Introduction

Idiopathic pulmonary fibrosis (IPF) is a progressive and fatal pulmonary disease with an annual cumulative prevalence of 18.2 cases per 100,000 persons in America [1] while a median survival of only 3–5 years [2, 3]. IPF is characterized pathologically by proliferation and differentiation of lung fibroblasts. Cigarette smoking, age and gender, chronic viral infections, etc., are considered as risk factors of IPF [4]. In the past decade, the understanding of IPF pathogenesis has shifted from an inflammatory-driven process [5] to the hypothesis of aberrant activation of the alveolar epithelial cells [6]. Nevertheless, the exact etiology of IPF remains unclear.

In recent years, it has been observed that IPF patients are often additionally diagnosed with diabetes mellitus (DM), resulting in increased research interest in the correlation between these two diseases. Firstly, two earlier case-controls studies [7, 8] were successively conducted in Japan but obtained opposite conclusions. Subsequent clinical observational studies [9, 10] suggested that DM was likely to increased the risk of IPF, which inspired experiments related to antidiabetic treatment for IPF. Interestingly, using mice animal models, it was revealed that metformin could reverse established lung fibrosis [11,12,13,14,15], however, a post hoc analysis [16], which investigated the effect of combinations therapy in IPF (pirfenidone/pirfenidone + metformin), concluded that metformin had no effect on clinically relevant outcomes. Furthermore, another pooled analysis demonstrated that metformin might increase the risk of disease progression when in combination with proton pump inhibitors, angiotensin II receptor blockers or thyroid medications [17].

Due to the conflicting results in the existing studies, it remains controversial whether DM is truly correlated with IPF. Hence, in the present study, we conducted a meta-analysis and systematic review, aiming to assess the association between DM and the incidence of IPF.

Materials and methods

We performed meta-analysis and wrote this report referring to the Meta-analysis Of Observational Studies in Epidemiology (MOOSE) proposal [18] and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement [19].

Literature search

We searched databases including Medline, Cochrane Library, Embase, Web of Science and China National Knowledge Infrastructure. The following items were searched in databases as keywords or random words: “pulmonary fibrosis”, “diabetes”, “risk factors” and such searches were additionally filtered for articles published in any language leading up to September 30, 2020 (Complete search strategy presented in Appendix 1).

Inclusion criteria and exclusion criteria

Case–control studies or cohort studies were selected. The case groups were all diagnosed with IPF in accordance with clinical history, High-Resolution Computed Tomography (HRCT), and when available, lung biopsy. Also, a calculated measure of association between DM and IPF was required. Studies focusing on progression or prognosis of IPF and studies lacking general information about control groups were excluded.

Data extraction

Two researchers (L.B. and L.Z.) managed data extraction independently, reviewing the title, abstract, and full text of each article, and discussed or consulted a third researcher (T.P.) when disputations arose. The following are included: (1) basic information of each study including author, publication year, study design, etc.; (2) characteristics of case and control groups; (3) diagnostic methods of DM and IPF; (4) the number of diabetics in case and control groups; and (5) potential sources of biases.

Quality assessment

The Newcastle–Ottawa Scale (NOS) was used for quality assessment of included studies, covering three domains: selection of groups, comparability of groups and ascertainment of exposure [20]. The NOS score ranges from 0 to 9 stars and studies that receive 5 stars or more are regarded as high quality. We evaluated the diagnostic criteria of IPF and DM in each study for the possibility of selection bias. Cigarette smoking, age, gender, environmental exposure and genetic factors, which may induce IPF and bring about information bias, were deemed as covariates and all taken into consideration when estimating whether control subjects were adequately selected.

Statistical analysis

In our meta-analysis, odds ratio with 95% confidence interval (95% CI) was used as the effect measure. Heterogeneity was assessed by I2 statistics and random effect model was chosen when heterogeneity was significant (I2 > 50%), otherwise, fixed effect model was selected. Forest plots were used to display the results from individual studies and pooled estimates, and P < 0.05 were regarded as statistically significant. Trial Sequential Analysis (TSA) was used for estimate of evidence size and reliability of the conclusion [21, 22]. We also performed sensitivity and subgroup analyses to assess resources of heterogeneity. Meta-regression and Egger’s test [23] were utilized for bias analysis. Data analysis was performed using RevMan 5.3, Stata 12, and TSA 0.9 beta.

Results

Study selection and characteristics

As is briefly illustrated in Fig. 1, out of the 1528 articles reviewed, 9 studies [7,8,9,10, 24,25,26,27,28] from 5 countries finally met our eligibility. All studies were case–control and distinguished as high-quality by NOS assessment. General population was selected as control groups in six studies [7, 8, 10, 25,26,27], one study included only healthy volunteers [24], and the remaining two [9, 28] included patients with other chronic pulmonary diseases. IPF was diagnosed based on clinical history, HRCT, and lung biopsy while diagnosis of DM could be established with any objective method such as fasting blood glucose or simple by clinical symptoms combined with clinical history. More details are displayed in Table 1.

Fig. 1
figure 1

The flow diagram of study selection

Table 1 Characteristics of Included Studies

Meta-analysis

A total of 5096 IPF cases and 19,095 control subjects were involved in the analysis (Fig. 2), which suggested a significant association between DM and IPF (OR 1.65, 95% CI 1.30–2.10; P < 0.0001), based on statistical reliability verified by the subsequent trial sequential analysis (Fig. 3). The heterogeneity was significant (I2 = 68%) with no obvious sources of biases found among the sensitivity analyses (Fig. 4). Therefore, we performed subgroup analyses to investigate factors that possibly contribute (Table 2). The result remained consistent in separate analyses regardless of community or hospital controls, and irrespective of diagnostic criteria of IPF/DM or characteristics of control groups (healthy subjects, general population or patients with pulmonary disorders). When smoking status, age, and gender were accounted for in case and control groups, the heterogeneity still existed as before.

Fig. 2
figure 2

Forest plot of individual and pooled effect of odds ratio for DM in IPF. IPF idiopathic pulmonary fibrosis, DM diabetes mellitus, M–H Mantel–Haenszel

Fig. 3
figure 3

Trial sequential analysis for the 9 studies providing the calculation of correlativity between DM and IPF. The diversity-adjusted required information size (RIS) was 47,415 patients, with α = 5% (two-side), power of 80%, relative risk reduction of 35% and incidence of 3% in the control arm. The cumulative Z curve got across both the conventional boundary (green line) and the trial sequential monitoring boundary (blue line), indicating that the information size was sufficient and the outcome was reliable

Fig. 4
figure 4

Sensitivity analyses of the primary meta-analysis. Every odds ratio was located between 1.30 and 2.10 while none of 95% confidence intervals crossed the invalid line “1”, signifying that the result was stable

Table 2 Subgroup analysis

Bias analysis

All elements that could lead to IPF were considered as potential sources of biases. Firstly, cigarette smoking is a known risk factor for IPF [29] and in the five included studies [7, 24, 26,27,28], smokers or ex-smokers were much more in case groups than in controls. In the following subgroup analysis (Table 2), when we selected the other four studies [8,9,10, 25] in which the smoking habits between case and control groups were similar, the association remained statistically significant (OR 1.52, 95% CI 1.05–2.19; P = 0.02), which coincided with the outcome of the meta-regression (Fig. 5, P = 0.351), suggesting that smoking was unlikely to distort the final results. Next, considering that age and gender are related to IPF despite inexplicable reasons [4], seven studies [7, 8, 10, 24,25,26,27] in which these two factors were well-balanced were used to inform another analysis, yet again concluding that DM correlated with IPF (OR 1.43, 95% CI 1.16–1.76; P = 0.0007). Still, there remained two potential sources of biases that the included literatures did not sufficiently address (Table 3). In light of a multicenter case–control study [30], environmental exposure was likely responsible for the incidence of IPF, which has been gradually acknowledged in recent years. Though three studies [8, 9, 24] took this into consideration, unfortunately, only one study matched case and control groups. However, in that single study [9], DM was proven to be the most dangerous factor for IPF in the logistic regression model (OR 4.3, 95% CI 1.9–9.8). Genetic factor, which was considered as one of the covariates according to the guidelines by ATS/ERS [4], was not referred to in any study except one [24]. Additionally, case and control groups in this study were regrettably unmatched. Thus, the impact it had on final conclusions was difficult to evaluate. Publication bias existed (P = 0.016 < 0.05) judged by the Egger’s test (Fig. 6).

Fig. 5
figure 5

Meta-regression analysis presenting the impact of smoking on relevance between DM and IPF. The X-axis represents the ratio of proportion of smokers in IPF group to that in control group in each study (this was used to represent the influence of smoking in each study), while the Y-axis accordingly represents the ratio of the morbidity of DM in IPF group to that in control group (this was used to represent the association between DM and IPF). In the meta-regression analysis, no significant linear correlation was found (P = 0.351 > 0.05) between Y-axis and X-axis, which indicated that smoking didn’t affect the association between DM and IPF

Table 3 Potential sources of bias
Fig. 6
figure 6

Egger’s publication bias plot for included studies

Discussion

Main findings and clinical inspiration

In the present meta-analysis, it was revealed that the prevalence of diabetes was increased markedly in IPF cases compared with controls, which suggests that DM and IPF might be positively associated. However, we noticed that a latest study [31] came to the opposite conclusion. Since all of the included studies are retrospective case–control studies, which are easily affected by recall bias and additionally, the interpretation of the outcome is in the limitation of the significant heterogeneity, which could not be satisfactorily explained, we believe that our conclusion still need further evidence.

Interestingly, a recent review [32] clarified the common features between IPF and pulmonary complications in diabetics. These include clinical characteristics (significant reductions in FVC, FEV1 [33] and DLCO [34,35,36,37,38]), HRCT imaging (the frequently presented UIP pattern [39, 40]) and histopathological changes (thickening of the basal lamina of lung capillaries [41, 42], increased amount of collagen in the alveolar walls [43], etc.), all of which indicated that IPF and diabetes are closely related. This conclusion validated the findings of our meta-analysis as well as equipping them with biological plausibility. However, it is still unclear whether a causal relationship exists between DM and IPF.

Therefore, understanding the exact pathological mechanisms is crucial; namely, how persistent hyperglycemia, a known characteristic of diabetes, gradually contributes to the pulmonary lesions. Studies found that a high glucose concentration could result in nonenzymatic glycation with the ultimate formation of advanced glycation end products (AGEs), which may target type IV collagen in the alveolar basement membrane, thicken the basal lamina both in epithelial and capillary of alveoli and eventually lead to a decrease in pulmonary elasticity and compliance [44,45,46]. This hypothesis has become recognized as an explanation for the pathological abnormalities of interest, including injured pulmonary function in diabetic individuals. Furthermore, some investigators hold the view that oxidative stress (OS), which refers to an imbalance between free radicals and antioxidants in the body, is intimately connected with the onset of IPF. On one hand, OS can directly enhance nonenzymatic glycation [47], but on another, OS participates in the activation of nuclear factor-kappaB (NF-κB) [48], which is presumably the central part in initiating processes of alveolitis. One study [49] shows that inhibiting the activation of the transcription factor NF-κB could reduce lung injury and fibrosis. Hürdag et al. [50] discovered that OS could decrease superoxide dismutase (SOD), increase nitric oxide synthase (NOS), and contribute to overproduction of nitric oxide (NO) and peroxynitrite (ONOO), potentially giving rise to damage of lung tissue and ultimately pulmonary fibrosis [51, 52]. In addition, inflammatory cytokines play a crucial role, among which transforming growth factor-beta1 (TGF-β1) attracts the most attention. TGF-β1 was found overexpressed in hyperglycemia, which has been documented to promote proliferation and differentiation of fibroblasts, activation of myofibroblasts and deposition of extracellular matrix (ECM) [53,54,55,56,57], all of which will eventually bring about lung fibrosis. In recent years, the relationship between telomere length, DM and IPF has attracted attention. Elevated glucose and increased oxidative stress might interfere with telomerase function, thereby leading to shortened telomere length [58] and a mendelian randomisation study [59] inferred a causal link between shorter telomere length and higher risk of IPF.

Although the possible pathophysiologic mechanisms do explain the disease process, we acknowledge that multiple factors may induce pulmonary fibrosis and that it is also indeterminate to what extend IPF is affected by diabetes. Thus, to solidify the association between DM and IPF, the beneficial effect of antidiabetic therapy should be established. Rangarajan et al. [12] demonstrated that metformin could reverse lung fibrosis in a bleomycin model via AMPK activation, which is also the potential mechanism of metformin in diabetes [60]. The discovery indicated a certain connection between these two diseases and provided a potential evidence on possible benefit of anti-diabetics for the treatment of IPF. Nevertheless, successful therapies in animal models was not particularly efficacious in human studies and a post hoc analysis [16] showed no advantages of metformin when in combination therapy with pirfenidone. Possible explanations may include that AMPK activation is only relevant to certain IPF phenotypes, or insufficient drug concentrations in the lung [61]. Consequently, the effect of antidiabetic treatment in IPF remains uncomfirmed up to now. Since persistent hyperglycemia might participate in the occurrence of pulmonary fibrosis, perhaps future researches could investigate the suitable threshold of blood glucose for IPF and figure out whether timely and effective hypoglycemic therapy would prevent the incidence of the disease. We expect such findings could further strengthen the evidence linking IPF with DM.

Strengths and limitations

In our meta-analysis, we screened literatures in strict accordance with inclusion and exclusion criteria, designed the study with high quality, and finally demonstrated that DM and IPF are very likely interrelated. Nevertheless, there remain several limitations in our study. Firstly, the heterogeneity was significant, and a reasonable interpretation is still absent. The wide range of prevalence of DM (from 10 to 61% in IPF groups and from 3 to 43% in control groups) in the studies has attracted our attention, which could be responsible for major heterogeneity; we assumed this to be secondary to multiple factors including diagnostic criteria of diabetes (subjective or objective methods) and IPF (differences of diagnostic criteria in different guidelines), selection of populations (with or without underlying diseases), and regional differences. However, the internal causal relationship has not been established to date. Secondly, given that the result is based on case–control studies, which are susceptible to confounding factors, the potential sources of biases have always been a focus. Even though influence of smoking, age, and gender was ruled out in our bias analysis, other covariates such as genetic factor might also cloud the association between DM and IPF, especially considering that at least 30% of patients have predisposing genetic factors which could increase the risk of pulmonary fibrosis [62,63,64]. Besides, two studies [10, 27] from the UK are based on the THIN database (in Gribbin’s research, IPF cases are from the period 1991–2003 while in Dalleywat’s, cases are from 2000 to 2011), which may be one of the contributory reasons of the notable publication bias, and repeated cases might cause type I errors (false positive conclusions).

Conclusion

The association between diabetes and IPF was briefly referred to in ATS/ERS clinical practice guideline updated in 2011 [65], but not in the newest edition [4] perhaps owing to the contradiction of existing evidence. Although our study suggested that IPF and DM might be relevant, the causal relationship cannot be completely established up to now. Thus, more evidences are still required.