Finger Versus Earlobe Pulse Oximetry During the 6-Minute Walk Test in Interstitial Lung Disease

Absolute agreement in minimum SpO2 during the six-minute walk test between the finger and earlobe sensors, quantified with the intraclass correlation coefficient (two-way mixed-effects model, absolute agreement) and its 95% confidence interval. ICC values are interpreted as below 0.50 poor, 0.50 to 0.75 moderate, 0.75 to 0.90 good, and above 0.90 excellent agreement.

Agreement in minimum SpO2 during the six-minute walk test between the finger and earlobe sensors, assessed with Bland-Altman analysis: mean bias (earlobe minus finger) with its 95% confidence interval and 95% limits of agreement, evaluated against a prespecified clinically acceptable difference of 3 percentage points. The direction of the bias is reported, as the finger is hypothesised to register lower values than the earlobe.

Proportion of participants whose classification of significant exertional desaturation changes between the finger and earlobe sensors at the threshold of minimum SpO2 below 88% (American Thoracic Society indication threshold for ambulatory oxygen). Agreement is assessed with the Cohen kappa coefficient (Landis-Koch interpretation) and the difference in proportions with the McNemar test for paired data. At this threshold the McNemar test and the raw reclassification proportions are the primary read, as kappa can be unstable when prevalence is extreme.

Proportion of participants whose classification of significant exertional desaturation changes between the finger and earlobe sensors at the threshold of minimum SpO2 below 90% (British Thoracic Society indication threshold for ambulatory oxygen). Agreement is assessed with the Cohen kappa coefficient (Landis-Koch interpretation) and the difference in proportions with the McNemar test for paired data.

Proportion of participants whose classification changes between the finger and earlobe sensors at the prognostic threshold of a fall in minimum SpO2 of 4% or more from baseline. Agreement is assessed with the Cohen kappa coefficient (Landis-Koch interpretation) and the difference in proportions with the McNemar test for paired data.

Comparison of the agreement between sensor locations (intraclass correlation coefficient and Bland-Altman) in autoimmune-associated ILD versus non-autoimmune ILD, with interstitial pneumonia with autoimmune features (IPAF) described separately. Where sample size permits, agreement is also described, exploratorily and without formal hypothesis testing, by major ILD subtype (for example idiopathic pulmonary fibrosis, hypersensitivity pneumonitis, or connective tissue disease-associated ILD), and by autoimmune subtype (for example systemic sclerosis or rheumatoid arthritis).

Comparison, among enrolled participants, of the rate of invalid or poor-quality (low-signal) readings between the finger and earlobe sensors during the six-minute walk test, across the eight measurement timepoints. At each timepoint the paired heart rate from the two sensors is used as a signal-quality control: a discordant heart rate flags poor signal or sensor disconnection and identifies invalid SpO2 readings, so that agreement is assessed only on simultaneous, valid readings. The per-participant difference in the rate of invalid readings between sensor locations is reported with its 95% confidence interval and the McNemar test for paired data.

Clinical, vascular and functional factors associated with discordance between sensor locations, including ILD type (autoimmune-associated versus other), autoimmune subtype (systemic sclerosis versus other), fibrosing versus non-fibrosing ILD, pathological capillaroscopy, Raynaud phenomenon, skin pigmentation, cardiovascular comorbidity, and pulmonary function (FVC and DLCO). For each discordance criterion (SpO2 below 88%, SpO2 below 90%, and a fall of 4% or more) a logistic regression model is fitted with the patient as the unit of analysis, complemented by mixed-effects logistic regression (GLMM) with the measurement timepoint as a predictor and a random intercept per participant. Missing data from incomplete follow-up are handled by maximum likelihood; penalised regression (L1/L2) is considered if events are sparse.

Agreement between sensor locations in SpO2 at baseline, maximum, per-minute intervals during the walk, and recovery, assessed with the intraclass correlation coefficient and Bland-Altman analysis.

Exploratory analysis of systematic differences between the finger and earlobe in SpO2. The paired difference (earlobe minus finger) is summarised with its 95% confidence interval (paired t test, or Wilcoxon signed-rank test with a Hodges-Lehmann interval, as appropriate). Passing-Bablok regression is used to detect constant and proportional differences between the two locations, with 95% confidence intervals for the intercept and the slope.

Exploratory comparison of the agreement of finger and of earlobe SpO2 with baseline arterial oxygen saturation, in the subgroup of participants undergoing arterial blood gas sampling.

Exploratory description of the direct and indirect costs associated with invalid or poor-quality finger readings. No prespecified cost model is applied.