We conducted a prospective diagnostic accuracy study.

Patients were recruited in primary care clinics and in the paediatric pulmonology unit of a tertiary care hospital. The inclusion criteria were: (a) age 6–14 years, (b) clinical diagnosis of asthma (typical episodic symptoms, reversible with asthma medication, with no symptoms, signs or history suggestive of a different diagnosis), (c) history of asthma symptoms or asthma treatment in the past 12 months. We included patients with asthma of any severity. We excluded children with relative contraindications for spirometry.17

The initial visit (V0) involved collection of data on demographic characteristics, recent history of exacerbations, visits and hospitalization due to asthma, asthma treatment step based on the definitions of the Global Initiative for Asthma,18 performance of lung function tests and assessment of asthma control. The second visit (V1), 12 weeks later, entailed documentation of the course of the disease between V0 and V1 (need of on-demand primary care or emergency department [OD/ED] visit, exacerbation requiring the use of systemic corticosteroids [SAE], hospital admission due to asthma) and any changes in the step of asthma treatment between V0 and V1, in addition to repeated assessment of asthma control. The clinical data were obtained through interviews with the primary caregivers of the patients and the review of health records. Patients underwent lung function studies in 7 consecutive months (October-April) in the course of dedicated visits unrelated to their clinical management, and the paediatricians in charge of the patients were not informed of the results until patients had attended V1.

Participants received instructions to refrain from using any asthma medication in the 18 h preceding lung function testing, and the tests were always performed in the same order8: RO → spirometry → bronchodilator (BD) responsiveness test. High-efficiency viral and bacterial filters were used during testing, and the equipment was calibrated daily.

Respiratory oscillometry tests were performed with a Tremoflo C-100 system (Thorasys Thoracic Medical Systems, Montreal, Canada) in adherence to the standards established by the ERS.8 The criteria applied to consider a test valid were the acquisition of at least 3 acceptable replicates with a coefficient of variation of the resistance at 5 Hz (R5) of 15% or less. For the analysis, we used the mean values for R5, the difference between the resistance at 5 and 19 Hz (R5−19) and the area under the curve of the reactance (AX), expressed as z-scores calculated with the reference equations published by Ducharme et al.19 Higher resistance and reactance area values are indicative of greater impairment of lung function, and were considered abnormal when their respective z scores were greater than 1.645.20

Spirometry tests were performed with a Pneumotrac spirometer with Spirotrac-5 software (Vitalograph Ltd, Buckingham, United Kingdom) according to the recommendations of the ATS/ERS.17 We collected data on the following variables: FVC, forced expiratory in the first second of expiration (FEV1), the FEV1/FVC ratio and the FEF25-75. Tests were considered valid if 2 or more acceptable manoeuvres were achieved with a difference of 0.15 L or less between the largest FEV1 and FVC values.17 We expressed the results as z scores applying the reference equations of the Global Lung Initiative,21 and considered values abnormal when the z score was less than −1.645.20

The BD responsiveness test consisted in the administration of 4 doses of salbutamol with an inhaler (100 μg/puff), through an OptiChamber Diamond chamber (Philips Respironics, Eindhoven, Netherlands), followed by repetition of the spirometry test 20 min later. We expressed changes in FEV1 and FVC as the percent change relative to the individual’s predicted value, and considered an increase greater than 10% as a significant BD response, in adherence with the most recent recommendations of the ERS/ATS.20

We classified patients based on the lung function phenotypes proposed by Sorkness et al.15:

• 1)

  AT: FVC z score < −1.645 or change in FVC with BD > 10%.

• 2)

  AFL: FEV1/FVC z score < −1.645 and not meeting criteria for AT.

• 3)

  Normal: not meeting criteria for neither AT nor AFL phenotypes.

Asthma control was evaluated by means of the CAN questionnaire (carer version),22 which assesses asthma control in the past 4 weeks through 7 items rated on a Likert scale ranging from 0 to 4 points. The total score in the questionnaire ranges from 0 to 36 points, and lower scores indicate better control of asthma. A CAN score of 8 or greater is indicative of poor asthma control. In order to collect data with the same instrument through the entire age range under study, we did not use the child versions of the CAN.

The primary endpoint of the study was the development of SAEs between V0 and V1. The secondary outcomes included the need for OD/ED visits between V0 and V1 and poor asthma control at the time of V1.

We summarised the data as mean and standard deviation (SD) or median and interquartile range (IQR) as applicable. We compared categorical variables by means of the χ2 test and quantitative variables by means of the Kruskal–Wallis test.

We assessed the predictive power of RO and spirometry parameters by means of receiver operating characteristic (ROC) curves, measuring the area under the curve (AUC) with the corresponding 95% confidence interval (CI). We calculated the sensitivity, specificity, positive likelihood ratio (LR+) and negative likelihood ratio (LR−) in relation to the primary endpoint (SAE) for RO and spirometry parameters with a lower bound of the 95% CI of the AUC greater than 0.5 and for the AFL and AT phenotypes.

To control for the effect of potential confounders on the predictive power of RO and spirometry parameters and the AL/AT phenotypes, we calculated adjusted odds ratios (aORs) for logistic regression models in which we included the following adjustment variables: age, sex, maintenance treatment at the time of V0, change in dose between V0 and V1 and poor asthma control in V0. We selected these adjustment variables to replicate the analysis conducted by Fielding et al.7

We considered results with an α of less than 0.05, an AUC whose 95% CI excluded 0.5 and LR+/LR− whose 95% CIs excluded 1 as statistically significant.

We calculated that we needed a sample of 139 patients to estimate sensitivity and specificity values of 90% with a precision of 5% and a 95% level of confidence. Since we expected that valid lung function tests would not be achieved for every participant and that some would be lost to follow-up, we increased the calculated sample size by 10%.

The study was approved by the Clinical Research Ethics Committee of the coordinating hospital. The parents of the participants received information regarding the study in writing and signed an informed consent form.

Of all the RO parameters, the only one that was associated with an increased risk of SAE in the medium term was the AX z-score, but it had a very low predictive value and was inferior compared to spirometry. The AT phenotype had a high specificity and its detection was associated with an increased risk of SAE in the medium term, but due to its low sensitivity, the possibility of SAE cannot be ruled out when this phenotype is not present.

According to a recent meta-analysis,12 RO could be useful in identifying children at risk of asthma exacerbations. However, the evidence included in this review came primarily from cross-sectional studies,23–26 and to date, only 3 prospective studies have been published with data on the association of RO findings and the future risk of poor asthma control in children.27–29 These studies were conducted in small samples (54–72 patients) of school-aged children and adolescents27 or pre-school children28,29 and studied the association of RO findings with worsening asthma control 8–12 weeks after27,29 or the risk of moderate (but not severe) exacerbations in the following 12 months.28 They found an association between high R5, AX and R5−20 values and worsening asthma control27,29 and between high R5 values (unexpectedly, along with a decreased R5−20 values) and the risk of exacerbations in the 12 following months.28 Although one study included some RO results standardised according to previously published reference values,28 all used absolute R5, AX and R5−20 values. This is a significant source of confusion in the analysis of these studies, as both the RO parameters and the risk of poor outcomes were inversely associated with age. Our findings in relation to RO parameters would also have been more salient if we had used absolute values, especially for R5 and R5−19, but they would have certainly been biased. The only RO parameter associated with the risk of SAE was the AX z-score, but it had a low predictive value.

As regards spirometry, an individual patient data meta-analysis that included 7 trials in children and adolescents7 also found that the FEV1 in isolation could not predict the development of SAEs in the next 3 months, while the FEV1/FVC and the change in FEV1 had some predictive value. The review did not consider the FEF25-75 nor the response to BDs in the analysis. However, in our study the FEV1/FVC was only slightly superior to the FEV1 in predicting SAEs, and the FEF25-75 performed better than either of those two parameters.

The usefulness of FEF25-75 in clinical practice was questioned in an influential article30 that reported a very low probability (<3%) of finding abnormal FEF25-75 values in children and adults in spirometry tests with normal FEV1/FVC and FVC values, and that when it happened, it was usually due to poor technique (submaximal inhalation or FVC). Other studies in children31,32 have confirmed these findings. As the authors acknowledged,31 these studies are limited due to their retrospective design and not analysing correlations with the clinical manifestations. In our sample, 4.9% had a low FEF25-75 combined with normal FEV1/FVC and FVC values. When we reviewed the results of these patients, we did not find any indication of submaximal inhalation (all achieved manoeuvres of the maximum quality, with reproducibility of the FEV1 and FVC values within ≤ 0.10 L, better than required). However, none of these patients had severe exacerbations, and therefore they were all “false positives” and did not contribute to the predictive value of the FEF25-75.

On the other hand, we did not find the reduction in the FEF25-75 to be a phenomenon specific to the AT phenotype (small airway disease). A low FEF25-75 was much more frequent in patients with the AFL versus the AT phenotype. This was consistent with the findings of Sorkness et al.,15 whose study did not find an association between the FEF25-75 and the AT phenotype. Thus, if the predictive power of the FEF25-75 is greater compared to other parameters it is because it decreases both when there is air trapping and when there is only airflow limitation, and therefore is not specific for small airway dysfunction.

We used the definition of AT given by Sorkness et al.,15 but there were some differences in the AT phenotype in their study compared to ours: it was more frequent (14%), it was usually associated with a reduced FEV1/FVC (62%) and several cases were classified as AT only on account of the post-BD change in FVC when they had a baseline FVC in the normal range. However, there was excellent agreement between the studies in the age at onset of asthma in each phenotype, which was lower in the AT phenotype. Sorkness et al. found an association between AT and the frequency of exacerbations in the past 12 months. The yield of these phenotypes had not been studied in a prospective study before. We found that the AT phenotype seemed to be a specific marker of the risk of SAE. Although its low sensitivity limits its usefulness in clinical practice, the detection of the AT phenotype in a correctly performed spirometry test in a child with asthma is indicative of an increased risk of SAE in the medium term.

(1) Although the sample was larger compared to other prospective studies that have investigated the performance of RO,27–29 its size was not sufficient to make precise estimations for the parameters under study. Other factor that contributed to this was the low incidence of exacerbations that resulted from the low burden of asthma due to the COVID-19 pandemic.33 (2) The study was designed to assess the individual yield of RO and spirometry, and we did not study their usefulness in multivariate models for the prediction of exacerbations. (3) We did not include the response to BDs in the analysis of RO, as its association with the different lung function phenotypes in asthma patients has not been established, and there are still controversial aspects in relation to its application, such as which measure is best for the purpose of analysing post-BD changes (absolute value, % of baseline value, change in z-score), or the possibility of paradoxical responses due to its dependence on volume (with apparent worsening in children with the AT phenotype that improves with BD administration).8 (4) There is no evidence on the temporal trends of the AT/AFL phenotypes in individuals with asthma, and these phenotypes could change over time, so our findings can only be extrapolated to the 12-week interval that follows the identification of either of these phenotypes.