This internal discrepancy limits precision, but the reported direction is consistent: recorded cases rose from roughly 8,000 in 2014 to more than 10,000 in both 2020 and 2023, an increase of well over 25% in less than a decade. The findings are based on registered deaths and reported circumstances rather than clinical assessments, and the authors caution that under-reporting, misclassification and incomplete documentation make the true burden difficult to quantify. The analysis found that females accounted for more than half of the deaths in the result section, with 53,315 female, 47,468 male and four transgender deaths recorded, while the abstract separately gives a total of 100,787.

Female deaths consistently exceeded male deaths from 2015 onward; in 2021, for example, the NCRB recorded 5,655 female and 5,077 male suicide deaths among minors, plus a negligible number of transgender cases. The pattern should not be interpreted as showing that boys face little risk. The article notes that adolescent boys may encounter sexual abuse, stigma and major barriers to seeking help, while girls can face gender-based disadvantage, coerced marriage, early pregnancy and restrictive social expectations.

It also reports broader estimates that India’s suicide rate among people aged 10–24 is 25 per 100,000, compared with a global figure of 13 per 100,000, and says young women have nearly six times the rate reported for young men. These contextual figures were not newly estimated by the present descriptive analysis. Official records listed family-related problems as the most frequently reported circumstance, accounting for 25,049 deaths during 2014–2023.

Examination failure was the largest specifically defined academic cause, with 12,988 deaths, while love affairs were also a prominent category. The body of the paper gives 12,223 deaths for love affairs, although its abstract lists 12,988 for both academic failure and love affairs, creating another discrepancy in the published material. The authors say this pattern should not be read as evidence that psychiatric conditions are unimportant.

Instead, it may reveal how stigma, moral disapproval of suicide, administrative practices and the trivialisation of young people’s distress can obscure mental health problems in official records. To interpret the official categories, the researchers used the Socio-Ecological Model, which places risk and protection across individual, interpersonal, community and societal levels. At the individual level, the framework includes age, gender, personal vulnerability and mental health factors.

Family dynamics, parental conflict, peer relationships and relationships with teachers operate at the interpersonal level, while schools, neighbourhoods and community support systems shape exposure to distress at the community level. Wider determinants include socioeconomic conditions, cultural norms, media portrayals and state-level factors. The study’s design was descriptive: it tabulated age group, gender, cause, year and state or union territory, then organised the findings within this framework.

It did not directly test which factor caused an individual death, compare matched populations, estimate the independent effect of a risk factor or establish the effectiveness of an intervention. The analytical approach is useful for locating potential intervention points, but it does not demonstrate that the listed circumstances caused the deaths. The authors argue that prevention should combine life-skills education, family interventions, school-based mental health support, community capacity building and national policy reform.

They also call for screening in schools and primary care, evidence-based psychological and family interventions, structured postvention after attempts or deaths, and stronger means-restriction measures. India has established several policy and service mechanisms relevant to child and adolescent mental health. The Mental Healthcare Act of 2017 replaced the criminalisation of suicide attempts under Section 309 of the Indian Penal Code and enshrined a right to mental healthcare.

Other measures include the National Mental Health Policy, the District Mental Health Programme, the Rashtriya Kishore Swasthya Karyakram, the School Health Programme under Ayushman Bharat, the National Suicide Prevention Strategy, the Tele-MANAS helpline launched in 2022, and Manodarpan, which provides psychological support in schools. The paper also highlights Karnataka’s SURAKSHA Project, a pilot combining community suicide surveillance with structured post-attempt care. However, child-specific provisions remain uncommon, financing is described as inadequate, and implementation is fragmented, with many promising initiatives remaining small in scale.

Access to specialist care is particularly constrained. The article reports that India has fewer than 0.3 psychiatrists per 100,000 people, with even fewer psychologists, psychiatric social workers and school counsellors, and most professionals concentrated in urban areas. Families in rural communities may face long journeys, transport costs, lost wages and disrupted schooling before specialist care is available.

Primary-care providers can consequently become the only realistic first contact, while teachers, community health workers and local institutions may encounter distress before formal mental health services do. The authors describe Tele-MANAS, online cognitive-behavioural therapy, gatekeeper training and community-based support as promising ways to bridge gaps, particularly in resource-poor settings, but they also stress that evidence on which programmes actually prevent child and adolescent suicide remains limited. Robust surveillance is a central gap in the evidence base.

National statistics do not capture suicide attempts, suicidal ideation or detailed differences among India’s cultural and social groups, while official cause-of-death categories frequently leave little information about mental health, sexual abuse, digital exposure or other contextual risks. The study recommends longitudinal surveillance and self-harm registries that could identify changing patterns of risk and resilience, as well as more rigorous evaluations of school, family and digital interventions. The authors stress the need for culturally relevant programmes co-produced with children, adolescents, families and educators, especially in rural and resource-limited settings.

They also identify LGBTQIA+ young people, adolescents living with mental health or substance use disorders, and those exposed to parental incarceration, sex work or domestic violence as groups requiring more inclusive evidence, without implying that surveillance alone can prevent suicide. For families affected by a death or attempt, the study calls for grief counselling, follow-up care and coordinated school crisis protocols, arguing that structured postvention can help people cope with trauma and reduce contagion. Means restriction is another priority, with the article citing bans on highly hazardous pesticides as having contributed to declines in self-poisoning and recommending stronger implementation.

The central message is not that schools, helplines or families can solve the problem alone, but that coordinated action must connect them with accessible treatment, community surveillance, trained non-specialist workers and sustained policy reform. The research presents suicide among Indian minors as a public mental health crisis shaped by interacting clinical, family, educational and structural pressures. Its descriptive data do not prove why individual deaths occurred or demonstrate that a particular intervention will reduce the total, but they show a large, widening recorded burden, frequent gaps in cause documentation and major shortcomings in the services intended to protect children in distress.

Subject of Research: Descriptive analysis of suicide among Indian children and adolescents using NCRB mortality data and a socio-ecological framework.

Article Title: Suicide among children and adolescents in India as a public mental health priority

Article References: Ramakrishnan, T., Deac, A. A., Vivekanandhan, S., Salisbury, T. T., Ramasamy, D. P., Cherian, A. V., Madegowda, R. K., & Kannuri, N. K. (2026). Suicide among children and adolescents in India as a public mental health priority. Discover Mental Health, 6(1), Article 207. https://doi.org/10.1007/s44192-026-00478-5

Image Credits: AI Generated

DOI: 10.1007/s44192-026-00478-5

Keywords: child suicide, adolescent suicide, India, NCRB data, suicide prevention, mental health services, academic stress, family conflict, data quality, socio-ecological model, Suicide, children

The research, published in Discover Artificial Intelligence, combined audited financial statements and central-bank supervisory information covering 2015–2024 with survey responses collected from chief risk officers and chief information officers in 2024. The initial pool contained 198 institutions. Researchers excluded 23 with incomplete or disrupted financial data, 12 undergoing major mergers, acquisitions or restructurings that would disrupt time-series continuity, and 21 with survey completeness below 80 per cent.

The final sample included commercial banks, specialised lenders, fintechs and insurers across the five countries. The stated observation period captured the gradual implementation of Basel III, the expansion of machine-learning credit scoring and fraud detection, and the COVID-19 disruption of 2020–2021. Despite financial records being available across the decade, the estimation was cross-sectional, using one recent complete observation per institution supplemented by the 2024 survey.

Its dependent variable, PERF_AI, combined two measured dimensions of risk performance: a 60 per cent-weighted reduction in the non-performing-loan ratio attributed to AI-enhanced credit scoring and a 40 per cent-weighted improvement in the area under the credit-scoring discrimination curve. Weights were derived through confirmatory factor analysis on a pilot subset of 40 institutions. Data quality combined completeness, reliability and freshness, with respective weights of 40, 35 and 25 per cent.

Governance maturity combined audit processes and oversight committees, each weighted 35 per cent, with reporting practices weighted 30 per cent. Regulatory compliance came from supervisory assessments and was cross-validated against institutional self-evaluations. Researchers estimated a multivariate ordinary least-squares model containing AI adoption, data quality, governance maturity, institution size and regulatory compliance.

The AI-adoption variable was the logarithm of AI investment as a proportion of total assets. Diagnostic tests did not indicate significant departures from linearity, normality, constant error variance or residual independence; variance-inflation factors ranged from 1.143 to 2.217, well below a threshold of 5 used to flag problematic multicollinearity. Cook’s distance reached a maximum of 0.087, below the stated influence threshold of 1.

The model explained 53.1 per cent of the variation in the composite performance score, with an adjusted R-squared of 0.516 and an F-statistic of 35.24. These diagnostics support the stability of the reported associations, but they do not resolve possible reverse causality or unmeasured confounding. AI adoption had the largest estimated association with risk-management performance.

Its coefficient was 8.312, with a 95 per cent confidence interval from 4.078 to 12.546 and a p-value below 0.001. The corresponding bivariate correlation with the performance measure was 0.68, also the strongest pairwise relationship reported. Because the predictor was log-transformed, the coefficient should not be interpreted as a percentage return on AI spending or as evidence that every additional unit of expenditure produces an 8.312-point performance gain.

Instead, with data quality, governance, regulatory compliance and size held constant, greater AI investment intensity was associated with a higher composite score. The wide confidence interval indicates substantial variation among institutions in the observed relationship. Data quality was also positively associated with performance.

A one-unit increase on the data-quality index corresponded to an estimated 0.318-unit increase in the composite score, with a 95 per cent confidence interval from 0.174 to 0.462. The index captured whether data were complete, reliable and fresh, qualities directly relevant to models that learn from historical information and assess transactions in near real time. The sample nevertheless showed wide differences: the mean data-quality score was 0.612, with a standard deviation of 0.148 and a range from 0.210 to 0.920.

Risk-governance maturity had a positive coefficient of 0.241, with a 95 per cent confidence interval of 0.107 to 0.375 and a p-value of 0.001. The measure covered audit processes, oversight committees and reporting practices. Survey information associated higher-scoring institutions with technical model validation, cross-functional review of model outputs, approval of major recalibrations and independent governance audits.

The study also identifies governance problems that such structures are intended to address, including model drift, overfitting to historical patterns, discriminatory outputs and optimisation against imperfect performance proxies. However, the cross-sectional design does not directly establish that governance preserves early AI gains over time, nor does it establish that oversight alone causes better performance; the result is an adjusted statistical association among institutions. Regulatory compliance had a positive coefficient of 0.187, with a 95 per cent confidence interval of 0.071 to 0.303 and a p-value of 0.002.

The mean compliance score was 0.673 and displayed the clearest country pattern: approximately 0.82 for Moroccan institutions, 0.71 for Egyptian institutions, 0.63 for Tunisian institutions, 0.52 for Algerian institutions and 0.39 for Libyan institutions. The measure was sourced from central-bank supervisory assessments and cross-validated against self-reports, with a reported correlation of 0.812. In a separate model combining Tunisia, Algeria and Libya, the regulatory coefficient was no longer significant at p = 0.124, a result the authors linked to the smaller subsample and limited regulatory variation.

Institution size, represented by the logarithm of total assets, had no statistically significant association with performance after the other variables were included. Its coefficient was 0.021, with a 95 per cent confidence interval from minus 0.014 to 0.056 and a p-value of 0.245. The size coefficient also remained non-significant in country and institution-type subsamples.

The null size result suggests that large balance sheets did not independently explain performance once AI adoption, data quality, governance and regulation were considered. The paper’s proposed mechanisms for this result include legacy-system complexity and slow organisational decision-making at large incumbents, but the cross-sectional analysis cannot directly demonstrate those pathways. The paper groups data quality, governance maturity and regulatory compliance into an AI Performance Enabling Ecosystem, describing them as mutually reinforcing institutional conditions associated with AI-driven risk performance.

The framework shifts attention from algorithm acquisition alone to the environment in which models operate. Reliable inputs matter because poor historical or current data constrain what a model can learn and how well its outputs can be applied. Governance provides structures for validation, recalibration, fairness monitoring and accountability, while supervisory frameworks can set expectations for documented, auditable risk systems.

The results are consistent with these mechanisms, but they do not show that AI “works best” only under one specific combination or establish causal amplification through a formal interaction model. In an augmented specification, the AI-adoption-by-data-quality interaction was positive but only marginally significant, at a coefficient of 2.14 and p = 0.063, leaving the relationship for testing in a larger sample. Several limitations constrain interpretation.

The cross-sectional structure prevents the use of within-institution fixed effects, lagged regressors or other designs needed for stronger causal identification, and the authors explicitly describe the estimates as robust statistical associations consistent with a causal interpretation rather than proof of causal direction. AI investment could follow management quality or other institutional advantages, and survey self-reporting may introduce social-desirability bias. The Libya subgroup contained only three institutions, preventing meaningful country-specific inference, while weighting choices in composite indicators introduce unavoidable measurement decisions.

The observational data also cannot determine whether particular AI applications caused reductions in non-performing loans, because that component depends on a benchmark attributed to AI-enhanced scoring. Despite these limits, the study offers a consistent regional picture in which AI adoption, strong institutional conditions and better measured risk performance occur together, while scale alone does not explain the cross-institution variation.

Subject of Research: Cross-sectional associations between AI adoption, institutional conditions and financial risk-management performance at 142 North African banking and financial institutions.

Article Title: Artificial intelligence in financial risk management empirical evidence from Morocco and North Africa

Article References: Attahiri, A., & Makhroute, M. (2026). Artificial intelligence in financial risk management empirical evidence from Morocco and North Africa. Discover Artificial Intelligence, 6(1), Article 1197. https://doi.org/10.1007/s44163-026-02233-9

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02233-9

Keywords: artificial intelligence, financial risk management, North African banking, cross-sectional study, data quality, risk governance, regulatory compliance, credit scoring, ordinary least squares, emerging markets, Artificial, intelligence

Researchers led by Bradley Rowland, MD, of Wake Forest University School of Medicine and Atrium Health Wake Forest Baptist Medical Center, analyzed electronic health record data from May to December 2024 covering 197,305 eligible adult inpatient encounters across 18 hospitals in the Atrium Health system, an integrated network spanning three states in the Southeastern United States. The network includes two quaternary care centers and numerous regional hospitals, providing significant demographic, geographic, and socioeconomic diversity. The system deployed the Accountable Health Communities Health-Related Social Needs Screening Tool, the instrument available through CMS, with minor modifications adding two response options per domain: patient declined to answer and patient unable to answer. The patient’s primary admission nurse completes the screen as part of the admission intake forms in the electronic health record, and nursing staff received an online training module before implementation and a second module ten months later.

The study’s first major finding concerns data quality, evaluated using two complementary frameworks. Under the harmonized data quality framework, which assesses conformance, completeness, and plausibility against internal standards, the screening responses satisfied all three domains, meeting minimum standards for real-world electronic health record data analysis. Validation against an external gold standard was not possible because other institutions use heterogeneous screening tools and inpatient screening evidence remains limited. A secondary semantic framework interrogating whether data collected for operational purposes is fit for research use revealed a notable weakness: the Housing Instability domain of the AHC-HRSN tool allows 2,048 possible unique response combinations, far more than a care team member could reasonably navigate during an inpatient encounter, and the answer choices were not sufficiently distinct to warrant their use. The researchers suggest modifying that domain’s implementation to achieve higher impact with lower response burden.

Screening performance was impressively high overall. Of the 197,305 eligible encounters, 172,519, or 87.4 percent, had at least one social needs element assessed, with consistent rates across all months of the study period. But the headline number conceals substantial variation between sites. Hospitals in Market B, comprising one quaternary center and twelve regional hospitals, completed screens in 92.1 percent of eligible encounters, while Market A, with one quaternary academic center and four regional hospitals, reached only 73.4 percent. The gap between the two largest hospitals in the study was maintained over time and was not significantly affected by the post-implementation training module. Despite a unified tool and standardized training, screening rates varied widely across sites, and the authors note that qualitative work is needed to understand these site-level differences.

Among patients who were screened, 12.7 percent, or 21,900 encounters, registered positive for at least one social need, while 80.1 percent were negative and 7.2 percent were inconclusive due to incomplete responses, refusals, or inability to answer. At the patient level, 12.4 percent screened positive for at least one category. Breaking down the positivity rates by domain, 7.4 percent of patients registered positive on transportation questions, 5.5 percent on food insecurity, 4.1 percent on utilities, 3.9 percent on housing insecurity, and 1.4 percent on partner violence. Refusals and inability-to-answer responses were infrequent, at less than 1 percent and less than 3 percent respectively, suggesting that patients are generally willing to discuss social needs with their nurses.

The correlation structure of the screening data provides important technical validation. Using Cramer’s V, a measure of association between categorical variables ranging from 0 to 1, the researchers found moderate to strong correlations between questions within the same domain, such as food insecurity questions correlating at 0.87. But correlations between different domains were weaker, such as 0.44 between food and housing insecurity. This pattern is clinically meaningful: it confirms that questions within a domain hang together coherently, while also demonstrating that a positive answer in one domain does not reliably predict positive answers in others. The practical implication is that screening all five domains is essential, because patients with transportation needs are not necessarily the same patients facing food insecurity or housing instability.

The clinical characteristics of patients screening positive aligned well with established knowledge about social vulnerability. Positive screens were more likely among younger patients, males, Black or African American individuals, Medicaid recipients, and those with higher social vulnerability as measured by the CDC’s Social Vulnerability Index, a composite census-based measure incorporating socioeconomic, demographic, and housing-related factors. Patients screening positive had a median Social Vulnerability Index of 0.74 compared with 0.67 for negative screens, and were more likely to have Medicaid coverage at 23.1 percent versus 13.7 percent. Patients with active online portals were less likely to screen positive, as were patients without recent hospitalizations. These patterns strengthen the validity of the tool, because the patient populations registering positive screens are precisely those the literature identifies as facing limited social resources.

On clinical outcomes, the picture is more nuanced. Among the 160,117 encounters with a positive or negative result, the overall 30-day readmission rate was 11.4 percent. Patients screening positive had higher raw readmission rates, 14.2 percent versus 11.0 percent for negative screens, and also showed higher annual hospitalization rates. However, after adjusting for age, sex, race, patient portal access, insurance class, Social Vulnerability Index, and prior year admissions using a mixed-effects logistic regression model with patient and hospital site as random intercepts, the significant difference in readmission rates disappeared. This suggests that the social needs captured by the screen overlap substantially with other clinical and socioeconomic risk factors already known to predict readmission, and that the incremental predictive value of the screening data for this particular outcome remains to be demonstrated.

The authors frame their findings as a functional first step in understanding the value of social needs screening across healthcare systems, particularly in the Southeastern United States. The diverse patient population and variety of care environments, from rural to suburban settings, lend hope that even smaller systems may benefit from these investigations. From a data quality standpoint, the researchers conclude that the AHC-HRSN tool and their implementation approach can be considered a ready option for other health systems contemplating deployment, with the caveat that the Housing Instability domain warrants modification. The implementation process appears sustainable, and the clinical correlations validate the screening instrument at significant scale.

Yet the study also highlights the gap between collecting social needs data and acting on it meaningfully. When patients screen positive, an automated alert is sent to the care coordination team, which then refers patients to FindHelp.Org, a zip code-based online resource platform. Whether such referrals translate into improved health outcomes remains an open question, and the authors acknowledge that insights attendant to this work remain preliminary. The CMS mandate presents an opportunity for health systems to provide more holistic and personalized care, and may offer improved risk stratification for important clinical outcomes. But as the researchers caution, for this potential to be realized, health systems must verify and validate their screening data and leverage it in a meaningful way to benefit patients, otherwise the requirement risks becoming yet another metric for metric’s sake. The 87.4 percent screening rate achieved across 18 hospitals demonstrates that large-scale implementation is feasible; the remaining challenge is ensuring that the data collected actually changes what happens to patients after they leave the hospital.

Subject of Research: Health systems research on social determinants of health screening implementation

Article Title: Evaluating the Implementation of Social Determinants of Health Screening Across 18 Hospitals

Article References: Rowland, B., Gee, J., Obermiller, C., Thorpe, A., Hernandez, J., Osiecki, L., Silber, R., & Houston, T. (2026). Evaluating the Implementation of Social Determinants of Health Screening Across 18 Hospitals. Journal of General Internal Medicine. https://doi.org/10.1007/s11606-026-10662-6

Image Credits: AI Generated

DOI: 10.1007/s11606-026-10662-6

Keywords: social determinants of health, health-related social needs, hospital screening, CMS mandate, electronic health records, data quality, health equity, readmission rates, Atrium Health, Journal of General Internal Medicine, Evaluating, Implementation

The researchers, led by Shahlokhon Gafurova of Bukhara State University in Uzbekistan, call this body of data quality-oriented big data: the subset of an organization’s data assets that is acquired, integrated, and analyzed to support decisions about the quality of products, services, and processes. Its defining feature, they argue, is consequence. When ordinary enterprise data is wrong, the result may be a misdirected advertisement. When quality-oriented data is wrong, the result is a defective product shipped to customers, a safety signal missed, or a supplier wrongly cleared. This distinction matters because the properties that make big data valuable, namely volume, velocity, and variety, are the same properties that make it difficult to manage. Variety in particular means that quality-relevant data arrive in incompatible schemas, units, granularities, and time bases, so the apparently simple act of asking a question across sources presupposes a substantial integration effort.

The paper positions itself against the rapidly growing literature on Quality 4.0, a term describing the alignment of quality management practices with the technological capabilities of Industry 4.0. That literature has produced valuable insights into definitions, readiness factors, and implementation barriers, but the authors contend it has a systematic blind spot: it treats the data lifecycle as a black box. The Quality 4.0 literature generally presumes that relevant data exist, that they can be brought together, and that they are trustworthy enough to feed analytical models. Research in data management suggests none of these presumptions is safe. Data integration remains one of the hardest and most expensive problems in enterprise computing, and machine learning pipelines in production settings fail far more often because of data problems than because of modeling problems. Field studies of high-stakes artificial intelligence have documented data cascades, in which small compromises in data work compound into serious downstream failures.

To close this gap, the authors propose a layered conceptual architecture that organizes the lifecycle of quality-oriented big data into five interconnected layers. Four of them form a vertical value chain: multi-source acquisition, integration and data fusion, data quality assurance, and analytics and knowledge extraction. The fifth layer, data governance and ethics, is deliberately drawn as a cross-cutting spine rather than a stage, because governance decisions constrain and enable every other layer, from what may be collected to how model outputs may be used. The architecture should be read through two flows. The upward flow is the value flow, in which raw signals gain meaning through fusion, gain trustworthiness through assurance, and gain decision relevance through analytics. The downward flow is the feedback flow, in which analytical findings redirect acquisition, refine fusion, and recalibrate assurance. Together, the two flows make the architecture a loop, consistent with the closed-loop character of quality work.

At the base of the architecture sits acquisition, which the authors insist is a design problem rather than a passive recording problem. Three design decisions carry particular weight. The first is coverage: which processes, machines, and product families are instrumented, and which remain dark. Coverage decisions create the blind spots within which quality problems can develop unobserved, and they are rarely revisited once made. The second is representativeness: whether captured data reflect the full operating envelope of the process, including rare but consequential conditions. The machine learning literature is unambiguous that collection-stage bias is among the most damaging and least visible defects a dataset can carry, because no downstream procedure can recover information that was never captured. The third is metadata at capture: recording the provenance, calibration state, and context of each stream at the moment it is acquired. Each characteristic failure of this layer, from blind spots to orphaned data whose meaning is unrecoverable, is invisible at the point where it is committed and expensive at the point where it is discovered.

The second layer converts many streams into one coherent body of evidence, and for quality-oriented data two integration tasks stand out. The first is product genealogy: linking every unit of output to the materials, machines, parameters, operators, and environmental conditions that produced it, across systems that were never designed to share keys. Genealogy is the backbone of root-cause analysis, recall management, and supplier quality, and it is an entity-resolution problem of considerable difficulty. The second is temporal alignment: placing measurements from sources with different clocks, sampling rates, and latencies onto a common time base. Quality analytics is intensely temporal, because cause precedes effect along a process flow, and misaligned time bases manufacture spurious lead-lag relationships. Both failure modes are insidious because they pass silently through statistical checks: the integrated data are internally well formed, they are simply wrong about the world.

The third layer, data quality assurance, is described as the keystone of the entire structure, the point at which the pipeline either does or does not convert integrated data into trustworthy data. The authors distinguish two shifts that separate assurance in their architecture from traditional data cleansing. The first is from episodic to continuous: in a closed-loop setting fed by streams, quality assurance cannot be a project conducted before analysis but must be a monitoring function that runs alongside the pipeline, validating new data against expectations and raising alarms on drift. The second is from generic to task-conditional. Fitness for use is relative to the use: a real-time process control application is exquisitely sensitive to timeliness and far more tolerant of isolated gaps, while a model trained to predict field failures is sensitive to completeness and representativeness across months of history and indifferent to millisecond latency. The paper maps seven quality dimensions, including accuracy, completeness, consistency, timeliness, validity, traceability, and representativeness, to the specific ways each degrades in multi-source pipelines and the quality decisions each degradation corrupts.

The theoretical core of the paper is a mediation claim stated in falsifiable form. The prevailing intuition, encouraged by the readiness literature, is that acquiring big data capability, meaning infrastructure, tools, and skills, improves organizational quality. The authors argue instead that a central part of this relationship runs through a specific intermediary: data quality. Capability contributes to better quality outcomes in large part because, and to the extent that, it delivers trustworthy data to the analytical layer; where the data foundation is weak, much of the capability accumulates as cost rather than converting into quality. The claim is testable: the mediation account predicts that a substantial share of the capability-to-outcome association should attenuate once data quality is introduced as an intermediate variable. Four additional propositions formalize boundary conditions, task-conditional effects, and the moderating role of governance, each stated so that it could in principle be contradicted by firm-level or pipeline-level evidence.

The governance and ethics layer answers a different kind of question: not what can be done with quality-oriented big data, but what may and should be done, by whom, and under whose accountability. Governance earns its cross-cutting position for a structural reason: it is the only layer whose outputs are constraints and mandates for all the others. Stewardship assignments determine who is accountable for coverage decisions, standards for identifiers determine whether the genealogy problem is tractable, and approval requirements determine which models may act on the process. Ethics enters through the specific character of quality-oriented data, because process data are also people data: operator identifiers, shift patterns, and performance traces are embedded in the same streams that describe machines and materials, so quality analytics borders on workforce surveillance and must be designed deliberately to respect that border. The authors argue that ethical principles take effect only when translated into decision rights, review procedures, and accountable roles.

For organizations, the architecture suggests a different investment logic than the one currently prevailing. The typical pattern is to begin at the analytics layer by procuring tools and then to discover the data layers below as obstacles. If the propositions hold, organizations would be better served by the reverse sequence: auditing acquisition coverage and metadata practice first, making genealogy and temporal alignment explicit engineering objects second, instituting continuous, task-conditional quality monitoring third, and letting the analytical portfolio grow at the pace the assured foundation supports. The authors are careful to note that the framework is conceptual and awaits empirical validation, and that testing the mediation claim will require longitudinal designs capable of ruling out reverse causality and common causes such as quality culture. But if the central claim survives scrutiny, the implication is striking: as quality decisions become more automated and more consequential, the trustworthiness of the data beneath them stops being a technical detail and becomes part of the substance of quality itself, making better managed data one of the most direct paths to better organizational quality.

Subject of Research: A conceptual architecture for managing quality-oriented big data across acquisition, fusion, quality assurance, analytics, and governance in data-driven organizations

Article Title: Managing quality-oriented big data: a conceptual architecture for acquisition, fusion, and analytics in data-driven organizations

Article References: Gafurova, S., Rahmatov, S., Zakhidova, U., Rakhimova, D., & Kabilova, G. (2026). Managing quality-oriented big data: a conceptual architecture for acquisition, fusion, and analytics in data-driven organizations. Frontiers in Big Data, 9, Article 1977331. https://doi.org/10.3389/fdata.2026.1977331

Image Credits: AI Generated

DOI: 10.3389/fdata.2026.1977331

Keywords: big data, data quality, Quality 4.0, data governance, data fusion, data integration, machine learning systems, quality analytics, data-driven organizations, data management, Industry 4.0, data-centric AI

The research, led by Christian Selinger of the Swiss Tropical and Public Health Institute in collaboration with the National Malaria Control Program (NMCP) of Côte d’Ivoire and partners, tackles a problem that has long troubled malaria evaluators. Randomized controlled trials remain the gold standard for establishing efficacy, but they are expensive and slow. Real-world evidence drawn from routine health information systems is cheaper, locally specific, and immediately useful to decision-makers—but it typically lacks the control groups and counterfactuals that retrospective impact evaluations critically require. The Ivorian team set out to quantify exactly how much these data quality limitations distort the picture.

The setting was a pair of districts, Nassian and Sakassou, where IRS campaigns were carried out in 2020 and 2021 with support from the US President’s Malaria Initiative (PMI). Two neighboring districts, Dabakala and Beoumi, served as unsprayed controls. Because routine data quality was deemed insufficient, PMI and the NMCP also re-digitized patient records directly from health facility consultation registers covering September 2018 to April 2022. This created a rare opportunity: the same malaria burden, recorded twice—once through the national District Health Information Software (DHIS2) database, in place since 2015, and once through carefully curated registry abstraction.

The researchers first asked which factors explained the discrepancies between the two data sources. Using mixed-effect linear models on log-transformed incidence differences, with a random intercept for each health district and backward stepwise selection, they found that the presence of the first IRS campaign, test positivity rates, and increased coverage of intermittent preventive treatment for pregnant women were statistically associated with a narrowing gap between the datasets. Health facility reporting rates entered the picture once environmental and entomological variables were added to the model. In other words, the very act of spraying—and the diagnostic and reporting behaviors surrounding it—left fingerprints on the surveillance record itself.

Correlation analysis sharpened this concern. When the team compared monthly incidence against remote-sensed covariates from the ERA5 reanalysis—rainfall, temperature, humidity, vegetation cover—alongside entomological measures such as biting rates and the entomological inoculation rate (EIR), a stark asymmetry emerged. Incidence derived from the re-digitized PMI data tracked expected seasonal patterns and correlated significantly with biting rates (coefficients of 0.53 and 0.57) and with surveillance quality indicators. The DHIS2 data, by contrast, showed no significant correlation with any entomological variable and only weaker links to rainfall and humidity. The authors conclude that routine DHIS2 incidence likely does not faithfully represent transmission dynamics at monthly scales, and that coarser temporal aggregation may be more appropriate for such data.

The heart of the study lies in its causal impact quantification. The team employed Bayesian structural time series analysis, a method that combines state-space modeling with Bayesian feature selection through spike-and-slab priors—probabilistic switches that decide which covariates belong in the regression at each iteration. The model is trained on pre-intervention data to forecast a counterfactual: what incidence would have been in the absence of spraying. Causal impact is then the difference between that counterfactual and the observed post-intervention incidence, averaged over one year, with causality deemed established if fewer than 5 percent of 10,000 simulated counterfactual draws fall below the observed values.

The results are striking in their divergence. When the model was trained only on incidence from the sprayed districts themselves, DHIS2 data suggested a 30 percent reduction in Nassian (95% credible interval 11–43 percent) and 36 percent in Sakassou (20–48 percent), while PMI data showed no significant impact. When incidence from the control districts was incorporated into model training, the estimates converged toward modesty: 12 percent (3–19 percent) for Nassian and 19 percent (10–27 percent) for Sakassou using the high-quality PMI data, versus 30 percent (19–38 percent) and 43 percent (34–50 percent) using DHIS2. The low-quality data thus yielded impact estimates roughly two and a half times larger than those based on carefully validated records. Notably, the PMI-based figures of 12 and 19 percent align well with a previously published independent analysis reporting reductions of 16 and 15.8 percent in the same districts, whereas the DHIS2-based estimates resemble the 44 and 47 percent reductions reported in DHIS2-based evaluations from Ghana and Uganda.

Which covariates drove these discrepancies? The posterior regression coefficients—extracted only when they maintained a consistent sign across 95 percent of model runs—told a coherent story. Incidence from control districts carried consistently positive coefficients with high inclusion probability, meaning that rising transmission in unsprayed neighbors raises the counterfactual and thereby tempers the apparent impact of spraying. For PMI data, health facility reporting rates showed consistently negative coefficients: periods of improved reporting during the intervention would lower the counterfactual and thus shrink the estimated effect. For DHIS2 data, test positivity rates, the number of active health facilities, and rainfall carried positive coefficients, suggesting that fluctuations in these surveillance-related quantities were the main drivers of the inflated impact estimates. The authors caution that these coefficients are high-dimensional and interdependent, so causality likely reflects a combination of factors rather than any single one.

The analysis extended to entomology with remarkable results. Using the same Bayesian framework on EIR data—excluding incidence covariates because they are statistically dependent on EIR—the team estimated a strong, statistically significant causal impact of IRS: roughly a 70 percent decrease in the entomological inoculation rate averaged over the first post-intervention year. This corroborates end-of-spray reports documenting more than 80 percent vector mortality on treated walls. Transmission modeling suggests that a 90 percent EIR reduction in highly endemic settings like Nassian and Sakassou would translate into about a 38 percent decline in clinical incidence under active case detection, hinting that entomological surveillance could serve as a surrogate metric for intervention impact—though the authors warn against generalizing from the limited spatial and temporal scope of entomological sampling.

The implications reach well beyond Côte d’Ivoire. The authors argue that scrutinizing test positivity rates and health facility reporting completeness before running impact analyses can prevent overly optimistic conclusions, and that carefully chosen control districts remain indispensable for factoring out confounders that surrogate variables like rainfall cannot capture. They acknowledge limitations: their causal graph remains elementary, modeling relationships between covariates was beyond scope, and findings may not extrapolate to health systems with stronger active surveillance or near-universal care-seeking, where baseline detection is less vulnerable to reporting gaps. Still, with malaria funding under strain—IRS in these very districts was halted after 2022 when PMI support ended—the message is urgent. Overestimating intervention impact leads to misallocated resources that could instead strengthen the surveillance systems on which all future evaluations depend. As climate change alters vector dynamics and malaria seasonality, the integration of remote-sensed data and rigorous counterfactual modeling with honest, high-quality routine data may determine whether the next decade of malaria control progress stalls or resumes.

Subject of Research: The influence of routine malaria surveillance data quality on estimates of indoor residual spraying impact in Côte d'Ivoire

Article Title: Evaluating indoor residual spraying in Côte d’Ivoire: the role of routine malaria data quality

Article References: Selinger, C., Appeti, K. D., Thawer, S., Angoa, G., Aïmain, S. A., Assi, S. B., Pothin, E., Hilton, E. R., & Tanoh, A. M. (2026). Evaluating indoor residual spraying in Côte d’Ivoire: the role of routine malaria data quality. International Journal of Public Health, 71, Article 1608766. https://doi.org/10.3389/ijph.2026.1608766

Image Credits: AI Generated

DOI: 10.3389/ijph.2026.1608766

Keywords: malaria, indoor residual spraying, Côte d'Ivoire, data quality, Bayesian structural time series, DHIS2, surveillance, vector control, entomological inoculation rate, counterfactual modeling, President's Malaria Initiative, health information systems

The review, led by Khurram Shahzad of Government College University in Lahore with collaborators from Alfaisal University, Prince Sultan University, and the University of Salford, screened 3,280 records pulled from eleven databases including Scopus, Web of Science, PubMed, and MEDLINE. After removing duplicates, ineligible studies, and non-English publications, only thirty peer-reviewed articles and conference proceedings survived the full-text gauntlet. The team followed PRISMA 2020 guidelines, registered the protocol on the Open Science Framework, and applied a ten-item CASP-informed quality appraisal to each surviving study, with inter-rater agreement checked using Cohen’s Kappa. The result is less a triumphal catalog of AI successes than a field guide to the gap between laboratory performance and real-world deployment.

The central technical finding concerns sentiment analysis itself. Sentiment analysis classifies the emotional polarity, attitudes, and opinions expressed in text, whereas misinformation detection asks whether a claim is factually accurate or misleading. These are fundamentally different tasks, and the review is emphatic that conflating them is a methodological trap. A false claim about a vaccine can be written in cheerful, confident language; a genuine warning about a drug interaction can sound alarming and negative. Both factual and misleading health information may carry positive or negative sentiment, which means emotional polarity alone cannot establish truthfulness. The authors position sentiment-derived signals as an auxiliary feature that must be combined with contextual semantics, source credibility, behavioral patterns, temporal information, and evidence-based verification before any judgment about accuracy can be made.

What does move the needle, according to the synthesis, is a cluster of five enabling factors. Advanced deep learning architectures, including transformer models and large language models, provide contextual reasoning that older lexical and statistical approaches cannot match. Feature engineering, the craft of extracting sentiment scores, named entities, health-specific lexicons, and semantic representations from raw text, remains foundational; high-quality data improved classification performance across transformer-based and machine learning frameworks alike. Real-time analytics, powered by cloud infrastructure and streaming pipelines, allow systems to flag emerging misinformation narratives during fast-moving public health events rather than cataloging them after the fact. And human-centered verification, where domain experts review automated outputs before they inform decisions, emerged as a critical safeguard against the unsupported inferences that even state-of-the-art models confidently produce.

The challenges side of the ledger is equally detailed. Annotation limitations top the list: high-quality labeled health misinformation datasets are scarce, manual annotation struggles to keep pace with rapidly evolving narratives, and inconsistencies between annotators degrade model training. Sampling and algorithmic biases can skew sentiment estimates derived from social media data, reducing the representativeness of any downstream detection system. Informal language, slang, emojis, and code-switching complicate processing, particularly in non-English settings, and mixed or neutral sentiments increase misclassification risk. These are not peripheral annoyances; they strike at the training data on which every detection model depends.

Scalability constraints form the second major barrier. Large language models demand substantial memory and processing power, real-time analytics introduce latency trade-offs between speed and accuracy, and cloud-based deployments carry infrastructure costs and external dependencies that many health systems cannot sustain. During health emergencies, when misinformation volumes spike, resource requirements can overwhelm the very systems meant to respond. Domain adaptation compounds the problem: models trained on one misinformation context lose accuracy when applied to different health topics, populations, or linguistic environments, and distinguishing misinformation from legitimate expressions of psychological distress remains contextually fraught.

Ethical and sustainability concerns round out the picture. Analyzing sensitive patient-generated and mental health data raises privacy questions that current governance frameworks do not consistently answer. Unreliable automated reasoning, hallucinated outputs from large language models, and inadequate external knowledge sources erode trust in AI-based detection. Cyber-security vulnerabilities, algorithmic bias in public health decision-making, and the absence of comprehensive infodemic surveillance frameworks all constrain long-term adoption. The review notes that operational integration into existing healthcare systems, coordination among healthcare decision-makers, policymakers, and technical teams, and cross-platform consistency remain unsolved organizational problems rather than purely technical ones.

From this synthesis the authors derive a five-component conceptual framework: a multi-source data acquisition layer to capture heterogeneous health information across social media, news portals, healthcare repositories, and IoT-enabled systems; an intelligent data management system handling cleaning, normalization, multilingual processing, and feature engineering; an AI-driven analytical engine integrating deep learning, sentiment signals, knowledge graphs, and real-time analytics; a human validation layer incorporating explainable AI, evidence retrieval, and expert oversight; and adaptive governance covering ethics, privacy, bias mitigation, regulatory compliance, and continuous model updating. The components are interdependent rather than sequential, and the authors are explicit that the framework is evidence-informed but not empirically validated; its effectiveness awaits prototype implementation and field testing.

One of the review’s more consequential observations is that data quality may matter as much as, or more than, algorithmic sophistication. Recent studies reporting strong performance from deep learning and LLM-based architectures still depend on representative, balanced, accurately annotated, domain-specific datasets. Increasing model complexity cannot compensate for weaknesses in annotation, contextual representation, or domain adaptation. This challenges a purely technology-centered reading of the misinformation problem and suggests that future research should prioritize computationally efficient, context-adaptive approaches over raw predictive performance, especially in healthcare settings with limited resources.

The review also documents a clear historical shift in the field. Earlier studies emphasized linguistic, statistical, and behavioral features for classification, while recent work increasingly employs BERT, transformer models, and large language models with retrieval-augmented generation for evidence-grounded analysis. Detection itself has moved from static classification of already-disseminated content toward streaming analytics, continuous surveillance, anomaly detection, and predictive assessment, a shift particularly relevant during health emergencies when narratives mutate rapidly. The authors caution that real-time capability is an operational requirement to be balanced against scalability and resource constraints, not an inherent marker of system superiority. For policymakers and health officials, the practical message is blunt: investing in advanced analytics without matching investment in datasets, human expertise, verification workflows, and governance will produce systems that look impressive in benchmarks and fail when it matters.

Subject of Research: Big data analytics and sentiment analysis for detecting health misinformation on social media

Article Title: Big data analytics and sentiment analysis for health misinformation detection: a systematic literature review of influencing factors, implementation challenges and a data-driven adoption framework

Article References: Shahzad, K., Javeed, A. M. D., Ahmad, S., Latif, M., & Saleem, K. (2026). Big data analytics and sentiment analysis for health misinformation detection: a systematic literature review of influencing factors, implementation challenges and a data-driven adoption framework. Frontiers in Big Data, 9, Article 1948398. https://doi.org/10.3389/fdata.2026.1948398

Image Credits: AI Generated

DOI: 10.3389/fdata.2026.1948398

Keywords: health misinformation, big data analytics, sentiment analysis, systematic literature review, deep learning, large language models, real-time analytics, human verification, data quality, AI ethics, public health surveillance, Frontiers in Big Data

The dataset was assembled by researchers at Rey Juan Carlos University in Madrid, working with a dermatologist from Hospital Universitario de Móstoles, and covers the complete image sets of three widely used public repositories: ISIC-2019 with 19,000 images, Derm7pt with 1,011 images, and PH2 with 200 images. For every one of the 20,211 images, the dataset records whether each of seven artifact categories is present, producing 141,477 individual binary labels. The contribution is not the images themselves, which remain with their original repositories, but the standardized, expert-reviewed artifact metadata layered on top of them, together with the annotation pipeline and dataset statistics released alongside.

Why does this matter? Prior work has shown that artifacts are not merely cosmetic nuisances. Studies have linked surgical skin markings and scale bars in dermoscopic images to measurable drops in the diagnostic performance of market-approved deep learning systems for melanoma recognition. Artifacts can alter the color and texture patterns that both clinicians and algorithms depend on, and they act as confounding factors that models can latch onto for the wrong reasons—learning, for example, that a dark corner correlates with a particular diagnosis rather than anything biologically meaningful. Earlier efforts to annotate artifacts, such as manual labeling of small subsets of ISIC 2018 and the Dermoscopic Atlas, demonstrated the problem but covered only a fraction of available data, and later classifier-generated labels lacked expert validation.

DermArtifactDB tackles the scale problem with a hybrid, multi-phase pipeline. In the first phase, images were collected from the three source repositories, none of which originally included artifact labels. The second phase deploys an ensemble of artifact-specific detectors: six categories—dark corners, gel border, gel bubbles, ruler markings, ink markings, and patches—are handled by classical image-processing techniques operating in multiple color spaces such as HSV, YCrCb, and CIELAB, while hair detection uses an ensemble of three deep learning segmentation networks (Attention U-Net, Dense U-Net, and Attention Dense U-Net) chosen for their ability to capture the morphological complexity of hair filaments. For each artifact, three complementary detectors run in parallel, each tuned to a different visual aspect such as color, geometry, or intensity, and a majority-voting rule accepts an artifact as present when at least two of the three detectors agree.

The technical details reveal how demanding reliable detection can be. Ruler markings are found with the probabilistic Hough transform, which identifies straight line segments; because a ruler edge typically appears as fragmented segments rather than one continuous line, the method groups segments sharing a common orientation within a tolerance of about ten degrees. Gel bubbles are isolated through Canny edge detection and morphological filtering, then filtered by a circularity index and an area window between 50 and 1,000 pixels, with hair masks suppressing false contours. Patch detection first segments the lesion itself using a Double U-Net model so that lesion pixels can be excluded, then searches for the saturated warm and cool chromatic tones characteristic of adhesive markers. Dark corners are flagged by analyzing the proportion of dark pixels along border rows and columns in each image quadrant, a thresholding scheme sensitive enough to catch subtle vignetting from the dermatoscope lens.

The third phase brings in a transformer. The team adapted C-Tran, a general multi-label classification transformer, pairing a ResNet-101 backbone pretrained on ImageNet with a Transformer encoder that models dependencies between artifact categories—for instance, the tendency of gel borders and gel bubbles to appear together. Training used a masked-label strategy, in which a random subset of labels is hidden so the model must infer them from context, together with a custom imbalance-aware loss called Class-Adaptive Focal with Confidence, designed to sharpen sensitivity to rare and visually subtle artifacts. In head-to-head comparisons against convolutional baselines including ResNet-101, DenseNet-121, and EfficientNet-B0, the transformer achieved the best scores on all four evaluation metrics: mean average precision, micro-averaged F1 and recall, and macro-averaged class-wise F1.

Neither automated stage is trusted alone. Whenever the classical ensemble and the transformer disagreed on any artifact category, the image was flagged for human review by two independent annotators—a computer vision researcher and a board-certified dermatologist—who examined the original image at full resolution, compared both sets of predictions, and consulted the transformer’s attention maps to locate the evidence behind each prediction. Before any joint discussion, the annotators independently labeled all 10,711 artifact-level decisions submitted for review, reaching 94.3 percent overall agreement with a Cohen’s kappa of 0.88. Agreement was perfect for patches and nearly perfect for ruler markings and dark corners, while gel border and gel bubbles proved hardest to separate from each other and from lookalike structures such as milia-like cysts within lesions. Of the flagged decisions, 5.7 percent required joint discussion, and three-quarters of those involved the two gel-related categories.

The resulting statistics expose how differently artifacts distribute across collections. In ISIC-2019, hair appears in 56.8 percent of images, gel bubbles in 37.2 percent, and dark corners in 32.7 percent, and it is the only dataset where ruler markings are frequent at 20.3 percent—a reflection of its large-scale, multi-center acquisition. PH2, captured under controlled conditions at a Portuguese hospital, is dominated by dark corners in 92 percent of images, with no ruler markings or patches at all. Derm7pt, drawn from an atlas of dermoscopy, shows a striking 76.2 percent prevalence of gel bubbles and almost no dark corners at 1.7 percent. Every dataset exhibits artifact co-occurrence, confirming that multi-artifact images are the norm rather than the exception in real clinical acquisition.

The authors are candid about the limits of their method. Agreement between the two automated approaches does not guarantee correctness, since both can fail in the same way on subtle artifacts, and such cases are invisible to the disagreement statistic. On the fully reviewed PH2 dataset, about 5.5 percent of the decisions on which the two systems agreed turned out to be inconsistent with the expert reference—roughly the same order as the 4.1 percent disagreement rate. Random audits of agreed labels in ISIC-2019 and Derm7pt corrected 2.76 and 3.33 percent of decisions respectively. The labels are therefore presented as a high-quality reference for benchmarking rather than absolute ground truth. The annotations are also image-level only: they say whether an artifact is present, not where it is, which rules out tasks like artifact segmentation or inpainting-based removal for now.

Released under FAIR data principles and documented with the Datasheets for Datasets framework, with annotations matched to original images by their unique identifiers, DermArtifactDB arrives at a moment when the field is reckoning with dataset bias and the brittleness of diagnostic AI. By quantifying artifact prevalence, co-occurrence, and diversity at scale, it gives developers the tools to stratify test sets, debias training data, and build preprocessing pipelines that detect and mitigate artifacts before they distort a diagnosis. The natural next step, the authors note, is region-level annotation—masks and bounding boxes for a representative subset—which would open the door to localization-aware artifact removal. For now, the hidden flaws of dermoscopic imaging finally have a systematic, searchable record.

Subject of Research: Multi-label annotation of acquisition artifacts in public dermoscopic image repositories for skin lesion analysis

Article Title: DermArtifactDB: a multi-label dataset for artifact annotation in public dermoscopic image repositories

Article References: Gómez-Martínez, V., Chushig-Muzo, D., Polimón-Olabarrieta, I., & Soguero-Ruiz, C. (2026). DermArtifactDB: a multi-label dataset for artifact annotation in public dermoscopic image repositories. Frontiers in Big Data, 9, Article 1866430. https://doi.org/10.3389/fdata.2026.1866430

Image Credits: AI Generated

DOI: 10.3389/fdata.2026.1866430

Keywords: dermoscopy, artifact detection, skin cancer, machine learning, medical imaging, dataset annotation, deep learning, transformer models, computer-aided diagnosis, melanoma, data quality, image segmentation

The context for this work is the ongoing crisis of data quality in internet-based research. Crowdsourcing platforms and open online surveys have made it possible to recruit thousands of participants quickly and cheaply, but that convenience has attracted a shadow economy of fraudulent respondents. Recent reviews of online studies published in 2022 found that many researchers still lack systematic practices for evaluating data quality, and a growing literature documents the scale of the problem: survey bots that mimic human responding, workers who misrepresent their identities or eligibility, and participants who speed through questionnaires without reading a single item. For longitudinal studies, which depend on the same people returning again and again over weeks or months, the stakes are even higher, because every fraudulent or disengaged recruit who drops out represents wasted resources and statistical power.

Howard and Peetz took advantage of a natural experiment embedded in their own research program. The study drew on existing data from a longitudinal financial intervention study of Canadian young adult women, with a total sample of 1,257 participants who had been recruited through three different streams. The first stream involved active videoconference verification: prospective participants were invited to a one-on-one Zoom meeting in which a researcher confirmed their identity and eligibility before enrollment. The second stream used passive screening, in which respondents were considered eligible if they possessed valid e-mail addresses and their contact information was not definitively fraudulent. The third stream consisted of pre-screened members of Prolific.com, a commercial participant platform that maintains its own vetting procedures. Because the researchers preregistered their hypotheses and analysis plan before compiling the data, the comparison was planned rather than post hoc, lending credibility to the findings.

The methodological design was unusually thorough. The researchers evaluated participants on seven distinct measures of retention, including whether they initiated the study, whether they completed the first wave, and whether they remained enrolled through subsequent waves over the course of the intervention. They also assessed six measures of engagement with the study tasks, including the time participants took to complete surveys, their interaction with the intervention activities, and indicators of honest responding such as attention checks. This dual focus matters because retention and engagement are conceptually distinct: a participant can show up for every session while clicking through the tasks mindlessly, or engage deeply once but then vanish. By separating these outcomes, the study could pinpoint exactly what videoconference verification does and does not buy a researcher.

The results on retention were unambiguous. Participants who had been verified over videoconference showed better rates of study initiation than those recruited through either passive screening or the commercial platform, and their longitudinal retention was superior as well. More than 90 percent of the videoconference-verified group completed all study waves, a remarkable figure for a multi-wave intervention study, where attrition rates of 30 to 50 percent are common in the literature. The passive screening stream, by contrast, almost certainly retained some fraudulent respondents among those who never responded to invitations or failed to show up for their Zoom appointments, since the initial e-mail screening was deliberately inclusive and only excluded contacts whose information was definitively fake.

Yet the engagement findings complicate the story. On measures of engagement, the three recruitment streams looked largely the same. The researchers found no meaningful differences in time to survey completion, in how participants engaged with the intervention activities, or in indicators of honest responding. In other words, videoconference verification did not produce participants who worked faster, tried harder on the intervention tasks, or answered more honestly once they were in the study. What it produced was participants who stayed. The distinction is important for researchers deciding how to allocate limited screening resources: if the primary threat to a study is attrition rather than careless responding, verification appears to be an efficient tool, but it is not a substitute for in-study attention checks and engagement measures.

The psychological mechanisms behind this effect are worth considering. One plausible explanation involves the sunk-cost dynamic documented in decades of behavioral economics research: participants who have invested effort in a Zoom appointment, shown their face to a researcher, and established a personal connection may feel more committed to seeing the study through. A brief face-to-face interaction transforms an anonymous transaction into a social relationship, and social obligations are notoriously harder to abandon than click-through commitments. There may also be a self-selection component, since people willing to schedule and attend a video call are likely those with more stable schedules and greater intrinsic motivation, meaning the verification session functions partly as a filter for reliable participants rather than only as a commitment device.

The study also speaks to a live debate about who bears responsibility for data quality. Some methodologists have argued that the burden for high-quality online data collection lies with researchers rather than recruitment platforms, and this study supports that view in a specific way: even participants drawn from a pre-screened commercial platform did not match the retention of videoconference-verified recruits, suggesting that platform-level vetting does not capture the commitment benefits of personal contact. At the same time, the authors note practical constraints. Prolific imposes maximum completion times once a survey begins, and videoconference verification requires researcher time that scales linearly with sample size, making it most practical for longitudinal intervention studies where each participant’s continued presence is worth a substantial investment. For massive cross-sectional surveys, the calculus may differ.

Limitations deserve honest acknowledgment. The sample consisted of Canadian young adult women recruited for a financial literacy intervention, so generalization to other populations, topics, and study designs remains to be demonstrated. The comparison across recruitment streams was not a randomized experiment at the level of the individual participant, since people self-selected into their recruitment pathways, and the researchers used existing data from a study originally conducted for a contract with the Financial Consumer Agency of Canada, which was not involved in the present analysis. The preregistration also contained a minor error about the recruitment start date, which the authors transparently corrected. None of these caveats undermines the central finding, but they frame it as strong evidence from a quasi-experimental comparison rather than definitive causal proof.

For the research community, the takeaway is both simple and consequential: engagement, or at least the kind that matters most for longitudinal research, begins at recruitment. A five-minute video call before enrollment appears to select and commit participants in ways that e-mail screening and platform pre-screening cannot, cutting attrition to levels that would make most longitudinal researchers envious. As fraudulent respondents grow more sophisticated and artificial intelligence makes automated survey-taking easier than ever, methods that exploit genuinely human signals, like showing up on camera and talking to a real person, may prove among the most robust defenses available. The study’s data and analysis code are publicly available on the Open Science Framework, allowing other researchers to scrutinize and build on the approach. In an era when the cheapest research methods are also the most vulnerable to contamination, Howard and Peetz’s results suggest that a modest investment of researcher time at the front door pays dividends all the way through the final wave of data collection.

Subject of Research: Comparing videoconference eligibility verification with passive online screening for participant retention and engagement in longitudinal online research

Article Title: Engagement begins at recruitment: Comparing videoconference verification to passive online screening for participant retention and study engagement

Article References: Howard, A. L., & Peetz, J. (2026). Engagement begins at recruitment: Comparing videoconference verification to passive online screening for participant retention and study engagement. Behavior Research Methods, 58(11), Article 318. https://doi.org/10.3758/s13428-026-03194-8

Image Credits: AI Generated

DOI: 10.3758/s13428-026-03194-8

Keywords: online research, participant retention, videoconference verification, survey bots, data quality, longitudinal studies, Prolific, participant engagement, recruitment screening, Behavior Research Methods, attrition, research methods

The research team set out to answer a deceptively simple question: what do international and European Union-level documents actually require of medical AI systems whose behavior determines whether a patient receives the right diagnosis? To find out, they searched and screened relevant regulatory and ethical documents in April and May 2025, then subjected the material to a thematic evaluation followed by a formal ethical analysis. Rather than treating every guideline as an isolated text, the researchers wanted to understand the framework as a whole, mapping how rules issued by different bodies fit together, where they overlap, and where they leave dangerous silence. The result is one of the most comprehensive ethical maps to date of how diagnostic AI is supposed to behave before it ever touches a patient.

What emerged from the analysis was a framework organized into five thematic categories, each addressing a different dimension of an AI system’s behavioral performance. The first category concerns the data basis, the foundation on which every diagnostic algorithm stands. Regulations and ethical recommendations in this area demand data standardization, systematic data assessment, and demonstrable data quality. This makes intuitive sense: an image-classification model trained on poorly standardized scans, inconsistent labeling, or datasets that fail to represent the diversity of real patients will inherit those flaws and reproduce them at scale, potentially misdiagnosing entire populations. By anchoring requirements at the data level, regulators are attempting to ensure that the raw material of machine learning is trustworthy before a single parameter is trained.

The second category, reliability, addresses whether a system actually performs as intended. Here the framework calls for rigorous assessments, verification of the intended performance, robustness against variation, accuracy, and compliance with applicable standards. Reliability is where engineering meets ethics most directly, because a diagnostic model that is 95 percent accurate in a curated laboratory dataset may perform far worse on the noisy, heterogeneous images of everyday clinical practice. Robustness requirements push developers to demonstrate that their systems degrade gracefully rather than catastrophically when confronted with unfamiliar inputs, and compliance checks are meant to ensure that claimed performance matches real-world behavior rather than existing only in marketing materials.

The third category tackles one of the most debated problems in medical AI: comprehensibility. The framework centers on explainability, understandability, and traceability, reflecting a growing consensus that a black-box diagnosis is ethically problematic even when it is statistically accurate. If a radiologist cannot understand why an algorithm flagged a lesion, or if the system’s reasoning cannot be traced after the fact, then meaningful human oversight becomes nearly impossible. Explainability requirements are designed to give clinicians enough insight into a model’s behavior to accept or reject its output responsibly, while traceability ensures that when something goes wrong, investigators can reconstruct what the system did and why. Without these properties, accountability in the event of misdiagnosis would dissolve into technological opacity.

The fourth category concerns the disclosure of the system itself, encompassing transparency, operational disclosure, and collaborations. This dimension asks whether the people who use and are affected by diagnostic AI actually know what they are dealing with. Transparency obligations push developers and healthcare providers to reveal that an AI system is in use, what its limitations are, and how it was validated. Operational disclosure extends this to the practical conditions under which the system should run, and requirements around collaborations address the often tangled relationships between technology companies, hospitals, and research institutions that shape how these tools reach patients. Disclosure, in the ethical analysis, is not merely bureaucratic paperwork; it is the precondition for informed consent and for preserving the trust on which the doctor-patient relationship depends.

The fifth and final category is usability, regulated through requirements concerning resources and competency. A perfectly designed diagnostic algorithm can still cause harm if it is deployed in a hospital lacking the computational infrastructure, staffing, or training to use it correctly. Competency requirements recognize that clinicians need education to interpret AI outputs appropriately, to recognize when a system is operating outside its validated scope, and to resist the subtle automation bias that can lead human experts to defer uncritically to machine judgments. Resource requirements acknowledge that equitable access to well-implemented AI is itself an ethical issue, since institutions with fewer means risk falling behind in diagnostic quality.

Crucially, the researchers then tested this five-part framework against the four classical principles of medical ethics: autonomy, beneficence, non-maleficence, and social justice. Their conclusion was cautiously positive. The framework, they found, could support all four principles, ensuring that patients retain meaningful choice, that AI genuinely benefits health outcomes, that harm is minimized, and that the advantages of the technology are distributed fairly. But the support comes with a significant caveat. The framework permits a wide range of technically compliant measures that vary dramatically in their actual effectiveness. Two systems could both satisfy every regulatory checkbox while offering very different levels of real protection, which means formal compliance does not automatically translate into ethical safety.

The most alarming findings concern security. The study identifies an insufficient technical feasibility of providing security measures against malicious data alteration, combined with the dynamic, constantly evolving development of attack and defense mechanisms, as a severe challenge to maintaining a compliant, safe, and secure diagnostic AI system. In practical terms, this means that the images feeding diagnostic algorithms can potentially be manipulated, and that defenders are locked in a moving race with attackers whose techniques evolve faster than standards can be updated. A corrupted or adversarially altered medical image could steer an algorithm toward a wrong diagnosis while appearing entirely ordinary to human reviewers. Because such manipulation strikes at the integrity of the system’s inputs and outputs simultaneously, the researchers warn that it could jeopardize all four ethical principles at once: undermining patient autonomy through deceived decision-making, defeating beneficence by producing harmful outputs, violating non-maleficence by introducing preventable harm, and straining social justice if certain patient groups bear disproportionate risk.

The study’s overall conclusion is sobering but constructive. AI systems for image-based diagnostics, the authors find, are governed by essentially the same ethical and regulatory framework as general medical AI systems, despite the distinctive challenges that imaging data presents, including its high variability and the particular vulnerabilities of visual models. That equivalence may be convenient for regulators, but it means the framework was not specifically engineered for the threats that image-based diagnostics face. The authors argue that sufficient protection through a regulatory framework is necessary, one that responds directly to the challenges of image-based AI diagnostic systems and safeguards both the underlying data and the intended performance of the models. As hospitals worldwide continue to deploy these tools at an accelerating pace, the research serves as a timely warning: the rules exist, they are broadly sensible, but unless they evolve to address security gaps and the variability of compliant implementations, the promise of AI-assisted diagnosis will rest on foundations that a determined adversary, or simply an unrepresentative dataset, could quietly shake.

Subject of Research: Ethical and regulatory evaluation of behavioral performance requirements for AI systems in image-based medical diagnostics at the international and European Union level

Article Title: The ethical and regulatory framework of medical artificial intelligence (AI) system’s behavioral performance for image-based diagnostics: An ethical evaluation of the International and European Union Scope

Article References: Parchmann, N., Orzechowski, M., Braun, D., & Steger, F. (2026). The ethical and regulatory framework of medical artificial intelligence (AI) system’s behavioral performance for image-based diagnostics: An ethical evaluation of the International and European Union Scope. PLOS Digital Health, 5(10), e0001781. https://doi.org/10.1371/journal.pdig.0001781

Image Credits: AI Generated

DOI: 10.1371/journal.pdig.0001781

Keywords: medical AI, image-based diagnostics, ethics, regulation, European Union, data quality, explainability, transparency, reliability, security, patient autonomy, PLOS Digital Health

Process mining is the discipline of reconstructing business processes from the digital footprints left in information systems. Every event log must record at least three things: a case identifier that ties events to a single process instance, a timestamp, and an activity label naming the step that occurred. Those activity labels are the backbone of nearly every process mining technique, which makes them a tempting target for anyone asking the classic data science question: how much does garbage in really cost? The researchers framed their work around a simple research question: how do errors on activity labels affect the performance of discovery, conformance checking, and predictive monitoring?

To answer it, the team leaned on a well-established taxonomy of event log imperfection patterns developed by Suriadi and colleagues, focusing on the four that corrupt categorical labels. Distorted labels arise from typos, simulated in the experiments through five operations such as deleting a character, inserting a random one, swapping two characters, substituting a neighboring keyboard key, or flipping case. Polluted labels occur when extra information, such as a resource identifier or a random string, gets glued onto the activity name. Homonymous labels use one name for several genuinely different steps, while synonymous labels describe the same step with different names, a common artifact of merging logs from multiple systems. Using scripts built on the FLAWD language, the researchers injected each error type separately into clean logs at controlled error ratios, from as low as 2 percent of events up to 50 percent for the prediction experiments.

The experimental design cleverly mirrors real-world scenarios. For process discovery, models were built from dirty logs and then tested against clean logs, simulating a hospital that mines a process from error-prone records at one site and applies the resulting model at another site with clean data. For conformance checking, the direction was reversed: models discovered from clean logs were used to replay dirty logs, reflecting organizations that hold a trusted reference model and audit live, imperfect data. For predictive monitoring, long short-term memory neural networks with two layers of 128 nodes were trained on noisy logs and evaluated on clean test data across three tasks: next activity prediction, outcome prediction, and remaining time prediction. The experiments drew on widely used real-life logs from the Business Process Intelligence Challenges of 2011 and 2015, two synthetic datasets that allowed controlled injection of homonyms and synonyms, and a real log from the Italian judicial system.

The starkest result came from process discovery. Fitness, which measures how well a discovered model can replay the clean log, barely budged, because most traces in a dirty log remain clean. Precision, however, collapsed. Error patterns that inject many new labels, namely random distortions and random pollution, drove precision close to zero with as little as 2 percent of events affected, because the discovery algorithm responds to the flood of novel trace variants by producing wildly permissive, underfitting models. Patterns that introduce only a handful of new labels, such as homonyms and synonyms, were far gentler. The practical implication is uncomfortable: cleaning for discovery must be nearly perfect to help, since even a small residue of random errors can destroy model precision, and analysts rarely know the true error rate in advance.

Conformance checking told a much more forgiving story. When a clean reference model was replayed against dirty logs, fitness declined only modestly, by roughly 10 percent in the worst case at a 10 percent error ratio, while precision stayed essentially constant. In many real settings, the authors argue, such degradation may be negligible for decision-making. This asymmetry carries a clear managerial message: it is far riskier to discover a process model from dirty data than to check dirty data against a trustworthy model. The quality of the reference process model, not the cleanliness of every incoming log, is the true fulcrum of reliable process mining.

Predictive process monitoring showed the sharpest differentiation among tasks. Next activity prediction, a multi-class classification problem in which label errors directly corrupt the target classes, suffered massive degradation that scaled with the number of new labels each error pattern introduced. Random pollution was worst, typos came next, and systematic pollution was milder, with performance at a 30 percent error ratio falling to roughly 30 percent of the clean-data baseline. Outcome prediction and remaining time prediction, by contrast, proved remarkably robust, with errors in mean absolute error generally staying within 10 percent of baseline. The researchers attribute this to the models leaning on numerical temporal features, such as time since case start and time since the previous event, rather than the exact wording of activity labels. In a striking twist under mixed error conditions, the LSTM networks eventually learned to discount the noisy categorical features entirely and rely on the unpolluted time series, stabilizing their predictions.

A real-world case study in the Italian civil judicial system grounded these findings outside the laboratory. The telematic case management system SICID defines more than 1,400 event types whose abstract and textual descriptions frequently diverge, creating naturally polluted and ambiguous labels. Training a remaining time predictor on 24,465 finished cases yielded a mean absolute error of 212 days with standardized clean codes versus 269 days with polluted labels, a difference of nearly two months in predicting trial durations. For a justice system under constitutional pressure to deliver timely hearings, that gap is not an academic footnote; it directly affects resource planning and the fairness of proceedings.

Perhaps the most counter-intuitive guidance from the study concerns cleaning priorities. Because errors that multiply the number of distinct activity labels, especially random typos and random pollution, cause the most damage, the authors recommend prioritizing cheap, automated fixes for random errors over expensive expert-driven efforts to resolve synonyms and homonyms, which introduce few new labels and correspondingly little harm. Techniques borrowed from tabular data cleaning, such as the Sorted Neighborhood deduplication method, edit distance metrics, word embeddings, transformer-based semantic similarity, and domain ontologies, can all be deployed to detect and repair label errors. Conformance checking itself can even serve as a diagnostic, with alignment discrepancies pinpointing which labels need targeted repair.

The study is candid about its limits: it relies on the Heuristic Miner discovery algorithm and LSTM-based predictors, restricts itself to imperative process models, and uses random uniform error injection that may not capture every real-world failure mode. Yet the authors see no strong theoretical reason why other algorithms would behave fundamentally differently. What emerges is a practical knowledge base for anyone designing process mining pipelines: know your downstream task before you clean, spend your data quality budget where new labels multiply, and treat a trusted process model as your best defense against dirty data. In an era when organizations increasingly hand their operations to data-driven analysis, this work is a reminder that the humble activity label, the word naming each step of a process, can be the difference between insight and illusion.

Subject of Research: The impact of activity label errors in event logs on process mining tasks such as process discovery, conformance checking, and predictive process monitoring

Article Title: On the impact of low-quality activity labels on data-driven process analytics

Article References: Comuzzi, M., Kim, S., Ko, J., Cappiello, C., Salamov, M., & Pernici, B. (2026). On the impact of low-quality activity labels on data-driven process analytics. International Journal of Data Science and Analytics, 22(1), Article 340. https://doi.org/10.1007/s41060-026-01334-z

Image Credits: AI Generated

DOI: 10.1007/s41060-026-01334-z

Keywords: process mining, event logs, data quality, activity labels, process discovery, conformance checking, predictive process monitoring, data cleaning, LSTM, machine learning, imperfection patterns, business process management