|Disease severity assessment in epidemiological studies: accuracy and reliability of visual estimates of Septoria leaf blotch (SLB) in winter wheat|
M. El Jarroudi (1), L. Kouadio (2), C. Mackels (1), B. Tychon (1), P. Delfosse (3), C. H. BOCK (4). (1) Univ of Liege, Arlon, Belgium; (2) Agriculture and Agri-Food Canada | Agriculture et Agroalimentaire Canada, Lethbridge, Alberta, Canada; (3) Centre de Recherche Public - Gabriel Lippmann, Belvaux, Luxembourg; (4) USDA-ARS-SEFTNRL, Byron, GA, U.S.A.
The accuracy and reliability of visual assessments of SLB severity by raters (i.e. one plant pathologist with extensive experience and three other raters trained prior to field observations using standard area diagrams and DISTRAIN) was determined by comparison with assumed actual values obtained by digital image analysis. Initially analyses were performed using SLB severity over the full 0-100% range; then, to explore error over short ranges of the 0-100% scale, the scale was divided into sequential 10%-increments based on the actual values. Lin’s concordance correlation (LCC) analysis demonstrated that all raters were accurate when compared over the whole severity range (LCC coefficient (ρc)= 0.92-0.99). However, agreement between actual and visual SLB severities was less good when compared over the short intervals of the 10×10% classes (ρc= -0.12-0.99), demonstrating that agreement will vary depending on the actual disease range over which it is compared. Inter-rater reliability between each pair of raters over the full 0-100% range (correlation analysis r= 0.970-0.992, <i>P</i><0.0001), and inter-class correlation coefficient (ρ≥ 0.927) were very high. This study provides new insight into using a full range of actual disease severity vs limited ranges to ensure a realistic measure of rater accuracy and reliability, in addition to contributing to the ongoing debate on the use of visual disease estimates based on the 0-100% ratio scale for epidemiological research.