Search Results (2,091)

Background/Objectives: Rituximab has improved outcomes in pemphigus vulgaris (PV), but treatment response remains heterogeneous and no clinically validated biomarker currently supports reliable individual-level prediction. Oral lesions frequently represent the earliest manifestation of PV and contribute substantially to disease burden and longitudinal assessment. This perspective aimed to synthesize candidate biomarkers and organize them within a conceptual, task-specific artificial intelligence (AI) framework for future rituximab outcome modeling. Methods: A structured literature search and narrative synthesis evaluated genetic and pharmacogenomic, immunological and serological, cellular, clinical, and oral phenotypic variables potentially relevant to rituximab response. Variables were considered according to biological rationale, available evidence, temporal availability, and potential use in pretreatment prognosis or post-treatment response monitoring. No patient-level dataset was analyzed, and no model was trained or validated. Results: Candidate pharmacogenomic variables include FCGR3A rs396991, FCGR2A rs1801274, and IL6 rs1800795, although evidence derives mainly from autoimmune diseases other than PV. IL10, TNF, TNFRSF13B, and population-specific HLA variables remain exploratory. Anti-desmoglein autoantibodies, BAFF, cytokine profiles, CD19⁺ B-cell depletion, CD27⁺ memory B-cell repopulation, regulatory T-cell dynamics, natural killer cell phenotypes, and clinical and oral phenotypic variables may provide complementary information, but none currently demonstrates sufficient independent predictive validity. The proposed RTX-AI Response Framework distinguishes the Pretreatment Prediction Component from the Post-Treatment Response Monitoring Component and assigns no numerical weights, probabilities, thresholds, or treatment recommendations. Conclusions: The framework should be interpreted exclusively as a conceptual research architecture. Prospective multicenter data collection, standardized biomarker assessment, task-specific model development, calibration, external validation, and evaluation of clinical utility are required before potential clinical implementation. Full article

Mountain biking over rough terrain exposes riders to vibrations that may affect comfort, health, bicycle control, and performance. This study analyzed the effect of wheel diameter on vibration transmission and performance during a short uphill and downhill test over rocky terrain. Forty-nine highly trained male mountain bikers completed repeated trials using two equivalent hardtail mountain bikes with 26- and 29-inch wheels. Vibrations were recorded with eight triaxial accelerometers placed on the bicycle and rider, and performance was assessed using an electronic photocell timing system. Both wheel sizes showed a similar vibration pattern, with lower root mean square (RMS) acceleration values at the helmet and coccyx and higher values at the wrists and ankles. However, wheel diameter significantly influenced vibration transmission. During the uphill test, the 26-inch bicycle produced higher accelerations at the coccyx and rear hub, whereas during the downhill test, the 29-inch bicycle transmitted greater vibrations to the rider’s wrists and ankles. No significant differences were observed between wheel sizes in the time required to complete either the uphill or downhill tests. These findings suggest that, in short and highly irregular uphill and downhill sections, 29-inch wheels do not necessarily improve performance over 26-inch wheels, although they modify the distribution of vibrations transmitted to the rider. Full article

Background: Falls are the most common hospital-acquired adverse event, leading to extended hospitalization, loss of independence, disability, and premature death. Routine fall risk assessments are time-consuming, even with limited factors. An AI-derived fall prediction model can provide more comprehensive and comparably accurate risk predictions quickly and as often as needed. Objective: To develop and validate a fall prediction model for fall risk in adult inpatients. Methods: Patient records from 2016 were extracted from the adult inpatient database, including information from the Electronic Inpatient Medication Records, SAP, and Hospital Incident Reporting System. The sample consisted of 1506 cases (1:5 faller to non-faller). The fall prediction model was trained using the following four variables: demographics, diagnosis, medications, and surgery. Data sources included the hospital’s data repository, integrating admission/discharge, pharmacy, laboratory, and incident reports. Results: The support vector machine model performed best among all tested models, achieving an AUC of 0.803, recall of 0.816, and precision of 0.440. In the validation cohort (978 patients: 163 fallers and 815 non-fallers), the fall prediction model demonstrated moderate-to-good discrimination (AUC 0.79), with accuracy of 0.67, sensitivity of 0.46, and specificity of 0.86. Compared with the nursing four-item fall risk assessment, which showed lower discrimination (AUC 0.65, accuracy 0.65, sensitivity 0.58, specificity 0.72), the fall prediction model had better specificity and overall discrimination, though the nursing tool was more sensitive in identifying fallers. Conclusions: The fall prediction model using demographics, diagnoses, medication, and surgery data predicts falls risk effectively. It enables timely, accurate risk assessments and supports preventive interventions, saving nurses’ time for direct patient care. Full article

Background/Objectives: Delirium is prevalent among intensive care unit (ICU) patients and is associated with prolonged ventilation, longer ICU stays, increased mortality, and post-discharge cognitive impairment. Because most current prediction tools were developed using European cohorts, their transportability to other ICU settings remains unclear. The device support burden has received limited attention, and preprocessing data leakage remains a secondary methodological concern. Methods: This prospective study enrolled consecutive adults from the Internal Medicine ICU of Ankara Etlik City Hospital. Fifty predictors from electronic health records were documented at ICU admission or within the first 24 h. Device support variables were assessed as a distinct domain using ablation analysis. We developed an elastic net regularized logistic regression model and validated it using 5 × 4-fold nested cross-validation. Preprocessing was restricted to the training folds, and gradient boosting was used as a comparison method. We also evaluated a fixed exploratory 11-predictor reduced model. Results: Delirium was detected in 50.8% of patients. The full regularized model achieved an area under the receiver operating characteristic curve (AUROC) of 0.696 (95% CI: 0.613–0.767). The fixed exploratory reduced model had a higher apparent AUROC of 0.756 (95% CI: 0.684–0.819) than the other models. This apparent advantage was absent when feature selection was repeated within each outer training fold (nested reduced model AUROC 0.700). The full-model calibration slope was 0.741 (95% CI: 0.345–1.265). This confidence interval was wide and included the value of 1.0. The fixed reduced-model slope was 0.235 (95% CI: 0.068–0.886), indicating that the predicted probabilities were too extreme. Ablation analysis showed that device support variables added predictive value to clinical and laboratory variables, whereas environmental variables alone did not significantly discriminate between the two groups. Conclusions: In this single-center Turkish ICU cohort, regularized logistic regression showed moderate performance in predicting delirium. Nested feature selection did not reproduce the fixed reduced model’s apparent discrimination advantage. Device support variables are best interpreted as markers of care complexity rather than direct causal factors. External validation and recalibration are required before clinical use. Because delirium onset and device support start times were unavailable, the model estimated delirium risk across the ICU stay from predictors documented at ICU admission or within the first 24 h and was not time-anchored. Full article

Background: Immune checkpoint inhibitors (ICIs) have transformed the treatment of advanced melanoma, but the comparative cardiac safety of cytotoxic T-lymphocyte-associated antigen 4 (CTLA-4)-containing regimens versus programmed cell death protein 1 (PD-1)-only regimens in real-world populations remains poorly characterized. We compared cardiac outcomes between melanoma patients exposed to CTLA-4 blockade (ipilimumab) and those treated with anti-PD-1 therapy alone (nivolumab or pembrolizumab) using a large multicenter electronic health record database. Methods: This retrospective cohort study used de-identified data from the TriNetX Research Network (111 US healthcare organizations). Adults with melanoma (ICD-10-CM C43) who received ipilimumab (CTLA-4 exposed) were compared with those who received nivolumab or pembrolizumab without ipilimumab (anti–PD-1 only); both groups were therefore treated with immune checkpoint inhibitors, isolating the effect of CTLA-4 exposure. Propensity score matching (1:1) balanced demographics, cardiovascular comorbidities, and baseline antiarrhythmic use. The primary outcome was a cardiac conduction disorder composite; outcomes were assessed at 1 and 3 years. Results: Of 7313 CTLA-4-exposed and 11,481 anti-PD-1-only patients, 6841 matched pairs were analyzed (all standardized mean differences <0.03). CTLA-4 exposure was associated with a significantly higher incidence of cardiac conduction disorders that was already present at 1 year (3.8% vs. 2.1%; RR 1.82; 95% CI 1.48–2.24; p < 0.001) and persisted at 3 years (5.1% vs. 3.7%; RR 1.36; 95% CI 1.16–1.60; p < 0.001). Atrioventricular block was higher at both 1 year (RR 2.10; 95% CI 1.59–2.78) and 3 years (RR 1.49; 95% CI 1.20–1.85), and complete heart block was markedly increased at 3 years (0.5% vs. 0.2%; RR 3.28; 95% CI 1.67–6.43; p < 0.001). Heart failure was modestly higher with CTLA-4 exposure at both timepoints (1-year RR 1.35; 3-year RR 1.17). Conclusions: Among melanoma patients treated with immune checkpoint inhibitors, CTLA-4-containing therapy is associated with a higher burden of cardiac conduction disorders, including a roughly three-fold excess of complete heart block, that is evident within the first year and sustained thereafter. Because the CTLA-4-exposed cohort comprised both ipilimumab monotherapy and nivolumab plus ipilimumab and the subgroup analyses localized the excess risk to the combination regimen, this association reflects CTLA-4-containing (predominantly combination) therapy rather than ipilimumab monotherapy in isolation. These findings support electrocardiographic surveillance beginning during, not only after, treatment for patients receiving CTLA-4-containing immunotherapy. Full article

Background/Objectives: Birth size is a population-level marker of fetal growth, maternal health, and perinatal care. In Kazakhstan, long-term national and regional patterns in newborn anthropometry have not been jointly examined across birth weight, birth length, the Rohrer index, low birth weight (LBW), and macrosomia. This study aimed to quantify national annual trends in these five indicators for 2013–2022, compare region-specific trajectories, and examine whether the two ends of the birth-weight distribution changed in different directions. Methods: This retrospective ecological time-trend study used de-identified individual newborn records (birth weight, birth length, sex, birth date, and region; 3,483,810 live births, 2013–2022) from the Electronic Register of Inpatients of the Ministry of Health of the Republic of Kazakhstan. These individual-level records were used to calculate the Rohrer index and to apply exclusion criteria at the newborn level, but national trend models were deliberately fitted on the 10 annual observations, and regional heterogeneity was tested with a single region × year interaction model rather than individual births or independent per-region regressions to avoid pseudoreplication. Results: Mean birth weight increased from 3429.5 g (2013) to 3518.4 g (2022), and mean birth length increased from 52.5 to 53.3 cm. Recorded LBW prevalence declined from 3.09% to 1.46%, while macrosomia increased from 12.59% to 16.02%. Pooled regional birth weight ranged from 3392.6 g to 3548.6 g, LBW from 0.9% to 5.3%, and macrosomia from 11.7% to 18.1%; region-specific trends after false-discovery-rate correction showed that only North Kazakhstan Region had a statistically robust increase in LBW. Conclusions: Official registry data for Kazakhstan, analyzed at the individual level and aggregated for trend modeling, showed a decade-long increase in mean birth weight and length, an observed decline in recorded LBW prevalence, and an increase in macrosomia prevalence, with statistically confirmed heterogeneity in regional trajectories. Because this study lacked gestational age or maternal characteristics, these findings should be interpreted as ecological surveillance patterns that support continued monitoring of both ends of the birth-weight distribution, rather than as evidence of individual-level changes in fetal growth or maternal metabolic risk. Full article

Background: CMV is the most common congenital infection worldwide, affecting 0.2–6% of live births and causing significant morbidity and mortality. Although most infections are asymptomatic, severe cases can lead to fetal demise. Objective: To quantify and characterize confirmed CMV infection among fetal demises in a large pediatric hospital in the US from 2013 to 2024. Methods: A retrospective electronic health record (EHR) review of pregnancies with CMV testing, delivery summaries, and confirmed fetal CMV infection via tissue or autopsy linkage. Results: During the review period, 61,973 pregnancies were recorded, with 1173 resulting in fetal demise. Among 1670 pregnancies tested for CMV, 859 had positive testing (e.g., CMV IgG, CMV IgM, CMV DNA), including 96 demises, 94 (11%) of which we had complete delivery records. Four (4%) of these fetal demises were directly attributable to CMV based on positive CMV testing from amniocentesis, fetal tissue, or placenta. Most demises occurred in the second or third trimester (mean gestational age 28.4 ± 6.6 weeks), while the four CMV-attributable cases averaged 23.7 ± 3.7 weeks. Among 93 fetal demise pregnancies in which maternal CMV IgM testing was performed, 9 (9.7%) had a positive CMV IgM result. All confirmed CMV-attributable fetal demises occurred among these CMV IgM-positive pregnancies. Conclusion: These findings underscore CMV’s role in perinatal mortality and highlight the need for improved screening and prevention strategies. Full article

Background/Objectives: Cystic echinococcosis (CE) remains endemic in many regions worldwide, including Jordan. It is one of the three zoonotic infections caused by the tapeworm of the genus Echinococcus. The epidemiology of CE in Jordan remains poorly defined, with systematic national investigations being absent for the past three decades. International Classification of Diseases (ICD) is a universal coding system for disease archiving within healthcare facilities. This coding system is utilized by the Hakeem program, a health informatics system that includes all Ministry of Health (MOH) hospitals and medical centers in Jordan. In this study, we assessed the frequency, plausibility, internal consistency, and administrative limitations of CE-related ICD codes within the MOH database, which have never been tested, and determined the validity of using the administrative data for epidemiological studies of CE in the future. Methods: A retrospective database study was conducted, using 12 echinococcosis-related ICD codes for electronic health record review. Data were retrieved from the Hakeem program. Unique patient entries recorded between 2010 and 2024 were extracted, and descriptive analytical methods were applied. Results: Over a 15-year period, 45,702 echinococcosis-related entries were reported across 178 healthcare facilities, with Amman accounting for the majority (59.1%) of reported entries. Entry numbers rose in the first 10 years (2010–2019) before a measurable decline and fluctuations in the following 5 years (2020–2024). Entries spanned all age groups, with middle-aged adults being the most frequently reported age group (42.45%). Females accounted for 74.4% of entries, reflecting a marked female predominance among coded entries. The most commonly reported ICD code was B67.8 (unspecified echinococcosis of the liver, 64.22%), followed by B67.5 (Echinococcus multilocularis infection of the liver, 33.57%). Confirmatory diagnostic data could not be extracted from the coding system. As a result, clinical validation could not be performed. Conclusions: The high frequency of echinococcosis-related entries suggests artificial inflation of CE-related codes, along with other data anomalies, reflecting internal inconsistencies and epidemiological implausibility of the administrative data within the Hakeem program. MOH should implement a dual, system-user-level plan to improve CE-related coding in the future. Full article

Background and Objective: Accurate medication allergy documentation is essential for patient safety. This study evaluated discrepancies and documentation quality between medication allergy records in the primary care information system (AP-Madrid) and the Hospital Clinical Information System (HCIS) at a tertiary care hospital. Methods: A retrospective cohort study included 223 patients with medication allergy records during 2022. Agreement between systems was assessed using Cohen’s kappa, and documentation quality was classified as adequate, moderate, or poor based on four predefined parameters. Results: Agreement between AP-Madrid and HCIS was poor (Cohen’s κ = −0.25). Of the 223 patients, medication allergy records were documented exclusively in HCIS for 94 patients, exclusively in AP-Madrid for 36 patients, and in both systems for 88 patients, while 5 patients had medication allergy information identified only in free-text clinical documentation. Documentation quality was predominantly poor in both HCIS (60.4%, n = 134) and AP-Madrid (61.9%, n = 138). Beta-lactam antibiotics and non-steroidal anti-inflammatory drugs (NSAIDs) were the most frequently implicated therapeutic groups. Two positive medication re-exposure events were identified. Conclusions: Substantial discrepancies and poor documentation quality were identified between primary care and hospital medication allergy records, highlighting limited interoperability between the two systems. Improving the completeness, consistency, and standardized exchange of medication allergy information across healthcare settings may enhance patient safety and continuity of care. Full article

Home enteral nutrition (HEN) is essential for patients with a functioning gastrointestinal tract who cannot meet nutritional needs orally, yet outpatient management remains complex, resource-intensive, and supported by a limited evidence base. Artificial intelligence (AI) may augment HEN care by extending monitoring, education, risk assessment, documentation support, and operational coordination into the home environment. Consistent with the narrative review format, this article uses a pragmatic, transparent synthesis of influential HEN-specific literature, relevant clinical nutrition evidence, and background knowledge from adjacent fields, including home healthcare, chronic disease management, oncology nutrition, telehealth, and software regulation. Direct evidence in established HEN populations remains scarce; therefore, most AI applications should be considered hypotheses or early implementation opportunities rather than proven standards of care. The strongest near-term opportunities are clinician-supervised patient education, symptom triage, adherence support, remote monitoring, and workflow automation. Predictive analytics, smart pumps, and precision enteral prescription tools are promising but require prospective HEN-specific validation, interoperability with electronic health records and home-infusion systems, reimbursement pathways, and governance safeguards. Key barriers include dataset bias, limited external validation, alert fatigue, privacy and regulatory concerns, unclear accountability, digital equity, cost uncertainty, and the risk of dehumanizing care. AI should be viewed as a complement to multidisciplinary HEN expertise. Priorities for the near future include HEN registries, standardized outcomes, prospective validation, pragmatic implementation trials, health-economic evaluation, and transparent oversight that preserves clinician accountability and patient-centered care. Full article

Background/Objectives: Childhood-onset systemic lupus erythematosus (cSLE) is a chronic, multisystem autoimmune disease that is associated with more severe organ involvement, more intensive drug therapy, and increased long-term organ damage compared with adult-onset disease. The objectives of this study were to evaluate the performance of widely used small area-level multidimensional indicators of neighborhood disadvantage in a mixed urban–rural cSLE cohort against disease outcomes. Methods: This retrospective cohort study utilized electronic health records to identify and evaluate pediatric patients with cSLE across a single center in South Carolina between 1 January 2020 and 31 December 2024. Primary outcomes included disease activity at diagnosis, measured by Systemic Lupus Erythematosus Disease Activity Index 2000 (SLEDAI-2K); the achievement of a low lupus disease activity state (LLDAS) by the last visit; the development of major organ involvement; and the rates of unplanned hospitalizations and emergency department visits. The associations of the census tract-level Area Deprivation Index (ADI), Social Vulnerability Index (SVI), and Childhood Opportunity Index (COI) with clinical presentation and outcomes were estimated using generalized linear models and logistic regression. Results: A total of 85 patients with cSLE were included, of which 76% reported being of Black race, and 28% lived in rural areas. Lupus disease severity was inconsistent across the metrics of area-level social vulnerability, neighborhood deprivation, child opportunity, and rurality. Patients who lived in more socially vulnerable communities were more likely to attain LLDAS during follow-up, while those who lived in lower-opportunity areas were more likely to develop CNS lupus. Baseline disease activity, renal involvement, and healthcare use were not significantly associated with area-level social disadvantages or rurality. Conclusions: In this single-center, mixed urban–rural cohort of children with cSLE, indicators of neighborhood-level disadvantage and rurality alone explained little variation in disease severity and disease control overall. The utility of available small area-level metrics is likely context-dependent and influenced by regional features, population characteristics, and the outcomes being studied. Future work should integrate individual- and area-level factors across diverse settings to better understand the drivers of health disparities. Full article

Background: Artificial intelligence (AI) and machine-learning (ML) prognostic models are increasingly developed for inflammatory bowel disease (IBD), yet their reported performance and clinical readiness remain inadequately appraised. Methods: Following PRISMA 2020 and a registered protocol, we searched PubMed, Web of Science, IEEE Xplore, and arXiv (January 2012–January 2026) for studies developing or validating prognostic models in Crohn’s disease or ulcerative colitis. Two reviewers independently screened the studies, extracted data, and assessed risk of bias using PROBAST + AI; discrimination was summarized by area under the curve (AUC) and stratified by validation type. Results: Of the 3050 records, 111 studies were included. Treatment response was the most common target; laboratory data and electronic health records were the most frequent modalities. Across 83 studies, the median AUC was 0.850; externally validated models reached 0.870 versus 0.845 for internal-only and 0.790 for cross-validation-only. External validation was reported in 29.7%, calibration in 14.4% and analysis code in 3.6%; the analysis domain was the leading source of bias. Conclusions: The evidence base maps reported discrimination rather than demonstrated clinical readiness. Until calibration, decision-curve utility, and transportability are reported alongside external validation, clinical deployment remains premature. Full article

Background/Objectives: To describe therapy-associated oral stomatitis and perianal mucosal lesions in pediatric oncohematology patients and compare healing time according to the treatment approach used in routine clinical practice. Methods: This retrospective observational study was conducted at Meyer Children’s Hospital, Florence, Italy. Electronic health records of patients admitted to the Oncology or Bone Marrow Transplantation units between 1 January and 31 December 2024, were screened. Eligible patients were aged 0 to 18 years and had documented oral stomatitis and/or perianal mucosal lesions with sufficient clinical and nursing documentation to determine lesion localization, treatment, clinical evolution, and healing time. Demographic, clinical, pharmacological, nutritional, and nursing variables were extracted. Oral mucositis severity was classified using the World Health Organization Oral Toxicity Scale. Results: Among 525 screened records, 89 eligible patients aged 5 months to 18 years were included in the final analysis. Lesions were distributed across three mutually exclusive categories: oral-only (39.3%), perianal-only (42.7%), and combined oral–perianal involvement (18.0%). Most oral cases were grade 2 to 4. Oral stomatitis treated with Mucosamin^® Spray healed faster than conventional therapy (3.5 ± 0.8 vs. 6.7 ± 1.1 days; mean difference, −3.21 days; 95% CI, −3.75 to −2.67; p < 0.001). Perianal fissures treated with Mucosamin^® Rectal Gel also showed a shorter healing time than conventional therapy (3.4 ± 0.8 vs. 6.1 ± 0.8 days; mean difference, −2.76 days; 95% CI, −3.30 to −2.21; p < 0.001). Conclusions: Mucosamin^® Spray and Rectal Gel were associated with shorter documented healing times than conventional approaches in this real-world pediatric cohort. These findings should be interpreted as preliminary and confirmed in controlled prospective multicenter studies. Full article

Trichostrongylid nematodes are globally distributed gastrointestinal parasites of major veterinary and, in some cases, zoonotic importance. Accurate identification often requires an integrative approach because morphology alone may be compromised by specimen damage, cryptic diversity, and dimorphic species complexes. In the Philippines previous work has been largely limited to coproscopy, with no published DNA-based studies on adult trichostrongylids or scanning electron microscopy (SEM) documentation. Here, we provide an integrative characterization of Haemonchus contortus and Trichostrongylus axei recovered from a single goat in Mindanao, Philippines, in 2023. Identification was based on light microscopy, SEM, morphometry, and sequence data from the cox1 and ITS regions. Phylogenetic analyses and a haplotype network for H. contortus placed the Philippine material within a widely distributed lineage complex with weak geographic structuring. Because the study is based on parasites from a single host, the findings should be interpreted as a taxonomic and faunistic record rather than evidence for broader epidemiological or population-level patterns. This study provides the first molecular record of trichostrongylid nematodes from the Philippines and the first SEM-based documentation of H. contortus in the country. Broader sampling is needed to assess parasite diversity, distribution, genetic variation, and potential veterinary and public health relevance nationwide. Full article

Background: Artificial intelligence has significantly improved the diagnosis of retinal diseases using fundus and optical coherence tomography. However, the pathway between the accuracy of the models and their clinical application is still unclear. Objective: The objective of this systematic review is to overcome this gap by correlating the quality of methodology with clinical preparation in the current studies and to critically examine the translation readiness possibility. Methods: The PRISMA-based review clearly addresses the six research questions in an organized manner. A comprehensive search is conducted across six electronic databases, resulting in a total of 883 records. After rigorous screening, a final set of 43 fundus- and OCT-based articles published between 2020 and August 2025 that passed a comprehensive eligibility criterion was selected. The selected studies are assessed using the custom AI-specific risk-of-bias assessment tool and the CLAIM checklist to measure the quality of technical reporting. The reviewer’s agreement is calculated using Cohen’s kappa, interrater reliability, and the intraclass correlation coefficient. Results: Although convolutional and hybrid deep-learning-based studies are often reported to have an accuracy of more than 95.0%, only 14.3% are truly clinically validated, and 23.1% do not include code or implementation data details. Reviewers’ agreement scores are from moderate to high (fundus-based ICC(2, 1) = 0.911, CI = [0.896, 0.923]) and (OCT based ICC(2, 1) = 0.643, CI = [0.422, 0.792]), supporting the reliability of the quality assessments. In order to combine these dimensions, we come up with a composite Technical-Clinical Integrity Index, which measures data transparency, rigor of validation, and reporting of performance. Conclusions: Retinal disease diagnosis has achieved high performance with low clinical transfer. This review provides the guidelines of a Translational Readiness Framework that can be used by researchers, clinicians, and health analytics teams to assess and implement credible artificial intelligence systems in ophthalmology. Full article