Source – https://www.news-medical.net/
The overwhelming increase in critically ill COVID-19 patients, who urgently require intensive care units (ICU) and emergency departments (ED), has challenged healthcare systems worldwide.
The unprecedented and large number of patients, especially in the regions where the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) pandemic has hit badly, has created an immediate need for novel approaches to dealing with the issue. One such approach is the application of artificial intelligence (AI). Now, in a new research paper published on the medRxiv* server, an international team of scientists has systematically reviewed and critically appraised the current evidence on AI applications for COVID-19 in intensive care and emergency settings, focusing on methods, reporting standards, and clinical utility.
Artificial intelligence uses computational methods to replicate human intelligence. Two branches of AI, namely, deep learning and machine learning, are involved in the automatic development of computer programs through experience. In medical research, various regression models such as logistic, linear, or Cox regressions are used to develop AI-based applications. These models are the simplest form of machine learning. However, more recently, advanced and complex forms of AI, i.e., machine learning, including neural networks, random forest models, and support vector machines, are becoming more popular in medical research.
Prior research has shown that AI could assist with the automated monitoring of patients in intensive care and emergency settings, prognostication, and optimal allocation of staff. Previous systematic reviews have also revealed the issues concerning the quality of COVID-19 prediction models developed for the disease’s diagnosis and prognosis. The research concluded that the machine learning studies’ limitations are inadequate sample size and insufficient validation of predictions. In the current scenario where the rate of hospitalization is exceptionally high, researchers mainly focus on the use of AI for optimization of ICU bed usage.
Presently, not much information is available about AI’s role as a decisive technology in the clinical management of COVID-19 patients in ICU and emergency settings. Thereby, scientists have systematically analyzed the existing documents involving the application of AI for COVID-19 patients admitted to the intensive care unit of a hospital. They have also focused on the clinical efficacy, various methods, and reporting standards associated with the emergency settings.
For this study, a thorough review of the literature, available in IEEE Xplore, Scopus, Embase, ACM Digital Library, CINAHL, and PubMed, was conducted from inception pandemic to 1st October 2020. The literature review was done across papers published in different languages. All the articles were associated with the application of AI for COVID-19 patients, healthcare resources in the intensive care unit, emergency, prehospital settings, and health care workers. For the predictive modeling studies, researchers have used various tools such as the prediction model risk of bias assessment tool (PROBAST) and a modified transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD).
Among the fourteen studies that were analyzed, eleven developed predictive AI-based diagnostic models. Two of the three remaining studies showed the development of lung segmentation software (based on deep learning) used for prognosis, and the remaining study was associated with the optimization in the ICU.
All of these studies were assessed to be at a high risk of bias. Some of the common drawbacks of these studies were poor handling of missing data, weak validation of models, small sample sizes, and failure to account for censored participants. Among the studies, the most common source of bias that commonly prevailed was the inadequate sample size. A small sample size leads to the risk of over-fitting and model optimism. Missing data also leads to significant error in the model, and ideally, the percentage of the missing variable must be reported. In the case of diagnostic studies, bias was introduced while using the reverse transcription-polymerase chain reaction (RT-PCR) test as many a time a false-negative report arises in the diagnostic model and validation study. Additionally, poor reporting on model calibration, development of proper guidelines, and lack of accurate reports of predictor assessment have failed to validate the research in clinical settings.
The current systematic review has shown that despite the rapid development of novel technologies to contain the COVID-19 pandemic, there is a shortage in AI-based applications for clinical applicability. A valuable improvement in the development and deployment of AI applications in emergency settings could help combat the current situation, which requires optimal emergency resources usage. Integration of new AI-specific reporting guidelines such as SPIRIT-AI and CONSORT-AI into research would help develop novel AI-based applications. Such applications would help the health care system to fight the COVID-19 pandemic and other future pandemics. Researchers have emphasized the need for interdisciplinary collaboration between AI developers and medical experts.