predictive model
2 articles
Artificial intelligence-based techniques for predicting outcomes in COVID-19 patients
Introduction
Currently, extensive research has shown that almost all published prediction models are poorly studied and have significant limitations, leading to their predictive performance often being overestimated. Additionally, there is still no universally accepted scoring system, primarily due to the need for adaptation to heterogeneous patient samples (including patient numbers, clinical profiles, and risk factors) and/or ongoing differences in the organization of healthcare systems across various countries.
Material and methods
This is a narrative literature review. A bibliographic search was conducted in the PubMed, Hinari, SpringerLink, National Center for Biotechnology Information, and Medline databases. Articles published between 2000 and 2024 were selected based on keyword combinations such as "artificial intelligence", "prediction model", "algorithm", "machine learning", and "COVID-19". Information on machine learning predictive models was selected and processed to identify characteristics that can be used to predict diagnosis, severity, length of hospital stay, ICU admission, treatment, vaccination, and mortality in COVID-19 patients. After processing the data according to the search criteria, 125 full-text articles were identified. The final bibliography includes 52 relevant sources, which were considered representative of the literature on this synthesis article topic.
Results
Artificial intelligence techniques are increasingly being used to predict outcomes in COVID-19 patients, particularly in estimating mortality among individuals infected with SARS-CoV-2, which can rapidly and effectively support clinical decision-making. According to the analysis of multiple studies, strong predictors of mortality in COVID-19 patients include advanced age, male gender, comorbidities, reduced levels of calcium, albumin, red blood cells, and oxygen saturation, as well as lymphopenia, elevated blood urea nitrogen, creatinine, lactate dehydrogenase, D-dimers, neutrophils, interleukin-6, procalcitonin, bilirubin, ferritin, aspartate aminotransferase, and troponin.
Conclusions
Artificial intelligence techniques provide potential advantages over conventional assessment methods. The information obtained from machine learning and deep learning algorithms, including easily accessible and interpretable data, can assist healthcare workers in making accurate decisions for the appropriate and timely care of COVID- 19 patients. This can improve patient outcomes, reduce the burden on healthcare systems, and ultimately decrease mortality rates.
Model predicting the onset of antiepileptic drug resistance in women of reproductive age with epilepsy: analytical study
Introduction
In attempt to find an answer regarding the possible scenarios of epilepsy evolution in women of reproductive age (e.g. worsening, remission, antiepileptic drug resistance, status epilepticus occurence), preferably - objective, based on simple, replicable, observable indicators that can be included in a mathematical probability estimation model, could significantly improve their quality of life and increase the effectiveness of prescribed treatments.
Materials and methods
Bidirectional, cohort, descriptive-analytical study, conducted between 2016-2020. Primary data were collected in the Diomid Gherman Institute of Neurology and Neurosurgery, the State Hospital of Republic of Moldova and the Excellence Private Medical Institution. Out of 366 unique parameters, which were recorded in the 159 patients enrolled in the study at each visit (total, 4 documentation visits over 5 years period), 10 parameters were selected for multivariate analysis, considered relevant for predicting clinically significant outcomes. Criteria for parameter relevance were: reaching p≤0.1 in univariate analysis, easy documentation. Subsequently, testing for multicollinearity (calculation of variance inflation factor) and the contribution of each parameter in the formula was performed using the Akaike informativeness criteria. The performance of the developed predictive models was expressed by the area under the ROC curve, positive and negative prognostic power. Statistical analysis: GraphPad Prism, v. 9 trial (Graph Pad Software, Boston, USA).
Results
Age at onset of the disease 14.0±6.3 years; age at first referral to specialist 24.0±7.2 years. The developed predictive model, based on 3 parameters (depressive state, annual frequency of seizures, presence of brain lesions on MRI) has a positive predictive value of 83%, negative of 62%, with an area under the ROC curve of 0.72 (95%CI = 0.56 to 0.88) and a probability of occurrence of 96%.
Conclusions
Depressed patients with documented structural lesions on MRI and a high frequency of epileptic seizures have a progressive, significant risk (an OR of 5.3-24.0) of developing resistance to antiepileptic drugs.