In a recent study posted to the medRxiv* preprint server, an interdisciplinary team of researchers conducted an open, prospective pilot feasibility analysis through artificial intelligence (AI)-based platform to provide clinical decision support on coronavirus disease 2019 (COVID-19) outcomes.
Study: ARTIFICIAL INTELLIGENCE TOOLS FOR EFFECTIVE MONITORING OF POPULATION AT DISTANCE DURING COVID-19 PANDEMIC. RESULTS FROM AN ITALIAN PILOT FEASIBILITY STUDY (RICOVAI-19 STUDY). Image Credit: elenabsl/Shutterstock
Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2)-related symptoms and disease course pose an enormous burden on the healthcare facilities. During the COVID-19 pandemic, e-telemonitoring was recommended to reduce the pressure on the over-whelming healthcare systems and limit access to the emergency department (ED). Healthcare big data analysis use is increasing, as represented by the explosion of the internet of medical things (IoMT).
*Important notice: medRxiv publishes preliminary scientific reports that are not peer-reviewed and, therefore, should not be regarded as conclusive, guide clinical practice/health-related behavior, or treated as established information.
This study was designed to estimate the application and integration of a dedicated AI-based support system with the territory and hospital intervention plan during the COVID-19 pandemic.
Study design
In this study, the researchers enrolled 129 subjects living in Offagana, Italy, with known or suspected SARS-CoV-2 infection between March 2021 to October 2021. The subjects were over 18 years of age, of which 60 were males and 69 were females, and were monitored for 21 days at home.
The team used the RICOVAI-19 monitoring system, which enabled augmented decision and consisted of an application software accessible through smartphones and a multi-parameter sensor medical device that allowed insertion of clinical parameter values. A dashboard telemedicine platform was used to visualize the clinical stability index (CSI) based on the AI of each patient, which represented the whole journey of the patient. This system supported good clinical practice with an AI-based decision support algorithm solution.
During the study, four phases were taken into consideration:
1. subject enrolment by the general physician (GP);
2. delivery of AI applicative based smartphone and sensor to the subject;
3. activation of the enrolled subject;
4. monitoring for 21 days.
The patients enrolled in the monitoring campaign filled out a questionnaire through a smartphone application twice a day. In the questionnaires, patients answered a set of pre-defined questions regarding the presence and severity of symptoms attributable to SARS-CoV-2 infection, presence of comorbidities, risk factors, and exposure history to any suspected or confirmed SARS-CoV-2 cases. After the questionnaire completion, the application measured the clinical parameters and calculated the CSI, which enabled patient health monitoring and the identification of the negative evolution of COVID-19.
For the machine learning method, the researchers performed a linear discriminant analysis to predict the algorithm of the CSI. A widespread and powerful predictive machine learning tool – decision trees and a popular algorithm classification and regression trees (CART) were used to predict discrete or continuous variables in the study.
Findings
The researchers analyzed that 40% of the enrolled subjects adhered to AI-based digital applications and were contacted by their GP. Among the 158 recruits, 82% participated in RICOVAI-19 experimentation. The mean value of the use and access of digital apps was >60%.
The researchers observed that due to the continuous training on AI, >95% of the enrolled patients experienced CSI. During the monitoring period, 7386 surveys were performed in recruited patients with an average of three surveys daily. About 0.1% of cases had different CSI compared to the results of the RICOVAI-19.
It was found that AI-based digital applications led to a successful increase of more than three interactions per patient with their GPs. There was a significant association of this application with the treatment regimen and provided clinical parameters to GP electronically. From the hospital perspective, there was a significant impact of the digital application-based interactions, with >3 surveys per subject, and provided clinical parameters to an in-hospital specialist every time.
The researchers focused on the efficacy of the CSI during the patient journey from admission to the possible acute event treatment. Out of the 128 enrolled subjects, only one patient was admitted to the ED on family doctor recommendation but not based on the CSI result derived from the AI. Further, the patient was discharged with a negative COVID-19 test from the ED.
The researchers hypothesized that at a high observed CSI value between 7 to 10 (stable patient), there was no indication to admit the patients to the ED. The remaining 128 subjects were home monitored, and none were hospitalized.
Conclusion
The findings of the study showed the beneficial effect of the CSI in predicting clinical categories of the patients and the identification of those requiring ED admissions amid SARS-CoV-2 infection. The AI-based RICOVAI-19 platform enabled quality care of patients and a virtuous decrease in the clinical workload of the healthcare workers during the COVID-19 pandemic.
The current study highlighted the impact of the AI-based software applications on the digital data collected from the patients and doctors and how they can enable earlier diagnosis and timely management of SARS-CoV-2-infected individuals and help track COVID-19 outbreaks by implementing novel approaches for public awareness.
*Important notice: medRxiv publishes preliminary scientific reports that are not peer-reviewed and, therefore, should not be regarded as conclusive, guide clinical practice/health-related behavior, or treated as established information.
Researchers classify post-COVID subtypes and uncover vaccination’s protective role