May Artificial Intelligence take health and sustainability on a honeymoon? Towards green technologies for multidimensional health and environmental justice

Global Bioethics, Journal Year: 2024, Volume and Issue: 35(1)

Published: March 11, 2024

The application of Artificial Intelligence (AI) in healthcare and epidemiology undoubtedly has many benefits for the population. However, due to its environmental impact, use AI can produce social inequalities long-term damages that may not be thoroughly contemplated. In this paper, we propose consider impacts applications medical care from One Health paradigm global health. From health justice, rather than settling a short fleeting green honeymoon between sustainability caused by AI, it should aim lasting marriage. To end, conclude proposing that, upcoming years, could valuable necessary promote more interconnected health, call cost transparency, increase responsibility.

Language: Английский

---

Racial underrepresentation in dermatological datasets leads to biased machine learning models and inequitable healthcare

Journal of Biomed Research, Journal Year: 2022, Volume and Issue: 3(1)

Published: Dec. 31, 2022

Clinical applications of machine learning are promising as a tool to improve patient outcomes through assisting diagnoses, treatment, and analyzing risk factors for screening. Possible clinical especially prominent in dermatology many diseases conditions present visually. This allows model analyze diagnose using images data from electronic health records (EHRs) after training on datasets but could also introduce bias. Despite applications, artificial intelligence has the capacity exacerbate existing demographic disparities healthcare if models trained biased datasets.

Language: Английский

---

Systematic reviews of machine learning in healthcare: a literature review

Expert Review of Pharmacoeconomics & Outcomes Research, Journal Year: 2023, Volume and Issue: 24(1), P. 63 - 115

Published: Nov. 13, 2023

The increasing availability of data and computing power has made machine learning (ML) a viable approach to faster, more efficient healthcare delivery.

Language: Английский

---

Uncovering Clinical Risk Factors and Predicting Severe COVID-19 Cases Using UK Biobank Data: Machine Learning Approach

JMIR Public Health and Surveillance, Journal Year: 2021, Volume and Issue: 7(9), P. e29544 - e29544

Published: Sept. 30, 2021

COVID-19 is a major public health concern. Given the extent of pandemic, it urgent to identify risk factors associated with disease severity. More accurate prediction those at developing severe infections high clinical importance.Based on UK Biobank (UKBB), we aimed build machine learning models predict or fatal infections, and uncover involved.We first restricted analysis infected individuals (n=7846), then performed population level, considering no known infection as controls (ncontrols=465,728). Hospitalization was used proxy for A total 97 variables (collected prior outbreak) covering demographic variables, comorbidities, blood measurements (eg, hematological/liver/renal function/metabolic parameters), anthropometric measures, other smoking/drinking) were included predictors. We also constructed simplified (lite) model using 27 covariates that can be more easily obtained (demographic comorbidity data). XGboost (gradient-boosted trees) predictive performance assessed by cross-validation. Variable importance quantified Shapley values (ShapVal), permutation (PermImp), accuracy gain. dependency interaction plots evaluate pattern relationships between outcomes.A 2386 477 cases identified. For analyses within our achieved area under receiving-operating characteristic curve (AUC-ROC) 0.723 (95% CI 0.711-0.736) 0.814 0.791-0.838) respectively. The top 5 contributing (sorted ShapVal) severity age, number drugs taken (cnt_tx), cystatin C (reflecting renal function), waist-to-hip ratio (WHR), Townsend deprivation index (TDI). mortality, features testosterone, cnt_tx, waist circumference (WC), red cell distribution width. involving whole UKBB population, AUCs fatality 0.696 0.684-0.708) 0.825 0.802-0.848), same identified both outcomes, namely, WC, WHR, TDI. Apart from above, C, TDI, cnt_tx among 10 across all 4 analyses. Other diseases ranked ShapVal PermImp type 2 diabetes mellitus (T2DM), coronary artery disease, atrial fibrillation, dementia, others. "lite" models, performances broadly similar, estimated 0.716, 0.818, 0.696, 0.830, similar including sex (male), T2DM.We numerous baseline severe/fatal XGboost. example, central obesity, impaired function, multiple cardiometabolic abnormalities may predispose poorer outcomes. useful level susceptible facilitating targeted prevention strategies. risk-prediction tool available online. Further replications in independent cohorts are required verify findings.

Language: Английский

---

eXtreme Gradient Boosting-based method to classify patients with COVID-19

Journal of Investigative Medicine, Journal Year: 2022, Volume and Issue: 70(7), P. 1472 - 1480

Published: July 18, 2022

Different demographic, clinical and laboratory variables have been related to the severity mortality following SARS-CoV-2 infection. Most studies applied traditional statistical methods in some cases combined with a machine learning (ML) method. This is first study date comparatively analyze five ML select one that most closely predicts patients admitted COVID-19. The aim of this single-center observational classify, based on different types variables, adult COVID-19 at increased risk mortality. infection was defined by positive reverse transcriptase PCR. A total 203 were between March 15 June 15, 2020 tertiary hospital. Data extracted from electronic medical record. Four supervised algorithms (k-nearest neighbors (KNN), decision tree (DT), Gaussian naïve Bayes (GNB) support vector (SVM)) compared eXtreme Gradient Boosting (XGB) method proposed excellent scalability high running speed, among other qualities. results indicate XGB has best prediction accuracy (92%), precision (>0.92) recall (>0.92). KNN, SVM DT approaches present moderate (>80%), (>0.80) (>0.80). GNB algorithm shows relatively low classification performance. greatest weight predicting C reactive protein, procalcitonin, glutamyl oxaloacetic transaminase, pyruvic neutrophils, D-dimer, creatinine, lactic acid, ferritin, days non-invasive ventilation, septic shock age. Based these results, solid candidate for correct

Language: Английский

---

COVID-19, Cation Dysmetabolism, Sialic Acid, CD147, ACE2, Viroporins, Hepcidin and Ferroptosis: A Possible Unifying Hypothesis

F1000Research, Journal Year: 2022, Volume and Issue: 11, P. 102 - 102

Published: Feb. 28, 2022

Background:iron and calcium dysmetabolism, with hyperferritinemia, hypoferremia, hypocalcemia anemia have been documented in the majority of COVID-19 patients at later/worse stages. Furthermore, complementary to ACE2, both sialic acid (SA) molecules CD147 proved relevant host receptors for SARS-CoV-2 entry, which explains viral attack multiple types cells, including erythrocytes, endothelium neural tissue. Several authors advocated that cell ferroptosis may be core final degenerative mechanism.Methods: a literature research was performed several scientific search engines, such as PubMed Central, Cochrane Library, Chemical Abstract Service. More than 500 articles were retrieved until mid-December 2021, highlight available evidence about investigated issues.Results: based on data, we highlighted few pathophysiological mechanisms, associated virus-based cation multi-organ attack, mitochondria degeneration ferroptosis. Our suggested elucidated pathological sequence is: a) spike protein subunit S1 docking sialylated membrane glycoproteins/receptors (ACE2, CD147), S2 fusion lipid layer; b) morpho-functional changes due consequent electro-chemical variations viroporin action, induce an altered ion channel function intracellular accumulation; c) additional iron concentration deregulated hepcidin-ferroportin axis, higher hepcidin levels. Viral invasion also affect erythrocytes/erythroid precursors, endothelial cells macrophages, through SA receptors, relative hemoglobin iron/calcium dysmetabolism. AB0 blood group, hemochromatosis, or environmental elements represent possible factors individual susceptibility COVID-19. Conclusions: our analysis confirms combined role molecules, CD147, viroporins determining dysmetabolism infected by SARS-CoV-2. The channels electrochemical gradients pivotal virus entry subsequent immune-inflammatory erythrocyte/hemoglobin alterations.

Language: Английский

---

Harnessing the power of AI: Advanced deep learning models optimization for accurate SARS-CoV-2 forecasting

PLoS ONE, Journal Year: 2023, Volume and Issue: 18(7), P. e0287755 - e0287755

Published: July 20, 2023

The pandemic has significantly affected many countries including the USA, UK, Asia, Middle East and Africa region, other countries. Similarly, it substantially Malaysia, making crucial to develop efficient precise forecasting tools for guiding public health policies approaches. Our study is based on advanced deep-learning models predict SARS-CoV-2 cases. We evaluate performance of Long Short-Term Memory (LSTM), Bi-directional LSTM, Convolutional Neural Networks (CNN), CNN-LSTM, Multilayer Perceptron, Gated Recurrent Unit (GRU), (RNN). trained these assessed them using a detailed dataset confirmed cases, demographic data, pertinent socio-economic factors. research aims determine most reliable accurate model cases in region. were able test optimize deep learning with each displaying diverse levels accuracy precision. A comprehensive evaluation models’ discloses appropriate architecture Malaysia’s specific situation. This supports ongoing efforts combat by offering valuable insights into application sophisticated timely case predictions. findings hold considerable implications decision-making, empowering authorities create targeted data-driven interventions limit virus’s spread minimize its effects population.

Language: Английский

---

Medical Imaging Applications of Federated Learning

Diagnostics, Journal Year: 2023, Volume and Issue: 13(19), P. 3140 - 3140

Published: Oct. 6, 2023

Since its introduction in 2016, researchers have applied the idea of Federated Learning (FL) to several domains ranging from edge computing banking. The technique’s inherent security benefits, privacy-preserving capabilities, ease scalability, and ability transcend data biases motivated use this tool on healthcare datasets. While reviews exist detailing FL applications, review focuses solely different applications medical imaging datasets, grouping by diseases, modality, and/or part body. This Systematic Literature was conducted querying consolidating results ArXiv, IEEE Xplorer, PubMed. Furthermore, we provide a detailed description architecture, models, descriptions performance achieved how compare with traditional Machine (ML) models. Additionally, discuss highlighting two primary forms techniques, including homomorphic encryption differential privacy. Finally, some background information context regarding where contributions lie. is organized into following categories: architecture/setup type, data-related topics, security, learning types. progress has been made within field imaging, much room for improvement understanding remains, an emphasis issues remaining concerns researchers. Therefore, improvements are constantly pushing forward. highlighted challenges deploying provided recommendations future directions.

Language: Английский

---

Artificial Intelligence Quality Standards in Healthcare: A Rapid Umbrella Review (Preprint)

Journal of Medical Internet Research, Journal Year: 2024, Volume and Issue: 26, P. e54705 - e54705

Published: April 4, 2024

Background In recent years, there has been an upwelling of artificial intelligence (AI) studies in the health care literature. During this period, increasing number proposed standards to evaluate quality AI studies. Objective This rapid umbrella review examines use a sample systematic articles published over 36-month period. Methods We used modified version Joanna Briggs Institute method. Our approach was informed by practical guide Tricco and colleagues for conducting reviews. search focused on MEDLINE database supplemented with Google Scholar. The inclusion criteria were English-language reviews regardless type, mention abstract, during For synthesis, we summarized issues noted these drawing set standards, harmonized terms used, offered guidance improve future Results selected 33 between 2020 2022 our synthesis. covered wide range objectives, topics, settings, designs, results. Over 60 approaches across different domains identified varying levels detail spanning life cycle stages, making comparisons difficult. Health applied only 39% (13/33) 14% (25/178) original from examined, mostly appraise their methodological or reporting quality. Only handful mentioned transparency, explainability, trustworthiness, ethics, privacy aspects. A total 23 standard–related There recognized need standardize planning, conduct, address broader societal, ethical, regulatory implications. Conclusions Despite growing assess studies, they are seldom practice. With desire adopt domains, practitioners researchers must stay abreast adapt evolving landscape apply

Language: Английский

---

A deep learning model for clinical outcome prediction using longitudinal inpatient electronic health records

medRxiv (Cold Spring Harbor Laboratory), Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 23, 2025

ABSTRACT Objective Recent advances in deep learning show significant potential analyzing continuous monitoring electronic health records (EHR) data for clinical outcome prediction. We aim to develop a Transformer-based, Encounter-level Clinical Outcome (TECO) model predict mortality the intensive care unit (ICU) using inpatient EHR data. Materials and Methods TECO was developed multiple baseline time-dependent variables from 2579 hospitalized COVID-19 patients ICU mortality, validated externally an ARDS cohort (n=2799) sepsis (n=6622) Medical Information Mart Intensive Care (MIMIC)-IV. Model performance evaluated based on area under receiver operating characteristic (AUC) compared with Epic Deterioration Index (EDI), random forest (RF), extreme gradient boosting (XGBoost). Results In development dataset, achieved higher AUC (0.89–0.97) across various time intervals EDI (0.86–0.95), RF (0.87–0.96), XGBoost (0.88–0.96). two MIMIC testing datasets (EDI not available), yielded (0.65–0.76) than (0.57–0.73) (0.57–0.73). addition, able identify clinically interpretable features that were correlated outcome. Discussion outperformed proprietary metrics conventional machine models predicting among non-COVID-19 patients. Conclusions demonstrates strong capability While further validation is needed, has serve as powerful early warning tool diseases settings. LAY SUMMARY units (ICUs), accurately estimating risk of death crucial timely effective medical intervention. This study new AI algorithm, (Transformer-based, model), which uses continuously after admission, update predictions hourly basis. trained over 2,500 designed analyze types collected during patient’s stay. tested TECO’s against widely used tool, other methods, such XGBoost, three patient groups: COVID-19, (acute respiratory distress syndrome), sepsis. consistently showed better earlier methods. Additionally, identified key indicators associated making its more clinicians. These findings suggest could become valuable helping doctors monitor patients’ take action range critical situations.

Language: Английский

---