Evaluating blood oxygen saturation measurements by popular fitness trackers in postoperative patients: A prospective clinical trial

Figures

• 

  Figure 1. Study flow chart Top : (I) During screening of 288 patients, 87 of these met exclusion criteria. (II) Informed consent was obtained by 201 patients, of whom 112 patients could be included in the measurements. Bottom : For the SaO2 benchmark (left ), 347 valid measurements (IIIa) were obtained, which after outlier removal led to 343 measurement pairs to be considered in our statistical analysis (IVa). Regarding the SpO2 benchmark, 706 valid measurements (IIIb) yielded 694 pairs to be evaluated. Of note, data acquisition in (IIIa, IVa) and (IIIb, IVb) are based on the same patient cohort recruited in (I, II).

• 

  Figure 2. Bland-Altman plots comparing SaO2 measurements by ABG to the SpO2 readings of each of the investigated devices, including TPO Following the visualization proposed by Bland and Altman, scatterplots showing the real errors of the measurements (y axis: SpO2 measurements minus SaO2 reference) stratified by the mean of each measurement pair (x axis). Dashed horizontal lines mark the bias (B), i.e., the arithmetic average of all real errors with the limits of agreement (LoA) as determined by an offset of ±2 times the standard deviation (SD). Error bars show the 95% confidence interval (CI) for the bias and both LoA. For the ease of comparison, data points are color-coded, specifically for each of the devices: TPO = orange (top-left); Apple = red (top-right); Garmin = blue (bottom-left); Withings = green (bottom-right).

• 

  Figure 3. Linear correlation assessment of the blood oxygen saturation measurements comparing the investigated devices to ABG Scatterplots localize each of the paired measurements (x,y) by the SaO2 reference value obtained by ABG (x) and the corresponding SpO2 measurement of the benchmarked device (y). The black solid line depicts the linear regression model, with the 95% confidence interval shaded in gray. Color codes for the devices: TPO = orange (top-left); Apple = red (top-right); Garmin = blue (bottom-left); Withings = green (bottom-right).

• 

  Figure 4. Agreement of SpO2 measurements between fitness trackers and TPO Bland-Altman diagrams (upper ) and scatterplots (lower ) assess the agreement between the SpO2 readings obtained by fitness trackers to the SpO2 reference values defined by TPO measurements. Due to the discrete nature of the SpO2 measurements, multiple data points coinciding at the same coordinates are visualized by circles with varying diameters. The black solid line depicts the linear regression model, with the 95% confidence interval shaded in gray. Color codes for the fitness trackers: Apple = red (left); Garmin = blue (center); Withings = green (right).

• 

  Figure 5. Analysis of potential confounders Patients were segregated in different cohorts according to their attributes classified by variables of different nature (x axis), to assess potential influences on the fitness trackers readings (y axis). In all diagrams, the colors identify the device: TPO = orange; Apple = red; Garmin = blue; Withings = green. (A) the real errors are stratified by perfusion index. (B and C) characteristics of the ABG analysis. (D and E) boxplot visualisations of the absolute errors binned by categorical classifications of the patient attributes. (F) barplots contrasting the dropout rate in non-vs. shivering patients after surgery. ∗∗p < 0.01; ∗∗∗p < 0.001 (Fisher’s Exact Test).