The prospective Health, Alcohol and Psychosocial factors in Eastern Europe (HAPIEE) cohort study has been designed to investigate the risk factors for high rates of mortality and cardiovascular diseases in Eastern European countries (e.g., Czechia, Poland, and Lithuania) (36). It includes randomly selected people with a mean age of 59 ± 7.3 years from population registers in urban centres in Czechia, Poland, and Lithuania (N = 26,746). The centres of data collection were located in seven towns in Czechia and in big cities such as Krakow in Poland and Kaunas in Lithuania. Data on age, sex, health status, medical examination, lifestyle, and socioeconomic and psychosocial factors were collected during 2002–2005. The follow-up survey in the Czechia and Poland and the baseline survey in Lithuania were conducted in 2005–2008 with the use of face-to-face computer-assisted personal interviews combined with clinical examination.

The follow-up time was estimated based on deaths occurring until the end of 2020 in Czechia, until 31 July 2017 in Poland, and until 31 March 2019 in Lithuania. Persons with complete follow-up data were included in the study. Participants were censored on the date of CVD event or the end of the study depending on data availability for each country.

All participants provided written informed consent. The study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Joint UCL/UCLH Committees on the Ethics of Human Research (Committee Alpha), reference 99/0081; the Ethics Committee of the Kaunas Medical University (reference P1-09/2005); and Ethics Committee at the National Institute of Public Health, Prague (reference 2002-01-08/P1).

Spirometry was performed using a Micro-Medical Microplus spirometer. Participants with acute pulmonary infections and illnesses (e.g., vomiting and nausea), recent surgical procedures, and cardiovascular conditions [e.g., myocardial infarction (MI) and stroke] were excluded from testing (37). Two or more measurements of FEV1 and FVC within 150 ml variation were considered for the study (37). For each participant, the highest values of FEV1 and FVC were selected for further analysis.

Predicted values of FEV1 and FVC were obtained for all participants with age and height as main predictors separately for men and women. Z-scores of FEV1 and FVC were calculated using the Global Lung Initiative (GLI) equations (38). In addition, FEV1 was standardised by height and defined as FEV1 divided by height cubed (FEV1/height³). All three predicted spirometry indices were also expressed at levels of tertiles of their distribution for further analyses. These spirometry indices were selected based on our previous analyses demonstrating a strong relationship between these measures and all-cause and CVD mortalities in our population (24, 39).

The primary outcomes were fatal and non-fatal CVD events, defined as a composite event of cardiovascular mortality including deaths due to coronary heart disease, heart failure, and stroke, and non-fatal CVD events including myocardial infarction (MI) and stroke. The outcome was defined for people without a previous history of coronary heart disease, myocardial infarction, and stroke. Dates of death were obtained from the national death registers in each country. All registers have shown a complete coverage of deaths (36). Cause-specific mortality was based on underlying causes of death, which were determined according to the selection and application rules of ICD-10 maintained by the World Health Organization (WHO). Participants were followed-up until the incident case of stroke or MI, CVD death, or end of the registration period.

Data on covariates were obtained from questionnaires and medical examination (36). The selection of predictors was based on the original FRS model including age, sex, systolic blood pressure (SBP), hypertension and the use of anti-hypertensive therapy, current smoking, diabetes, total cholesterol (TC), and high-density lipoprotein (HDL) concentrations (3). Smoking status was classified as current in case of currently and regularly smoking at least one cigarette per day. SBP was measured three times, with a 2-min interval between measurements using a digital blood pressure monitor (Omron M5-I). Cholesterol concentrations were measured in fasting venous blood samples with an enzymatic method. This method was calibrated and validated. We also identified the following self-reported comorbidities: stroke, myocardial infarction, ischaemic heart disease, hypertension (defined as measured blood pressure >140/90 mmHg and/or self-reported treated hypertension), diabetes (treated and/or untreated), asthma, and chronic obstructive pulmonary disease (COPD).

All analyses were performed with Stata (Version 17; StataCorp). Descriptive statistics are presented as means with standard deviations (SDs) or frequencies with proportions.

The FRS scores were derived using the STATA module based on validated predictors and coefficients from the FRS Cox proportional hazards regression model (40). Subsequently, FRS scores were categorised into low CVD risk (FRS < 10%), intermediate CVD risk (FRS 10%–19%), and high CVD risk (FRS ≥ 20%). The FRS model was used for people between 30 and 75 years that is within the age range of HAPIEE cohort participants. Spirometry indices were entered into the model as continuous variables and as categorical variables accounting for the level of lung function impairment with highest tertile as a reference. Linear trend was explored by adding polynomials and splines. Based on the original FRS methods, the association of predictors alone and with addition of spirometry with the risk of CVD events was estimated using Cox proportional hazards regression models stratified by country. We used the robust variance estimator to account for possible interactions between groups and multiple comparisons. Proportional hazards assumptions were confirmed by exploring parallelism of log-negative and log-estimated survival curves for each covariate (Supplementary Figures S3–S5). Hazard ratios (HRs) with their corresponding 95% confidence intervals (CIs) were also estimated by crude (included sex and age) and confounder-adjusted models. Using these data, we assessed whether each type of spirometry indices had an independent predictive value and provided additional information on the risk beyond that predicted by the FRS model.

The discriminative ability of each model was assessed with Harrell’s C-statistic applying cross-validation of predictive powers in training and testing sets (41). However, this measure was less sensitive to compare multiple models with varying number of predictors; therefore, we also employed Bayesian information criterion (BIC) and likelihood ratio (LR) for nested models as evaluation criteria for quantifying added value of spirometry indices (42, 43). Calibration was assessed with the Grønnesby and Borgan test (43–45).

We also conducted separate sensitivity analysis by fitting flexible survival models that provide smooth estimates of the baseline cumulative hazards to allow non-proportional effects of age, sex, and spirometry indices (46–49).

Altogether 26,746 individuals were recruited at baseline, of whom 14,061 met the inclusion criteria (Figure 1). Spirometry tests were performed only on random 50% of respondents in the second year of baseline survey in Poland resulting in a reduced sample size. In addition, non-response in clinical examination among 15%–20% of Czech and Polish participants was another reason for missing data. Data regarding the comparison between countries and spirometry groups are provided in Supplementary Tables 1–2 respectively.

In total, 1,690 CVD events occurred and 987 people died from other causes during the average of 13 years of follow-up (Table 1). CVD events included 463 incident MI, 381 incident strokes, and 846 CVD deaths. Those with CVD events were older (mean age 62 ± 7 years) with a higher prevalence of cardiovascular risk factors and reduced spirometry values (Table 1).

The FRS model classified the majority of the study sample as having a low cardiovascular risk (68%), and 18% and 14% of persons comprised the intermediate and high cardiovascular risk groups, respectively (Table 2). Among people with low FRS risk, a substantial proportion were women (78.4%), and the highest number of men were present in the intermediate-risk group (Supplementary Figure S1A). The distribution of CVD risk factors increased in a dose–response manner from low to high CVD risk groups (Table 2). The Z-score standardised spirometry indices showed a similar trend with slight decrease in values towards the high-risk group, whereas the highest values of lung function measured by FEV1/ht³ was seen in the intermediate-risk group (Table 2). The overall ability of the FRS model to discriminate events from non-events was good and comparable with other populations (C-statistic: 0.714).

The distribution of the cardiovascular risk categories for the FRS model alone and with addition of spirometry indices (FEV1/ht³ tertiles) is outlined in Supplementary Figure S2A. The incidence rates for CVD events per 1,000 person-years progressively increased across all FRS risk groups and from highest to lowest FEV1/ht³ tertiles (Supplementary Figure S2A). As shown in Supplementary Figure S2A, the overall incidence rate of CVD events in the moderate FRS risk group increased from 15.9 per 1,000 person-years in people with no/low lung function impairment to 23.4 per 1,000 person-years in people with severe lung function impairment. The difference in rates was more pronounced in the high-risk CVD group, where the incidence of CVD events was higher than predicted even in people with decreased spirometry indices at modest levels (Supplementary Figure S2A).

Trends were similar after adjustment for all components of the Framingham risk score with elevated compared to predicted relative risk of CVD events associated with decreasing lung functions (Supplementary Figure S2B).

The probabilities of CVD event-free survival according to baseline tertiles of FEV1/height³, FEV1, and FVC (Z-standardised) spirometry indices are presented in Supplementary Figures S4, S5. All spirometry indices remained independent predictors of CVD events after adjustment for all FRS components, with similar effect estimates of continuous measures (per 1 SD) of FEV1/ht³ [hazard ratio (HR): 1.10; 95% CI: 1.05–1.16], FEV1 (HR: 1.12; 95% CI: 1.08‒1.16), and FVC (HR: 1.12; 95% CI: 1.08‒1.18) (Table 3). However, the effect of the continuous variable was driven largely by the lowest tertile of the lung function measure. No significant improvements of the original FRS model in the C-statistic were observed in either model (C-statistic for original FRS model: 0.714 vs. for all spirometry indices: 0.715) (Table 3). However, the LR chi-square test and BIC suggested the superiority of models with the addition of spirometry. All models with continuous spirometry indices had a better fit. Both LR and BIC suggest that the model based on Z-transformed FEV1 measurement had a stronger contribution than other spirometry indices (LR: 1,079.8 and BIC: 29,663.6) (Table 3). In terms of calibration, models with added FEV1-based spirometry measures showed the improved separation between observed and expected risks of CVD events in the moderate- and high-risk groups (Supplementary Table S3 and Figure S6).

Table 4 shows relative risks of cardiovascular events by FRS groups according to increasing tertiles of spirometry indices. People with poorer lung function had a higher risk of CVD events relative to healthy individuals in all categories of FRS risk. In comparison between spirometry indices, the strongest association with CVD events was seen in the model stratified by FEV1/ht³ tertiles. Among persons with low FRS, the risk of CVD events was already elevated in people with intermediate lung function impairment (HR: 1.23; 95% CI: 1.01‒1.50) and doubled for those in the lowest FEV1/ht³ tertile (HR: 2.35; 95% CI: 1.96‒2.81).

Sensitivity analysis comparing survival models with relaxed proportional hazard assumptions for age, sex, and FEV1/ht³ are outlined in Supplementary Tables S4–S6 and Supplementary Figures S7, S8. Despite the significant time-varying effect of age, it did not improve the overall performance of the model. Similar results were observed in models with age interactions. Overall, the majority of women had a low risk of CVD events; therefore, the additive prediction of spirometry indices between sexes did not yield a valid comparison.

Several limitations deserve to be acknowledged. First, the HAPIEE cohort cannot be seen as being fully representative for Eastern Europe. It only includes several urban centres and does not include data on people from rural areas, and thus it does not cover the whole population of the included countries. Second, the overall response rates were moderately high, and participation in the clinical examination was lower among people from the Czechia and Poland. This may have introduced a selection bias. In addition, data on some risk factors including spirometry were not complete. Both these issues might lead to the underestimation of CVD risk in our study. Third, while the study protocols were identical, and the follow-up time was balanced between countries, the self-reported information in the questionnaire may be a source of reporting bias. Fourth, the composite outcome in our study did not include non-fatal CVD events related to peripheral vascular disease and heart failure, which were part of the original FRS model. This limitation may potentially underestimate our findings.

Fifth, the addition of spirometry indices to the conventional FRS model resulted in small increases in the ability to classify risk, as measured by the C-statistic. On the other hand, the improvement in model performance was detected as measured by the LR test and BIC. Such statistics must always be interpreted in the context of clinically meaningful results.

Finally, the numbers of women in the intermediate and high FRS risk groups were too small for investigations with sufficient statistical power; therefore, the comparison results between sexes were not performed in main analysis. Sensitivity analysis did not show any particular trend in the model comparison between men and women (Supplementary Table S5).