Study sample

This investigation included Framingham Offspring Study participants who attended their eighth quadrennial examination cycle when they underwent routine echocardiography and pulmonary function tests (n = 2423). For the prospective component relating pulmonary function indices to incident HF, participants with prevalent HF were excluded (n = 45), resulting in a sample of 2378 eligible for the longitudinal analyses. Approval from the Boston University Medical Center Institutional Review Board was obtained for the study protocol and all study participants provided written informed consent.

Pulmonary and echocardiographic measurements

For the pulmonary function tests (PFT), the variables FEV1, FVC and DLCO were assessed using spirometry based on a standardized protocol described previously [15]. For FEV1 and FVC, spirometry was performed three times with training of participants to exhale as hard and fast as possible. The highest FEV1 and FVC values from the three trials were used. For the DLCO measurements, tracer carbon monoxide was inhaled into the Spirometer with a deep breath (at least 90% of their vital capacity) and participants held their breath for 10–12 seconds. The difference between the inhaled and exhaled carbon monoxide was used to calculate the DLCO. This was done twice at least 4 minutes apart to allow for carbon monoxide wash out. The average of the two values was used to determine the DLCO.

For the echocardiographic measurements, the variables LVEF, LAEF, E/e’, E/A, LAD, and LV mass were assessed using transthoracic two-dimensional echocardiography with pulsed wave Doppler and tissue Doppler imaging based on a standard protocol [16]. LV systolic function was determined by calculating LVEF using the method of de Simone complemented by the visual assessment of LV systolic function [17]. LV diastolic dysfunction was determined by E/e’ (ratio of peak blood flow velocity across the mitral value to peak tissue Doppler movement of the mitral valve annulus in early diastole). We also evaluated the E/A ratio (ratio of peak blood flow velocity from LV relaxation in early diastole [the E wave] to peak blood flow velocity in late diastole caused by atrial contraction [the A wave]) [18, 19]. End diastolic parasternal long axis M-mode measurements were used to calculate LV mass using a validated equation: LV mass (grams) = 0.8 x [1.04 x (LVID+SWT+PWT)³−(LVID)³]+0.6, using LV internal diameter (LVID), LV septal wall thickness (SWT), and LV posterior wall thickness (PWT) [20]. LAD was determined using M-mode measurements for the end systolic anterior-posterior LA diameter. LA maximum and minimum volumes were determined by averaging values from the area-length method in the 2 and 4 chamber views in order to calculate the LA emptying fraction using [LA_max−LA_min]/LA_max*100 [21–23].

Covariate definitions

Data were collected during examination cycle 8 (2005–2008) using standardized protocols. Fasting blood samples were collected for blood concentrations of total and high-density lipoprotein cholesterol. All blood samples are stored at - 80°C until assayed [24]. Diabetes was defined as a fasting glucose ≥126mg/dL or use of medications to lower blood glucose. Medication history for prescribed antihypertensive and anti-diabetic agents were self-reported and reviewed by physicians. Participants self-reported if they had smoked in the year before the Heart Study examination to determine current smoking status. An average of two systolic and diastolic blood pressure measurements taken 5 minutes apart on the right arm of the seated participant by a single physician was used for analysis.

Outcomes of interest

The outcome of interest for this investigation was incidence of HF and its subtypes HF HFpEF and HFrEF. Participants were under surveillance from the time of their eighth examination until the development of a HF event or censoring. A review panel of three physicians adjudicated each HF event. The Framingham criteria were used to define a diagnosis of clinical HF [25]. LVEF at time of diagnosis was used to classify HF as HFpEF (LVEF ≥50%) or HFrEF (LVEF <50%) [17].

Statistical methods

We used multivariable-adjusted linear regression models to relate FVC, FEV1, and DLCO (dependent variables, separate model for each) to LVEF, LAEF, E/e’, E/A, LAD, and LV mass (independent variables, separate model for each). Models were first adjusted for age, sex, smoking, and height due to their reported relations with FVC, FEV1 and DLCO [26, 27] and further adjusted for weight, history of myocardial infarction (MI), diabetes, total cholesterol/high-density lipoprotein cholesterol (HDL), systolic blood pressure (SBP) and hypertension treatment. A Bonferroni-adjusted p-value of <0.0027 (0.05/18, to account for number of variables tested) was used to account for multiple statistical testing.

Cox proportional hazards regression models were used to assess the relations between pulmonary function test variables (DLCO, FEV1, and FVC; separate model for each) and time to HF after confirming that the assumption of proportionality of hazards was met. Models were adjusted for age, sex, height, weight, smoking, and history of MI, diabetes, total cholesterol/HDL, SBP, and hypertension treatment. Multivariable-adjusted splines were generated to assess for potential nonlinearity of the associations of DLCO, FEV1 and FVC with incidence of HF. Cox proportional hazards regression models for DLCO, FEV1 and FVC were repeated for incidence of HF among non-smokers, as well as for HFrEF and HFpEF separately. Kaplan Meier curves were generated to depict the unadjusted association of tertiles of DLCO, FEV1, and FVC with risk of HF and a log-rank test was evaluated. Analyses were performed in SAS version 9.4. A two-sided p-value of <0.05 was considered statistically significant.

Principal findings

First, we observed that DLCO, FEV1 and FVC were positively associated with LAEF and LVEF, while DLCO and FVC were inversely associated with E/A. Second, DLCO, FEV1, and FVC were inversely associated with risk of HF adjusting for potential confounders; these associations remained significant even after limiting analyses to non-smokers. Third, DLCO, FEV1 and FVC were inversely associated with risk of HFpEF, while DLCO was also inversely associated with risk of HFrEF.

Comparison with the literature

Pulmonary dysfunction, a progressive process in HF, is a common finding and it has been used as a prognostic factor in clinical HF [5, 8, 28–37]. However, its potential relevance as a predictor of preclinical HF has not been explored. We observed that DLCO, FEV1 and FVC were all inversely related to HF risk, similar to prior studies [5, 6]. HF is a progressive disorder, with patients advancing through the stages of HF. Most of what is known about the lungs in HF has been studied in symptomatic patients (ACC/AHA stages C and D HF); it is well recognized that LV systolic and diastolic dysfunction may be associated with increased LV filling pressures, which can cause pulmonary venous congestion [38]. Likewise, LA size is frequently a marker of LV dysfunction and is a risk factor for stroke and mortality [39, 40]. E/e’ and E/A, which are markers of LV diastolic dysfunction have also been used as indirect markers of LV filling pressures [41, 42].

The observations that DLCO, FEV1, and FVC are positively associated with LVEF, LAEF and inversely with E/e’ and E/A suggest that markers of systolic and diastolic cardiac function may be associated with subclinical alterations in pulmonary function. Subclinical cardiac alterations may influence pulmonary vascular hemodynamics, resulting in pulmonary airway and parenchymal remodeling and subsequent alterations in lung function on spirometry.

Prior studies showed that FEV1 and FVC are inversely associated with HF risk and it was proposed that pulmonary dysfunction caused cardiac dysfunction through a variety of mechanisms (systemic inflammation, oxidative stress, changes in intrathoracic pressure) [5, 6]. The foregoing studies were limited in that they did not have baseline echocardiograms for participants, and did not evaluate non-smokers separately.

Implication of findings

Our investigation suggests that asymptomatic cardiac dysfunction may be associated with lung function alterations, and such alterations may, in turn, influence the risk of developing and manifesting clinical symptoms of HF. These results suggest that cardiac dysfunction may be a precursor of pulmonary abnormalities even in the absence of HF. Current guidelines suggest against the use of pulmonary function testing in asymptomatic patients for COPD, and to our knowledge it has not been evaluated as a screening test for HF [43]. Routine echocardiographic screening for HF has not been recommended because it is impractical and expensive, and requires skilled technicians and interpreters [44]. However, pulmonary function testing is unlikely to prove to be a useful or cost-effective tool for HF screening in the community either. Nonetheless, presence of pulmonary function abnormalities in asymptomatic non-smokers may alert clinicians to the presence of occult lung disease, and also, when the former is excluded, possible subclinical HF or alterations in cardiac function.

Strengths and limitations

The strengths of our investigation include the large sample size of this community-based cohort and the availability of standardized routine PFT and echocardiograms, as well as the continuous surveillance of individuals for the incidence of HF. Some limitations must be noted. Causality of the observed associations cannot be inferred, given our observational study design. Our sample of predominantly white individuals of European ancestry may limit the generalizability of our findings to other races/ethnicities not studied.