Search PubMed⌕ Search

SEARCH · Search PubMed

Results for “data disaggregation”

Search indexed PubMed citations on genomics, clinical trials, systematic reviews and public health. Explore titles, authors and supplied subject terms, then open the PubMed record.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

788 recordsLinked to original sources
PubMed2026

Cancer statistics for Asian American, Native Hawaiian, and Pacific Islander people, 2026.

BACKGROUND: Cancer statistics for Asian American and Native Hawaiian and Pacific Islander (NHPI) people are usually aggregated, masking substantial variation within this heterogeneous population. Herein, the American Cancer Society reports cancer incidence and survival for 8 Asian American and 3 NHPI ethnic groups. METHODS: The authors used population-based cancer registry data from the National Cancer Institute's Surveillance, Epidemiology, and End Results program, for Asian American and NHPI ethnic groups from 2000 through 2022. RESULTS: During 2018-2022, overall cancer incidence ranged from 218.3 per 100,000 Kampuchean people to 474.5 per 100,000 Native Hawaiian people, which was 1.5 times higher than the rate for the aggregated Asian American and NHPI population (307.3 per 100,000). High incidence among Native Hawaiian people is largely driven by the highest rates of female breast, colorectal, and prostate cancers, whereas infection-related cancers were highest among Asian American ethnic groups. For example, liver and stomach cancer incidence is highest among Vietnamese (22.2 per 100,000) and Korean people (17.8 per 100,000), respectively, both of which were nearly twice that in Native Hawaiian people (12.9 and 9.6 per 100,000, respectively). Native Hawaiian and Samoan women are twice and 3 times as likely, respectively, to be diagnosed with uterine corpus cancer as aggregated Asian American and NHPI women or White women. Five-year relative survival ranges from 42% in Laotians to 74% in Asian Indians/Pakistanis, with largest differences for colorectal (43% in Laotians to 72% in Asian Indians/Pakistanis) and prostate (63% in Kampucheans to 97% in Japanese) cancers. CONCLUSIONS: Wide variation in cancer risk within the Asian American and NHPI population highlights the critical need for disaggregated data to effectively target cancer prevention and control interventions.

Adolescent↗
PubMed2026

Sex-stratified mortality trends in preterm birth complications in Sierra Leone: progress, persistence, and equity implications.

BACKGROUND: Preterm birth complications remain a leading cause of neonatal mortality in Sierra Leone, despite recent health system gains. Evidence on long-term sex-specific disparities in mortality due to preterm birth complications is limited, constraining equitable neonatal care planning. OBJECTIVE: To examine two‑decade trends in sex‑stratified mortality from preterm birth complications using standardized equity indicators. METHODS: We conducted a retrospective longitudinal analysis of sex-disaggregated mortality estimates from the World Health Organization (WHO) Global Health Estimates (GHE), accessed through the WHO Health Equity Assessment Toolkit (HEAT), Built-in Database Edition (Version 6.0). Mortality rates per 100,000 population were extracted for 2001, 2006, 2011, 2016, and 2021. Inequality was assessed using absolute difference (D), relative ratio (R), population attributable risk (PAR), and population attributable fraction (PAF). RESULTS: Mortality declined substantially between 2001 and 2021 for both males (85.1-49.3 per 100,000) and females (71.2-39.9 per 100,000). Male mortality remained consistently higher across all years, with relative ratios indicating approximately 20-25% excess mortality among male neonates. Absolute inequalities narrowed modestly over time, whereas relative inequalities remained largely unchanged. PAR and PAF remained close to zero throughout the study period. Wider uncertainty intervals in earlier years reflected limited empirical data availability. CONCLUSION: Although preterm mortality declined over two decades, a persistent male disadvantage remained in Sierra Leone. These findings highlight the importance of integrating sex-disaggregated equity monitoring into neonatal policies and programmes. Future research should evaluate strategies to reduce the persistent excess mortality among male neonates while sustaining overall improvements in neonatal survival and progress toward Sustainable Development Goal 3.2.

Humans↗
PubMed2026

Access to maternity services for women asylum seekers and refugees: A transnational document analysis of international, European regional, and United Kingdom governance.

Women asylum seekers and refugees face persistent barriers to maternity care (antenatal, intrapartum and postnatal care) across high-income countries, yet the upstream governance shaping access remains under-examined. Although legally distinct, both groups share protection-seeking experiences and are addressed jointly in governance documents. This study examined and synthesised how international (macro), European regional (meso), and United Kingdom (UK, micro) governance documents frame and operationalise maternity service access. Sixty-four documents were analysed using the READ framework. Inductive analysis of macro and meso documents identified six access dimensions: universal coverage; cultural and linguistic adaptation; rights-based approaches; multi-agency collaboration; data, monitoring and accountability; and quality of care. These dimensions structured assessment of UK governance, with jurisdictions rated strong, moderate or weak. Alignment was fragmented: Wales, Scotland and Northern Ireland exempted asylum seekers from charging, whereas England retained charging provisions. Multi-agency collaboration was consistently articulated, yet none of the 35 UK government documents focused on maternity access for this population, and none required outcome monitoring disaggregated by asylum or refugee status. UK governance appears coordinated in form but fragmented in substance. UK-wide minimum standards and routine recording of these data, with safeguards against immigration-related use, could strengthen coherence and accountability and improve visibility of inequities.

Refugees↗
PubMed2026

A narrative systematic review of definitions and diagnostic criteria for disordered eating and eating disorders in type 1 diabetes.

AIMS/HYPOTHESIS: Type 1 diabetes and disordered eating (T1DE) affects 8-37.1% of adults and is associated with high rates of morbidity and mortality. The absence of a standardised case definition of T1DE and its severity hinders effective screening, diagnosis and treatment. This systematic review aimed to (1) synthesise existing case definitions and diagnostic criteria for T1DE in adults and (2) identify key characteristics to inform consensus for future diagnostic criteria. METHODS: A systematic review was conducted following the Preferred Reporting Items for Systematic reviews and Meta-Analysis (PRISMA) guidelines. Eligible studies involved adults (≥18 years) with type 1 diabetes assessing disordered eating; paediatric studies, mixed samples without disaggregated data, non-empirical designs and non-English publications were excluded. PubMed, MEDLINE, EMBASE, CINAHL and PsycINFO were searched up to November 2025 for peer-reviewed studies involving adults with T1DE. Qualitative and quantitative data on definitions, diagnostic criteria and assessment tools were extracted. Study quality was appraised using a modified Graphical Appraisal Tool for Epidemiological studies (GATE) checklist. Due to heterogeneity of data, a narrative synthesis of findings was performed to describe current definitions of T1DE. RESULTS: Sixty-one studies met the inclusion criteria, with a pooled sample of 111,208 participants (76% women) from over 22 countries. T1DE was defined using a heterogeneous array of terms, diagnostic frameworks and assessment tools (29 distinct methods). The Diabetes Eating Problem Survey-Revised (DEPS-R) was the most used questionnaire, but many studies relied on criteria adapted from general eating disorder classifications or generic questionnaires. Approximately three-quarters of the studies assessed insulin omission behaviours, but the operationalisation of the cognitions for insulin omission varied widely. Beyond physiological markers such as HbA1c and BMI, studies explored various diabetes-related and psychological constructs, although often considering diabetes and disordered eating separately rather than as an integrated condition. CONCLUSIONS/INTERPRETATION: This systematic review highlights the lack of a unified, evidence-based definition of T1DE, resulting in inconsistent screening, diagnostic and reporting practices. Establishing clear, consistent, evidence-based diagnostic criteria and screening questionnaires for T1DE is critical to improving early detection and developing targeted interventions. These findings provide a foundation for refining T1DE definitions as a stepping stone to an international consensus definition. STUDY REGISTRATION: PROSPERO registration no. CRD420250223622 FUNDING: King's College London and King's College Hospital through the KMRT KCH Joint Research Committee studentship. This work was also conducted as part of the National Institute for Health Research (NIHR; CS-2017-17-023)-funded STEADY project (Safe management of people with Type 1 diabetes and EAting Disorders studY). NZ's salary was part-funded by the NIHR via the NIHR Clinician Scientist award to MS; JT and KI are part-funded by the NIHR Mental Health Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King's College London. MS was funded through her NIHR Clinician Scientist Fellowship (CS-2017-17-023).

Humans↗
PubMed2026

An introductory practical guide to secondary data analysis in pediatric urology.

INTRODUCTION: Secondary data analysis (SDA) has become an increasingly important approach in pediatric urology, enabling the study of long-term outcomes, care variation, and disparities in populations with chronic or congenital urologic conditions. With the growing availability of large datasets, a structured approach to designing and conducting SDA studies is increasingly relevant. OBJECTIVES: To provide an introductory, practical guide to SDA in pediatric urology by (1) summarizing commonly used data sources with representative studies, (2) outlining a stepwise approach to designing and executing SDA studies, and (3) highlighting key methodological considerations, limitations, and opportunities for future work. STUDY DESIGN: Narrative review of existing literature and commonly used datasets relevant to pediatric urology, including administrative claims, hospital encounter databases, clinical registries, electronic health record networks, and population-based surveys. RESULTS: Data sources differ in scope, clinical granularity, longitudinal follow-up, and representativeness, and each is suited to specific research questions. We present a practical workflow for SDA, including dataset selection, cohort definition, and analytic planning. Linkage across datasets can provide a more comprehensive view of care patterns and outcomes, although feasibility is influenced by legal, technical, and data-quality constraints. DISCUSSION: SDA enables population-level analyses and the study of rare conditions that are challenging to evaluate through single-center or prospective designs. However, careful cohort definition, feasibility assessment, and awareness of data limitations are essential to ensure validity and interpretability. CONCLUSION: SDA provides a scalable, cost-efficient framework for generating meaningful evidence in pediatric urology. Continued efforts to harmonize data elements, improve linkage infrastructure, and support cross-institution collaboration will enhance the quality and impact of future research. This article provides a practical framework and examples to support the design and execution of SDA studies.

Humans↗
PubMed2026

Computational metabolomics at scale: from open data to insight.

Metabolomics data are currently generated at scale thanks to the evolution of technologies that have led to marked improvements in the number of metabolites detected, spanning all chemical classes. These data are increasingly submitted to public repositories for data reuse, integration, and interpretation. Despite the availability of public resources and associated computational tools, the field still lacks a widely adopted, consistent data and analytics infrastructure capable of transforming this wealth of information into scientific insight. Indeed, the metabolomics field is just now scratching the surface of being able to harness the power of new computational technologies. In this review, we summarize discussions from the "Dagstuhl-Seminar 24181 Computational Metabolomics: Towards Molecules, Models, and their Meaning" with a focus on public data availability, open data standards, data and knowledge integration, and education. Our goal is to raise awareness and adoption of the latest open science resources while highlighting key areas needing further development.

Metabolomics↗
PubMed2026

Assessing the public health impact of routinely collected electronic healthcare record data in NICE guidelines: A systematic review of CPRD research.

OBJECTIVES: Evidence used in NICE guidance has traditionally prioritised randomised controlled trials, but increasing availability of electronic health record (EHR) data has expanded opportunities for real-world evidence. The Clinical Practice Research Datalink (CPRD) is a commonly used UK primary care EHR resource, yet the extent to which CPRD studies have informed NICE guidelines in the past decade is unclear. STUDY DESIGN: The systematic review was conducted in accordance with PRISMA guidelines. METHODS: We conducted a systematic review of CPRD studies in PubMed, MEDLINE, and Embase published between 04/16-09/25. For each eligible CPRD study, targeted searches of NICE guidelines were performed to identify explicit citations in NICE guidelines. Two reviewers screened and extracted data independently, resolving disagreements by consensus or third reviewer. Guideline information, number of guidelines over time, type of guidelines, and disease area guidelines (using British National Formulary (BNF) chapters) were described. RESULTS: 7181 records were identified. After de-duplication, 2704 unique CPRD studies were screened against NICE guidelines. Of these, 92 CPRD-based studies met inclusion criteria and were cited across 67 NICE documents. The annual number of NICE guidelines citing CPRD studies increased between 2016 and 2025; 1.5% of identified guidelines published in 2016 and 27.7% in 2025. The guideline citing the most CPRD studies was cancer related. The most common types of guidelines included clinical guidelines (49.3%) and technology appraisals (32.8%). Guidelines made up 12 different BNF categories, most frequently central nervous system related (23.9%; n = 16). CONCLUSION: Observational CPRD studies are increasingly referenced in NICE guidelines across multiple disease areas, supporting the growing role of EHR data in national guideline development.

Clinical studies↗
PubMed2026

Data-centric, robust, and explainable multimodal deep learning for clinical decision support: A systematic review.

PURPOSE: Multimodal deep learning is increasingly proposed for clinical decision support (CDS) under a "data-centric" framing that prioritizes label quality, missing-modality robustness, distribution shift, calibration, and explainability. Prior reviews have examined multimodal medical AI, CDS, and data-centric methods separately, but none address their intersection. We mapped the modalities, fusion strategies, and data-centric and explainability techniques used in this recent literature, quantified how often each is implemented rather than merely mentioned, assessed deployment-relevant evidence (external validation, clinical-outcome measurement, equity), and formally appraised study-level risk of bias. METHODS: Following the PRISMA 2020 statement (PROSPERO CRD420261427815; registered retrospectively), we screened 150 records and included primary, clinical, multimodal studies that applied machine or deep learning to a decision-support task and reported at least one quantitative result. Two reviewers screened and extracted data with consensus adjudication. Each study was coded against pre-specified operational definitions, separating implemented or empirically evaluated techniques from those only mentioned. Study-level risk of bias was assessed with PROBAST + AI. Synthesis was narrative. RESULTS: Thirty-one studies met inclusion; 30 (97%) were published between 2024 and 2026, with a median of three modalities (range 2-6), most commonly structured EHR (71%) and imaging (39%). Data-centric techniques were frequently reported (74-84% across label-noise, distribution-shift, calibration, missing-modality and class-imbalance handling; equity 61%). However, external validation was reported in only 4/31 studies (13%), a clinical or provider outcome in 3/31 (10%), and no study reported routine deployment. Overall risk of bias was high in 27/31 studies (87%), driven by the analysis domain. CONCLUSION: Within this recent, self-selected slice of the field, technical robustness and explainability techniques are widely reported but rarely validated out-of-distribution or against clinical outcomes, and the underlying evidence is at high risk of bias. Progress requires external multi-site validation, clinical-outcome measurement, formal bias appraisal, and adherence to AI reporting standards (e.g., TRIPOD + AI) before deployment can be justified.

Deep Learning↗
PubMed2026

Recovering membrane interaction kinetics of single molecules from 3D tracking data.

Interactions between cytosolic biomolecules and the bacterial inner membrane are fundamental to many cellular processes, yet directly measuring their binding kinetics in living cells remains challenging. Conventional 2D single-molecule tracking analyses can be insufficient, particularly when membrane association does not markedly alter the diffusion rate. Here, we present a method to recover membrane interaction kinetics from 3D single-molecule trajectories in rod-shaped bacteria. Using simulated 3D tracking data, we identify membrane-associated motion by quantifying how well short trajectory segments follow the circular curvature of the cell membrane. The resulting measure is further analyzed using a hidden Markov modeling framework, enabling robust discrimination between cytosolic and membrane-bound states and capturing the dynamics of state transitions without requiring diffusion-rate changes or direct colocalization with membrane markers. This work establishes a general framework for extracting membrane interaction kinetics from 3D single-molecule tracking data in live bacteria and highlights the value of realistic microscopy simulations for quantitative interpretation and systematic bias assessment.

Kinetics↗
PubMed2026

Privacy, security, and reliability risks of artificial intelligence in healthcare: a systematic review of empirical evidence.

BACKGROUND: Artificial intelligence (AI) is increasingly integrated into healthcare information systems, supporting clinical decision-making, imaging analysis, and predictive modeling. While these applications offer operational and clinical benefits, they also introduce emerging risks to patient privacy, data security, and system reliability. OBJECTIVE: To systematically review empirical evidence on privacy breaches, security vulnerabilities, and misuse associated with AI applications in healthcare settings. METHODS: PubMed, Embase, Web of Science, Scopus, IEEE Xplore, and ACM Digital Library were searched for empirical studies published between January 2015 and November 2025 that evaluated AI use or misuse in clinical diagnosis, treatment, or decision-making. Two reviewers independently screened studies and extracted data using a standardized form. Findings were synthesized narratively due to heterogeneity in study designs, AI methods, and reported outcomes. RESULTS: Of 7,285 records identified through database searches and 205 through citation screening, 22 empirical studies met the inclusion criteria, spanning multiple clinical domains and data modalities, predominantly medical imaging applications. Five recurring threat categories were identified: patient re-identification, membership inference, unauthorized access and adversarial exploitation, input manipulation, and misuse or overinterpretation of AI outputs. Across studies, AI models were shown to encode latent biometric signals across diverse data types, limiting the effectiveness of traditional anonymization and synthetic data approaches. Adversarial attacks and input manipulation were also shown to compromise diagnostic performance and system integrity. CONCLUSION: This systematic review provides empirical evidence suggesting that contemporary AI systems in healthcare introduce privacy and security risks that may challenge traditional assumptions about data protection. These findings underscore the need for privacy- and security-by-design approaches and governance frameworks that address risks across the AI lifecycle.

Humans↗
PubMed2026

Effects of H-coil TMS on suicidality in major depression: A secondary analysis of data from a multisite randomized trial comparing accelerated to once-a-day stimulation.

Suicide is the 10th leading cause of death in US adults. Standard once-daily repetitive transcranial magnetic stimulation (rTMS) can reduce suicidal ideation. Yet, antidepressant and anti-suicidal effects often take several weeks to emerge, while rapid improvement is often required. Accelerated TMS has been proposed as a strategy to hasten therapeutic response. A recent FDA-regulated multicenter trial evaluated accelerated intermittent theta burst Deep TMS with the H1-coil versus standard high-frequency Deep TMS in MDD. Both groups demonstrated high remission and response rates for depression, with the accelerated protocol showing non-inferiority and a shorter time to remission. The goal of this exploratory secondary analysis was to evaluate the impact of these two H-coil TMS dosing paradigms on suicidal ideation. The Scale for Suicide Ideation (SSI), as well as suicidality items of HDRS, MADRS and CUDOS were collected and analyzed. On all scales, both accelerated and standard Deep TMS protocols were associated with meaningful reductions in suicidality. The accelerated protocol achieved a faster onset of improvement. Comparison between the timeline of improvement in suicidality and in overall depressive symptoms found a trend for faster improvement in suicidality, especially with the accelerated protocol. These findings highlight the importance of treatment frequency in determining time to clinical benefit and support the use of scalable accelerated protocols for patients requiring more rapid symptom relief.

Humans↗
PubMed2026

Building phenotypic character matrices for phylogenetic inference: exploration of 35 years of practice.

Recent methodological development in phylogenetic inference has focused predominantly on molecular data. However, renewed interest in other data types, particularly morphological data, has followed from the increased recognition of the power of total evidence and tip-dating approaches, including fossil data, for inference of time-scaled trees and rates of evolution. However, attention has largely focused on the improvement of models of morphological evolution and other analytical tools with much less discussion about data acquisition itself. Here we review past and current practice for describing and collecting morphological data for phylogenetic inference. We present a systematic review of 164 phylogenetic analyses conducted over the last 35 years and focused on a diverse group of extinct arthropods: trilobites. Trends in increasing matrix size, data type, and coding strategy are evident. Where present, polymorphic characters have been predominantly derived from discretized continuous characters, although increasingly practitioners are utilizing alternative approaches for the treatment of quantitative characters. Not surprisingly, traditional indices that describe character consistency are highly correlated with matrix size but show surprising variation at different taxonomic scales. More recent attempts to describe data quality using information theory imply that characters can have high information content even if data are missing for many tips, providing support against the exclusion of characters because of missing data. In consideration of this, as well as advances in the study of developmental biology and variational complexity, we identify several avenues for increasing the quality and quantity of morphological data going forward.

Phylogeny↗
PubMed2026

Relationship Between Number of Acute Pancreatitis Episodes and Risk of New-onset Diabetes in the U.S.: A Real-world Data Analysis.

INTRODUCTION: Acute pancreatitis (AP) is a common inflammatory disorder that is associated with increased risk for diabetes mellitus (DM). It remains unclear whether recurrent acute pancreatitis (RAP) is associated with further increased risk of incident DM. This study aims to investigate the association between RAP and incident DM using real-world data. METHODS: We conducted a retrospective cohort study using the MerativeTM MarketScan&#xae; claims database (2016-2023), identifying patients with AP and no prior history of DM at baseline. The primary exposure of interest, RAP, was defined as one or more episodes of AP occurring &#x2265;90 days after the index AP diagnosis, whereas one episode of AP referred to a single episode of AP (SAP) with no subsequent recurrence within 90 days following the index event. A multivariable stratified Cox proportional hazards regression models were used to determine the association between RAP and incident DM, identified using ICD-10 codes. RESULTS: In total, 16,184 individuals with AP (mean [SD] age: 45.8 [12.3]) contributed 40,712 person-years of follow-up, during which 1,477 incident cases of DM were documented. Individuals with RAP had an increased risk of incident DM compared with those with a SAP(adjusted HR, 1.92; 95% CI, 1.61-2.29). The risk increased significantly with the frequency of RAP. In comparing the modifying effect of patient demographics and comorbidities, a stronger association between RAP and incident DM was observed in females (adjusted HR, 2.44; 95% CI, 1.87-3.19) than in males (adjusted HR, 1.64; 95% CI, 1.30-2.07; Pinteraction=0.03). Also, stronger associations were observed among younger patients (18-46&#xa0;y) (adjusted HR=2.56; 95% CI, 1.97-3.31) and among non-tobacco abuse (adjusted HR=2.19; 95% CI, 1.81-2.65), with significant interactions for all comparisons (Pinteraction<0.05. CONCLUSIONS: In this real-world study, RAP was associated with an increased risk of incident DM. Our findings highlight an opportunity for glycemic monitoring and proactive management of patients with RAP to mitigate their risk of developing DM.

AP↗
PubMed2026

Identification Matters: How Data Sharing Affects Pupil Honesty and Engagement in Universal School Well-Being Assessments.

PURPOSE: Universal well-being assessments in schools may support early identification of pupils needing mental health support. However, little is known about how privacy and confidentiality concerns influence pupils' acceptability of assessments and willingness to engage authentically. This study examined how hypothetical identification, where responses are linked to pupils and shared with key stakeholders, affects pupils' anticipated honesty and engagement, and whether known help-seeking barriers predict negative responses. METHODS: Cross-sectional data were collected from 12,377 primary (ages 8-10) and secondary pupils (ages 11-17) across 55 schools in England. Pupils reported whether their responses would change if identifiable and shared with school staff, parents/guardians, or external professionals. Responses indicating reduced honesty or likelihood of disengagement were coded as negative. Predictors were examined using mixed-effects logistic regression models, including demographics, school connectedness, and mental well-being. RESULTS: Identification and data sharing influenced pupils' anticipated engagement, particularly in secondary schools. Identification by school staff elicited the highest proportion of negative responses in both phases, whereas external professionals elicited the fewest. Most primary pupils reported they would respond authentically, while a larger proportion of secondary pupils indicated they would respond less honestly or disengage when responses were identifiable and shared. Across primary and secondary samples, low well-being, low school connectedness, and being female were associated with greater likelihood of negative response. DISCUSSION: Pupils' anticipated engagement with well-being assessments is shaped by who accesses their data, with marked developmental differences. Strengthening trust, privacy, and connectedness, and supporting pupils' autonomy, may improve the acceptability and response accuracy.

Humans↗
PubMed2026

Reliability-aware hierarchical learning for Chagas disease screening from 12-lead ECGs: tackling label uncertainty and class imbalance.

Objective.Chagas disease, a neglected tropical disease (NTD) with significant cardiovascular impact, remains underdiagnosed in resource-limited regions. Electrocardiogram (ECG) screening offers a low-cost tool for detecting cardiac involvement, yet algorithm development is challenged by label noise, data scarcity, and the latent nature of infection. This study proposes a robust ECG-based screening framework that explicitly addresses these constraints.Approach.We introduce aReliability-Aware Hierarchical Learningstrategy that calibrates supervision according to data provenance, prioritizing serology-confirmed labels over noisy self-reports. To mitigate data scarcity, we compare a specialized convolutional neural network (CNN) trained from scratch with a transfer learning approach based on a Spatio-Temporal ECG foundation Model (FM). Performance is evaluated across varying data scales, and the representation structure is analyzed to interpret model behavior.Main results.On the official hidden test set of the George B. Moody PhysioNet/Computing in Cardiology Challenge 2025, our approach achieved a Challenge Score of 0.163. We observe that while the specialized CNN performs competitively in data-rich regimes, the FM exhibits superior robustness in extreme low-resource settings. Furthermore, performance reaches a plateau imposed by underlying disease physiology. Bimodal score distributions suggest that models distinguish established cardiomyopathy from indeterminate infection, which remains electrophysiologically indistinguishable from healthy controls.Significance.These findings clarify both the potential and intrinsic limits of ECG-based AI screening for NTD-associated cardiac involvement. Reliability-aware supervision and data-efficient transfer learning provide a practical framework toward scalable and clinically meaningful ECG screening systems in resource-constrained environments.

Humans↗