METFORMIN USE IS ASSOCIATED WITH DECREASED ASTHMA EXACERBATIONS IN ADOLESCENTS AND YOUNG ADULTS

Abstract

RATIONALE:

Metformin is a commonly used anti-diabetes medication with suggested anti-inflammatory and anti-oxidative effects. Metformin use has been associated with lower risk of asthma exacerbations and hospitalizations in adults. Here, we aimed to evaluate how asthma exacerbation rates changed after adolescents and young adults were prescribed metformin, and to learn if those changes were related to metformin prescription adherence.

METHODS:

Using secondary data of patients between 12–20 years old with asthma diagnosis and a metformin prescription from the Arkansas All Payers Claim Database and Arkansas School BMI database, we estimated the change in annualized asthma exacerbation rates after metformin prescription. We also evaluated the association of prescription adherence to the changes in those rates using univariate and multivariate regression models.

RESULTS:

464 patients met inclusion criteria. Outpatient exacerbation rates decreased after metformin prescription (13.4% only before vs 7.8% only after, p=0.009), and the annualized rate decreased more after metformin prescription as adherence increased (rank r = −0.165, p<0.001). After adjusting for potential confounders – age, sex, BMI, and inhaled corticoid steroid use – the strength of the association was attenuated.

CONCLUSIONS:

Asthma exacerbation rates decreased after metformin prescription, but a larger sample of patients who have experienced exacerbations and including patients with asthma who have not been prescribed metformin is needed to better know whether these decreases are driven by metformin use.

1. INTRODUCTION

Asthma is the leading chronic disease in children affecting about 8.4% of all children in the United States.¹ Prevalence of overweight/obesity has dramatically increased in children and adults in the last several years. Previous studies have established a link between obesity and asthma. Obesity is a common comorbidity among children with asthma. Being overweight or obese is associated with a higher risk of asthma.² Moreover, people with obesity and asthma have more asthma-related symptoms, more frequent exacerbations, and lower quality of life.^3–5 Children with obesity and asthma also exhibit reduced response to asthma controller therapies, and thus new asthma therapies are needed for this population.

Metformin is the most commonly prescribed medication for the treatment of type 2 diabetes mellitus (T2DM) worldwide. It is also widely used off-label for weight management, pre-diabetes, non-alcoholic fatty liver disease (NAFLD) and polycystic ovarian syndrome (PCOS). In addition to anti-diabetic effects, metformin has anti-inflammatory effects through eotaxin and AMPK pathways resulting in a decrease in oxidative stress and chronic inflammation.^6,7 It also decreases inflammatory cytokines tumor necrosis factor-alpha (TNF-α), interleukin (IL)-6, and IL-1, and induces the production of anti-inflammatory cytokines IL-4 and IL-10.^8,9 Moreover, it may reduce eosinophilic airway and pulmonary inflammation in murine models of obesity.¹⁰ In clinical studies in adults, metformin has been associated with improved clinical outcomes of lower respiratory diseases such as chronic obstructive pulmonary disease (COPD) and asthma, as well as reduced asthma-related hospitalizations and asthma exacerbations, reduced oxygen needs, and hospitalizations.^11–13

It is plausible that metformin use may have a favorable effect on adolescents and young adults with asthma. However, to date there have been no studies of metformin and asthma clinical outcomes in the pediatric population. The objective of this study is to evaluate the change in asthma exacerbations after adolescent children and young adults with asthma are prescribed metformin, and whether those changes are associated with adherence to metformin prescriptions. We hypothesized that metformin use is associated with decreased asthma exacerbations after first metformin prescription and this association is stronger with better adherence.

2. METHODS

2.1. Data sources

We used the Arkansas All-Payer Claims Database (APCD) to capture demographic data, medical and pharmacy claims. The APCD contains information from nearly 50 commercial payers and also compiles data from Arkansas Medicaid and Arkansas Medicare beneficiaries. In total, over 80% of the insured Arkansas population (more than 2.5 million unique individuals) are represented in the APCD. Five years of data were compiled (2013-2018) at the time of this study, containing more than 2 billion records.

We captured body mass index (BMI) from the Arkansas School BMI database. Arkansas State launched Arkansas Act 1220 of 2003 to address childhood obesity in the state. One element of the Act requires public schools to annually collect height and weight measures on students at even grades. Thus, for students who progress through the grades as expected, BMI measurements are spaced 2 years apart. If a student repeated a grade, the BMI measurements could be as close as 1 year apart (repeating an even-numbered grade) or 3 years apart (repeating an odd-numbered grade). At the time of this study, a 15-year longitudinal data file was constructed (2003/04 - 2017/18).

2.2. Study population and design

Figure 1 shows the establishment of the study population. From the APCD database, we identified 41,818 patients ages ≥12 years and <21 years with asthma. We identified those having asthma using the International Classification of Diseases, 9th, and 10th, Revision, Clinical Modification (ICD-9-CM and ICD-10-CM) codes: patients with asthma had at least one inpatient or two outpatient diagnoses of asthma (ICD-9-CM: 493.x or ICD 10-CM: J45.x). From this initial set, we excluded all patients with the following diagnoses: bronchiectasis, cystic fibrosis, ciliary dyskinesia, childhood cancer, inflammatory bowel disease, bronchopulmonary dysplasia, cerebral palsy, and dysphagia/aspiration at any time. Among those remaining, we identified 583 patients with a metformin prescription. We excluded patients if all metformin prescription was before 12 years of age, if the patient didn’t have at least 3 months of data before the first metformin prescription, or if the first metformin prescription was after age 20. From the remaining 511 patients, we were able to merge 464 with the School BMI database, and those patients are included in our final analysis. The index date is the date of the first metformin prescription. The pre-index period is defined as the time from insurance enrollment date or Jan 1, 2013 (APCD start date) if enrollment date is before Jan 1, 2013, to the index date (not including). The post-index period was defined as the time from the index date (including) to insurance disenrollment date or June 30,2018 if patient has active enrollment. The baseline characteristics of the study cohort were captured 1 year before the index date. The study was approved by the University of Arkansas for Medical Sciences Institutional Review Board.

Figure 1.

Flow chart of the study population

2.3. Study outcomes

Asthma exacerbation outcomes.

Asthma exacerbations in the outpatient, emergency room (ER), and inpatient settings were captured in the APCD database. Outpatient exacerbations were defined as having an oral steroid prescription associated with a clinical encounter for asthma. ER exacerbations were defined as having an ER asthma exacerbation diagnosis code and an oral steroid prescription. Inpatient exacerbations were defined as a hospital admission with an asthma exacerbation diagnosis code or oral steroid prescription following a hospital admission. Annualized asthma exacerbation rates were calculated from the division of the number of exacerbations by the pre- or post-index period in years.

The utilization of Metformin is assessed by measuring its refill rate, which is determined by dividing the total sum of days’ supply by the pre- or post-index period in days.

2.4. Statistical analyses

We summarized patients’ characteristics with percentages for categorical variables and with means ± standard deviations (SD) or medians (interquartile ranges or IQRs) for continuous variables. We used both standard nonparametric methods and bootstrapped regressions to determine whether annualized exacerbation rates changed from the pre- to post-index period. We also used bootstrapped regressions to determine whether pre- to post-index changes were related to adherence to metformin prescription alone and after adjusting for these potential modifiers: age, sex, BMI z-score, and inhaled corticoid steroid use (yes/no). We report p-values for the standard nonparametric tests and bias-corrected accelerated bootstrap confidence intervals for the bootstrap regressions. Our significance level was 0.05.

We performed these analyses on all 464 patients (whole sample) and again on the 136 patients who experienced at least one exacerbation in the pre- and/or post-index periods (reduced sample). The analyses were performed in R version 4.2.2 with bootstrap results coming from the boot.ci() function within the boot package, version 1.3-28. The statistical code and data are available upon request.

3. RESULTS

3.1. Patient characteristics

A total of 464 patients with asthma diagnoses and a metformin prescription were identified in the defined age group in the APCD database after merging with the school BMI database and excluding those who met the exclusion criteria.

Table 1 shows the baseline characteristics of all patients in the cohort (n=464). The median (IQR) age of the patients was 16 (14-17) years. Overall, the patients were 25% male, 37% were Hispanic, 30% were white, and 20% were Black or African American. Regarding asthma and allergy-related characteristics, 17% were prescribed ICS, 31% was diagnosed with allergic rhinitis, and 5.2% were diagnosed with atopic dermatitis. Regarding the metabolic profile, BMI z-scores were 2.04 (1.19 - 2.53); about 19% had type 2 diabetes and again 19% had acanthosis nigricans;12% had hypertension; and very few had dyslipidemia or non-alcoholic fatty liver. Pre-index period was 31 (12-49) and post-index period was 29 (15-46) months.

Table 1.

Clinical characteristics of the study population

CHARACTERISTICS	N = 4641
Demographics
Age	16.0 (14-17)
Gender (Males)	115 (25%)
Ethnicity
Hispanic or Latino	172 (37%)
Not Hispanic or Latino	250 (54%)
Unknown	42 (9.1%)
Race
White	139 (30%)
Black or African American	94 (20%)
Asian	4 (0.9%)
Unknown	214 (46%)
Other	13 (2.8%)
Asthma and Allergy Profile
ICS Use	81 (17%)
Allergic rhinitis	143 (31%)
Atopic Dermatitis	24 (5.2%)
Metabolic Profile
BMI z-score	2.04 (1.19 - 2.53)
Type 2 Diabetes	87 (19%)
Acanthosis Nigricans	89 (19%)
Hypertension	56 (12%)
Non-Alcoholic Fatty Liver	10 (2.2%)
Dyslipidemia	30 (6.5%)

¹Median (IQR); n (%)

Among the 464 patients, 29.3% experienced exacerbations: 14.2% experienced exacerbations before their first metformin prescription (the index date), 10.3% after, and 4.7% both before and after. The remaining 70.7% experienced no exacerbations either before or after their index date.

3.2. Pre- to post-index changes in exacerbation rates

Table 2 provides the mean pre- and post-index date annualized exacerbation rates for outpatient, emergency room, and inpatient events, as well as the sum of these (i.e., overall); the mean change from pre- to post-index date is also provided for each type of event, with negative values indicating a decrease in the post-index period.

Table 2.

Annualized exacerbation rates^a for each type of exacerbation during the pre- and post-index periods, along with the within-individual change in those rates (Pre – Post index) for both the whole (n = 464) and reduced sample (n = 136).

Sample	Outcome	Outpatient	Emergency room	Inpatient	Overall
Whole	Pre-index (mean ± SD)	10.8 ± 37.7	7.6 ± 74.2	0.1 ± 3.1	18.2 ± 83.9
	Post-index (mean ± SD)	6.7 ± 28.2	5.8 ± 37	0.4 ± 4.7	12.7 ± 50
	Pre – Post index (mean ± SD)	−3.8 ± 38.5	−1.8 ± 51.9	0.2 ± 3.3	−5.5 ± 65.2
	Pre – Post index (95% CI b)	−3.8 (−7.5, −0.3)	−1.8 (−10.0, 1.4)	0.2 (0.1, 0.9)	−5.5 (−13.1, −0.8)
Reduced	Pre-index (mean ± SD)	36.8 ± 62.4	26 ± 135.8	0.5 ± 5.8	62.2 ± 146.3
	Post-index (mean ± SD)	23 ± 48.6	19.8 ± 66.6	1.3 ± 8.7	43.4 ± 85
	Pre – Post index (mean ± SD)	−12.8 ± 70.5	−6.2 ± 96.1	0.8 ± 6	−18.8 ± 119.6
	Pre – Post index (95% CI b)	−12.8 (−24.7, −1.1)	−6.2 (−34.6, 4.4)	0.8 (0.2, 3.0)	−18.8 (−43.5, −1.7)

^aRates were multiplied by 100 to reduce leading 0s.

^bBootstrapped confidence intervals not containing 0 indicate the change in rates were significantly different from 0 at the 0.05 level.

For outpatient exacerbation events, 13.4% of the 464 patients experienced exacerbations only in the pre-index period and 7.8% only in the post-index period (McNemar’s p=0.009). Outpatient rates were significantly lower in the post-index period; with this evidenced in the whole and reduced samples (Wilcoxon signed rank p=0.023; Table 2). Regressing the rate changes on prescription adherence estimated non-significant downtrend for both samples in the simple linear regressions and multiple linear regressions (Table 3). However, the rank correlation, which can detect monotonic decreasing/increasing relationships rather than strictly linear ones, revealed the post-index decreases were more pronounced for those who had higher adherence for both the whole (r = −0.165, p < 0.001) and reduced samples (r = −0.228, p = 0.009). After adjusting the rate changes for the potential confounders, though, these rank correlations with adherence were no longer significant: r = −0.029 (p = 0.550) for the whole sample and r = −0.146 (p = 0.102) for the reduced sample.

Table 3.

Estimated effects^a (and 95% confidence intervals^b) of metformin adherence on each type of annualized exacerbation rate for both the whole and reduced sample. The simple linear model had only adherence as the explanatory variable; the multiple linear model included the linear effects of age, sex, BMI z-score, and the use of inhaled corticoid steroids.

Sample	Model	Outpatient	Emergency room	Inpatient	Overall
Whole	Simple linear	−2.1 (−4.9, 1.5)	−1.3 (−8.2, 1.5)	−0.1 (−0.2, −0.0)	−3.8 (−9.9, 0.2)
	Multiple linear	−2.1 (−4.8, 0.9)	−0.2 (−4.3, 2.8)	−0.1 (−0.4, −0.0)	−2.7 (−7.5, 1.4)
Reduced	Simple linear	−6.4 (−14.8, 4.6)	−4.2 (−26.1, 4.8)	−0.2 (−0.8, 0)	−11.7 (−30.4, 1.3)
	Multiple linear	−5.9 (−13.4, 3.1)	−0.7 (−14.7, 9.1)	−0.4 (−1.4, −0.0)	−7.5 (−21.5, 5.4)

^aEffects are expressed as the expected difference in rate changes between those who had (essentially) no adherence and those who had 100% adherence; these differences were multiplied by 100 to reduce leading 0s.

^bBootstrapped confidence intervals not containing 0 indicate that adherence was significantly related to the change in rates at the 0.05 level.

For emergency room exacerbation events, 4.3% of the 464 patients experienced exacerbations only in the pre-index period and 5.6% only in the post-index period (McNemar’s p=0.376). Even though 6 more patients had emergency room exacerbations in the post- compared to the pre-index period, the annualized rates were estimated to be lower in the post-index period, but not significantly lower: this was for both the whole and reduced samples (Wilcoxon signed rank p = 0.508; Table 2). Both the linear and multiple linear regressions estimated steeper decreases in post-index rates with higher prescription adherence in both samples, but none of these results was statistically significant (Table 3). Simple rank correlations gave some indication that post-index decreases were more pronounced with higher adherence for both the whole (r = −0.103, p = 0.029) and reduced samples (r = −0.172, p = 0.051). After adjusting the rate changes for the potential confounders, these rank correlations with adherence were no longer significant: r = 0.079 (p = 0.106) for the whole sample and r = 0.052 (p = 0.564) for the reduced sample.

For inpatient exacerbation events, only 4 of the 464 patients experienced exacerbations: one with events in the pre- and post-index periods, and the other 3 all in the post-index period. With so few patients experiencing inpatient events, the standard nonparametric tests (McNemar’s, Wilcoxon signed-rank, rank correlation) found no statistical evidence that metformin changed annualized rates from pre-index levels. The bootstrap results seemed to indicate otherwise (Tables 2 and 3), but since none of the 4 patients experienced an inpatient event only in the pre-index period, the bootstrap could not extend inferences beyond those patients.

For all exacerbation events (inpatient, ER, and outpatient), 14.2% of the 464 patients experienced exacerbations only in the pre-index period and 10.3% only in the post-index period (McNemar’s p=0.092). Overall rates were estimated to be lower in the post-index period with mixed statistical evidence in both the whole and reduced samples: the Wilcoxon signed rank test had p=0.076 but the bootstrapped confidence intervals excluded 0 by a small margin (Table 2). Regressing the overall rate changes on prescription adherence estimated steeper decreases in post-index rates as prescription adherence increased, but not significantly so: this for both samples and both linear regressions. Rank correlations of adherence with the overall rate changes indicated improvement in rates from pre-index periods with increased adherence: r = −0.174 (p<0.001) for the whole sample and r = −0.196 (p=0.023) for the reduced sample. However, after adjusting the overall rate changes for the potential confounders, these correlations were attenuated to 0: r = −0.021 for the whole sample and r = −0.097 for the reduced (both p>0.274).

4. DISCUSSION

In this study, we explored if metformin has any favorable association with asthma exacerbations in adolescent children and young adults with asthma. This study is novel as it is the first to examine this effect in a large pediatric population. Our results suggests that outpatient and, consequently, overall exacerbations decrease after metformin prescription and that better adherence correlates with more improvement. After adjusting for potential confounders, estimated improvements (whether slopes in the regressions or rank correlations) were attenuated, but in the same direction. We note that the post-index rates improved over pre-index rates as both age and ICS use increased (results not shown). We observed weaker evidence of improvement in emergency room exacerbations and no significant evidence for inpatient exacerbations which could be due to rarity of the events.

Our findings are consistent with studies in adults with asthma. Li et al.¹¹ conducted a retrospective cohort study using the Taiwan National Health Insurance Research Database in adults. They reported a lower risk of asthma-related hospitalization and asthma exacerbations in patients with asthma and diabetes. Moreover, Mendy et al.¹⁴ analyzed data from National Health and Nutrition Examination Survey to examine the association of metformin use with the risk of mortality from chronic lower respiratory diseases (CLRD), including asthma and chronic obstructive pulmonary disease (COPD). Using Cox proportional regression analysis, they reported a decreased risk of CRLD mortality with metformin prescription. Similarly, Wu et al.¹⁵ showed that metformin use was associated with fewer exacerbations and improved quality of life in adult patients with asthma and COPD overlap.

The beneficial effects of metformin in patients with asthma are conceivable due to its more than one favorable biologic mechanism. Metformin is shown to suppress eosinophilic inflammation in murine model of chronic asthma and reduce peribronchial fibrosis, smooth muscle layer thickness, and mucin secretion in bleomycin-induced acute lung injury mouse model, suggesting anti-inflammatory and anti-tissue remodeling properties.¹⁶ Another study showed significant reduction in tissue eosinophil infiltration through inhibition of the TNF-α-induced inflammatory signaling and NF-κB-mediated iNOS expression in lung tissue of obese mice.¹⁰ Metformin is also shown to reduce the rate of cytoskeleton of airway smooth muscle (ASM) cells and inhibit ASM cell proliferation.¹⁷ Moreover, metformin prevented vagally induced airway hyperreactivity in male rats on high-fat diet, in addition to inhibiting weight gain, fat gain, and increased insulin.¹⁸

Metformin’s anti-inflammatory and favorable effects on the respiratory system are partially attributed to weight loss as weight loss is shown to result in improved small airway function, decreased systemic inflammation, and the number of mast cells in the airways.¹⁸ In clinical practice, metformin is used for weight loss as an off-label medication. Our study didn’t have the capability to evaluate the weight/BMI change after metformin use.

Asthma remission is common in adolescence, reported up to 65% in a longitudinal study.¹⁹ We speculate that time can be a confounder in reduced annual exacerbation rate in our study. In our study, we observed signals of greater improvement with better metformin adherence, suggesting that metformin potentially plays a role in reducing exacerbations.

As strengths of this study, we accounted for the within-individual correlation between pre- and post-index periods by taking the difference in the two rates. We also treated the primary data (changes in exacerbation rates and prescription adherence) as they were – continuous – by using regression, as opposed to dichotomizing these on arguably arbitrary cut-points. Because the residuals from the regressions were not normal (they had very long tails), we did not rely on normal-based inferences from the regressions, but rather used bootstrap methods to estimate confidence intervals. Finally, we used standard nonparametric methods to further support our conclusions.

There are limitations to this study. Regarding the effective sample size: though we had data from 464 patients, 70% of these provide no information on whether metformin changes, for better or worse, exacerbation rates. Our effective sample size for all events was thus 136, and this number decreased as we examined each type of exacerbation rates (outpatient, emergency room, and inpatient). The closest BMI measurement to index date is chosen for patient characteristics so it doesn’t precisely reflect the BMI at the initiation of metformin. We were unable to assess other relevant confounders’ effects on asthma exacerbations such as socioeconomic status, smoking exposure, allergens, infections, air pollution, dietary habits, and changes in BMI. Also, the actual medication adherence is unknown because a record of prescriptions filled might not mean a prescription is taken by the patient. Finally, the information regarding the severity of asthma and chronic symptoms was not available in the claim database.

In conclusion, metformin use is associated with improved asthma exacerbations among adolescents and young adults with asthma. In order to delve deeper into the link between metformin use and asthma exacerbations, it is necessary to conduct research with a larger sample size of patients who experienced exacerbations and to compare these to a control group of patients with asthma who have not been prescribed metformin.