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 ChromeDownload Microsoft Edge
Blog

Early risk assessment of COVID-19 patients using clinical AI technology

March 17, 2021

Sensyne Health researchers in partnership with Chelsea and Westminster Hospital NHS Foundation Trust develop models that can forecast a COVID-19 patient’s need for intensive care and ventilation and the risk of mortality.

Read the full paper here.


Worldwide more than 100 million cases and over 2 million death of COVID-19 have been reported to date [1]. A major clinical challenge associated with the condition is the dramatic difference in disease severity between patients. While around 80% of COVID-19 patients only show mild or asymptomatic infection, 14% will suffer severe symptoms and 6% will require critical care [2]. Of the patients that become critically ill, most will need intensive care or mechanical ventilation. Identifying those patients at risk of severe disease early during their hospital journey can enable clinicians to increase monitoring, focus care efforts and initiate treatment earlier, which may improve therapy success.


An early warning model for severe COVID-19

Sensyne Health together with Chelsea and Westminster Hospital NHS Foundation Trust have studied how state-of-the-art machine learning can help to predict severe COVID-19. When patients first present with symptoms in hospital, a variety of clinical information, vital signs and lab tests are routinely collected. The belief amongst the Sensyne Health and Chelsea and Westminster research group was that these data hold the key to forecast how the disease will progress and that information available during the first few hours of a patient’s hospital stay would indicate whether they will need intensive care or mechanical ventilation.


What can we learn from clinical data?

The research used anonymised electronic patient data from Chelsea and Westminster Hospital NHS Foundation Trust, totalling 1235 records of confirmed COVID-19 positive patients aged between 18 and 100 years. The data set and that number continues to increase as the data is updated regularly. These updates allow for continuous improvements to the model as the COVID-19 situation in the UK evolves. While patient records are carefully anonymised and do not contain any personal details, they comprise basic demographic information, vital signs, laboratory tests and clinical observations.

The data was fed into Sensyne Health’s SENSE™ clinical AI platform, asking the algorithm to predict a patient’s need for critical care and mechanical ventilation and their risk of mortality from the clinical variables it thought were most useful. For each prediction task, a variety of models were trained, including standard methods like logistic regression and random forest classifiers, as well as artificial neural networks.

What the models taught us

Despite the relatively small data set (Sensyne Health machine learning models typically work with thousands of records) the models did surprisingly well. After being trained on patient data, they accurately predicted both the need for intensive care and mechanical ventilation. Performance of this type of model is typically judged by inspecting the rate of true positive predictions over false positive predictions.

In such graphs, a poor model forms a straight line on the diagonal. By contrast, the Sensyne Health’s algorithms occupy the upper left corner of the graph, indicating good performance. The models can indeed forecast future clinical events from early patient data.

Once trained with patient data, the models were asked which clinical characteristics of COVID-19 patients signal poor prognosis. In all models, patient age plays an important role. But beyond this already recognised observation, variables indicating a patient’s oxygen status are enriched among important predictors. Knowledge of what features the models use and their alignment with known indicators of poor patient outcomes can increase clinicians’ trust when using the algorithm.

Next steps

This progress in studying COVID-19 was encouraging. The early research showed that it is feasible to predict disease severity from early patient data, collected during the first few hours after a patient arrives at the hospital. Building on this study, Sensyne Health and Chelsea and Westminster Hospital NHS Foundation Trust developed SYNE-COV™ and SYNE-OPS-1™, by improving on the machine learning models and leveraging recent data of COVID-19 patients. SYNE-COV and SYNE-OPS-1 offer AI-powered decision support and proactive patient management for COVID-19 to clinicians and the NHS.  


References
  1. COVID-19 situation update worldwide, as of week 4, updated 4 February 2021. European Centre for Disease Prevention and Control https://www.ecdc.europa.eu/en/ geographical-distribution-2019-ncov-cases.
  2. Anderson, R. M., Heesterbeek, H., Klinkenberg, D. & Hollingsworth, T. D. How will country-based mitigation measures influence the course of the COVID-19 epidemic? The Lancet 395, 931–934 (2020).

Stefan Heldt
Lead Machine Learning Researcher, Sensyne Health
Blog

Early risk assessment of COVID-19 patients using clinical AI technology

March 17, 2021

Sensyne Health researchers in partnership with Chelsea and Westminster Hospital NHS Foundation Trust develop models that can forecast a COVID-19 patient’s need for intensive care and ventilation and the risk of mortality.

Read the full paper here.


Worldwide more than 100 million cases and over 2 million death of COVID-19 have been reported to date [1]. A major clinical challenge associated with the condition is the dramatic difference in disease severity between patients. While around 80% of COVID-19 patients only show mild or asymptomatic infection, 14% will suffer severe symptoms and 6% will require critical care [2]. Of the patients that become critically ill, most will need intensive care or mechanical ventilation. Identifying those patients at risk of severe disease early during their hospital journey can enable clinicians to increase monitoring, focus care efforts and initiate treatment earlier, which may improve therapy success.


An early warning model for severe COVID-19

Sensyne Health together with Chelsea and Westminster Hospital NHS Foundation Trust have studied how state-of-the-art machine learning can help to predict severe COVID-19. When patients first present with symptoms in hospital, a variety of clinical information, vital signs and lab tests are routinely collected. The belief amongst the Sensyne Health and Chelsea and Westminster research group was that these data hold the key to forecast how the disease will progress and that information available during the first few hours of a patient’s hospital stay would indicate whether they will need intensive care or mechanical ventilation.


What can we learn from clinical data?

The research used anonymised electronic patient data from Chelsea and Westminster Hospital NHS Foundation Trust, totalling 1235 records of confirmed COVID-19 positive patients aged between 18 and 100 years. The data set and that number continues to increase as the data is updated regularly. These updates allow for continuous improvements to the model as the COVID-19 situation in the UK evolves. While patient records are carefully anonymised and do not contain any personal details, they comprise basic demographic information, vital signs, laboratory tests and clinical observations.

The data was fed into Sensyne Health’s SENSE™ clinical AI platform, asking the algorithm to predict a patient’s need for critical care and mechanical ventilation and their risk of mortality from the clinical variables it thought were most useful. For each prediction task, a variety of models were trained, including standard methods like logistic regression and random forest classifiers, as well as artificial neural networks.

What the models taught us

Despite the relatively small data set (Sensyne Health machine learning models typically work with thousands of records) the models did surprisingly well. After being trained on patient data, they accurately predicted both the need for intensive care and mechanical ventilation. Performance of this type of model is typically judged by inspecting the rate of true positive predictions over false positive predictions.

In such graphs, a poor model forms a straight line on the diagonal. By contrast, the Sensyne Health’s algorithms occupy the upper left corner of the graph, indicating good performance. The models can indeed forecast future clinical events from early patient data.

Once trained with patient data, the models were asked which clinical characteristics of COVID-19 patients signal poor prognosis. In all models, patient age plays an important role. But beyond this already recognised observation, variables indicating a patient’s oxygen status are enriched among important predictors. Knowledge of what features the models use and their alignment with known indicators of poor patient outcomes can increase clinicians’ trust when using the algorithm.

Next steps

This progress in studying COVID-19 was encouraging. The early research showed that it is feasible to predict disease severity from early patient data, collected during the first few hours after a patient arrives at the hospital. Building on this study, Sensyne Health and Chelsea and Westminster Hospital NHS Foundation Trust developed SYNE-COV™ and SYNE-OPS-1™, by improving on the machine learning models and leveraging recent data of COVID-19 patients. SYNE-COV and SYNE-OPS-1 offer AI-powered decision support and proactive patient management for COVID-19 to clinicians and the NHS.  


References
  1. COVID-19 situation update worldwide, as of week 4, updated 4 February 2021. European Centre for Disease Prevention and Control https://www.ecdc.europa.eu/en/ geographical-distribution-2019-ncov-cases.
  2. Anderson, R. M., Heesterbeek, H., Klinkenberg, D. & Hollingsworth, T. D. How will country-based mitigation measures influence the course of the COVID-19 epidemic? The Lancet 395, 931–934 (2020).

Stefan Heldt
Lead Machine Learning Researcher, Sensyne Health
Blog

Early risk assessment of COVID-19 patients 
using clinical AI technology

Early risk assessment of COVID-19 patients using clinical AI technology

March 17, 2021

Sensyne Health researchers in partnership with Chelsea and Westminster Hospital NHS Foundation Trust develop models that can forecast a COVID-19 patient’s need for intensive care and ventilation and the risk of mortality.

Read the full paper here.


Worldwide more than 100 million cases and over 2 million death of COVID-19 have been reported to date [1]. A major clinical challenge associated with the condition is the dramatic difference in disease severity between patients. While around 80% of COVID-19 patients only show mild or asymptomatic infection, 14% will suffer severe symptoms and 6% will require critical care [2]. Of the patients that become critically ill, most will need intensive care or mechanical ventilation. Identifying those patients at risk of severe disease early during their hospital journey can enable clinicians to increase monitoring, focus care efforts and initiate treatment earlier, which may improve therapy success.


An early warning model for severe COVID-19

Sensyne Health together with Chelsea and Westminster Hospital NHS Foundation Trust have studied how state-of-the-art machine learning can help to predict severe COVID-19. When patients first present with symptoms in hospital, a variety of clinical information, vital signs and lab tests are routinely collected. The belief amongst the Sensyne Health and Chelsea and Westminster research group was that these data hold the key to forecast how the disease will progress and that information available during the first few hours of a patient’s hospital stay would indicate whether they will need intensive care or mechanical ventilation.


What can we learn from clinical data?

The research used anonymised electronic patient data from Chelsea and Westminster Hospital NHS Foundation Trust, totalling 1235 records of confirmed COVID-19 positive patients aged between 18 and 100 years. The data set and that number continues to increase as the data is updated regularly. These updates allow for continuous improvements to the model as the COVID-19 situation in the UK evolves. While patient records are carefully anonymised and do not contain any personal details, they comprise basic demographic information, vital signs, laboratory tests and clinical observations.

The data was fed into Sensyne Health’s SENSE™ clinical AI platform, asking the algorithm to predict a patient’s need for critical care and mechanical ventilation and their risk of mortality from the clinical variables it thought were most useful. For each prediction task, a variety of models were trained, including standard methods like logistic regression and random forest classifiers, as well as artificial neural networks.

What the models taught us

Despite the relatively small data set (Sensyne Health machine learning models typically work with thousands of records) the models did surprisingly well. After being trained on patient data, they accurately predicted both the need for intensive care and mechanical ventilation. Performance of this type of model is typically judged by inspecting the rate of true positive predictions over false positive predictions.

In such graphs, a poor model forms a straight line on the diagonal. By contrast, the Sensyne Health’s algorithms occupy the upper left corner of the graph, indicating good performance. The models can indeed forecast future clinical events from early patient data.

Once trained with patient data, the models were asked which clinical characteristics of COVID-19 patients signal poor prognosis. In all models, patient age plays an important role. But beyond this already recognised observation, variables indicating a patient’s oxygen status are enriched among important predictors. Knowledge of what features the models use and their alignment with known indicators of poor patient outcomes can increase clinicians’ trust when using the algorithm.

Next steps

This progress in studying COVID-19 was encouraging. The early research showed that it is feasible to predict disease severity from early patient data, collected during the first few hours after a patient arrives at the hospital. Building on this study, Sensyne Health and Chelsea and Westminster Hospital NHS Foundation Trust developed SYNE-COV™ and SYNE-OPS-1™, by improving on the machine learning models and leveraging recent data of COVID-19 patients. SYNE-COV and SYNE-OPS-1 offer AI-powered decision support and proactive patient management for COVID-19 to clinicians and the NHS.  


References
  1. COVID-19 situation update worldwide, as of week 4, updated 4 February 2021. European Centre for Disease Prevention and Control https://www.ecdc.europa.eu/en/ geographical-distribution-2019-ncov-cases.
  2. Anderson, R. M., Heesterbeek, H., Klinkenberg, D. & Hollingsworth, T. D. How will country-based mitigation measures influence the course of the COVID-19 epidemic? The Lancet 395, 931–934 (2020).

Stefan Heldt
Lead Machine Learning Researcher, Sensyne Health
Blog

Early risk assessment of COVID-19 patients 
using clinical AI technology

Early risk assessment of COVID-19 patients using clinical AI technology

Sensyne Health researchers in partnership with Chelsea and Westminster Hospital NHS Foundation Trust develop models that can forecast a COVID-19 patient’s need for intensive care and ventilation and the risk of mortality.

Read the full paper here.


Worldwide more than 100 million cases and over 2 million death of COVID-19 have been reported to date [1]. A major clinical challenge associated with the condition is the dramatic difference in disease severity between patients. While around 80% of COVID-19 patients only show mild or asymptomatic infection, 14% will suffer severe symptoms and 6% will require critical care [2]. Of the patients that become critically ill, most will need intensive care or mechanical ventilation. Identifying those patients at risk of severe disease early during their hospital journey can enable clinicians to increase monitoring, focus care efforts and initiate treatment earlier, which may improve therapy success.


An early warning model for severe COVID-19

Sensyne Health together with Chelsea and Westminster Hospital NHS Foundation Trust have studied how state-of-the-art machine learning can help to predict severe COVID-19. When patients first present with symptoms in hospital, a variety of clinical information, vital signs and lab tests are routinely collected. The belief amongst the Sensyne Health and Chelsea and Westminster research group was that these data hold the key to forecast how the disease will progress and that information available during the first few hours of a patient’s hospital stay would indicate whether they will need intensive care or mechanical ventilation.


What can we learn from clinical data?

The research used anonymised electronic patient data from Chelsea and Westminster Hospital NHS Foundation Trust, totalling 1235 records of confirmed COVID-19 positive patients aged between 18 and 100 years. The data set and that number continues to increase as the data is updated regularly. These updates allow for continuous improvements to the model as the COVID-19 situation in the UK evolves. While patient records are carefully anonymised and do not contain any personal details, they comprise basic demographic information, vital signs, laboratory tests and clinical observations.

The data was fed into Sensyne Health’s SENSE™ clinical AI platform, asking the algorithm to predict a patient’s need for critical care and mechanical ventilation and their risk of mortality from the clinical variables it thought were most useful. For each prediction task, a variety of models were trained, including standard methods like logistic regression and random forest classifiers, as well as artificial neural networks.

What the models taught us

Despite the relatively small data set (Sensyne Health machine learning models typically work with thousands of records) the models did surprisingly well. After being trained on patient data, they accurately predicted both the need for intensive care and mechanical ventilation. Performance of this type of model is typically judged by inspecting the rate of true positive predictions over false positive predictions.

In such graphs, a poor model forms a straight line on the diagonal. By contrast, the Sensyne Health’s algorithms occupy the upper left corner of the graph, indicating good performance. The models can indeed forecast future clinical events from early patient data.

Once trained with patient data, the models were asked which clinical characteristics of COVID-19 patients signal poor prognosis. In all models, patient age plays an important role. But beyond this already recognised observation, variables indicating a patient’s oxygen status are enriched among important predictors. Knowledge of what features the models use and their alignment with known indicators of poor patient outcomes can increase clinicians’ trust when using the algorithm.

Next steps

This progress in studying COVID-19 was encouraging. The early research showed that it is feasible to predict disease severity from early patient data, collected during the first few hours after a patient arrives at the hospital. Building on this study, Sensyne Health and Chelsea and Westminster Hospital NHS Foundation Trust developed SYNE-COV™ and SYNE-OPS-1™, by improving on the machine learning models and leveraging recent data of COVID-19 patients. SYNE-COV and SYNE-OPS-1 offer AI-powered decision support and proactive patient management for COVID-19 to clinicians and the NHS.  


References
  1. COVID-19 situation update worldwide, as of week 4, updated 4 February 2021. European Centre for Disease Prevention and Control https://www.ecdc.europa.eu/en/ geographical-distribution-2019-ncov-cases.
  2. Anderson, R. M., Heesterbeek, H., Klinkenberg, D. & Hollingsworth, T. D. How will country-based mitigation measures influence the course of the COVID-19 epidemic? The Lancet 395, 931–934 (2020).

Stefan Heldt
Lead Machine Learning Researcher, Sensyne Health
Arrange to meet us
Blog

Early risk assessment of COVID-19 patients 
using clinical AI technology

March 17, 2021

Sensyne Health researchers in partnership with Chelsea and Westminster Hospital NHS Foundation Trust develop models that can forecast a COVID-19 patient’s need for intensive care and ventilation and the risk of mortality.

Read the full paper here.


Worldwide more than 100 million cases and over 2 million death of COVID-19 have been reported to date [1]. A major clinical challenge associated with the condition is the dramatic difference in disease severity between patients. While around 80% of COVID-19 patients only show mild or asymptomatic infection, 14% will suffer severe symptoms and 6% will require critical care [2]. Of the patients that become critically ill, most will need intensive care or mechanical ventilation. Identifying those patients at risk of severe disease early during their hospital journey can enable clinicians to increase monitoring, focus care efforts and initiate treatment earlier, which may improve therapy success.


An early warning model for severe COVID-19

Sensyne Health together with Chelsea and Westminster Hospital NHS Foundation Trust have studied how state-of-the-art machine learning can help to predict severe COVID-19. When patients first present with symptoms in hospital, a variety of clinical information, vital signs and lab tests are routinely collected. The belief amongst the Sensyne Health and Chelsea and Westminster research group was that these data hold the key to forecast how the disease will progress and that information available during the first few hours of a patient’s hospital stay would indicate whether they will need intensive care or mechanical ventilation.


What can we learn from clinical data?

The research used anonymised electronic patient data from Chelsea and Westminster Hospital NHS Foundation Trust, totalling 1235 records of confirmed COVID-19 positive patients aged between 18 and 100 years. The data set and that number continues to increase as the data is updated regularly. These updates allow for continuous improvements to the model as the COVID-19 situation in the UK evolves. While patient records are carefully anonymised and do not contain any personal details, they comprise basic demographic information, vital signs, laboratory tests and clinical observations.

The data was fed into Sensyne Health’s SENSE™ clinical AI platform, asking the algorithm to predict a patient’s need for critical care and mechanical ventilation and their risk of mortality from the clinical variables it thought were most useful. For each prediction task, a variety of models were trained, including standard methods like logistic regression and random forest classifiers, as well as artificial neural networks.

What the models taught us

Despite the relatively small data set (Sensyne Health machine learning models typically work with thousands of records) the models did surprisingly well. After being trained on patient data, they accurately predicted both the need for intensive care and mechanical ventilation. Performance of this type of model is typically judged by inspecting the rate of true positive predictions over false positive predictions.

In such graphs, a poor model forms a straight line on the diagonal. By contrast, the Sensyne Health’s algorithms occupy the upper left corner of the graph, indicating good performance. The models can indeed forecast future clinical events from early patient data.

Once trained with patient data, the models were asked which clinical characteristics of COVID-19 patients signal poor prognosis. In all models, patient age plays an important role. But beyond this already recognised observation, variables indicating a patient’s oxygen status are enriched among important predictors. Knowledge of what features the models use and their alignment with known indicators of poor patient outcomes can increase clinicians’ trust when using the algorithm.

Next steps

This progress in studying COVID-19 was encouraging. The early research showed that it is feasible to predict disease severity from early patient data, collected during the first few hours after a patient arrives at the hospital. Building on this study, Sensyne Health and Chelsea and Westminster Hospital NHS Foundation Trust developed SYNE-COV™ and SYNE-OPS-1™, by improving on the machine learning models and leveraging recent data of COVID-19 patients. SYNE-COV and SYNE-OPS-1 offer AI-powered decision support and proactive patient management for COVID-19 to clinicians and the NHS.  


References
  1. COVID-19 situation update worldwide, as of week 4, updated 4 February 2021. European Centre for Disease Prevention and Control https://www.ecdc.europa.eu/en/ geographical-distribution-2019-ncov-cases.
  2. Anderson, R. M., Heesterbeek, H., Klinkenberg, D. & Hollingsworth, T. D. How will country-based mitigation measures influence the course of the COVID-19 epidemic? The Lancet 395, 931–934 (2020).

Stefan Heldt
Lead Machine Learning Researcher, Sensyne Health