Asthma is a common chronic condition in childhood and is characterised by chronic airway inflammation which leads to symptoms of cough, wheeze and shortness of breath which relapse and remit over time.1 2 Asthma guidelines recommend that current asthma symptoms should guide decision-making for asthma treatment.3–8 Some guidelines also recommend longitudinal measurements of spirometry in children aged over 5 years should be made to help guide decision-making.4 5 7 8 Only two guidelines, now at least 10 years old, indicate how spirometry should be used to guide asthma treatment in children (ie, treatment should be increased when forced expiratory volume in 1 s (FEV1) is less than 80% of predicted)4 5 and one guideline suggests that a change between clinic visits of more than 12% in FEV1 is excessive7 but does not say if or how treatment should change.

It remains unclear what magnitude of change in spirometry (eg, FEV1) is indicative of a clinically significant change in children with asthma. A relative rise of more than 10%9 or 12%10 predicted FEV1 from baseline is considered a positive change after inhaling a bronchodilator within a single test occasion, but this magnitude of change is not necessarily generalisable to repeated FEV1 measurements made 3–4 months apart for a couple of reasons. First, the time interval between measurements is different (20 min compared with 3–4 months). Second, there is no evidence in children to say by how much FEV1 changes over 3–4 months; a 12% cut-off7 is supported by a small study of 47 adult patients where FEV1 was measured at weekly intervals over 9–10 weeks.11 One study reported that the coefficient of variation (CV) in paired measurements of FEV1 in children was 4.3% for paired values measured within a 2-hour assessment, rising to 8.3% for paired values measured over an interval of 1–4 weeks.12 A second study measured FEV1 twice daily over 2 weeks and concluded that variability of more than 11.8% in FEV1 was likely to be clinically relevant.13 Data on the variability of spirometry indices in children with well-controlled asthma over intervals of months will inform clinical practice and may assist asthma specialists who are known to be uncertain about the role of monitoring asthma by FEV1 measurements.14

The aim of this study was to determine the variability of FEV1 in children with controlled asthma by applying three different methods of expressing change in FEV1 over 3-month intervals. We also sought to determine the influence of regression to the mean15 for each of the three methods of expressing FEV1 change; regression to the mean is a short-term consequence of the normal variability of FEV1 which complicates the interpretation of longitudinal FEV1 measurements.

Data from children with a clinical diagnosis of asthma collected in five independent clinical trials were obtained for this secondary data analysis study.16–20 In each study, prebronchodilator FEV1 and asthma control status were measured at baseline and approximately 3-month intervals over the course of 616 or 12 months,17–20 resulting in up to five measurements from an individual (0, 3, 6, 9, 12 months). Data from up to four consecutive 3-month periods were analysed and expressed as variability (ie, difference from the mean), rather than reproducibility (ie, the extent to which a test result can be reproduced).

Fritsch et al 16 undertook a study of 47 children with asthma attending a hospital asthma clinic in Vienna, Austria and collected data (including FENO, asthma symptom score and history of recent exacerbations) at 6-week intervals over 6 months. Petsky et al 17 recruited 63 children from hospital clinics in Australia and Hong Kong, and data were collected on eight occasions over 12 months (1, 2, 3, 4, 6, 8, 10 and 12 months). Szefler et al 18 recruited 546 participants from the community in the USA and collected postrandomisation information over 46 weeks including at 3 months, 6 months, 8 months and 10 months. Peirsman et al 19 recruited 99 participants with persistent asthma attending hospital asthma clinics across Belgium and collected data at 3-month intervals over 12 months. Turner et al 20 recruited 509 children with asthma from the hospital across the UK and collected data every 3 months over a year. The treatment algorithms in FENO-guided and standard practice arms in each randomised controlled trial (RCT) were different to other RCTs and are summarised in online supplemental table 1.

FEV1 was measured using spirometry according to the American Thoracic Society/European Respiratory Society standards.21 Measured FEV1 values were standardised to per cent predicted (%FEV1) and z score (zFEV1) using the Global Lung Function Initiative reference equations.22 A conditional change score (Zc) in FEV1 was calculated using a formula derived from children without any respiratory condition.23 Zc adjusts for factors associated with FEV1 variability, that is, baseline FEV1, the child’s age and time interval between measurements. Additionally, in the absence of data to calculate Zc in children with asthma, Zc was calculated using data from these five populations. Data from each of the 10 pairwise comparisons of FEV1 measurements were included, that is, 0–3 months, 3–6 months, 6–9 months, 0–6 months, 0–9 months, 0–12 months, 3–9 months, 3–12 months, 6–12 months and 9–12 months. Asthma control was determined with either the Asthma Control Test (ACT, for children aged >12 years),24 Child ACT (CACT, for children aged 4–11 years)25 or a study-specific questionnaire in two studies.16 17 For ACT and CACT, control was defined by a score of >19. Visits with missing FEV1 values and individuals with only a single measurement were excluded from analyses.

The mean (with SD), median (with IQR) or proportion (expressed as a percentage) of baseline characteristics for the combined and individual populations were presented. The intraclass correlation coefficient (ICC), a measure of correlation between multiple measurements within the same individual, was calculated for %FEV1 using a linear mixed effects model with random slope and intercept. The ICC was calculated with and without accounting for the level of asthma control to determine the impact of control on variability (online supplemental table 2). Online supplemental table 3 presents the coefficients for deriving Zc from children with asthma and from a healthy population.23 Variability for each of the paired comparisons of FEV1 measurements and an average over the entire pairwise dataset was calculated for the following outcomes:

Within-subject CV of %FEV1, which was calculated as the within-subject SD divided by the individual’s mean.

Bland-Altman limits of agreement (LoAs) for absolute change in FEV1.

Bland-Altman LoAs for relative change in %FEV1.

Bland-Altman LoAs for absolute change in zFEV1 (relative change in z-scores was not calculated because z-scores are already standardised).

95% prediction limits (mean±1.96×SD) of the Zc derived from a population of children without a chronic respiratory condition.22

95% prediction limits (mean±1.96×SD) Zc derived from children with asthma within the present dataset (see online supplemental table 2 for details).

Online supplemental table 2 gives fuller detail of how the variables presented in this paper were derived. All analyses were carried out in R software ((https://www.R-project.org/)).

Data were available from 1264 individuals (table 1). There were differences between the cohorts whose data contributed to the overall dataset for age, ethnicity, treatment with long-acting beta-agonist or leukotriene receptor antagonist, proportion controlled at randomisation and median FeNO (table 1). There were also differences across the five populations for FEV1 z score at baseline and at 12 months but not on the intervening occasions (online supplemental table 4). There were no differences in FEV1 z score at any time between individuals in separate trial arms (online supplemental table 5).

Characteristics of children recruited in the five trials and whose data contributed to this study

After excluding data from individuals with only one FEV1 measurement or missing FEV1 measurements, and who were uncontrolled on one or more occasions when FEV1 was measured, there were 881 individuals with a total of 3338 individual measurements and 5184 pairs of measurements (ie, an individual with two measurements gives one pair of measurements, while an individual with three measurements gives three pairs of measurements, etc). Online supplemental figure 1 describes how data for the present analyses were identified from the original 1264 individuals. Of the 881 individuals whose data were analysed, there were 303 (34%) who contributed 5 paired FEV1 measurements with 242 (27%) contributing 4, 183 (21%) contributing three and 153 (17%) contributing 2 paired measurements. Younger children had fewer controlled visits than older children.

The ICC was 0.81, ie 81% of the variance in %FEV1 was explained by between-individual differences. The ICC was similar when adjusting for level of asthma control (ICC=0.80) and varied between 0.56 and 0.89 between the five populations contributing data (online supplemental table 4). As uncontrolled asthma did not account for the within-individual variability of %FEV1, all study visits were used to define the measures of variability.

The variability of FEV1 between measurements was consistent across all time intervals for each of the three outcomes (table 2). When the analysis was restricted to paired FEV1 measurements made over only a 3-month interval the results were comparable to paired FEV1 measurements made over 3, 6, 9 and 12 months intervals (table 2). Each unit change in FEV1 z score was equivalent to a Zc 1.45 and an absolute change in FEV1% of 11.6%. The LoAs for a change in %FEV1 were wide, typically between ±20% for absolute change and ±27% for relative change. Online supplemental figure 2 presents the Bland-Altman plots. The conditional change was similar to an absolute change in zFEV1; the limits for the conditional change score were wider. When the 687 episodes where asthma was uncontrolled were included in the analysis the variability for all comparisons became slightly wider; the CV for FEV1 was 6.0 (compared with 5.2) and LoAs for difference in FEV1% values typically ±21% for absolute change and ±28% for relative change (table 1). Online supplemental table 6 provides full details of the variability when all children were considered and online supplemental figure 3 shows how the 1166 individuals included in the analysis were identified. Conditional change, but not relative change in %FEV1, absolute change in %FEV1 or absolute change in zFEV1 mitigated the influence of baseline FEV1 on subsequent change in FEV1, also termed regression to the mean (figure 1).

Variability of FEV1 between each pair of measurements in children

Average change in FEV1 (variability) between two visits against baseline FEV1 using (A) relative change in %FEV1, (B) absolute change in %FEV1, (C) absolute change in zFEV1, (D) change score derived in health. In all cases, except change score, there is a regression to the mean observed. FEV1, forced expiratory volume in 1 s.

In a population of 881 children with controlled asthma symptoms, the LoAs for absolute %FEV1 were ±~20%, meaning that a child’s %FEV1 would need to rise or fall by at least 20% to be considered outside the limits of a statistically different change. A similar result was seen when children with uncontrolled asthma were included in the analysis. However, we note that relatively large distribution-based estimates of variability such as the LoAs do not necessarily imply that smaller changes are not clinically relevant.26 Assuming a ratio of one unit change in FEV1 z score was equivalent to a change in FEV1% of 11.6% and a Zc of 1.45, a significant equivalent change in FEV1 z score (±~1.7) and slightly smaller Zc (±~2.3) were also close to or outside the range which might be expected as within normal variation (±1.96). A benefit of using Zc over the change in FEV1 z score is that the former is influenced less by age and FEV1 at first measurement. Collectively, these results support the recommendation asthma treatment should primarily be guided by symptoms,3–8 and that change in FEV1 for children with asthma may be best expressed as Zc,9 and not as a change in %FEV1 or change in FEV1 z score. Research is now required to determine what change in Zc is most clinically relevant.

One finding from this study is that it may not be valid to extrapolate results from a bronchodilator response, where an FEV1% change of 10%9 or 12%10 is considered clinically relevant, to define a change in FEV1% over a 3-month interval in children. We demonstrate that an absolute change in FEV1% of up to 20, or a relative change in FEV1% of up to 27%, can occur in children whose asthma symptoms are controlled at both measurements. An additional finding of note was that coefficients for Zc in the present population were similar to those for children without asthma,23 suggesting that the variability between healthy and controlled asthma is similar. In contrast with standardisation of cross-sectional FEV1 measurements, where there is a large international reference population,22 there are relatively few studies describing longitudinal change in FEV1 within populations of children (with or without asthma). Standardisation of longitudinal FEV1 measurements with a large international reference would allow more meaningful interpretation of longitudinal results.

Our results are mostly consistent with the limited literature describing variation in longitudinal measurements of FEV1. In a study of 47 adults, Pennock et al 11 report the CV for absolute change in FEV1 was 13, and that a significant difference would be an absolute change of 20% (ie, CV×1.65). A study of 7885 children with no chronic respiratory condition23 also reported a CV of 5.2 for FEV1%, identical to the present study. A third study of 232 children aged 7-yeas reported a CV of 4.3% for paired FEV1 measurements within the same assessment and of 8.3% for paired FEV1 measurements made over an interval of 1–4 weeks12. The higher CV in the study by Strachan12 compared with the present study may be explained by the former study recruiting younger children who have higher variability in spirometry.

LoAs were reported as the main outcome, and an alternative could have been to report the less conservative variability range, which for FEV1 was ±8.6% (ie, CV×1.65); this change would notably be smaller that cut-offs currently considered to be clinically relevant.9 10 We use LoAs since this can be derived for all three outcomes (ie, change in FEV1%, change in FEV1 z score and Zc), and therefore, allows direct comparison. An analysis which related change in FEV1 to clinical outcomes such as loss of asthma control and asthma exacerbation would give useful answers to the question ‘what is a clinically significant change in FEV1’. An additional consideration for future research is that different cut-offs for change may be appropriate for different clinical settings and the pretest probability for example, urgent versus planned assessment.

The conditional change score has been advocated as a method to identify significant change in FEV19 and has been described in healthy children and those with cystic fibrosis.23 The correlation (r) between repeated FEV1 measurements and both time and age that were observed in health and stable cystic fibrosis was also observed here in children with controlled asthma (online supplemental table 1), supporting its use as a measure of change in children with asthma. While time did not significantly influence the variability of FEV1 in this study, the maximum time between measurements was 12 months, whereas the original change score23 was developed in data of up to 5 years between measurements, for which the influence of time may be more relevant.

Each of the different methods of expressing change in FEV1 has advantages and disadvantages. Per cent predicted has historically been used to interpret pulmonary function measurements and is familiar to clinicians, patients and families. However, absolute and relative changes in per cent predicted are prone to bias, especially in children with lower lung function; an absolute fall of 10% from a baseline of 100% may be less relevant than a similar fall from a baseline of 70%, and a fall of 100 mL from a baseline of 5 L may be considered measurement variability but is a 10% fall from a baseline of 1 L. Change in z score avoids the bias of using FEV1% but exhibits regression to the mean, as we have demonstrated. The conditional change score avoids both bias and regression to the mean, but is not easy to calculate, however manufacturers of pulmonary function devices could imbed this approach into their software and reports.

The study has notable limitations which should be considered when interpreting the data. First, our analysis used more than one pair of FEV1 measurements from the same individual, and the assumption that paired FEV1 measurements from the same individual measured over different periods were independent may not be valid. However, using only one paired set of FEV1 measurements per person would have greatly reduced the amount of data available. Second, we were not able to compare variation between younger and older age groups since the former was less likely to be controlled and thus more likely to be excluded from the analysis; if variation was greater for a younger age group this might be due to a smaller sample size and/or greater inherent variability compared with an older age group. Third, some children may have had an asthma exacerbation between episodes when FEV1 was measured, and this might have temporarily reduced FEV1 values; based on a prior analysis of this population27 we assumed that any change in FEV1 associated with an exacerbation will have resolved by the time FEV1 was measured. A fourth limitation of our work is that we did not adjust for multiple testing. Our findings are descriptive and adjustment for multiple testing is not possible, for example, adopting a lower p value. Multiple testing is a particularly relevant consideration for statistical inference, for example, finding false positive results, and our study does not describe significant associations. A further limitation of our analysis is that it has not considered the effect of treatment changing during the period of follow-up. We also assumed that each paired observation was independent for the purposes of these analyses, but it is possible that repeatability may be dependent on an individual (ie, some children may have more variable FEV measures for each of their pairwise comparisons compared with children whose FEV is more stable). Further, the variability pattern within an individual may be an important prognostic indicator.

In summary, our findings demonstrate relatively wide LoAs in paired FEV1 measurement over time intervals that are relevant to asthma care. The conditional z score for change mitigated bias introduced from baseline FEV1 and age and may be more appropriate than change in FEV1% or FEV1z for assessing change in FEV1 in children with asthma. These data support current guidelines which recommend that asthma treatment should primarily be guided by symptoms and not by change in spirometry.