Abstract
INTRODUCTION: COVID19 has challenged the already established predictive tools in critical care There is a plethora of unintegrated, untested, underpowered risk-stratification tools, derived using traditional linear modelling with limited predictability for the wider population Artificial neural network approaches could help developing tools which can be integrated into EMRs METHODS: We have developed a proof of concept algorithm based on our proprietary artificial neural network (ANN) methodology using 8 markers derived from clinical and laboratory data of 40 critical care patients obtained within 24 hours of ICU admission The methodology uses a proprietary stepwise ANN algorithm with extensive regularisation and cross validation to identify an optimised panel of markers having maximum sensitivity and specificity The crossvalidation strategy does not rely on large data sets but rather analyses smaller data sets in parallel In this process, markers are added sequentially, and their classification performance monitored The optimised marker panel is then incorporated into a tuned classifier further optimising performance The final stage takes the validated classifier and converts it to a piece of software RESULTS: Our ANN derived clinical risk-stratification model predicts ICU mortality with a sensitivity of 96% and a specificity of 94% The optimal performance was reached after adding eight parameters: LDH, Bilirubin, Heart rate, platelet count, sex, age, white blood cell count and number of co-morbidities The model resulted in two misclassifications: The one false positive case was in the training set and one false negative in the test set The model AUC overall was 0 98 with a corrected threshold of 0 54 CONCLUSIONS: During the pandemic, significant amounts of health data covering admission, monitoring and outcome have been collected Routine ICU data capture is well developed and standardised Our preliminary results support the use of the ANN derived algorithm for mortality prediction, derived from commonly available clinical data Our model will need external validation in larger datasets from diverse geographic regions, however has the potential to be integrated in electronic clinical information systems to aid resource utilisation and recruitment of high-risk patients to clinical trials
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.