Kellgren & Lawrence grading in the check cohort: methods matter

Purpose: Using Kellgren & Lawrence (KL), radiographic osteoarthritis (ROA) is usually defined as KL Grade≥2, however KL≥1 has been suggested as a better threshold for investigating early ROA. A challenge is that classifying osteophytes as doubtful or definite is subjective. One factor in particular that may influence score is whether images are read at a single time point vs. simultaneously across multiple time points. Understanding the extent to which these two approaches influence KL score could inform study design and interpretation. We aimed to evaluate KL scores assigned to knee and hip radiographs at a single time point in comparison to scores assigned to the same radiographs when read together with follow-up radiographs, in individuals with recent onset knee and/or hip ROA symptoms in primary care. We then explored how KL scores assigned using the two approaches influenced longitudinal study outcomes. Methods: The Cohort Hip and Cohort Knee (CHECK) is a prospective multicenter cohort study of 1002 individuals, eligible if aged 45-65 years with recent onset of knee and/or hip pain or stiffness for which they had not yet sought medical care, or had done so for the first time not more than six months prior to enrollment. Radiographs of knees and hips were acquired at baseline, then at two, five, eight, and ten years follow-up. For the knee, we assigned KL scores using semi-flexed weightbearing (WB) postero-anterior radiographs. For the hip, we did so using WB antero-posterior radiographs. At baseline, images were scored (blinded to complaints) by a member of the CHECK Steering Committee (three rheumatologists, one professor/physical therapist, one rehabilitation physician, one physician-researcher, one biologist, and one physiotherapist). These scores were never made available at subsequent readings. Independently of the Steering Committee scores, all baseline and follow-up images were scored by research assistants trained by an experienced musculoskeletal radiologist and general practitioner (GP). We previously reported inter-rater reliability of the research assistants compared to the GP (prevalence and bias adjusted kappa 0.6 for the knee, 0.8 for the hip). They assigned scores at each visit with access to previous images and scores. After the 10 year follow-up, final baseline scores were assigned with access to up to five follow-up images and previous scores. Data integrity was monitored by the GP and an independent researcher, and scores were adjusted if appropriate (e.g. missing score but radiograph available, score suggesting KL regression over time). We first described the prevalence of each KL score in the sample using the two approaches (single time point vs. multiple time points), and then defined ROA using two definitions, early (KL≥1) or established (KL≥2). We then compared how the two approaches affected baseline ROA prevalence using mixed effects Poisson regression with robust estimates of variance. Next, we compared how the two approaches and two ROA definitions at baseline influenced relative risk (RR) for undergoing joint replacement by the end of the study, using similar models. Finally, we compared the RR of developing incident established ROA for a baseline score KL1 vs. KL0, using the two scoring approaches. Results: Of 1002 participants, 792 (79%) were women, the mean (SD) age was 55.9 (5.2) years, and BMI was 26.2 (4.0) kg/m2. KL scores differed between the two approaches: 20% of knee and 25% of hip images scored KL0 at baseline were assigned higher scores once read by research assistants with radiographs at multiple follow-up time points; there was a 50% increase in knee KL1 scores, 140% for hip; and a 150% increase in knee KL2 scores, 230% for hip (Tables 1 & 2). Using the multiple time points approach resulted in 1.7 and 2.4 times the prevalence of early (KL≥1) and established (KL≥2) knee ROA, respectively; and 2.5 and 2.9 times the prevalence of early and established hip ROA, respectively, compared to the single reading approach. RR for incident knee replacement (KR) in early ROA compared to no ROA was higher using the second approach (RR 12.3 vs. 8.5), and this approach captured the largest number of KRs; for established ROA, the RR was lower (RR 5.0 vs. 17.9) and captured the fewest number of KRs. RR for hip replacement (HR) did not change substantially between approach for early ROA, but it did capture 21 more HRs using the second approach; and for established ROA the RR again was lower. Confidence intervals overlapped for all comparisons. RR for incident established knee and hip ROA for baseline KL1 compared to KL0 was slightly and significantly higher using the second approach for both joints, RR 1.7 and 2.0, respectively. Conclusions: Compared to a single reading, when scoring radiographs together with follow-up images, the number of assigned KL0 scores decreased and KL1/2 scores increased, leading to more individuals being classified as having both early and established ROA. While arguably a source of bias, it may be that seeing later images (and progression of ROA over time) increases the reader’s confidence in classifying baseline osteophytes. We acknowledge, however, this could also relate to individuals with different training and experience assigning scores. Though not significant, reclassifying KL0 to KL1 improved predictions of joint replacement in early ROA, and reclassifying KL1 to KL2 reduced predictions in established ROA. Regardless of approach, defining ROA as KL≥1 instead of KL≥2 in our younger cohort with recent onset of symptoms captured more future joint replacements, even though change in RR was not consistent.View Large Image Figure ViewerDownload Hi-res image Download (PPT)