Metabolic and weight components’ impact on incident asthma using a real-world cohort

Introduction

Obesity impacts 40% of adult patients with asthma and 60% of patients with severe asthma in the United States (US).¹ The prevalence of asthma increases from 7.1% in lean adults to 11.1% in obese adults.² Estimates project that nearly half of US adults will be obese by 2030,³ with an expected 250,000 new asthma cases related to obesity per year.⁴ A meta-analysis of over 300,000 adults found a dose-response relationship between obesity and incident asthma yet failed to consider common comorbidities of obesity including metabolic syndrome (MetS), defined by having three of the following five conditions together: abdominal obesity, hypertriglyceridemia, low High Density Lipoprotein (HDL) cholesterol, high blood pressure or high fasting glucose.⁴ The increasing prevalence of obesity requires that we clarify the role of obesity and MetS in asthma incidence.

Individuals with asthma and comorbid obesity have greater morbidity related to disease, with a 4- to 6-fold greater risk of hospitalization, more frequent exacerbations, and increased symptoms compared to lean individuals with asthma.⁵ Obesity is associated with reduced efficacy of inhaled corticosteroids, the mainstay of asthma treatment.⁶ Metabolic dysregulation (MetD) is often employed in electronic health records (EHRs) as a surrogate for MetS, as these records may not contain the more precise ascertainment of individual metabolic components. For example, body mass index (BMI) is used in place of more traditional adiposity measures such as waist circumference or DEXA scans as these measurements are infrequently ascertained in clinical practice and therefore largely absent in EHRs.⁷ Similarly, diabetes replaces fasting glucose in MetD as it is often unclear whether the measured glucose in a patient record is truly fasted.⁸ Recent work highlighted the impact of MetD, particularly glucose homeostasis, insulin resistance, and lipid levels on asthma outcomes. Individuals with hemoglobin (Hb)A1c levels in the pre-diabetic range had higher rates of asthma exacerbations than those with HbA1c in the normoglycemic range.⁹ Insulin resistance mediates the association of Th1 polarization with pulmonary function, whereas BMI mediates the association of monocyte activation with pulmonary function.¹⁰ In adults with asthma, low levels of HDL was associated with lower lung function parameters.¹¹ Total cholesterol and non-HDL were inversely correlated with current asthma.¹² As such, there are multiple hypothesized mechanisms by which obesity and metabolism contribute to asthma: dietary and metabolic alterations in immune cell and airway function, obesity-induced changes in chest wall mechanics, and altered oxidative stress.¹³ Determining the contribution of obesity and individual MetD components to asthma incidence and subsequent asthma risk is imperative to develop targeted preventative measures for adult-onset asthma.

The existing epidemiologic studies on the contribution of obesity and individual MetD components to asthma incidence are conflicting.¹⁴ A Norwegian study of over 23,000 adults found that abdominal circumference and impaired glucose metabolism, but not other Met D components, were associated with incident asthma; weight or BMI were not included in the analysis.¹⁵ Conversely, a US cohort study of over 4,000 adults found that obesity, defined by BMI, in the third decade of life predicts asthma incidence in females, but not in males, after 25 years of follow up. The presence of MetD, captured in the third decade of life, in females did not associate with incident asthma after the model was adjusted for obesity (BMI). When stratified for individual MetD components, only abdominal adiposity associated with incident asthma in females.¹⁶ Changes to BMI and the presence of MetD components beyond the third decade of life were not evaluated. The relative contribution of obesity and features of MetD to asthma incidence in adulthood remains unclear. Identifying which clinical factors contribute to asthma incidence will provide targets for asthma prevention and insight into mechanisms of asthma inception in adulthood.

Methods

We performed a real-world landmark analysis of over 90,000 patients from the Vanderbilt University Medical Center (VUMC) synthetic derivative (SD), a de-identified copy of the Electronic Health Record (EHR).^17,18 Because the deidentification process includes date shifting as well as scrubbing of Health Insurance Portability and Accountability Act (HIPAA) related information, the SD resource is not a direct copy of the VUMC EHR and is considered non-human subjects data.^17,18 As such the SD is optimized for research and the longitudinal nature of the resource makes is ideal for assessing metabolic parameters. This unique study design allows for the evaluation of the independent influence of body weight and specific MetD components on asthma incidence. The study protocol was reviewed and received non-human subject designation from the VUMC Institutional Review Board.

Study setting.

We used a longitudinal, data rich retrospective Vanderbilt CardiOvascular and Multiple MetabOlic Disease in Obesity Resource (COMMODORE) cohort of patients derived from the VUMC SD (eFigure 1). The SD contains clinical data on approximately 3.2 million patients from VUMC, a tertiary care academic medical center in Nashville, TN, USA, and the affiliated health system outpatient clinics. These data include: basic demographics, laboratory values, patient medication data, clinical care notes, and Current Procedural Terminology (CPT) and International Classifiers of Disease (ICD v9 and 10) codes. We restricted our analyses to patients ≥ 18 years of age with comprehensive capture of cleaned anthropometric measures,¹⁹ metabolic variables including height, weight, HDL cholesterol, low-density lipoprotein (LDL) cholesterol, triglycerides (TGs), and cardiovascular phenotypes.^20,21 To minimize potential impacts associated with pregnancy, such as weight gain or gestational diabetes, data collected 9 months preceding or following a pregnancy-related ICD code or laboratory test are excluded, as described previously.²² Study specific variables were also included such as smoking history which was captured using Systematized Nomenclature of Medicine clinical terms (SNOMED CT) found within the smoking fields in the EHR. These data were extracted from the EHR in August 2019 by a programmer with extensive experience in EHR-data extraction and cleaning.

Study population and design.

A time-to-asthma diagnosis analysis was conducted after a predetermined three-year inclusion period (eFigure 2A). The study design used a three-year inclusion period, landmark, to identify patients who frequently use the VUMC healthcare system, e.g. medical home. The three-year inclusion period was identified based on visit density and was determined to adequately capture MetD components (diabetes, hypertension, high triglycerides, low HDL) and weight while also providing substantial follow-up during the time-to-event outcome period to minimize the impacts of patient loss due to lack of follow-up. Subjects were included if they had four measurements of height and weight collected over the three-year period with approximately one year (± six months) separating collection of these data points to generate a medical home cohort. The outcome period was defined as the remaining recorded time following the 3-year landmark period (eFigure 2A, 2B). The date of entry into the study (t₀) was defined as the date of first weight measurement. Weights closest to the 12-month interval were selected in the case of multiple values within the range and weights within 9-months of a potential pregnancy, identified by ICD or lab, were excluded.

Asthma cases were identified from the EHR using ICD-9/10 codes, starting with 493* or J45*. Subjects were required to have two or more asthma codes on two separate visits and one asthma medication prescription from eTable 1 within six months of an asthma code to be included. Two separate reviewers, one database expert and one clinical asthma expert, reviewed the same record set to verify cases. Indeterminate cases were reviewed as a group with the clinical senior author before inclusion or exclusion decisions. Performance of the algorithm and the inter-reviewer correlation were determined. Individuals with an asthma diagnosis before t₀ or during the landmark period or that met an exclusion criterion were excluded (eFigure 3). Exclusions represent disease states that can mimic asthma and lead to over/misdiagnosis.

The MetD components including diabetes, hypertension (HTN), HDL <40 mg/dL in males or <50 mg/dL in females, or TG >150 mg/dL were determined for all subjects during the landmark period. Diabetes and HTN were identified from the EHR using previously published phenotyping algorithms.²³ For laboratory values, outlier values beyond physiologic range, such as HDL >200 mg/dL or TG >1200 mg/dL were excluded. Each MetD component was categorized as being chronic (present at baseline, t₀), developed during the three-year landmark period (t₀ through t₃), or free from development throughout the landmark period. The risk of asthma development during the outcome period for each MetD component was evaluated based on chronic, developed, or free status. Due to variable impacts of gender on these components based, we performed secondary gender-stratified analyses.

Study end point.

The primary end point for this study was incident asthma during the outcome period.

Statistical methods.

Descriptive statistics for incident asthma versus those free from asthma were compared using Wilcoxon tests for continuous variables or Pearson’s chi-square tests for categorical variables. Cox regression analyses were conducted to examine how weight and MetD component impact the risk of developing asthma, and adjusted for age, gender, race, height, smoking status, and individual MetD components. To evaluate the impact of weight independent of MetD, a Cox model was used to estimate 10-year attributable asthma risk by weight at t₃ based on the population’s median individual, e.g. age 51.2 years, White race, and female sex, with no MetD components. Hazard ratios (HRs) and their 95% confidence intervals were plotted to visualize their effects on the risk of asthma. Burden analysis of the individual metabolic derangements (i.e., number of metabolic syndrome components at t₃) were analyzed using a Cox proportional hazard model and subsequent analysis of variance to calculate hazard ratios and 95% confidence intervals. In addition, interactions between weight and gender, and metabolic syndrome components and gender were tested using Cox regression analyses to examine whether gender modifies the effect of weight and metabolic syndrome components on the risk of developing asthma. P-values < 0.05 were considered nominally statistically significant, while p-values less than <0.003 achieved Bonferroni corrected significance. The Bonferroni correction is based on the number of variables assessed in the univariate analyses, e.g. 16, and is calculated by dividing the standard significance threshold, 0.05, by the number of tests, 16, resulting in a threshold of p=0.003. All tests two-tailed and statistical analyses were performed with R version 3.3.1.

Results

Population characteristics.

The sample studied consisted of 90,081 patients with available anthropometric measurements and follow-up after inclusion and exclusion criteria, including COPD, were met (eFigure 1). This medical home population is generally reflective of the larger VUMC population (eTable 2) except the longer average patient record length for included subjects [vs. 2.8 (0.5-7.0) years for the entire population]. The number of preexisting asthma cases (n=2,880) in our cohort results in a life-time prevalence of asthma of 0.3% and aligns with rates reported in similar adult cohorts in the southeastern US.¹⁶ The included population was predominantly White and female with a median outcome period of approximately eight years with high rates of developed hypertriglyceridemia, low HDL, HTN, and diabetes during the landmark period (Table 1, eFigure 2B).

Table 1:

Subject demographics and clinical characteristics stratified by the development of asthma during the outcome period.

	Asthma
	No (n=89,245)	Yes (n=836)	p-value
Minimum age, years	50.6 (36.5; 62.4)	42.2 (30.4; 53.3)	<0.001
Gender, female	53,866 (60.4)	633 (75.7)	<0.001
Race			<0.001
Black	9,060 (10.2)	121 (14.5)
White	77,026 (86.3)	688 (82.3)
Other	3,159 (3.5)	27 (3.2)
Height, cm	168.2 (162.6; 177.8)	166.5 (160.0; 174.0)	<0.001
Weight, kg
Weight t0, kg	80.3 (66.7; 95.7)	79.8 (66.7; 95.3)	0.948
Weight t1, kg	80.7 (67.1; 95.8)	80.7 (67.6; 95.7)	0.607
Weight t2, kg	81.1 (67.6; 96.2)	81.0 (67.6; 97.5)	0.528
Weight t3, kg	81.2 (67.7; 96.6)	81.7 (68.7; 98.3)	0.193
Weight change t3-t0, kg	0.9 (−2.8; 4.8)	1.5 (−2.0; 5.5)	<0.001
Record length*, years	7.5 (5.2; 10.7)	8.1 (4.1; 10.2)	<0.001
Years to outcome	----	5.7 (4.0; 8.7)
History of tobacco use, yes	14,495 (16.2)	63 (7.5)	<0.001
Metabolic dysregulation components
Diabetes mellitus			<0.001
Free	74,159 (83.1)	744 (89)
Chronic	4,792 (5.4)	30 (3.6)
Developed	10,294 (11.5)	62 (7.4)
Hypertension			<0.001
Free	45,854 (51.4)	508 (60.8)
Chronic	12,420 (13.9)	102 (12.2)
Developed	30,971 (34.7)	226 (27)
High-density lipoprotein†			0.398
Free	74,899 (83.9)	695 (83.1)
Chronic	2,489 (2.8)	32 (3.8)
Developed	11,857 (13.3)	109 (13.1)
Triglycerides‡			0.872
Free	76,224 (85.4)	709 (84.8)
Chronic	2,400 (2.7)	24 (2.9)
Developed	10,621 (11.9)	103 (12.3)

Data are presented as the median (lower; upper quartile) or number (%)

^*Median patient record length from t₃ to last International Classifiers of Disease Billing code date.

^†< 50 mg/dL in males and < 50 mg/dL in females

^‡> 150 mg/dL

Asthma outcome and univariate risk models.

836 (0.93%) patients developed asthma during the outcome period with the median time to event of 5.7 years. The asthma algorithm had > 90% positive and negative predictive values, sensitivity, and specificity. Patients who developed asthma were more frequently younger, female, Black, non-smokers, with a history of diabetes, without a history of HTN, and experienced more weight gain during the landmark period (Table 1). A Cox-proportional hazards model evaluated the impact of weight on asthma risk, independent of other MetD, by modeling the median individual of the population with no MetD components. This model demonstrates an increase in 10-year asthma risk beginning around 70.5 kg, which is consistent with a BMI of 25 kg/m² (Figure 1A). Based on this model, individuals with a BMI > 25 kg/m² have a 10-year attributable risk associated with obesity of 15.4% for an asthma diagnosis.

Figure 1. 10-year attributable asthma risk by weight at landmark t₃.

Attributable risk for (A) total population and (B,C) male/female gender adjusted to population medians (height-1.69m, age-51.9 years, female (total only), White race, and no smoking, hypertension, diabetes, high density lipoprotein <50 mg/dL, and triglycerides >150 mg/dL ≤ t₃).

Multivariable models for asthma risk.

In the multivariate Cox proportional hazard model for asthma risk, younger age, female gender, absence of smoking history, and the presence of diabetes were significantly associated with risk for incident asthma (Figure 2). When comparing the relative importance of these predictors, younger age, female gender, and presence of diabetes were the greatest drivers of risk for asthma. Diabetes present at t₀ was independently associated with increased risk of incident asthma (HR_adj = 1.85, 95% CI: 1.27 – 2.71, p=0.0002) after adjusting for all other covariates. No other MetD components were associated with an increased risk of incident asthma. There was not a significant association between increasing number of MetD components and increased risk of incident asthma (eTable 3). The incident asthma population had a lower smoking history (7.5%) compared to the asthma free (16.2%) population. However, 25.8% of the patients excluded from the study for having asthma before t₃ had a smoking history.

Figure 2. Independent effects on 10-year asthma risk.

Adjusted HRs calculated comparing third to first quartiles, i.e., age (63.4; 38.1 years), height (177.8; 162.6 cm), weight at t₀ (96.2; 67.1 kg), weight at t₁ (96.2; 67.7 kg), weight at t₂ (96.7; 67.9 kg), weight at t₃ (96.9; 68.0 kg).

Differences in asthma risk based on EHR-reported gender.

Gender is an independent predictor of incident asthma, and MetD presentation and risk vary by gender. We performed secondary gender-stratified analyses to understand risk considering the individual MetD components. Asthma was associated with younger age and absence of HTN in both women and men. Women with asthma were more frequently Black with greater weight gain, diabetes, and low HDL. Men with asthma more frequently had hypertriglyceridemia (eTable 4). In gender-stratified analyses, weight at t₃ in females was associated with increased risk for asthma development (Figure 1B). In the Cox model evaluating the impact of weight alone, weight in males did not impact incident asthma (Figure 1). Only younger age was a significant independent predictor of incident asthma in both men and women in multivariable models. While diabetes and smoking history were independent predictors in women, elevated TG was an independent predictor in men. Subsequent analysis indicated that metformin was prescribed more frequently in individuals with diabetes who did not develop asthma than those who did (43.4% vs 19%, p-value <0.0001, eTable 5).

Discussion

In this medical home cohort optimized for the longitudinal evaluation of MetD components, we found that a diabetes diagnosis and weight adjusted for height at the time of cohort entry were each independently associated with the increased risk of incident asthma. Studies have shown that modeling weight adjusted for height has less bias and improved accuracy for predicting disease risk compared to BMI.¹⁹ Unlike previous studies that used single time point exposures for MetD components, our study design allowed us to evaluate the impact of variable exposure length of each MetD component, e.g. prior to study inclusion at t₀ or developed during the three-year landmark period, and identify opportunities to reduce risk.

These findings advance the existing medical literature¹⁶ and further support clinically actionable intervention periods for asthma prevention in adulthood.^9,10,16,24 Consistent with the literature, our findings show that insulin resistance as manifested in diabetes and excess weight are key independent drivers of incident asthma risk in adulthood. The evaluation of variable exposure length for diabetes suggests reversing newly diagnosed diabetes or the prevention of diabetes among individuals at risk for diabetes may prevent adult incident asthma. The dose-response relationship between chronic and developed diabetes and risk of incident asthma suggests a latency period. Recognition, through HbA1c testing, and the early intervention for pre-diabetes with metformin or lifestyle strategies should be tested as adult asthma prevention strategies.

Our secondary gender stratified analyses affirm that MetD confers variable risk by gender. Weight in females was associated with increased risk for asthma development, whereas weight in males did not impact incident asthma. These findings are consistent with previous studies showing that obesity predisposes to asthma, particularly amongst women.² High TGs was a significant independent predictor of incident asthma in men. Intervention strategies that prevent weight gain or promote weight loss, particularly among women before their sixth decade of life, should be studied for the prevention of adult incident asthma.

Previous literature suggests potential medication effects on asthma exacerbation rates. We investigated medication exposures and glycemic control in patients with diabetes. Notably, metformin was prescribed more frequently and HbA1c values were lower in patients with diabetes who did not develop asthma than those who did. Our findings of increased metformin prescription in patients who do not develop asthma together with previous work that identified an association between metformin and reduced risk for incident asthma suggest a potential protective therapeutic effect.^25,26 Metformin, which improves insulin resistance, improves asthma control and was associated with fewer exacerbations and increased quality of life in patients with asthma.^9,27–29 Metformin use was also associated with reduced mortality from lower respiratory tract infections and reduced risk for incident asthma in patients with diabetes.^25,26,30 Randomized controlled trials are needed to establish whether metformin can prevent incident asthma in adulthood.

It is also possible that metformin usage characterizes a diabetes phenotype associated with better glycemic control, which may be inversely related to incident asthma risk. The difference in HbA1c values between those with diabetes at the start of the landmark period among those with and without incident asthma are supportive. Our data supports the treatment of diabetes, glycemic control, and clinical phenotype may impact the risk conferred by diabetes on adult incident asthma. Future studies are needed to clarify whether factors associated with glycemic control are more relevant to different asthma phenotypes.

Smokers experience poorer asthma control likely due to altered airway inflammation and corticosteroid insensitivity, and quitting smoking can improve symptoms and lung function.³¹ The association between cigarette use and new onset asthma is less clear.³² We observed that nonsmoker status was associated with asthma incidence in this landmark analysis. Our longitudinal study design, which excluded prevalent asthma prior to and during the landmark period, excluded many smokers with associated asthma. Our finding that significantly more patients excluded for having asthma before t₃ had a history of smoking (25%) compared to the asthma free (16.2%) or incident asthma (7.5%), suggests that the risk associated with smoking is conferred before study inclusion. Future work using a younger population would be required to assess this relationship. It is also important to note the smoking phenotype in the EHR may be impacted by incomplete capture of a subject’s smoking history including variables related to packs per day and length of tobacco use. While EHR-based smoking history is imprecise, any possible smoking history misclassification should be equally distributed throughout the population. Additionally, previous work in the VUMC EHR dataset has linked documented smoking to lung cancer outcomes suggesting that while these data are imperfect, they are generally consistent with expected exposures.³³

A strength of the COMMODORE dataset is the inclusion of clinically relevant anthropometric and metabolic variables. Confounders in metabolic data, like pregnancy, are reduced. The landmark analysis uniquely allows for the estimation of event risk probabilities in each group based on membership during the landmark time. We used this longitudinal study design to evaluate the impact of time-dependent development of metabolic abnormalities on the outcome of incident asthma, a novel contribution to the literature.

Although EHR-based landmark analyses such as this one can offer insight into the temporal association of MetD components with asthma incidence, the approach has intrinsic limitations. Asthma over diagnosis is a real-world issue impacting 30% of clinical asthma cases;³⁴ obesity does not increase the rate of asthma over diagnosis.³⁵ In our study design, one would anticipate that any over diagnosis would impact prevalent asthma cases excluded and incident asthma cases identified. Requirements for bronchodilator responsiveness or methacholine-induced airway hyperresponsiveness to establish an asthma diagnosis can lead to underdiagnosis if these tests are not repeated over time and with adequate medication tapers.³⁶ Further, objective measures of lung function are largely underutilized and not uniformly available with most of these measures requiring the patient to attend a specialty clinic where care access may be influenced by various social determinants of health.³⁷ Because of this, the requirement for lung function in asthma diagnosis remains an area of debate.^38,39 Previous work has demonstrated that ICD-based algorithms for asthma have similar sensitivity as the gold-standard HEDIS criteria but dramatically improved specificity.⁴⁰ As such we employed a similar ICD-based algorithm in our system to define our study cohort.

Direct measures of insulin resistance, visceral adiposity, waist circumference, and clean fasting glucose data are largely unavailable in EHR datasets. The classic definition of MetS could not be used; we used a diabetes diagnosis consistent with other studies.⁴¹ The study is limited to patients who received care at a tertiary care, academic medical center health system consistent with a medical home cohort. This potential selection bias reduced issues relating to loss to follow-up as most included patients demonstrate substantial follow-up both in terms of record length and visit frequency.

As with any landmark analysis, since separation of groups does not follow random allocation, it is possible that non-measured confounders bias the results of this study.⁴² Due to inherent data sparsity of the EHR, it is also possible that individuals that developed metabolic abnormalities during the three-year landmark period had these abnormalities previously. It is also possible the “chronic” status was diagnosed shortly before being included in the landmark period and was therefore not classified as “developed”. Additionally, there is risk of false positive outcomes given the multiple comparison performed in this analysis, however p-values less than <0.003 achieved Bonferroni corrected significance.

Conclusions

Asthma development in adulthood has few known modifiable risk factors. This real-world landmark analysis found that diabetes and weight, but not other MetD components, are potentially modifiable risk factors for asthma development in adulthood. Prospective controlled trials of diabetes prevention and weight management strategies to prevent adult incident asthma are warranted.

Abbreviations used:

MetD

metabolic dysregulation

MetS

metabolic syndrome

EHR

electronic health record

Hb

hemoglobin

BMI

body mass index

HDL

high density lipoprotein

VUMC

Vanderbilt University Medical Center

COMMODORE

Vanderbilt CardiOvascular and Multiple MetabOlic Disease in Obesity Resource

SD

synthetic derivative

CPT

Current Procedural Terminology

ICD

International Classification of Diseases

LDL

low density lipoprotein

TG

triglycerides

SNOMED CT

Systematized Nomenclature of Medicine clinical terms

HTN

hypertension

HR

hazard ratio