The collection, assembled as a Research Topic titled Advances in Mathematical Modelling for Infectious Disease Control and Prevention, brings together eight studies that span an extraordinary range of biological scales and methodological traditions. Edited by Fahad Al Basir of Asansol Girls’ College in India and Khalid Hattaf of Hassan II University of Casablanca in Morocco, the Research Topic pursues three interrelated goals: understanding the mechanisms underlying infectious diseases, predicting their dynamics, and ensuring effective control and prevention. According to the editors, recent advances in the field have produced models that account for biological delays, coinfection, disease progression, vector-host interaction, and the constraints of healthcare systems, while machine learning, artificial intelligence, and time series analysis now offer powerful tools for analyzing large-scale epidemiological datasets.

At the mechanistic end of the spectrum, Elaiw and colleagues investigated the within-host infection dynamics of a simultaneous HIV and HBV infection using a nonlinear delayed system model. The inclusion of delay parameters, which capture the lagged nature of viral dynamics, makes the system considerably more realistic than instantaneous formulations. By analyzing the stability and long-term behavior of the modeled system, the study demonstrates how delay differential equations can reveal whether a coinfection settles into a persistent state or is driven toward clearance, a question of direct clinical relevance for patients carrying both viruses.

Mathematics can also connect infection dynamics to their long-term consequences, as shown by Amanso and coauthors, who built a model for human papillomavirus infections and cervical cancer prevention. Their framework incorporates disease progression and interventions including vaccination, awareness campaigns, screening, and treatment. Using sensitivity analysis and an optimal control approach, they explored the impact of intervention measures, identifying strategies that minimize both infection and cancer incidence at acceptable cost. The work is a striking example of how mathematical models can bridge the gap between a viral infection today and a malignancy decades later, giving policymakers a quantitative basis for investing in prevention.

The scope of the Research Topic extends beyond human medicine into agriculture. Collins and colleagues extended infectious disease modeling to plant pathology with a study of African cassava mosaic virus and East African cassava mosaic Cameroon virus, two pathogens that devastate a staple crop across much of the developing world. Their model considers disease dynamics and control in the presence of both infected plants and the whiteflies that transmit the viruses, making the study directly relevant to the agricultural and food industries. Because cassava feeds hundreds of millions of people, plant-virus modeling of this kind carries implications for food security that are every bit as serious as those of human epidemic models.

A second cluster of studies turns to the interaction between disease spread and the resilience of healthcare systems. Shaale and coauthors examined the effects of healthcare breakdown on nosocomial infections, constructing a model that explores how reduced healthcare capacity can contribute to rising hospital-acquired infections through weakened infection control measures and the persistence of pathogens in the hospital environment. The importance of this work is especially pronounced because infection control cannot be viewed separately from the healthcare environment itself. Models that explicitly incorporate the capacity of the healthcare system, the editors note, can yield additional insight into infection control problems that capacity-blind models miss entirely, a lesson underlined by the strains placed on hospitals during recent global emergencies.

While mechanistic models focus on the process of transmission, forecasting models aim to predict future epidemic behavior, and the Research Topic features several computational contributions in this vein. Demongeot and colleagues studied epidemic peak forecasting for COVID-19 and dengue by relying on the index of dispersion of new cases, a fairly simple statistical measure. Their research underscores the capabilities of parsimonious approaches while pointing to the importance of surveillance and reporting systems, since even the simplest forecasting statistic is only as good as the case data feeding it.

More sophisticated signal-processing and machine-learning techniques appear in the work of Feng and coauthors, who developed an approach to long-term forecasting of influenza epidemics using spectral filtering and a frequency-aligned loss function. By aligning the training objective with the frequency content of epidemic time series, the authors showed the effectiveness of applying tools from signal processing to traditional epidemiological forecasting, capturing seasonal patterns that standard loss functions can obscure. In a related contribution, Wang and colleagues compared time-series foundation models, namely TimesFM and Moirai, with ARIMA and LSTM networks in their ability to forecast hand, foot and mouth disease. The comparison drew on data from Korea, Singapore, and Chongqing and involved various forecasting horizons, and its findings show that the effectiveness of forecasting depends on the model, the horizon, and the specifics of the data, with no single architecture dominating across all settings.

The final contribution addresses a problem that is sometimes neglected in mathematical modeling: making a model sustainable for public health practice. Thakur and coauthors reflected on lessons learned from the development of the California Communicable Diseases Assessment Tool, known as CalCAT, a public health dashboard built on mathematical modeling. CalCAT illustrates that the usefulness of an infectious disease model goes far beyond its mathematical structure. Proper implementation, the authors argue, involves a reliable data pipeline, transparency of uncertainty estimates, validation against observed outcomes, and sustained interaction with users. If a model is not sustainable and not capable of being used by working epidemiologists, its usefulness is questionable, however elegant its equations may be.

Taken together, the eight studies reveal two complementary trends shaping the future of the field. The first is increasing biological and mathematical complexity, with models incorporating coinfections, time delays, disease progression, vector-host dynamics, and healthcare system limitations. The second is computational convergence, as machine learning, foundation models, time series analysis, and decision-support platforms are folded into the modeling workflow. The editors suggest that future infectious disease modeling will benefit most from the combined use of both traditions: mechanistic models contribute biological plausibility and interpretability, while data-driven models help discover patterns in large datasets. Yet uncertainty remains essential, because epidemiological measurements are subject to incomplete surveillance, reporting delays, changes in behavior and policy, and pathogen adaptation. The central theme of the collection is clear: mathematical models perform at their best when theory and data are combined in the context of real-world public health issues, and continued progress will depend on state-of-the-art techniques, high-quality data, realistic assumptions, proper validation, interdisciplinary collaboration, and clear communication with the practitioners who rely on these models for preparedness and prevention.

Subject of Research: Infectious Disease Modelling

Article Title: Editorial: Advances in mathematical modelling for infectious disease control and prevention

Article References: Al Basir, F., & Hattaf, K. (2026). Editorial: Advances in mathematical modelling for infectious disease control and prevention. Frontiers in Applied Mathematics and Statistics, 12, Article 2016937. https://doi.org/10.3389/fams.2026.2016937

Image Credits: AI Generated

DOI: 10.3389/fams.2026.2016937

Keywords: mathematical modelling, infectious disease, epidemiology, machine learning, public health, forecasting, Editorial, Advances, mathematical, modelling, infectious, disease

The study utilizes a panel dataset from the World Bank’s World Development Indicators, covering 45 Sub-Saharan African countries from 2001 to 2023. This dataset comprises 1,035 observations, providing a robust foundation for modeling electrification trajectories. Ethiopia serves as the primary case study, allowing for a detailed examination of a country with significant electrification challenges and potential. The framework compares several machine learning algorithms, including Linear Regression, Random Forest, XGBoost, and LightGBM, to determine which model offers the best balance of accuracy and interpretability for policy application.

Model validation was conducted using both leave-one-country-out and temporal-split methods to ensure robustness. The results indicated that Linear Regression achieved the lowest error rates, with a mean absolute error of 1.93 percentage points across the 45 held-out countries. For Ethiopia specifically, the error was 2.04 percentage points in the cross-country validation and 1.33 in the temporal split. This performance outperformed a persistence baseline in 33 of the 45 countries, with a median improvement in mean absolute error of 15.9 percent. The simplicity and accuracy of the linear model suggest that complex non-linear relationships may not be necessary for capturing the primary drivers of electrification in this context.

A key component of the framework is the use of SHapley Additive exPlanations, or SHAP, to provide interpretability. SHAP values reveal that prior-year access is the dominant feature in predictions, a finding consistent with the autoregressive specification and the concept of infrastructure path dependence. This means that a country’s current level of electrification is the strongest predictor of its future progress. While other socioeconomic factors play a role, the inertia of existing infrastructure creates a strong momentum effect. Understanding this path dependence is crucial for policymakers, as it highlights the importance of maintaining and expanding existing grids rather than starting from scratch in new areas.

The selection of Linear Regression as the scenario model was further justified through multicollinearity screening, coefficient-sign verification, and scenario-sensitivity checks. These diagnostic assessments ensure that the model’s predictions are not driven by spurious correlations or unstable coefficients. By using an interpretable model, the study allows policymakers to understand exactly how changes in input variables affect the forecasted electricity access rates. This transparency is essential for building trust in the model’s outputs and for designing targeted interventions that address specific bottlenecks in the electrification process.

Under three distinct policy scenarios—Slow, Business-as-Usual, and Accelerated—the model projects that Ethiopia will reach between 67.5 and 69.9 percent electricity access by 2030. This projection leaves a significant gap of 30 to 32 percentage points to achieve universal access. The analysis suggests that macroeconomic improvements alone are unlikely to close this gap within the remaining timeframe. Instead, structural interventions are required, including the extension of the national grid, reduction of connection costs, and the deployment of off-grid solar solutions. These findings align with broader trends in the region, where economic growth has not always translated directly into expanded energy access without deliberate policy action.

The framework’s open and potentially transferable nature offers a valuable resource for other regions facing similar challenges. By providing a reproducible workflow, the study enables researchers and policymakers to adapt the methodology to different contexts, pending validation in other areas. The use of publicly available data from the World Bank ensures that the model can be replicated and updated as new data becomes available. This accessibility is particularly important for developing countries that may lack the resources to develop proprietary forecasting tools. The open-source approach also facilitates collaboration and peer review, enhancing the reliability and applicability of the findings.

The implications of this research extend beyond Ethiopia to the broader Sub-Saharan African region. The 45 countries included in the panel represent a wide range of electrification trajectories, from near-zero access in countries like Somalia and South Sudan to near-universal access in Mauritius and South Africa. This diversity allows the model to capture the full spectrum of challenges and opportunities in continental-scale policy analysis. The findings suggest that a one-size-fits-all approach is insufficient; instead, policies must be tailored to the specific path dependence and structural constraints of each country. The framework provides a tool for such tailored analysis, enabling policymakers to simulate the impact of different interventions and prioritize those with the highest potential for closing the access gap.

As the world moves closer to the 2030 deadline for Sustainable Development Goal 7, the need for accurate and interpretable forecasting tools becomes increasingly critical. This study demonstrates that machine learning can be effectively applied to development policy when combined with rigorous validation and explainability techniques. The emphasis on interpretability ensures that the model’s insights are actionable, providing clear guidance on the structural changes needed to achieve universal electricity access. By bridging the gap between advanced data science and practical policy analysis, the framework offers a promising path forward for addressing one of the most pressing development challenges of our time.

Subject of Research: An interpretable machine learning framework for SDG 7 electricity access forecasting and policy scenario analysis in sub-Saharan Africa

Article Title: An interpretable machine learning framework for SDG 7 electricity access forecasting and policy scenario analysis in sub-Saharan Africa

Article References: File, S. G. (2026). An interpretable machine learning framework for SDG 7 electricity access forecasting and policy scenario analysis in sub-Saharan Africa. Discover Sustainability. https://doi.org/10.1007/s43621-026-05053-0

Image Credits: AI Generated

DOI: 10.1007/s43621-026-05053-0

Keywords: interpretable, machine, learning, framework, electricity, access, forecasting, policy, scenario, analysis, sub-Saharan, Africa

Despite extended advances in monitoring over the last decades, anticipating eruptions remains challenging. Seismic, geodetic, and gas precursors often herald unrest, yet forecasts remain uncertain because genuine precursor signals vary across volcanoes and are difficult to identify robustly across eruptive and non-eruptive periods. Some eruptions occur with subtle or absent warnings, and most volcanoes lack sufficient recorded events to identify reliable patterns. Recent data-driven approaches have improved the detection of subtle patterns in monitoring data, producing likelihood of eruption forecasts that can be evaluated under retrospective conditions. However, most studies focus on pre-eruptive intervals without comparing against non-eruptive repose, and objective metrics that account for missed eruptions and false alarms over long periods are necessary to evaluate forecast reliability. This gap highlights the need for a framework that explicitly balances the short-term costs of precautionary action against the avoided losses from catastrophic events.

In a study published in Nature Communications, researchers Alberto Ardid, David Dempsey, Garry McDonald, and Shane Cronin introduce a Potential Economic Value (PEV) framework to volcanology. This framework evaluates forecast utility by balancing precautionary actions against avoidable catastrophic losses, such as mass casualties. Originally developed for meteorology and aviation, and now applied in flood, drought, and wildfire risk management, PEV balances the short-term costs of precautionary actions against the avoided losses when forecasts correctly anticipate eruptions. The study demonstrates the PEV framework on multi-year eruption forecasts derived from a machine-learning model at five volcanoes that collectively span a spectrum of eruption frequencies, monitoring histories, and socio-economic settings. Two cases, Whakaari and Ontake, provide documented human and economic losses, enabling retrospective evaluation of the framework, while three additional applications serve as stress tests for generalizability.

The retrospective analysis of Whakaari shows that the forecast model could have reduced cumulative costs by more than 30% relative to a no-forecasting baseline. This result was obtained using a decision threshold of 0.6, which represents a deliberately precautionary choice that favors reducing missed eruptions at the expense of more frequent false alarms. In a separate optimization analysis, the value-maximizing threshold near 0.85 to 0.90 was associated with up to approximately 220 million USD in modeled avoided losses. That estimate should not be attributed to the more precautionary 0.6 threshold. The PEV curve exhibits a characteristic hump shape, where low thresholds result in frequent false alarms dominating expected costs, while high thresholds lead to missed eruptions driving losses. Monte Carlo ensemble sampling suggests that while absolute PEV values shift under different cost assumptions, the location of the optimum remains relatively stable near a threshold of 0.85 to 0.90. This stability suggests that threshold selection can be robust to substantial uncertainty in economic valuation.

Ontake provides a complementary test of the PEV framework, representing a high-exposure, land-based volcanic setting with public access and minimal structural damage. A retrospective scenario analysis suggests that a short-term forecast could have substantially reduced exposure-related losses while requiring repeated precautionary alerts over the analysis period. If a forecasting method had been available and acted upon with evacuation planning, the analysis shows that a forecast at Ontake could have reduced human and exposure-related losses by approximately 90%, avoiding short-term impacts despite 40 false alarms. This finding is consistent with the Whakaari results, indicating that when catastrophic human losses dominate the cost structure, even imperfect forecasts with high false-alarm rates provide substantial value over the long term. The contrast between Whakaari and Ontake illustrates how PEV responds to differences in exposure structure and institutional control, with Ontake showing that precautionary actions are limited to temporary closure of public trails and summit access.

The study also tests the generalizability of PEV to volcanoes lacking recent major loss events but facing credible high-impact scenarios. In high-exposure alpine settings like Mount Ruapehu, where lahars threaten people working in and visiting ski fields, short-term forecasts could reduce expected losses by enabling temporary access restrictions. Although only two minor eruptions occurred during the 14-year observation period at Ruapehu, ensemble modeling indicates that precautionary access restrictions could reduce expected losses by more than 40% under plausible assumptions about the probability that a missed eruption coincides with high exposure. Sensitivity analysis reveals that balanced thresholds maintain positive PEV across wide cost ranges, whereas risk-averse thresholds eliminate value in the absence of recent catastrophic losses. This illustrates a practical challenge: in the absence of recent damaging eruptions to calibrate perceived risk, high alert thresholds may seem cautious, yet they increase expected losses by allowing rare but catastrophic events to be missed.

Applications to Copahue and Bezymianny further demonstrate framework generalizability across volcanic systems with contrasting dominant avoidable loss pathways. At Bezymianny, short-term forecast value is driven primarily by aviation risk mitigation, as explosive eruptions frequently generate ash clouds that intersect major trans-Pacific air routes. In this context, precautionary actions consist mainly of airspace management and flight rerouting rather than evacuation of exposed populations. In contrast, the Copahue application models hypothetical evacuation and access-control costs involving nearby communities, people, livestock, and infrastructure in the Chile–Argentina border region. The analysis shows that positive forecast value is not confined to a single optimal threshold, but arises across a range of intermediate thresholds where reducing the likelihood of rare, high-impact misses outweighs the cumulative cost of precautionary actions.

To assess how PEV depends on forecast model, the researchers evaluated the metric on three different models: a machine-learning model tailored to a specific volcano, a model that uses transfer-learning to pool precursor information from multiple volcanoes, and a benchmark forecast based on unmodified seismic data. For Whakaari, the generalized model yielded the highest PEV and greatest cost savings, with the tailored model performing slightly worse. The same comparison for Copahue and Bezymianny indicates that generalized models achieve the highest positive economic value, despite training on data from other volcanic systems. This suggests that models incorporating generalized multi-volcano precursory signals offer operational advantages for data-limited settings where local eruption histories are insufficient. These comparisons suggest potential benefits from transferring precursor information between volcanoes, but the small set of case studies does not establish universal performance across volcanic systems.

Operational challenges remain in applying the PEV framework, as the decision threshold that maximizes expected value can only be identified retrospectively. In practice, agencies must choose alert thresholds under uncertainty, informed by imperfect knowledge of future eruption behavior, exposure, and societal tolerance for false alarms versus missed events. The researchers recommend that agencies adopt a robust initial threshold that maintains positive PEV across a broad range of plausible cost assumptions, rather than attempting to identify a value-maximizing threshold a priori. Balanced thresholds sustain positive PEV across wide cost-ratio ranges, making them robust initial choices. This approach parallels earthquake early warning practice, where alert thresholds are set below the target damage threshold to bound miss probability at a policy-specified ceiling, accepting a measurable increase in false alerts. As forecast-outcome data accumulate, thresholds can be refined through Bayesian updating of the posterior expected loss, following established methods in flood and weather forecasting.

Successful implementation of PEV-informed forecasting requires addressing both technical and behavioral dimensions. Research on warning compliance shows that repeated false alarms can erode public response by reducing perceived credibility and willingness to act, a phenomenon commonly referred to as the cry-wolf effect. However, evidence from risk-communication research suggests that transparent communication of uncertainty can sustain credibility even when forecasts are imperfect. A central behavioral challenge is managing societal response to frequent precautionary warnings, which can be addressed by reframing alerts as indicators of elevated risk rather than definitive eruption statements. Operationally, this behavioral framing must be supported by alert designs that translate high-frequency forecast outputs into warnings that are actionable but not overwhelming. The study concludes that the PEV framework provides a structured means of comparing alternative warning strategies under uncertainty, supporting the use of more precautionary warning thresholds provided that the resulting increase in alert frequency is managed in ways that maintain public compliance and trust.

Subject of Research: Socio-economic evaluation of data-driven volcanic eruption forecasting systems

Article Title: Socio-economic value of data-driven eruption forecasts to balance false alarms against catastrophic loss

Article References: Ardid, A., Dempsey, D., McDonald, G., & Cronin, S. (2026). Socio-economic value of data-driven eruption forecasts to balance false alarms against catastrophic loss. Nature Communications, 17(1), Article 9356. https://doi.org/10.1038/s41467-026-77242-4

Image Credits: AI Generated

DOI: 10.1038/s41467-026-77242-4

Keywords: volcanology, machine learning, risk management, seismic monitoring, economic impact, disaster preparedness, forecasting, public safety

The study, a retrospective time-series analysis spanning nearly a decade, drew on the District Health Information System, the digital backbone of Uganda’s public health surveillance. From January 2016 through 2025, the researchers compiled monthly counts of stroke cases and stroke deaths reported across the country. Because raw case counts can rise simply because more people live in Uganda each year, the team estimated monthly population figures using exponential growth interpolation and folded them into their statistical models as an offset term. That step allowed them to separate genuine increases in disease risk from the arithmetic of a growing population.

The modeling framework was Bayesian negative binomial regression, a choice suited to monthly count data that are typically overdispersed, meaning the variance exceeds the mean. The models yielded incidence rate ratios with 95 percent confidence intervals, quantifying how much the risk of stroke incidence or death changed month over month. For incidence, the ratio was 1.005 (95 percent CI: 1.004 to 1.006); for mortality, it was 1.006 (95 percent CI: 1.005 to 1.007). Both figures carry p-values below 0.001. The numbers look small, but their effect compounds relentlessly: a monthly increase of half a percent, sustained for years, translates into a substantially larger burden over a decade.

To understand what was driving the rise, the researchers turned to decomposition analysis using the Kitagawa method, a classical demographic technique that partitions change in a rate into contributions from population composition and from the rates themselves. Applied to the model estimates, the decomposition delivered an unambiguous verdict: the epidemiological effect was roughly twice as large as the population effect. In other words, most of the growth in stroke cases and deaths stems from rising incidence rates, not merely from more Ugandans being alive to fall ill. That distinction matters enormously for policy, because it implies that prevention, not demographic inevitability, holds the key to reversing the trend.

Before trusting the trends, the team subjected the data to rigorous diagnostic checks. Stationarity was assessed with unit root tests, and autocorrelation in the residuals was evaluated with the Ljung-Box test, standard procedures for confirming that a time series behaves in a statistically coherent way. Only after these checks did the researchers proceed to forecasting, using the exponential smoothing state-space model, known as ETS, which generates predictions with 80 percent prediction intervals that capture the plausible range of future values rather than a single point estimate.

The projections are sobering. By December 2028, monthly stroke cases are forecast to reach approximately 927, with an 80 percent prediction interval spanning 709 to 1,156. Monthly deaths are projected at roughly 126, with an interval of 94 to 160. The width of those intervals reflects genuine uncertainty in extrapolating health-system data, but even the lower bounds represent a marked escalation from current levels. If the trend holds, Uganda’s hospitals and stroke units will face a caseload in 2028 that dwarfs what they managed even a few years ago.

Stroke is already a leading cause of death and disability worldwide, and the global burden has shifted decisively toward low- and middle-income countries, where prevention infrastructure, acute care capacity, and rehabilitation services are often thinnest. Uganda exemplifies the pattern. Hypertension, diabetes, and other cardiovascular risk factors are rising across sub-Saharan Africa as diets, activity levels, and urbanization change, yet many patients never receive a diagnosis until a catastrophic vascular event occurs. The new findings give that familiar narrative a hard quantitative edge: the epidemic is not approaching, it is already underway and accelerating.

The authors are explicit about the implications. In the absence of effective prevention measures, the stroke burden will continue to climb through 2028 and likely beyond. They call for strengthened primary prevention, meaning population-level control of blood pressure, better detection and management of diabetes and atrial fibrillation, tobacco control, and dietary interventions. They also urge expanded acute stroke care capacity, including timely imaging, thrombolysis, and stroke units, which remain scarce across much of the region. And they emphasize improved surveillance, since the DHIS2 facility data that made this analysis possible still capture only part of the true disease picture, particularly among patients who never reach a health facility.

What makes the study notable beyond Uganda is its methodological clarity. By combining Bayesian count models, demographic decomposition, and formal forecasting within a single analytical pipeline, the researchers produced evidence that is both statistically defensible and directly actionable. The finding that epidemiological change outpaces population growth by a factor of two is the kind of result that can redirect national health planning: it says the crisis is behavioral and clinical, not demographic, and therefore preventable.

For a country whose health system already strains under infectious disease burdens alongside a growing tide of noncommunicable conditions, the forecast is a warning with a deadline. Every month of inaction adds roughly half a percent to stroke risk, compounding silently across a population of tens of millions. The mathematics of the trend is unforgiving, but so is the opportunity: interventions that bend the incidence curve even slightly would avert thousands of cases and deaths over the coming years. Uganda’s stroke epidemic, the data make clear, will be shaped by what happens now, not by fate.

Subject of Research: Rising stroke burden and forecasts in Uganda

Article Title: Rising stroke burden in Uganda: A national time-series analysis and forecasts, 2016–2028

Article References: Nakasendwa, S., Mutyoba, J. N., Ario, A. R., Omondi, E., Bashemera, B. B., Kakooza, F., Najjemba, L., Nabende, J. N., Guwatudde, D., & Kiragga, A. (2026). Rising stroke burden in Uganda: A national time-series analysis and forecasts, 2016–2028. PLOS One, 21(10), e0360963. https://doi.org/10.1371/journal.pone.0360963

Image Credits: AI Generated

DOI: 10.1371/journal.pone.0360963

Keywords: stroke, Uganda, time-series analysis, forecasting, noncommunicable diseases, public health surveillance, DHIS2, Bayesian regression, epidemiology, stroke prevention, health systems, sub-Saharan Africa

Brantingham has spent two decades studying car theft, but the Kia and Hyundai episode presented an unusually rich case. The surge began in several U.S. cities, including Los Angeles, before spreading nationwide. To understand it, his interdisciplinary team combined ideas from foraging ecology, social learning theory, criminology, and methods drawn from mathematical modeling and computer science. Working with data from the Los Angeles Police Department, the researchers built forecasts showing that the theft wave is not a short-lived spike tied to a fleeting internet trend, but a slow-moving phenomenon governed by the life cycle of the vehicles themselves.

The root of the problem lies in a manufacturing decision. From 2011 to 2022, Kia and Hyundai left engine immobilizers out of 20 car models. These electronic devices, standard anti-theft equipment in most modern vehicles, prevent an engine from starting without the correct key. Without one, stealing the affected cars became a relatively low-skilled crime. During the COVID-19 pandemic, when cars sat unattended on the street for long stretches and school and business closures left young people with more unstructured time, even inexperienced thieves had hours to experiment. The technique they learned was simple: pry apart the steering wheel casing, expose the starter, and turn the ignition with a plain USB cable.

Social media then amplified what had begun as a localized discovery. Videos tagged Kia Boys or Kia Challenge spread across YouTube and TikTok, some offering step-by-step instructions for the thefts. But Brantingham notes that even clips without explicit instructions were enough to signal to naive thieves that these models could be taken in minutes. Social media, he observed, is remarkably effective at creating experts when the required skill is low. The platforms did not invent the vulnerability, but they broadcast it, standardized it, and glamorized the joyrides that followed.

Perhaps the most counterintuitive finding concerns traditional media. The evidence from Los Angeles indicates that the vulnerability was initially discovered without the help of social media, and the same pattern held in Milwaukee, the birthplace of the Kia Boys posts, as well as in Denver. Print and cable news coverage of the theft wave may itself have encouraged copycats, the researchers suggest, extending the crime spree to places that viral videos never reached directly. The interplay between old and new media, rather than either alone, appears to have shaped the trajectory of the epidemic.

One of the most striking insights from the study is geographic. Crime hotspots, the researchers found, move with the underlying new and used car market. New vehicles tend to be sold in more affluent areas, held for a couple of years, and then resold. As cars age and depreciate, each change of hands shifts them toward less affluent neighborhoods. A vulnerable vehicle that began its life on an affluent street in Los Angeles will, some twenty to twenty-five years later, likely end its life in a far less prosperous one. The same pattern held for Honda Civics and Toyota Camrys, frequent targets in the 1980s, 1990s, and 2000s, and it will hold for the Kia and Hyundai models as well.

For people living in those less affluent areas, Brantingham explained, the crime hotspot comes to them. The movement is not about offenders relocating from one neighborhood to another, he said, but about ripe criminal targets migrating through the used car market. This means that by the time higher-income areas of Los Angeles have long forgotten the Kia Boys vulnerability, lower-income communities will still be absorbing the aging, unprotected vehicles as they drift toward the end of their life cycles. The burden of the crime wave, in other words, is not static; it travels with the cars.

The forecast that thefts will continue through at least 2042, and likely through 2050, follows directly from this logic. Because the vulnerable models remain on the road and continue to change hands, the pool of easy targets persists for decades. The research, funded by the Air Force Office of Scientific Research, offers a quantitative window into how social media shapes offline crime and how long the shadow of a single manufacturing decision can stretch across a vehicle fleet.

There are practical lessons for consumers. Brantingham points to the demonstrated success of electronic engine immobilizers, noting that Toyota introduced them on the Camry in 1992 and Honda on the Civic in 2001, with marked reductions in theft. His advice is straightforward: verify that any car you buy comes with an immobilizer as standard equipment, and if you own an older model without one, buy a steering wheel lock and use it correctly. Owners of vulnerable Kia and Hyundai models should also take advantage of manufacturer recalls that improve anti-theft technology, or consider trading in the vehicle for one with modern protections.

Policy has begun to catch up as well. A class action lawsuit brought by 36 states against Kia and Hyundai led to a settlement in 2025 in which both companies agreed to equip all future cars with industry-standard engine immobilizer anti-theft technology, provide ignition cylinder protectors, and offer some monetary compensation for cars damaged by thieves. For Brantingham, the episode carries a broader significance beyond one brand or one crime wave. We live in one of the safest times ever recorded, he noted, yet crime remains incredibly common, and documenting how people behave in complex online and offline environments is core to a better understanding of the human condition.

Subject of Research: Forecasting the impact of the viral Kia Boys social media trend on Kia and Hyundai car thefts in the United States

Article Title: Car owners beware! Lessons from the ‘Kia Boys’ crime spree

Article References: Car owners beware! Lessons from the ‘Kia Boys’ crime spree. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: Kia Boys, car theft, Hyundai, Kia, social media, engine immobilizer, UCLA, criminology, forecasting, used car market, TikTok, COVID-19

The researchers, led by Zhiheng Luo and Yunting Liu of the Affiliated Guangdong Second Provincial General Hospital of Jinan University, confronted a constraint familiar to every public health agency: a routine two-day reporting delay between when a case occurs and when it appears in the surveillance record. Rather than pretending this lag does not exist, they designed their forecasting pipeline around it. Their XGBoost model, an extreme gradient boosting algorithm, was tasked with predicting the increment in cases seven days ahead, using only information that would actually be observable at forecast time. This reporting-lag-aware design is the study’s central technical contribution, because most published forecasting models implicitly assume the forecaster can see the latest true case counts, an assumption that collapses in real-world operations.

The data spanned November 1, 2023 to July 31, 2026 and included physician-recorded or laboratory-supported cases of influenza, COVID-19 and Mycoplasma pneumoniae, alongside fever clinic visit counts and meteorological variables. The team engineered their predictors with care: most count-derived features were normalized to ratios, which stabilizes them when overall attendance fluctuates, while the latest observable case count was deliberately retained as a level anchor. This hybrid specification lets the model distinguish between a genuine rise in infections and a mere surge in care-seeking behavior, a distinction that absolute counts alone cannot make. The XGBoost forecast was then shrinkage-calibrated and combined with a rolling seasonal autoregressive integrated moving average, or SARIMA, forecast, blending machine learning flexibility with classical time series stability.

Performance was assessed over 363 scored target dates within a 371-day calendar evaluation window, using a battery of metrics including root mean square error, mean absolute error, R-squared, directional accuracy, alert sensitivity, positive predictive value and prediction-interval coverage. The framework achieved RMSE values of 16.50, 12.64 and 6.14 cases in the total population, adults and children respectively, with corresponding R-squared values of 0.640, 0.694 and 0.353. Directional accuracy, the proportion of days on which the model correctly predicted whether activity would rise or fall, ranged from 67.2 percent to 73.5 percent across the three series. For identifying days on which activity reached or exceeded twice the development-period median, a proxy for alert-worthy surges, sensitivity ranged from 58.1 percent to 78.4 percent.

The statistical comparisons were revealing. Using Diebold-Mariano tests, which compare forecast accuracy between two methods on the same target dates, the framework significantly outperformed persistence, the naive rule that tomorrow’s increment equals today’s, in the total and adult series, with p-values of 0.001 and 0.003. However, improvements over a simple seven-day moving average or a rolling SARIMA forecast were not statistically significant. In children, no comparison reached statistical significance at all, likely reflecting the smaller and noisier case counts in that age group. The practical improvement over persistence amounted to roughly one case per day in mean absolute error, a gain the authors describe as modest supplementary situational awareness rather than a transformative forecasting capability.

Deep learning comparators, including long short-term memory networks and Transformer models, were also evaluated, alongside an oracle-weather SARIMAX benchmark that used future meteorological information as an upper reference. The fact that gradient boosting on carefully engineered routine features held its own against these architectures underscores a recurring lesson in applied forecasting: with limited training data and strong autoregressive structure, simpler models with thoughtful feature design often match or beat more complex ones. The study followed the TRIPOD+AI reporting guidelines and used tree-structured Parzen estimator optimization for hyperparameter tuning, with SHAP values available for interpreting feature contributions.

Perhaps the most important finding emerged from the extended evaluation period, during which the absolute-count specification of the model performed noticeably worse, with RMSE rising to 20.32 and R-squared falling to 0.453. Over that same interval, fever clinic attendance declined by 31.5 percent while the case-to-visit ratio increased by 31.6 percent. This divergence in care-seeking behavior destabilized the count-based predictors the model relied upon, demonstrating that forecasting performance depends critically on the stability of the underlying data-generating process. A model trained when patients flock to fever clinics may misjudge activity when they stay home, even if the infection itself behaves similarly. This vulnerability to distributional shift is a fundamental challenge for all surveillance forecasting systems, not just this one.

The authors are candid about the limits of their work. The study used data from a single sentinel hospital, and the retrospective design means the model was not prospectively registered or tested before analysis. They explicitly recommend prospective multi-site validation before broader use, noting that performance under shifts in care-seeking behavior remains the key uncertainty. The framework’s value proposition is also framed carefully: it provides supplementary seven-day situational awareness from routine data, not a replacement for the judgment of epidemiologists or for laboratory-based confirmation. In an era when machine learning forecasts are often oversold, this kind of measured evaluation is itself a public service.

For the science of epidemic forecasting, the Guangzhou study offers a template worth copying. It shows how to build a forecasting pipeline that respects the operational reality of reporting delays, how to specify predictors that are robust to attendance fluctuations, and how to evaluate performance with statistical rigor against meaningful baselines rather than straw men. It also demonstrates the discipline of publishing null or modest results: the finding that a sophisticated model barely beats a seven-day moving average is as informative as any headline-grabbing accuracy claim. As respiratory pathogens continue to circulate in overlapping waves of influenza, COVID-19 and Mycoplasma pneumoniae, tools like this one, honest about their lag-aware design and their limits, may become quiet but useful fixtures in the surveillance toolkit, provided they are validated across sites and monitored as care-seeking behavior evolves.

Subject of Research: Machine learning forecasting of respiratory disease surveillance under reporting delays

Article Title: Reporting-lag-aware machine learning framework for short-term forecasting of respiratory disease activity: a sentinel hospital study in Guangzhou

Article References: Luo, Z., Duan, W., Lin, Y., Wang, H., Liu, H., & Liu, Y. (2026). Reporting-lag-aware machine learning framework for short-term forecasting of respiratory disease activity: a sentinel hospital study in Guangzhou. BMC Infectious Diseases. https://doi.org/10.1186/s12879-026-14653-0

Image Credits: AI Generated

DOI: 10.1186/s12879-026-14653-0

Keywords: machine learning, disease surveillance, forecasting, reporting delay, XGBoost, respiratory infections, Guangzhou, sentinel hospital, Reporting-lag-aware, machine, learning, framework

The research, published open access in the International Journal of Data Science and Analytics, introduces OFTER, short for online forecasting via temporal embedding regression. Developed by Nikolas Michael, Mihai Cucuringu and Sam Howison, the pipeline is tailored for financial problems centered on mid-sized multivariate time series, such as forecasting next-day returns and traded volumes for exchange traded funds. Its core ingredients are two non-parametric models that the machine learning community has largely overlooked: k-nearest neighbors and generalized regression neural networks, integrated with a dimensionality reduction component and a novel correlation-driven weighting scheme.

The central obstacle the team had to overcome is the curse of dimensionality, the phenomenon whereby the amount of data required to estimate an unknown function grows exponentially with the dimensionality of the data set. Both k-nearest neighbors and generalized regression neural networks, despite being simple, fast methods requiring only a single hyperparameter, suffer badly from this problem when applied to high-dimensional feature matrices. OFTER attacks it from two directions. First, it projects the data into a lower-dimensional space using principal component analysis, retaining a fixed proportion of the total variance. Second, instead of the standard Euclidean norm, it employs a thresholded weighted norm driven by each feature’s predictive potential, so that dimensions carrying little information about the target are effectively silenced when measuring the similarity between observations.

The weighting scheme rests on what the authors call the one-sided maximal correlation, a modified version of the maximal correlation coefficient. Rather than transforming both a feature and the target to maximize their association, the method transforms only the feature, which is what a regression setting demands. The transformation is computed in a restricted function class of Bernstein polynomials, a basis justified by the Stone-Weierstrass theorem and familiar from computer-aided geometric design. The authors derive a closed-form solution via a Lagrangian optimization and prove in an appendix that the resulting transformation is consistent, converging at a rate of order N to the minus one half. Features whose transformed correlation with the target falls below a threshold are discarded, and the squared correlations of the survivors become the weights of the norm used for neighbor selection.

OFTER also builds on a hybrid tradition in time series modeling. Before any regression takes place, the target is tested for stationarity with an augmented Dickey-Fuller test and differenced if necessary, then screened for autoregressive and moving average effects. If such effects are identified, a grid search over ARIMA orders, selected by the Bayesian Information Criterion, decomposes the target into a time series signal and a residual component. The regression machinery then forecasts the residuals, in the spirit of earlier work combining ARIMA models with machine learning corrections.

What makes the pipeline genuinely online is the way its embedding evolves. Recomputing a principal component analysis at every time step would cost on the order of n d squared plus d cubed operations, which is prohibitive for streaming data. Instead, OFTER exploits rank-one updates of the eigensystem: when a new observation arrives, the updated covariance matrix differs from the old one by a single outer product, and the new eigenvalues are the roots of a secular equation that can be solved efficiently using a convexifying method. The authors also handle the re-centering that occurs as the running mean drifts, expressing it as two further rank-one updates. The result is an embedding that refreshes itself at a complexity of only order p squared d per iteration, where p is the reduced dimension.

Hyperparameters are handled by dynamic model selection rather than fixed choices. The pipeline runs generalized regression neural networks with a whole candidate set of smoothing parameters, k-nearest neighbors with a candidate set of neighbor counts, and a standard ordinary least squares model, all in parallel. At each time step, the model with the best cumulative performance over a recent lookback window is chosen to make the forecast, with ties split evenly among the winners. For return forecasting, the loss function is economic rather than statistical: minus the profit and loss of a hypothetical trading strategy based on the forecasted signs.

The empirical case is striking. On synthetic multivariate models, including linear, nonlinear and vector autoregressive generators, OFTER configurations generally matched or beat linear models, random forests, XGBoost and support vector machines, and were competitive with LSTM and GRU networks at a fraction of the running time. On real data covering ten ETFs, with features drawn from multi-day returns, absolute returns, log-volume changes, the three Fama-French daily market factors, the returns of 49 industry portfolios and the VIX index, OFTER outperformed all benchmark methods for the volume forecasting task in seven out of ten cases, with Diebold-Mariano p-values of zero up to four decimal places. For market excess returns, three of four OFTER strategies produced statistically significant Sharpe ratios, a test the authors run with an adjustment for the skewness and kurtosis of returns, and none of the benchmark methods achieved significance except one Bi-LSTM configuration in the highest-confidence portfolios.

Beyond raw forecasting accuracy, the pipeline yields two byproducts that practitioners may find as valuable as the predictions themselves. A feature importance score can be derived analytically by measuring the sensitivity of the embedding-space distance to each original feature, giving the user a transparent account of which inputs drive the forecast. And because the method measures distances between observations in the embedding, any newly arrived time point whose nearest neighbors are still far away can be flagged as an outlier. When the authors applied this test to their ETF data with a lookback of 600 days, the flagged days clustered around historically significant events: the aftermath of the September 11 attacks, the 2002 market downturn, the 2007 drop in US and China stocks, the October 2008 crash, the May 2010 flash crash, the eve of Black Monday in 2011, and the November 2016 US election.

The authors are careful about the limits of their claims. OFTER struggles to discover nonlinear interaction effects between covariates, a general weakness of models and data sets of this size, and the financial evaluation focuses on predictive signal quality rather than fully implementable trading performance, with transaction frictions such as price slippage and market impact left to future work. They also note that the one-sided maximal correlation metric is of independent interest as a fast additional step for other simple regression methods. But the broader message is unambiguous: while deep learning holds significant promise for time series forecasting, traditional methods carefully integrating mainstream tools remain very competitive alternatives, with the added benefits of scalability and interpretability. For the quant facing a mid-sized, low signal-to-noise data set and a live market clock, the simplest tools, thoughtfully combined, may once again be the sharpest ones on the desk.

Subject of Research: Machine Learning

Article Title: Online forecasting for financial time series

Article References: Michael, N., Cucuringu, M., & Howison, S. (2026). Online forecasting for financial time series. International Journal of Data Science and Analytics, 22(1), Article 344. https://doi.org/10.1007/s41060-026-01302-7

Image Credits: AI Generated

DOI: 10.1007/s41060-026-01302-7

Keywords: financial time series, online forecasting, machine learning, quantitative finance, dimensionality reduction, k-nearest neighbors, time series analysis, OFTER, Online, forecasting, financial, time

The three regularizers introduced in the paper each address a known failure mode of deep forecasting models in clinical settings. Frequency-aware embedding shrinkage discourages the model from relying on embedding features that appear rarely in the training data, which helps prevent overfitting to unusual measurement patterns that may not generalize across hospitals with different charting practices. Historical group-entropy regularization encourages the model to distribute its attention across relevant historical windows rather than collapsing onto a narrow slice of the past, a behavior that can make forecasts brittle when a patient’s trajectory departs from recent trends. Shock-alignment regularization is perhaps the most clinically motivated of the three: ICU deterioration often announces itself through abrupt transitions rather than gradual drift, and models trained on mean-squared error alone tend to smooth over these transitions, producing forecasts that are accurate on average but useless precisely when they matter most. By explicitly aligning the model’s predictions with abrupt state changes during training, the authors nudge the learning process toward the regime shifts that carry the greatest clinical consequence.

State-balanced sliding-window sampling similarly addresses a structural imbalance in ICU data. Patients spend most of their stays in relatively stable physiological states, so a naive sampling strategy that draws training windows uniformly from the record will overwhelmingly expose the model to stable periods. The model then becomes well calibrated for the common case but poorly trained on the transitions that precede clinical events. By balancing the composition of training windows across patient states, the sampling scheme ensures that the model sees enough examples of both stability and change to learn the boundary between them. This echoes a well-established principle in imbalanced learning, but applying it at the level of temporal windows rather than class labels is a design choice tailored to the forecasting task.

The decision to train on MIMIC-IV and validate on the MIMIC-III CareVue subset and eICU deserves attention because of how the cohorts were constructed. The CareVue subset was deliberately curated by PhysioNet to exclude patients who appear in MIMIC-IV, which means the external validation on that cohort tests generalization across both time and patient identity rather than merely re-scoring familiar individuals. The eICU database, meanwhile, aggregates data from many hospitals across the United States with heterogeneous charting systems, staffing patterns, and patient populations. Performance that holds up on eICU therefore provides some evidence that the learned representations are not simply memorizing the idiosyncrasies of a single institution’s data pipeline. The gap between development and validation cohorts in this study is unusually rigorous for the field, where many published models are trained and tested on overlapping samples from the same database.

The choice of eight forecasting targets spans the physiological variables that dominate ICU monitoring: heart rate, blood pressure measurements, respiratory rate, temperature, oxygenation, and laboratory values such as creatinine. Creatinine is particularly consequential because it is the primary laboratory marker for kidney function and the basis for acute kidney injury staging under standard clinical criteria. Forecasting creatinine accurately over a horizon of hours could, in principle, give clinicians a head start on detecting kidney injury before laboratory values confirm it. The finding that OmniTFT ranked second for creatinine, behind a task-specific comparator, is an honest result that illustrates the trade-off inherent in the shared-backbone design: a single framework that handles many targets will sometimes concede a small margin to a specialist model tuned for one. The authors’ argument is that the marginal loss on any single target is outweighed by the operational simplicity of maintaining one system across the monitoring panel.

The retrospective acute kidney injury analysis is the study’s most clinically oriented experiment, and its design warrants careful reading. The authors compared risk discrimination for incident moderate-to-severe AKI using two inputs: the complete observed record alone, and the observed record augmented with OmniTFT’s forecast of the subsequent twenty-four hours. The improvement in AUROC from 0.693 to 0.763, with a DeLong test p-value of 0.009, indicates that the forecasted future adds information beyond what the observed past contains. This is a nontrivial claim, because a forecasting model could in principle merely repackage information already present in the observed data. The demonstrated gain suggests the model extracts predictive signal from temporal patterns that a static risk score applied to the observed record does not capture. The screening framing is also notable: when thirty percent of patients were flagged for review, the forecast-augmented approach identified 73.3 percent of subsequent events versus 57.8 percent for the observed record alone. For a triage workflow, that difference translates into meaningfully fewer missed cases at a fixed review capacity.

Several caveats temper these findings. The AKI analysis was retrospective and conducted in a disease-enriched cohort, meaning the prevalence of the outcome was substantially higher than in a general ICU population. Discrimination metrics such as AUROC are known to shift with prevalence, and a model that performs well in an enriched cohort may show different operating characteristics when deployed for unselected screening. The authors themselves frame the result as motivation for broader external validation rather than as evidence of deployment readiness. Additionally, the comparison against frozen pretrained foundation models such as Chronos-2 and TimesFM 3.0 reflects a rapidly evolving landscape; these general-purpose time-series models were not fine-tuned on ICU data, so the comparison establishes a floor rather than a ceiling for what foundation-model approaches might achieve with domain adaptation.

The multi-seed evaluation with ten random training runs and a fixed patient partition addresses a problem that has plagued the machine learning literature on clinical prediction: sensitivity to random initialization. Deep learning models can exhibit substantial variance in performance across seeds, and single-run results sometimes overstate a method’s reliability. Reporting results across ten seeds, along with component ablations showing the contribution of each regularizer and the sampling strategy, gives readers a more honest picture of expected performance and of which design elements carry the weight. Ablation studies of this kind are increasingly expected by reviewers in the medical AI field, where concerns about fragile or irreproducible results have prompted calls for standardized reporting.

The broader context for this work is the growing effort to build proactive rather than reactive clinical decision support. Traditional early-warning scores in the ICU, such as those based on vital sign thresholds, flag deterioration only after physiological changes have already registered in the chart. Forecasting models aim to move the detection point earlier in time, giving clinicians a window to intervene before a crisis fully develops. Whether such forecasts improve outcomes in practice depends on many factors beyond model accuracy, including how alerts are presented, how clinicians respond to them, and whether the alert burden contributes to alarm fatigue, a well-documented problem in critical care. The authors are appropriately measured on this point, positioning OmniTFT as a forecasting component that could be integrated into decision support systems rather than as a standalone intervention.

The heterogeneity problem that motivates the shared-backbone design is worth situating in the data landscape of modern EHRs. Vital signs in the ICU are typically recorded at intervals ranging from hourly to once per minute depending on the patient’s acuity, while laboratory measurements are drawn on clinical indication, producing irregular and sparse series. Oxygenation indices derived from ventilator settings add another layer of complexity. Most prior forecasting work has addressed one variable family at a time, which produces a proliferation of single-purpose models that are difficult to maintain, validate, and integrate. A unified framework that preserves target-specific temporal grids while sharing learned representations offers an engineering path toward systems that forecast an entire monitoring panel coherently. This systems-level argument, as much as the headline accuracy numbers, may prove to be the study’s most durable contribution.

Finally, the study’s grounding in publicly available databases supports reproducibility in a field where proprietary data often limits independent verification. MIMIC-IV, MIMIC-III, and eICU are accessible to credentialed researchers under PhysioNet’s data use agreement, which means other groups can attempt to replicate the reported results, stress-test the ablation findings, and benchmark alternative architectures on identical partitions. The explicit separation of development and external validation cohorts, the multi-seed protocol, and the published code conventions of the Frontiers in Digital Health venue together set a methodological standard that future EHR forecasting studies would do well to follow. As clinical decision support systems mature from retrospective demonstrations toward prospective evaluation, studies with this level of validation discipline will form the evidentiary foundation on which safe deployment ultimately depends.

Subject of Research: OmniTFT: an EHR-driven multi-target forecasting model for ICU clinical decision support systems

Article Title: OmniTFT: an EHR-driven multi-target forecasting model for ICU clinical decision support systems

Article References: Xu, W., Dai, Y., Yang, Y., Zhang, Z., Loza, M., Zhang, W., Cui, Y., Zeng, X., Park, S.-J., & Nakai, K. (2026). OmniTFT: an EHR-driven multi-target forecasting model for ICU clinical decision support systems. Frontiers in Digital Health, 8, Article 1965956. https://doi.org/10.3389/fdgth.2026.1965956

Image Credits: AI Generated

DOI: 10.3389/fdgth.2026.1965956

Keywords: OmniTFT, EHR-driven, multi-target, forecasting, model, clinical, decision, support, systems, scientific research

The study drew on monthly and annual reported case counts from China’s Public Health Science Data Center, covering 192 consecutive monthly observations for each virus from January 2005 to December 2020. Reported incidence was calculated as cases per 100,000 permanent residents, using year-end population figures from the Hainan Provincial Bureau of Statistics as the denominator. To place the provincial data in context, the team also extracted national Global Burden of Disease (GBD) estimates for China spanning 1990 to 2023, and they obtained annual tourist visit statistics from the provincial tourism department. No missing or duplicate observations were identified, and no study years were excluded, giving the analysis a complete and unbroken record.

Joinpoint regression, a technique that identifies inflection points in a time series and estimates the annual percent change (APC) within each segment, revealed a distinctly phased pattern for reported hepatitis B incidence in Hainan. Rates fell from 2005 to 2008 with an APC of −9.10 percent per year, rose from 2008 to 2018 with an APC of +10.50 percent per year, and then showed a non-significant decline through 2020. Across the full sixteen years, the average annual percent change (AAPC) was +3.37 percent, a statistically significant overall increase. Reported hepatitis C followed a different course: it climbed continuously from 2005 to 2015 at a striking +32.10 percent per year, then declined at −10.90 percent per year through 2020, yielding an overall AAPC of +14.54 percent. By contrast, the national GBD estimates showed overall AAPCs of −0.54 percent for hepatitis B and −1.52 percent for hepatitis C, both modestly downward over their respective periods.

The divergence between provincial reported data and national modeled estimates is a central finding. The authors caution that the two data sources differ in definition and origin, so national GBD trends serve only as contextual references. Still, the mid-period rise in reported hepatitis B in Hainan is unlikely to reflect a simple failure of vaccination strategy. China added hepatitis B vaccination to its childhood immunization program decades ago, and studies from Guangzhou and Xiamen have documented declining reported incidence in those cities. The Hainan increase may instead reflect expanded screening, greater detection of previously acquired chronic infections, repeated healthcare encounters, rising healthcare accessibility, and changes in reporting-system sensitivity, layered on top of genuine transmission dynamics and increasing vaccine-derived protection in younger birth cohorts. China continues to carry a large reservoir of chronic hepatitis B infection, and insufficient diagnosis, treatment, and follow-up remain major barriers to elimination.

For hepatitis C, the pre-2015 surge in reported incidence plausibly mirrors improvements in testing capacity, strengthening of the reporting infrastructure, and increased screening that uncovered previously undiagnosed infections. The post-2015 decline coincided with broader gains in blood safety, awareness, screening, and access to antiviral therapy, including the transformative arrival of direct-acting antivirals (DAAs), which can cure chronic hepatitis C infection in most treated patients. Yet the authors note that Hainan-specific information on testing and treatment coverage was unavailable, so the analysis cannot determine whether these factors caused the observed decline. A large pool of undiagnosed infections may continue to limit progress, since DAAs alone cannot reach patients who are never screened or linked to care.

Seasonality, often a signature of environmentally driven pathogens, was conspicuously absent for both viruses. The team applied seasonal-trend decomposition using Loess (STL), an additive method that splits each monthly series into trend, seasonal, and remainder components with a twelve-month seasonal period. Kruskal–Wallis tests comparing the distributions of the STL seasonal components across calendar months found no statistically significant differences for hepatitis B (H = 11.10, P = 0.435) or hepatitis C (H = 2.81, P = 0.993). Descriptive seasonality indices were highest in July for hepatitis B (107.8) and April for hepatitis C (107.4), and lowest in February for both (78.3 and 73.9, respectively). The February trough may relate to reduced healthcare seeking and delayed testing during the Spring Festival holiday, but the authors stress that these fluctuations likely reflect reporting variation rather than seasonal transmission, and that long-term trends and population-structure changes carry greater public health relevance than isolated monthly peaks.

To project future trajectories, the researchers fitted seasonal autoregressive integrated moving average (SARIMA) models to the monthly series. For hepatitis B, the selected model was SARIMA (0,1,2) × (2,1,0) with a twelve-month seasonal cycle, which had the lowest Akaike information criterion among candidates and passed the Ljung–Box residual autocorrelation test at lag 12 (Q = 10.771, P = 0.549). For hepatitis C, SARIMA (1,1,1) × (0,0,1) was chosen because an alternative with slightly lower information criteria retained significant residual autocorrelation; the selected model showed acceptable diagnostics at lag 24 (Q = 31.162, P = 0.074). Forecasting to 2030, the annualized projected reported hepatitis B incidence was 12.00 per 100,000 population, with a 95 percent prediction interval spanning roughly 0.00 to 121.20, numerically close to the study-defined reference value of 12.20. The hepatitis C projection was 22.54 per 100,000 (approximately 0.00 to 39.10), about 6.4 times the study-defined reference value of 3.55. The reference values were defined as 10 percent of the corresponding 2015 reported incidence rates, echoing the WHO framework of a 90 percent reduction in new infections by 2030, but the authors insist they are not equivalent to WHO elimination targets.

The wide prediction intervals, generated from 10,000 simulated forecast paths, signal substantial long-term uncertainty, and the authors repeatedly warn that the projections are conditional extrapolations of historical reporting patterns. A declining trend, they note, does not necessarily imply proximity to elimination. Hepatitis B control depends on sustained immunization coverage, prevention of mother-to-child transmission, adult diagnosis and treatment, and management of chronic infection, whereas hepatitis C elimination hinges on active screening, rapid linkage to treatment after diagnosis, and improved treatment coverage. Insufficient screening, incomplete linkage to care, declining treatment uptake, and ongoing transmission in high-risk populations may all weaken control efforts even in the DAA era.

Sensitivity analyses highlighted the disruptive fingerprint of the COVID-19 pandemic year. When Joinpoint models were refitted excluding 2020, the overall AAPC rose from 3.37 to 4.96 percent per year for hepatitis B and from 14.54 to 16.43 percent per year for hepatitis C, indicating that the 2020 observations deviated from prior reporting patterns and pulled the long-term trend estimates downward. Out-of-sample SARIMA forecasts built on 2005–2019 data predicted a mean monthly reported hepatitis B incidence of 15.11 per 100,000 for 2020, but the observed value was 10.98, some 27.3 percent lower; for hepatitis C, the forecast of 1.92 exceeded the observed 1.74 by 9.5 percent. Reduced healthcare seeking, testing, and reporting during pandemic-control measures, consistent with findings from China, Spain, and Italy, likely explain these shortfalls rather than any sudden improvement in prevention.

Finally, the study explored ecological associations between annual tourist visits, which grew from roughly 17 million in 2005 to about 83 million in 2019 before falling to approximately 56 million in 2020, and reported hepatitis incidence. In raw annual analyses, tourist visits correlated strongly with both reported hepatitis B (r = 0.871) and hepatitis C (r = 0.879), both P < 0.001, and the associations persisted after excluding 2020. However, because both series shared pronounced long-term trends, the team repeated the correlations on first-differenced and linearly detrended data. Only the hepatitis B association survived (r = 0.567 after differencing, P = 0.028; r = 0.691 after detrending, P = 0.003), while the hepatitis C association vanished (r = 0.070, P = 0.804 after differencing). The authors interpret this cautiously: tourism volume may reflect concurrent changes in population mobility, healthcare use, screening activity, socioeconomic development, and reporting capacity rather than any direct causal effect on transmission. The analyses were aggregated and exploratory, lacking individual-level data on age, sex, vaccination status, transmission route, testing, and treatment. The authors conclude that Hainan should maintain hepatitis B immunization and screening while prioritizing hepatitis C screening, diagnosis, and linkage to treatment, supported by continued surveillance among mobile and key populations as the province’s free trade port development accelerates population mobility.

Subject of Research: Infectious Disease Epidemiology

Article Title: Trends and forecasting of reported hepatitis B and hepatitis C incidence in Hainan Province, China, 2005–2020

Article References: Wang, J., Li, S., Fu, C., Su, Y., Fu, H., Tian, Y., & Lan, C. (2026). Trends and forecasting of reported hepatitis B and hepatitis C incidence in Hainan Province, China, 2005–2020. Frontiers in Epidemiology, 6, Article 1899759. https://doi.org/10.3389/fepid.2026.1899759

Image Credits: AI Generated

DOI: 10.3389/fepid.2026.1899759

Keywords: hepatitis B, hepatitis C, Hainan Province, time-series analysis, SARIMA forecasting, viral hepatitis surveillance, China, Trends, forecasting, reported, hepatitis, incidence

A team of Finnish researchers has now built the tool that could change how the world judges its forecasts. Zhi-Song Liu of LUT University, Michael Boy of LUT University and the University of Helsinki, and Risto Makkonen of the Finnish Meteorological Institute have created TailWeather, a global dataset of climatological extreme weather events described in a preprint currently under review for the journal Earth System Science Data. The dataset covers every land area on Earth from 1981 through 2022 and into early January 2023, resolving the planet into a grid of cells 0.25 degrees on a side, roughly 25 to 28 kilometers at mid-latitudes. For each cell and each day, TailWeather labels whether the weather qualified as a heatwave, a cold wave, heavy precipitation, or extreme wind, and for each month it flags meteorological drought.

The technical insight behind TailWeather is deceptively simple. Rather than defining extremes by their consequences, which requires knowing where damage was recorded and reported, the researchers defined them statistically, relative to each location’s own climate. Every grid cell is compared against its local 1991 to 2020 climatology, so a heatwave in Helsinki is measured against Helsinki’s normal summer, not against the tropics. This means a temperature that would be unremarkable in Death Valley can be flagged as an extreme event in Lapland, because what matters for infrastructure, agriculture, and human physiology is the deviation from what a place is built to expect.

Each label in the dataset carries two layers of information. The first is an ordinal severity tier, a ranking of how far into the statistical tail an event falls. The second is a continuous intensity score expressed relative to the local climate, which allows users to shift the threshold defining what counts as extreme without ever recomputing the underlying climatology. This design choice matters enormously for reproducibility. A researcher studying moderate heat stress and another studying lethal heat can both work from the same dataset, simply choosing different cutoffs within the score’s resolution, and their results will remain directly comparable.

To validate that these statistical tails actually correspond to real-world harm, the team compared TailWeather labels against EM-DAT, one of the most widely used international disaster catalogs. The comparison produced a striking and consistent pattern: for all five hazard types, the stricter the tail definition, the more strongly the events were enriched in recorded disaster impacts. At the best-balanced threshold, heatwave tails were 9.3 times more likely to coincide with recorded disasters than average weather, while extreme wind tails showed a 1.6-fold enrichment. Cold waves, heavy precipitation, and drought fell between these bounds. Rarity, in other words, is not a perfect proxy for damage, but it is a meaningful one, and the relationship strengthens as events become more exceptional.

Yet the validation also exposed a sobering blind spot in the global disaster record. The researchers calculated that 80.1 percent of the world’s land population lives in regions where EM-DAT is too sparse to test the link between climatological extremes and recorded harm locally for at least one hazard type. Disaster catalogs are shaped by reporting capacity, media attention, and political priorities, and vast stretches of Africa, Asia, and South America remain effectively unmapped in terms of documented weather impacts. This is why the authors emphasize that TailWeather complements rather than replaces disaster catalogs: it provides a physically consistent target everywhere, including the places where human records of suffering are thinnest.

The dataset’s most viral finding, however, concerns the AI models themselves. When the researchers tested leading machine-learning forecasting systems against TailWeather’s extreme event labels, the results were humbling. Models that achieve remarkable accuracy on ordinary days, hitting that sub-kelvin temperature target at day one, reached a critical success index below 0.5 on heatwave tails, meaning they missed or falsely flagged more than half of the extreme heat events. And their performance was worst of all for extreme wind, the hazard that drives much of the structural damage in storms. The models that look brilliant on paper may be quietly failing exactly when it matters most.

This weakness is not accidental. Machine-learning weather models are trained to minimize average error across billions of examples, and the mathematics of that optimization naturally prioritizes the frequent, well-sampled conditions that dominate the training data. Rare extremes are, by definition, underrepresented, and the loss functions that drive learning offer little reward for nailing a once-in-a-decade storm. Traditional numerical models, which solve physical equations rather than learn statistical patterns, have their own biases but are not subject to this particular failure mode in the same way. The result is that the AI revolution in weather forecasting, while genuinely transformative for routine prediction, may be building its impressive averages on a foundation that crumbles under extreme conditions.

The implications ripple outward across the entire weather enterprise. Insurance companies pricing catastrophe risk, governments planning heat-action plans, energy grid operators stress-testing supply, and humanitarian organizations prepositioning relief all depend on knowing when forecasts can be trusted and when they cannot. If the leading models systematically underpredict the very events that cause the most damage, then every downstream decision built on those forecasts inherits that blind spot. TailWeather gives these users, and the model developers themselves, a standardized global benchmark for measuring tail skill, the forecasting equivalent of a crash test for the AI era.

The dataset is openly available, distributed under a Creative Commons Attribution 4.0 license, and hosted with an interactive computing environment that allows researchers to explore the labels directly in a browser. The authors frame TailWeather as a reusable physical target for studying and evaluating weather extremes, one that can serve training, validation, and benchmarking across the rapidly growing ecosystem of machine-learning forecast systems. As climate change pushes more of the planet’s weather into unprecedented territory, and as AI models increasingly become the default tool for prediction, the ability to test forecasts specifically against the rare and the extreme is no longer a luxury. It is the difference between knowing the average weather and being prepared for the days that change everything.

Subject of Research: TailWeather: from tail to extremes, a global climatological dataset for machine-learning weather forecasting

Article Title: TailWeather: from tail to extremes, a global climatological dataset for machine-learning weather forecasting

Article References: Liu, Z.-S., Boy, M., & Makkonen, R. (2026). TailWeather: from tail to extremes, a global climatological dataset for machine-learning weather forecasting. https://doi.org/10.5194/essd-2026-792

Image Credits: AI Generated

DOI: 10.5194/essd-2026-792

Keywords: TailWeather, tail, extremes, global, climatological, dataset, machine-learning, weather, forecasting, scientific research