Association of metformin with lung function decline and mortality in fibrotic interstitial lung disease: an observational cohort study

Abstract

Many patients with fibrotic interstitial lung disease (ILD) also have comorbid diabetes, for which they take metformin. Metformin use has been shown to have lung fibrosis attenuation potential based on laboratory research but studies in human populations have yielded inconsistent results. To investigate this, 2,226 patients with fibrotic ILD from the large, prospective Canadian Registry for Pulmonary Fibrosis were analyzed for lung function at baseline and up to 2-year follow up. Joint modeling was used to assess the association of baseline metformin use with longitudinal lung forced vital capacity change and survival (death or lung transplant), adjusting for age, sex, smoking pack-years, baseline lung function, medication use. Results showed that baseline metformin use was not significantly associated with attenuation of forced vital capacity decline in the overall interstitial lung disease cohort, or in the subgroups based on presence of a diagnosis of idiopathic pulmonary fibrosis or diabetes. Kaplan–Meier and Cox analyses suggested a trend toward increased mortality or transplant risk associated with metformin use in all cohorts, although this was not statistically significant after adjustment for other variables. In our study, metformin did not significantly alter lung function decline in fibrotic ILD despite laboratory evidence of antifibrotic properties.

Introduction

Fibrotic interstitial lung disease (ILD) is frequently associated with diabetes [1]. Metformin, a common treatment of diabetes, has potential to decrease lung fibrosis by activating AMP-activated protein kinase [2, 3], which blocks myofibroblast transdifferentiation induced by transforming growth factor-beta [4]. This possibility is supported by multiple laboratory studies [3, 5, 6]; however, findings in humans from claims-based studies, post hoc analyses of clinical trial datasets, and retrospective cohorts have been inconsistent [7–9]. Given the scant literature and contradictory findings regarding the potential benefit of metformin in ILD, our objective was to examine the potential antifibrotic role of metformin in its ability to attenuate lung function decline.

Methods

Patients with fibrotic ILD from the prospective eight-center CAnadian REgistry for Pulmonary Fibrosis (CARE-PF) [10] were included provided the following were available: i) physician-reported data on medications used at the time of initial assessment in an ILD clinic; ii) baseline forced vital capacity percent-predicted (FVC%) measurement within 90 days of clinic assessment; and iii) at least one more FVC% measurement over the subsequent 2-year follow-up period. The mean number of FVC% measurements per patient was 5 ± 2, and the mean duration between consecutive PFT measurements was 162 ± 97 days. There were no exclusion criteria. FVC% and diffusion capacity of the lung for carbon monoxide percent-predicted (DLCO%) were measured as part of routine care. Percent-predicted values were derived using the reference equations in use at the local testing centres at the time of measurement; a single standardized prediction equation was not applied across the entire cohort. Dates of death and lung transplant were obtained from the clinical record. Informed consent and research ethics approval were obtained for this study in accordance with the Declaration of Helsinki (coordinating centre: Providence Health Care and University of British Columbia Research Ethics Board H19-01989).

Joint modeling analysis was used to evaluate the association of baseline metformin use with longitudinal FVC% change, integrating a linear mixed-effects model for FVC% change with a Cox proportional hazards model for death or transplant. The linear mixed-effects models evaluated the interaction of baseline metformin with time of follow-up for each FVC% measurement. Fixed variables adjusted for in multivariable models included baseline age, sex, smoking pack-years, baseline DLCO%, non-metformin antidiabetic medication usage, and antifibrotic/immunosuppressant usage. Non-metformin antidiabetic usage was defined as a dichotomous variable indicating whether the person has ever used a non-metformin antidiabetic drug. This was included to control for the positive or negative effects non metformin agents may have on ILD disease progression. Antifibrotic/immunosuppressant usage was a single dichotomous variable defined as 3 months of any oral antifibrotic or immunosuppressant drug, at least 1 dose of rituximab, or at least 3 doses of IV cyclophosphamide. Kaplan Meier survival curves and Cox proportional hazards models were used to determine the association between baseline metformin use and time to death or lung transplant within the 2-year study period. The Cox proportional hazards model was adjusted for baseline age, sex, baseline FVC%, and baseline DLCO%. The presence or absence of diabetes was excluded from the models due to high collinearity (0.72) with baseline metformin use. Analyses were performed for the full cohort and repeated in subgroups of patients with idiopathic pulmonary fibrosis (IPF) and non-IPF fibrotic ILD. A subgroup analysis was also conducted which only included patients who have a diagnosis of diabetes. All analyses were conducted using R Statistical Software [11].

Results

There were 2,226 included patients, with 582 (26%) having IPF (Table 1). After censoring for death and lung transplant, mean follow-up was 597 days with 84.4% of patients alive and not transplanted at 1 year. The non-IPF group comprised of 843 patients with connective tissue disease-associated ILD, 187 with hypersensitivity pneumonitis, and 614 with unclassifiable ILD. The IPF subgroup had a greater proportion of males (75% vs. 42%), ever-smokers (77% vs. 57%), presence of cardiovascular disease (defined as coronary artery disease, congestive heart failure, or atrial fibrillation; 27% vs. 14%), and baseline metformin users (13% vs. 9%), but a smaller proportion of prednisone users (4.8% vs 24%). Baseline age and lung function were similar between IPF and non-IPF groups (age 70±9 years vs. 60±13 years, FVC% 80±19% vs. 78±20%, DLCO% 57±18% vs. 62±20%, respectively). The diabetes subgroup included 350 patients, with 124 (35%) having IPF.

Table 1.

Baseline characteristics at initial visit of IPF and Non-IPF cohorts

Variable	IPF	Non-IPF
Sample size	582 (26%)	1644 (74%)
Age, years	70 ± 9	60 ± 13
Male sex	438 (75%)	688 (42%)
Ever-smoker	447 (77%)	935 (57%)
Current smoker	26 (4.5%)	85 (5.2%)
Smoking pack-years	16 (0.6 to 35)	2 (0 to 21)
Diabetes	124 (21%)	226 (14%)
Cardiac disease	155 (27%)	223 (14%)
Chronic kidney disease	26 (4.5%)	75 (4.6%)
Metformin	74 (13%)	147 (8.9%)
FVC%	80 ± 19	78 ± 20
FVC, litres	2.96 ± 0.84	2.79 ± 0.94
Number of FVC measurements per patient	5 ± 2	5 ± 2
DLCO%	57 ± 18	62 ± 20

Abbreviations: DLCO% diffusing capacity of the lung for carbon monoxide percent-predicted, FVC forced vital capacity, FVC% forced vital capacity percent-predicted, IPF idiopathic pulmonary fibrosis

Data are reported as mean ± standard deviation, number (%), or median (interquartile range)

¹The reported follow-up duration includes time up to death or lung transplant, set to a maximum of 2 years from baseline

Baseline metformin use was not associated with change in FVC% in the full ILD cohort on unadjusted or multivariable analysis (Fig. 1). Trends differed in the IPF and non-IPF subgroups but without any statistical significance. In the IPF subgroup, there was no statistically significant difference in rate of FVC% decline between baseline metformin users and non-metformin users. Similarly, in the non-IPF subgroup there was no statistical significance difference between FVC% decline between baseline metformin users and non-metformin users. Similar non-significant trends were also observed when stratified for only those patients with a diabetes diagnosis. There was no significant difference in a sensitivity analysis that also adjusted for current smoking status.

Fig. 1.

A Annual change in FVC% over a 2-year period between baseline metformin and non-metformin usage. B Annual change in FVC% over 2-year period between baseline metformin and non-metformin usage for only patients with a diagnosis of diabetes. C Cox proportional hazards regression models estimating hazard ratios for mortality (death or lung transplant) associated with metformin and non-metformin users over a two-year period. D Kaplan–Meier survival curves showing probability of death or lung transplant over the same 2-year period. Parts (A) and (B) are adjusted for baseline age, sex, smoking pack-years, baseline DLCO%, non-metformin antidiabetic medication usage, and antifibrotic/immunosuppressant usage. Part (C) is adjusted for baseline age, sex, baseline FVC%, and baseline DLCO%. Part (D) is unadjusted. Abbreviations: CI, confidence interval; DLCO%, diffusing capacity of the lung for carbon monoxide percent-predicted; FVC, forced vital capacity; FVC%, forced vital capacity percent-predicted; ILD, interstitial lung capacity; IPF, idiopathic pulmonary fibrosis

Kaplan–Meier survival curves demonstrated time to death or lung transplant over the 2-year follow-up period (Fig. 1). Baseline metformin use was associated with a greater risk of death or lung transplant on unadjusted Cox regressions. On multivariable Cox regressions, the associations between baseline metformin use and risk of death or transplant were no longer statistically significant with confidence intervals crossing the null in all 3 groups.

Discussion

In this large prospective cohort, we showed no clear association of baseline metformin use with an attenuation in lung function decline in patients with fibrotic ILD, and with a trend toward increased mortality. This finding is consistent with the findings of a pooled post-hoc analysis of 624 patients with IPF from the placebo arms of 3 recent clinical trials of pirfenidone suggesting no benefit of metformin on lung function or mortality [7]. Similarly, another separate post-hoc analysis of the same pirfenidone trials with addition of both arms of an interferon gamma-1b trial found no association of metformin with disease progression of IPF [8]. These studies contrast with a claims-based study of 3,599 adults with IPF and concomitant type 2 diabetes that showed metformin use was associated with a decrease in all-cause mortality and hospitalization; however, this benefit may be attributable to the study's focus on diabetic patients and the expected impact of metformin on typical consequences of diabetes [9].

Although there is an increased risk of diabetes in patients with IPF [1], recent Mendelian randomization studies assessing potential causal relationships between diabetes and IPF have had inconsistent results [12, 13]. This has resulted in a lack of clarity on whether the association between diabetes and IPF is due to shared risk factors (e.g., older age, cigarette smoking) or whether there could be a direct biological connection. Additionally, they may share a biological connection, but metformin may only act on and potentially improve diabetes-related outcomes, as opposed to universally improving both IPF and diabetes outcomes. Our finding that baseline metformin use is not clearly associated with outcomes in ILD further supports the likelihood that diabetes and ILD are independent biological processes that simply possess shared risk factors, consistent with other recent perspectives [14].

Our study has several limitations inherent to its observational design, most notably prohibiting clear evaluation of possible causation due to the inability to account for residual confounding in a non-randomized study design. While we initially had hoped to include diabetes in the model as a fixed variable, there was sufficient collinearity with baseline metformin use such that this was not possible. However, we were able to account for this variable, by stratifying the analysis by only patients who have a diagnosis of diabetes, which gave comparable results to the main analysis. In addition, the association between baseline metformin use and increased risk of mortality in patients with fibrotic ILD is likely confounded by indication of diabetes which may offset any mortality benefit of metformin as patients with more severe diabetes may have intensified metformin use. However, the lack of improvement in FVC% trajectory with metformin when controlling for diabetes further suggests a lack of treatment effect of metformin on pulmonary fibrosis. Joint modeling likely further reduces the risk of spurious associations by accounting for measurement error and shared random effects between longitudinal (FVC%) and survival processes [15]. In addition, only baseline medications are available in CARE-PF given the logistical challenges in obtaining reliable start and stop dates for all medications throughout the entirety of follow-up. We were therefore unable to account for changes in metformin dose and use during follow up which may further cause confounding by indication as patients with worsening ILD may have discontinued metformin. This could artificially associate worsening ILD with non-metformin use and strengthen the trend between non-metformin use and mortality, though findings were not significant. However, by controlling for non-metformin antidiabetic medication, this may also control for more severe diabetes as these agents are typically used in more poorly controlled diabetes. Lastly, more frequent and regularly scheduled pulmonary function tests would ideal; however, this is a real-world cohort for which we analyzed clinically available FVC% measurements that may have occurred at irregular time intervals with a variable number of available tests between patients.

In summary, this prospective cohort suggests that metformin does not significantly attenuate lung function decline in patients with fibrotic ILD. Most notably, we expand upon previous analyses by using joint models that better account for the dual risks of death and transplant and by including a large cohort of patients with non-IPF fibrotic ILD. Our study adds to the growing literature that suggests a lack of association between metformin and attenuation of lung function decline in fibrotic ILD despite its in vivo antifibrotic properties, which may be further clarified by a placebo-controlled clinical trial of metformin.

Abbreviations

CARE-PF

CAnadian REgistry for Pulmonary Fibrosis

DLCO%

Diffusion capacity of the lung for carbon monoxide percent-predicted

FVC%

Forced vital capacity percent-predicted

ILD

Interstitial lung disease

IPF

Idiopathic pulmonary fibrosis

Authors’ contributions

CJR conceptualized the study. LB, VM, and CJR designed the study. DA, JHF, KAJ, MK, HM, VM, DCM, and CJR acquired the data. LB, VM, and YHK analyzed and interpreted the data. LB and VM drafted the manuscript. All authors substantively revised the work, approved the submitted version, and agreed to be personally accountable for their own contributions and ensure that questions related to the accuracy or integrity of any part of the work, even ones in which the author was not personally involved, are appropriately investigated, resolved, and the resolution documented in the literature.

Funding

HM received a research grant from the University of British Columbia.

Boehringer Ingelheim provided funding for the CAnadian REgistry for Pulmonary Fibrosis (CARE-PF) registry and did not have any influence on the study design, data analysis, and manuscript writing.

Data availability

The datasets analysed for the current study are not publicly available due to the absence of research ethics board approval for this form of data sharing.

Declarations

Ethics approval and consent to participate

Informed consent and research ethics approval were obtained for this study (coordinating centre: Providence Health Care and University of British Columbia Research Ethics Board H19-01989).

Consent for publication

Not applicable.

Competing interests

Boehringer Ingelheim provided funding for the CAnadian REgistry for Pulmonary Fibrosis (CARE-PF) registry and did not have any influence on the study design, data analysis, and manuscript writing.