Two cytokines, FGF-basic and Monocyte-inducible protein (MIP)-1β, detected in less than 25% of NS and WO tear samples, were excluded from subsequent analyses. The remaining 25 cytokines were detected in at least 80% of patient samples. NS and WO tear comparisons across the entire patient group are shown in
Table 2. Scatter plots for most cytokines show good agreement, with significant NS versus WO correlations for all 25 cytokines, 19 showing correlation coefficients > 0.75 (
R 2 ≥ 0.56). However, scatter plots do not reveal the agreement between WO cytokine levels and their corresponding NS levels. Apart from higher means for NS tear cytokine levels, paired
t-tests and signed rank tests show significant distributional differences between NS and WO tear samples for nearly all cytokines. This is due to a broader range of NS tear values, in particular at the upper end. For any cytokine showing a significant distributional difference but no Bland–Altman plot bias, the WO cytokine level is simply underestimating the NS level by a constant value and is a viable “proxy” measurement. However, when there is significant Bland–Altman slope bias, the ability of WO tears to replace NS is questionable (
Table 2). Nine cytokines fall into this category: IL-8, IL-1β, VEGF, Granulocyte-colony stimulating factor (G-CSF), IL-9, IL-15, IL-6, IL-7, and IP-10. As an example, NS versus WO scatter plots for IL-2 and IL-6 (
Figs. 1,
2, respectively) both show reasonably strong, statistically significant (
Table 2), correlations. However, the corresponding Bland–Altman plots (
Figs. 3,
4), demonstrate stronger agreement between tests for IL-2 than for IL-6. There are more data points outside the ±2 standard deviation Bland–Altman limit for IL-6, and there is a significant Bland–Altman plot bias (slope) for IL-6, but not IL-2 (
Table 2).