Long-term exposure to common air pollutants may significantly raise the risk of developing substance use disorders, while living surrounded by greenery appears to offer a measure of protection, according to a large new population-based study drawing on nearly a quarter of a million participants from the UK Biobank.

The research, published in BMC Psychiatry, analyzed data from 247,138 UK Biobank participants who were free of substance use disorders at baseline, tracking them over a median follow-up period of 10.9 years. During that time, 4,982 participants went on to develop a substance use disorder, providing the researchers with a robust dataset to examine the relationship between environmental exposures and mental health outcomes.

The study, led by Huan Liu and Bolun Cheng of Xi’an Jiaotong University in China, along with colleagues including Feng Zhang, represents one of the most comprehensive investigations to date into how the air people breathe and the greenness of their neighborhoods may influence vulnerability to addiction. The findings arrive at a moment when urban planners and public health officials are increasingly recognizing that the built and natural environments shape health in ways that extend well beyond respiratory and cardiovascular disease.

To assess air pollution exposure, the team employed Land Use Regression modeling to estimate concentrations of five pollutants at each participant’s residential address: fine particulate matter known as PM2.5, coarse particulate matter classified as PM2.5−10, the larger particulate fraction PM10, nitrogen dioxide (NO2), and nitrogen oxides (NOX). These pollutants, produced primarily by traffic, industry, and combustion sources, have long been associated with a range of physical health problems, but their role in psychiatric and behavioral conditions has remained far less certain.

For residential greenness, the researchers turned to the 2005–2007 Land Cover Map, measuring three distinct aspects of the environment within a 300-meter buffer around each home: the proportion of green space, the presence of a domestic garden, and the extent of natural environment. This tripartite approach allowed the investigators to distinguish between different types of vegetation and their potentially different effects on health.

Using Cox proportional hazards models, the team calculated hazard ratios with 95 percent confidence intervals to quantify the association between each exposure and the risk of developing a substance use disorder. The results were striking in their consistency. All five air pollutants were positively associated with incident SUD in fully adjusted models that controlled for a wide range of potential confounders.

Fine particulate matter, or PM2.5, showed the strongest effect, with a hazard ratio of 1.111 (95 percent CI: 1.081–1.142), followed by PM10 with a hazard ratio of 1.070 (95 percent CI: 1.054–1.086). These figures indicate that higher exposure to particulate pollution was associated with a measurable increase in the likelihood of developing a substance use disorder over the follow-up period, even after accounting for demographic, socioeconomic, and behavioral factors.

On the protective side of the ledger, two of the three greenness measures emerged as significant. Green space was inversely associated with SUD risk, showing a hazard ratio of 0.997 (95 percent CI: 0.996–0.998), and the natural environment measure produced an identical finding, with a hazard ratio of 0.997 (95 percent CI: 0.995–0.998). While these effect sizes appear modest on a per-unit basis, they represent population-level signals across a very large cohort, suggesting that even incremental increases in neighborhood greenness may contribute to reduced vulnerability.

Perhaps the most intriguing finding of the study concerns the interaction between pollution and greenery. The researchers tested whether residential greenness modified the association between air pollutants and SUD risk by introducing cross-product terms into their models and evaluating them with likelihood-ratio tests. The results pointed to a buffering role for domestic gardens specifically. Among participants in the high-exposure subgroup, the presence of a domestic garden attenuated the positive associations between pollutants and SUD, with cross-product terms of 0.979 (95 percent CI: 0.963–0.996) for PM2.5 and 0.987 (95 percent CI: 0.980–0.995) for PM10.

In practical terms, this suggests that having access to a private garden may partially offset some of the harm associated with living in areas of elevated particulate pollution, offering a form of environmental resilience that public health advocates have long hypothesized but rarely demonstrated with prospective cohort data.

The associations were not uniform across the population. The positive relationships between air pollution and substance use disorder risk were most pronounced among males, participants aged 65 and older, and individuals with a history of smoking. These findings highlight the importance of considering demographic and behavioral modifiers when assessing the health consequences of environmental exposures, and they suggest that certain subgroups may benefit disproportionately from interventions aimed at reducing pollution or increasing green cover.

To ensure the robustness of their findings, the team conducted a series of sensitivity analyses. These included restricting the sample to long-term residents to minimize exposure misclassification from relocation, using alternative outcome definitions based on ICD-10 codes F10-F19 and F11-F19 to capture different ranges of substance-related diagnoses, and recomputing greenness measures at a wider 1000-meter buffer around each residence. The core conclusions held across these variations, strengthening confidence in the results.

The study was performed in accordance with the ethical principles of the Declaration of Helsinki and was registered with the UK Biobank under application number 46478. Ethical approval for the UK Biobank was granted by the North West Multi-centre Research Ethics Committee under approval reference 11/NW/0382. All participants provided written informed consent before any study-related procedures were initiated.

The work was supported by the China Postdoctoral Science Foundation, the Natural Science Basic Research Plan in Shaanxi Province of China, and the Fundamental Research Funds for the Central Universities. Corresponding authors are Bolun Cheng and Feng Zhang, both affiliated with Xi’an Jiaotong University, with Zhang also holding a position in the Department of Psychiatry at the First Affiliated Hospital of Xi’an Jiaotong University.

The findings come amid growing scientific interest in the neurological and psychiatric consequences of air pollution. Particulate matter, particularly PM2.5, is small enough to cross the blood-brain barrier and has been implicated in neuroinflammation, oxidative stress, and alterations in brain structure and function. These biological mechanisms could plausibly influence reward processing, impulse control, and stress reactivity, all of which are central to the development and perpetuation of substance use disorders.

Conversely, greenness has been associated with stress reduction, improved mental well-being, greater opportunities for physical activity and social interaction, and lower levels of psychological distress, each of which may buffer against the onset of addictive behaviors. The current study adds a significant piece of evidence to this emerging picture by demonstrating the relationship prospectively in one of the world’s largest and most deeply phenotyped biobanks.

The authors conclude that long-term exposure to ambient air pollutants was associated with a higher risk of incident substance use disorders and that residential greenness showed protective associations, with the most pronounced patterns observed among older males and individuals with a smoking history. They argue that the findings support integrating greenness exposure into environmental frameworks for the prevention of substance use disorders, a recommendation that carries implications for urban design, housing policy, and environmental regulation.

As cities around the world continue to grapple with the twin challenges of air quality and mental health, studies of this scale provide an evidence base for interventions that address both simultaneously. The prospect that something as tangible as a garden could offer measurable protection against the psychiatric consequences of polluted air adds a compelling dimension to ongoing debates about how the environments we build shape the minds that inhabit them.

The article is published open access under a Creative Commons Attribution 4.0 International License, permitting broad sharing and adaptation with appropriate credit to the original authors and source. It was received on 08 December 2025, accepted on 28 August 2026, and published on 09 September 2026.

When surgeons remove a portion of a cancerous lung, one of the most frustrating complications they can face is air. A persistent leak from the lung’s cut surface or staple lines — known as prolonged air leak — keeps chest tubes in place far longer than anyone would like, extends hospital stays, inflates costs, and elevates the risk of infection and other downstream problems. Now, a research team in China has developed a machine learning tool that predicts, with striking accuracy, which older lung cancer patients are most likely to develop this complication before it ever occurs. The study, published in BMC Cancer, combines cardiopulmonary exercise testing data with an interpretable artificial intelligence model and could reshape how thoracic surgical teams prepare for and manage one of the most common postoperative challenges in lung surgery.

The research was led by Zixiao Wang, Jiabao Ding, Huagang Liang and colleagues at the First Hospital of Qinhuangdao, in collaboration with Kai Wang of Tianjin Chest Hospital. It addressed a well-recognized gap in thoracic surgical practice. While prolonged air leak — typically defined as air leaking from the chest drainage system for more than five days after surgery — is among the most frequent complications following video-assisted thoracoscopic anatomical lung resection, the prediction models available to clinicians have historically lacked rigorous external validation and, critically, interpretability. Surgeons have been asked to trust risk scores without understanding where those risks come from, a barrier that has limited clinical uptake of computational tools in the operating room planning process.

To build their model, the investigators enrolled patients aged 60 and older with lung cancer who underwent thoracoscopic anatomical lung resection — either lobectomy, the removal of an entire lung lobe, or segmentectomy, the removal of a smaller anatomical segment. The study was designed as a dual-centre cohort investigation with both retrospective and prospective components. A total of 471 patients treated at Qinhuangdao First Hospital between January 2024 and June 2025 formed the model development cohort. A separate group of 140 patients from Tianjin Chest Hospital, enrolled prospectively between June and August 2025, provided the external validation set — a crucial test of whether the model generalized beyond the institution where it was built.

The methodology followed the emerging best-practice playbook for clinical machine learning. After preliminary data processing, the development cohort was randomly split into a training set containing 70 percent of patients and an internal validation set holding the remaining 30 percent. Because prolonged air leak affects only a minority of surgical patients, the training data were inherently imbalanced, with far more non-leak cases than leak cases — a condition that can fool machine learning algorithms into simply predicting that no one is at risk. The researchers countered this with SMOTE, the Synthetic Minority Over-sampling Technique, which generates synthetic examples of the underrepresented class to rebalance the dataset and force the algorithms to learn the genuine signature of risk.

Feature selection was performed using LASSO regression — least absolute shrinkage and selection operator — a statistical technique that penalizes model complexity and drives the coefficients of irrelevant variables to zero, effectively filtering a large panel of candidate variables down to the most predictive few. From this process, four predictors of prolonged air leak emerged: a history of emphysema, the smoking index, a previous history of thoracic surgery or chest trauma, and the ventilatory equivalent for carbon dioxide, abbreviated VE/VCO₂. This last variable is drawn directly from cardiopulmonary exercise testing, a stress test performed on a bicycle or treadmill that measures how the heart, lungs, and circulatory system respond together to increasing exertion. VE/VCO₂ reflects the efficiency of ventilation during exercise — how much air a patient must move to eliminate a given amount of carbon dioxide — and elevated values signal impaired pulmonary gas exchange and often underlying parenchymal lung disease.

With the four features in hand, the team constructed six competing machine learning models: logistic regression, support vector machine, decision tree, extreme gradient boosting, and artificial neural networks, among others. Each was evaluated using a battery of metrics including the area under the receiver operating characteristic curve, or AUC, the F1 score, and the Brier score, which measures the accuracy of probabilistic predictions. The winner was XGBoost, an ensemble method that builds a sequence of decision trees, with each new tree trained to correct the residual errors of its predecessors. In internal validation, XGBoost achieved an AUC of 0.903 (95 percent confidence interval, 0.834 to 0.960) — well within the range conventionally considered excellent discrimination — along with the highest F1 score of 0.717 and the lowest Brier score of 0.092 among all six candidates.

The external validation was arguably the study’s most important test. A model can memorize the quirks of its home institution’s patient population, equipment, and surgical practices, only to falter elsewhere. Yet when applied to the 140 previously unseen patients from Tianjin Chest Hospital, the XGBoost model maintained an AUC of 0.800 (95 percent confidence interval, 0.656 to 0.913), with a sensitivity of 0.700, a specificity of 0.833, and a Brier score of 0.133. In practical terms, the model correctly flagged seven out of ten patients who went on to develop prolonged air leak while correctly reassuring more than eight out of ten of those who did not — performance that, if replicated in routine practice, would give surgeons a meaningful window to intervene preemptively.

Interpretability was treated not as an afterthought but as a core design requirement. The researchers applied SHAP — Shapley Additive Explanations — a technique borrowed from cooperative game theory that assigns each feature a contribution value for every individual prediction. Rather than presenting clinicians with an opaque probability, SHAP reveals how much each of the four variables pushed a given patient’s risk up or down, and in which direction. Through this analysis, VE/VCO₂ emerged as the single most influential feature, followed by the other three predictors ranked by correlation. The finding is physiologically plausible: patients whose lungs ventilate inefficiently under the stress of exercise are precisely those whose lung tissue is most fragile and least likely to seal spontaneously after surgical manipulation. A history of emphysema, a heavy smoking history, and prior chest surgery or trauma each independently compound this tissue vulnerability.

The clinical utility of the work extends beyond prediction to prevention. Armed with a preoperative risk estimate, surgical teams can tailor their approach for high-risk patients. Options include more meticulous staple-line reinforcement, the use of sealants or pledgeted sutures on fragile lung tissue, choosing segmentectomy over lobectomy where oncologically appropriate, planning for ambulatory drainage systems on discharge, or scheduling earlier follow-up to catch and manage leaks before they become complicated. To lower the barrier to bedside use, the team built a web-based risk calculator that allows clinicians to enter a patient’s four variables and receive an individualized probability of prolonged air leak — no programming expertise required.

The significance of incorporating cardiopulmonary exercise testing into a surgical prediction model deserves particular emphasis. Traditional preoperative pulmonary assessment in lung cancer surgery has relied on static spirometric measures such as forced expiratory volume in one second, or FEV1, and the diffusing capacity for carbon monoxide, DLCO. These tests measure lung function at rest and provide only a partial picture of cardiopulmonary reserve. CPET, by contrast, captures the integrated response of multiple organ systems under physiological stress, and the VE/VCO₂ slope in particular has gained recognition as a sensitive marker of ventilatory inefficiency in chronic obstructive pulmonary disease and heart failure. The present study suggests that this dynamic measure carries prognostic information for surgical outcomes that static tests do not fully capture — a finding with implications well beyond air leak prediction.

Several caveats frame the work appropriately. The study population was limited to Chinese patients aged 60 and above undergoing video-assisted thoracoscopic resection, and generalization to younger patients, other surgical techniques such as robotic-assisted resection, or different health care systems remains to be demonstrated. The external validation cohort, while methodologically essential, comprised 140 patients, and wider multicenter prospective trials would strengthen confidence in the model’s performance. The published article is an accepted, peer-reviewed author manuscript shared early ahead of the final version of record. The study was funded by the Beijing Kechuang Medical Development Foundation, with the funder playing no role in study design, analysis, or manuscript preparation, and the authors declare no competing interests.

Even so, the study stands as a compelling example of where predictive medicine is heading. By pairing a rigorous dual-centre validation strategy with explainable artificial intelligence and a physiologically grounded predictor, the researchers have moved machine learning in thoracic surgery closer to genuine clinical utility rather than proof-of-concept novelty. For the growing population of older patients facing lung cancer surgery, a simple preoperative test and a few clicks in a web calculator may soon determine who needs extra surgical vigilance — and spare many of them the days of tethered chest drainage that have long been an accepted, if unwelcome, part of recovery.

In one of the most comprehensive investigations of its kind, researchers have taken a major step toward untangling the biology underlying suicide risk, moving the field beyond long-standing debates about whether suicidal behavior is simply a consequence of psychiatric illness. By integrating large-scale genetic data with cutting-edge approaches that map how genes act in specific brain cells and circuits, an international team has identified shared genetic liability that cuts across psychiatric diagnoses while also pinpointing druggable biological pathways that could one day inform prevention strategies. The study, published in Translational Psychiatry, offers a sweeping synthesis of how inherited risk factors may translate into molecular vulnerabilities, and it does so with a level of resolution that previous candidate-gene approaches could never achieve.

Suicide remains one of the leading causes of death worldwide, claiming more than 700,000 lives each year according to global health estimates. Despite decades of research, clinical prediction remains stubbornly imprecise, and biological models of suicidal behavior have historically been hampered by small sample sizes, inconsistent findings, and a tendency to conflate suicide risk with the psychiatric disorders with which it so often co-occurs. Major depression, bipolar disorder, schizophrenia, and substance use disorders all carry elevated suicide risk, yet most people with these conditions never attempt suicide, while some individuals who die by suicide have no diagnosed psychiatric illness at all. This paradox has long suggested to researchers that suicidal behavior may possess partially distinct biological underpinnings, and the new study set out to test that hypothesis with unprecedented statistical power.

At the heart of the work lies genome-wide association study methodology, or GWAS, which scans hundreds of thousands to millions of genetic variants spread across the entire genome in very large groups of people, comparing individuals who have experienced suicidal thoughts or behaviors with those who have not. Each variant is tested for statistical association with the outcome, and the results can be combined into polygenic risk scores, numerical summaries of a person’s inherited liability based on the cumulative small effects of thousands of variants. Because no single common variant exerts a large effect, the field depends on aggregating data across massive biobanks and cohorts, a strategy that has only become feasible in the past decade as datasets containing hundreds of thousands of individuals became available.

The research team took particular care to address a central confound: psychiatric disorders themselves are highly heritable, and any genetic signal linked to suicide could simply reflect inherited liability to depression or schizophrenia rather than to suicide specifically. To disentangle this, the investigators applied sophisticated statistical techniques, including genetic correlation analysis and conditional modeling, which allow researchers to ask how much genetic overlap remains between suicidal behavior and psychiatric conditions after accounting for their shared heritability. Their findings revealed a complex architecture. Substantial genetic correlations confirmed that suicide risk shares a meaningful fraction of its inherited basis with major psychiatric disorders, supporting the idea of a common pool of genetic liability. Yet the analyses also uncovered signals that persisted beyond what could be explained by those disorders, pointing to biological mechanisms that may be more specific to the emergence of suicidal thoughts and actions.

A key conceptual innovation of the study is its framing of “shared genetic liability” as a bridge between diagnoses. Rather than treating suicide as an epiphenomenon of any single disorder, the researchers modeled it as a transdiagnostic phenomenon, one whose risk emerges partly from inherited vulnerabilities that cut across diagnostic categories. This perspective aligns with a growing movement in biological psychiatry that emphasizes symptom dimensions and mechanisms over traditional diagnostic boundaries. From this vantage point, the genetic variants that raise risk for depression, PTSD, and schizophrenia may do so partly because they converge on shared molecular pathways, and those same pathways may govern an individual’s threshold for suicidal crises regardless of which diagnosis they carry.

Identifying genetic variants, however, is only the first step. A GWAS signal typically points to a genomic region rather than a specific causal gene or mechanism, since many associated variants fall in non-coding stretches of DNA that regulate gene activity from a distance. To translate statistical associations into biological insight, the team deployed a battery of functional genomics approaches. Transcriptome-wide association studies, known as TWAS, integrate gene expression data to identify genes whose genetically predicted expression levels are linked to suicide risk. Tissue and cell-type enrichment analyses then ask whether the implicated genes are unusually active in particular organs or cell populations. The results converged strikingly on brain tissue, and more specifically on neuronal populations, providing support for the idea that inherited suicide risk is mediated largely through gene regulation within the central nervous system.

The study’s most clinically provocative findings concern druggable biology. Using databases that catalog the protein targets of approved and investigational medications, the researchers examined whether genes implicated in suicide risk were over-represented among known drug targets. The analysis identified several classes of candidate targets, including components of neurotransmitter signaling systems, ion channels, and intracellular signaling cascades that are already the focus of pharmacological development for other conditions. This “druggable genome” approach, which has gained traction in genetics research across many diseases, offers a pragmatic shortcut for translational science: rather than waiting decades for entirely new therapeutic classes to be invented, researchers can prioritize existing or in-development compounds whose molecular targets intersect with genetically implicated pathways.

Among the pathways receiving attention in the analysis were those involving glutamatergic and GABAergic neurotransmission, the primary excitatory and inhibitory systems of the brain, which have repeatedly surfaced in studies of mood disorders and stress response. The endocannabinoid system and inflammatory signaling pathways also appeared in the convergent analyses, consistent with a broader literature linking immune dysregulation and allostatic load to suicidal ideation. Importantly, the authors emphasize that these findings identify hypotheses rather than ready-made treatments. Genetic association with a drug target does not guarantee that modulating that target will reduce suicide risk, and any therapeutic application would require extensive preclinical validation and carefully designed clinical trials. Nevertheless, the work provides a principled, data-driven map for where to look, replacing decades of largely failed candidate-gene speculation with a genome-wide evidence base.

The methodological rigor underpinning the study deserves emphasis, because the genetics of suicidal behavior has a troubled history. Early candidate-gene studies, which typically examined a handful of variants in genes related to serotonin signaling, produced findings that largely failed to replicate and in some cases were later shown to reflect statistical artifacts, small samples, and publication bias. The field has since undergone a reckoning, and modern efforts adhere to far stricter standards: preregistered analyses, very large and ancestrally diverse cohorts, replication in independent samples, and conservative multiple-testing corrections that demand genome-wide significance thresholds on the order of five times ten to the negative eighth power. The new study reflects these lessons, applying sensitivity analyses to guard against confounding from psychiatric comorbidity and examining whether findings generalize across ancestral groups, an issue of considerable importance given that most genetic discovery to date has been conducted in populations of European ancestry.

The implications for clinical practice, while still distant, are worth considering. Polygenic risk scores for suicide-related outcomes currently explain only a small fraction of individual differences in risk, far too little to be used as stand-alone screening tools, and their predictive performance is degraded in populations not represented in the discovery data. Researchers in the field are careful to stress that genetic risk is not destiny: suicidal behavior arises from an intricate interplay of inherited vulnerability, mental illness, trauma, substance use, social isolation, access to lethal means, and the availability of support. What genetics offers is not a crystal ball but a window into mechanism, and mechanistic knowledge is the raw material from which better biomarkers, risk stratification tools, and ultimately targeted interventions are built. In the near term, the druggable pathway findings are most valuable as a guide for laboratory science, directing neurobiologists toward molecular systems whose manipulation might alter suicide-relevant phenotypes in animal models and cellular assays.

Looking forward, the study signals where the field must go next. Sample sizes for suicide-specific genetic analyses remain smaller than those for major psychiatric disorders, and expanding representation across ancestries is essential if the benefits of this research are to be shared equitably. Integrating genetic data with longitudinal measures of life stress, neural imaging, and digital phenotyping may eventually allow researchers to model how inherited liability interacts with environmental triggers across the lifespan. Functional experiments in induced pluripotent stem cell-derived neurons and brain organoids could test whether the implicated genes genuinely alter stress-response circuitry or resilience. And randomized trials in patients carrying high genetic liability, though ethically and logistically complex, represent the logical endpoint of the druggable-target approach.

What the study ultimately delivers is a reframe. Suicide risk, in this analysis, is not an amorphous clinical outcome but a measurable, partly heritable trait whose genetic architecture is now being resolved with the same tools that have transformed our understanding of heart disease, diabetes, and schizophrenia. By demonstrating that shared genetic liability links suicide risk across diagnostic boundaries and by mapping that liability onto specific, pharmacologically tractable biology, the researchers have converted a long-standing clinical puzzle into a concrete research agenda. The path from genetic correlation to effective prevention is long, and no genetic finding will soon replace the human work of crisis intervention, connection, and care. But for a problem that has defied biological explanation for generations, the identification of shared, druggable mechanisms represents a genuine and hard-won advance, one that may shape suicide research for years to come.