The seven selected progenies had lower plant heights of 1.02–1.64 m, compared with 2.01 m for the Olour parental genotype. Their stem girth ranged from 49.56 to 70.81 mm, versus 93.56 mm in the parent, while internodal length ranged from 1.82 to 2.55 cm, compared with 3.45 cm. The selected plants also had higher fresh- and dry-weight bark-to-wood ratios than the parent, apart from the specific exceptions reported for OP-17–58 on a fresh-weight basis and OP-17–40 and OP-17–58 on a dry-weight basis. These measured differences formed the basis for classifying the seven progenies as promising dwarf phenotypes.

Physiological and biochemical measurements also distinguished the selected group. The promising progenies recorded photosynthetic rates of 4.49–5.46 μmol CO2 m−2 s−1, compared with 6.29 μmol CO2 m−2 s−1 for Olour, along with lower stomatal density and decreased chlorophyll levels. The abstract reports lower transpiration rates and higher levels of total phenolics, flavonoids and antioxidant capacity in the selected progenies, with the stated compound-level exceptions. Positive associations were found between plant height, stem girth, internodal length and chlorophyll content. Bark-to-wood ratio, total phenolics, flavonoids and total antioxidant capacity were negatively associated with vigour, indicating statistical relationships rather than demonstrated causes of dwarfing.

Multivariate analysis showed moderate to high variation among the half-sib progenies. The first four principal components accounted for 76.1% of the total variation, with the first component explaining 45.86%. Two-way hierarchical clustering separated the progenies into dwarf, semi-vigorous and vigorous groups and placed the same seven selected progenies in the dwarf cluster. The authors identified reduced height, shorter internodes, lower stem girth, higher bark-to-wood ratio, and elevated phenolics and antioxidant capacity as traits associated with dwarfing. However, the abstract does not establish the effects of these plants on orchard yield, fruit quality, climate resilience or scion growth, nor does it report multi-site field validation. The selected progenies are therefore promising candidates for further rootstock evaluation and genetic research, not yet proven replacements for established mango rootstocks.

Subject of Research: Evaluation of dwarfing traits among 28 Olour mango half-sib progenies

Article Title: Dissecting dwarfness in half-sib progenies of Olour mango through morpho-physiological and biochemical characterization

Article References: Shivam, Singh, K., Prakash, J., Singh, S. K., Hussain, Z., Lama, A., Sharma, N., Shukoor, A., Dedhia, L., Singh, A., Pradhan, S., Kuldeep, & Raghav, D. (2026). Dissecting dwarfness in half-sib progenies of Olour mango through morpho-physiological and biochemical characterization. BMC Plant Biology. https://doi.org/10.1186/s12870-026-09801-2

Image Credits: AI Generated

DOI: 10.1186/s12870-026-09801-2

Keywords: Olour mango, mango rootstocks, half-sib progenies, plant dwarfing, morphological traits, physiological traits, biochemical traits, principal component analysis, hierarchical clustering, Dissecting, dwarfness, half-sib

The flipped classroom is an appealing testing ground for this kind of question because it loads nearly all of the time-management burden onto the student. In the course the researchers studied, a compulsory Programming Languages I class for second-year computer education students in Ankara, materials went online a week before each Tuesday session. Students were expected to work through recorded lectures, readings, and curated videos on their own, then arrive in class for a short diagnostic quiz, instructor-led clarification of difficult constructs like if-else statements and loops, and hands-on coding exercises. After class came a second quiz and a lab assignment with a four-day submission window. With no lecturer standing over them in the preparation phase, the course effectively measures how well each student regulates their own time, which is precisely where procrastination does its damage.

The methodological design is what sets the study apart from the long line of papers linking procrastination to poor grades. Akkurt and Somyürek used an explanatory sequential mixed-methods approach with 45 students. First came the behavioral data: every submission timestamp across twelve weeks was converted into a sequence of states, first-day submission, mid-week submission, deadline-day submission, or no submission at all. Those sequences were then fed into sequence dissimilarity analysis using the TraMineR package in R, followed by hierarchical clustering with Ward’s algorithm. The Silhouette method confirmed a two-cluster solution with a dendrogram coefficient of 0.90, indicating unusually clean separation between the groups.

The clusters were stark. Thirty-seven students, 82 percent of the class, fell into the Procrastinator profile, characterized by chronic last-day submissions and a noticeable rate of non-submissions. Only eight students, 18 percent, formed the Non-Procrastinator group, whose submissions clustered early and mid-week and whose failures to submit were nearly non-existent. The temporal analysis added an important wrinkle: during the first two weeks, both groups submitted early at high rates, a novelty effect that faded fast. From the third week onward the two profiles diverged sharply, and first-day submissions across the whole term eventually declined to just 22 percent, down from 96 percent in the opening fortnight.

The performance gap between the clusters was substantial. A Mann-Whitney U test found a statistically significant difference in final exam scores, with an effect size of r = .42. Non-Procrastinators averaged 58.44 points while Procrastinators averaged 33.08, a difference of more than 25 points on the exam. In a cumulative subject like programming, where each week’s logic builds on the last, the researchers argue that this is exactly what spaced practice theory would predict. Students who distributed their work across the four-day window were effectively practicing coding in multiple sessions, while deadline-day submitters compressed everything into a single massed effort that the literature consistently associates with weaker retention.

Then came the psychometric layer. All 45 students completed the 19-item Academic Procrastination Scale, which showed strong internal consistency in this sample with a Cronbach’s alpha of 0.922. A TwoStep cluster analysis in SPSS sorted the students into three groups by self-reported tendency: low, with a mean score of 33.89; moderate, at 51.52; and high, at 69.87. A Kruskal-Wallis test confirmed that final exam performance differed significantly across these groups, and post-hoc comparisons showed the low-procrastination group significantly outperformed the high-procrastination group with an effect size of r = .52. Procrastination, in other words, behaves as a continuum rather than a binary trait, and both ends of it show up in the gradebook.

The most revealing result, however, is what happened when the two classification systems were laid side by side. They overlapped only partially. Every single student in the high-procrastination group did indeed delay their online submissions, which suggests the questionnaire is a reliable flag for the most entrenched cases. But behavioral procrastination also appeared among 33 percent of the moderate group and 16 percent of the low group, students whose self-images did not match their timestamps. The authors conclude that self-reports and submission logs capture related but distinct aspects of the same phenomenon, and that relying on questionnaires alone would overlook students whose procrastination is situational, task-dependent, or simply not recognized by the students themselves.

The interviews with 13 students explain why the same outward behavior can have different roots. One theme, labeled Time Planners, gathered students who finish early to avoid anxiety and produce careful work. One participant explained that leaving things to the last minute creates difficulty, especially with factors beyond their control, which could cost points. The second theme, Last-minute Crammers, was dominated by a single code: habit. One student described how the transition from high school to university broke an early-finishing routine and installed a procrastination habit in its place. Another offered the study’s most sobering line about what deadline-driven work actually does to learning, saying they do not feel like they really learn, they just complete the task to meet the deadline and the information goes in one ear and out the other.

From these patterns the researchers propose a five-stage support framework for flipped courses: early identification through LMS indicators, profile differentiation using both logs and check-ins, targeted support matched to the underlying cause, continuous monitoring across assignment cycles, and adjustment of intensity based on how students respond. The timing detail matters. Because the behavioral divergence emerged from week three, the authors suggest monitoring becomes most informative after the initial weeks, when early enthusiasm gives way to stable patterns. They are careful, though, to frame the whole framework as a hypothesis. The study was observational, the strategies were never experimentally tested, and the authors explicitly state that dashboards, reminders, and intermediate deadlines cannot yet be claimed to reduce procrastination or raise achievement.

The limitations are honestly cataloged and worth keeping in mind. Forty-five students in a single course taught by one instructor is a context-specific sample, not a population. There was no traditional-classroom comparison group, so the submission patterns cannot be attributed to the flipped format itself. Causality runs in neither proven direction, since prior achievement, programming skill, workload, and motivation could all influence both submission timing and exam performance. The two behavioral clusters were badly unequal in size, and timestamps record when work was submitted, not when it was started; a student who begins early and revises until the deadline looks identical to a crammer in the log data. What the study does establish is a methodological point with real consequences for learning analytics: the digital exhaust of a course, read longitudinally and triangulated with self-reports and interviews, exposes the temporal anatomy of procrastination in a way no end-of-term questionnaire can. For instructors running demanding, cumulative courses, the third week of the semester may be the moment to start watching the timestamps.

Subject of Research: Academic procrastination and assignment submission behavior in a flipped undergraduate programming course

Article Title: Academic procrastination and assignment submission patterns in a flipped programming course

Article References: Akkurt, O., & Somyürek, S. (2026). Academic procrastination and assignment submission patterns in a flipped programming course. Discover Education, 5(1), Article 838. https://doi.org/10.1007/s44217-026-02111-7

Image Credits: AI Generated

DOI: 10.1007/s44217-026-02111-7

Keywords: academic procrastination, flipped classroom, learning analytics, programming education, assignment submission, LMS log data, mixed methods, self-regulated learning, hierarchical clustering, academic achievement, spaced practice, higher education

The research team started from a simple observation. HCBS is not a single intervention but a bundle of distinct services, and older adults combine them in very different ways depending on their health, finances, and living situations. Rather than treating all HCBS consumers as one homogeneous group, the investigators used a statistical technique called hierarchical clustering to sort them into natural groupings based on their actual patterns of service use. Hierarchical clustering works by repeatedly merging individuals or groups whose service-use profiles are most similar, building a tree-like structure that reveals how many distinct consumer types genuinely exist in the data. This method, common in genomics and market research, is less familiar in health services research, which makes its application here notable.

The study population consisted of 775 Medicaid HCBS consumers in Minnesota aged 65 and older, drawn from data collected in 2017 and 2018. The researchers linked administrative claims data from the Minnesota Department of Human Services with responses from the National Core Indicators Adult Consumer Survey, a standardized instrument that asks service recipients directly about their experiences. This linkage is what gives the study its power: instead of relying on provider reports or administrative proxies, the quality-of-life measure comes from the consumers themselves, including those living with dementia.

Quality of life was measured using a validated index that captures three domains: care experiences, security, and autonomy. Care experiences reflect how people feel about the help they receive; security concerns safety and stability in daily life; and autonomy captures the ability to make choices and maintain independence. Together, these dimensions form a composite score that can be compared across individuals and groups. The team then used linear regression to test two things: whether mean quality-of-life scores differed between the consumer clusters, and whether, within each cluster, scores differed between people with dementia and people without it.

The clustering analysis identified four distinct groups of HCBS consumers, defined by different combinations of the seven service types tracked in the claims data: home healthcare, non-medical transportation, personal care assistance, case management, adult day care, durable medical equipment, and homemaker services. The clusters were not interchangeable. They differed in their racial and ethnic composition, in the proportion of members living with dementia, and in their levels of functional dependence—the degree to which people needed hands-on help with everyday activities. In other words, the statistical method surfaced real, socially meaningful segments of the HCBS population, not arbitrary mathematical artifacts.

Then came the finding that gives the study its punch. When the researchers compared mean quality-of-life scores across the four clusters, the differences were essentially nil. The regression coefficients ranged between 0.00 and 0.02 percentage points, and none approached statistical significance. Within each cluster, the comparison between consumers with dementia and those without told the same story: coefficient estimates between −0.01 and 0.04 percentage points, again with no significant differences. For a field that often assumes dementia fundamentally alters how people experience and evaluate their care, this null result is striking.

The authors draw a practical conclusion from these numbers: consumer-reported quality of life may serve as a valid quality indicator uniformly across all HCBS, regardless of whether a recipient has a dementia diagnosis. That matters because states increasingly use quality measures to manage Medicaid long-term care programs, set payment rates, and hold providers accountable. If quality-of-life scores had varied wildly between service-use patterns or between diagnostic groups, policymakers would need to interpret scores differently for different populations—a statistical headache that invites unfair comparisons. The Minnesota data suggest that a single yardstick can be applied across the board.

The findings also carry a subtler message about equity. Because the clusters differed in racial and ethnic composition and in dementia prevalence, one might have expected systematic gaps in how well the service system serves different groups. The absence of quality-of-life differences between clusters offers cautious reassurance that, at least on this consumer-reported dimension, the HCBS system in Minnesota was delivering comparable perceived value across very different client profiles. The authors frame this as evidence that service providers may be able to meet the unique needs of diverse client bases, though the cross-sectional design means the study cannot prove that services caused these outcomes—only that, at a single point in time, self-reported well-being looked similar across groups.

Several caveats deserve attention. The study covers a single state and a single program era, so generalizing to other Medicaid systems requires care. The data come from 2017 and 2018, before the disruptions of the COVID-19 pandemic reshaped home-based care delivery. And clustering, while powerful, depends on the service categories recorded in claims data; unmeasured supports—informal family caregiving above all—could shape quality of life in ways the analysis cannot capture. The research was supported by the National Institute on Aging under award numbers RF1AG069771 and R01AG069771, and the authors note that the sponsor had no role in the design or interpretation of the study.

Even with those limits, the study lands at a moment when the demographic math is unforgiving. Populations are aging rapidly, nursing home costs continue to climb, and states are under pressure to expand home-based alternatives. If consumer-reported quality of life proves to be a stable, comparable metric across the full spectrum of HCBS users—with and without dementia—then states gain a practical tool for monitoring a fragmented system with one consistent instrument. The next step, which the Minnesota team’s methodology makes feasible, is to track these clusters over time and test whether shifts in service mix actually move the needle on the well-being that older adults themselves report. For now, the message is quietly radical: in this large Medicaid sample, what mattered least was which services you used or whether you had dementia—what the system delivered, by its recipients’ own account, was a broadly similar quality of life.

Subject of Research: Patterns of home- and community-based service use and quality of life among older adults with and without dementia

Article Title: Clusters of home- and community- based service use and association with quality of life in older adults with and without dementia

Article References: Jutkowitz, E., Parikh, R. R., Wolf, J. M., Duan, Y., Langworthy, B. W., & Shippee, T. P. (2026). Clusters of home- and community- based service use and association with quality of life in older adults with and without dementia. BMC Geriatrics. https://doi.org/10.1186/s12877-026-08442-2

Image Credits: AI Generated

DOI: 10.1186/s12877-026-08442-2

Keywords: home- and community-based services, dementia, quality of life, Medicaid, older adults, hierarchical clustering, long-term care, geriatrics, health services research, Minnesota, personal care assistance, aging population

Lipophilicity, conventionally expressed as the logarithm of the partition coefficient between n-octanol and water, log P o/w, is one of the most influential parameters in early drug discovery. It governs how readily a molecule crosses cell membranes, binds to plasma proteins, and distributes through tissues, thereby shaping absorption, distribution, metabolism, excretion, and toxicity. According to Lipinski’s rule of five, orally administered small-molecule candidates should ideally fall between 0 and 5 on this scale. Values above that window are frequently associated with poor aqueous solubility, excessive plasma protein binding, tissue accumulation, receptor promiscuity, and elevated in vivo toxicity, while insufficient lipophilicity undermines the same pharmacokinetic machinery from the opposite direction.

Measuring that property experimentally has traditionally relied on the OECD shake flask method, a liquid-liquid extraction procedure that is slow, material-hungry, and unforgiving of impure samples. Reversed-phase HPLC has emerged as a faster and more economical alternative, requiring only tiny amounts of compound and tolerating impurities. Because different stationary phases, from conventional C18 silica to polymer-based supports, present distinct hydrophobic surfaces and residual silanol populations, running a candidate across several columns yields a richer picture of lipophilic behavior than any single system can provide.

In the new work, the researchers dissolved each of the thirty-one cis- and trans-styrylbenzoxazolone derivatives and injected them onto three different C18 columns operated with either methanol-water or acetonitrile-water mobile phases, generating five distinct chromatographic systems. Isocratic elution ran from 95 percent down to 60 percent organic modifier at 0.75 ml/min and 35 °C, with a diode array detector recording at 254, 274, 292, and 305 nm. Because direct measurement in pure water would demand prohibitively long run times, the team measured retention factors at each eluent composition and extrapolated the linear or quadratic relationship between log k and the organic fraction back to zero organic modifier, yielding the chromatographic hydrophobicity index log kw.

Choosing between the linear Snyder-Soczewinski model and its quadratic extension is not trivial, since the quadratic coefficients proved vanishingly small across all systems. The authors therefore applied Mandel’s test, the IUPAC-recommended Fisher F-test comparison of residual variances, to decide which regression form best described each analyte. After extending the mobile phase interval with additional 65 and 75 percent compositions, the test confirmed linear behavior for only a handful of compound-system pairs, including S55-Z in system A, S67-Z in system B, S44-E in system C, and S48-E in system D. Notably, no derivative required a linear model on system E, whose endcapped stationary phase suppresses unwanted secondary interactions and treats all the stilbenes uniformly.

The solvent itself left a clear imprint on the results. Methanol, a protonic solvent capable of hydrogen bonding, strengthens interactions between analyte and the hydrophobic C18 chains, prolonging retention, whereas aprotic, low-polarity acetonitrile competes more effectively in the mobile phase and elutes compounds faster. Accordingly, log kw values were consistently higher in the three methanol-water systems than in their acetonitrile counterparts. System C, built on a polymer-based matrix with wide 300 Å pores, showed the weakest retention, since large pores reduce diffusional obstruction and ease desorption, and the polymeric backbone changes the interaction mechanism altogether. System D, also polymeric and wide-pored, similarly produced the lowest retention values among the acetonitrile systems.

Molecular geometry mattered as well. In reversed-phase HPLC, retention mirrors the way a molecule threads itself between nonpolar stationary phase chains, a process favored by planar, elongated structures and nonpolar fragments. The trans, or E, isomers, being more extended than their cis counterparts, were expected to retain more strongly, and indeed every E-isomer outperformed its Z-partner on system A. Only scattered inversions appeared on the other systems, and two isomer pairs, S61-Z/S62-E on system D and S77-Z/S78-E on system E, converged to identical extrapolated log kw values despite different retention profiles across the mobile phase compositions.

On the computational side, the team used SwissADME to gather predictions from five algorithms: XLOGP3, an atomistic method with corrective factors and a knowledge-based library; WLOGP, a purely atomistic fragmental approach; MLOGP, a topological model built on thirteen molecular descriptors; SILICOS-IT, a hybrid of twenty-seven fragments and seven topological descriptors; and iLOGP, a physics-based method deriving free solvation energies in octanol and water through a Generalized-Born and surface area model. Their arithmetic mean forms the consensus clog P. The consensus values ranged narrowly from 3.10 to 3.24, placing every derivative comfortably within the Lipinski window. Compounds bearing three methoxy groups on ring A were the most polar, while those with a single methoxy group were the most lipophilic, and most E-isomers edged out their Z-analogues, with the largest isomer gap appearing at position 7 of the styryl residue.

Correlating experiment with theory revealed a decisive pattern. The methanol-water systems A, C, and E produced statistically significant agreement between clog P and chromatographic lipophilicity at p < .05, while the acetonitrile systems ranged from moderate (system B, p ≈ .05) to poor (system D, p > .05). The strongest correspondences appeared on the silica-based columns with methanol eluents, marking those configurations as the most reliable predictors of lipophilicity for this compound family, and singling out the polymer-based, wide-pore system D as the least suitable for these analytes.

Chemometrics then consolidated the picture. Hierarchical clustering analysis, visualized as a dendrogram built from a bottom-up merging algorithm and validated with the Sneath-Sokal dissimilarity index, split the compounds into two main clusters that loosely tracked the position of the styryl fragment on ring B. All 3,5-dimethoxy derivatives fell into cluster I, whereas the 3,4-dimethoxy analogues, with one exception, grouped into cluster II, which also harbored most of the Z-isomers and the lower-lipophilicity members of the set. Principal component analysis, using the Kaiser eigenvalue criterion and a statistically validated loading threshold based on p-values rather than arbitrary cut-offs, extracted two components explaining 41.98 and 23.69 percent of total variance. The first component behaved as a silica factor dominated by systems A, B, and E, while the second, shaped by the wide-pored polymeric systems C and D, acted as a polymeric factor. The score biplot confirmed the clustering and flagged S44-E and S74-E as the most strongly retained outliers across all five systems. Together, the chromatographic, computational, and chemometric results deliver a robust framework for steering future ADME-oriented optimization of these anticancer candidates.

Subject of Research: Lipophilicity characterization of novel combretastatin A-4 stilbene derivatives by reversed-phase HPLC and computational prediction

Article Title: Investigation of lipophilicity of new stilbene derivatives with chromatographic and computational approaches

Article References: Georgieva, S., Voyslavov, T., Chanev, C., & Petrov, O. (2026). Investigation of lipophilicity of new stilbene derivatives with chromatographic and computational approaches. Pharmacia, 73, Article e200443. https://doi.org/10.3897/pharmacia.73.e200443

Image Credits: AI Generated

DOI: 10.3897/pharmacia.73.e200443

Keywords: lipophilicity, combretastatin A-4, stilbene derivatives, reversed-phase HPLC, log P, benzoxazolone, chemometrics, principal component analysis, hierarchical clustering, ADME, drug discovery, SwissADME

The research team began by assembling a candidate pool of quality indicators drawn from existing evaluation instruments and the published literature on health information quality. This starting library reflected the accumulated wisdom of information science and health communication research, covering familiar concepts such as accuracy, authority, and objectivity. But rather than imposing expert judgments from the top down, the researchers then designed a questionnaire survey aimed squarely at mobile social media users themselves. Their goal was to capture user perception as the organizing principle of the evaluation framework, on the reasoning that an assessment scheme which ignores the expectations of its end users risks being technically sound but practically irrelevant.

Methodologically, the study combined two analytical strategies. First, respondents were asked to rate the importance of each candidate indicator, allowing the researchers to rank the indicators according to user suggestions. Second, the team applied hierarchical clustering analysis, a statistical technique that groups items based on the similarity of their response patterns, to identify natural dimensions of indicator classification. Hierarchical clustering builds a tree-like structure in which each indicator starts as its own cluster and neighboring clusters are merged step by step according to a distance measure, so the final grouping emerges from the data rather than from preconceived categories. The researchers then refined these statistical clusters with professional practice, producing a framework that is both empirically grounded and operationally usable.

The result is a five-dimensional organization of health information quality indicators for mobile social media. Four of the dimensions concern the health information content itself: relevance, timeliness, credibility, and professional value. Relevance captures whether the information matches a user’s actual health situation and needs. Timeliness reflects how current the content is, a critical attribute in a field where guidance evolves rapidly. Credibility speaks to the trustworthiness of the claims and their sources, while professional value addresses whether the content carries genuine clinical or scientific substance rather than superficial wellness talk. The clustering analysis, supplemented by expert input, showed that these content-oriented properties form coherent, distinguishable axes along which users implicitly evaluate what they read on their phones.

Beyond the content dimensions, the framework incorporates platform-level and service-level indicators that reflect the distinctive affordances of mobile social media, such as navigation, personalized recommendation, and the presence or absence of advertising. This is precisely where conventional evaluation criteria fall short. Traditional health information quality tools were built for websites, printed materials, or institutional sources, and they have no natural way to account for algorithmic feeds, recommendation engines, or the commercial pressures embedded in platform design. By grounding the indicator set in user perception, the study ensures that these platform-specific factors receive the same analytical attention as the classical content attributes, producing an evaluation model that matches the environment in which people actually encounter health information.

One of the most practically valuable findings of the study concerns demographic consistency. Subgroup analysis revealed no statistically significant differences across demographic groups in the perceived importance of several key indicators, including authentication, objectivity, author transparency, reference source, and readability. In other words, users across different ages, backgrounds, and population segments appear to agree that knowing who authored a post, whether the author is credentialed, what evidence is cited, and how easily the text can be understood are all matters of shared concern. This consensus simplifies the task of platform designers and regulators, because it identifies a common baseline of quality expectations that can be applied universally without tailoring to specific demographic segments.

The same consistency extended to another cluster of indicators. Users from different groups showed aligned recognition of the importance of navigation, personalized recommendation, content relevance, the absence of advertising, content focus, and the universality of health programs. These indicators describe the experience of finding and consuming health content: whether users can move through information efficiently, whether recommendations match their needs, whether the content stays on topic, and whether commercial distractions interfere with the health message. The demographic agreement on these points suggests that platforms can pursue improvements in these areas with confidence that the changes will be valued broadly across their user base, rather than serving only a narrow slice of the audience.

Not every indicator, however, enjoyed universal endorsement. The study found significant differences among user groups in the perceived importance of multimedia richness, the degree to which health information is presented through varied formats such as video, audio, images, and interactive elements. This divergence is instructive. It implies that the optimal presentation format for health content is not a one-size-fits-all decision: some users may rely heavily on visual and video formats to understand medical concepts, while others may prefer text-based information or may be indifferent to format altogether. For content creators and health communicators, this finding argues for audience-aware production strategies, in which the choice of media format is calibrated to the target population rather than assumed to be universally beneficial.

The broader implications of the work reach into platform governance, health communication practice, and future research. By identifying a structured set of evaluation indicators and organizing them into five dimensions, the researchers have given platform operators, public health agencies, and information quality auditors a user-validated yardstick for assessing the health content that flows through mobile social media. Moreover, the demonstration that perceptions vary partially across user groups leads the authors to a clear conclusion: future optimization of health information quality can be carried out in a targeted manner for different user groups. Rather than applying uniform quality interventions, platforms and health authorities could tailor their efforts, strengthening, for example, credibility signals for all users while adjusting multimedia presentation strategies for specific segments.

The study, conducted in accordance with the Declaration of Helsinki with ethical approval from the Ethics Committee of the Second Xiangya Hospital of Central South University and informed consent from all participants, was supported by grants from the 2023 Key R&D Projects in Hunan Province and a 2023 Teaching Reform Research Project of Ordinary Higher Education Institutions in Hunan Province. As health misinformation continues to spread at the speed of a share button, frameworks like this one shift the conversation from vague alarm about bad information to concrete, measurable criteria grounded in what users themselves expect. The message for the digital health era is clear: quality evaluation must live where the information lives, on the mobile screens of the people it is meant to protect.

Subject of Research: User-perception-based evaluation indicators for health information quality on mobile social media

Article Title: Research on health information quality evaluation indicators for mobile social media based on user perception

Article References: Luo, A., Xu, L., Xu, Q., Nie, Y., Hu, D., Xin, Z., Dai, P., Zhao, X., Pan, Z., & Dai, P. (2026). Research on health information quality evaluation indicators for mobile social media based on user perception. BMC Public Health. https://doi.org/10.1186/s12889-026-29799-4

Image Credits: AI Generated

DOI: 10.1186/s12889-026-29799-4

Keywords: mobile social media, health information quality, evaluation indicators, user perception, public health, health informatics, hierarchical clustering, health communication, e-health, misinformation, personalized recommendation, BMC Public Health

The problem the researchers set out to solve is a familiar one in modern natural language processing. Transformer-based summarization models such as BART, PEGASUS, and T5 are remarkably fluent, but they face a persistent trade-off between brevity and completeness. Longer summaries tend to score better on automatic benchmarks like ROUGE because they overlap more with reference texts, yet they offer limited value to instructors and administrators who need rapid, actionable insights. Shorter summaries are more practical but often omit critical details or misrepresent the breadth of learner experiences. Compounding the issue, most existing systems apply a single pretrained model uniformly across all input content, treating highly informative sentences and semantically redundant ones in exactly the same way.

ClusSumm’s central innovation is to separate content selection from content generation. Rather than feeding an entire corpus of reviews into a summarizer and hoping for the best, the framework first organizes the material. Review texts are cleaned of URLs, punctuation noise, digits, and non-English tokens, then segmented into sentences and grouped by K-Means clustering so that semantically similar sentences end up together. The optimal number of clusters is determined automatically using the silhouette coefficient, a statistical measure that rewards clusters that are both internally compact and well separated from one another. The researchers searched cluster counts from 2 to 18 and selected the configuration yielding the highest silhouette score.

Within large clusters, where redundancy is most likely to accumulate, the framework applies TF-IDF sentence ranking. This classic information retrieval technique scores each sentence by combining how frequently its terms appear within the sentence with how rare those terms are across the wider corpus. Sentences with high scores are retained; low-ranked ones are filtered out before any generation takes place. This pre-selection step is what allows the downstream models to concentrate their generative capacity on genuinely informative content instead of diluting it across repetitive praise and boilerplate recommendations.

The framework then deploys what the authors call a model-adaptive assignment strategy. Large clusters, which serve as a practical proxy for input complexity, are routed to PEGASUS, a model known for strong abstractive capabilities on complicated inputs. Smaller, simpler clusters are handled by DistilBART, a computationally efficient distilled variant. Finally, the individual cluster summaries are merged into one coherent output by a hierarchical BART-based fusion module. According to the authors, this coupling of cluster-size-adaptive model assignment with hierarchical fusion within a single pipeline is what distinguishes ClusSumm from earlier hierarchical approaches that applied one model uniformly across all clusters.

To test the framework, the team first had to build the data. Because no publicly available dataset offered sufficient course-level review information across multiple MOOC domains, they constructed one through systematic web scraping of Coursera using Selenium WebDriver with Chrome automation. The collection covered eight domains: Artificial Intelligence, Machine Learning, Deep Learning, Natural Language Processing, Python Programming, Java Programming, Web Development, and Generative AI. Roughly 1,000 course links were gathered per domain, producing approximately 8,000 course records, each containing quantitative metadata such as ratings, enrollment figures, and like ratios alongside up to three learner reviews. The full collection effort took around eighty hours, and the curated dataset has been released publicly on Kaggle to support reproducibility.

The researchers benchmarked eight state-of-the-art transformer models against ClusSumm, including BART, PEGASUS, T5, FLAN-T5, DistilBART, LED, LongT5, and BigBird PEGASUS. All models were run in a zero-shot configuration using their standard Hugging Face settings, with no domain-specific fine-tuning, a choice the authors say reflects realistic deployment scenarios where labeled training data is scarce. The results revealed striking differences in compression behavior. LED produced the shortest summaries at a summarization ratio of just 3.50 percent, but paid a devastating price in quality, recording the lowest ROUGE-1 F-measure at 6.47 percent. At the other extreme, PEGASUS generated the least compressed summaries at 57.49 percent, achieving the highest F-measures largely by retaining vast portions of its input.

ClusSumm carved out a distinctive position. It achieved a summarization ratio of 12.65 percent, the second most concise of all nine systems evaluated, while maintaining ROUGE-1 precision of 98.31 percent and ROUGE-2 precision of 90.12 percent. In other words, nearly everything the framework chose to include was relevant, even though it included far less than its competitors. The authors are careful to note that ClusSumm’s recall and F-measure scores were lower than several baselines, and they characterize the framework as offering high precision under strong compression rather than overall superiority across every dimension. Still, the findings suggest that high ROUGE recall can sometimes reflect extensive text retention rather than effective abstraction, and that extreme compression without structured content selection, as LED demonstrated, severely degrades semantic coverage.

Perhaps the most consequential methodological choice in the study is its evaluation framework. Conventional assessments rely almost exclusively on ROUGE metrics, which measure overlap with reference summaries but say nothing about what actual users value. The researchers instead derived a composite Overall Score by surveying 30 participants with backgrounds in computer science, education technology, and instructional design, who ranked precision, recall, F-measure, and summarization ratio by perceived importance. Respondents assigned the highest weight to the summarization ratio, reflecting a consistent preference for concise outputs. Under this user-calibrated scoring, PEGASUS ranked first with 87.97 points, followed by BART-large-cnn and DistilBART, while ClusSumm ranked sixth overall with 74.10 points but earned the highest contribution from the compression component at 34.94 points. LongT5 and BigBird PEGASUS fared worst, suggesting that long-sequence architectures alone do not guarantee strong performance on heterogeneous, semantically sparse learner feedback.

The study’s limitations are acknowledged candidly by its authors. All models were evaluated without domain-specific fine-tuning, and no component-wise ablation was performed to isolate the individual contributions of clustering, sentence filtering, and hierarchical fusion. The reliance on ROUGE introduces lexical bias, and semantic similarity measures, human evaluation, and statistical significance testing were left for future work. Even so, the findings offer a practical blueprint for educational platforms drowning in feedback. By clustering semantically similar reviews, filtering redundancy before generation, and matching model capacity to input complexity, ClusSumm demonstrates that structured preprocessing and hierarchical summarization can turn unstructured learner commentary into decision-ready summaries, supporting instructors, course developers, and researchers in improving course quality at a scale that manual analysis could never match.

Subject of Research: Automated summarization of MOOC learner reviews using hierarchical clustering and transformer models

Article Title: ClusSumm a user centered hierarchical clustering and adaptive transformer framework for efficient summarization of MOOC learner reviews

Article References: Tanzin, M. A. A. A., & Mashrafi, M. J. (2026). ClusSumm a user centered hierarchical clustering and adaptive transformer framework for efficient summarization of MOOC learner reviews. Discover Artificial Intelligence, 6(1), Article 1461. https://doi.org/10.1007/s44163-026-02491-7

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02491-7

Keywords: MOOC reviews, text summarization, ClusSumm, hierarchical clustering, transformer models, K-Means, TF-IDF, PEGASUS, BART, ROUGE metrics, learning analytics, natural language processing

The scale of the underlying dataset is one of the study’s most striking features. The team examined primary and secondary care costs for the financial year 2018/19 among 10,119,490 adult patients registered with general practices contributing to the Clinical Practice Research Datalink and linked to Hospital Episode Statistics, a cohort representing roughly 23 percent of the adult population of England. Because the data capture both general practice contacts and hospital activity, the researchers could compute a genuinely comprehensive picture of each patient’s annual cost to the system, valued in 2018 prices using established unit cost references. High-cost patients were defined as those falling in the top one percent of total care costs, a threshold widely used in health economics because spending is so heavily concentrated at the extreme of the distribution.

The numbers illustrate that concentration vividly. High-cost patients had average annual costs of £17,485, with a median of £14,609 and an interquartile range running from £12,028 to £19,633. The rest of the population, by contrast, averaged just £653 per year, with a median of only £103 and an interquartile range of £14 to £352. In other words, the typical high-cost patient cost the system well over a hundred times as much as the typical patient, and even the cheapest quarter of the high-cost group outspent the most expensive quarter of everyone else by a wide margin. Understanding who these patients are, the authors argue, is therefore central to any serious attempt at resource allocation, case management and health system planning.

Methodologically, the study is a careful exercise in unsupervised learning, the branch of machine learning that searches for structure in data without any predefined labels. The researchers clustered patients on the basis of diagnoses, age, sex, deprivation measured by the Index of Multiple Deprivation, and ethnicity. They compared multiple algorithms, including approaches built on multiple correspondence analysis and t-distributed stochastic neighbour embedding for dimensionality reduction, before settling on hierarchical clustering as the optimal method. That choice was not arbitrary: the team evaluated candidate solutions using established validity and stability indices, including the Davies–Bouldin Index, the Calinski–Harabasz index and the Adjusted Rand index, which reward clusters whose members are internally similar, well separated from other clusters, and reproducible under resampling.

The winning solution produced nine clusters with clearly differentiated multimorbidity and sociodemographic profiles. Mean age across the clusters ranged from 56 to 79 years, a reminder that high costs are not confined to the very old. Mean annual costs per cluster ranged between £15,792 and £19,107, with confidence intervals indicating that these differences were estimated precisely. The spread of clinical pictures was equally wide. Several core profiles recurred across different clustering approaches and across subgroup analyses, suggesting they represent genuine and robust archetypes within the high-cost population rather than statistical artefacts of a single algorithm.

Among those recurring archetypes, four stand out. Younger people with liver disease and mental health conditions formed one recognisable group, a combination that clinicians will recognise as associated with complex social circumstances and frequent emergency attendance. Patients with nodal metastases, meaning cancers that have spread to lymph nodes, formed another, as did patients with prostate cancer complicated by nodal metastases. The fourth recurring profile comprised older people with cardiovascular disease and dementia, a group whose costs typically accumulate through repeated admissions, long hospital stays and heavy reliance on both primary and secondary care. That the same profiles emerged repeatedly, including in analyses restricted to in-year decedents and to patients with the highest unplanned care costs, strengthens the case that these are stable, clinically meaningful segments.

The study also stratified its analyses by age, examining age-stratified subgroups alongside the whole population. This matters because the drivers of high cost plausibly differ across the life course: a 58-year-old with alcoholic liver disease and depression uses services very differently from an 80-year-old with heart failure and cognitive decline. By confirming that the core clinical profiles persisted across these stratifications, the researchers provide evidence that segmentation is not simply a proxy for age. The consistency of the clusters across methods and subgroups was assessed formally, and the authors report that the hierarchical solution performed best on the validity indices they applied, giving the nine-group taxonomy a firmer empirical footing than ad hoc clinical categorisations.

Why does this matter for policy? Population health management, the approach increasingly adopted by integrated care systems in England, depends on being able to identify groups of patients whose needs are similar enough to be addressed by a common service model. If the top one percent of patients were homogeneous, a single intensive case management programme might suffice. The new analysis shows the opposite: a younger cohort dominated by liver disease and mental illness needs addiction, psychiatric and social care integration, while an older cohort with dementia and cardiovascular disease needs falls prevention, medication review and coordinated discharge planning. Designing services for a blended average of these groups risks serving none of them well. Cluster-based segmentation, the authors suggest, could inform the targeting of case management programmes, the allocation of budgets and longer-term workforce and capacity planning.

The research also carries implications for how health systems think about equity. By incorporating deprivation and ethnicity into the clustering alongside diagnoses, the analysis acknowledges that cost is shaped by social context, not just by disease burden. Patients in deprived areas often present later, have more multimorbidity at younger ages, and generate higher unplanned care costs, so a segmentation that ignores socioeconomic characteristics would systematically misdescribe the population. The study was approved by the Clinical Practice Research Datalink’s independent scientific advisory committee and used anonymised patient data under ethical approval from the Health Research Authority, and the authors declare no competing interests. Funding came from the Health Foundation and the National Institute for Health and Care Research, with the funders having no role in study design, analysis or the decision to publish.

Limitations and next steps remain. The analysis reflects a single financial year, 2018/19, before the disruptions of the pandemic reshaped waiting lists and care patterns, and cost data capture NHS spending rather than the full social care picture. Unsupervised clustering, however carefully validated, identifies statistical structure rather than causal mechanisms, so the clusters describe who high-cost patients are rather than why their costs are high. Even so, the study offers one of the most comprehensive segmentations of a national high-cost population attempted anywhere, drawing on a cohort of over ten million linked primary and secondary care records. Its central message is likely to travel well beyond England: the costliest patients are a heterogeneous collection of recognisable clinical archetypes, and health systems that learn to see them as distinct groups, rather than as an undifferentiated burden, will be better placed to design care that is both more effective and more humane.

Subject of Research: Machine learning segmentation of high-cost patients in the English NHS

Article Title: Segmentation of high-cost patients in England: insights from a machine learning analysis

Article References: Wang, S., Anselmi, L., Sutton, M., Kontopantelis, E., Beaney, T., & Anderson, M. (2026). Segmentation of high-cost patients in England: insights from a machine learning analysis. BMC Medicine. https://doi.org/10.1186/s12916-026-05233-9

Image Credits: AI Generated

DOI: 10.1186/s12916-026-05233-9

Keywords: high-cost patients, machine learning, hierarchical clustering, multimorbidity, NHS, health economics, population health management, CPRD, Hospital Episode Statistics, patient segmentation, healthcare costs, England

The research, which forms part of Pham’s doctoral work at National Economics University in Hanoi, analyzes a balanced panel of 135 Vietnamese listed industrial firms covering the period from 2016 to 2024. That window is analytically valuable because it spans years of robust pre-pandemic expansion, the sharp disruption of COVID-19, and the uneven recovery that followed. By holding the sample constant across all nine years, the authors ensure that every firm contributes equally to the analysis, avoiding the selection biases that can creep in when companies enter or exit a dataset mid-study. The result is a comprehensive portrait of how the financial anatomy of Vietnam’s industrial listed sector has evolved through one of the most turbulent stretches in modern economic history.

What sets the study apart methodologically is its three-stage analytical framework, which begins with a technique rarely applied to corporate finance in emerging markets: compositional data analysis. Financial statements are inherently compositional in nature, because the components of a balance sheet or income statement are constrained to sum to a fixed total. Treating raw ratios without accounting for this constraint can produce misleading statistical artifacts, a problem long recognized in fields like geology and ecology but often ignored in financial research. By applying compositional data methods, the authors respect the true geometry of financial statement data, ensuring that the relative structure of assets, liabilities, and equity is analyzed in a mathematically coherent way.

Building on that foundation, the researchers employed Ward’s hierarchical clustering to sort firm-year observations into recurring financial profiles. Ward’s method is a clustering algorithm that merges observations in a way that minimizes the increase in within-group variance at each step, producing compact and internally coherent groups. Rather than forcing firms into arbitrary categories defined by regulators or rating agencies, the clustering lets the data speak for itself, revealing natural groupings based on the relative composition of each firm’s financial statements. The analysis identified four distinct financial profiles, ranging from financially robust firms to the most vulnerable companies whose balance sheets signal elevated structural risk.

The second stage of the framework tackles a persistent problem in performance research: the proliferation of accounting ratios. Analysts can compute dozens of measures covering profitability, leverage, liquidity, efficiency, and growth, but these indicators overlap heavily and often move together in ways that obscure rather than clarify a firm’s true condition. Pham and Dao addressed this by applying principal component analysis to eleven accounting ratios, distilling them into three composite indices that capture the dominant dimensions of variation across the sample. The first index summarizes overall Performance, the second captures Financial Risk, and the third measures Management Efficiency. Principal component analysis works by identifying linear combinations of the original variables that explain the maximum possible variance, effectively compressing a crowded dashboard of ratios into a small set of orthogonal summary scores that retain most of the informational content.

With firms classified into profiles and composite indices constructed, the third stage estimated the relationships between financial risk, management efficiency, and performance using interactive two-way fixed-effects regression models. Fixed-effects models control for time-invariant characteristics of each firm and for shocks common to all firms in a given year, which helps isolate the genuine association between the explanatory indices and performance from confounding factors such as managerial quality or industry-wide cycles. The models included interaction terms between the indices and profile membership, allowing the strength of each relationship to differ across the four financial profiles. Standard errors were clustered at the firm level to account for serial correlation in the residuals of individual companies over time.

The headline findings are both intuitive and consequential. Financial risk was negatively associated with performance across all four profiles, confirming that firms carrying heavier structural financial burdens tend to underperform regardless of their position on the risk spectrum. Crucially, however, the strength of this adverse relationship was not uniform. The most financially vulnerable profile exhibited the strongest negative association, suggesting that risk is not merely a drag on performance but an increasingly punishing one as a firm’s balance sheet deteriorates. This nonlinearity carries a clear implication for investors and policymakers: the marginal cost of additional leverage or liquidity strain is highest precisely where firms can least afford it.

Management efficiency told a different story. The composite index of efficiency was positively associated with performance in several of the identified profiles, and the relationship was significantly stronger in the largest profile of firms. In other words, operational discipline, asset utilization, and cost control appear to translate into superior performance most powerfully for the broad middle of Vietnam’s industrial listed sector, where competitive pressures reward managerial skill most directly. For the smallest and most vulnerable groups, the payoff to efficiency gains appears more muted, possibly because structural financial fragility overwhelms the benefits of good management, or because such firms face constraints that prevent efficient practices from being fully exploited.

Robustness testing is where the study earns particular credibility. The authors re-estimated their models using Driscoll-Kraay inference, a technique designed to handle cross-sectional dependence in panel data, where shocks to one firm may spill over to others through supply chains, credit markets, or common exposure to macroeconomic conditions. They also considered lagged profile membership, addressing the concern that a firm’s current performance might influence its profile classification rather than the other way around. The main patterns remained broadly consistent under both checks, indicating that the core conclusions are not artifacts of a particular statistical assumption or of simultaneity between classification and outcomes.

The broader significance of the research extends beyond Vietnam’s borders. Emerging markets often lack the deep institutional infrastructure of developed economies, making it harder for investors, creditors, and regulators to distinguish genuinely resilient firms from those that merely look healthy on headline metrics. A profile-based approach offers a complementary lens: instead of asking whether a firm beats a uniform sector-wide benchmark, analysts can ask whether a firm’s financial structure, risk exposure, and management efficiency combine in ways characteristic of resilient or fragile archetypes. As Vietnam continues to attract foreign investment and its industrial firms take on longer-term sustainability-oriented projects, the capacity to match assessment methods to financial profiles may prove essential for channeling capital toward the companies best positioned to sustain growth through the next decade of global economic uncertainty.

Subject of Research: Financial performance assessment of Vietnamese listed industrial firms using compositional data analysis and principal component analysis

Article Title: Evaluating the financial performance of Vietnamese listed industrial firms using compositional data analysis and PCA indexing

Article References: Pham, V. H., & Dao, T. T. B. (2026). Evaluating the financial performance of Vietnamese listed industrial firms using compositional data analysis and PCA indexing. Discover Sustainability. https://doi.org/10.1007/s43621-026-04976-y

Image Credits: AI Generated

DOI: 10.1007/s43621-026-04976-y

Keywords: Vietnam, financial performance, compositional data analysis, principal component analysis, financial risk, management efficiency, hierarchical clustering, panel data, listed firms, industrial sector, corporate financial resilience, financial econometrics

The research, published in BMC Endocrine Disorders, enrolled 516 patients with diabetes aged 18 to 80 years across multiple hospitals, including Beijing Hospital, Capital Medical University Beijing Tiantan Hospital, Renmin Hospital of Wuhan University, Qilu Hospital of Shandong University, and Hebei Provincial People’s Hospital. Rather than dividing patients simply into type 1 and type 2 diabetes, the team applied a modified version of the Ahlqvist-based classification, a hierarchical clustering approach that sorts individuals with diabetes into biologically distinct subgroups based on variables such as age at diagnosis, body mass index, glycemic control, and insulin resistance. The subgroups include a severe insulin-deficient diabetes form, known as SAID, and a mild obesity-related diabetes form, known as MOD, among others. The idea is deceptively simple: if diabetes is really several different diseases wearing the same label, then its complications should also carry subtype-specific signatures.

To read those signatures, the investigators turned to corneal confocal microscopy, or CCM, a non-invasive imaging technique that can visualize living nerve fibers in the cornea at near-cellular resolution without taking a single biopsy. From these images, the researchers quantified key morphological parameters, most notably corneal nerve fiber density, or CNFD, a measure of how many main nerve fibers traverse a defined area of the corneal surface. Because small unmyelinated nerve fibers are damaged early in diabetic neuropathy — and because the cornea is the only place in the living body where such fibers can be directly observed — CCM has emerged in recent years as one of the most promising window-on-the-disease tools in neurology and ophthalmology alike.

The results revealed genuine heterogeneity across the Ahlqvist-based subtypes. Corneal nerve fiber density showed an overall between-subtype difference, with numerically lower values in patients classified as having severe insulin-deficient diabetes and numerically higher values in those with mild obesity-related diabetes. That pattern is biologically plausible: severe insulin deficiency is thought to expose nerves to a particularly hostile metabolic environment, while the milder, obesity-driven form may spare small fibers for longer. Yet the study’s authors are careful about how far this finding can be pushed. When the team applied corrected post hoc comparisons — the rigorous pairwise statistical tests required after an initial analysis of variance — no individual subtype pair reached statistical significance. The overall trend, in other words, is suggestive rather than definitive.

This honesty about statistical limits runs through the entire paper and is part of what makes it notable. The researchers also examined whether corneal nerve parameters tracked with nerve conduction studies, the electrical tests that measure how fast signals travel along peripheral nerves. Here they encountered a surprise: the associations between CCM parameters and median nerve motor conduction velocity pointed in an unexpected inverse direction, the opposite of what the team had anticipated. Rather than forcing an interpretation onto this anomaly, the authors explicitly advise caution, noting that the finding requires further validation before any conclusions can be drawn. Similarly, no significant associations emerged between the neuropathy disability score, a clinical measure of neurological impairment, and any of the corneal nerve parameters.

Where the study found its strongest and most consistent signal was in the relationship between corneal nerves and diabetic retinopathy, the microvascular eye disease that remains a leading cause of blindness in working-age adults. Across the entire cohort of 516 patients, lower standardized CCM parameters were consistently associated with higher odds of having diabetic retinopathy. Crucially, these associations held up even in sensitivity models that added further statistical adjustment for diabetes duration, HbA1c — the standard marker of long-term blood sugar control — and body mass index. That robustness matters, because diabetes duration and glycemic control are the classic drivers of retinal disease, and any new marker must prove its worth independent of them.

Intriguingly, the same did not hold for diabetic kidney disease. When the researchers tested the association between corneal nerve parameters and diabetic kidney disease, or DKD, they found no significant relationship. This divergence is scientifically interesting because both retinopathy and nephropathy are traditionally grouped together as microvascular complications of diabetes. If corneal small-fiber damage tracks with retinal involvement but not kidney involvement, it hints that the small nerve fibers of the cornea may share developmental, structural, or metabolic vulnerabilities with the microvasculature of the retina specifically — two tissues that, unlike the kidney, are classically described as lacking direct sympathetic innervation. The finding suggests that corneal nerve imaging markers may be a particularly faithful mirror of what is happening in the retina, potentially allowing clinicians to gauge retinal microvascular risk from a quick, painless scan of the eye’s surface.

The subtype-specific regression analyses, by contrast, remain exploratory. The authors attribute this caution to limited event numbers in some subgroups — a common challenge in clustering-based diabetes research, where certain subtypes may contain relatively few patients with a given complication. Small subgroup sizes make it difficult to detect associations with confidence, and the team is transparent that these findings should be treated as hypothesis-generating rather than practice-changing. The study was retrospectively registered in the Chinese Clinical Trial Registry on April 1, 2025, and was funded by the Beijing Municipal Science and Technology Commission and the Administrative Commission of Zhongguancun Science Park.

The broader significance of the work lies in its convergence of two major trends in diabetes research. The first is the Ahlqvist paradigm itself, first proposed by Swedish researchers in 2018, which reframed diabetes as a cluster of five subtypes with distinct prognoses and complication risks. Since then, researchers worldwide have tested whether these subtypes predict complications differently, and the Chinese study extends that question into the territory of small-fiber neuropathy — a complication that is notoriously difficult to diagnose early because its symptoms, such as burning pain and loss of sensation, often appear only after substantial nerve loss has occurred. The second trend is the maturation of corneal confocal microscopy from a research curiosity into a candidate surrogate endpoint, with international consortia pushing for its adoption in clinical trials of nerve-regenerating therapies.

By combining these two frontiers, the study raises the possibility of a two-layered precision approach: first classifying a patient’s diabetes by its biological subtype, then using corneal nerve imaging to monitor that patient’s specific microvascular and neurological risk over time. The authors themselves frame the path forward clearly, calling for longitudinal studies with comprehensive nerve conduction study-based definitions of distal symmetric polyneuropathy to confirm their exploratory findings. Cross-sectional snapshots, however large, can only show that two things coexist; longitudinal follow-up is needed to show whether corneal nerve loss actually precedes and predicts retinal disease, and whether the subtype differences seen here translate into different rates of complication progression. Until then, the study stands as an early but carefully documented map of how the invisible nerves of the eye differ from one form of diabetes to another — and a reminder that some of the most revealing diagnostics of the future may require nothing more than looking, with the right optics, at the surface of the eye.

Subject of Research: Corneal nerve morphology across Ahlqvist-based diabetes subtypes and its association with microvascular complications

Article Title: Corneal nerve morphological differences across Ahlqvist-based diabetes subtypes: a multicenter cross-sectional study

Article References: Li, Z., Liu, X., Wang, W., Zhong, L., Gao, L., Hou, X., Chen, S., Li, H., Yu, X., Huang, J., Pan, Q., Wu, J., & Guo, L. (2026). Corneal nerve morphological differences across Ahlqvist-based diabetes subtypes: a multicenter cross-sectional study. BMC Endocrine Disorders. https://doi.org/10.1186/s12902-026-02524-3

Image Credits: AI Generated

DOI: 10.1186/s12902-026-02524-3

Keywords: diabetes subtypes, Ahlqvist classification, corneal confocal microscopy, corneal nerve fiber density, diabetic neuropathy, small fiber neuropathy, diabetic retinopathy, diabetic kidney disease, hierarchical clustering, microvascular complications, nerve conduction studies, precision medicine

The research team, led by Ashiqur Rahman Khan and colleagues, built their predictive models using the Nigeria Malaria Indicator Survey 2021, a nationally representative dataset that captures demographic, household, and socioeconomic characteristics across the country. This choice of data source is significant. In many resource-limited settings, rapid diagnostic tests are the frontline screening tool, but they are not perfect. False-negative results can send an infected child home without treatment, and delays in confirmatory microscopy can postpone care until the disease has progressed. A statistical model that flags high-risk children using information already collected in routine surveys could serve as an early warning layer that complements, rather than replaces, conventional diagnostics.

The methodological pipeline began with 43 candidate variables drawn from the survey. Through careful preprocessing and statistical dependency testing, the team whittled this list down to 13 statistically significant predictors. This dimensionality reduction is more than a housekeeping step; it guards against overfitting, the phenomenon in which a model memorizes noise in the training data rather than learning genuine patterns that generalize to new populations. By retaining only variables with demonstrated statistical relevance, the researchers improved the odds that their models would perform reliably on unseen data.

Five machine learning approaches were then trained and compared: Logistic Regression, a Decision Tree, a Support Vector Classifier, Extreme Gradient Boosting, and H2O AutoML, a platform that automatically searches across many model configurations to find strong candidates. Evaluation followed rigorous practice, using stratified train-test splitting to preserve class proportions and 10-fold cross-validation to ensure that performance estimates were not artifacts of a single lucky split. Because missing a true malaria case is far more dangerous than a false alarm, the researchers emphasized the F2-score, a metric that weights recall, the ability to catch actual positives, more heavily than precision.

On the full dataset, the H2O AutoML Generalized Linear Model achieved the highest F2-score at 79.38 percent, with a recall of 83.08 percent, meaning it correctly identified more than four out of five children who truly belonged in the high-risk category. XGBoost, the gradient-boosted ensemble famous for dominating tabular data competitions, posted the highest test accuracy at 75.54 percent. The contrast between these two results illustrates a classic tension in medical machine learning: overall accuracy treats false positives and false negatives as equally costly, while a recall-oriented metric reflects the clinical reality that a missed infection can be fatal.

One of the study’s most interesting innovations addresses a subtle but important problem: regional confounding. Nigeria’s malaria epidemiology varies dramatically by geography, driven by differences in rainfall, ecology, and intervention coverage. A model that simply learns regional identifiers risks encoding where a child lives rather than the underlying risk factors, producing a tool that may not transfer to new regions or that obscures the true drivers of risk. To mitigate this, the team performed Agglomerative Hierarchical Clustering on the eight most important predictors after deliberately excluding regional identifiers, allowing the data to reveal natural population subgroups on its own.

The clustering analysis yielded two optimal population subgroups, with a Silhouette Score of 0.3939, a moderate but acceptable indication that the two clusters are internally coherent and distinct from one another. Rather than treating the entire nation as a homogeneous population, the researchers then trained cluster-specific models. The results validated the strategy: in Cluster 1, the H2O Generalized Linear Model delivered the best recall-oriented performance with an F2-score of 83.31 percent, while in Cluster 2, XGBoost took the lead with an F2-score of 83.25 percent. Both cluster-specific figures exceed the best overall-dataset F2-score, demonstrating that accounting for population heterogeneity measurably improves predictive performance.

The implications for public health practice in Nigeria and beyond are considerable. Malaria control programs already collect rich household survey data through instruments like the Malaria Indicator Survey, yet these data are often analyzed only in aggregate for policy planning. This study shows that the same data can be repurposed into an operational risk-stratification tool. Health workers equipped with a model trained on such data could prioritize households for bed net distribution, targeted testing campaigns, or preemptive treatment, stretching scarce resources further in settings where every diagnostic kit and every clinic visit counts.

The choice of H2O AutoML as the top performer also carries practical weight. Automated machine learning platforms reduce the expertise barrier, allowing public health analysts without deep data science backgrounds to retrain models as new survey waves arrive. In a country where malaria transmission dynamics shift with seasons and intervention campaigns, a framework that can be refreshed regularly, without a team of specialists hand-tuning algorithms, is far more likely to be sustained in practice than a bespoke model that decays as conditions change.

Caveats remain, as they do with any model built on survey data rather than clinical measurements. The predictors capture demographic, household, and socioeconomic conditions, not parasitological evidence, so the framework identifies risk rather than confirming infection, and it is designed to complement rapid diagnostic tests and microscopy, not to replace them. Still, the study demonstrates a robust and automated pathway toward early malaria risk assessment in the populations that need it most. If validated prospectively in field settings, cluster-aware AutoML screening could become a quiet but powerful addition to the toolkit fighting one of the world’s oldest and deadliest diseases.

Subject of Research: Machine learning-based early malaria risk prediction in Nigerian minors using non-clinical survey data

Article Title: Early malaria risk screening in Nigerian minors using AutoML and cluster-based analysis of non-clinical survey data

Article References: Khan, A. R., Mahbub, N., Aziz, R. A., Siddique, M. H., & Sayeem, Z. I. (2026). Early malaria risk screening in Nigerian minors using AutoML and cluster-based analysis of non-clinical survey data. PLOS Digital Health, 5(9), e0001736. https://doi.org/10.1371/journal.pdig.0001736

Image Credits: AI Generated

DOI: 10.1371/journal.pdig.0001736

Keywords: malaria, Nigeria, AutoML, machine learning, risk prediction, hierarchical clustering, XGBoost, public health, Malaria Indicator Survey, rapid diagnostic tests, children under five, resource-limited settings