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.

Even years after the acute phase of the COVID-19 pandemic faded from daily headlines, one of its most stubborn legacies continues to shadow millions of working adults: post-COVID conditions, the cluster of persistent symptoms now broadly known as long COVID. A new cross-sectional study from Taiwan, published in BMC Psychology, offers a sobering snapshot of what these conditions mean for mental health and the capacity to work, even among people who were highly vaccinated and infected predominantly during the milder Omicron era. The findings challenge a lingering assumption that vaccination and less virulent variants have fully neutralized the long-term occupational and psychological costs of SARS-CoV-2 infection.

The research, led by Jhe-Ping Lin of the Department of Occupational Medicine at Linkou Chang Gung Memorial Hospital in Taoyuan City, together with Chia-Cheng Chang, Chen-Chang Yang, Ting-An Yang and corresponding author Yu-Chung Tsao, set out to answer a question that has become increasingly urgent for occupational health practitioners worldwide: how do employed adults who report persistent post-COVID symptoms differ, in measurable psychological and functional terms, from infected peers who consider themselves fully recovered? Much of the existing long COVID literature emerged from populations infected with earlier viral variants, before widespread vaccination, and focused heavily on physical sequelae. Relatively less is known about the psychological and occupational footprint of post-COVID conditions under contemporary epidemiological conditions, which is precisely the gap the Taiwanese team sought to fill.

The study design was straightforward but methodologically deliberate. Employed adults with prior confirmed SARS-CoV-2 infection were recruited through clinical and occupational-health channels using non-probability sampling, a practical approach for reaching workers with documented infection histories. All participants completed a battery of validated questionnaires assessing four core domains: health-related quality of life, fear of COVID-19, insomnia severity and work ability. The investigators then divided the cohort into two groups based on whether participants reported persistent symptoms consistent with post-COVID conditions, and statistically compared the two groups on every psychological and occupational measure. This symptom-based categorization mirrors the approach increasingly favored internationally, since no single laboratory biomarker can yet reliably define long COVID.

The results were striking in their consistency. Participants reporting post-COVID conditions demonstrated significantly lower quality of life, with a mean difference of 13.9 points on the quality-of-life instrument compared with fully recovered peers. They also scored higher on the Fear of COVID-19 Scale, a validated measure of pandemic-related anxiety, and reported greater insomnia severity, suggesting that disturbed sleep is a common companion of persistent post-viral illness. Perhaps most consequential for employers and policymakers, the post-COVID group showed reduced scores on the Work Ability Index, a well-established instrument that quantifies how well a worker perceives themselves able to meet the physical and mental demands of their job. The mean difference of 3.3 points on this index may sound modest, but in occupational medicine research, shifts of this magnitude on the Work Ability Index are considered clinically meaningful and are associated with elevated risks of reduced productivity, prolonged sickness absence and early exit from the labor force.

To appreciate why these numbers matter, it helps to understand what each instrument measures. Health-related quality-of-life scales capture how illness erodes daily functioning across physical, emotional and social dimensions; a 13.9-point deficit represents a substantial perceived deterioration in wellbeing, not a marginal statistical blip. The Fear of COVID-19 Scale taps illness-related anxiety that can itself become disabling, feeding avoidance behaviors and hypervigilance. Insomnia severity indices quantify difficulties falling asleep, staying asleep and feeling restored by sleep, and chronic insomnia is strongly linked to depression, impaired concentration and workplace accidents. The Work Ability Index synthesizes a worker’s self-assessed physical and mental capacity, diagnosed illnesses and resources relative to job demands, making it one of the most predictive single measures available for future work disability. That all four domains moved in the same negative direction among the post-COVID group paints a coherent picture of a population under compounding psychological and functional strain.

Crucially, these associations emerged within a population that reflects the current reality of the pandemic rather than its early, pre-vaccine chapter. Taiwan maintained exceptionally high vaccination coverage, and the overwhelming majority of infections in the study cohort occurred during the Omicron-dominant period, when the variant itself was generally associated with less severe acute disease. The persistence of measurable psychological and occupational burden under these comparatively favorable conditions suggests that the risk of long-term impairment has been reduced but not eliminated. For the millions of workers globally who have been infected with Omicron-lineage viruses despite vaccination, the Taiwanese data imply that lingering symptoms remain a genuine occupational health concern rather than a relic of earlier pandemic waves.

The authors are careful about what their study can and cannot claim. Because the design is cross-sectional, capturing participants at a single point in time, causal relationships cannot be inferred. It remains possible, for instance, that pre-existing anxiety, poor sleep or low work ability influenced how participants recalled and reported persistent symptoms, rather than the symptoms driving the psychological burden. The non-probability recruitment strategy, while effective for reaching the target population, also limits generalizability; workers who chose to participate in an occupational health study may differ systematically from those who did not. Self-reported symptom status, the standard for long COVID research given the absence of a definitive diagnostic test, introduces additional subjectivity. The researchers openly acknowledge these constraints and call for longitudinal studies to clarify recovery trajectories and functional outcomes over time, which would allow investigators to determine whether reduced work ability precedes, follows or merely co-occurs with post-COVID symptomatology.

Even with those caveats, the study carries practical implications that extend well beyond Taiwan. The authors argue that mental health assessment deserves a central place in post-COVID follow-up care, not as an afterthought but as a routine component of clinical evaluation. This recommendation aligns with a growing body of international evidence linking long COVID to depression, anxiety and sleep disorders, and it reframes post-COVID conditions as a psychosomatic and occupational problem in equal measure with a physical one. For occupational physicians, the message is that a worker presenting with persistent fatigue or brain fog should not only be evaluated for physical sequelae but also screened for insomnia, pandemic-related fear and diminished work capacity. For employers, the findings underscore the value of graded return-to-work programs, workplace accommodations and access to psychological support, since untreated mental health burdens may prolong functional impairment even after physical symptoms begin to subside.

The policy stakes are considerable. Post-COVID conditions affect a substantial fraction of infected individuals worldwide, and because the condition disproportionately impairs the ability to sustain full-time work, its economic ripple effects touch social insurance systems, disability programs and national productivity. Studies conducted in vaccinated, Omicron-era populations are especially valuable for forecasting the forward-looking burden, since they describe the risk profile relevant to new infections rather than historical ones. By demonstrating a 13.9-point quality-of-life deficit and a 3.3-point reduction in work ability under exactly these modern conditions, the Taiwanese study provides quantitative anchors that health economists and policymakers can use to model the ongoing costs of long COVID in the workforce.

The research also highlights the importance of studying long COVID outside the Europe-North America axis, where much of the literature has concentrated. Taiwan’s pandemic trajectory, characterized by early containment followed by a large, late Omicron wave in a highly vaccinated population, offers a distinct natural experiment. The fact that measurable occupational and psychological impairment still emerges in this setting strengthens the argument that post-COVID conditions are a global phenomenon requiring coordinated surveillance, standardized assessment tools and sustained investment in rehabilitation services. The study received no external funding and was approved by the Institutional Review Board of the Chang Gung Medical Foundation, with all participants providing informed consent under protocols compliant with the Declaration of Helsinki.

For now, the study’s central contribution is to quantify, with validated instruments, the invisible toll that persistent post-viral symptoms exact on working adults: poorer perceived health, greater fear, worse sleep and a weakened sense of capacity to meet job demands. As health systems worldwide transition from emergency response to long-term management of COVID-19’s aftermath, evidence of this kind, gathered under the most favorable epidemiological conditions yet seen, serves as a reminder that the pandemic’s occupational and psychological dimensions are far from over. The authors’ call for longitudinal follow-up is the natural next step, and its findings will help determine whether the burdens documented here fade with time or harden into chronic disability for a vulnerable segment of the working population.