This was a retrospective, observational study of infant SF₆-MBW data from previously described cohorts of healthy infants (Basel-Bern Infant Lung Development (BILD) cohort (18, 19) and infants diagnosed with CF (Swiss Cystic Fibrosis Infant Lung Development (SCILD) cohort (20); Supplementary Table S1). The reference examiner (MO) selected part of the dataset to ensure a minimum number of measurements with artefacts present, while the additional measurements were randomly selected to obtain a total of 200 SF₆-MBW trials per setup (Spiroware and WBreath; Table 1). Each dataset included measurements from study visits at six weeks of age (BILD cohort) and six weeks or one year of age (SCILD cohort). The Ethics Committee of the Canton of Bern, Switzerland approved the study protocol (B201901072, PB_2017-02139) and parents gave written consent.

MBW measurements were performed during natural, non-REM sleep at six weeks of age and under sedation with chloral hydrate (75 mg/kg; per rectum) at one year of age in accordance with current ATS/ERS standards (1). While sleeping in supine position with head in midline, infants breathed through a facemask (open cuff silicone mask, size 1; Draeger AG, Luebeck, Germany). Flow and molar mass (MM) signals were measured by an ultrasonic flowmeter [Exhalyzer D, Eco Medics AG; with either WBreath 3.28.0 (ndd Medizintechnik AG) or Spiroware 3.2.1 software (Eco Medics AG)] using 4% SF₆. The operator aimed for 2–3 valid trials per subject and setup.

Per measurement occasion, one trial was included in the study dataset for analysis. Calculation of outcome parameters, LCI and FRC, were performed in the software versions 3.52.3 for WBreath (14), and 3.3.1 for Spiroware (16, 21). For visual quality control, we developed custom Python scripts to display data gathered by both setups by reading the exported text files of each trial. Raw data containing text files in Spiroware (A-files) consisted of unprocessed flow, molar mass (MM), oxygen (O₂), and carbon dioxide (CO₂) signals (13). In WBreath, raw data consisted of corrected flow and MM signals (14) (ASCII export) after applying signal processing which includes BTPS-correction, temperature simulation, and a step-response correction according to standard protocols (14). Besides the flow and volume trace, we additionally displayed the following signals in Python: for Spiroware files tracer gas, O₂, and CO₂ concentrations; for WBreath mainstream MM and a computed tracer gas concentration signal. We programmed an automated identification of the critical test phases for both washin [Figure 1, blue area (A)] and washout [Figure 1, green area (B)] defined as five breaths before washin/washout start to five breaths after reaching the test-end criterion [i.e., 1/40th of the starting tracer gas-concentration)]. Further, we developed a heat map visualization based on the tidal volume difference of each breath to the median tidal volume over the measurement. Within the flow signal trace, we developed the option of automated sigh detection, and observer-based identification of irregular breathing and breath hold assisted by visual quality criteria.

Based on current ATS/ERS consensus recommendations (1), we developed a step-wise workflow for the systematic assessment of visual quality criteria (Supplementary Summary). All SF₆-MBW trials were preliminarily evaluated, categorized, and annotated by the reference examiner (MO; Table 1). The washin and washout phase within each trial were assessed separately. Excluded trials may feature more than a single artefact in the critical phase of the washin or washout, whereby the examiners were instructed to identify all detected artefacts. For an acceptable trial and the calculation of outcomes, all qualitative criteria had to be met within the critical test phase of the washout. Outside the critical test phase, the following qualitative criteria could deviate and, depending on the individual case, the washout still be classified as acceptable: sigh, irregular tidal breathing, and breath hold. An error was defined as empty or erroneous signal trace(s) and led to exclusion of the trial.

Three blinded reviewers (MBW experienced: YS, ID; MBW novice: NK) evaluated the same set of 400 MBW trials using the tool and following the quality check-list (Supplementary Summary). For each trial, the reviewers assessed the washin and washout separately and documented their (i) decision to accept/reject, (ii) identified visual artefacts, and (iii) reason for exclusion using the developed tool. In a next step, we compared the interrater agreement for the decision to accept/reject between the reference (MO) and each of the three reviewer (YS, ID, NK) using Cohens Kappa (κ) and between all reviewers using Brennan and Predigers Kappa coefficient (22). We interpreted the Kappa coefficient between 0.41 and 0.60 as moderate, 0.61 and 0.80 as substantial, and 0.81 and 1.0 as almost perfect agreement. Interrater agreement was then compared between (a) the setups (WBreath and Spiroware), and (b) between healthy infants and infants with CF. The reported percent agreement is the percentage of all identically rated decisions (to accept or to reject the washin or the washout) among the reference and the three reviewers. Finally, we examined whether the systematic quality control had an influence on LCI and FRC as main outcome parameters of MBW. For this, we compared mean LCI/FRC of all acceptable washouts between the reviewers using paired t-tests per setup, and per disease group. Additionally, the reference examiner (MO) investigated the total time required to (i) boot the analysis software (WBreath, Spiroware, and the developed visual QC tool) and import raw data, and (ii) perform visual quality control and document (a) the decision to accept/reject, (b) identified visual artefacts, and (c) the reason for exclusion in a subset of five files per setup (randomly chosen from the main dataset). The total time required to perform a visual quality control and document the findings and decision to accept/reject was compared by paired t-tests (Supplementary Table S2). Statistical analysis was performed using STATA 16 (StataCorp, College Station, USA) and GraphPad Prism 9 (GraphPad Software, San Diego, USA). A p ≤ 0.05 was considered significant.

The assessment of infant SF₆-MBW measurements can be divided into (i) the analysis of gathered raw data (signal processing and outcome calculation), (ii) a numeric QC based on pre-defined thresholds (e.g., equilibrium of exogenous washin gas, or reaching a tracer gas concentration below 2.5% at the end of the washout), and (iii) a visual QC for artefacts (e.g., leaks, sighs, irregular breathing patterns, or breath holds; Table 2). We developed a software package for the visualization of (raw Spiroware and signal processed WBreath) signal traces supporting a systematic assessment of pre-defined visual QC criteria, which is now available to researchers (https://doi.org/10.6084/m9.figshare.22193737.v2). A detailed summary on applied criteria is provided in the online Supplementary.

The user interface features four sections: (i) signal traces of flow, volume, and tracer gas over time, as well as MM (for WBreath data) and O₂ and CO₂ (for Spiroware data), (ii) display options (e.g., zoom and a heat map visualization of tidal volume per breath), (iii) systematic assessment options (for both washin and washout) to accept/reject, identify artefacts (leak, sigh, irregular breathing pattern, breath hold), and comment, and (iv) additional information for the operator [e.g., automatic identification of sighs, deviations in step response-correction, or MM steps between phases (Figure 1)]. A color-coded critical phase for washin (blue) and washout (green), as well as the achievement of the 2.5% criterion (1), provide the operator with visual support (automatic phase detection; Figure 1).

The characteristics of the study population for both setups are summarized in Supplementary Table S1. While part of the dataset were selected by the reference examiner (to ensure a minimum number of measurements with artefacts present), the additional measurements were randomly selected to obtain 200 SF₆-MBW trials per setup (Table 1). For WBreath, the study population included more healthy children than patients with cystic fibrosis, while for Spiroware the groups were comparable. On average, the patients with cystic fibrosis were older than the healthy controls.

The reference dataset included a comparable amount of acceptable and unacceptable trials in both setups (Table 1). The most common artefacts leading to trial exclusion in both setups were leak and sigh (Table 1). Using the newly programmed tool for the systematic visual quality assessment by the three independent reviewers, the interrater agreement ranged from 81.1% to 86.3% among the reviewers (kappa Spiroware washout: 0.637; kappa WBreath washout: 0.653; Table 3). The interrater agreement for the decision to accept/reject the washout between all individual reviewers was similar in WBreath (82.7%) compared to Spiroware (81.8%). Comparison of the interrater agreement (decision to accept/reject the washout) between healthy children and children with CF showed no difference within the setups (WBreath p = 0.055, Spiroware p = 0.261; Table 3). For both infant MBW setups, the time required to perform visual quality control and documentation of findings (e.g., the decision to accept/reject) was substantially shorter when using the newly developed tool compared to the standard analysis software WBreath or Spiroware (paired t-test, n = 5, WBreath p < 0.001; Spiroware p = 0.031; Supplementary Table S2).

In both setups LCI was higher in the CF group compared to the healthy children (Supplementary Table S3). Systematic visual quality control had no substantial influence on test results. There was no significant difference in LCI and FRC of acceptable trials between the reviewers, neither per setup nor per disease group within the setups (Supplementary Table S3).

The newly developed tool provides a framework for swift and systematic visual assessment of SF₆-MBW measurements in infants, with straightforward documentation and good agreement among experienced and novice users. While the tool can read Spiroware raw data (A-files) and perform all underlying signal processing steps, we were not able to read original WBreath raw files (.brw files) in all cases. Therefore, a signal analysis in the WBreath software including a data export of corrected signals remains necessary before data visualization with our tool is possible. Our reference dataset consisted of one trial per measurement occasion, thus hampering the evaluation of intra-test variability. However, the newly developed tool will allow a systematic approach with a simple and user-friendly application interface.