The team, led by co-corresponding authors Dong Keon Yon of Kyung Hee University College of Medicine and Jiseung Kang, classified children as prenatally exposed if their mothers received two or more opioid prescriptions during pregnancy. That threshold was chosen to distinguish sustained exposure from a single, isolated dispensing event. The exposed children were compared in two ways: against unexposed children from the general population, and, in a more stringent test, against their own unexposed siblings, a design that partially controls for the family environment and shared genetics.

The headline result was striking. Compared with unexposed children, those with prenatal opioid exposure had a 28 percent higher rate of hospital admission, with an adjusted incidence rate ratio of 1.28 and a 95 percent confidence interval of 1.28 to 1.29. When the comparison was restricted to siblings within the same family, the association shrank to an adjusted rate ratio of 1.08, with a confidence interval of 1.07 to 1.10, but it did not disappear. That pattern, the authors argue, is consistent with a genuine contribution of the drug itself alongside a substantial contribution from factors shared within households.

The study also uncovered a clear dose-response relationship, one of the criteria epidemiologists look for when weighing whether an association is causal. Children whose mothers received low doses of opioids showed an adjusted rate ratio of 1.19, while those whose mothers received high doses showed a ratio of 1.40. In other words, the more opioid exposure a fetus experienced, the greater the subsequent risk of hospital admission in childhood. The elevated rates spanned multiple categories of illness, including infections as well as respiratory, gastrointestinal, and genitourinary conditions, a breadth that points toward possible effects on the developing immune system rather than damage confined to a single organ system.

Perhaps the most consequential finding was temporal. The association between prenatal opioid exposure and hospital admission was strongest in the earliest years of life and was most pronounced up to age five. Joint first author Yesol Yim described the fading of the signal as the study’s biggest surprise. The excess risk was clearest in the first year of life and had largely disappeared after age five, a pattern she interpreted as evidence of a window of vulnerability in early infancy rather than a lifelong effect, and one in which pediatric follow-up can make a real difference. For clinicians, that time-limited risk profile transforms an abstract statistical association into an actionable surveillance strategy.

The biological rationale for such a link is plausible, though not yet proven. Prenatal opioid exposure has been associated in prior research with disruptions to infant development and to immune system function, mechanisms that could plausibly increase susceptibility to the infections and respiratory illnesses that dominated the admission data. Opioid prescriptions and doses have risen considerably in clinical practice over recent decades, and the scale of prenatal exposure is far from trivial. In the United States, data have shown that between 14 and 22 percent of women filled an opioid prescription during pregnancy, making even a modest relative increase in childhood hospitalizations a meaningful public health concern.

What sets the South Korean study apart is the unusual cleanliness of its exposure data. As co-corresponding author Jiseung Kang explained, opioids in South Korea are obtained almost entirely by prescription, so the researchers could examine prescribed opioid exposure in pregnancy with unusual precision. Following nearly three million children for up to fourteen years gave the team the statistical power to examine dose, timing, and specific causes of admission separately, and Kang argued that the consistency of the pattern across all of these analyses is what makes the finding difficult to dismiss. Prior studies in this field have been hampered by confounding factors, limited sample sizes, or restricted follow-up durations, shortcomings this cohort was designed to overcome.

The authors themselves urge caution, and their caveats deserve as much attention as the headline numbers. The study relied on electronic health records, which record prescriptions written and dispensed but may not reflect the medication a patient actually took. Unmeasured factors shared within families, such as parental health behaviors, socioeconomic stressors, or genetic predispositions, may still partly explain the observed associations, and the narrowing of the effect in the sibling comparison is direct evidence of how much such factors matter. The sibling analysis, while powerful, cannot eliminate influences that vary between siblings born at different times or under different circumstances.

The research team was careful to frame its findings as a call for better care rather than a prohibition. Joint first author Hyeon Jin Kim emphasized that the message is not that pregnant women should never receive opioids, since for some patients pain relief is essential. Instead, the recommendation is that opioids in pregnancy be prescribed at the lowest effective dose for the shortest possible time, and that children born after such exposure may benefit from closer routine health checks, particularly during infancy when the risk was found to be highest. Co-corresponding author Dong Keon Yon echoed that framing, noting that the dose-dependence points to a possible biological effect while acknowledging that shared family factors clearly explain part of what the study observed.

The study, an observational analysis of nationwide health insurance data rather than a randomized trial, was funded by South Korea’s Ministry of Science and ICT, Ministry of Health and Welfare, National Research Foundation, and Ministry of Education, none of which had a role in the study design, data collection, analysis, or manuscript preparation. The authors declared no competing interests. For families and clinicians, the practical takeaway is measured: prenatal opioid exposure, as captured by two or more maternal prescriptions, is associated with a modestly elevated, dose-dependent, and age-limited risk of childhood hospital admission, concentrated in infancy and driven largely by infections and respiratory illness. That risk, while real, is neither uniform nor permanent, and it appears most amenable to mitigation through vigilant postnatal monitoring in the first years of life, when the window of vulnerability is open and pediatric care can respond.

Subject of Research: The association between prenatal opioid prescription exposure and childhood hospital admission rates in a nationwide South Korean birth cohort.

Article Title: Opioid prescriptions during pregnancy linked to higher rates of childhood hospital admission

Article References: Opioid prescriptions during pregnancy linked to higher rates of childhood hospital admission. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: opioids, pregnancy, prenatal exposure, hospital admission, pediatrics, PLOS Medicine, South Korea, birth cohort, epidemiology, childhood infections, dose-response, electronic health records

Ultra-processed foods, defined under the widely used NOVA classification system as industrial formulations assembled largely from substances derived from foods plus additives, flavorings, and cosmetic ingredients, now account for more than half of the energy intake among U.S. adults. Their rapid displacement of freshly prepared meals has coincided with rising rates of several chronic diseases, and observational studies have repeatedly linked high UPF consumption to elevated risks of cardiovascular disease, type 2 diabetes, and certain cancers. Colorectal cancer has been of particular interest because diet is one of its few modifiable risk factors, and because incidence in younger adults has been climbing for reasons that remain poorly understood. Yet the new analysis indicates that the processing category itself may be a poor predictor of cancer risk when the nutritional character of the foods within it is taken into account.

To untangle these relationships, the research team turned to one of the largest dietary cohorts ever assembled: the NIH-AARP Diet and Health Study. Between 1995 and 1996, 461,682 adults aged 50 to 71 years completed a detailed food frequency questionnaire documenting their habitual eating patterns. The researchers used these responses to estimate each participant’s intake of ultra-processed foods and then followed the cohort through 2018, a median of 18.8 years, during which incident colorectal cancer cases were identified through linkage with cancer registries. This combination of scale and follow-up duration gives the analysis statistical power that few dietary studies can match, allowing the investigators to detect even modest associations between food groups and cancer risk.

The headline result was striking in its simplicity: when all ultra-processed foods were considered together, there was no association between total UPF intake and colorectal cancer incidence. Participants who ate the most UPF overall were not more likely to develop colorectal cancer than those who ate the least. This null finding stands in contrast to the monolithic framing that has dominated public discussion of ultra-processed foods, in which the category is often treated as a single, uniformly hazardous block of the food supply. The data suggest that this framing obscures more than it reveals, because the health consequences of a food depend on what the food actually is, not merely on how it was made.

When the researchers disaggregated the UPF category into its constituent food groups, the picture changed dramatically. Higher intake of whole grains, fiber, dairy products, and calcium derived from ultra-processed sources was associated with a lower incidence of colorectal cancer, while meat from ultra-processed sources was associated with a higher incidence. In other words, an ultra-processed whole grain bread and an ultra-processed sausage sit at opposite ends of the risk spectrum despite sharing the same processing label. The associations observed within these food groups were consistent with long-standing dietary guidance for colorectal cancer prevention, which emphasizes whole grains, fiber, dairy, and calcium while recommending limits on red and processed meat.

The team also performed a substitution analysis that offers a practical takeaway for readers. Replacing 10 percent of gram weight from ultra-processed food intake with minimally processed foods was associated with a 2 percent lower incidence of colorectal cancer. While the magnitude of this effect is modest, it reinforces the value of shifting the overall balance of the diet toward fresh and minimally processed options. The result also implies that the potential benefit of reducing UPF consumption lies less in avoiding the processing category as a whole and more in the specific nutritional substitutions that such a shift typically involves, such as trading processed meats for vegetables, legumes, and whole foods.

These findings arrive at a moment when ultra-processed foods have become one of the most hotly debated topics in nutrition science and public health policy. Several countries have considered front-of-package warnings or taxation schemes targeting the UPF category, and recent dietary guidelines in some nations have begun to explicitly recommend limiting ultra-processed foods regardless of their nutritional profile. The new study does not settle that policy debate, but it does provide a caution against treating the NOVA processing category as a proxy for healthfulness. A food’s position within the processing spectrum, the data suggest, is a far less reliable guide to colorectal cancer risk than the food’s actual composition, particularly its content of whole grains, fiber, dairy, calcium, and meat.

The researchers are careful to acknowledge the boundaries of their evidence. The study cohort consisted primarily of older, white Americans, and the findings cannot be generalized with confidence to younger, more diverse populations whose dietary patterns and cancer risk profiles may differ substantially. In addition, UPF intake was measured only once, at baseline in the mid-1990s, a period when ultra-processed foods were less ubiquitous in the American food supply than they are today. As eating habits changed and new UPF products became available over the following decades, the single baseline measurement may have underestimated participants’ actual long-term intake, potentially biasing the results in unpredictable directions. Future studies with repeated dietary assessments and more diverse cohorts will be needed to confirm whether the food-group pattern observed here holds in other populations and eras.

Loftfield summarized the implications in a statement accompanying the publication: in this large population of older U.S. adults, the team did not observe an association between total ultra-processed food intake and colorectal cancer incidence, but when the investigators examined food groups such as whole grains, dairy, and meat by their UPF or non-UPF source, the associations aligned with current dietary guidance for colorectal cancer prevention, such that higher whole grain and dairy intakes and lower meat intakes were associated with lower colorectal cancer incidence. The study, funded by the Intramural Research Program of the National Cancer Institute, with authors from the United States, France, the Netherlands, Brazil, and Australia, ultimately suggests that ultra-processed foods should not be treated as a monolith. For readers seeking to reduce their colorectal cancer risk, the message is refreshingly familiar: fill the plate with whole grains, fiber, dairy, and calcium, keep processed meat to a minimum, and let the food itself, rather than its processing label, guide the choice.

Subject of Research: Association between ultra-processed food intake and colorectal cancer incidence in the NIH-AARP Diet and Health Study cohort

Article Title: Total ultra-processed food intake not linked to colorectal cancer in large US cohort study

Article References: Total ultra-processed food intake not linked to colorectal cancer in large US cohort study. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: ultra-processed foods, colorectal cancer, NIH-AARP Diet and Health Study, PLOS Medicine, National Cancer Institute, dietary risk factors, whole grains, processed meat, NOVA classification, cohort study, nutrition epidemiology, cancer prevention

The study comes at a pivotal moment. Emerging evidence has suggested that immunity conferred by typhoid conjugate vaccines, or TCVs, may wane over time, raising concerns that a single dose in infancy might not provide lasting protection. This uncertainty has complicated efforts to craft vaccination policies, because the optimal timing of a first dose and the potential value of boosters depend heavily on how quickly protection fades. To resolve the question, the WHO brought together modeling teams operating different approaches, including two agent-based models that simulate individual people and their interactions, and two compartmental models that track populations moving between susceptible and infected states. Comparing multiple models allows researchers to distinguish findings that are robust from those that are artifacts of any single model’s structure.

The modeling teams calibrated their simulations against harmonized age-specific incidence data drawn from settings classified as medium, high, and very high typhoid incidence. This fitting process ensured that each model reproduced the observed patterns of who gets sick and at what ages, which is essential because the best age for vaccination depends on when disease risk peaks in a population. The researchers then tested an array of TCV schedules under two contrasting scenarios for vaccine waning, one slow and one fast. They evaluated routine vaccination at different ages, the addition of booster doses, and the inclusion of catch-up campaigns that would vaccinate older children beyond the routine target age, all assessed over a ten-year time horizon.

The headline finding concerns a strategy that combined several elements: routine vaccination at nine months of age, a catch-up campaign reaching children up to fifteen years old, and a booster dose administered five years after the primary dose. Across the incidence settings examined, this comprehensive approach was predicted to deliver the greatest reduction in disease, cutting cases by a median of 40 to 68 percent. The wide range reflects variation across the different models and settings, but the consistency of the direction of effect across all four models is notable. Catch-up campaigns appear to be a particularly valuable component, because vaccinating older children who currently carry a substantial share of transmission accelerates the decline in disease and provides indirect protection to unvaccinated individuals through reduced circulation of the bacterium.

Cost-effectiveness analysis added a crucial economic dimension to the findings. The team employed the net-monetary-benefit framework, a standard approach in health economics that converts health gains into monetary terms and compares them against intervention costs. Because real-world estimates of the costs of typhoid illness and the risk of dying from it vary widely, the researchers ran their economic evaluations under two representative scenarios, one with high costs of illness and high case fatality risk and one with low values for both. This sensitivity analysis matters enormously, because the economic case for vaccination is far stronger when typhoid kills more frequently and treatment imposes greater financial burdens on households and health systems.

The economic results revealed clear patterns tied to disease burden. In medium incidence settings, introducing TCV with a catch-up campaign proved cost-effective at willingness-to-pay thresholds above 1,250 dollars per disability-adjusted life-year averted, but only when the costs of illness and case fatality risk were high. In high incidence settings, the same strategy was cost-effective even when costs and fatality risk were low, underscoring how a heavier disease burden strengthens the economic argument for vaccination. Most striking were the very high incidence settings, where introducing TCV at either nine months or two years of age was predicted to be cost-saving outright, meaning the vaccine would pay for itself through averted treatment costs and productivity losses regardless of the scenario tested.

The question of boosters received a nuanced answer that hinges on the pace of immune waning. If vaccine protection fades slowly, the standard schedule of a single dose at nine months with catch-up performs well without needing boosters in most settings. But if waning is fast, the models pointed toward a different strategy: delaying the primary vaccination to two or even five years of age, depending on the age at which disease incidence peaks in the community. The logic is straightforward. If protection will not last, giving it too early wastes the most valuable years of immunity on ages when disease risk is comparatively low. In very high incidence settings, however, adding a booster dose at five years of age was cost-effective at most willingness-to-pay values across all scenarios, suggesting that countries with intense transmission should plan for boosters from the outset.

One of the most reassuring aspects of the analysis is the degree of agreement among the four models despite their structural differences. Agent-based models and compartmental models make different simplifying assumptions about how immunity, mixing patterns, and transmission dynamics operate, so when they converge on similar predictions, confidence in the conclusions rises substantially. The researchers reported that model predictions for both health impact and cost-effectiveness were fairly robust to uncertainty in parameter values and model structure. This robustness is exactly what policymakers need, because it suggests the recommendations are not hostage to the idiosyncrasies of any particular modeling approach.

The authors are careful to acknowledge the limits of their work. The available data on indirect vaccine protection, the herd immunity effects that arise when vaccination reduces transmission, remain limited, and this gap constrains how precisely the models can predict population-level benefits. Other simplifying assumptions, including the representation of waning immunity and the treatment of age-specific contact patterns, also introduce uncertainty. These limitations do not undermine the central conclusions, but they highlight the value of continued surveillance as countries introduce TCVs, since real-world post-introduction data will allow the models to be refined and the waning question to be settled definitively.

For global health decision-makers, the study delivers a practical framework rather than a single universal prescription. The optimal typhoid vaccination strategy depends on three key local factors: the incidence rate of typhoid, the case fatality risk, and the waning rate of vaccine protection. Countries with very high transmission can proceed confidently with early routine vaccination and boosters, knowing the investment will likely save money. Countries with moderate transmission face a more contingent calculus that depends on the severity of disease and treatment costs. As more countries introduce TCVs and longer-term effectiveness data accumulate, the models compared in this analysis will provide an evolving evidence base for the WHO’s updated recommendations, helping ensure that every dose of typhoid vaccine is delivered at the age and in the schedule that delivers the greatest benefit.

Subject of Research: Modeling the impact and cost-effectiveness of typhoid conjugate vaccine schedules across transmission settings

Article Title: Evaluating the impact and cost-effectiveness of typhoid conjugate vaccine schedules across diverse settings: A multi-model comparison

Article References: Wenger, C. G. C., Grantz, K. H., Menkir, T. F., Muellenmeister, A. M., Pithawala, Z., Hutubessy, R., Mogasale, V., Kraay, A. N. M., Scott, N., Abeysuriya, R. G., Andrews, J. R., Gauld, J., Lo, N. C., & Pitzer, V. E. (2026). Evaluating the impact and cost-effectiveness of typhoid conjugate vaccine schedules across diverse settings: A multi-model comparison. PLOS Medicine, 23(10), e1005280. https://doi.org/10.1371/journal.pmed.1005280

Image Credits: AI Generated

DOI: 10.1371/journal.pmed.1005280

Keywords: typhoid fever, typhoid conjugate vaccine, vaccination schedules, mathematical modeling, cost-effectiveness, WHO, immunity waning, booster doses, catch-up campaigns, public health, PLOS Medicine, disability-adjusted life-years

The study team, led by Hyeon Jin Kim and Dong Keon Yon, constructed a population-based birth cohort by linking mothers to their children through South Korea’s National Health Insurance Service database, following families from January 2009 through December 2023. Because the insurance system covers virtually the entire population, the dataset captures an extraordinarily complete picture of medical care. In total, the cohort encompassed 2,236,167 mothers and 3,013,262 children, with individual children followed for as long as 14 years depending on their birth year. Prenatal opioid exposure was defined conservatively: a mother had to have received at least two opioid prescriptions within a single trimester of pregnancy to count as exposed.

Among the 2,928,186 children included in the primary analysis, 118,671 had documented prenatal opioid exposure while 2,809,515 did not. Roughly half of each group was male, and the researchers adjusted their statistical models for an unusually thorough list of potential confounders, including maternal age at delivery, infant sex, region of residence, household income, birth season, parity, a severe maternal morbidity score, the number of outpatient visits and hospital admissions in the year before pregnancy, preterm birth, low birth weight, delivery type, year of delivery, plurality, and maternal histories of substance use and neuropsychiatric disorders. This level of adjustment matters, because mothers who receive opioids during pregnancy differ from those who do not in many ways that could independently affect their children’s health.

The headline result was stark. In the propensity score-matched cohort, where exposed and unexposed children were balanced on dozens of measured characteristics, children with prenatal opioid exposure were hospitalized at rates 28 percent higher than their unexposed peers, with an adjusted incidence rate ratio of 1.28 and a confidence interval of 1.28 to 1.29. In absolute terms, that difference translates into a substantial number of additional hospital admissions across a population of this size, spanning everything from chest infections to broken arms. The sheer scale of the cohort gives the estimate unusual precision, and the association held after every adjustment the researchers applied.

What elevates the study above many of its predecessors is its use of sibling comparisons, a design that partially controls for the stable genetic and environmental factors shared by brothers and sisters. When the researchers compared siblings discordant for prenatal opioid exposure, essentially asking whether a child exposed in utero was hospitalized more often than a sibling from the same mother who was not exposed, the association shrank considerably, from a 28 percent excess to an 8 percent excess. Crucially, however, it did not disappear. The attenuated estimate, with a rate ratio of 1.08, remained statistically significant, suggesting that shared familial factors explain part but not all of the observed link.

The dose-response pattern added further weight to the findings. Children whose mothers received low cumulative opioid doses showed a 19 percent higher hospital admission rate, while those whose mothers received high doses showed a 40 percent higher rate. A similar gradient emerged with the duration of prescriptions. In epidemiology, a dose-dependent relationship is often taken as evidence that an association is not merely an artifact of confounding, since confounders rarely track so neatly with the exposure of interest. The researchers also found that the association weakened as children aged but remained detectable through age five, hinting at a biological effect that fades over time rather than a permanent vulnerability.

Perhaps the most striking aspect of the results is their breadth. Prenatal opioid exposure was associated with increased hospital admission rates across multiple distinct categories of illness: infections, non-infectious respiratory conditions, non-infectious gastrointestinal conditions, oral cavity diseases, injuries and external causes, and non-infectious genitourinary conditions. A pattern this diffuse does not point to a single damaged organ system. Instead, it is consistent with the hypothesis, advanced in earlier experimental work, that prenatal opioid exposure may disrupt the development of the infant immune system, leaving children broadly more susceptible to the ordinary illnesses and accidents of early childhood rather than to any one specific disease.

The biological plausibility of such a mechanism rests on well-established pharmacology. Opioids cross the placenta, and fetal brain tissue is rich in opioid receptors from early in gestation. Animal studies have shown that prenatal opioid exposure can alter immune cell development, inflammatory signaling, and stress-response systems, any of which could plausibly influence a child’s resilience to infection or recovery from illness. Human studies of neonatal opioid withdrawal syndrome have long documented immediate effects on newborns, but evidence about consequences that emerge months or years later has been sparse and often confounded by the social circumstances that accompany opioid use. A cohort of nearly three million children offers a rare opportunity to separate signal from noise.

The authors are careful about the limits of their data. Prescription records show what pharmacies dispensed, not what pregnant women actually ingested, and adherence may differ between those who fill multiple prescriptions and those who do not. Residual confounding also cannot be excluded: no statistical model can fully account for unmeasured differences in health, pain conditions, or family circumstances between women who receive opioids during pregnancy and those who do not. The sibling analysis partially addresses this concern, but siblings share only some of the relevant environment, and the attenuation observed in that analysis is itself a reminder that much of the raw association may reflect familial background rather than the drug itself.

Even with those caveats, the study’s practical implications are relatively clear. The authors suggest that the findings could inform postnatal monitoring and management strategies for children born to mothers who received opioid prescriptions during pregnancy, potentially flagging these children for closer follow-up in the vulnerable early years when the excess risk is most pronounced. The results do not argue that opioids should be withheld from pregnant women in severe pain, a decision that belongs to patients and their clinicians weighing individual circumstances. What they do provide is a population-level evidence base for a question that has grown urgent as opioid prescribing has expanded globally: that fetal exposure to these medications may leave a measurable, if modest and fading, imprint on childhood health that deserves attention from both researchers and pediatricians.

Subject of Research: The association between prenatal opioid exposure and childhood hospital admission risk

Article Title: Maternal opioid prescriptions and subsequent risk of hospital admission in offspring in South Korea: A nationwide birth cohort study

Article References: Kim, H. J., Yim, Y., Nehs, C. J., Park, J., Kim, E. S., Cho, S. J., Fond, G., Boyer, L., Kang, J., & Yon, D. K. (2026). Maternal opioid prescriptions and subsequent risk of hospital admission in offspring in South Korea: A nationwide birth cohort study. PLOS Medicine, 23(10), e1004829. https://doi.org/10.1371/journal.pmed.1004829

Image Credits: AI Generated

DOI: 10.1371/journal.pmed.1004829

Keywords: prenatal opioid exposure, birth cohort, hospital admissions, South Korea, National Health Insurance Service, propensity score matching, sibling comparison, dose-response, immune system development, pediatric health, pharmacoepidemiology, PLOS Medicine

Respiratory syncytial virus, or RSV, is the leading cause of bronchiolitis, a viral infection of the small airways that sends hundreds of thousands of infants to hospitals across Europe each winter. For decades, pediatricians had little to offer beyond supportive care, since a previous antibody product, palivizumab, was reserved for high-risk infants because of its high cost and the need for monthly injections. Nirsevimab changed that calculus. A single dose provides protection lasting at least five months, covering an entire RSV season, and clinical trials demonstrated strong efficacy against medically attended lower respiratory tract infection in healthy term and preterm infants alike.

When nirsevimab finally became available in Europe in late 2023, national and regional health authorities adopted it at different speeds and with different strategies. Some countries launched universal immunization programs targeting all infants entering their first RSV season, while others restricted initial campaigns to infants born during or just before the season, and a few had no program at all during the study window. This patchwork of implementation created an unintended natural experiment, and a large consortium of European pediatric emergency researchers seized the opportunity to measure what happened.

The research team, led by Léa Lenglart and Naim Ouldali with the EPISODES Study Group, assembled routinely collected data from 27 pediatric emergency departments spanning 12 European countries, covering the period from January 2018 through March 2024. They included every case of bronchiolitis diagnosed in infants younger than 12 months, yielding a total of 107,088 episodes. Regions were classified as intervention or control depending on whether nirsevimab had been implemented, and immunization coverage figures were drawn from official regional reports. Urinary tract infections in the same age group served as a control outcome, a standard technique to detect whether observed changes reflect a real intervention effect or some unrelated shift in emergency department attendance.

The analytical approach was a controlled interrupted time-series design, a method well suited to evaluating population-level interventions introduced at a known point in time. The models accounted for underlying seasonality and pre-existing temporal trends, allowing the researchers to estimate how much of the change in monthly bronchiolitis cases during the intervention period, from October 2023 through March 2024, could be attributed to the antibody program rather than to the ordinary rhythm of winter epidemics. Estimates were stratified by age group, separating infants aged 0 to 3 months, who face the highest risk of severe disease, from those aged 3 to 12 months.

The results were striking in both their magnitude and their heterogeneity. Reductions in bronchiolitis visits ranged from essentially no change, with one estimate of plus 0.6 percent, to a remarkable 61 percent decline, with a confidence interval stretching from minus 67.8 to minus 35.2 percent. Infants in the youngest age band generally showed the largest reductions, consistent with the biology of RSV, since the highest rates of hospitalization and intensive care admission occur in the first months of life. Meanwhile, bronchiolitis trends in control regions without nirsevimab programs remained stable, and urinary tract infection visits showed no comparable decline, strengthening the argument that the observed reductions were tied to the immunization campaigns rather than a general drop in emergency presentations.

Perhaps the most consequential finding was the strength of the statistical relationship between coverage and benefit. Across regions and age groups, the correlation between nirsevimab coverage and the percentage reduction in bronchiolitis visits was negative and pronounced, with a correlation coefficient of minus 0.86 and a p-value of 0.001. In practical terms, regions that immunized a larger share of their infant population saw proportionally larger drops in emergency department visits. The relationship held across both age strata, suggesting that the antibody’s protective effect translated into population benefit wherever it reached enough infants, whether they were immunized at birth during the season or through catch-up campaigns targeting older babies.

Why did coverage differ so much across Europe? The study points to implementation choices as the decisive variable. Countries and regions that embedded nirsevimab into routine newborn care, offering the dose in maternity wards before discharge, achieved far higher uptake than those relying on later appointments or opportunistic vaccination during other visits. Administrative hurdles, supply constraints, differences in eligibility criteria, and varying levels of parental and clinician acceptance all likely contributed. The authors emphasize that the antibody’s intrinsic efficacy, already demonstrated in randomized trials, was never in question; what varied was the fraction of the infant population that actually received it in time to be protected when the virus arrived.

The researchers are careful about the limits of their design. As an ecological study, it examines populations rather than individuals, so causality cannot be definitively inferred. Unmeasured regional differences, such as changes in testing practices, health-seeking behavior, viral circulation intensity, or concurrent public health measures, may have contributed to some of the observed variability. The interrupted time-series models adjust for seasonality and trend, and the stability of control outcomes argues against simple artifacts, but the authors acknowledge that residual confounding remains possible. Still, the consistency of the findings across 27 departments, multiple countries, and two age groups, together with the tight coverage-response relationship, makes a compelling case that immunization coverage was a key driver of the reductions.

The implications for health policy are immediate. RSV prevention has entered a new era, with nirsevimab programs now expanding and maternal RSV vaccines adding another layer of protection for newborns. This study demonstrates that the public health return on these interventions depends not only on the biology of the product but on the machinery of delivery. Regions that achieved high coverage among both season-born and out-of-season infants reaped reductions in bronchiolitis visits of a scale rarely seen for a respiratory disease, easing pressure on emergency services during the busiest weeks of the winter. For health authorities weighing how to structure their own programs, the European experience of 2023–2024 offers a clear lesson: the fastest, broadest, and earliest possible reach into the infant population is what converts a highly effective antibody into a visible decline in disease. As nirsevimab programs mature and additional seasons of data accumulate, researchers will be watching to see whether the coverage-response relationship holds, whether protection extends to severe outcomes such as intensive care admission, and whether the substantial regional variability in implementation narrows as systems learn from one another.

Subject of Research: Association between nirsevimab immunization coverage and reductions in infant bronchiolitis emergency department visits across 12 European countries

Article Title: Association between nirsevimab coverage and pediatric emergency department visits for bronchiolitis in 12 European countries: An interrupted time-series analysis

Article References: Lenglart, L., Titomanlio, L., Alberti, I., Almeida, L., Aupiais, C., Akyüz Özkan, E., Barrett, M., Basmaci, R., Birbilen, A., Borensztajn, D., Schönenberger, C. B., Bressan, S., Buonsenso, D., Campos, T., Castanhinha, S., Chiaretti, A., Claret, G., De Zan, F., Durnin, S., … the EPISODES Study Group (2026). Association between nirsevimab coverage and pediatric emergency department visits for bronchiolitis in 12 European countries: An interrupted time-series analysis. PLOS Medicine, 23(9), e1005223. https://doi.org/10.1371/journal.pmed.1005223

Image Credits: AI Generated

DOI: 10.1371/journal.pmed.1005223

Keywords: nirsevimab, RSV, bronchiolitis, infants, pediatric emergency departments, immunization coverage, interrupted time-series, Europe, PLOS Medicine, monoclonal antibody, respiratory syncytial virus, public health

The study, led by Oliver S. Simmons and Ilaria Dorigatti together with colleagues at Mexican research institutions, set out to fill two important gaps in the understanding of dengue epidemiology. First, while earlier work had mapped long-term average spatial patterns of dengue risk in Mexico, nobody had systematically characterised how transmission intensity shifts from year to year across the country. Second, the relative transmissibility of the four dengue virus serotypes, DENV-1, DENV-2, DENV-3 and DENV-4, remains poorly characterised both in Mexico and globally, despite the fact that these serotypes differ genetically and can dominate outbreaks at different times and places. Understanding these differences matters because infection with one serotype confers long-lasting immunity only against that serotype, while secondary infection with a different serotype is a well-established risk factor for severe disease.

To address these questions, the researchers employed catalytic models, a class of mathematical models originally developed to describe age patterns of infection for vaccine-preventable diseases. These models use the age distribution of cases, combined with case counts over time, to estimate the force of infection without requiring population-wide serological surveys. Crucially, the team developed a new serotype-specific extension of these established models, allowing them to estimate how intensely each of the four serotypes was transmitting in each state and year. This methodological advance means that routinely collected case-notification data, which health ministries already gather as part of standard surveillance, can be repurposed to yield insights that would otherwise require expensive and logistically demanding seroprevalence studies.

The analysis confirmed that dengue transmission in Mexico is far from uniform. Large spatial heterogeneities emerged, with some states experiencing consistently higher force of infection than others, reflecting differences in climate, urbanisation, mosquito abundance, human mobility and surveillance capacity. Equally notable were the temporal patterns: transmission intensity fluctuated considerably from year to year, underscoring that a single snapshot of average risk can be misleading for planning purposes. In years of intense transmission, a state’s population may experience a rapid accumulation of immunity, only for that same population to remain relatively untouched in subsequent years. This year-to-year volatility means that cumulative estimates of infection burden can conceal periods of both quiescence and explosive epidemic growth.

The serotype-specific findings are among the most significant contributions of the study. DENV-1 and DENV-2, according to the model estimates, have historically circulated at high levels across Mexico, making them the dominant drivers of the country’s dengue burden during the study period. In contrast, DENV-4 exhibited low transmission intensity throughout the eight years analysed. Perhaps most concerning for public health planners was the evidence of increasing DENV-3 transmission intensity in several states in recent years, a trend that coincided with large outbreaks. Because many younger Mexicans may never have been exposed to DENV-3, its re-emergence creates a large susceptible population at risk of primary infection, while older cohorts previously exposed to this serotype decades ago face the possibility of secondary infection with a now-circulating different serotype, a scenario associated with elevated risk of severe dengue.

The mechanistic underpinnings of these serotype-specific patterns are complex. Dengue virus transmission depends on the Aedes aegypti mosquito vector, whose abundance and susceptibility to infection can vary, as can the intrinsic fitness of each viral serotype. Differences in the force of infection between serotypes may reflect a combination of viral characteristics, such as replication efficiency in the mosquito and human host, and herd immunity dynamics, whereby a serotype that circulated widely in the past leaves fewer susceptible individuals available for future transmission. The modelling framework cannot fully disentangle these mechanisms on its own, but by generating comparable, state-level estimates of serotype-specific transmission intensity, it provides the empirical foundation upon which such mechanistic questions can be addressed in future work.

The study also highlights the crucial role of diagnostics in making serotype-specific inference possible. Extensive reverse transcription polymerase chain reaction, or RT-PCR, testing during the study period allowed many cases to be attributed to a specific infecting serotype, and the authors emphasise that newly available rapid diagnostic tests capable of discerning the infecting serotype could further improve this picture. As these tests become more widely deployed, surveillance systems in Mexico and elsewhere will be better positioned to track which serotypes are circulating in near real time, information that is directly relevant to decisions about when and where to deploy interventions. The timing is particularly significant given that Mexico has introduced dengue vaccination programmes, and vaccine effectiveness and deployment strategy can depend on the local transmission intensity and the serotype composition of circulating viruses.

The authors are careful to note an important limitation: the serotype-specific estimates may be affected by heterogeneities in how cases are reported across states and over time. Surveillance intensity, health-seeking behaviour, access to healthcare and diagnostic practices all vary geographically and can shift between years, and disentangling genuine changes in transmission from artefacts of the reporting system remains a task for future validation studies. Nevertheless, the scale of the dataset, spanning 833,629 cases across 27 states and eight years, lends considerable weight to the broad patterns identified, and the consistency of the findings with known epidemiological events, such as the DENV-3 resurgence coinciding with large outbreaks, supports their plausibility.

The wider context makes this work especially timely. Mexico and the Americas region as a whole have experienced rising dengue incidence in recent years, and there are widespread concerns that changing climatic conditions, including warmer temperatures and shifting rainfall patterns, may expand the geographic range and seasonal window of Aedes mosquito activity, thereby enhancing transmission in the future. In this environment, accurate estimates of who remains susceptible to which serotype, and where and when transmission is most intense, become essential inputs for targeting vaccination, vector control and clinical preparedness. The methods developed in this study contribute to a better understanding of the past and current burden of dengue infection and can help refine assessments of the potential impact of existing and new interventions, from next-generation vaccines to Wolbachia-based biocontrol strategies.

Ultimately, the study demonstrates the power of mining routinely collected surveillance data with appropriately sophisticated models. Rather than relying solely on costly serological surveys, public health authorities can extract detailed, serotype-resolved pictures of transmission intensity from the case notifications they already gather, provided that diagnostic testing is sufficiently extensive and that reporting heterogeneities are acknowledged and, where possible, corrected for. As dengue continues to expand its footprint across the Americas under the combined pressures of urbanisation, connectivity and climate change, approaches of this kind offer a practical route toward the granular, timely evidence base that controlling this evolving threat will demand.

Subject of Research: Serotype-specific dengue virus transmission intensity across Mexican states estimated from case-notification data using catalytic models

Article Title: Overall and serotype-specific dengue virus transmission intensity in Mexico, 2016–2023: A modelling study of case-notification data

Article References: Simmons, O. S., Vicco, A., Martínez-Vega, R. A., Ramos-Castañeda, J., & Dorigatti, I. (2026). Overall and serotype-specific dengue virus transmission intensity in Mexico, 2016–2023: A modelling study of case-notification data. PLOS Medicine, 23(9), e1004874. https://doi.org/10.1371/journal.pmed.1004874

Image Credits: AI Generated

DOI: 10.1371/journal.pmed.1004874

Keywords: dengue, DENV-1, DENV-2, DENV-3, DENV-4, force of infection, catalytic models, Mexico, serotypes, surveillance, PLOS Medicine, epidemiology

The scale of the discrepancy is easiest to see with the glucagon-like peptide-1 receptor agonists, the class of drugs behind the global weight-loss revolution. A recent synthesis of real-world evidence found that between 20 and 50 percent of patients discontinue GLP-1 receptor agonists within the first year. Many who stay on treatment use substantially lower doses than those studied in trials, and the weight loss observed in clinical practice is consistently lower than in randomized controlled trials. Crucially, among highly adherent patients, outcomes approach the trial results, which suggests the drugs themselves are not the problem. The drivers of early discontinuation are familiar ones: gastrointestinal side effects, cost, insurance barriers, and supply shortages that have repeatedly left pharmacies without stock.

The consequences of stopping, however, are not uniform, and this is where the stakes become far higher than a number on a bathroom scale. For patients prescribed GLP-1 receptor agonists solely for weight management, discontinuation primarily results in weight regain. But the same drugs also reduce the risk of cardiovascular events and clinically important kidney disease outcomes compared with placebo. Because these protective benefits depend on continued exposure to the drug rather than conferring durable protection after withdrawal, a patient who stops treatment forfeits not only the weight loss but also reductions in morbidity and mortality. The prescription, in other words, is a subscription to benefit, and cancellation carries a clinical price.

A similar shortfall appears with anticoagulants, medications whose value is about as unambiguous as it gets. A systematic review and meta-analysis of nearly 595,000 patients with atrial fibrillation found that only two-thirds maintained good adherence to direct oral anticoagulants, the modern stroke-preventing drugs that require far less monitoring than their predecessor warfarin. Among those who did not adhere, the risk of stroke increased by 39 percent. That finding quantifies the distance between prescribed therapy and delivered benefit even for medications with straightforward dosing, and it reframes nonadherence from a vague behavioral concern into a measurable contributor to preventable strokes and deaths.

A different kind of adherence failure is unfolding with the recombinant zoster vaccine, which protects against shingles and requires only two doses separated by two to six months. A nationwide study of more than 726,000 US adults found that adherence to the recommended schedule was just 72 percent, with completion rates markedly lower among racial and ethnic minorities, younger adults, and people with lower household income. What makes this particularly consequential is emerging evidence about what the vaccine might do beyond preventing a painful rash. A large-scale longitudinal analysis of US health records found that herpes zoster vaccination was associated with a reduced risk of dementia, with stronger protection in people who received both doses of the recombinant vaccine compared with a single dose. A target trial emulation among nearly 510,000 skilled-nursing facility residents similarly found the vaccine was associated with reduced dementia risk. If these findings are confirmed by ongoing trials, every person who fails to complete the two-dose series may be forgoing not only protection against shingles but a potential modification of dementia risk.

The underlying problem itself has barely changed in fifty years of research, and that is precisely the point. Marcum’s argument centers on the opportunity cost of nonadherence, the benefit forgone when effective therapy goes untaken. The logic is straightforward: when adherence rates stay static while treatment effects grow, the benefit lost through nonadherence grows with them. The numbers illustrate the leap in therapeutic power. In a network meta-analysis, semaglutide reduced body weight by 11.4 percent from baseline compared with 3.1 percent for the older drug orlistat. Direct oral anticoagulants reduce stroke or systemic embolism by 19 percent compared with warfarin. The recombinant zoster vaccine cuts shingles risk by 92 percent, versus 51 percent for its live attenuated predecessor. But these benefits are realized only under the conditions of trial participation, where drugs are free and adherence is monitored. In practice, where patients bear cost-sharing and juggle competing demands, the attenuation of benefit is substantial.

The economics of the problem are genuinely awkward. With therapies costing tens of thousands of dollars per year, improving adherence increases short-term expenditures, a tension that makes payers uneasy even though the downstream costs of preventable hospitalizations and disease progression are far greater. Nonadherence itself is also heterogeneous, ranging from cost-related gaps in treatment to unintentional lapses in complex regimens, and effective solutions must account for that variation. Nor is every discontinuation a failure. Stopping a therapy after an informed discussion of adverse effects with a healthcare provider is shared decision-making, and conflating it with nonadherence distorts both research and policy.

What can be done? Established interventions such as patient education, simpler dosing schedules, and improved communication remain necessary but demonstrably insufficient. A Cochrane review of 182 randomized trials found that even the most effective adherence interventions, typically bundles of education, counseling, and ongoing support from health system professionals, produced only modest improvements in adherence or clinical outcomes. At the bedside, clinicians can treat adherence as a clinical decision point: counseling patients on the cardiorenal benefits at risk when a GLP-1 receptor agonist is stopped, checking persistence at every anticoagulant renewal, and stressing the importance of completing the vaccine series. But individual effort has limits, and the evidence says those limits arrive quickly.

That is why Marcum calls for a system-level reconceptualization of adherence as a determinant of therapeutic value, meaning interventions that change the environment in which prescriptions are written, filled, and consumed. For GLP-1 receptor agonists, that means addressing the structural barriers of cost, insurance coverage, and supply that drive discontinuation. For chronic medications, it means expanding pharmacists’ authority to manage therapy through collaborative practice agreements and transition-of-care programs, and aligning cost-sharing with therapeutic value through value-based insurance design. For multi-dose vaccines, it means integrating completion tracking into electronic health records. The proof that such approaches can work already exists: a randomized trial of nearly 6,000 patients after myocardial infarction eliminated copayments for cardiovascular medications, improving adherence by 4 to 6 percent and reducing rates of first major vascular events without increasing total health spending.

Compounding everything is the fact that nonadherence is hard to detect in routine practice. Clinicians consistently underestimate it, and the available measures, including pharmacy refill records, self-report, and electronic health record documentation, each capture only a partial picture. Artificial intelligence may eventually help close this detection gap, though rigorous evidence that such tools improve clinically meaningful endpoints remains limited. The bottom line, Marcum argues, is that the limiting factor in realizing the full value of modern therapeutics is no longer only pharmacology; it is human behavior, reinforced by systems that make adherence difficult. Until the same rigor applied to drug discovery is applied to coverage design, pharmacist-led programs, health information technology, and implementation research, medicine will keep paying for the promise of transformative therapies while realizing only a fraction of their benefit.

Subject of Research: Medication nonadherence and the widening gap between clinical trial efficacy and real-world effectiveness

Article Title: The widening medication adherence gap

Article References: Marcum, Z. A. (2026). The widening medication adherence gap. PLOS Medicine, 23(9), e1005237. https://doi.org/10.1371/journal.pmed.1005237

Image Credits: AI Generated

DOI: 10.1371/journal.pmed.1005237

Keywords: medication adherence, GLP-1 receptor agonists, anticoagulants, herpes zoster vaccine, dementia risk, obesity treatment, stroke prevention, health systems, value-based insurance design, pharmacist-led programs, PLOS Medicine, real-world evidence

The study drew on 6,896 adults from the National Health and Nutrition Examination Survey (NHANES) between 2005 and 2018, with survey weights applied so the findings reflect the broader US adult population. Rather than betting on a single algorithm, the researchers staged a computational tournament: a two-stage machine learning framework evaluated more than 100 candidate survival models, ranging from classical Cox regression to modern ensemble methods. The winner was chosen through Pareto front optimization, a technique borrowed from multi-objective engineering that refuses to sacrifice one performance dimension for another. Instead of picking the model with the best score on a single metric, the approach identifies models that sit on the mathematical frontier where no competitor can improve accuracy without degrading another quality such as calibration or simplicity. The eventual champion was, somewhat surprisingly, a simple Cox model, which achieved a mean concordance index of 0.893, a strikingly high figure for mortality prediction in a free-living population.

From that optimized model, the team distilled CKMAI, a single number that integrates a diverse set of inputs: clinical laboratory measurements, nutritional assessments, lifestyle factors, and social determinants of health. The philosophy is that aging in CKM syndrome is not written in any one biomarker but in the coordinated drift of dozens of variables that together describe how far along the heart-kidney-metabolic axis a person has traveled. The index was then stress-tested against three of the best-known competitors: PhenoAge, the Klemera-Doubal method (KDM), and the cardiometabolic index (CMI). Across every outcome and every time horizon, CKMAI came out ahead. For all-cause mortality, its time-dependent area under the curve reached 0.893 at three years, 0.907 at five years, and 0.890 at ten years. For cardiovascular mortality, the numbers climbed even higher, peaking at 0.937 at both five and ten years, meaning the index could discriminate between those who would and would not die of cardiovascular causes with remarkable precision.

Discrimination is only half the story, however, and the researchers went to considerable lengths to prove the index was not merely ranking people correctly but also assigning them the right absolute risks. Reclassification metrics told a consistent story: net reclassification improvement and integrated discrimination improvement were statistically significant at all time points, confirming that CKMAI moved patients into more accurate risk categories compared with existing indices. Calibration, assessed with the Greenwood-Nam-D’Agostino test, showed no significant departure between predicted and observed mortality, and decision curve analysis demonstrated superior net benefit, a measure of how much real-world clinical value a risk model delivers when used to guide decisions. In practical terms, a physician using CKMAI to decide who needs intensified intervention would make better decisions than one relying on PhenoAge, KDM, or CMI, with fewer false alarms and fewer missed high-risk patients.

One of the most intriguing findings concerns the shape of the relationship between biological aging and death. Using restricted cubic splines and two-piecewise regression, the team found that risk does not climb smoothly with CKMAI. Instead, there are inflection points: 61.587 for all-cause mortality, 60.732 for cardiovascular mortality, and 34.092 for high-risk CKM status. Below each threshold, risk rises steeply; above it, the curve flattens. This counterintuitive pattern suggests that the most dangerous phase of CKM-related aging is the transition zone, the period when a person’s index first crosses into the elevated range. Clinically, this reframes the goal of screening: the patients who most urgently need intervention may be those just approaching the threshold, not those already deep into the high-risk zone, whose trajectory may in some respects already be set.

The index also revealed a startling interaction with conventional biological age. When CKMAI and PhenoAge were analyzed together, the researchers documented a super-additive interaction, quantified by a relative excess risk due to interaction of 15.06 (95% confidence interval 8.14 to 38.44). In plain language, the combined effect of being old by both clocks is far worse than the sum of the two risks separately. A person who is chronologically aged by PhenoAge and additionally accelerated on the CKM-specific axis occupies a biological state that neither measure alone can capture. This finding offers a mechanistic clue about why CKM syndrome progresses so heterogeneously: the syndrome’s damage may compound with, rather than merely parallel, the broader aging process.

To map that heterogeneity directly, the team applied unsupervised clustering to the combined aging and metabolic data and uncovered six distinct aging-metabolic phenotypes within the population. The most alarming was a frail elderly cluster whose members faced a hazard ratio of 11.74 for mortality compared with the reference group, an eleven-fold elevation in risk. The existence of these discrete phenotypes suggests that CKM syndrome is not a single disease with a single trajectory but a family of biological states, some of which are far more lethal than others. Identifying which phenotype a patient belongs to could eventually guide how aggressively clinicians pursue kidney protection, cardiovascular risk reduction, or metabolic control.

Perhaps the most unexpected thread in the study is psychological. Using weighted bootstrap mediation analysis, the researchers found that depression partially mediates the link between CKMAI and its outcomes. The proportion of the association explained by depression was 5.6 percent for all-cause mortality, 8.2 percent for cardiovascular mortality, and 11.1 percent for high-risk CKM status. In other words, some of the excess mortality carried by an accelerated CKM aging signature travels through the mind: biological aging in this syndrome appears to foster depressive states, which in turn worsen outcomes, likely through behavioral, inflammatory, and neuroendocrine pathways. The mediation is partial, but its consistency across all three outcomes suggests that mental health may be an underappreciated lever in breaking the cycle of cardiovascular-kidney-metabolic decline.

The findings held up under scrutiny. Prespecified subgroup analyses and multiple sensitivity analyses confirmed robustness, and temporal validation within NHANES supported the model’s stability over time. A preliminary external validation in a hospital-based Chinese cohort of 261 patients also supported the index’s ability to identify high-risk CKM status, hinting at generalizability beyond the US population. The authors are candid about the study’s central limitation: CKMAI has not yet been validated in geographically independent cohorts with complete mortality follow-up, and until that happens, its performance estimates carry some uncertainty. Still, the combination of national representativeness, methodological rigor, and early external signals makes a strong case.

If the results replicate, the implications could be substantial. CKMAI is built from data that clinics already collect, bloodwork, nutrition, lifestyle, and social context, which makes it a practical, implementable tool rather than a laboratory curiosity. A single number that flags patients crossing the mortality inflection point, sorts them into one of six aging-metabolic phenotypes, and even points toward depression screening could reshape how preventive care is delivered to the millions of people whose hearts, kidneys, and metabolisms are failing together. As cardiovascular-kidney-metabolic syndrome becomes one of the defining health challenges of an aging century, an aging clock tuned to its specific biology may prove to be exactly the instrument medicine has been missing.

Subject of Research: A machine learning-derived biological aging index for mortality prediction and risk stratification in cardiovascular-kidney-metabolic syndrome

Article Title: A machine learning-derived aging index for risk stratification and mortality prediction in cardiovascular-kidney-metabolic syndrome: A retrospective cohort study

Article References: Zhu, Z., Ren, K., Wang, D., Lv, Y., Jin, H., Zhang, L., & Wang, Y. (2026). A machine learning-derived aging index for risk stratification and mortality prediction in cardiovascular-kidney-metabolic syndrome: A retrospective cohort study. PLOS Medicine, 23(10), e1005078. https://doi.org/10.1371/journal.pmed.1005078

Image Credits: AI Generated

DOI: 10.1371/journal.pmed.1005078

Keywords: biological aging, machine learning, cardiovascular-kidney-metabolic syndrome, mortality prediction, NHANES, risk stratification, PhenoAge, depression, Cox regression, epidemiology, precision medicine, PLOS Medicine

Active case finding, or ACF, involves systematically screening community members for tuberculosis regardless of whether they show symptoms, rather than waiting for patients to arrive at clinics. The World Health Organization endorses the approach because untreated, infectious tuberculosis silently sustains transmission in crowded urban settings. But screening entire populations is expensive, and in Blantyre the usual proxy for deciding where to screen—clinic notification data—can be misleading. Neighborhoods with better health infrastructure often report more cases simply because residents seek care more readily, masking areas where the true burden of undiagnosed disease may be highest.

The innovation at the heart of the study is the use of interferon-gamma release assays, or IGRAs, conducted on convenience samples of young children attending primary care. Because young children have had little time to accumulate old infections, a positive immune response signals recent transmission, making neighborhood-level estimates of the annual risk of tuberculosis infection, or ARTI, a sensitive indicator of where the force of infection is strongest. In the Blantyre survey, ARTI estimates across 33 urban neighborhoods ranged from zero to nine percent, revealing substantial heterogeneity in a city of roughly 800,000 people where adult HIV prevalence reaches 14 percent.

To translate these surveillance signals into policy, the team built a Markov microsimulation model that tracked simulated cohorts of 10,000 individuals over their entire lifetimes, updating health status weekly. The model captured HIV progression through CD4 cell counts, tuberculosis-related lung damage through FEV1 percentages, and the consequences of early versus late diagnosis, including post-tuberculosis sequelae that increasingly dominate the disease’s long-term burden. Costs were assessed from both health system and societal perspectives, incorporating transport, food, accommodation, and lost income, and all future costs and benefits were discounted at three percent per year.

The researchers compared three scenarios developed with Malawi’s National Tuberculosis and Leprosy Elimination Program: passive case finding alone, untargeted ACF delivered uniformly across all neighborhoods, and targeted ACF prioritizing neighborhoods from highest to lowest ARTI. The screening algorithm mirrored real program practice, using symptom questionnaires, digital chest X-ray, and Xpert MTB/RIF testing, with mobile diagnostic units deployed into communities. The intervention itself was assumed to cost about seven US dollars per person screened, based on a previous cluster-randomized trial in Blantyre.

The results were striking. Under the assumption that ARTI strongly predicts true tuberculosis prevalence, targeting just half the population identified roughly 80 percent of all individuals with tuberculosis detectable through screening, compared with only 50 percent under untargeted screening of the same number of people. For individuals actually diagnosed through ACF, life expectancy improved by an average of 3.3 years for those without HIV and 2.1 years for those with HIV, compared with passive detection alone. Each index case found through ACF was estimated to avert 0.88 downstream infections, adding a further 5.08 life-years and 8.98 disability-adjusted life years, or DALYs, averted through reduced transmission.

Efficiency gains translated directly into economics. Targeted ACF achieved a cost of about 400 US dollars per DALY averted, with a 95 percent credible interval of 20 to 1,000 dollars, while untargeted ACF cost about 700 dollars per DALY averted. Even when the predictive power of ARTI was weakened in sensitivity analyses, targeted screening retained its advantage, with cost-effectiveness ratios rising only modestly. Cluster-level analyses showed that high-ARTI neighborhoods generally achieved lower incremental cost-effectiveness ratios, supporting a phased scale-up strategy that begins in the highest-transmission areas and expands outward as budgets allow.

Yet the study is candid about a sobering reality: even the targeted strategy may not meet conventional cost-effectiveness thresholds for Malawi, which published estimates place between roughly 205 and 274 dollars per DALY averted, with a recent government report proposing an even stricter benchmark of 66 dollars. Against a per capita health expenditure of about 41 dollars in 2023, ratios of 400 to 700 dollars per DALY represent a substantial claim on limited resources. The authors note, however, that such thresholds may undervalue infectious disease interventions, since they capture neither the longer-term population benefits of interrupted transmission, the mitigation of future drug-resistant disease, nor progress toward elimination goals.

Sensitivity analyses reinforced this nuance. When the researchers assumed a higher baseline prevalence of 452 per 100,000, drawn from Malawi’s 2013–2014 national survey, cost-effectiveness ratios fell by roughly half, to as low as 200 dollars per DALY averted under strong targeting. Assuming ACF reduced transmission among detected cases by 75 percent rather than 50 percent cut ratios by about 30 percent. Conversely, excluding transmission benefits entirely inflated the ratios two-and-a-half to threefold, underscoring how much of ACF’s value depends on its epidemic-level effects rather than individual treatment outcomes alone.

The findings arrive amid debate about how best to find missing tuberculosis cases. The TREATS trial in Zambia and South Africa found no significant reduction in tuberculosis prevalence after a community-wide intervention combining HIV testing and symptom-based screening, but the Blantyre model differs in key respects: it incorporated chest X-ray screening for people without symptoms and allowed fine-scale geographic targeting of high-transmission neighborhoods. The authors caution that their conclusions rest on assumptions—particularly the uncertain quantitative link between childhood immunoreactivity and adult disease prevalence—and that the cost of collecting immunoreactivity survey data was excluded because the surveys were embedded in routine primary care visits. They argue that low-cost methods for gathering these data, and empirical validation of the ARTI-prevalence relationship, are essential before wider adoption. If those hurdles can be cleared, the study suggests that a simple blood test in toddlers could become an unexpected compass for one of global health’s most expensive hunts.

Subject of Research: Cost-effectiveness of tuberculosis active case finding targeted using childhood immunoreactivity survey data in Blantyre, Malawi

Article Title: Potential health impacts and costs of active case finding guided by Mycobacterium tuberculosis immunoreactivity survey results in Blantyre, Malawi: A mathematical modeling study

Article References: Kim, S., Can, M. H., Rickman, H. M., Phiri, M. D., Nliwasa, M., Mwenyenkulu, T. E., Mbendera, K., Verguet, S., Castro, M. C., Corbett, E. L., MacPherson, P., Cohen, T., & Menzies, N. A. (2026). Potential health impacts and costs of active case finding guided by Mycobacterium tuberculosis immunoreactivity survey results in Blantyre, Malawi: A mathematical modeling study. PLOS Medicine, 23(10), e1005272. https://doi.org/10.1371/journal.pmed.1005272

Image Credits: AI Generated

DOI: 10.1371/journal.pmed.1005272

Keywords: tuberculosis, active case finding, Malawi, mathematical modeling, annual risk of infection, interferon-gamma release assays, cost-effectiveness, HIV, surveillance, Blantyre, PLOS Medicine, public health

The team, led by Damir Zhakparov of the Swiss Institute of Allergy and Asthma Research together with collaborators including Kathleen Moriarty, Michael Levin, Carol Hlela, and Katja Baerenfaller, chose their study population with deliberate care. The AmaXhosa people of South Africa share a common ethnogenetic background, yet some children grow up in rural communities while others live in urban environments. Previous research has documented differences in atopic dermatitis prevalence and allergy sensitisation patterns between African populations, and between urban and rural settings more broadly. That combination, a shared genetic heritage paired with divergent environmental exposures, makes the AmaXhosa an unusually powerful natural experiment for disentangling the environmental and immune mechanisms that underlie the disease.

Methodologically, the study is notable for how it wrangled a genuinely complex dataset. The researchers re-analysed a previously established multimodal collection of measurements spanning four layers: environmental features describing each child’s surroundings, plasma cytokine levels reflecting immune signalling, antibody profiles including allergen-specific and total immunoglobulin E, and transcriptomic data capturing gene expression. Rather than treating these layers separately, they applied machine learning to each modality individually and then integrated them. Their toolkit included GeneSelectR, a workflow for identifying informative genes, SHAP values, an explainability technique that reveals which features drive a model’s predictions, and DIABLO, an integration method designed to find correlated patterns across different data types.

The environmental and antibody analyses produced an early clue. When the models predicted which children had atopic dermatitis, they relied on the combined effects of environmental features together with higher levels of allergen-specific and total IgE antibodies. IgE is the antibody class most closely associated with allergic responses, and its elevation in children with eczema fits the long-standing clinical observation that the skin disease frequently precedes or accompanies food allergies, asthma, and allergic rhinitis. But the SHAP-based explainability analysis showed that the antibody signal did not stand alone; it gained predictive power in combination with the environmental context in which each child was growing up.

The transcriptomic layer proved the richest single source of information. From the gene expression data, the researchers identified a subset of 560 genes whose activity discriminated between children with and without atopic dermatitis. These genes then served as the input for the downstream integration analyses. The size of this discriminating set underscores how genetically and molecularly diffuse the disease is: rather than a handful of culprit genes, the eczema phenotype appears to emerge from a broad shift in cellular programs, a pattern consistent with the involvement of skin barrier function, immune activation, and inflammatory signalling pathways.

The centrepiece of the study came when the team integrated all the modalities at once. Using DIABLO, they identified three multimodal clusters associated with disease status. The first cluster was linked to the healthy phenotype and was composed of environmental features found primarily in the rural setting. Crucially, these rural environmental features correlated with plasma cytokine levels and with the expression of autophagy-related genes. Autophagy is the cellular housekeeping process by which cells degrade and recycle damaged components, and it has been implicated in skin barrier maintenance and immune regulation. The suggestion that rural exposures might be associated with protective cytokine profiles and enhanced autophagy gene expression offers a molecular echo of the hygiene hypothesis, the idea that certain early-life exposures train the immune system away from allergic disease.

The two remaining clusters told a different story. Both were associated with atopic dermatitis, but in distinct ways. One was characterised by correlations between allergen-specific and total IgE antibodies and two cytokines, MCP-4 and TARC. Both of these signalling molecules are chemokines involved in recruiting immune cells to inflamed tissue, and TARC in particular is a well-established marker of the type 2 immune responses that dominate allergic disease. The second susceptibility cluster was defined by a transcriptomic feature signature associated with the atopic dermatitis endotype, the term researchers use for biologically distinct subtypes of a disease that may look similar on the surface. In other words, the study did not find a single molecular route to eczema but at least two, one immunological and antibody-driven, the other transcriptional.

The authors are candid about the limitations of their work. The explainable machine learning framework they employed is exploratory in nature, meaning the clusters and feature associations it surfaces are hypotheses to be tested rather than definitive causal pathways. The study also lacked validation in an independent cohort, a standard benchmark for observational research of this kind, so the identified signatures must be replicated before they can inform clinical practice. Observational designs, however carefully analysed, cannot by themselves prove that rural environments cause protection or that elevated IgE causes disease; they establish associations whose direction and mechanism require further study.

Even with those caveats, the findings carry significant weight for the field. Atopic dermatitis affects a substantial share of children worldwide and is often the first step in the so-called atopic march, the progression from eczema to food allergy and asthma. Understanding which children are on which molecular trajectory could eventually support earlier and more targeted interventions. The identification of a protective cluster rooted in rural environmental exposures, cytokine profiles, and autophagy gene expression points to concrete biological hypotheses about how environment gets under the skin, potentially through immune training and cellular maintenance pathways that could one day be mimicked therapeutically.

There is also a methodological legacy. The researchers emphasise that their feature-selection and integration workflow, combining GeneSelectR, SHAP explainability, and DIABLO integration, provides a reusable framework for analysing complex multimodal datasets in biomedical research more broadly. As studies increasingly collect layered data spanning genomes, transcriptomes, immune measurements, and environmental exposures, the bottleneck is often not data generation but interpretation. By demonstrating how machine learning can be made explainable enough to yield biologically interpretable clusters from such data, the team offers a template that extends well beyond eczema, and their AmaXhosa cohort stands as a reminder that the most informative answers may come from populations whose environments, as much as their genes, tell the story.

Subject of Research: Machine learning identification of protective and susceptibility clusters in paediatric atopic dermatitis among AmaXhosa children

Article Title: Multimodal protective and susceptibility clusters in paediatric atopic dermatitis: A machine learning-based, data-driven observational study

Article References: Zhakparov, D., Lunjani, N., Schmid, M., Moriarty, K., Roquero, D., Dreher, A., Heldstab-Kast, J. I., Nadeau, K. C., Akdis, C., Levin, M., Hlela, C., Sokolowska, M., O’Mahony, L., & Baerenfaller, K. (2026). Multimodal protective and susceptibility clusters in paediatric atopic dermatitis: A machine learning-based, data-driven observational study. PLOS Medicine, 23(9), e1004917. https://doi.org/10.1371/journal.pmed.1004917

Image Credits: AI Generated

DOI: 10.1371/journal.pmed.1004917

Keywords: atopic dermatitis, eczema, machine learning, AmaXhosa, pediatrics, IgE antibodies, cytokines, transcriptomics, autophagy, rural environment, immune regulation, PLOS Medicine