Abstract
Background
Respiratory oscillometry can inform early-life lung function, but evidence supporting its use in Guatemala is limited. We assessed the validity and reliability of respiratory oscillometry in 3-year-old Guatemalan children and whether these improve with quality control (QC).
Methods
During home visits, we measured respiratory oscillometry resistance (R) and reactance (X) at 7 Hz (R7 and X7), the frequency dependence of resistance (R7–R19) and the area of reactance (AX) in 666 children participating in the Household Air Pollution Intervention Network trial. Standard measurement QC removed artefacts using commercial software and selected the first three measurements with an R7 coefficient of variation of ≤15%. Alternative investigators’ methods varied in QC procedures. We evaluated validity as the proportion of variability in tests predicted by a child's height (R2) and, in a subset of 50 children, as agreement with test results after manual QC by a paediatric pulmonologist. Reliability was assessed by the intraclass correlation coefficient (ICC) of repeated tests.
Results
We collected oscillometry results from 537 children. Height explained 5–15% of the variability in R7, X7 and AX, but none in R7–R19. The correlation between standard and manually cleaned tests ranged from 0.66 to 0.97, depending on the parameter. Of the first tests, between-day ICCs using the standard method for R7, R7–R19, X7 and AX were 0.68, 0.45, 0.60 and 0.71, respectively, and the best between-day ICCs using an alternative method were 0.75, 0.57, 0.69 and 0.74, respectively.
Discussion
We demonstrated good validity and reliability of respiratory oscillometry measurements collected at home from preschool-aged Guatemalan children. Investigators’ automated QC methodology performed equally to or better than standard methodology.
Shareable abstract
Oscillometry can be measured in rural Guatemala in 3-year-old children with moderate to high validity and reliability. Investigators’ quality control methods offer flexibility, transparency and alignment with the technical guidelines. https://bit.ly/4rBqWpH
Introduction
Chronic lung diseases are leading causes of death and disability globally. In 2021, in low- and middle-income countries, chronic lung diseases were responsible for 4–7% of deaths and 2–4% of disability-adjusted life-years [1]. According to the developmental origins of health and disease theory, chronic lung diseases may originate in utero or during childhood when the rapidly developing and growing lungs are exposed to environmental hazards [2, 3]. The study of risk factors for early-life onset of chronic diseases requires reliable age-appropriate lung function tests that can be performed in settings where vulnerable populations live.
Respiratory oscillometry is an alternative to spirometry, considered suitable for children because it is performed during tidal breathing [4, 5]. Oscillometry measures the impedance of the respiratory system by superimposing external oscillating pressures of 4–50 Hz on breathing. The returning pressure and flow waves are used to measure the impedance (Z) of the respiratory system, composed of resistance (R) and reactance (X), measured as pressure divided by flow.
R reflects frictional pressure losses as air travels through the airways. It is interpreted as the internal airway diameter (calibre), with higher values of R associated with narrowed airways [6, 7]. Lower-frequency oscillatory signals (e.g. R at 7 Hz (R7)) theoretically travel to greater depths within the respiratory system than higher frequencies, which are thought to primarily reflect the large airways (e.g. R at 19 Hz (R19)). Therefore, the difference between R7 and R19 (R7–R19), named the frequency dependence of resistance, is believed to represent the smaller airways [8]. X evaluated at low frequencies (e.g. X at 7 Hz (X7)) measures the ability of the respiratory system to store energy. It represents the elastic properties or compliance of the lungs expressed in negative values, with less compliance associated with lower X. Because capacitive energy is stored in the small airways, it is considered that X7 represents the small airways [8]. X evaluated at high frequencies measures the inertial forces of the air moving through the airways, or the inertance, of the respiratory system. It is expressed in positive values. The frequency at which the elastance and inertance forces are equal in magnitude is called the resonant frequency (Fres). The area of reactance (AX) is the area inscribed by the X curve between the lowest measured frequency and the Fres.
Oscillometry measurement requires children to breathe quietly through a mouthpiece. Perturbations to the airflow, by obstructing the flow or losing the seal around the mouthpiece, result in artefacts. The reference standard for selecting artefact-free breaths is manual revision by a pulmonary expert. Still, experts are not available in all settings, and manual revision is a resource-intensive task. Alternatively, commercial software for removing artefacts exists, but it is proprietary and does not follow all the guidelines from the European Respiratory Society (ERS) for quality control [6]. Other automated methods, such as those based on the shape of the breaths, are not easily accessible [9–12]. Novel approaches are needed to optimise oscillometry measurement in children and to improve access to pulmonary function testing in low- and middle-income countries.
We were interested in investigating the effect of early-life exposure to household air pollution on childhood lung function in the Household Air Pollution Intervention Network (HAPIN) trial. In this trial, 800 pregnant women in Guatemala were randomised to receive a cooking stove intervention with liquefied petroleum gas or to continue cooking as usual with biomass fuels until the child was aged 1 year [13–16]. However, the ability to measure potential subclinical effects can be affected by loss of power from measurement errors caused by artefacts. The challenges of using oscillometry in epidemiological studies and the need to report oscillometry procedures have been noted before [17]. To better understand the extent of measurement error in our setting, we aimed to estimate the validity and reliability of oscillometry measurements in 3-year-old children in rural Guatemala. Furthermore, we explored whether applying an investigator-coded quality control methodology enhances validity and reliability, compared with commercial methods.
Methods
Population and study design
We recruited 3-year-old Guatemalan children who participated in the HAPIN randomised controlled trial [13–16]. The study was approved by the institutional review boards at Emory University in Atlanta, Georgia, USA (00089799) and Universidad del Valle de Guatemala in Guatemala City, Guatemala (146-08-2016). Children were visited at home on two consecutive days during the early study period (February 2022 to May 2022, n=96) or on a single day during the late study period (May 2022 to July 2023, n=588). We assessed for acute infectious symptoms and performed two oscillometry tests at each household visit.
Lung function measurement
We measured R7, X7, R7–19 and AX with the Tremoflo C-100 Airwave Oscillometry System (THORASYS Thoracic Medical Systems Inc., Montreal, QC, Canada) at a multifrequency range of 7–41 Hz. For each test, we collected repeated measurements within minutes, stopping when we achieved an R7 coefficient of variation (CoV) of <15% with at least four measurements or a maximum of eight measurements.
To perform the measurements, we placed a mouthpiece in the child's mouth and instructed them to form a seal around it with their teeth and lips to avoid air leaks, to put their tongue underneath the mouthpiece and to avoid swallowing. We blocked the child's nose with a clip or, when the clip was not tolerated, with the child's or caretaker's fingers. We started each measurement when the child was breathing regularly. During measurements, children sat in a chair or on the caretaker's lap. They sat in an upright posture with the chin angled upwards 15° while the operator supported the child's cheeks. A team of two operators, including one nurse, conducted the measurements. The equipment was calibrated following the manufacturer's instructions before each testing session. Measurements were 20 s in duration.
For each 20-s measurement, the oscillometer recorded, at 0.004-s intervals, the volume, flow and pressure associated with the participant's respiratory cycle. The oscillometer software uses the flow and pressure signals to calculate Z, R and X every 0.1 s for each frequency and to derive the child's respiratory frequency. The 0.1-s data points were averaged to obtain measurement summaries, which were averaged to obtain test results (figure 1). Since two tests were collected each day, we evaluated the reliability and validity of using a single test (i.e. the first test of the day) or the average of the two tests.
FIGURE 1.
Schema of respiratory oscillometry test summary computation from pressure, flow and volume signals. In this example, tests were calculated from the average of three measures. Each vertical line in the time series represents the start of a breath.
R7–R19 was calculated at the measurement level by subtracting R19 from R7. AX was calculated at the measurement level as the area above the linear step function starting at X7 and moving across each prime frequency until the Fres was reached. Fres is the frequency at which X crosses 0. If X did not cross 0, the linear step function was extended to 41 Hz.
Quality control methods
Quality control was conducted in two steps. First, in the 0.1-s time-resolved series, measurement periods with artefacts were filtered out. Second, we selected measurements to average. At each step, we compared alternative criteria for inclusion.
The alternative methods for step 1 consisted of the use of all the 0.1-s data points in a measurement (Unfiltered), artefact removal with a commercial proprietary method integrated into the Tremoflo software (p/n 101653) (Commercial) or artefact removal (AR) with one of three distinct methods developed by our team and implemented in R (AR1–AR3). A measurement was considered valid when ≥70% of the 0.1-s data points for the Commercial method, or ≥4 valid breaths for AR1–AR3, remained after artefact removal. Table 1 describes the alternatives and compares them with the technical standards of the ERS [6].
TABLE 1.
Comparison of alternative algorithms for artefact removal (AR)
| Characteristic | Commercial | AR1 | AR2 | AR3 | Manual |
|---|---|---|---|---|---|
| Type of algorithm | Proprietary commercial algorithm | Investigators’ algorithm coded in R | Investigators’ algorithm coded in R | Investigators’ algorithm coded in R | Visual inspection by a paediatric pulmonologist |
| Inclusion of an indicator variable of filtering status | No | Yes | Yes | Yes | Yes |
|
Criteria for identification of irregular breaths# ERS guidelines:
|
None | Incomplete breaths at the start or end of measurements | Incomplete breaths at the start or end of measurements | Incomplete breaths at the start or end of measurements |
|
|
Criteria for identification of leaks ERS guidelines: sudden large decreases in |Z|. Changes in flow time and volume time may be subtle |
None | None | R7 or X7 remained between −1 and 1 for >15% of the breath | R7 or X7 remained between −1 and 1 for >15% of the breath | Too much short-term variation around the trend line of the respiratory flow is associated with a drop of R7 and/or a spike of X7 |
|
Criteria for identification of obstructions ERS guidelines:
|
R outside of 3 standard deviations from the mean | R7, X7 or Z7 outside 3 standard deviations from the measurement mean, iterated three times [25] | R7, X7 or Z7 outside 3 standard deviations from the measurement mean, iterated three times [25] | R7, X7 or Z7 outside 3 standard deviations from the test mean, iterated three times [25] | Visual identification of outlier values in R7, X7 or Z7 |
|
Exclusion intervals ERS guidelines: exclude individual breaths or affected segments |
0.1-s periods | Breaths | Breaths | Breaths | Breaths |
| Valid measures | ≥70% valid data points and ≥2 valid breaths | ≥4 valid breaths | ≥4 valid breaths | ≥4 valid breaths | ≥4 valid breaths |
ERS: European Respiratory Society; Z: impedance; R7: resistance measured at 7 Hz; X7: reactance measured at 7 Hz; R: resistance; Z7: impedance measured at 7 Hz. #: The onset of a breath was identified from the volume signal as the time point where the slope changed from negative to positive.
In addition, a paediatric pulmonologist certified in paediatric pulmonology by the American Board of Pediatrics manually removed artefacts from 1 day of measurements (two tests) of a random sample of 49 participants. Manual artefact removal was conducted in a password-protected web application developed by our team solely for this purpose. The web application displayed a time-series image of the volume, flow, R7, X7 and Z7 of each measurement. Measurements were presented to the pulmonologist in a random order. The pulmonologist reviewed the time series and flagged irregular breaths and breaths with obstructions or leaks (table 1). We considered this method our reference standard for artefact removal.
Selection among repeated 20-s measurements was approached using four alternative methods. For the first method (All), we used all valid measurements without selection. For the second method (CoV15), we selected the first three measurements that yielded an R7 CoV of ≤15%, as recommended in the ERS technical standards [6]. For the third method (Closest3), we selected the three measurements with the closest values for each parameter. For the fourth method (Distance15), we removed 15% of the measurements with the largest absolute distance from the test average for each parameter.
Statistical analysis
Children who had an acute lower respiratory infection (ALRI) within 7 days preceding the visit, defined as difficulty breathing or fast breathing, were excluded from analysis. Difficulty breathing was determined by means of a caregiver report or nurse observation. Fast breathing was defined as a caregiver report of short, rapid breaths attributed to a “chest problem” or a measured respiratory rate of ≥40 breaths per minute in children with caregiver-reported cough or difficulty breathing.
Assessment of validity
We assessed criterion validity by evaluating the extent to which a child's height predicted the oscillometry test results of day 1. The predictability was estimated with the coefficient of determination (R2) obtained from simple linear regression models of test results regressed on height. This analysis leverages the recognition that height is a strong predictor of lung function.
We also assessed validity by estimating the inter-method intraclass correlation coefficient (ICC) of the results of tests after using each automated artefact removal method versus the reference method, which is manual artefact removal by a paediatric pulmonologist. We restricted the analysis to the first test per child that underwent manual cleaning. We estimated the ICCs with linear mixed-effects models with random child intercepts and a fixed effect for method. The 95% confidence intervals of the ICCs were constructed with a parametric bootstrap over 1000 samples. The bias in the automated methods was also reported.
Assessment of reliability
We estimated within-day reliability from repeated tests performed sequentially on the first day of testing and between-day reliability from repeated tests performed on consecutive days on a subgroup of children during the early study period. Within-day reliability was estimated by the ICCs calculated in linear mixed-effects models with a random intercept per child and a fixed effect for test order using tests conducted on the first day of testing. The linear mixed-effects models for estimating between-day ICCs had a random intercept per child and a fixed effect for day order. In a sensitivity analysis of within-day reliability, we also included a fixed effect for study period (early versus late). The 95% confidence intervals of the ICCs were constructed with a parametric bootstrap over 1000 samples.
All analyses were conducted in the R programming language version 4.4.0. The mixed-effects models and parametric bootstraps were implemented with the lme4 package [18]. The investigators’ R functions for quality control are available on GitHub (https://github.com/lgrajeda/2025-OSM-QC.git), Zenodo [19] or upon request to the corresponding author.
Results
Child participation
A total of 684 children were reached at home for oscillometry testing. Of these, 666 (97%) were free of ALRI in the preceding 7 days, and, of these, 537 (81%) were tested (figure 2). Of the 537 children tested, 80 were enrolled between February and May 2021, during which 2 days of testing were conducted. Of these 80 children, 76 (95%) had a test on the second day. The main reason some children did not participate in oscillometry testing was a lack of cooperation, which occurred in 125 (19%) children visited on day 1, and 3 (4%) children visited on day 2.
FIGURE 2.
Child participation in oscillometry testing.
Child characteristics
Of the 537 children, 55% (n=294) were male; the mean±sd age was 38±1 months, weight was 13±1 kg and height was 88±4 cm (supplementary table S1). In the preceding 7 days, 98 (18%) had had a cough, 24 (5%) had had a fever and 49 (9%) had had diarrhoea. Of the 537 children, 73 (14%) had four tests, 3 (1%) had three tests, 446 (83%) had two tests and 15 (3%) had one test.
Oscillometry data
In total, we collected 1 009 992 0.1-s intervals, 5580 measurements and 1208 tests (table 2). The number of measurements per test was three or fewer for 28 (2%) tests, four for 814 (67%) tests, five for 172 (14%) tests and between six and eight for 194 (16%) tests. Depending on the quality control method, we obtained test results for 85–99% of the 1208 tests collected, and at least one test result was available for 95–99% of the 537 children (table 2). Following quality control, the means and standard deviations of test results were lower than those before quality control (supplementary table S2).
TABLE 2.
Total data points and data points remaining after applying alternative quality control approaches
| Data level | Unfiltered,#,¶ n | Automated, n (%) | Manual,+ n/N (%) | |||
|---|---|---|---|---|---|---|
| Commercial¶ | AR1¶ | AR2¶ | AR3¶ | |||
| All § | ||||||
| 0.1-s intervals | 1 009 992 (100) | – | 578 088 (57) | 480 207 (48) | 461 848 (46) | 37 182/80 727 (46) |
| Breaths | 71 314 (100) | – | 37 810 (53) | 30 969 (43) | 29 680 (42) | 2523/5873 (43) |
| Measures | 5580 (100) | 5229 (94) | 5230 (94) | 4805 (86) | 4463 (80) | 358/447 (80) |
| Tests | 1208 (100) | 1194 (99) | 1199 (99) | 1194 (99) | 1194 (99) | 96/98 (98) |
| Children | 537 (100) | 528 (98) | 532 (99) | 529 (99) | 528 (98) | 49/49 (100) |
| CoV15 | ||||||
| Selected measures | 3543 (63) | 3537 (63) | 3465 (62) | 3255 (58) | 3090 (55) | 240/447 (54) |
| Tests | 1181 (98) | 1179 (98) | 1155 (96) | 1085 (90) | 1030 (85) | 80/98 (82) |
| Children | 526 (98) | 522 (97) | 524 (98) | 513 (96) | 508 (95) | 48/49 (98) |
| Closest3 | ||||||
| Selected measures | 3585 (64) | 3546 (64) | 3507 (63) | 3363 (60) | 3177 (57) | 246/447 (55) |
| Tests | 1195 (99) | 1182 (98) | 1169 (97) | 1121 (93) | 1059 (88) | 82/98 (84) |
| Children | 528 (98) | 522 (97) | 524 (98) | 517 (96) | 512 (95) | 48/49 (98) |
| Distance15 ƒ | ||||||
| Selected measures | 4714 (84) | 4419 (79) | 4435 (79) | 4050 (73) | 3733 (67) | 298/447 (67) |
| Tests | 1170 (97) | 1169 (97) | 1165 (96) | 1139 (94) | 1109 (92) | 91/98 (93) |
| Children | 528 (98) | 521 (97) | 524 (98) | 523 (97) | 520 (97) | 49/49 (100) |
AR: artefact removal; CoV15: coefficient of variation of ≤15%; Closest3: the three measurements with the closest values for each parameter; Distance15: 15% of the measurements with the largest absolute distance from the test average for each parameter were removed. #: No artefact filtering method applied; ¶: denominators correspond to the number of data points before artefact removal and measurement selection; +: denominators include only the unfiltered data subjected to manual artefact removal;§: no measurement selection algorithm applied; ƒ: data for resistance measured at 7 Hz (R7).
R7 coefficient of variation
The paediatric technical standard for valid oscillometry tests recommends that the CoV for R7 from at least three measurements should be ≤15% [6]. Before quality control, of the first 537 tests on the first day, 349 (65%) met this criterion, and of the 537 children, 473 (88%) had at least one of two tests on the first day that met this criterion. After quality control, the proportion of first-day first tests with R7 CoV ≤15% ranged from 76% with AR1-All to 100% with various methods, while the proportion of children who produced at least one test with R7 CoV ≤15% ranged from 93% (498/537) with AR1-All to 98% (524/537) with Unfiltered-CoV15 or Unfiltered-Closest3 (supplementary table S3).
Predictability of oscillometry measurements from participant height
Before quality control, height explained only a small proportion of the variability in first-day, first-test oscillometry results, with R2 of 0.04 for R7, 0.00 for R7–R19, 0.00 for X7 and 0.01 for AX. Following application of the investigators’ quality control methods, R2 values increased for R7, X7 and AX (R7 range: 0.08–0.13, X7 range: 0.05–0.11, AX range: 0.07–0.11). However, the highest R2 values for R7, X7 and AX were observed in a subsample of 49 children with manual artefact removal. By contrast, the predictability of R7–R19 remained low across all quality control methods, with the highest R2 (0.08) observed when manual quality control was applied, and all valid measures were averaged to derive test results (Manual-All) (figure 3 and supplementary table S4).
FIGURE 3.
Proportion of the variability of an oscillometry test explained by the child's height by quality control method in single tests (first test of day 1) and two-test average (average of up to two tests on day 1). The coefficient of determination (R2) was estimated from simple linear regression models of oscillometry parameters as the dependent variable and children's height as the independent variable. The models for the manual artefact removal (AR) algorithm included tests from a random sample of 49 children; all other models included test results from up to 537 children. CoV15: coefficient of variation of ≤15%; Closest3: the three measurements with the closest values for each parameter; Distance15: 15% of the measurements with the largest absolute distance from the test average for each parameter were removed.
Inter-method reliability
Test results had better agreement with those manually cleaned (reference standard) after automated cleaning than before. For R7 and R7–R19, the ICC point estimates were higher when using the investigators’ methods (AR1–AR3) than when removing artefacts with the commercial software. Moreover, the mean differences between test results derived using AR1–AR3 and those obtained with the reference standard were smaller than the corresponding difference observed for test results with artefacts filtered using the Commercial software (table 3). The difference is a measure of the bias in the evaluated method when compared with the reference standard.
TABLE 3.
Inter-method intraclass correlation coefficient (ICC) and difference between oscillometry tests produced with each artefact removal (AR) method and the manual method
| Oscillometry parameter | Unfiltered (n=49) | Commercial (n=49) | AR1 (n=49) | AR2 (n=49) | AR3 (n=49) |
|---|---|---|---|---|---|
| ICC (95% CI) # | |||||
| R7 | 0.65 (0.46–0.79) | 0.82 (0.70–0.90) | 0.87 (0.79–0.93) | 0.90 (0.83–0.94) | 0.90 (0.84–0.94) |
| R7–R19 | 0.50 (0.26–0.67) | 0.66 (0.47–0.79) | 0.87 (0.78–0.92) | 0.86 (0.76–0.92) | 0.84 (0.73–0.90) |
| X7 | 0.71 (0.54–0.83) | 0.84 (0.75–0.91) | 0.84 (0.74–0.91) | 0.81 (0.69–0.89) | 0.79 (0.66–0.87) |
| AX | 0.87 (0.78–0.92) | 0.97 (0.95–0.98) | 0.97 (0.94–0.98) | 0.97 (0.94–0.98) | 0.98 (0.97–0.99) |
| Difference (95% CI) ¶ | |||||
| R7 | 1.64 (0.89–2.39) | 0.84 (0.37–1.32) | 0.18 (−0.19–0.55) | 0.38 (0.06–0.71) | 0.25 (−0.07–0.57) |
| R7–R19 | 1.49 (0.84–2.14) | 0.93 (0.51–1.34) | 0.26 (−0.01–0.52) | 0.35 (0.08–0.61) | 0.41 (0.12–0.69) |
| X7 | −0.83 (−1.27– −0.40) | −0.82 (−1.25– −0.39) | 0.00 (−0.28–0.28) | −0.51 (−0.82– −0.20) | −0.51 (−0.84– −0.18) |
| AX | 12.23 (5.44–19.03) | 5.92 (2.90–8.95) | 1.46 (−1.88–4.79) | 4.32 (1.03–7.61) | 1.82 (−0.51–4.15) |
No measurement selection algorithm was applied. Models are restricted to the first test on the first day. R7: resistance measured at 7 Hz; R7–R19: frequency dependence of resistance; X7: reactance measured at 7 Hz; AX: area of reactance. #: ICC between the test values obtained after applying the quality control method listed in the column and the test values obtained after manual artefact removal; ¶: average of the differences between the test values obtained after applying the quality control method listed in the column and the test values obtained after manual artefact removal.
Within-day reliability
Depending on quality control criteria, the within-day ICCs ranged from 0.77 to 0.86 for R7, from 0.51 to 0.74 for R7–R19, from 0.42 to 0.84 for X7 and from 0.67 to 0.89 for AX (figure 4 and supplementary tables S5 and S6). The magnitude of the ICCs was not sensitive to adjustments for study period (supplementary table S9).
FIGURE 4.
Within-day reliability of test results from day 1 by quality control method. R7: resistance measured at 7 Hz; R7–R19: frequency dependence of resistance; X7: reactance measured at 7 Hz; AX: area of reactance; CoV15: first three measurments with R7 coefficient of variation ≤15%; Closest3: the three measurements with the closest values for each parameter; Distance15: 15% of the measurements with the largest absolute distance from the test average for each parameter were removed; AR: artefact removal.
Between-day reliability
The between-day ICCs of the first tests before quality control were 0.37 (95% CI 0.17–0.55) for R7, 0.17 (95% CI 0.00–0.37) for R7–R19, 0.06 (95% CI 0.00–0.28) for X7 and 0.15 (95% CI 0.00–0.35) for AX (figure 5). The highest between-day ICC was obtained after AR3–Distance15 was applied. These ICCs were 0.75 (95% CI 0.63–0.84) for R7, 0.57 (95% CI 0.39–0.72) for R7–R19, 0.69 (95% CI 0.54–0.80) for X7 and 0.74 (95% CI 0.61–0.84) for AX. By contrast, the Commercial–CoV15 ICCs were 0.68 (95% CI 0.54–0.79) for R7, 0.45 (95% CI 0.25–0.61) for R7–R19, 0.60 (95% CI 0.43–0.72) for X7 and 0.71 (95% CI 0.57–0.80) for AX. The use of two-test averages, instead of single tests, increased the between-day ICCs (figure 5 and supplementary table S7).
FIGURE 5.
Between-day reliability by quality control method for single tests (first test) and the two-test average (average of up to two tests). R7: resistance measured at 7 Hz; R7–R19: frequency dependence of resistance; X7: reactance measured at 7 Hz; AX: area of reactance; CoV15: first three measurments with R7 coefficient of variation of ≤15%; Closest3: the three measurements with the closest values for each parameter; Distance15: 15% of the measurements with the largest absolute distance from the test average for each parameter were removed; AR: artefact removal.
Temporal shifts
In general, we observed lower oscillometry values in R7, R7–R19 and AX and higher oscillometry values in X7 on the second day of testing than on the first, and during the later study period than in the earlier period. The mean difference (in cmH2O·s·L−1) in the first tests between consecutive days was −0.37 (95% CI −0.84–0.10) for R7, −0.18 (−0.53–0.17) for R7–R19, 0.22 (−0.15–0.58) for X7 and −7.16 (−13.58– −0.70) for AX when using Commercial–CoV15, and −0.28 (95% CI −0.69–0.12) for R7, −0.21 (−0.52–0.10) for R7–R19, 0.08 (−0.22–0.37) for X7 and −0.95 (−6.96–5.05) for AX when using AR3–Distance15 (supplementary table S8). The mean difference in child respiratory frequency between days 2 and 1 was −0.90 breaths per minute (95% CI −1.85–0.04). The mean differences (in cmH2O·s·L−1) in the first tests between study periods were −0.29 (95% CI −0.96–0.38) for R7, −0.25 (−0.57–0.07) for R7–R19, 0.49 (0.06–0.91) for X7 and −4.77 (−14.96–5.42) for AX when using Commercial–CoV15, and −0.55 (95% CI −1.17–0.06) for R7, −0.46 (−0.79–−0.13) for R7–R19, 0.29 (−0.09–0.68) for X7 and −0.45 (−10.44–9.53) for AX when using AR3–Distance15. The mean difference in child respiratory frequency between the late and early study periods was 2.54 breaths per minute (95% CI 1.06–4.03).
Discussion
We evaluated the validity and reliability of oscillometry tests conducted in the homes of 3-year-old children living in rural Guatemala. We tested multiple alternative quality control methods for upper-airway artefact removal and measurement selection. We found that respiratory oscillometry can be performed with good validity and reliability for R7, X7 and AX, and with moderate reliability for R7–R19. In general, reliability was higher when using the AR3 artefact removal method combined with Distance15 for measurement selection instead of the Commercial–CoV15, which is currently the standard method. AR3 filters out breaths with obstructions identified as R7, X7 or Z7 values more than 3 standard deviations from the test mean, and breaths with leaks identified as R7 or X7 values close to 0 for >15% of the breath. Distance15 filtered out measurements that had the largest absolute distance between the observed oscillometry value within a single measurement and the average of all valid measurements within a test. Averaging two tests per child increased the between-day reliability further.
Oscillometry has been recommended as a feasible lung function technique for preschool-aged children [4]. The main challenge we encountered was obtaining a child's cooperation to perform the test; however, cooperation seemed to improve as children became more familiar with the procedure. On the first day, 19% of the children were uncooperative and could not be tested, compared with 14% on the second day. Encouragingly, the majority of the tests conducted were valid, between 85% and 99%, depending on the quality control method. A feasibility study conducted in Spain among asthmatic children aged 6–14 years reported a similar proportion of successful tests, with 83% (n=154) of children providing a valid test [5]. Although obtaining a child's cooperation at this early age can be challenging, once achieved, oscillometry is feasible [20].
In 49 children, we found manual artefact removal by a paediatric pulmonologist to be the most valid quality control method for R7 and AX of the methods we compared. Our results showed that oscillometry results that underwent manual control were more highly associated with height than the other tests. Manual artefact removal consisted of a thorough visual inspection of breaths and the removal of irregular breaths and breaths with signs of flow obstructions or flow leaks. This justifies our use of manual quality control as a reference standard to validate automated quality control procedures.
Height is the main predictor variable used in population reference equations for lung function [21]. We used the proportion of variability in oscillometry explained by height (R2) as a relative indicator of criterion validity across parameters and quality control methods. In our cohort, height explained a modest proportion of variability in R7, X7 and AX, and less in R7–R19. This pattern is consistent with Canadian reference equations in children aged 3–17 years, where R2 was higher for R7 and X7 (0.68 and 0.59, respectively) and lower for R7–R19 (0.33) [21]. The lower R2 values in our study (0.08–0.28 after quality control) than in the Canadian study probably reflect the narrower height range of our population (74–99 cm versus 100–189 cm). Because R2 is constrained by variability in height, its absolute magnitude is less informative than the relative difference observed across parameters and quality control methods.
Between-day ICCs estimated in our study were similar in magnitude to those estimated in a reliability study of 48 healthy Australian children aged 8–11 years, despite these children being older and perhaps more cooperative with the oscillometry technique than our young population. In the Australian study, the ICCs between tests collected 2 weeks apart were 0.81 for R6 and 0.73 for X6. In that study, tests were constructed from three 16-s measurements that underwent complete breath manual quality control [22]. Importantly, between-day ICCs should be considered in sample size planning for epidemiological exposure-response studies, as random outcome measurement error inflates residual variance and reduces statistical power [23].
Commercial–CoV15, the current standard method, differs from our proposed artefact removal (AR1–3) and measurement selection (Closest3 and Distance15) methods in some features that may help explain the lower reliability we observed with Commercial–CoV15. First, artefact removal excluded periods corresponding to whole breaths, while the commercial method excluded parts of breaths containing artefacts. Higher ICCs of tests that underwent whole-breath compared with partial-breath quality control were also observed in the reliability study of 48 Australian children [22]. Second, in harmony with the ERS technical standards, AR2 and AR3 include criteria to remove leaks manifested as dips of R7 or spikes of X7; by contrast, the commercial algorithm does not account for leaks, which may bias measurements towards lower R and less negative X values [6]. Third, artefact removal methods identified artefacts in the R7, X7 and Z7 signals, but the commercial used R7 only. Lastly, CoV15 selects the first three measures that produce a CoV for R7 of ≤15%; however, we believe quality control should be applied by parameter [24].
We observed lower R7, R7–R19 and X7 values and higher (less negative) X7 values on the second day of testing than on the first, and during the later study period than in the early period, although not all of these differences were statistically significant. We observed no difference in mean respiratory frequency between days, but a greater respiratory frequency in the late study period. Because identical quality control algorithms were applied consistently throughout the study and manual review was conducted in random order, these differences are unlikely to reflect changes in quality control procedures. Between-day differences may instead relate to greater child familiarity with the procedure (e.g. reduced tongue interference and improved mouth seal), leading to fewer artefacts on day 2. Between-period differences could be explained by increasing nurse experience over time (e.g. improved coaching) or measurement on the second day of consecutive days of visits later in the study. Within-day ICCs remained stable after adjustment for study period, suggesting that within-day repeatability was not affected. However, because repeated testing on consecutive days was conducted only in the early study period, between-day reliability estimates should be interpreted cautiously. If there were an increase in nurse experience over time, this could result in a reduction in within-child variance and potentially lead to higher ICCs in the later period. As the children grow older and become more familiar with the technique, we anticipate that these logistical issues may be resolved.
A key strength of our quality control algorithms is that they were developed in accordance with ERS technical standards and can be adapted to other study settings and age groups with only minor modifications (e.g. multifrequency range, CoV threshold). However, they are currently configured to process data exports from the Tremoflo device; therefore, implementation with other oscillometry systems would require additional adjustment. This practical barrier may be substantial, as many research groups lack the programming expertise to adapt R code to different data export formats. Nevertheless, the algorithms are publicly available on GitHub to facilitate transparency, reproducibility and collaborative development. Broader adoption of this feasible technique for young children will require access to user-friendly, standardised quality control tools that can be applied across devices and enable valid comparisons between systems [6].
This study has several limitations. First, due to resource limitations, we were unable to revisit the homes of children who, during the first visit, had ALRI documented within 7 days before testing. However, these children comprised only 3% of all the children visited. Second, repeated testing on consecutive days was performed only on a subset of the cohort who turned 3 years between February and May 2022. Oscillometry results in this early period differed modestly from those observed in the later period. Although within-day ICCs were unchanged after adjustment for study period, between-day ICCs were estimated exclusively in this early-period subgroup and therefore may not fully reflect reliability in the overall cohort. Third, some estimates had a low precision. A larger sample size in a future study will address this limitation. Fourth, with our measurement protocol, which restricted data collection to four measurements when an R7 CoV of <15% was achieved, we were unable to evaluate the effect on the validity and reliability of increasing the number of measurements per test. However, we addressed a similar question by doubling tests. Fifth, the narrow height range in our cohort most likely produced low R2 values. Extending the age range of participants, and thus the height, will permit a more robust evaluation of criterion validity. Sixth, manual artefact removal was performed by a single paediatric pulmonologist, which may have introduced variability and attenuated validity estimates. Future validation studies should incorporate multiple independent reviewers and a formal assessment of inter-rater reliability.
Conclusion
In a cohort of 3-year-old children living in Guatemala, we performed oscillometry measurements with moderate to good validity and reliability during home visits after quality control. Validity and reliability when applying the investigators’ methods were similar to or higher than those with commercial artefact removal followed by measurement selection based on CoV15. The investigators’ methods followed the recommended technical standards, included other parameters besides R7 and have the added advantages of being open source and providing transparency and flexibility to investigate better ways of optimising quality control.
Acknowledgements
We thank the participants of the HAPIN trial and our staff at Universidad del Valle de Guatemala: Erick Alarcón, Rosalbina Cisneros, Libny Monroy, Waldemar Nájera, Alejandro Polanco, Alexander Ramirez and Carla Trinidad.
Footnotes
Provenance: Submitted article, peer reviewed.
Ethics statement: We performed oscillometry on 3-year-old children during household visits. The study was approved by the institutional review boards at Emory University in Atlanta, Georgia, USA (00089799), and Universidad del Valle de Guatemala in Guatemala City, Guatemala (146-08-2016).
Author contributions: T. Clasen, J. Peel and J.P. McCracken acquired the funding. T. Petrovick, E.D. McCollum and J.P. McCracken were responsible for the conceptualisation and design. L.M. Grajeda, T. Petrovick, K. Castellanos, A. Díaz-Artiga, M.R. López, A. Barraza-Villarreal, E.D. McCollum and J.P. Mcracken acquired the data. L.M. Grajeda and K. Castellanos curated the data. L.M. Grajeda, E.D. McCollum and J.P. McCracken were responsible for data analysis. L.M. Grajeda, L.M. Thompson, C.M. Eckhardt, Y. Shen, J. Knight, L.P. Naeher, E.D. McCollum and J.P. McCracken were responsible for data interpretation. All authors were responsible for writing, reviewing and editing. L.M. Grajeda, E.D. McCollum and J.P. McCracken had full access to and verified the data, and L.M. Grajeda had the final responsibility to submit for publication.
Conflict of interest: C.M. Eckhardt reports stock or stock options with Merck & Co. and has been an employee of Merck & Co. since July 2025. All other authors report no conflicts of interest.
Support statement: This study was supported by the National Institutes of Health (R01ES033530 (principal investigators: T. Clasen and J. Peel) and R01HL163256 (principal investigator: J.P. McCracken)) and the University of Georgia (J.P. McCracken). Funding information for this article has been deposited with the Open Funder Registry.
Supplementary material
Please note: supplementary material is not edited by the Editorial Office, and is uploaded as it has been supplied by the author.
Supplementary material
01238-2025.SUPPLEMENT
Data availability
Contact John P. McCracken (John.McCracken@uga.edu) to request access to the de-identified data used in this analysis. The investigators’ R functions for quality control of oscillometry measurements are available on GitHub (https://github.com/lgrajeda/2025-OSM-QC.git) and Zenodo (https://doi.org/10.5281/zenodo.18745348).
References
- 1.Institute for Health Metrics and Evaluation. Global Burden of Disease Study 2019 (GBD 2019) results. Date last accessed: 21 September 2023. https://vizhub.healthdata.org/gbd-results/
- 2.Stocks J, Hislop A, Sonnappa S. Early lung development: lifelong effect on respiratory health and disease. Lancet Respir Med 2013; 1: 728–742. doi: 10.1016/S2213-2600(13)70118-8 [DOI] [PubMed] [Google Scholar]
- 3.Kajekar R. Environmental factors and developmental outcomes in the lung. Pharmacol Ther 2007; 114: 129–145. doi: 10.1016/j.pharmthera.2007.01.011 [DOI] [PubMed] [Google Scholar]
- 4.Beydon N, Davis SD, Lombardi E, et al. An official American Thoracic Society/European Respiratory Society statement: pulmonary function testing in preschool children. Am J Respir Crit Care Med 2007; 175: 1304–1345. doi: 10.1164/rccm.200605-642ST [DOI] [PubMed] [Google Scholar]
- 5.Dominguez-Martin C, Cano A, Diez-Monge N, et al. Spirometry and respiratory oscillometry: feasibility and concordance in schoolchildren with asthma. Pediatr Pulmonol 2023; 58: 1896–1903. doi: 10.1002/ppul.26409 [DOI] [PubMed] [Google Scholar]
- 6.King GG, Bates J, Berger KI, et al. Technical standards for respiratory oscillometry. Eur Respir J 2020; 55: 1900753. doi: 10.1183/13993003.00753-2019 [DOI] [PubMed] [Google Scholar]
- 7.Kaminsky DA, Simpson SJ, Berger KI, et al. Clinical significance and applications of oscillometry. Eur Respir Rev 2022; 31: 210208. doi: 10.1183/16000617.0208-2021 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Brashier B, Salvi S. Measuring lung function using sound waves: role of the forced oscillation technique and impulse oscillometry system. Breathe (Sheff) 2015; 11: 57–65. doi: 10.1183/20734735.020514 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Bhatawadekar SA, Leary D, Chen Y, et al. A study of artifacts and their removal during forced oscillation of the respiratory system. Ann Biomed Eng 2013; 41: 990–1002. doi: 10.1007/s10439-012-0735-9 [DOI] [PubMed] [Google Scholar]
- 10.Pham TT, Leong PHW, Robinson PD, et al. Automated quality control of forced oscillation measurements: respiratory artifact detection with advanced feature extraction. J Appl Physiol (1985) 2017; 123: 781–789. doi: 10.1152/japplphysiol.00726.2016 [DOI] [PubMed] [Google Scholar]
- 11.Veneroni C, Acciarito A, Lombardi E, et al. Artificial intelligence for quality control of oscillometry measures. Comput Biol Med 2021; 138: 104871. doi: 10.1016/j.compbiomed.2021.104871 [DOI] [PubMed] [Google Scholar]
- 12.Marchal F, Schweitzer C, Demoulin B, et al. Filtering artefacts in measurements of forced oscillation respiratory impedance in young children. Physiol Meas 2004; 25: 1153–1166. doi: 10.1088/0967-3334/25/5/006 [DOI] [PubMed] [Google Scholar]
- 13.Clasen T, Checkley W, Peel JL, et al. Design and rationale of the HAPIN study: a multicountry randomized controlled trial to assess the effect of liquefied petroleum gas stove and continuous fuel distribution. Environ Health Perspect 2020; 128: 47008. doi: 10.1289/EHP6407 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Checkley W, Thompson LM, Sinharoy SS, et al. Effects of cooking with liquefied petroleum gas or biomass on stunting in infants. N Engl J Med 2024; 390: 44–54. doi: 10.1056/NEJMoa2302687 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Clasen TF, Chang HH, Thompson LM, et al. Liquefied petroleum gas or biomass for cooking and effects on birth weight. N Engl J Med 2022; 387: 1735–1746. doi: 10.1056/NEJMoa2206734 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.McCollum ED, McCracken JP, Kirby MA, et al. Liquefied petroleum gas or biomass cooking and severe infant pneumonia. N Engl J Med 2024; 390: 32–43. doi: 10.1056/NEJMoa2305681 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.McCollum ED, Checkley W, Balakrishnan K, et al. Fetal and early-life stove and fuel interventions and respiratory oscillometry lung function measures in preschool-aged children from Ghana. Am J Respir Crit Care Med 2024; 209: 625–626. doi: 10.1164/rccm.202311-2197ED [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Bates D, Mächler M, Bolker B, et al. Fitting linear mixed-effects models using lme4. J Stat Softw 2015; 67: 1–48. doi: 10.18637/jss.v067.i01 [DOI] [Google Scholar]
- 19.Grajeda L, Castellanos K, Petrovick T, et al. Assessing pulmonary function in 3-year-old children using respiratory oscillometry: a validation study in rural Guatemala. Investigator's R function for quality control, version 1. Zenodo. Date last accessed: 17 April 2026. Date last updated: 23 February 2026. 10.5281/zenodo.18745348 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Gheorghiu RM, Stan IV. Reflections of a paediatric pulmonologist: strategies for optimising lung function tests in preschool children. Breathe (Sheff) 2025; 21: 240178. doi: 10.1183/20734735.0178-2024 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Ducharme FM, Smyrnova A, Lawson CC, et al. Reference values for respiratory sinusoidal oscillometry in children aged 3 to 17 years. Pediatr Pulmonol 2022; 57: 2092–2102. doi: 10.1002/ppul.25984 [DOI] [PubMed] [Google Scholar]
- 22.Robinson PD, Turner M, Brown NJ, et al. Procedures to improve the repeatability of forced oscillation measurements in school-aged children. Respir Physiol Neurobiol 2011; 177: 199–206. doi: 10.1016/j.resp.2011.02.004 [DOI] [PubMed] [Google Scholar]
- 23.Baker D, Nieuwenhuijsen MJ. Environmental Epidemiology: Study Methods and Application. Oxford, Oxford University Press, 2008. [Google Scholar]
- 24.Hantos Z, Wu JKY, Dandurand RJ, et al. Quality control in respiratory oscillometry: reproducibility measures ignoring reactance? ERJ Open Res 2023; 9: 00070-2023. doi: 10.1183/23120541.00070-2023 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Schweitzer C, Chone C, Marchal F. Influence of data filtering on reliability of respiratory impedance and derived parameters in children. Pediatr Pulmonol 2003; 36: 502–508. doi: 10.1002/ppul.10359 [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Please note: supplementary material is not edited by the Editorial Office, and is uploaded as it has been supplied by the author.
Supplementary material
01238-2025.SUPPLEMENT
Data Availability Statement
Contact John P. McCracken (John.McCracken@uga.edu) to request access to the de-identified data used in this analysis. The investigators’ R functions for quality control of oscillometry measurements are available on GitHub (https://github.com/lgrajeda/2025-OSM-QC.git) and Zenodo (https://doi.org/10.5281/zenodo.18745348).