Respiratory outcomes of metformin use in patients with type 2 diabetes and chronic obstructive pulmonary disease

Few studies investigated the respiratory outcomes of metformin use in patients with coexistent type 2 diabetes mellitus (T2DM) and chronic obstructive pulmonary disease (COPD). We want to compare the long-term respiratory endpoints of metformin use and nonuse in patients with T2DM and COPD. This retrospective cohort study enrolled patients with T2DM and COPD from Taiwan’s National Health Insurance Program between January 1, 2000, and December 31, 2012. Main outcomes were hospitalized bacterial pneumonia, hospitalization for COPD, noninvasive positive pressure ventilation (NIPPV), invasive mechanical ventilation (IMV), and lung cancer. In total, 20,644 propensity score-matched metformin users and nonusers were assessed. The adjusted hazard ratios (95% confidence intervals) of metformin use relative to nonuse for bacterial pneumonia, hospitalization for COPD, NIPPV, IMV, and lung cancer were 1.17 (1.11–1.23), 1.34 (1.26–1.43), 0.99 (0.89–1.10), 1.10 (1.03–1.17), and 1.12 (0.96–1.30). Metformin use also exhibited significant dose–response relationship with respect to the risks of bacterial pneumonia, hospitalization for COPD and IMV. Consistent results were found in the sensitivity test. This nationwide cohort study demonstrated that in patients with T2DM and COPD, metformin use was associated with higher risks of pneumonia, hospitalization for COPD, and IMV. If patients with COPD use metformin, vigilance with regard to their pulmonary condition may be required.

Scientific RepoRtS | (2020) 10:10298 | https://doi.org/10.1038/s41598-020-67338-2 www.nature.com/scientificreports/ and prognosis of COPD through the consequences of hyperglycemia, including reduced respiratory function, chronic inflammation, and susceptibility to bacterial infection 9 . Effective treatment of diabetes may improve the prognosis of COPD. Because only few studies 10 have assessed which diabetes treatments are suitable for patients with COPD, most clinicians follow the T2DM guidelines to treat patients with COPD, with metformin typically adopted as the first-line treatment. Metformin, possibly through the activation of AMP-associated protein kinase (AMPK), can reduce the accumulation of advanced glycated end product, oxidative stress, systemic inflammation, and improve insulin resistance 11 . In addition, it can reduce airway inflammation 12 as well as improve respiratory muscle strength 13 and forced vital capacity 14 . One animal study revealed that metformin could reduce airway glucose permeability and limit hyperglycemia-induced bacterial growth 15 . Several studies have reported that the use of metformin in patients with T2DM and COPD can lower the risk of all-cause mortality [16][17][18] . Only one randomized clinical trial of metformin use studied nondiabetic patients with exacerbated COPD, and no significant difference in clinical outcomes was observed 19 ; but this study was a short-term (one month) trial with moderate number (n = 52) of patients. Because the evidence regarding metformin use in patients with COPD is so inconclusive, we conducted this retrospective cohort study to evaluate the long-term respiratory outcomes of metformin use in patients with T2DM and COPD.

Results
Participants. Of the 402,153 patients in the Longitudinal Cohort of Diabetes Patients (LHDB) who were newly diagnosed with T2DM and COPD between January 1, 2000, and December 31, 2012, 61,226 and 71,374 individuals were metformin users and nonusers, respectively. These users and nonusers constituted the cohorts of our study (Fig. 1). After propensity score matching to a 1:1 ratio, 20,644 metformin users and 20,644 nonusers were included in the outcome analysis. The two groups of patients were similar with respect to all covariates (Table 1). Considered jointly, the mean age of patients in these two groups was 63.9 years. The mean followup times for metformin users and nonusers were 5.01 years (standard deviation [SD] = 3.26) and 5.19 years (SD = 3.11), respectively.
The outcome of hospitalized bacterial pneumonia. During follow-up, 3,133 (15.18%) of the 20,644 metformin users and 3,106 (15.05%) of the 20,644 metformin nonusers were admitted for bacterial pneumonia. The incidence rates (IRs) were 37.8 and 34.5 per 1,000 person-years; the adjusted hazard ratio (aHR) of metformin use vs. nonuse was 1.17, with a 95% confidence interval (CI) of 1.11-1.23 and a p value < 0.0001 ( Table 2). The difference in the cumulative incidence of bacterial pneumonia between the metformin users and nonusers was illustrated using a Fine and Gray's sub-distribution hazard model ( Fig. 2A), which showed a higher risk of pneumonia in metformin users relative to nonusers. With respect to the risk of bacterial pneumonia, the aHRs for < 15, 15-29, and ≥ 30 cumulative defined daily dose (DDDs, relative to nonuse) were 1.09, 1.17, and 1.24, respectively (Fig. 3A); the aHRs of the prescribed daily doses of metformin, < 500, 500-749, and ≥ 750 mg/day (relative to nonuse), were 0.88, 1.17, and 1.20, respectively (Fig. 3B).
The outcomes of invasive and noninvasive ventilation. The IRs of noninvasive positive pressure ventilation (NIPPV) were 7.03 and 6.48 per 1,000 person-years for metformin users and nonusers, respectively; aHR was 0.99, with a 95% CI of 0.89-1.10 and p value of 0.95 ( Table 2). The IRs of IMV were 22.2 and 18.4 per 1,000 person-years for metformin users and nonusers, respectively; aHR was 1.10, with a 95% CI of 1.03-1.17 and p value of 0.001. The cumulative incidence of invasive mechanical ventilation (IMV, Fig. 2C) revealed that metformin users had a higher risk of IMV relative to nonusers. With respect to the risk of IMV, the aHRs of the cumulative DDDs of metformin use of < 15, 15-29, and ≥ 30 were 1.10, 1.07, and 1.13, respectively (Fig. 3A); the aHRs of the prescribed daily doses of < 500, 500-749, and ≥ 750 mg/day of metformin were 0.90, 1.12, and 0.96, respectively (Fig. 3B).
The outcome of lung cancer. The IRs of lung cancer were 3.6 and 3.0 per 1,000 person-years for metformin users and nonusers, respectively; the aHR was 1.12 (95% CI of 0.96-1.30), with p = 0.12 (

Discussion
Our study observed that metformin use in patients with T2DM and COPD increased the risks of bacterial pneumonia, hospitalization for COPD and use of IMV. The sensitivity test of metformin use for more than 6 months versus control also disclosed consistent results. The relative risks of the cumulative and prescribed daily dose of metformin seemed to have dose-response relationship. Patients with T2DM have an increased risk of respiratory infections which could lead to frequent exacerbations of COPD and worse outcomes 20 . Treatment with corticosteroids is also associated with a higher risk of pneumonia in patients with COPD 21 . An animal study suggested that the use of metformin could attenuate bacterial growth by reducing the glucose permeability of the airway 15 . However, our study observed that the use of metformin increased the risk of bacterial pneumonia. In this study, metformin use increased the risk of hospitalization for COPD, this severe exacerbation might increase the use of systemic steroids and add the hazard of bacterial pneumonia.     www.nature.com/scientificreports/ T2DM has been associated with progression and worse prognosis of COPD 9, 17 . Greater COPD exacerbation may result in greater airflow limitation, respiratory failure, or hospitalization. Three cohort studies have demonstrated that metformin use could reduce the risk of mortality [16][17][18] . Bishwakarma et al. 22 demonstrated that metformin use in patients with coexistent COPD and T2DM was associated with a lower risk of COPD-specific emergency room (ER) visits and hospitalizations, especially in patients with low-complexity COPD. In another study using randomized control of metformin use in nondiabetic patients with severe exacerbation of COPD, metformin use had no detectable effect on clinical outcomes or C-reactive protein 19 . Our study demonstrated that metformin users were associated with higher risk of the cumulative incidence of hospitalization for COPD. The Cox proportional hazards with a marginal model indicated that relative to nonuse, metformin use had an aHR of 1.34 (95% CI 1.26-1.43). The cumulative and prescribed daily doses of metformin also had dose-response trends with respect to the risk of hospitalization for COPD. The inconsistency among these 3 studies may be attributed to differences in sample sizes, patient ethnicities, and follow-up periods. Patients with acute exacerbated COPD often have low serum vitamin B12 levels 23 . Long-term metformin use was associated with lower serum vitamin B12 levels 24 , which may affect respiratory muscle function 25 , and have relationship with the exacerbation and hospitalization for COPD.
Oxygen therapy improves survival in patients with severe lung disease and hypoxemia. Mechanical ventilation is essential life-support for respiratory distress; the slow and sustained flow of air to distal airspaces minimizes airflow turbulence, reduces airway resistance, lowers the effort required for breathing, and relieve hypercapnic respiratory failure 26 . Sexton et al. 13 conducted a prospective observational study on 6-month metformin use in 17 participants with moderate to severe COPD and observed that metformin use was associated with reduced symptoms of dyspnea and improved health status. By contrast, our results indicated that metformin use increases the risk of IMV. This inconsistency is probably due to differences in research designs and sample sizes. Mitochondria account for most of the body's oxygen consumption in their production of adenosine triphosphate (ATP). Metformin can enter cells and inhibit complex I of the mitochondrial electron transport chain through organic cation transporter 1. This inhibition potentially reduces ATP production, which may lead to energy stress and potential mitochondrial dysfunction 27 . Metformin can diminish mitochondrial respiration in skeletal muscle 28 . Insufficient energy in respiratory muscle may affect pulmonary function, and the degree of mitochondrial dysfunction in the skeletal muscle was associated with the severity of diseases 29 . In addition, our study exhibited that metformin use increased the risks of bacterial pneumonia and hospitalization for COPD, both of which are prone to developing into hypercapnic respiratory failure and need the support of invasive mechanical ventilation.
Metformin has been reported have anticancer effects on many types of tumor. A meta-analysis of eight observational studies by Zhu et al. 30 suggested that metformin use can yield a significant 16% reduction in the risk of lung cancer. However, in another meta-analysis of 11 cohort and four case-control studies, Nie et al. 31 reported a null association between metformin use and lung cancer. Likewise, our study of metformin use in patients with T2DM and COPD revealed no significant association with lung cancer.
This study has some limitations. First, the national health insurance research database (NHIRD) does not provide data regarding family medical history, social economic status, education, body weight, smoking habits, alcohol-drinking habits, or physical activity, all of which may influence our investigated outcomes. Second, because NHIRD claims data do not include symptoms, signs, or the results of pulmonary functional tests, we could not calculate COPD severity scores. Thus, to balance the severity of COPD between metformin users and nonusers, we used clinical records (antibiotics use, oral corticosteroid use, hospitalization or ER visit) to assess the number of moderate and severe exacerbations of COPD. Third, much evidence favors positive roles for longacting β2 agonists (LABAs) and long-acting muscarinic antagonists (LAMAs) in reducing COPD exacerbations; most of our patients had moderate or severe exacerbations. However, inhaled LABA and LAMA use was not prevalent, which indicated that the treatment was suboptimal. We must educate our physicians, encourage them to adhere to the guideline of COPD treatment, to improve patient's care. Finally, indication bias might exist in the cohort study because patients and physicians may unconsciously select their preferred method of treatment. Therefore, we balanced multiple variables, such as basic demographics, comorbidity, severity of COPD and T2DM, and the use of various medications to minimize such bias.
In conclusion, our study observed that metformin use in patients with T2DM and COPD was associated with higher risks of bacterial pneumonia, hospitalization for COPD and use of IMV. However, because of some unmeasured or inevitable bias still exist in this cohort study, stringent prospective studies or randomized control clinical trials are warranted to verify our results.

Methods
Study design and participants. The NHIRD comprises health care data from approximately 99% of the population of Taiwan (approximately 23 million people) 32 . The NHIRD has encrypted data on date of birth; gender; area of residence; disease coding according to the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM); prescriptions; and clinical procedures. The LHDB is part of the NHIRD. The LHDB selected 120,000 newly diagnosed diabetes patients yearly from 1999 to 2012, and their medical records from 1996 to 2013 were collected.
We involved patients who had a record of T2DM and COPD in the LHDB between January 1, 2000 and December 31, 2012. Patients' study endpoints were withdrawal from the National Health Insurance program, occurrence of the outcome of interest, or December 31, 2013, whichever was earliest. The number of diagnoses of COPD (ICD-9-CM codes: 491, 492, and 496) was ≥ 2 for outpatients within 1 year or ≥ 1 diagnosis for hospitalization or ER visit. The criteria for the definition of COPD that use the ICD-9-CM in the NHIRD was validated in a previous study 33 . Moderate COPD exacerbation was indicated by prescription of antibiotics and/ or oral corticosteroid; severe COPD exacerbation was indicated by hospitalization or an ER visit 34 (Fig. 4). We identified potential explanatory variables, such as baseline characteristics, moderate or severe exacerbation of COPD, history of bacterial pneumonia, Charlson Comorbidity Index (CCI) 35 , Diabetes Complications Severity Index (DCSI) score 36 , and the use of specific classes of drugs (such as antidiabetics other than metformin, antihypertensive drugs, COPD drugs, statins, and aspirin). CCI and DCSI scores were calculated using patients' status within 1 year before the index date. Statistical analyses. To maximize comparability, a propensity score matching was used to balance user and nonuser groups with respect to known variables 37 . We determined propensity scores through nonparsimonious multivariable logistic regression, with receipt of metformin as the dependent variable. We incorporated 35 clinically relevant covariates into our analysis as independent variables (Table 1). We applied the nearestneighbors algorithm to match pairs, assuming that proportions of 0.995-1.0 were acceptable. For the main outcomes, we censored patients at either the time of events or end of the follow-up on December 31, 2013, whichever came earlier. Using the Fine and Gray's sub-distribution hazard model, we compared the cumulative incidence of events over time between metformin users and nonusers. We used a marginal structural Cox model, and robust sandwich standard error estimates to compare the outcomes while controlling for baseline covariates 38 . All analyses were done on the as-treated basis. We stopped following the metformin users when they discontinued metformin after the index date; on the contrary, the metformin non-users were censored when they started to use metformin after the index date. To assess dose effect, we analyzed the relative risks of bacterial pneumonia, hospitalization for COPD and IMV with regard to the cumulative DDD of metformin during the 90-day observational period (< 15, 15-29, and ≥ 30 DDD) and prescribed daily dose (< 500 mg, 500-749 mg, and ≥ 750 mg). DDD is a unit of measurement defined as the assumed average maintenance dose per day as the main indicated drug in an adult; the DDD is 2,000 mg for metformin. We report results as hazard ratios with 95% CI. A two-tailed p value less than 0.05 indicates significance. We used SAS statistical software (Version 9.4 for Windows; SAS Institute, Inc., Cary, NC, USA) for data analysis.