Enhancing the prediction of childhood asthma remission: integrating clinical factors with microRNAs

Capsule Summary:

The novel integration of baseline clinical and microRNA variables significantly improves the long-term individualized prediction of childhood asthma remission by early adulthood compared to using clinical variables alone.

To the Editor:

Approximately 15–65% of childhood asthmatics undergo remission of disease by early adulthood^1,2. Predicting the remission of childhood asthma is important for providing clinical risk stratification and anticipatory guidance to patients and their families. Previously, in the Childhood Asthma Management Program (CAMP), we used clinical variables to predict the remission of childhood asthma by early adulthood with high accuracy (area under the receiver operating characteristic curve (AUROC) 0.81) and found baseline forced expiratory volume in one second to forced vital capacity ratio (FEV₁/FVC) to be the greatest predictor of asthma remission². In addition, we previously used microRNA (miR, miRNA) networks to predict the remission of airway hyperresponsiveness in CAMP³. miRNAs are small non-coding single-stranded RNAs with excellent non-invasive biomarker potential. In this study, we integrate clinical with miRNA variables to increase the prognostic accuracy of predicting long-term childhood asthma remission.

CAMP was a double-blind placebo-controlled randomized trial of inhaled anti-inflammatory treatments for mild-to-moderate persistent asthma in children in North America⁴. From 1993 to 1995, 1041 children between the ages of 5–12 years were randomized to receive inhaled budesonide, nedocromil, or placebo for a mean 4.3 years, with an additional 13 years of observational follow-up. The clinical remission of childhood asthma was assessed at the first study visit on or after age 18 years². Clinical asthma remission was defined as: 1) no evidence of airflow obstruction with a FEV₁/FVC ≥ 80%, 2) no asthma exacerbations in the prior year, 3) no use of asthma medications in the prior year, and 4) no reported asthma symptoms in the prior year. Parental informed consent with subject assent was obtained, and the subjects were re-consented for each CAMP continuation study. This study was approved by each CAMP study center institutional review board.

miRNA-sequencing data in CAMP was obtained at enrollment prior to randomization in 491 subjects using published methods; 265 miRNA passed quality control⁵. R (Vienna, Austria) was used for all statistical analyses. The miRNA data were log₂-transformed and quantile normalized using the R preprocessCore package. From the 265 miRNAs that passed quality control, univariable logistic regression was first performed to select potential baseline miRNA predictors of asthma remission (Table I) that underwent further multivariable variable selection described below.

Table I.

Univariable and multivariable associations between microRNAs and asthma remission¹

microRNA	Univariable OR (95% CI)	P-value	Multivariable OR2 (95% CI)	P-value
miR-221-5p	0.82 (0.71 – 0.94)	0.004	0.61 (0.45 – 0.81)	< 0.001
miR-139-3p	0.84 (0.73 – 0.97)	0.02	-	-
miR-96-5p	1.17 (1.02 – 1.35)	0.03	1.24 (0.94 – 1.69)	0.14
miR-664a-5p	0.82 (0.69 – 0.98)	0.03	-	-
miR-199b-5p	1.45 (1.03 – 2.04)	0.03	1.63 (0.90 – 3.09)	0.11
miR-151b	1.29 (1.02 – 1.65)	0.04	-	-
miR-1307-3p	0.86 (0.74 – 1.00)	0.04	1.37 (0.96 – 2.00)	0.09
miR-148a-5p	0.77 (0.59 – 1.00)	0.0498	0.67 (0.44 – 1.00)	0.06

¹Odds ratios are for each two-fold increase in miRNA counts

²microRNAs in the multivariable model were selected using stepwise logistic regression

Out of 1041 children, 416 children had both baseline miRNA data and measurement of clinical asthma remission at early adulthood (range 18 to 23 years). Potential baseline clinical predictors of asthma remission were selected by performing univariable analyses on 23 clinical variables that had published associations with asthma susceptibility and/or remission⁶. The selection of these variables have been detailed in a prior publication². T-tests were used to compare continuous parametric variables, Wilcoxon rank-sum tests for continuous non-parametric variables, and χ² and Fisher’s exact tests for categorical variables. A missing data analysis was performed using the same 23 clinical variables. The variables associated with loss to follow-up of the asthma remission outcome (i.e., age, race, clinic site, income, exacerbations after exercise, exposure to tobacco smoke) and the differential acquisition of baseline miRNA data (i.e., race, treatment group, clinic site) were controlled for in the subsequent multivariable logistic regression model².

The clinical and miRNA variables statistically significant on univariable analyses were entered into a stepwise logistic regression model. In addition to the statistically significant variables from the missing data analysis, sex and miRNA batch effect were forced into the model. Continuous variables with a non-linear relationship with the log odds of remission were log transformed. The analysis was restricted to complete cases. Model selection was determined by Akaike information criterion (AIC), and Wald tests were used to evaluate individual variables. The receiver operating characteristic curve and comparisons of the AUROC were obtained using the R pROC and plotROC packages. Model fit and significance were evaluated using 10-fold cross-validation from the R caret package and permutation testing with 100,000 permutations.

Of 416 subjects, 59.6% were males and 40.4% females; 77.2% were White, 19.7% were Black, and 3.1% were Hispanic. The mean age of asthma onset was 3.9 (± 2.4) years, at enrollment in CAMP was 8.7 (± 2.1) years, and at assessment of asthma remission was 19.4 (± 1.6) years; 24.8% of subjects underwent asthma remission by early adulthood. On univariable analyses, baseline percent predicted FEV₁, FEV₁/FVC, the provocative methacholine concentration causing a 20% decrease in FEV₁ (PC₂₀), serum IgE level, serum eosinophil count, and pet ownership were associated with asthma remission status similar to our prior report (Supplemental Table I)². In addition, eight baseline miRNAs were associated with asthma remission on univariable analyses (Table I).

In the final model, baseline FEV₁/FVC and miR-221–5p were independent predictors of asthma remission by early adulthood. Each 10% increase in baseline FEV₁/FVC was associated with an 8.47 times odds of asthma remission (95% CI 4.02 – 20.80, P = 2.8 × 10⁻⁷), and each two-fold increase in baseline miR-221–5p expression was associated with a 39% decrease in the odds of asthma remission (OR 0.61, 95% CI 0.45 – 0.81, P = 9.6 × 10⁻⁴). The final model controlled for age, sex, race, treatment group, clinic, income, exposure to tobacco smoke, exercise induced bronchospasm, pet ownership, batch effect, miR-96–5p, miR-199b-5p, miR-1307–3p, and miR-148–5p. Five of the eight miRNAs that were significant on univariable analysis were included in the final model. Four of these miRNAs were included as covariables, and only one miRNA (miR-221–5p) was statistically significant on multivariable analysis (Table I). The AUROC of the final clinical and miRNA asthma remission prediction model was 0.90 (95% CI 0.85 – 0.95) (Figure 1), which is significantly improvement from the clinical prediction model without miRNAs (P = 0.001). On sensitivity analyses, removal of the four miRNA covariables from the model did not significantly alter predictive performance (AUROC 0.89, 95% CI 0.83 – 0.94, P = 0.20), whereas a model composed of only FEV₁/FVC and miR-221–5p had a lower, but still remarkable, predictive performance superior to using clinical factors alone (AUROC 0.85, 95% 0.79 – 0.91, P = 0.008). On cross-validation, the model was an excellent fit for the data (AUROC 0.82) and had statistically significant discriminative power (P < 0.001) on permutation testing.

Figure 1.

Receiver operating characteristic curve of the clinical and miRNA asthma remission prediction model

Overall, integrating baseline clinical factors with miRNAs resulted in improved model fit (AIC 194.1) and prediction performance (AUROC 0.90, P = 0.001) of childhood asthma remission compared to using clinical variables alone (AIC 752.1; AUROC 0.81)². Less severe baseline airway obstruction (i.e., higher FEV₁/FVC) and lower baseline miR-221–5p expression were independent long-term predictors of asthma remission by early adulthood. These results are consistent with our prior findings on the association of lung function trajectories, in particular FEV₁/FVC, with asthma remission². Predictive biomarkers may directly or indirectly be associated with disease pathophysiology, and experimental evidence supports our finding of the association between baseline miR-221 overexpression and long-term asthma persistence over a decade later. miR-221 has been shown to regulate the proliferation of and release of interleukin (IL)-6 from airway smooth muscle cells, promote IgE-mediated mast cell degranulation, stimulate IL-4 secretion from mast cells, and increase airway eosinophilic inflammation in asthma^7,8. Hence, the early induction of pro-inflammatory cytokines and type 2 immunity in the airway of asthmatics by miR-221 may promote long-term airway inflammation and remodeling leading to the persistence of asthma. The four miRNA covariables in the model have also been associated with immune regulation in the lung, and further investigation into their associations with asthma is needed⁹.

CAMP was a multi-ethnic population of mild-to-moderate persistent asthmatics from North America with detailed, long-term clinical follow-up to ensure the accurate classification of remission status. The strengths of this study include the mitigation of information bias through the control of variables associated with loss to follow-up and differential miRNA data acquisition and the evaluation of model significance and fit using permutation testing and cross-validation. Future external validation of the model is needed, but few similarly detailed childhood asthma populations with miRNA data and long-term clinical and lung function measurements exist for replication. Although CAMP was multi-ethnic, the majority of subjects were White and only mild-to-moderate childhood asthmatics were included, which may limit the generalizability of our results to other races or severe childhood asthma. This study provides cross-sectional baseline clinical and miRNA variables that can be obtained at a physician’s visit in childhood for the individualized long-term prediction of asthma over a decade later in adulthood, and our results demonstrate how the novel integration of clinical with miRNA variables can significantly improve prediction performance.