Association of Insulin Resistance With Radiographic Lung Abnormalities and Incident Lung Disease: The Framingham Offspring Study

Abstract

OBJECTIVE

Insulin resistance (IR) may be a risk factor for lung disease, but objective evidence is limited. We sought to define the relationship of longitudinal IR with radiographic imaging outcomes and examiner-identified incident lung disease in the Framingham Offspring Study.

RESEARCH DESIGN AND METHODS

Participants without baseline lung disease underwent repeated measurements of fasting insulin and glucose levels over an average period of 13.6 years, from which time-weighted average HOMA-IR was calculated. Each participant then underwent a cardiac gated whole-lung computed tomography scan, which was analyzed for the presence of emphysema, interstitial lung abnormalities (ILAs), and quantitative airway features. Incident lung disease was determined by a study examiner. The relationship of HOMA-IR to these outcomes was estimated in models adjusted for demographics, BMI, and lifetime smoking.

RESULTS

A total of 875 participants with longitudinal IR data and outcomes were identified. Their mean age was 51.5 years, and BMI was 26.7 kg/m². HOMA-IR was temporally unstable, with a within-person SD approximately two-thirds of the between-person SD. In adjusted models, a 1 SD increase in log(HOMA-IR) z score was associated with higher odds of qualitative emphysema (odds ratio [OR] 1.33; 95% CI 1.04–1.70), ILAs (OR 1.35; 95% CI 1.05–1.74), and modest increases in airway wall thickness and wall area percentage. These radiographic findings were corroborated by a positive association of HOMA-IR with incident lung disease.

CONCLUSIONS

IR is associated with radiographic lung abnormalities and incident lung disease. Deeper phenotyping is necessary to define mechanisms of IR-associated lung injury.

Graphical Abstract

Introduction

Insulin resistance (IR) is a systemic condition characterized by metabolic inflammation, heightened oxidative stress, and chronic hyperinsulinemia prior to the development of diabetes mellitus. Although the role of IR in predicting incident metabolic and cardiovascular disease has been widely confirmed, its relevancy to lung health is comparatively less understood.

Elevated IR has been consistently associated with “impaired” lung function as indicated by spirometry, specifically a lower forced vital capacity (1). Although evocative, this finding is nonspecific. Lower forced vital capacity can be caused by disorders not involving the lung and, even if attributed to the lung, cannot define the type or location of pathology.

Epidemiologic studies have implicated IR as a risk factor for chronic lung disease, but these reports are limited by single surrogate measurements of IR, uncertain temporality, and reliance on self-reported diagnoses (2,3). Insulin sensitivity also fluctuates over time and is influenced by short-term changes in diet or physical activity, increasing risk of misattribution. Translational studies have identified several mechanisms by which IR may lead to alterations in lung architecture, yet whether these findings are identifiable in humans is wholly unknown (4,5).

Here, we examine the relationship of longitudinal IR with qualitative and quantitative whole-lung computed tomography (CT) imaging findings and incident lung disease among participants in the Framingham Offspring Study. We postulate that elevated IR would be associated with imaging correlates consistent with pulmonary remodeling and corroborated with the development of lung disease.

Research Design and Methods

Cohort

The Framingham Heart Study (FHS) Offspring Cohort (hereafter, the Offspring Study) is an observational cohort study that enrolled direct descendants of FHS participants to investigate the epidemiology, risk factors, and familial heritability of cardiovascular disease. Participants in the Offspring Study were first enrolled in 1971 and were characterized with repeated study examinations approximately every 5 years (6).

We performed a secondary analysis using data from exams 5 through 8 (calendar years 1991–2008) and round 2 CT scans (calendar years 2008–2011). In this period, participants had repeated measurements of fasting insulin and glucose followed by cardiac gated whole-lung CT imaging. These scans were originally done to examine coronary artery calcifications and were later analyzed by Framingham investigators for radiographic pulmonary outcomes.

The cohort was defined as individuals who at baseline (exam 5) reported no chronic bronchitis or asthma symptoms and who had no chronic lung disease by a study examiner’s clinical impression. To ensure that insulin and glucose measurements reflected actual underlying IR and not confounding through drugs, we excluded participants using glucose-lowering medications during our study period and who did not have available IR measurements at all time points (Fig. S1).

Data were obtained from the National Heart, Lung, and Blood Institute Biologic Specimen and Data Repository Information Coordinating Center as recommended by the FHS coordinating center (7).

Primary Exposure

IR was represented by the HOMA-IR, calculated as fasting insulin (mU/L) × fasting glucose (nmol/L)/22.5, with higher levels representing worse IR (8). Fasting plasma insulin was measured at exam 5 by a Diagnostic Products Corporation radioimmunoassay, at exam 7 by a Linco radioimmunoassay, and at exam 8 by a Roche e411 electrochemiluminescence assay. To account for assay-dependent biases, we created a normalized measure of IR by log-transforming HOMA-IR and calculating its z score compared with the distribution of log-transformed HOMA-IR values from all normal-weight participants without diabetes who attended that exam, as done elsewhere (9,10). Therefore, IR at each visit was represented by a log-transformed HOMA-IR z score. The primary exposure for each participant was then calculated as their time-weighted average of this value.

As a sensitivity analysis, we also constructed parallel models using the following alternative measures of IR: the quantitative insulin sensitivity check index (QUICKI), insulin to glucose ratio (IGR), and triglyceride to glucose index (TyG) (11–13). For each of these measures, we computed time-weighted average z scores in the same manner as for HOMA-IR.

Imaging Outcomes

Between 2008 and 2011, men at least 35 years old and women at least 40 years old and who were not pregnant or breastfeeding underwent a gated whole-lung CT scan (VCT 64-slice positron emission tomography/CT scanner; GE Medical Systems) (14). Ancillary Framingham studies subsequently analyzed these images for radiographic airway and lung parenchymal outcomes, as described in the following sections. Each study was done separately and excluded varying participants based on image suitability. We included participants if they were included in at least one of the ancillary CT studies.

Qualitative Emphysema and Interstitial Lung Abnormalities

Presence and extent of emphysema and interstitial lung abnormalities (ILAs) were evaluated by three readers each (two radiologists and a pulmonologist for ILA, and three radiologists for emphysema) in a sequential reading manner (15,16). Emphysema was categorized as present or absent by visual determination, and the extent of emphysema was grouped based on the percentage of lung involved. These categories were represented as trivial (<5% of lung affected), mild (5–25% of lung affected), and moderate to severe (>25% of lung affected). ILA was categorized as absent, indeterminate, present without fibrosis, and present with fibrosis, which was collapsed to absent/indeterminate and present (17). As previously reported, imaging parameters resulted in artifacts that affected lung density–based estimates of emphysema (18). As a result, quantitative estimates of emphysema derived from densitometry are not presented in this analysis.

Airway Measurements

One segmental and one subsegmental airway from the right upper and lower lobes were measured, from which we examined the thickness of the airway wall (in mm) and wall area percentage (cross-sectional area of the airway wall divided by the total airway cross-sectional area, multiplied by 100) (19). The site for measurement was selected by manual inspection, with each participant having, at most, four measurements. Because there were insufficient measures per participant to calculate a reliable standardized within-person Pi10 (a measure of airway thickness for a theoretical airway with a 10-mm inner perimeter), we directly used the inner perimeter of the airway as a covariate in the prediction model to account for confounding by airway size (20).

Incident Lung Disease

At each follow-up exam, an examiner determined whether a participant had a list of chronic medical conditions, based on the participant’s questionnaire responses and physical examination. Incident chronic lung disease was determined if the examiner answered affirmatively to a suspected diagnosis of asthma, emphysema, chronic bronchitis, or other lung disease (excluding pneumonia). Because each condition was not uniformly asked about at all visits, incident lung disease was defined as having any of these conditions during the study period. Diagnoses were considered as possibly present were not included in the outcome definition.

Covariates

Because smoking is a critical confounder of the relationship of IR to pulmonary outcomes, we collected information regarding lifetime pack-years of smoking and smoking status at each exam. Participants who missed an intervening exam prior to baseline were assumed to be smoking at the same frequency as measured at their next visit. We also collected information on age, sex, and BMI at each exam. Few participants (n = 4) had missing BMI despite exam attendance, and those data were singly imputed based on their most prior BMI. All covariates were selected based on their likelihood of confounding the hypothesized causal pathway between IR and lung physiology and structure.

Statistical Analysis

The relationship of average log(HOMA-IR) z score with each CT imaging outcome and incident lung disease was estimated by multivariable logistic regression (for dichotomous outcomes) and linear regression (for continuous outcomes), adjusted for baseline age, sex, BMI, smoking status (current, former, and never), pack-years of smoking, and change in BMI and change in pack-years from baseline to exam 8. For airway measurements, we fit a mixed-effects model with a random intercept at the participant level adjusted for the same variables and each airway’s internal perimeter. Robust SEs were used for wall area percentage, due to evidence of heteroscedasticity. Marginal probability plots were created to visualize the covariate-adjusted relationship between IR and each outcome. Visualizations of log(HOMA-IR) z score changes over time were created with the Sankey module (21).

All statistical analyses were performed in Stata 15 (StataCorp, College Station, TX). A two-sided P value of <0.05 denoted statistical significance. This study was performed with approval of the Institutional Review Board of Baylor College of Medicine.

Data Availability

The data that support the findings of this study are available in the National Heart, Lung, and Blood Institute Biologic Specimen and Data Repository Information Coordinating Center (https://biolincc.nhlbi.nih.gov/studies/framoffspring).

Results

Cohort

A total of 875 participants who did not have a diagnosis of chronic lung disease at baseline, did not take glucose-lowering medications during the study, and had complete data on HOMA-IR with at least one CT outcome were identified (Fig. 1). Baseline age was 51.5 years (SD 8.4), and approximately half of participants were former smokers (Table 1). The mean participant was overweight with a BMI of 26.7 kg/m² (SD 4.5). Compared with that exam’s normal-weight, population without diabetes, participants in our study had average log(HOMA-IR) z scores of 0.66 (SD 1.1), corresponding to an IR within the 75th percentile.

Figure 1.

Participant flow diagram.

Table 1.

Participant characteristics (N = 875)

Characteristic	Values
Age, mean (SD), years	51.5 (8.4)
Female sex, n (%)	486 (56)
White race, n (%)	866 (99)
Baseline smoking status, n (%)
Never	355 (41)
Former	425 (49)
Current	95 (11)
Baseline pack-years, mean (SD)	10.3 (15.4)
BMI, mean (SD)	26.7 (4.5)
log HOMA-IR, z score
Mean (SD)	0.66 (1.1)
Range	−2.33 to 5.54
Years of follow-up, mean (SD)	13.6 (1.0)

Participants were followed for a mean of 13.6 years (SD 1.0). There was significant temporal heterogeneity in log(HOMA-IR) z score, with an average within-person SD of 0.70 compared with a between-person SD of 1.04. This longitudinal variability was highest among those in the highest tertile of HOMA-IR z score at baseline (within-person SD 0.81) but was substantial across the middle (SD 0.61) and lowest (SD 0.69) tertiles. Of the 291 participants whose HOMA-IR was in the middle tertile at baseline, 54 (18%) maintained this tertile through the study (Fig. S4).

Higher Average HOMA-IR Is Associated With Presence and Extent of Emphysema and ILA

Relationships of time-weighted average log(HOMA-IR) z score with study outcomes are shown in Table 2. Within the cohort, 140 individuals (17%) had emphysema and 93 individuals (11%) had ILA. In adjusted models, a 1-unit (1 SD) increase in average log(HOMA-IR) z score was associated higher odds of emphysema (odds ratio [OR] 1.33; 95% CI 1.04, 1.70) and ILAs (OR 1.35; 95% CI 1.05, 1.74) (Fig. 2).

Table 2.

Association of average log-transformed HOMA-IR z score and study outcomes

Outcome	Unadjusted (95% CI)	Adjusted* (95% CI)
Qualitative abnormalities, OR (n = 838)
Emphysema	1.21 (1.03, 1.43)	1.33 (1.04, 1.70)
ILAs	1.26 (1.04, 1.52)	1.35 (1.05, 1.74)
Quantitative airway measures, MD (n = 695)
Wall area percentage	0.79 (0.46, 1.11)	0.16 (0.01, 0.31)2
Wall thickness (mm)	0.015 (0.010, 0.020)	0.007 (0.001, 0.012)
Incident lung disease, OR (n = 875)	1.28 (1.07, 1.53)	1.37 (1.09, 1.74)

*Adjusted for baseline age, sex, smoking status, pack-years of smoking, BMI, airway internal perimeter (for airway measures), and change in pack-years and BMI from baseline to CT scan. ²The 95% CIs were estimated using robust SE due to heteroscedasticity.

Figure 2.

Association (with 95% CIs) of average log-transformed HOMA-IR z scores with predicted probability of emphysema, ILAs, and lung disease by clinical impression adjusted for age, sex, BMI, smoking status, baseline pack-years of smoking, and change in BMI and change in pack-years from baseline to exam 8.

Of the 140 participants who had emphysema, 24 (17%) had <5% involvement, 92 (66%) had 5–25% involvement, and 24 (17%) had >25% lung involvement. In an ordered logistic regression, a 1-unit increase in average log(HOMA-IR) z score was also associated with 41% higher odds of more extensive emphysema (95% CI, 12, 77) (Fig. S2).

Of the 93 participants with ILA, 64 (69%) had ILA without fibrosis and 29 (31%) had ILA with fibrosis. In an ordered logistic regression, a 1-unit increase in average log(HOMA-IR) z score was associated with 37% higher odds of having more severe ILA (95% CI 6, 76). There was no evidence of violation of the proportional odds assumption for either model (for likelihood ratio test, P = 0.23 and 0.11, respectively) (Fig. S3).

Higher Average HOMA-IR Is Associated With Modest Airway Wall Changes

A total of 695 participants had at least one airway that was suitable for measurement. Of these, 452 (65%) had four measurements taken. The overall mean wall area percentage was 55% (SD 7), and the mean wall thickness (SD) was 1.06 (0.10) mm. In the adjusted model, a 1-unit increase in log(HOMA-IR) z score was associated with a 0.16% higher mean wall area percentage and 0.007-mm thicker mean airway wall.

Higher Average HOMA-IR Is Associated With a Higher Risk of Incident Lung Disease

Of the 875 participants, 107 (12%) were assessed as having a new clinical diagnosis of lung disease. In adjusted models, a 1-unit increase in log(HOMA-IR) z score was associated with a 37% higher odds of incident lung disease (95% CI 9, 74). Participants suspected of developing lung disease were also more likely to have radiographic emphysema or ILA (OR 1.73; 95% CI 1.12, 2.68) and higher mean wall area percentage (mean difference [MD] 1.11; 95% CI 0.03, 2.20), but without differences in mean wall thickness (MD 0.004 mm; 95% CI −0.012 to 0.020 mm).

Association of Alternative Measures of IR to Study Outcomes

The associations of time-averaged QUICKI, IGR, and TyG with all study outcomes are presented in Table S1. Lower QUICKI was associated with a higher odds of emphysema, ILA, and examiner-suspected lung disease, and increased airway wall thickness. Higher IGR was associated with a higher odds of ILA and examiner-suspected lung disease. Finally, higher TyG was associated with a higher odds of examiner-suspected lung disease.

Conclusions

In participants in the Offspring Study, higher longitudinal IR was associated with subsequent emphysema, ILAs, and airway wall changes in a manner not explained by smoking, obesity, use of glucose-lowering medications, or self-reported lung disease. Our results show that IR is a longitudinal biomarker of lung impairment and reveals imaging correlates of potential IR-associated lung injury.

IR has been associated with a higher prevalence of self-reported asthma and chronic obstructive pulmonary disease and with a higher risk of incident wheezing and respiratory symptoms (2,3). Our analysis confirms and extends these reports by identifying relationships of IR with imaging changes associated with these conditions and with objective, examiner-assessed presence of chronic lung disease. To our knowledge, this is the first study to examine IR and pulmonary outcomes in this manner and the first to identify IR as a risk factor for ILA.

Previous reports from preclinical studies have proposed potential mechanisms that may explain this finding (4,5). In murine models, insulin exposure causes peribronchial collagen deposition and airway smooth muscle hypertrophy (22,23). IR enhances bronchial epithelial expression of TGF-β1, a central regulator of airway remodeling and fibrosis (24). Metabolic inflammation and oxidative and nitrosative stress associated with IR have been suggested to promote airways inflammation and subsequent emphysema (25–27).

The radiographic outcomes, including measures of emphysema, ILA, and airway abnormalities, highlighted in this analysis represent clinically significant markers of lung health. Previous studies have described that even mild emphysema (<5%) is associated with greater risk of death and disease progression (28). This mortality risk was higher among those with moderate (5–25%) and severe radiographic (>25%) emphysema, and the present study demonstrates that IR may drive emphysema severity. Additionally, a previous study that included participants from the FHS found that the ILA was associated with higher mortality risk (16). Finally, literature has demonstrated an association between quantitative airway wall thickness and greater risk for developing obstructive lung disease and experiencing exacerbations of underlying chronic lung disease, although we acknowledge that the difference associated with IR identified in this study is modest and of uncertain clinical significance (20,29).

An association between elevated IR, either assessed by HOMA-IR or through clinical surrogates, with an impaired spirometry pattern has been long reported (1,12,30,31). For example, elevated hemoglobin A_1c predicted membership in the lowest trajectory of lung function in the Fremantle Diabetes Study (32). Our findings suggest that a principal cause of this finding stems from parenchymal and airways disease. A possible unifying explanation for an impaired spirometry pattern with our imaging findings is small airways disease, which would be initially radiographically invisible but later manifest as ILA or emphysema. A follow-up study with more detailed metabolic, inflammatory, and small airway phenotyping is necessary to confirm this possibility and to disentangle factors that direct individuals preferentially down an obstructive versus fibrotic pathway. Although qualitative emphysema and ILA have been identified as clinically relevant imaging biomarkers, incorporation of quantitative techniques to assess small airways function and textural analysis of interstitial abnormalities is warranted (33,34).

The temporality of these results supports an inference that IR precedes pulmonary changes, but a reverse relationship or shared etiology remains possible. Participants were excluded if they had a prevalent diagnosis of lung disease, but because the study cohort was within the fifth decade of life and had a relatively high smoking background, it is plausible that some had undetected or subclinical lung disease at baseline. Also, although we carefully controlled for cumulative and time-varying smoking exposure and measures of adiposity, individuals with persistently elevated IR may harbor other unhealthy behaviors and face greater risk from harmful environmental exposures that can affect lung health. Investigation within younger populations is needed to corroborate these findings.

When examining alternative measures of IR, associations of QUICKI to study outcomes were maintained, but fewer associations with IGR and TyG were identified. These latter measures of insulin sensitivity are correlated with HOMA-IR but are comparatively less accurate, raising the possibility that the attenuated relationships in our study are due to greater misclassification (35,36). We note that the directionality of the point estimates for IGR and TyG are consistent with that of HOMA-IR.

The principal strengths of this study are a well-characterized study population, a parameterization of IR that incorporates information about its magnitude and longitudinal behavior, comprehensive availability of CT imaging metrics, and use of an objective examiner rather than self-report to detect lung disease. However, some limitations are noted. The Offspring Study enrolled a predominantly White population from a single geographic area, and our results may not be generalizable to the broader population. CT scans were not done at exam 5 to confidently exclude individuals with prevalent lung disease. Finally, differences in the assays used to measure insulin resulted in the need to normalize HOMA-IR, preventing direct comparisons with HOMA-IR values reported in other studies.

In conclusion, we found that longitudinally elevated IR is associated with CT changes, including a higher prevalence of emphysema and ILA, and clinical evidence of incident lung disease. These results suggest that IR, and in particular persistently elevated IR, may be a modifiable risk factor for lung health. Further study is needed to understand potential mechanisms of IR-associated lung injury and whether early pharmacologic intervention may prevent progression of lung disease.

This article contains supplementary material online at https://doi.org/10.2337/figshare.27325497.

Funding Statement

This work was supported by funding from the American Lung Association (grant 960953 to T.D.W.); the National Institutes of Health (grants K23HL151669 [to T.D.W.], K23HL164151 [to S.R.], and K23HL173570 [to V.T.]); and the Department of Veterans Affairs, Office of Research and Development, Center for Innovations in Quality, Effectiveness, and Safety (CIN 13-413).