External validation of an IOS-derived model for airflow obstruction and COPD in a community-based cohort

Abstract

A model based on impulse oscillometry (IOS) predicting airflow limitation and chronic obstructive pulmonary disease (COPD) has been developed in hospital-based populations. However, its diagnostic performance in community-recruited cohorts and its potential complementary value when combined with commonly used questionnaire-based tools remain to be evaluated. To externally validate the IOS-derived model for airflow obstruction (AO) and COPD in a community-based cohort, and to assess whether combining it with the COPD Screening Questionnaire (COPD-SQ) improves diagnostic accuracy compared to either tool alone. This was a cross-sectional analysis of baseline data from the ongoing Early Chronic Obstructive Pulmonary Disease (ECOPD) prospective cohort study. Participants completed pre-bronchodilator IOS tests, pre- and post-bronchodilator spirometry tests. The previously developed IOS-derived model (incorporating age, sex, height, weight, and IOS parameters) was applied to this Chinese population cohort. The COPD-SQ (7 items) was also administered. We defined a pre-bronchodilator FEV₁/FVC < 0.70 as airflow obstruction (AO) and a post-bronchodilator FEV₁/FVC < 0.70 as COPD. Receiver operating characteristic (ROC) curves were used to evaluate the diagnostic performance of the IOS-derived model, COPD-SQ, and their combination for AO and COPD. Differences in area under the curve (AUC) were compared using DeLong’s test. A total of 1858 participants with complete IOS parameters were analyzed. The IOS-derived model showed acceptable diagnostic accuracy for AO (AUC 0.826, 95% CI 0.808-0.844, sensitivity 67.4%, specificity 80.6%) and COPD (AUC 0.836, 95% CI 0.818-0.854, sensitivity 71.6%, specificity 78.1%) in this community-based cohort. The combination of the IOS-derived model and COPD-SQ demonstrated modestly higher diagnostic accuracy for AO (AUC 0.839 vs 0.826 for IOS-derived model or 0.784 for COPD-SQ, all P < 0.001) and COPD (AUC 0.851 vs 0.836 or 0.795, all P < 0.001) compared to either tool alone. In this community-recruited cohort with a relatively high prevalence of airflow limitation and COPD, the IOS-derived model exhibited good diagnostic performance and, when combined with the COPD-SQ, showed modestly improved accuracy for identifying AO and COPD compared to single-modality approaches. These findings support further evaluation of the model in unselected general populations and real-world primary care settings

Introduction

Chronic Obstructive Pulmonary Disease (COPD), the third leading cause of death, is associated with high rates of disability and mortality¹. The insidious onset of COPD, often with slight symptoms in mild-to-moderate COPD stages, frequently results in missed diagnosis and delayed intervention until advanced disease, where treatment efficacy is limited². Meanwhile, patients with mild-to-moderate COPD had a faster decline in lung function than those with advanced COPD^3,4. These findings underscore the paramount importance of early detection as a pivotal strategy to effectively intervene and mitigate disease progression in COPD.

The current clinical gold standard for diagnosing COPD is a post-bronchodilator FEV₁/FVC ratio of <0.70 using spirometry⁵. However, spirometry requires strict standardization and maximal expiratory effort, which can be challenging for elderly patients or those with severe dyspnea. Moreover, in China, spirometry is not widely available in primary care settings, and the technical quality is often suboptimal. Although community screening questionnaires are suitable for large-scale screening, they exhibit low sensitivity in detecting asymptomatic or mildly symptomatic individuals. Hence, there is a need for a portable, convenient, and time-efficient technique to screen patients with COPD.

Impulse oscillometry (IOS) is a technique that measures airway resistance through quiet breathing without requiring forceful exhalation. Clinically, IOS is often used as a complementary tool for spirometry tests and can effectively detect early abnormalities in COPD⁶. Recent studies have reported that IOS can serve as an effective tool for diagnosing COPD^7–10. A multicenter cross-sectional study conducted in Chinese hospitals found that the Fres parameter demonstrated a higher diagnostic capability for COPD compared to other individual parameters⁷. Another study also revealed that BMI-adjusted airway resistance is a potential predictor of spirometry-based COPD diagnosis⁸. However, most existing studies are based on small sample sizes and, considering the distribution of the COPD population, have not fully elucidated the clinical value of IOS in diagnosing COPD. A recent multicenter IOS study involving five hospitals explored the clinical utility of IOS in diagnosing COPD and proposed an IOS-derived diagnostic COPD model, and the external validation performance of this model for identifying post-bronchodilator COPD reached 0.828¹¹. Nevertheless, the study population was primarily recruited from hospital settings. It remains unclear whether IOS-derived diagnostic COPD model is applicable for screening stable COPD in community settings. Additionally, it is not yet known whether combining the IOS early screening model with community-based screening questionnaires can improve the efficacy of COPD screening.

Therefore, We conducted a cross-sectional analysis of baseline data from the ongoing ECOPD prospective cohort to address several key scientific questions: 1) Is the IOS-derived diagnostic COPD model applicable to community-based populations? 2) How does IOS-derived diagnostic COPD model compare with screening questionnaires in COPD screening, and can their combination enhance screening efficacy?

Study design and methods

Study Design and Population

We conducted a cross-sectional analysis of baseline data from the ongoing ECOPD prospective cohort. Its rationale and design have been previously reported¹². Individuals aged 40–80 years, including with COPD and without COPD, were consecutively recruited from the community settings in northern Guangdong Province, China, specifically involving counties such as Lianping (Heyuan) and Wengyuan (Shaoguan). These areas include a mix of rural and peri-urban communities. All individuals with post-bronchodilator FEV₁/FVC < 0.70 were invited to enroll in the ECOPD cohort. Among those with FEV₁/FVC ≥ 0.70, a random subsample (approximately one-quarter, selected chronologically based on spirometry completion order) was invited to participate. Willing participants then completed additional assessments. Subjects were excluded from the study based on the following criteria: 1) < 40 years or > 80 years old; 2) uncomplete IOS test, spirometry, and COPD Screening Questionnaire (COPD-SQ) at baseline; 3) respiratory infection or exacerbations within four weeks prior to screening; All participants provided written informed consent, and the study was approved by the Ethics Committee of the First Affiliated Hospital of Guangzhou Medical University (Approval No. 2018-53).

At baseline, individuals filled up the questionnaire, including demographic characteristics, respiratory symptoms smoking status, family history of respiratory diseases, occupational exposure, biomass exposure, and COPD-SQ, which was composed of seven items (age, smoking pack-years, body mass index, cough, dyspnea, family history of respiratory diseases and biomass exposure)¹³. COPD-SQ scores ≥16 was assessed as likely COPD.

Then these individuals underwent pre-bronchodilator IOS tests, pre-and post-bronchodilator spirometry tests. In detail, firstly, Pre-bronchodilator IOS (Masterscreen IOS; Jaeger, Höchberg, Germany) tests should be conducted based on European Respiratory Society (ERS) guidelines before spirometry test¹⁴. Quality control of IOS measurements followed European Respiratory Society (ERS) guidelines and manufacturer recommendations. At least three technically acceptable maneuvers were performed per participant, and measurements were accepted only if the within-session coefficient of variation (CoV) for resistance at 5 Hz (R5) was less than 10%¹⁴. IOS parameters included resistance of the respiratory system at 5 Hz (R5), resistance of the respiratory system at 20 Hz (R20), the difference between R5 and R20 (R5-R20), reactance of the respiratory system at 5 Hz (X5), area of reactance (AX), and resonant frequency (Fres). IOS-derived model was brought up in Liu et al study¹¹, and IOS-derived model was composed of nine items: age, sex, weight, height, R5, R20, Fres, X5, and AX. The equation for estimated pre-BD FEV₁/FVC: estimated FEV₁/FVC = −0.00137 × age (years) + 0.0364 × sex (0 if male, 1 if female) - 0.00304 × height (cm) + 0.00138 × body weight (kg) + 0.0858 × R5 (kPa.s.L⁻¹) - 0.107 × R20 (kPa.s.L⁻¹) - 0.00334 × Fres (Hz) + 0.182 × X5(kPa.s.L⁻¹) - 0.0254 × AX (kPa.L⁻¹) + 1.303.

Then spirometry (CareFusion, Yorba Linda, CA, USA) was conducted before and after administration of 400 µg salbutamol for over 20 min according to the ATS/ERS guidelines¹⁵. We defined a pre-bronchodilator FEV₁/FVC < 0.70 as airflow obstruction (AO) and a post-bronchodilator FEV₁/FVC < 0.70 as COPD¹¹.

Statistical analysis

Continuous variables are presented as mean ± standard deviation (SD) for normally distributed data, and as median (interquartile range, IQR) for non-normally distributed data. Categorical variables are expressed as number (percentage). We firstly performed Receiver operating characteristic curves (ROC) to verify the prediction AO and COPD value of participants in community using IOS-derived model, and we found the suitable cutoff of IOS-derived diagnostic COPD model in community. Correlation between actual and estimated pre-bronchodilator FEV₁/FVC using IOS tests was using Pearson’s correlation analyses.

Then we estimated and compared the differences of area under the curve (AUC) of IOS parameter, IOS-derived model, and COPD-SQ using Delong test. The combined model was created by fitting a logistic regression using two predictors: (1) the continuous predicted pre-bronchodilator FEV₁/FVC value from the IOS-derived model, and (2) the raw integer COPD-SQ score. The logistic model was of the form: logit(p) = β0 + β₁ × IOS predicted FEV₁/FVC + β₂ × COPD-SQ score. The predicted probability p from this logistic regression was then used to evaluate diagnostic performance (AUC, sensitivity, specificity) for both AO and COPD. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), balanced accuracy, positive likelihood ratio (+LR), and negative likelihood ratio (−LR) were calculated at the optimal cutoff. Calibration was assessed using the Hosmer-Lemeshow goodness-of-fit test (with 10 groups) and Bland-Altman analysis (difference between spirometry-measured and IOS-estimated pre-bronchodilator FEV₁/FVC plotted against the mean of the two measurements, with 95% limits of agreement calculated as mean difference ± 1.96 × standard deviation of the differences). Proportional bias was evaluated by linear regression of the differences on the means. Multiple comparisons of AUC values were adjusted using the Bonferroni correction. All statistical analyses were performed using R version 4.4.2, with a two-sided P-value < 0.05 considered statistically significant.

Results

Baseline characteristics

Figure 1 shows flow diagram of participants selection and exclusion from the baseline ECOPD cohort.The figure shows the step-wise process leading to the final analytic sample of 1858 participants with complete pre-bronchodilator IOS, spirometry, and COPD-SQ data. Exclusions were due to missing IOS data (n = 193) and missing COPD-SQ information (n = 4), the average age of overall participants was 61.23 (SD 8.17), 951 participants were diagnosed as AO and 831 participants were diagnosed as COPD. 838 participants had any symptoms, and 825 participants were current smokers (e-Table 1).

Fig. 1.

Flow chart of the participants in ECOPD cohort.

Prediction AO and COPD in community using IOS-derived model

Figure 2 shows the ROC curve for AO and COPD using the model-estimated FEV₁/FVC by IOS. AUC of 0.826 and 0.836 for predicting AO and COPD indicates that IOS-derived diagnostic COPD model was a very strong and clinically useful predictive COPD model. The optimal cutoff value for predicting both AO and COPD was the same at 0.73. Sensitivity and specificity for predicting AO were 67.4% and 80.6%, and for predicting COPD were 71.6% and 78.1% (Table 1). The estimated FEV₁/FVC using IOS was correlated with actual pre-bronchodilator FEV₁/FVC using spirometry (r = 0.715, P < 0.001) (Fig. 3). Calibration of the IOS-derived model was assessed using bootstrap-corrected calibration plots (B = 200 replications) and the Hosmer-Lemeshow goodness-of-fit test. As shown in e-Fig. 1, both panels demonstrated excellent agreement between predicted probabilities and observed proportions for AO and COPD. The Hosmer-Lemeshow test confirmed good fit for both outcomes (AO: χ² = 7.521, df = 8, p = 0.482; COPD: χ² = 11.134, df = 8, p = 0.194). Mean absolute error was very low (0.006 for AO and 0.011 for COPD), Bland-Altman analysis revealed a small mean difference of −0.055 (95% limits of agreement: −0.240 to +0.130) between spirometry-measured and IOS-estimated pre-bronchodilator FEV₁/FVC (e-Fig. 2). However, linear regression of the differences on the means demonstrated significant proportional bias (slope = 0.807, 95% CI 0.775-0.839, p < 0.001).

Fig. 3. Correlation between actual (spirometry-measured) and estimated (IOS -derived) pre-bronchodilator FEV₁/FVC in the ECOPD cohort.

The solid blue line represents the line of perfect agreement (y = x). The black dashed line represents the linear regression fit of the differences on the means (slope = 1.496, 95% CI 1.429–1.562; intercept = -0.417, 95% CI -0.466 to -0.368; r = 0.715, P < 0.001).

Fig. 2.

Receiver operating characteristic (ROC) curve for identifying airflow obstruction (AO) and chronic obstructive pulmonary disease using the model-estimated FEV₁/FVC by IOS. Abbreviations: FEV₁/FVC, forced expiratory volume in 1 s/forced vital capacity; AUC, area under the curve.

Table 1.

Model performance at various cut-offs for identifying spirometry-defined AO and COPD.

COPD subtype	AUC	Cut-off	Sensitivity (%)	Specificity (%)	Youden Index (%)	Balanced accuracy (%)	+LR	−LR	PPV (%)	NPV (%)
Spirometry-defined AO
IOS- derived model	0.826 (0.808 - 0.844)	<0.73	67.4 (64.4–70.3)	80.6 (77.9–83.0)	48.0	74.0	3.47	0.40	78.5 (75.5-81.1)	70.2 (67.4-72.9)
Spirometry-defined COPD
IOS- derived model	0.836 (0.818 - 0.854)	<0.73	71.6 (68.4–74.6)	78.1 (75.5–80.5)	49.7	74.8	3.27	0.36	72.6 (69.4–75.5)	77.3 (74.6-79.7)

AO airflow obstruction, COPD chronic obstructive pulmonary disease, PPV positive predictive value, NPV negative predictive value, +LR positive likelihood ratio, −LR negative likelihood ratio.

A Cut-off value with maximum Youden index.

The proportion of likely COPD in community using IOS-derived model and COPD-SQ

Figure 4 shows that the proportion of likely COPD in community using IOS-derived diagnostic COPD model < 0.73 and COPD-SQ ≥ 16. The COPD-SQ questionnaire and the IOS-derived model are both effective screening tools for identifying patients at high risk for COPD. Individuals with the likely COPD groups using IOS-derived diagnostic COPD model and COPD-SQ showing a significantly higher percentage of spirometry-defined AO and COPD diagnoses (72.3% and 66.5% for the questionnaire, 78.2% and 72.4% for IOS-derived model) compared to the unlikely COPD groups. The IOS-derived model demonstrates a slightly higher accuracy in predicting actual AO and COPD than the COPD-SQ questionnaire.

Fig. 4.

Bar chart showing the independent positive rates of COPD-SQ (cutoff ≥16) and IOS-derived model (“Likely COPD” based on estimated FEV₁/FVC < 0.73) in the full ECOPD community cohort.

IOS-derived model and COPD-SQ in predicting AO and COPD in community

We performed ROC analysis to compare the difference between IOS-derived model and COPD-SQ in predicting AO and COPD in community. Our findings showed that IOS-derived model had higher diagnostic accuracy of AO (AUC: 0.826 vs 0.784, p < 0.001) and COPD (AUC: 0.836 vs 0.795, p < 0.001) than COPD-SQ. The IOS-derived model combined with COPD-SQ questionnaire had higher diagnostic accuracy of AO (AUC IOS-derived model + COPD-SQ vs IOS-derived model: 0.839 vs 0.826, P < 0.001; IOS-derived model + COPD-SQ vs COPD-SQ: 0.839 vs 0.784, P < 0.001), and COPD (AUC IOS-derived model + COPD-SQ vs IOS-derived model: 0.851 vs 0.836, P < 0.001; IOS-derived model + COPD-SQ vs COPD-SQ: 0.851 vs 0.795, P < 0.001) than any single screening technology (Tables 2–3, e-table 2-3, Fig. 5). Meanwhile, the cutoff of COPD-SQ in predicting AO and COPD was similar, as 16.5 (e-table 4-5).

Table 2.

Model performance at various cut-offs for identifying airflow obstruction.

Model 1	Model 2	AUC1	AUC2	Difference	P-value	Adjusted P-value
IOS-derived model	COPD-SQ	0.826	0.784	0.042	<0.001	<0.001
R5 + R20 + AX + X5+Fres	COPD-SQ + IOS-derived model	0.758	0.839	-0.081	<0.001	<0.001
COPD-SQ	COPD-SQ + IOS-derived model	0.784	0.839	-0.055	<0.001	<0.001
IOS-derived model	COPD-SQ + IOS-derived model	0.826	0.839	-0.013	<0.001	0.001
COPD-SQ + R5 + R20 + AX + X5+Fres	COPD-SQ + IOS-derived model	0.825	0.839	-0.014	0.007	0.067

R5 resistance at 5 Hz; R20 resistance at 20 Hz; X5, reactance at 5 Hz; Fres resonant frequency; COPD-SQ COPD Screening Questionnaire; AUC area under the ROC curve.

The Bonferroni correction is a simple and conservative method used to adjust p-values when performing multiple comparisons.

Table 3.

Model performance at various cut-offs for identifying spirometry-defined chronic obstructive pulmonary disease.

Model 1	Model 2	AUC1	AUC2	Difference	P-value	Adjusted P-value
IOS-derived model	COPD-SQ	0.836	0.795	0.041	<0.001	<0.001
R5 + R20 + AX + X5+Fres	COPD-SQ + IOS-derived model	0.774	0.851	-0.077	<0.001	<0.001
COPD-SQ	COPD-SQ + IOS-derived model	0.795	0.851	-0.056	<0.001	<0.001
IOS-derived model	COPD-SQ + IOS-derived model	0.836	0.851	-0.015	<0.001	<0.001
COPD-SQ + R5 + R20 + AX + X5+Fres	COPD-SQ + IOS-derived model	0.843	0.851	-0.008	0.095	0.954

R5, resistance at 5 Hz; R20, resistance at 20 Hz; X5, reactance at 5 Hz; Fres, resonant frequency; COPD-SQ, COPD Screening Questionnaire; AUC, Area Under the ROC Curve; Adjusted P-value using Bonferroni Significant.

Fig. 5. ROC curve for identifying airflow obstruction and COPD using COPD-SQ questionnaire, IOS parameters, combination COPD-SQ questionnaire and IOS parameters, and IOS-derived model.

A Identifying airflow obstruction (pre-bronchodilator FEV1/FVC < 0.70). B Identifying COPD (post-bronchodilator FEV1/FVC < 0.70).

Discussion

Our study provides external validation of the previously developed IOS-derived model in a community-recruited cohort of adults aged 40-80 years. The model demonstrated acceptable discriminative performance for pre-bronchodilator airflow obstruction (AUC 0.826, 95% CI 0.808-0.844) and post-bronchodilator COPD (AUC 0.836, 95% CI 0.818-0.854) in this setting. When combined with the COPD-SQ questionnaire, the approach showed modestly higher diagnostic accuracy for both AO (AUC 0.839) and COPD (AUC 0.851) compared to the IOS-derived model or COPD-SQ alone (all P < 0.001).

IOS is a tool for measuring airway resistance that requires only quiet breathing. Due to these advantages, it is usually used as a supplementary tool for spirometry tests^16–19. Some studies recommend using IOS for screening Chronic Obstructive Pulmonary Disease. Chaiwong et al. conducted IOS and spirometry tests on patients with suspected COPD in Thailand, and found that the AUC of a single IOS parameter for COPD ranged from 0.80 to 0.84¹⁰. Similarly, Chen et al. suggested that BMI-adjusted airway resistance could serve as a potential predictor for spirometry-based COPD diagnosis⁸. Liu et al. reported that IOS may act as an alternative to spirometry in elderly COPD patients for assessing disease status⁹. A recent hospital-based cross-sectional study from China indicated that Fres z-score demonstrated the best diagnostic value for identifying COPD in a clinical setting⁷. However, it is important to note that these studies generally involved small sample sizes and may be subject to certain biases, which could affect the reliability and generalizability of their predictive value. A recent large-scale study across five Chinese hospitals, which included over 20,000 subjects, utilized IOS to create a predictive model for AO and COPD¹¹. The suitability of this model for screening stable COPD in community populations, however, was not established. To address this, we performed a community-based study in northern Guangdong. Our results indicate that the IOS-screening model maintains good predictive value in this setting. A key difference emerged in the optimal cutoff value: whereas Liu et al. proposed a cutoff of 0.70 for predicting COPD, our study found the cutoff to be 0.73 in our community cohort. We believed this discrepancy may be attributed to differences in the characteristics and composition of the study populations. Their research primarily enrolled individuals aged 20 and above who underwent pulmonary function testing in a hospital setting for various reasons—a population that may have been preselected due to respiratory concerns. In contrast, our study focused on community-dwelling adults aged 40–80, including both non-COPD individuals and patients with stable COPD, identified through community-based screening. Using a cutoff value of 0.73 derived from our analysis for predicting COPD, the IOS-based screening model identified 78.2% of cases with AO and 72.4% of COPD cases within our cohort. Alternative explanations cannot be excluded, such as potential differences in device calibration, operator technique variability, or analytical approaches (handling of outliers, data preprocessing, or regression modeling details). These factors highlight the importance of standardized protocols and external validation across diverse settings to confirm the stability of the cutoff value.

For the early screening of individuals with suspected COPD, questionnaires are the preferred initial method given their simplicity, convenience, and speed. In China, the COPD-SQ is a commonly used and validated questionnaire. It consists of seven questions assessing key factors such as age, smoking history (in pack-years), body mass index, presence of cough and dyspnea, family history of respiratory diseases, and exposure to biomass fuels¹³. The COPD-SQ questionnaire utilizes a cutoff score of ≥16 to define high risk for COPD. Its screening efficacy is well-supported, with prior literature reporting an AUC of 0.812. Due to this strong performance, it has been officially recommended as a screening tool in the Chinese COPD Prevention and Management Guidelines^20–23. We compared the screening COPD efficacy of the IOS-derived model and the COPD-SQ questionnaire in a community-based setting. Our results demonstrated that the IOS-derived model outperformed the COPD-SQ questionnaire alone in screening COPD. The IOS-derived model showed modestly higher AUC compared to COPD-SQ alone for both AO and COPD. This empirical improvement in diagnostic performance supports the complementary value of combining objective physiological measures from IOS with questionnaire-based risk assessment. Since the COPD-SQ relies on symptom reporting, its ability to identify patients in the early stages of the disease, when symptoms are often mild or absent, is inherently limited. Furthermore, our results revealed that the IOS-derived model had better screening COPD efficacy than the models using only IOS parameters. The observation results aligned with some existing research; For instance, Liu et al. found that single IOS parameters were less effective in diagnosing COPD than the IOS-derived model¹¹. COPD is a heterogeneous disease with diverse phenotypes²⁴. Some patients present predominantly with elevated airway resistance, while others exhibit emphysema and reduced lung elasticity⁶. R5 reflected total airway resistance, X5 and AX reflected peripheral airway function and lung elasticity¹⁸, which may only be sensitive to a specific phenotype, potentially missing others. In contrast, the IOS-derived model integrates multiple parameters that capture distinct pathophysiological features (both airway resistance and lung elasticity), enabling a more comprehensive and accurate identification of COPD patients with significant heterogeneity.

Supplementary analyses revealed no statistically significant difference in diagnostic performance between the full IOS-derived model and a simpler combination of COPD-SQ with raw IOS parameters (R5 + R20 + AX + X5 + Fres) (AO: AUC difference +0.001, adjusted p = 1.000; COPD: AUC difference −0.007, adjusted p =1.000). This suggests that, in this external community cohort, the incremental benefit from the regression optimization and demographic adjustments is minimal, with predictive power largely driven by the core IOS parameters themselves. From a clinical and practical standpoint, the simpler raw-parameter combination may offer advantages in ease of implementation and scalability, particularly in resource-limited primary care or community settings. However, the full IOS-derived model was theoretically optimized in the original development and may retain advantages in other populations, severity strata, or longitudinal contexts. As this study focused on external validation of the published model, we did not aim to compare alternative combination strategies. Future multi-center studies are needed to determine whether model complexity translates to meaningful clinical gains in diverse settings.

While the present study used a hospital-grade IOS device (Masterscreen IOS), portable oscillometry systems are increasingly available for community and primary care use. Examples include the tremoflo C-100 (Thorasys/MIR) (https://www.thorasys.com/c-100), Resmon Pro FULL (Restech) (https://restech.it/project/resmon-pro/), FOT devices (IOS500/600) in China (https://www.e-linkcare.cn/cpzx_hxly) and portable forced oscillation devices evaluated in recent studies^25,26. These systems require only tidal breathing, minimal training, and no forced maneuvers, making them technically feasible for general practice. Small-scale studies have demonstrated successful use of portable IOS/FOT in community settings, including primary care clinics and home monitoring^9,10,27,28. Acquisition costs for portable systems typically range from USD 5000 to 15,000 (depending on model and configuration, based on vendor quotes and market data 2024–2025) (e-table 6), with additional expenses for maintenance, calibration, and operator training. Reimbursement is available in some healthcare systems in China, but cost-effectiveness for routine COPD screening in low-prevalence community populations is an area for further evaluation. The current study does not assess real-world feasibility, compliance, or accuracy in primary care or community environments. These practical challenges highlight the emerging promise of IOS for widespread community screening, which requires dedicated prospective studies in unselected settings to optimize availability, operational reliability, cost-benefit ratio, and long-term performance. Importantly, our cohort was enriched with a relatively high prevalence of airflow limitation (51%) and COPD (45%), which differs substantially from unselected general community populations where COPD prevalence is typically much lower (approximately 8-10% in China)⁵. Therefore, the results primarily support the model’s diagnostic utility in community-dwelling individuals with varying degrees of respiratory impairment, rather than its direct application as a screening tool in low-prevalence settings. The observed improvement in accuracy with the combination of IOS-derived estimates and questionnaire scores suggests a potential complementary role for these non-invasive tools in identifying at-risk individuals, although the clinical magnitude of the incremental benefit remains modest and requires confirmation in broader, representative populations.

Our study has several limitations. First, as a single-center prospective cohort, it primarily serves as an external validation cohort to explore the diagnostic performance of the IOS-model in a community setting. Future multi-community studies are needed to further validate the generalizability of the IOS-derived model. Second, with the increasing use of portable oscillometry devices (airwave oscillometry (AOS) Tremoflosystem (Thorasys, Montreal, Canada), the Resmon Pro FULL; Restech Srl)) in community screening, differences in resistance parameters across various IOS devices must be considered^25,29. Widespread screening implementation will require cross-device comparison and validation of prediction models. Third, A key limitation is the geographic specificity of the ECOPD cohort to northern Guangdong Province (rural and peri-urban areas with relatively high biomass fuel exposure and moderate ambient/occupational air pollution). This may influence IOS parameters and COPD prevalence compared to urban or other regions of China, limiting generalizability. The diagnostic performance of the IOS-derived model and its combination with COPD-SQ thus requires validation in diverse environmental settings through future multi-regional studies. Finally, although several questionnaires (LFQ, CDQ, COPD-PS, PUMA, CAPTURE) are recommended by GOLD for COPD screening^20,30–33, we used the COPD-SQ specifically because it has been validated and is widely used in Chinese populations¹³. Fifth, the enriched composition of our cohort (high prevalence of airflow obstruction and COPD, the male predominance) limits the direct extrapolation of findings to true population-based screening in general communities with lower disease prevalence. Sixth, we did not assess the operational feasibility, cost, or accuracy of IOS deployment in general practice or community screening programs, which represents an important area for future research. Finally, The IOS measurements were performed pre-bronchodilator, yet the model uses the same pre-BD parameters to predict both pre-BD airflow obstruction and post-BD COPD, creating physiological asymmetry. This asymmetry may slightly affect the precision of post-BD COPD prediction, as pre-BD IOS captures a composite of reversible and irreversible changes, but the model still showed acceptable performance in this external cohort.

Conclusion

In this external validation in a community-based cohort enriched for airflow obstruction and COPD, the IOS-derived model demonstrated good diagnostic performance for both pre-bronchodilator AO (AUC 0.826) and post-bronchodilator COPD (AUC 0.836), with modest but statistically significant improvement when combined with COPD-SQ. These findings support the potential complementary role of IOS in enhancing diagnostic accuracy beyond questionnaire-based tools in similar settings. However, the enriched cohort design limits generalizability to unselected populations, and the observed performance requires prospective validation in broader, representative community cohorts to assess real-world clinical utility, feasibility, and impact on screening outcomes.

Author contributions

P.X.R., Y.M.Z., F.W., and L.F.L designed the project and planned the statistical analysis. L.F.L drafted and revised the paper. L.F.L., F.W., J.Q.P., G.Y.T, Z.S.D., C.Q.D., G.N.C, Q.W., K.N.Z, X.H.W, H.S.T., X.Q.H., J.F.L., S.L., S.Y.H.,S.T.C, C.L.Y, Y.Q.H and S.Q.Y collected and monitored the data collection. All authors approved the final draft of the manuscript for publication. L.F.L take responsibility for the integrity of the data and the accuracy of the data analysis. L.F.L is the study guarantors.

Data Availability

Need to seek the consent of the corresponding author.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41533-026-00506-5.