Case study

Improving stroke prediction

October 7, 2019

Overview

Stroke is a serious life-threatening condition that affects around 100,000 adults and children in the UK every year. Stroke happens when the blood supply to part of the brain is cut off. It is considered a medical emergency where the speed of treatment directly affects the amount of brain damage a patient may suffer. The current clinical guidelines used by clinicians in primary care for estimating and scoring individual stroke probability consider well- established risk factors such as high blood pressure, cholesterol, age, level of activity, smoking and diabetes.

Purpose

Electronic health records (EHR) have the potential to provide additional information that could be used to predict stroke risk, particularly when using AI and machine learning to research, contextualise and draw predictive conclusions from EHR data. EHR data has the added benefit that it is routinely collected in hospital at a higher frequency and scale than traditional well-established risk factor information.

The work

The Discovery Sciences team at Sensyne Health has developed a recurrent deep neural network algorithm for predicting the risk of ischaemic stroke on readmission. The method takes into consideration the established risk factors as well as EHR data such as vital signs, comorbidities, laboratory results and demographic information. Anonymised data from Sensyne Health’s partner Oxford University Hospitals NHS Foundation Trust was used to compare the performance of traditionally established risk factors and other conventional methods against a machine learning approach combining conventional methods with EHR data.

Outcomes

Sensyne Health’s new approach has shown promising preliminary results, outperforming both clinical standard and alternative machine learning methods.

References:

Goldstein, B. A. et al. (2017) ‘Opportunities and challenges in developing risk prediction models with electronic health records data: a systematic review’, Journal of the American Medical Informatics Association, 24(1), pp. 198–208.

Case study

Improving stroke prediction

October 7, 2019

Overview

Stroke is a serious life-threatening condition that affects around 100,000 adults and children in the UK every year. Stroke happens when the blood supply to part of the brain is cut off. It is considered a medical emergency where the speed of treatment directly affects the amount of brain damage a patient may suffer. The current clinical guidelines used by clinicians in primary care for estimating and scoring individual stroke probability consider well- established risk factors such as high blood pressure, cholesterol, age, level of activity, smoking and diabetes.

Purpose

Electronic health records (EHR) have the potential to provide additional information that could be used to predict stroke risk, particularly when using AI and machine learning to research, contextualise and draw predictive conclusions from EHR data. EHR data has the added benefit that it is routinely collected in hospital at a higher frequency and scale than traditional well-established risk factor information.

The work

The Discovery Sciences team at Sensyne Health has developed a recurrent deep neural network algorithm for predicting the risk of ischaemic stroke on readmission. The method takes into consideration the established risk factors as well as EHR data such as vital signs, comorbidities, laboratory results and demographic information. Anonymised data from Sensyne Health’s partner Oxford University Hospitals NHS Foundation Trust was used to compare the performance of traditionally established risk factors and other conventional methods against a machine learning approach combining conventional methods with EHR data.

Outcomes

Sensyne Health’s new approach has shown promising preliminary results, outperforming both clinical standard and alternative machine learning methods.

References:

Goldstein, B. A. et al. (2017) ‘Opportunities and challenges in developing risk prediction models with electronic health records data: a systematic review’, Journal of the American Medical Informatics Association, 24(1), pp. 198–208.

Case study

Improving stroke prediction

Improving stroke prediction

October 7, 2019

Overview

Stroke is a serious life-threatening condition that affects around 100,000 adults and children in the UK every year. Stroke happens when the blood supply to part of the brain is cut off. It is considered a medical emergency where the speed of treatment directly affects the amount of brain damage a patient may suffer. The current clinical guidelines used by clinicians in primary care for estimating and scoring individual stroke probability consider well- established risk factors such as high blood pressure, cholesterol, age, level of activity, smoking and diabetes.

Purpose

Electronic health records (EHR) have the potential to provide additional information that could be used to predict stroke risk, particularly when using AI and machine learning to research, contextualise and draw predictive conclusions from EHR data. EHR data has the added benefit that it is routinely collected in hospital at a higher frequency and scale than traditional well-established risk factor information.

The work

The Discovery Sciences team at Sensyne Health has developed a recurrent deep neural network algorithm for predicting the risk of ischaemic stroke on readmission. The method takes into consideration the established risk factors as well as EHR data such as vital signs, comorbidities, laboratory results and demographic information. Anonymised data from Sensyne Health’s partner Oxford University Hospitals NHS Foundation Trust was used to compare the performance of traditionally established risk factors and other conventional methods against a machine learning approach combining conventional methods with EHR data.

Outcomes

Sensyne Health’s new approach has shown promising preliminary results, outperforming both clinical standard and alternative machine learning methods.

References:

Goldstein, B. A. et al. (2017) ‘Opportunities and challenges in developing risk prediction models with electronic health records data: a systematic review’, Journal of the American Medical Informatics Association, 24(1), pp. 198–208.

Case study

Improving stroke prediction

Improving stroke prediction

Overview

Stroke is a serious life-threatening condition that affects around 100,000 adults and children in the UK every year. Stroke happens when the blood supply to part of the brain is cut off. It is considered a medical emergency where the speed of treatment directly affects the amount of brain damage a patient may suffer. The current clinical guidelines used by clinicians in primary care for estimating and scoring individual stroke probability consider well- established risk factors such as high blood pressure, cholesterol, age, level of activity, smoking and diabetes.

Purpose

Electronic health records (EHR) have the potential to provide additional information that could be used to predict stroke risk, particularly when using AI and machine learning to research, contextualise and draw predictive conclusions from EHR data. EHR data has the added benefit that it is routinely collected in hospital at a higher frequency and scale than traditional well-established risk factor information.

The work

The Discovery Sciences team at Sensyne Health has developed a recurrent deep neural network algorithm for predicting the risk of ischaemic stroke on readmission. The method takes into consideration the established risk factors as well as EHR data such as vital signs, comorbidities, laboratory results and demographic information. Anonymised data from Sensyne Health’s partner Oxford University Hospitals NHS Foundation Trust was used to compare the performance of traditionally established risk factors and other conventional methods against a machine learning approach combining conventional methods with EHR data.

Outcomes

Sensyne Health’s new approach has shown promising preliminary results, outperforming both clinical standard and alternative machine learning methods.

References:

Goldstein, B. A. et al. (2017) ‘Opportunities and challenges in developing risk prediction models with electronic health records data: a systematic review’, Journal of the American Medical Informatics Association, 24(1), pp. 198–208.

Arrange to meet us
Case study

Improving stroke prediction

October 7, 2019

Overview

Stroke is a serious life-threatening condition that affects around 100,000 adults and children in the UK every year. Stroke happens when the blood supply to part of the brain is cut off. It is considered a medical emergency where the speed of treatment directly affects the amount of brain damage a patient may suffer. The current clinical guidelines used by clinicians in primary care for estimating and scoring individual stroke probability consider well- established risk factors such as high blood pressure, cholesterol, age, level of activity, smoking and diabetes.

Purpose

Electronic health records (EHR) have the potential to provide additional information that could be used to predict stroke risk, particularly when using AI and machine learning to research, contextualise and draw predictive conclusions from EHR data. EHR data has the added benefit that it is routinely collected in hospital at a higher frequency and scale than traditional well-established risk factor information.

The work

The Discovery Sciences team at Sensyne Health has developed a recurrent deep neural network algorithm for predicting the risk of ischaemic stroke on readmission. The method takes into consideration the established risk factors as well as EHR data such as vital signs, comorbidities, laboratory results and demographic information. Anonymised data from Sensyne Health’s partner Oxford University Hospitals NHS Foundation Trust was used to compare the performance of traditionally established risk factors and other conventional methods against a machine learning approach combining conventional methods with EHR data.

Outcomes

Sensyne Health’s new approach has shown promising preliminary results, outperforming both clinical standard and alternative machine learning methods.

References:

Goldstein, B. A. et al. (2017) ‘Opportunities and challenges in developing risk prediction models with electronic health records data: a systematic review’, Journal of the American Medical Informatics Association, 24(1), pp. 198–208.