
Blog

Prediction of acute cardiovascular events from electronic health records

July 15, 2020

Sensyne Health, using anonymised patient data from Oxford University Hospitals NHS Foundation Trust, developed an algorithm to predict the probability of a patient having an acute cardiovascular event in the near future (for example 1, 3 or 12 months).

Cardiovascular diseases are associated with high mortality rates as well as long-lasting morbidities, accounting for 167,000 deaths in the UK each year. They represent a significant burden to the NHS and continued efforts to improve risk prediction are of great importance as they allow preventive interventions.

Existing clinical tools such as the QRISK2 score are often used in primary care settings to assess the 10-year cardiovascular risk of a patient. These tools are derived from population studies and consider well-established risk factors such as age, BMI, smoking, cholesterol levels, hypertension and diabetes. However, short-term predictions can be made following hospital encounter using secondary care electronic health records (EHR).

Fig. 1 – Illustration of acute cardiovascular event prediction. Previous patient observations are used as inputs for a deep-learning model which predicts future risk of ischaemic stroke or heart attack (myocardial infarction – MI) .

The new algorithm predicts the likelihood of a patient developing an ischaemic stroke or heart attack based on the EHR data recorded during their previous in-hospital observations.  Sensyne Health developed a deep learning-based approach which makes use of a recurrent neural network, with an attention mechanism, for multi-task prediction (prediction over several time horizons, from 1 month to 12 months after the last observation). The new algorithm leverages the full breadth of EHR data for prediction, including information on diagnoses, procedures, medications, encounters, demographics, laboratory values and vital signs.

The algorithm demonstrates advantages compared to traditional methods, when assessing risks over various prediction horizons. Through deep learning, the algorithm can analyze and integrate large amounts of data to predict one of two acute cardiovascular events and the attention mechanism provides clinicians with a better understanding of which specific data points at which time in the patient’s history contributed the most to the risk of a cardiovascular event.

Fig. 2 – Performance of different machine learning approaches when predicting the risk of MI within the next month. The rate of true positive predictions is shown with respect to the rate of false positives. Poor models form a straight line on the diagonal (shown dotted). The proposed method outperforms state-of-the-art techniques.
Fig.3 - Importance of clinical features for predicting the risk of MI. The contribution of each clinical variable to the overall prediction of a heart attack within 1 month is shown. Patient age, underlying comorbidities and selected laboratory results signal increase risk of heart attack.

This approach could be used as a risk stratification tool and has the potential to facilitate the targeting of preventive strategies to those most at-risk. The Sensyne Health team will be presenting the findings at the Healthcare Systems, Population Health, and the Role of Health-Tech workshop at the International Conference on Machine Learning (HSYS-ICML 2020).