Mylène Jansen

226 Chapter 11 (visits). However, in general the bias is small compared to the mean values and treatment effect that has been observed thus far. 8–10 Moreover, this bias becomes irrelevant when comparing differences in changes over time between groups, e.g. treatment arms. Still, when analyzing changes over time, it is strongly recommended to randomize the chronological order in which radiographs are analyzed, so this bias will not be of relevance. Although speculative, the systematic bias for bone density and osteophyte area may be caused by a gradual learning curve of the observer in identifying the outer and inner boundaries of the osteophytes and the edges of the bone ‘cartilage’ interface (black to white interface on the radiographs). Moving the small circles that determine JSW and bone density somewhat will likely not affect JSW significantly, but if the circle is placed slightly outside the actual (white) bone area, a small number of pixels could be dark-gray to black (background) and significantly impact the average gray value. It is remarkable that for many parameters, the SDD was lower (better) for severe OA patients analyzed within a month in this study than for mild OA patients from the original publication. However, the differences are not very large. Again the explanation may be found in a learning curve by analyzing KIDA images over the past 12 years. In this case the experience is in favor of the technique (reproducibility), instead of the time-dependent bias. The more important conclusion is that for most parameters, intra-observer variation is similar in severe OA patients compared to mild OA patients. Medial osteophyte areas seem to be the exception, and have a much bigger (worse) SDD for severe OA patients. For both medial osteophytes and lateral tibial osteophytes it was shown that the intra-observer variation depended on the osteophyte area, as bigger osteophytes, associated with more severe OA, and a higher KL grade results in a larger variation between measurements. This explains why, even if SDDs are not comparable between patients of different severities, the CVs are (as they are corrected for the mean osteophyte area, and with that partly for severity as well). Osteophytes did not only show a relatively high dependence on mean values, but also on whether the reanalysis was performed within a long or short period. All 4 osteophyte locations showed a clear floor effect in the complete dataset of 293 radiographs (Figure S3), which disappeared for the 98 radiographs reanalyzed within 1 month (Figure S8). This may also be explained by a learning curve, as these osteophytes were not recognized as osteophytes in the first reading (value 0) but were recognized as osteophytes in the second reading. Furthermore, while ICC improved for all parameters when reanalyzed within a month (compare Table 2 with Table 1), this effect was the most notable for the osteophyte measurements. Apparently the osteophytes are the parameters most sensitive to intra-observer variability. This may be explained by the fact that the values depend on a calculated area within a manually delineated boundary, a subjective action sensitive to a learning curve.