Danny Maupin

The growth of generative AI and easily available Open Access health datasets has transformed researcher productivity, leading to an explosion in publications that has in part been attributed to paper mills (organisations that provide manuscripts for payment) and other unethical actors. These entities are not, however, homogenous, and have a range of products and target markets. While the demand from China has received much attention, here we provide a case study of CDC WONDER, a dataset that has been exploited by a network of researchers reporting affiliations in Pakistan, the United States and the UK, potentially linked to medical residency driven demand from junior clinicians or trainees. The number of publications using CDC WONDER grew from 88 in 2021 to 1223 in 2025. Over the same time period, the proportion of papers reporting at least one author from Pakistan grew from 0.5% in 2021 to 27.2% in 2025, with unusually extensive collaboration networks. In some cases these works featured over 15 co-authors, often including representation from Western institutions, but in spite of this high level of resourcing only resulted in straightforward analyses of well-described conditions using publicly available data. The majority of these outputs additionally show evidence of being produced from a template, with formulaic titles and identical methods, for example using the same statistical model and platform (Joinpoint regression). Identifying papers produced by fast-churn workflows is essential to protect the integrity of the scientific literature from being flooded with low-quality research. This can be achieved through more proactive desk rejection of misleading and formulaic mass-produced submissions, and through better understanding of which use cases are appropriate for different Open Science resources. With the growing capabilities of AI to mass produce research, education will be essential to assist critical appraisal and preserve the benefits of Open Science.

Open research and data transparency are a bulwark against unethical activities, but can also introduce integrity risks. As with all public goods, freely available data can be exploited, and here we set out the case for the use of safeguarding practices.

Background Underemployment, working fewer hours than desired, has a negative impact on mental health. Research suggests that unemployment benefits can support the mental health of unemployed individuals, but it is unknown if a similar benefit occurs for underemployed workers. This study aims to assess the relationship between unemployment benefits and mental health for underemployed workers. Methods Data was analysed from 18 waves (June 2020-March 2021) of the COVID-19 survey, a sub-study of the Understanding America Study. The dataset included responses from individuals (n=1,241) who self-identified as employed before COVID-19. A meta-regression created pooled effect sizes measuring the impact of unemployment benefits on the change in mental health of underemployed workers with two-, four-, six-, or eight-week effect lags. Results Receiving unemployment benefits is associated with a trend towards reduced depression at four weeks (−0.13, 95%CI −0.25; 0.00) and reduced depression six weeks later (−0.14, 95%CI −0.23; −0.05). A significant reduction in anxiety scores was seen six weeks after receiving unemployment benefits (−0.18; 95%CI −0.34; −0.01). Results from gender were mixed with female workers noting a positive effect from unemployment benefits, and male workers showing no or even negative effects, particularly on anxiety. Conclusion These findings underscore the impact of unemployment benefits on underemployed workers’ mental health. Unemployment insurance, not a typically utilized policy strategy to improve mental health, may mitigate mental health repercussions of underemployment. Expanding benefits to include underemployed workers should be considered particularly during societal crises such as COVID-19. What is already known on this topic It is known that underemployment and employment insecurity can have negative impacts on mental and physical health, however it is not known whether providing benefits to underemployed individuals. What this study adds We now know that adding benefits to underemployed works can improve depression and anxiety, particularly in female workers. How this study might affect research, practice or policy This provides avenues for future research to address the impacts of benefits for underemployed workers, particularly for assessing causality. Further, it provides initial evidence that can be used to expand policy by including underemployed workers within unemployment benefits.

Background Redundant publication, the practice of submitting the same or substantially overlapping manuscripts multiple times, distorts the scientific record and wastes resources. Since 2022, publications using large open-science data resources have increased substantially, raising concerns that Generative AI (GenAI) may be facilitating the production of formulaic, redundant manuscripts. In this work we aim to quantify the extent of redundant publication from a single, large health dataset and to investigate whether GenAI can create submissions that evade standard integrity checks. Methods We conducted a systematic search for the years 2021 to 2025 (year to end-July) to identify redundant publications using the US Centers for Disease Control and Prevention National Health and Nutrition Examination Survey (NHANES) dataset. Redundancy was defined as publications analysing the same exposures associated with the same outcomes in the same national population. To test whether GenAI could facilitate creating these papers, we prompted large language models to write three synthetic manuscripts using redundant publications from our dataset as input, instructing them to maximise syntactic differences and evade plagiarism detectors. These three synthetic manuscripts were then tested using a leading plagiarism detection platform to assess their similarity scores. Findings Our search identified 411 redundant publications across 156 unique exposure-outcome pairings; for example, the association between oxidative balance score and chronic kidney disease using NHANES data was published six times in one year. In many instances, redundant articles appeared within the same journals. The three synthetic manuscripts created by GenAI to evade detection yielded overall similarity scores of 26%, 19%, and 14%, with individual similarity contributions below the typical 5% warning thresholds used by plagiarism detectors. Interpretation The rapid growth in redundant publications (a 17-fold increase between 2022 and 2024) is suggestive of a systemic failure of editorial checks. These papers distort meta-analyses and scientometric studies, waste scarce peer review resources and pose a significant threat to the integrity of the scientific record. We conclude that current checks for redundant publications and plagiarism are no longer fit for purpose in the GenAI era.

Background Law enforcement agencies typically conduct academy training to develop new officers. As these future officers are recruited from the general population, increases in physical workload during academy training can influence injury risk. This study explored the relationship between training load (TL) and injury risk among police officer recruits. Methods Data relating to injuries suffered, distance covered, physical fitness, and time spent in physical training were collected from 547 academy police recruits (431 male; 116 female). Course length varied between 20 and 22 weeks. A generalised linear mixed model was used to assess relationships between these variables and injury risk. The best fitting model was chosen using a stepwide approach with Akaike information criterion (AIC) and Bayesian information criterion (BIC) used for comparison. Results The best fitting model utilised weekly distance, week of training, and biological sex to predict injury (χ 2 = 38.3, p -value < 0.001). Higher weekly distances, earlier weeks of academy training, and female sex all resulted in higher probabilities of injury. Conclusions Rapid increases in TL (distance) during the transition from civilian to law enforcement recruit and lower fitness levels (resilience to TL) may lead to higher injury risk. The use of occupationally specific periodised, ability-based, training may lead to a more optimal TL for recruits, limiting overtraining while sufficiently developing fitness.

The monitoring and optimizing training load is a concept designed to decrease risk of injury while improving physical fitness and performance. This area has grown significantly in popularity, especially in the sporting world. However, these principles can be applied to other populations, such as tactical personnel who likewise are at increased risk of injury and engage in physically demanding occupational tasks that require sufficient fitness. Although monitoring the training load in tactical personnel may be effective, the critical differences between athletic and tactical populations necessitate occupationally specific load monitoring programs. This article aims to identify these barriers and suggests potential solutions for practitioners and researchers.

Survey response times (RTs) have hitherto untapped potential to allow researchers to gain more detailed insights into the cognitive performance of participants in online panel studies. We examined if RTs recorded from a brief online survey could serve as a digital biomarker for processing speed. Data from 9893 adults enrolled in the nationally representative Understanding America Study were used in the analyses. Hypotheses included that people’s average survey RTs would have a large correlation with an established processing speed test, small to moderate correlations with other cognitive tests, and associations with functional impairment. We also hypothesized that survey RTs would have sensitivity to various participant characteristics comparable to the established processing speed test’s sensitivity (eg, similar standardized means by sex). Overall, results support the validity and reliability of people’s average RTs to survey items as a digital biomarker for processing speed. The correlation between survey RTs (reverse scored) and the formal processing speed test was 0.61 (P < .001), and small to moderate associations with most other cognitive and functional status measures were observed. Sensitivity of survey RTs to various participant characteristics was nearly identical to the formal processing speed test. Survey RTs may be useful as proxies for processing speed.

Introduction: Cognitive assessments are useful in ascertaining dementia but may be influenced by patient characteristics. India's distinct culture and demographics warrant investigation into population‐specific cutoffs.Methods: Data were utilized from the Longitudinal Aging Study in India‐Diagnostic Assessment of Dementia (n = 2528). Dementia ascertainment was conducted by an online panel. A machine learning (ML) model was trained on these classifications, with explainable artificial intelligence to assess feature importance and inform cutoffs that were assessed across demographic groups.Results: The Informant Questionnaire of Cognitive Decline in the Elderly (IQCODE) and Hindi Mini‐Mental State Examination (HMSE) were identified as the most impactful assessments with optimal cutoffs of 3.8 and 25, respectively.Discussion: An ML assessment of clinician dementia ratings identified IQCODE and HMSE to be the most impactful assessments. Optimal cutoffs of 3.8 and 25 were identified and performed excellently in the overall sample, though did decrease in specific, more difficult‐to‐diagnose subgroups.Highlights: Pioneers use of explainable artificial intelligence in the diagnosis of dementia. Creates assessment cutoffs specific to the nation of India. Highlights differences in cutoffs across nations.

Background: The underdiagnosis of cognitive impairment hinders timely intervention of dementia. Health professionals working in the community play a critical role in the early detection of cognitive impairment, yet still face several challenges such as a lack of suitable tools, necessary training, and potential stigmatization. Objective: This study explored a novel application integrating psychometric methods with data science techniques to model subtle inconsistencies in questionnaire response data for early identification of cognitive impairment in community environments. Methods: This study analyzed questionnaire response data from participants aged 50 years and older in the Health and Retirement Study (waves 8-9, n=12,942). Predictors included low-quality response indices generated using the graded response model from four brief questionnaires (optimism, hopelessness, purpose in life, and life satisfaction) assessing aspects of overall well-being, a focus of health professionals in communities. The primary and supplemental predicted outcomes were current cognitive impairment derived from a validated criterion and dementia or mortality in the next ten years. Seven predictive models were trained, and the performance of these models was evaluated and compared. Results: The multilayer perceptron exhibited the best performance in predicting current cognitive impairment. In the selected four questionnaires, the area under curve values for identifying current cognitive impairment ranged from 0.63 to 0.66 and was improved to 0.71 to 0.74 when combining the low-quality response indices with age and gender for prediction. We set the threshold for assessing cognitive impairment risk in the tool based on the ratio of underdiagnosis costs to overdiagnosis costs, and a ratio of 4 was used as the default choice. Furthermore, the tool outperformed the efficiency of age or health-based screening strategies for identifying individuals at high risk for cognitive impairment, particularly in the 50- to 59-year and 60- to 69-year age groups. The tool is available on a portal website for the public to access freely. Conclusions: We developed a novel prediction tool that integrates psychometric methods with data science to facilitate “passive or backend” cognitive impairment assessments in community settings, aiming to promote early cognitive impairment detection. This tool simplifies the cognitive impairment assessment process, making it more adaptable and reducing burdens. Our approach also presents a new perspective for using questionnaire data: leveraging, rather than dismissing, low-quality data.