RapiD_AI as a novel disease and pandemic preparedness tool

In a current examine posted to the medRxiv* preprint server, researchers assessed the RapiD_AI framework for deployable synthetic intelligence (AI) for improved pandemic preparedness.

Study: RapiD_AI: A framework for Rapidly Deployable AI for novel disease & pandemic preparedness. Image Credit: cono0430/Shutterstock


The coronavirus disease 2019 (COVID-19) pandemic will most likely not be the final pandemic the human race will face. Research learning historic pandemics recorded from the 1600s to the current has revealed that the danger of the incidence of a COVID-19-like pandemic has a chance of two.63% yearly and 38% in a lifetime. While it’s tough to forestall such pandemics sooner or later, it’s beneath our management to be ready for his or her antagonistic impacts with acceptable preparedness. 

About the examine

In the current examine, researchers developed a framework known as RapiD_AI that might information the utilization of pre-trained neural community fashions as a tool for bettering pandemic preparedness.

The examine concerned three datasets obtained from the identical inhabitants: (1) DH – basic inpatient cohort from a pre-pandemic dataset collected between January 2016 and December 2019, (2) DW1- COVID-19 sufferers from the primary wave of the pandemic enrolled between March and July 2020, and (3) DW2- COVID-19 sufferers from the second wave of the pandemic enrolled between August 2020 and June 2021.

The observations corresponding to every affected person have been characterised as per a 77-dimensional function vector together with a label displaying a respiratory deterioration occasion inside 24 hours. The options comprised generally assessed laboratory parameters, important indicators, and variance over time.

The experimental course of utilized two job definitions that included affected person deterioration prediction duties. The first job was a respiratory deterioration prediction job TRD which was primarily based on the rise within the stage of oxygen assist required from stage zero or one to stage two or three or unplanned intensive care unit (ICU) admission. The second job was a basic deterioration prediction job TGD outlined as both the composite mortality consequence or ICU admission.

The experimental setup was primarily based on three situations: A, B, and C. In state of affairs A, the group demonstrated the worth of using historic information whereas pretraining RapiD_AI fashions with the background of a pandemic attributable to a novel disease. DH was used to pre-train deep studying fashions, DW1 was used to both prepare the benchmark neural networks or XGBoost fashions or fine-tune pre-trained networks, and DW2 was used as a held-out take a look at dataset to evaluate mannequin efficiency.  

Scenario B hypothesized that choosing essentially the most related pretraining fashions might facilitate the achievement of superior efficiency in comparison with retraining all historic information. This would additionally cut back the computational requirement of the pretraining course of. The group thought of historic examples that had main similarities to COVID-19 information.

Pretraining samples have been chosen by utilizing human expertise in figuring out 5 totally different disease clusters having various levels of similarity to the medical sample noticed for COVID-19 and utilizing a computational strategy that utilized tSNE to cluster all historic information with COVID-19 samples obtained within the preliminary three weeks of the pandemic. Additionally, state of affairs C replicated the state of affairs of a healthcare system that was going through a pandemic and had entry to deploy machine studying fashions.     


The examine outcomes confirmed that the pretraining deep neural community (DNN) fashions from state of affairs A improved their efficiency through the preliminary 20 weeks of the COVID-19 pandemic. Pretraining these DNN fashions improved efficiency by 110.87% relative and 41.71% absolute AUC within the first week and a 3.86% of absolute common AUC within the following 19 weeks of the pandemic.

Furthermore, the RapiD_AI outperformed the baseline XGBoost mannequin within the preliminary 4 weeks of the pandemic by 4.37% relative and 3.58% absolute AUC and the general common by 4.92% relative and 4.21% absolute AUC. These efficiency enhancements can translate into exceptional operational and medical advantages within the context of a international pandemic. The common acquire of 4.21% within the algorithm AUC implied a rise of as much as 1399 extra correct classifications weekly within the UK, which might additional result in improved affected person medical interventions.

Scenario B recognized that essentially the most often famous International Classification of Diseases-10 (ICD10) codes from the ten% of essentially the most related clusters have been I10, Z922, Z864, Z501, I489, N179, Z867, Z921, E119, N390. However, the group highlighted that the code frequency within the basic inhabitants might impression the composition of the often occurring ICD10 codes from chosen coaching clusters and that a number of ICD10 codes noticed for each affected person made it tough to determine the first analysis.

Scenario C resulted in an 11.93% relative and a 9.32% absolute AUC enchancment in efficiency over the preliminary two weeks of the pandemic in comparison with the XGBoost dataset coaching on solely weekly info. The efficiency acquire was constant over the preliminary 20 weeks, with a mean relative and absolute AUC enhance of seven.57% and 6.42%, respectively.


Overall, the examine highlighted the working of the RapiD_AI framework as a tool for pandemic preparedness together with the usefulness of machine studying throughout a pandemic.    

*Important discover

medRxiv publishes preliminary scientific reviews that aren’t peer-reviewed and, subsequently, shouldn’t be regarded as conclusive, information medical follow/health-related conduct, or handled as established info.


Recommended For You