Highlight
Deep learning-derived chest radiograph age acceleration was associated with higher odds of prevalent preserved ratio impaired spirometry (PRISm) and obstructive lung disease (OLD) in more than 231,000 Korean adults undergoing routine health screening.
Among participants with normal baseline spirometry, accelerated chest radiograph aging also predicted future development of PRISm and OLD over a median 4.3 years of follow-up.
The findings suggest that routine chest radiographs, when paired with AI-derived biologic aging metrics, may provide an opportunistic and scalable signal for subclinical pulmonary dysfunction in middle-aged adults.
The study is clinically provocative but should be interpreted in light of its retrospective design, single-country health-screening population, and reliance on prebronchodilator spirometric definitions.
Background
Routine chest radiography is one of the most commonly performed imaging tests in medicine, but its role in screening for early chronic lung disease has traditionally been limited. Structural abnormalities on chest radiographs often appear late, and conventional visual interpretation is not designed to quantify subtle, diffuse, or early parenchymal and airway changes. In parallel, spirometry remains the reference physiologic test for detecting airflow obstruction and impaired ventilatory function, yet it is not universally performed in asymptomatic adults and is often underused outside targeted respiratory evaluation.
Artificial intelligence has reopened the question of whether a standard chest radiograph contains latent information that extends beyond conventional imaging diagnosis. One emerging approach is the estimation of biologic or phenotypic age from imaging. The difference between AI-predicted age and chronological age, sometimes called age acceleration, may reflect cumulative cardiopulmonary and systemic injury that is not overtly visible to the human eye. If so, chest radiograph-derived age could function as an opportunistic biomarker for early disease detection.
This concept is particularly relevant to PRISm and OLD. PRISm, defined by reduced FEV1 with a preserved FEV1/FVC ratio, has gained increasing attention as a heterogeneous but clinically meaningful spirometric pattern associated with respiratory symptoms, metabolic dysfunction, cardiovascular risk, and progression to chronic obstructive pulmonary disease (COPD) in some individuals. OLD, typically operationalized by a reduced FEV1/FVC ratio, captures established airflow obstruction but also includes milder or undiagnosed disease in community populations. Identifying people at risk before symptoms or advanced impairment emerge remains an important unmet need.
Study Design
Kim and colleagues conducted a retrospective cohort study of Korean adults who underwent chest radiography and spirometry as part of regular health check-ups between 2006 and 2019. The study was published in Chest on June 3, 2026.
The investigators used baseline chest radiographs to derive a CXR-Lung-Risk score relative to chronological age. Participants were then categorized into three groups according to chest radiograph-derived age acceleration: decelerated aging, defined as less than -2.0 years; reference aging, defined as -2.0 to 1.9 years; and accelerated aging, defined as 2.0 years or greater.
Two related analyses were performed. First, the cross-sectional analysis examined associations between age acceleration and prevalent spirometric abnormalities at baseline. Prevalent PRISm was defined as FEV1 less than 80% predicted with FEV1/FVC 0.70 or greater. Prevalent OLD was defined as FEV1/FVC less than 0.70. Multivariable multinomial logistic regression was used to estimate adjusted odds ratios.
Second, the longitudinal analysis evaluated whether chest radiograph age acceleration predicted incident PRISm or OLD among participants with normal baseline spirometry. Follow-up spirometry data were assessed through December 31, 2022, and multivariable Cox proportional hazards models were used to estimate adjusted hazard ratios.
The baseline cohort included 231,278 participants with a mean age of 50.7 years, standard deviation 8.8 years, and 55.0% men. The longitudinal analysis included 104,158 participants with normal baseline spirometry and a median follow-up of 4.3 years.
Key Findings
Baseline prevalence of spirometric impairment
In the overall cohort, PRISm was present in 2.2% of participants, corresponding to 5,161 of 231,278 individuals. OLD was present in 4.6%, corresponding to 10,557 participants. These prevalence estimates are consistent with a relatively healthy, middle-aged health-screening population rather than a clinic-based respiratory cohort.
Compared with the reference group, individuals with accelerated chest radiograph aging had significantly higher odds of both major spirometric outcomes. For prevalent PRISm, the adjusted odds ratio was 1.37 with a 95% confidence interval of 1.28 to 1.46. For prevalent OLD, the adjusted odds ratio was 1.58 with a 95% confidence interval of 1.51 to 1.65.
These are modest but clinically credible effect sizes. They suggest that a chest radiograph appearing biologically older than the patient’s chronological age may be capturing structural or compositional changes linked to reduced lung function, even when those changes are not formally reported on standard radiologic interpretation.
Prediction of future pulmonary dysfunction
The longitudinal analysis adds the most clinically important dimension to the paper. Among 104,158 participants with normal baseline spirometry, accelerated chest radiograph aging was associated with a higher risk of subsequently developing both PRISm and OLD.
For incident PRISm, the adjusted hazard ratio was 1.37 with a 95% confidence interval of 1.27 to 1.48. For incident OLD, the adjusted hazard ratio was 1.28 with a 95% confidence interval of 1.20 to 1.37.
This temporal relationship strengthens the inference that chest radiograph age acceleration is not merely a correlate of already-detectable impairment. Instead, it may act as an early marker of vulnerability or preclinical disease trajectory. In practical terms, a routine chest radiograph obtained for health screening or other general medical purposes may contain prognostic information about future spirometric decline.
Consistency across subgroups
The reported associations were directionally consistent across subgroups defined by age, sex, smoking status, and body mass index. Most subgroup analyses remained statistically significant. That consistency matters because it argues against the signal being confined only to classic high-risk groups such as older smokers.
At the same time, subgroup consistency should not be mistaken for proof of uniform clinical utility. The absolute risk of future lung function decline likely still varies substantially by smoking exposure, adiposity, occupational history, environmental exposures, and baseline comorbidity. The AI-derived age signal is therefore best understood as complementary rather than substitutive.
Clinical interpretation of effect size
The magnitude of association observed here is not large enough to justify immediate disease labeling based on imaging alone. However, in a population-level screening context, even moderate risk enrichment can be useful if it helps identify candidates for confirmatory testing such as repeat spirometry, smoking cessation intervention, exposure review, or longitudinal surveillance.
Because chest radiography is already widely obtained, the appeal of this approach lies in opportunistic case-finding. It does not require a new imaging study, only additional algorithmic interpretation of an existing one. That practical advantage may prove more important than any single hazard ratio.
Mechanistic and Translational Perspective
Why might chest radiograph-derived age acceleration track with PRISm and OLD? Several biologically plausible explanations exist. A chest radiograph integrates information from lung volume, hyperinflation, thoracic cage morphology, cardiac silhouette, vascular markings, diaphragmatic contour, soft tissues, and subtle interstitial or airway-related patterns. Deep learning models may detect complex combinations of these features that collectively reflect cumulative cardiopulmonary aging.
For PRISm, the signal may relate to low lung volume states, obesity-associated mechanics, early interstitial change, respiratory muscle limitation, or systemic inflammatory and metabolic burden. For OLD, the model may be capturing signatures related to hyperinflation, chronic airway remodeling, reduced vascular caliber, or smoking-related thoracic changes before formal radiographic emphysema is recognized. These hypotheses remain inferential, but they are aligned with the multifactorial nature of both spirometric phenotypes.
The study also sits within a larger trend in radiology and respiratory medicine: moving from image interpretation focused on discrete lesions toward extraction of latent prognostic phenotypes. That shift is conceptually similar to coronary calcium scoring on nongated CT or body composition analysis on routine abdominal imaging. The present work suggests chest radiography could eventually contribute to risk stratification beyond tuberculosis screening, pneumonia detection, or incidental mass identification.
Strengths of the Study
The most obvious strength is scale. A cohort of 231,278 individuals provides excellent statistical precision and allows simultaneous evaluation of both cross-sectional and longitudinal associations. The use of a large health-screening population also makes the study relevant to preventive medicine and population health rather than only specialty referral practice.
A second strength is the inclusion of incident outcomes among participants with normal baseline spirometry. This helps distinguish prognostic association from simple cross-sectional correlation. The median follow-up of 4.3 years is sufficient to detect meaningful early transitions in spirometric status, although longer observation would be valuable.
Third, the study targets clinically relevant spirometric phenotypes. PRISm has often been underrecognized despite increasing evidence of adverse outcomes, and OLD remains highly prevalent and frequently undiagnosed in the community. Any tool that improves earlier recognition of either pattern warrants attention.
Limitations and Cautions
Several limitations temper immediate clinical implementation. First, the study was retrospective and observational. Residual confounding remains possible even with multivariable adjustment. Factors such as cumulative smoking dose, occupational exposure, biomass exposure, physical activity, socioeconomic status, respiratory symptoms, or undiagnosed cardiometabolic disease may partly contribute to both chest radiograph age acceleration and spirometric decline.
Second, the cohort consisted of Korean adults undergoing routine health check-ups. This may limit generalizability to other ethnic groups, age distributions, healthcare settings, or populations with heavier disease burden. External validation in diverse cohorts is essential before broad adoption.
Third, the abstract does not specify whether spirometry was prebronchodilator or postbronchodilator, but health-screening settings commonly use prebronchodilator testing. If so, some participants categorized as OLD may not meet strict postbronchodilator COPD criteria. Likewise, the fixed-ratio threshold of FEV1/FVC less than 0.70 can misclassify airflow obstruction at age extremes compared with lower-limit-of-normal approaches.
Fourth, PRISm is heterogeneous. It may reflect obesity, restriction, submaximal effort, early obstructive disease, or mixed physiology. A chest radiograph-derived aging signal may therefore be identifying a broad vulnerability phenotype rather than a disease-specific pathway. That is useful for screening but less informative for differential diagnosis.
Fifth, the algorithmic details matter. Clinical translation will depend on calibration, reproducibility across imaging hardware and acquisition protocols, handling of comorbid abnormalities, and transparency about model training and validation. Performance drift across institutions is a real concern in AI deployment.
Clinical Implications
This study does not support replacing spirometry with chest radiograph AI. Instead, it suggests a triage or enrichment role. In settings where chest radiographs are already obtained, an elevated chest age gap could identify people who merit further pulmonary assessment, especially if they have smoking exposure, exertional dyspnea, chronic cough, obesity, or cardiometabolic risk factors.
Potential use cases include executive health screening, occupational medicine, primary care preventive visits, and integrated radiology reporting systems. A future report might, for example, indicate that a chest radiograph-derived biologic age exceeds chronological age by several years and recommend confirmatory spirometry when clinically appropriate. Such a workflow would need careful validation to avoid overtesting and alert fatigue.
Importantly, the finding may be especially relevant to PRISm, a phenotype that often falls into a gray zone. Patients with preserved FEV1/FVC but reduced FEV1 may be overlooked if clinicians focus narrowly on obstruction. An imaging-derived risk signal could help bring this group to attention earlier, potentially prompting evaluation of obesity, restrictive physiology, cardiometabolic comorbidity, or evolving chronic lung disease.
Expert Commentary
The broader literature supports the clinical significance of both spirometric patterns studied here. GOLD 2025 continues to emphasize spirometry for diagnosing airflow obstruction and recognizes the importance of identifying at-risk individuals before advanced disease develops. Separately, cohort studies have shown that PRISm is associated with symptoms, exacerbation risk, cardiovascular disease, and increased mortality, although its natural history is heterogeneous.
From a radiology perspective, the paper is a reminder that routine imaging can be a source of quantitative biomarkers, not just descriptive findings. The challenge will be converting a statistically robust association into a clinically actionable pathway that improves outcomes. That requires prospective validation, threshold optimization, health economic assessment, and demonstration that AI-triggered follow-up actually changes diagnosis timing, treatment, or prognosis.
Conclusion
In this large Korean health-screening cohort, chest radiograph-derived accelerated aging was associated with both current and future PRISm and OLD. The signal was consistent across important demographic and clinical subgroups, and the longitudinal findings suggest that chest radiograph AI may detect subclinical pulmonary vulnerability before spirometric abnormality becomes evident.
The study is best viewed as an important step toward opportunistic lung health assessment using existing imaging infrastructure. It does not establish causality, does not replace spirometry, and does not yet define an implementation standard. But it does provide compelling evidence that a standard chest radiograph contains prognostic information relevant to pulmonary function, and that AI may help unlock it.
Next steps should include external validation in non-Asian populations, prospective studies with standardized spirometry protocols, integration with smoking and symptom data, and pragmatic trials testing whether AI-augmented chest radiography improves targeted case finding for early lung disease.
Funding and Registration
Clinical trial registration: N/A.
Funding: Not reported in the abstract provided. Readers should consult the full article for funding disclosures and conflict-of-interest details.
References
1. Kim H, Chang Y, Yoon SH, Ryu S. Chest Radiograph-Derived Age Acceleration as an Early Marker of Pulmonary Dysfunction in Middle-Aged Asians. Chest. 2026-06-03. PMID: 42242547.
2. Global Initiative for Chronic Obstructive Lung Disease. Global Strategy for the Diagnosis, Management, and Prevention of Chronic Obstructive Pulmonary Disease: 2025 Report.
3. Wan ES, Balte P, Schwartz JE, et al. Association between preserved ratio impaired spirometry and clinical outcomes in US adults. JAMA. 2021;326(22):2287-2298.
4. Marott JL, Ingebrigtsen TS, Çolak Y, Vestbo J, Lange P. Trajectory of preserved ratio impaired spirometry: natural history and long-term prognosis. Am J Respir Crit Care Med. 2020;201(12):1492-1502.
5. Bhatt SP, Balte PP, Schwartz JE, et al. Discriminative accuracy of FEV1:FVC thresholds for COPD-related hospitalization and mortality. JAMA. 2019;321(24):2438-2447.
