Fig 1.
MNREAD curve examples.
Fig 2.
Example of the standard deviation algorithm calculation on a typical MNREAD curve.
On iteration 1 (dark blue), the algorithm selects the first two sentences as plateau 1 (x-axis: 1.3 and 1.2 logMAR) and calculates a selection criterion for this plateau. Criterion plateau 1 = mean (reading speed plateau 1)– 1.96 x standard deviation (reading speed plateau 1) = 60.5–1.96 × 2.1 = 56.3 wpm. The point adjacent to plateau 1 (x-axis: 1.1 logMAR) was read at 60 wpm, which is faster than criterion plateau 1, indicating that this point belongs to the optimal plateau. A second iteration is then launched (light blue) with plateau 2 now encompassing the first three sentences and a new criterion calculation. Criterion plateau 2 = 60.3–1.96 × 1.5 = 57.3 wpm. Among the points adjacent to plateau 2, there is still a value higher than this criterion (y-axis: 59 wpm at x-axis: 0.9 logMAR), so the algorithm continues to iterate one sentence at a time, including reading speeds at 1.0 logMAR in plateau 3 and at 0.9 logMAR in plateau 4. The calculations stop with plateau 4, for which selection criterion is higher than any remaining points (criterion plateau 4 = 44.7 wpm). MRS is estimated as 57.2 wpm and a corresponding CPS of 0.9 logMAR.
Table 1.
Details of the ICC form chosen for analyses 1, 2 and 3.
Fig 3.
Box and whisker plots of estimated MRS (left) and CPS (right), grouped by raters and sorted in ascending order of expertise level (from 0 to 4). Boxes represent the 25th to 75th percentiles and whiskers range from min to max values. Medians (lines) and means (red cross) are also represented.
Table 2.
Self-reported score of expertise for our 7 raters.
Fig 4.
Bland–Altman plots showing agreement between SDev and NLME methods for both MRS (left) and CPS (right). x-axes represent the mean estimate for both methods; y-axes represent the estimate difference between the two methods (SDev—NLME). In both subplots, the red dashed lines represent the mean difference (i.e. bias) and the blue dashed lines represent the agreement limits (±1.96 SD). The dotted lines show the 95%CI of the limits; top and right histograms show the data distribution along the x- and y-axes respectively. Tables summarize the SDev and NLME average values as well as their ICC values of absolute agreement.
Table 3.
Absolute agreement (ICC values and their 95% confidence intervals) between CPS values estimated with the SDev and the NLME methods for five different definitions of CPS.
Fig 5.
Box and whisker plots showing MRS (left panel) and CPS (right panel) obtained with the two algorithms and the mean for all raters. The box represents 25th to 75th percentile with median line, the red + sign represents the mean and the whiskers represent minimum to maximum.
Table 4.
Absolute agreement (ICC values and their 95% confidence intervals) between CPS values estimated by the raters and with the NLME method for five different cut-off values of CPS.