Based on Equations 1, 2, we took deff = 5, α = 95%, p = 10.2%, r = 20%, and response rate = 90% to calculate the sample size. The referenced prevalence was derived from previous research (4).

During 2018–2019, a multistage stratified cluster sampling procedure was utilized to collect a representative sample from Gansu (Figure 1). Specifically, four monitoring sites in Gansu were selected by a probability proportional to size method (Longnan City, Jiuquan City, Qingyang City, and Gannan City). Using the same method, three towns were selected from each city, and more than two villages were further selected from each town. Subsequently, we selected a village with more than 150 households by the cluster random sampling method and randomly selected 100 families with members older than 40. Finally, the KISH table was used to select an eligible member from each family for investigation.

All participants were Chinese residents older than 40. Those physically unable to complete the spirometry examination were excluded (active tuberculosis, pregnancy, cardiovascular and cerebrovascular accidents in the past month, heart rate >120 beats/min or blood pressure >180/120 mmHg, related surgeries in the past 3 months, etc.) (16). The flowchart of screening participants was exhibited in Figure 2. Our study was approved by the ethics review committee of Guangzhou Medical University and Xi'an Jiaotong University Health Science Center.

A professional questionnaire was implemented to collect information about demographic characteristics, respiratory symptoms, and exposure to COPD-related risk factors (17). Specifically, the smokers were defined as those who had continually smoked for more than 6 months. The former smokers were defined as those who had quit smoking for more than 1 month when they were interviewed. Biomass was defined as wood, grass, animal dung, and crop waste. Coal was defined as coal and lignite. Occupational exposure was defined as exposure to dust or chemicals in the workplace for more than 1 year (4, 17). According to some references, the physiological alteration of the respiratory and circulatory systems began at altitudes >1,500 m (18–21). Thus, we chose 1,500 m to divide low and high altitudes.

Lung function was examined by EasyOne Spirometer (NDD Medizintechnik AG, Switzerland) following the operation of the American Thoracic Society and the European Respiratory Society (16). In brief, eligible participants were asked to estimate basic lung function and rechecked 15 min after inhalation of 400 μg salbutamol. Each test was acceptable only when the quality was rated A, B, or C. The overall proportion of A should be more than 70%.

The questionnaire and the spirometry examination were performed by trained staff from local medical systems.

According to the Global Initiative for Chronic Obstructive Lung Disease (GOLD) 2019, participants were diagnosed with COPD if forced expiratory volume in one second (FEV₁): forced vital capacity (FVC) was <70% after inhaling bronchodilators. The patients were categorized as the GOLD stage I (mild group: FEV₁ ≥80% predicted), GOLD stage II (moderate group: FEV₁ ≥50 to <80% predicted), GOLD stage III (severe group: FEV₁ ≥30 to <50% predicted), and GOLD stage IV (very severe group: FEV₁ <30% predicted). The modified Medical Research Council (mMRC) dyspnea scale was used to estimate the dyspnea of COPD patients (5).

In this study, another diagnostic criterion was available for sensitivity analysis: post-bronchodilator FEV₁/FVC < the lower normal limit (LLN). The LLN, predicted FEV₁ and FVC were calculated by specialized formulas that considered the characteristics of the Chinese population (22).

Continuous variables were described by mean [standard deviation (SD)] or median (lower quartile, upper quartile) and were compared by T-test or Mann–Whitney test. Categorical variables were described by rate [95% confidence interval (95% CI)] and were compared by Chi-square test or Fisher's exact test. The Cochran-Armitage test was used to compare the difference between two groups of ordinal data. The overall standardized prevalence of COPD was calculated based on the sample structure of the 2010 census of the Chinese population (23).

The 1:1 propensity score matching (PSM), with the caliper of 0.02 and in the nearest method, was implemented to balance the populations' characteristics between low and high altitudes. Subsequently, logistic regression models estimated odd ratios (Ors) and 95% CI to evaluate the association between high altitude and COPD risk (24).

Furthermore, the participants were randomly divided into the training and validation sets at the ratio of 6:4. In the training set, the univariable and stepwise logistic regression models were performed to screen the related factors to construct a nomogram for COPD (GOLD criterion). The nomogram was assessed by the receiver operating characteristic (ROC) curve, calibration curve, and decision curve analysis (DCA), which was reassessed in the validation set for internal validation.

Statistical analysis was performed by SPSS 26.0 and R 4.13. The result was considered statistically significant only when the two-sided P was < 0.05.

We interviewed 2,925 residents. Finally, 2,486 eligible participants were included in the analysis, of which 1,584 lived in low altitudes (mean: 1,190.56 m, range: 996–1,442 m) and 902 lived in high altitudes (mean: 2,684.92 m, range: 2,416–2,896 m). Compared to the low-altitude population, the high-altitude population showed significant differences in demographic characteristics except for body mass index (BMI), history of tuberculosis (TB), and coal as fuel (P < 0.05; Table 1).

There were 508 individuals diagnosed with COPD by post-bronchodilator FEV₁/FVC < 70%. The overall and overall standardized prevalences were 20.4% (18.7–22.0) and 19.7% (17.9–21.6). Taking 1,500 m as the boundary to divide low and high altitudes, we found that the prevalence in high altitudes was significantly higher than in low altitudes [high altitudes vs. low altitudes: 23.4% (20.7–26.4) vs. 18.8% (16.9–20.5), P = 0.006]. This tendency was also statistically significant in some subgroups (P < 0.05). The details were summarized in Table 2.

There were 538 individuals diagnosed with COPD by post-bronchodilator FEV₁/FVC < LLN. The overall and overall standardized prevalences were 21.7% (20.2–23.3) and 21.3% (19.5–23.3). The prevalence in high altitudes was significantly higher than in low altitudes [high altitudes vs. low altitudes: 24.7% (22.0–27.8) vs. 20.0% (18.0–22.1), P = 0.006]. This tendency was also statistically significant in some subgroups (P < 0.05). The details were summarized in Supplementary Table S1.

The formula calculating the predicted FEV₁ was suitable for the participants aged 40–81 years, so five patients older than 81 were excluded. Finally, 505 patients were included in the severity assessment. Among the eligible patients, the proportion of GOLD stage I, GOLD stage II, and GOLD stage III or IV was 67.1% (63.0–71.5), 28.7% (24.8–33.1), and 4.2% (2.6–5.9), respectively. The result of the mMRC dyspnea scale showed that the proportion of Grade 0, Grade 1, Grade 2, Grade 3, and Grade 4 was 55.1% (51.0–59.3), 17.3% (14.0–20.7), 16.7% (13.8–20.3), 5.3% (3.31–7.3), and 5.3% (3.31–7.3), respectively. Furthermore, compared to low-altitude patients, high-altitude patients had worse lung function and more serious dyspnea (both P < 0.05; Table 3, Supplementary Table S2).

Biomass as fuel and ethnic minorities almost occurred at high altitudes (Table 1). Taking them into PSM would enormously decrease the sample size. In addition, the distributions of ethnicity and biomass as fuel between cases and controls were similar to the altitude, which might lead to multicollinearity and conceal the actual association between altitude and COPD if we take them into the PSM-processed and multivariate logistic regression models. Thus, the two variables were not considered in calculating propensity score (PS) and the multivariate logistic regression model.

After matching, 699 participants were retained in each group; the two populations have similar distributions of PS and comparable characteristics (SMD < 0.2, P < 0.05; Figure 3, Supplementary Table S3). Further, logistic regression models indicated that high altitude increased COPD risk in both the criteria of FEV₁/FVC < 0.7 [crude OR: 1.323 (1.084–1.615); Adjusted OR: 1.408 (1.107–1.79); PSM OR: 1.516 (1.162–1.978)] and FEV₁/FVC < LLN [crude OR: 1.318 (1.084–1.602); adjusted OR: 1.544 (1.227–1.944); PSM OR: 1.599 (1.232–2.076); Table 4].

Participants were randomly divided into two sets. The training set had 1,491 individuals, while the validation set had 995 individuals. The difference in the characteristics between the two populations was not statistically significant (P > 0.05; Supplementary Table S4). The univariable logistic regression model first eliminated the factors that did not statistically associate with COPD. Then, the stepwise logistic regression model reserved altitude, age, sex, TB, coal as fuel, and smoking status for developing a nomogram (Table 5, Figure 4).

The ROC curve, calibration curve, and DCA were used to assess the efficiency of the nomogram. In the training set, the AUC was 0.722 (0.690–0.754); the predicted probabilities were almost identical to the actual probabilities; when the threshold probability ranged from 0.08 to 0.45, the clinical benefit was positive (Figures 5A–C). In the validation set, the AUC was 0.678 (0.634–0.722); the predicted probabilities also fluctuated around the actual probabilities; when the threshold probability ranged from 0.09 to 0.36, the clinical benefit was positive (Figures 5D–F). If removed altitude from the nomogram, the discrimination, calibration, and clinical benefit declined to some degree (Figures 5G–I).