NEW RADIOGRAPHIC GRADING SCALES FOR OSTEOARTHRITIS OF THE HAND

Transcription

1 1584 NEW RADIOGRAPHIC GRADING SCALES FOR OSTEOARTHRITIS OF THE HAND Reliability for Determining Prevalence and Progression DOUGLAS A. KALLMAN, FREDRICK M. WIGLEY, WILLIAM W. SCOTT, JR.. MARC C. HOCHBERG, and JORDAN D. TOBIN We developed and evaluated scales for grading the prevalence and progression of the individual radiographic features of osteoarthritis (OA) of the hand. Four equally time-spaced hand radiographs from 50 participants in the Baltimore Longitudinal Study of Aging, who were followed for at least 20 years, were read separately and blindly by 4 experienced, trained readers. Eleven hand joints were individually assessed for the presence of osteophytes, joint space narrowing, subchondral cysts, subchondral sclerosis, lateral deformity, and cortical collapse, and were also graded using the KellgreniLawrence scale, a global estimate of OA. The intraclass correlation coefficient was used to determine cross-sectional inter- and intrareader reliability. Interreader agreement on OA progression was evaluated using life-table analysis. Each of the grading scales for the individual radiographic features of OA of the hand, except for cysts, were cross-sectionally reliable between readers. Each reader was able to reproduce his original scores almost perfectly when grading the radio- From the Gerontology Research Center, Applied Physiology Section, the National Institute on Aging; and the Department of Medicine, Rheumatology Division at Francis Scott Key Medical Center, the Department of Radiology, Skeletal Radiology Section, and the Department of Medicine, Division of Molecular and Clinical Rheumatology, Johns Hopkins University School of Medicine, Baltimore, Maryland. Douglas A. Kallman, MD: Gerontology Research Center; Fredrick M. Wigley, MD: Rheumatology Division at Francis Scott Key Medical Center; William W. Scott, Jr., MD: Department of Radiology; Marc C. Hochberg, MD, MPH: Division of Molecular and Clinical Rheumatology; Jordan D. Tobin, MD: Gerontology Research Center. Address reprint requests to Jordan D. Tobin. MD, Gerontology Research Center Eastern Avenue, Baltimore, MD Submitted for publication March ; accepted in revised form July 25, graphs a second time. Using these scales, the different readers agreed on all the features of longitudinal progression of OA of the hand, except cysts. We believe that this method for grading hand radiographs for OA has proved to be reliable for both cross-sectional and longitudinal studies. The method of grading radiographic changes of osteoarthritis (OA), developed by Kellgren and Lawrence in 1957 (I) and adopted by the World Health Organization in 1961 (2), has been accepted as the gold standard for both cross-sectional ( 34) and longitudinal studies (7,8). However, the KellgreniLawrence scale assumes a sequence of events in the progression of OA that has never been validated by longitudinal studies. The basis of the KellgreniLawrence scale is the presence of osteophytes, which is considered the pathognomonic feature of OA. Joint space narrowing, subchondral sclerosis, cysts, and joint deformity, all signs of OA, are considered to be insufficiently specific changes, unless seen in conjunction with osteophytes (2). Experience dictates that in OA, joints may in fact be narrowed, sclerotic, and deformed, without the presence of osteophytes, but according to the KellgreniLawrence scale, if osteophytes are not present, the joint must be graded as negative for OA. Recently, there has been interest in developing grading systems for OA that yield more information than the KellgreniLawrence scale. Altman et a1 (9) analyzed some of the individual radiographic features of OA of the hand in 24 pairs of hand films from selected patients with a clinical diagnosis of OA who were followed for an average of 4 years. In the present study, we developed new grading scales and an atlas of standard findings for the individual radiographic fea- Arthritis and Rheumatism, Vol. 32, No. 12 (December 1989)

2 RADIOGRAPHIC GRADING SCALES FOR OA 1585 Table 1. Rating methods used in scales for grading individual features of osteoarthritis of the hand Feature Osteophytes Joint space narrowing* Subchondral sclerosis Subchondral cysts Lateral deformityt Collapse of central joint cortical bone Grade 0 = none 1 = small (definite) osteophyte(s) 2 = moderate osteophytefs) 3 = large osteophyte(s) 0 = none 1 = definitely narrowed 2 = severely narrowed 3 =joint fusion at at least 1 point 0 = absent 1 = present 0 = absent 1 = present 0 = absent 1 = present 0 = absent 1 = present * Scores are based on the amount of narrowing between bone end plates, not on osteophyte bridging. t Defined as malalignment of at least 15 degrees. tures of OA of the hand. These scales were tested for reliability using hand radiographs from 50 healthy men who had been followed for at least 20 years, and who were selected without knowledge of the presence of radiographically or clinically evident OA. In this study, we established the cross-sectional and longitudinal reliability of these new OA grading scales and compared the reliability of these scales with that of the KeHgrenlLawrence scale. SUBJECTS AND METHODS Subjects and radiographs. Subjects participating in the Baltimore Longitudinal Study of Aging (BLSA) routinely have posteroanterior left hand radiographs taken, without intensifying screens or magnification, at 24-year intervals. These apparently healthy men drawn from the community population have been described previously (10). For this study, radiographs from 50 subjects who had been followed for at least 20 years were selected based on the availability of 4 equally time-spaced left hand films of good quality. Radiographic scoring. On each hand radiograph, individual joints were graded for the presence and severity of 6 selected individual features of osteoarthritis: osteophytes, joint space narrowing, subchondral sclerosis, subchondral cysts, lateral deformity, and cortical collapse (Table I). Osteophytes and narrowing were differentiated into 3 grades, while sclerosis, cysts, lateral deformity, cortical collapse, and narrowing of the space of the first carpometacarpal joint were scored as either absent or present. Lateral deformity was defined as malalignment of at least 15 degrees. The joints were also graded using the Kellgren/ Lawrence scale (2). We modified the original descriptive terms of Kellgren and Lawrence (none, doubtful, minimal, moderate, severe) and developed a more detailed numerical rating based on an explanation of the scale written by Kellgren (2). For this study, the KellgredLawrence scale rating was 0 = none; 1 = doubtful; 2 = minimal-definite small osteophyte(s), possibly with minimal narrowing, cysts, or subchondral sclerosis; 3 = moderate-moderately sized osteophyte(s), or definite osteophyte(s) with moderate narrowing, cysts, or subchondral sclerosis; and 4 = severelarge osteophyte(s), or small or moderate osteophyte(s) with severe narrowing, cysts, subchondral sclerosis, or deformity. Grade 1 of our osteophyte scale (definite small osteophyte) corresponds to grade 2 of the KellgredLawrence scale. Grade 0 (none) of the osteophyte scale corresponds to either grade 0 (none) or grade 1 (doubtful) of the Kellgred Lawrence scale, i.e., doubtful osteophytes were graded 0 on our osteophyte scale. We also developed an atlas of standard radiographic findings illustrating the individual features of OA to be graded on the scale. This is to be used as a reference for training to achieve consistency in reading hand radiographs (Figures 1 and 2). Readers and training. Four readers (DAK, FMW, WWS, MCH), who each had previous experience using the KellgredLawrence scale, familiarized themselves with the scales for the individual features and with the radiographic atlas. Each reader separately scored a training packet of 20 radiographs from BLSA subjects not selected for this study; they then collectively discussed their results to derive a consensus on the scoring method. Reading radiographs. Radiographs selected for the study were assigned random numbers, and all identifying marks were obscured. Each reader read the 200 hand radiographs (50 subjects, 4 films each) one at a time, in random order. In addition, 1 film from each subject was randomly selected to be read again by each reader 1-2 months later. To help maintain consistency, the readers were encouraged to refer to the atlas of individual joint feature scales and the KellgredLawrence atlas. The 5 distal interphalangeal (DIP) and 4 proximal interphalangeal (PIP) joints, the first carpometacarpal (CMC1) joint, and the trapezioscaphoid (TS) joint were each scored individually using the scales for individual features and the KellgredLawrence scale. The CMCl and TS joints were not scored for cortical collapse and the TS joint was not scored for osteophytes and lateral deformity, because these findings are not characteristic of OA in these joints. Kellgren and Lawrence did not develop a method for scoring the TS joint. Data analysis. All 4 radiographs from each subject were used for analysis of longitudinal interreader agreement. The last radiograph taken of each subject was used for analysis of cross-sectional interreader agreement. Interreader agreement was higher for first radiographs, presumably because they showed a lower prevalence of disease. Therefore, data from last visits provided a more conservative estimate of agreement. For each individual feature scale and the Kellgren/Lawrence scale, the most severely affected and the number of affected DIP and PIP joints were also determined. The percent agreement for the 6 possible pairings of the 4 readers was calculated. The average percent agreement of the 6 pairs is presented here. Additionally, the

3 1586 KALLMAN ET AL Figure 1. Atlas of the joint space narrowing and osteophyte scales for grading osteoarthritis of the second to the fifth interphalangeal joints, the first interphalangeal joint, the first carpometacarpal joint, and the trapezioscaphoid joint. Narrowing of the first carpometacarpal joint is scored as 1 = present and 0 = absent. The trapezioscaphoid joint is not graded for osteophytes.

4 RADIOGRAPHIC GRADING SCALES FOR OA 1587 Figure 2. Atlas of lateral deformity (malalignment of at least 15 degrees), subchondral sclerosis, and cortical collapse, as shown in interphalangeal joints. These changes are scored as I = present and 0 = absent. overall percent agreement for a hand was determined for each scale by averaging the scores for all joints. Percent agreement is affected by the number of categories considered and by the prevalence of disease; therefore, a statistic that corrects for chance agreement was necessary to adequately assess the reliability of the readings. The weighted kappa statistic was developed to estimate reliability for nominal scale data (1 1). Weights can be chosen that set kappa equal to the intraclass correlation coefficient (ICC) (12). While kappa has the advantage of being able to be weighted so that widely disparate readings are granted no credit, in our data set, readings more than 1 grade apart were rare. Therefore, for this study, weighted kappa had no practical benefit over the ICC, and for this reason we have reported the ICC (13). Reliability was determined between the different readers (interreader) and between 2 readings by the same reader (intrareader). Longitudinal agreement was determined using lifetable analysis (SAS statistical software) (14). The most severely affected DIP and PIP joints, and the number of these joints involved, were used as the variables for longitudinal analysis. The progression of OA in the CMCl and TS joints was also determined. Progression of OA by 1 grade signified an event, and observations were right-censored if there was no progression during the followup period. A subject was excluded from life-table analysis if he started the study with a maximum score, and thus could not progress. The log rank test, which places more weight on longer survival times, and the Wilcoxon signed rank test, which places more weight on shorter survival times (15), were used to test whether the 4 readers were different in their assessment of progression using each of the scales. RESULTS Characteristics of the study population. Subjects ranged in age from 25.7 years to 68.1 years (mean 44.8) at the time of their first radiograph. The mean age at the last visit, which was used for the cross-sectional analysis of reliability, was 67.6 years. The length of followup varied from 20.0 years to 26.6 years, with a mean of 22.8 years. On average, radiographs were taken 7.6 years apart. Ninety-two percent of the subjects had at least some evidence of OA of the hand (KellgredLawrence grade 1) at the time of their last radiograph. Osteophytes and narrowing were the most common radiographic features, present in 81% and 88% of subjects, respectively. Sclerosis was found in 46%, while cysts were present in 33% of subjects. Lateral deformity and cortical collapse were less common findings, present in <20% of study participants. Cross-sectional interreader reliability. Percent agreement. The average percent agreement between pairs of readers (pairwise exact agreement) was highest for the dichotomous variables (present/absent) and ranged from 96% to 99% (Table 2). Of the scales with multiple categories, readers agreed most on the scores for osteophytes (86% agreement), followed by those for narrowing and the KellgredLawrence scale (Table 2). Pairwise agreement within 1 grade was nearly perfect for scores for narrowing and osteophytes, and was 93% for the KellgredLawrence scale. Reliability. Different readers were able to substantially reproduce each other s scores (interreader reliability) on the osteophyte (ICC = 0.71), narrowing (ICC = 0.70), sclerosis (ICC = 0.60), and Kellgred Lawrence scales (ICC = 0.74). Interreader reliability was less strong for scores for lateral deformity (ICC = 0.42) and cortical collapse (ICC = 0.56), and was weak for scores for cysts (ICC = 0.29). The ICC was also separately calculated for the DIP, PIP, CMC 1, and TS joint scores (Table 2). Reliability between readers was similar to the overall results for both the DIP and PIP joints, with only scores for collapse being markedly more reliable in the DIP joints. Interreader reliability was good for scores for all CMCl joint features except cysts. There was also good agreement between readers for narrowing and sclerosis scores in the TS joint. Cross-sectional intrareader agreement. Percent agreement. Readers agreed better with themselves at a second reading (intrareader percent agreement) than they did with other readers (interreader percent agreement) (Table 2). As with interreader percent agreement, the intrareader percent agreement was highest for the dichotomous scales, ranging from 98% to 99%. Of the multiple grade scales, the osteophyte scale had the highest intrareader percent agreement (93%). Readers agreed within 1 grade with their prior scores

5 1588 KALLMAN ET AL Table 2. Interreader (intrareader) agreement of scales for grading osteoarthritis of the hand* Scale ~~ Kellgrenl Osteophyte Narrowing Sclerosis cysts Deformity Collapse Lawrence % agreementt Pairwise exact 86 (93) 79 (88) 96 (98) 96 (99) 99 (99) 99 (99) 78 (85) Pairwise within 1$ 98 (99) loo (100) NA NA NA NA 93 (98) lntraclass correlation coefficient Overall (0.77) 0.70 (0.75) 0.60 (0.77) 0.29 (0.74) 0.42 (0.80) 0.56 (0.84) 0.74 (0.80) DIP joints (0.86) 0.74 (0.82) 0.54 (0.71) 0.27 (0.73) 0.46 (0.83) 0.67 (0.84) 0.74 (0.87) PIP joints# 0.73 (0.65) 0.66 (0.62) 0.62 (0.93) 0.32 (0.75) 0.33 (0.85) 0.41 (PA) 0.74 (0.68) CMCl joint 0.82 (0.87) 0.60 (0.80) 0.77 (0.83) 0.32 (PA) 0.60 (0.61) NA 0.81 (0.87) TS joint NA 0.74 (0.85) 0.69 (0.66) PA (PA) NA NA NA * NA = not applicable (for percent agreement because of the dichotomous variable; for intraclass correlation coefficient because items were not graded according to the scoring scale); PA = perfect agreement, but no agreement beyond chance; CMCl = first carpometacarpal; TS = trapezioscaphoid. t Expressed as the mean of 6 pairwise comparisons between readers; mean of scores for all 11 joints. $ Agreement within 1 grade. 0 Mean of scores for all 11 joints. ll Mean of scores for 5 distal interphalangeal (DIP) joints, including the interphalangeal joint of the thumb, # Mean of scores for 4 proximal interphatangeal (PIP) joints. (painvise within 1) 98-99% of the time for each of the multiple category scales. Reliability. Readers were able to substantially reproduce their own readings using all scales (intrareader reliability), with the ICC ranging from 0.74 to 0.84 (Table 2). Intrareader reliability was higher than interreader reliability. Intrareader reliability was also separately calculated for the DIP, PIP, CMC1, and TS joints (Table 2). For each of the joints, reliability between a reader s scores at a first and a second reading (intrareader reliability) was good for each of the grading scales (ICC ranging from 0.61 to 0.93). Intrareader agreement was higher for scores for the DIP joints than for the PIP joints on the Kellgred Lawrence, osteophyte, and joint space narrowing scales. Longitudinal interreader reliability. The progression of the highest degree of severity of DIP joint osteophytes and joint space narrowing, as determined by the 4 readers, is presented graphically in Figure 3. In general, the readers agreed very well on the percentage of the study population whose osteoarthritis progressed over a year period for all scales except that for cysts (Table 3). The readers agreed on the longitudinal scoring of deformity and collapse for all joints. There was scattered disagreement between readers when scoring progression on the osteophyte, narrowing, sclerosis, and KellgredLawrence scales. DISCUSSION In this study, new grading scales for the individual radiographic features of osteoarthritis have been shown to be reliable when used with both crosssectional and longitudinal data by trained readers from different specialties. These scales can be used to determine the prevalence of osteoarthritic joint abnormalities and may be useful for prospective longitudinal studies. We would like to emphasize several unique features of this study: 1) a normative aging population not selected for the presence of joint abnormalities, 2) lengthy followup, 3) development of an atlas of standard radiographic findings, and 4) reliability comparable with that of the KellgredLawrence scale. For this study, subjects participating in the Baltimore Longitudinal Study of Aging were selected purely on the availability of multiple radiographs of the hand, taken over at least 20 years, and not for any specific clinical or radiographic features. The BLSA is an ongoing, long-term evaluation of aging processes in normal volunteers from the community. The results of this study may be generalizable to other normative aging populations. Our findings established the longitudinal reliability of scales grading the individual features of OA over a long period of time. This is important because OA tends to progress very slowly. Earlier studies on

6 RADIOGRAPHIC GRADING SCALES FOR OA 1589 OSTEOPHYTE NARROWING Percent of 60 Su bjects wlthn00a 50 Progression i! 30 I]., -, L.-.,.,.,.,,I.,.,.,.,.,,, TIME (years) TIME (years) Figure 3. Radiographic evidence of progression of osteophytes and joint space narrowing in the study population over time, as determined by the 4 readers, each represented by a different line. OA = osteoarthritis. BLSA subjects showed that it took 11 years on average for OA to progress 1 KellgredLawrence scale grade for DIPjoints (8). Because subjects in this study were followed for twice this period, there was ample time to detect change. The reliability between readers for the Kellgren/Lawrence scale scores in this study was similar to that originally reported by Kellgren and Lawrence (1). When their data were tested, the ICC was 0.71 for the determination of the most affected DIP joint. In the current study, the ICC was 0.66 for the determination of the most affected DIP joint, and 0.74 when scores for the 5 DIP joints were averaged. The readers in our study reproduced each other s results using the scales Table 3. Longitudinal interreader agreement, determined by life-table analysis, of scales for grading osteoarthritis of the hand* Scale Joint Kellgreni measure Osteophyte Narrowing Sclerosis Cysts Deformity Collapse Lawrence Highest DIP? Agree Agree Disagree$ Disagree$ Agree Agree Agree Number DIPS Agree Agree Disagree$ Disagree$ Agree Agree Agree Highest PIP? Agree Disagree7 Agree Disagree$ Agree Agree Disagree# Number PIPS Disagree# Disagree7 Agree Disagree$ Agree Agree Disagree$ CMCl Agree Agree Agree Disagree7 Agree NA Agree TS NA Agree Agree Agree NA NA NA * Agree = agreement between readers (i.e., not significantly different, P > 0.05 by both the Wilcoxon and log rank tests); NA = not graded according to the scoring scale. See Table 2 for other definitions. t The highest degree of disease seventy in the joint group. $ P < 0.05, by log rank test and Wilcoxon signed rank test. 5 Number of involved joints in the joint group. 7 P < 0.05, by log rank test. # P < 0.05, by Wilcoxon signed rank test.

7 1590 KALLMAN ET AL for osteophytes, narrowing, and sclerosis with about the same level of reliability as for the Kellgren/ Lawrence scale. The scores for lateral deformity and cortical collapse were somewhat less reliable, while the interreader reliability for scoring cysts was low. The low agreement on scores for cysts might have been due to ambiguity on the part of readers about what constituted normal bony trabecular pattern versus abnormal subchondral cystic change. Readers agreed more with themselves than they did with other readers (intrareader reliability was higher than interreader reliability). This was not surprising because, despite training and the use of the radiographic atlas, each reader graded the radiographs according to his or her own inherent bias about what constituted a positive radiographic finding. In their original study, Kellgren and Lawrence also had higher intrareader reliability than interreader reliability (1). Readers agreed on the longitudinal progression of all individual joint features, except for cysts. Thus, while readers might only agree moderately well on the presence of a joint finding (for example, deformity [ICC = 0.42]), in general, they graded progression similarly. In fact, when progression rated on the KellgredLawrence scale for this study was compared with similar ratings done 13 years earlier by another reader (8), the results were in close agreement. The findings of this study suggest that it may be sufficient to have only 1 trained reader for longitudinal studies of OA, because different readers determined progression similarly and because the intrareader reliability was nearly perfect. Having more readers may be a more sensitive method of measurement, but given the funds necessary to perform large, prospective longitudinal studies, using 1 reader may be a cost-effective alternative. There are some interesting issues involved in the method of reading serial radiographs in longitudinal studies of OA. One method is to read pairs of radiographs placed side by side. Using this method, readers assign progression scores for time sequence and the magnitude of change between films. Although this method may increase sensitivity, it is theoretically possible that it will be affected by a time sequence bias. If progression of OA in 1 joint is seen unequivocally on 1 of the 2 radiographs, a reader will be able to ascertain the time sequence of the radiographs. With this knowledge, the reader may be biased toward grading other joints on the same film as showing progression. Another method is to read serial films one at a time, in random order. This method, while avoiding a time sequence bias, has the theoretical potential of being influenced by changes in hand positioning, such as rotation or finger flexion, which could give a false appearance of enlarging osteophytes or joint space narrowing. It appears that both methods of reading radiographs have potential biases. In the current study, radiographs were read individually in random order, to minimize the potential of time sequence bias and for efficiency in reading multiple films per subject. In summary, the scales for grading individual radiographic features of osteoarthritis of the hand (osteophytes, joint space narrowing, and sclerosis) that were developed in this study were reliable between trained readers from different specialties. In addition, each reader was able to reproduce his original scores almost perfectly when grading the radiographs a second time. The different readers agreed on the longitudinal progression of OA of the hand, as graded by these scales, for all joint features except cysts. Therefore, the method we have developed for grading hand radiographs for osteoarthritis has proved to be reliable for both cross-sectional and longitudinal studies. ACKNOWLEDGMENTS We thank Harry Carr, James Wood, and Kim Roadarmel for performing the hand radiographs; Paul Ciesla, Charlotte Adler, and Elsie Fields for preparing the photographic atlas; Kim Roadarmel for technical assistance; and the Baltimore Longitudinal Study subjects for their devotion to aging research. REFERENCES 1. Kellgren JH, Lawrence JS: Radiological assessment of osteoarthrosis. Ann Rheum Dis 16: , Kellgren JH: The Epidemiology of Chronic Rheumatism. Second edition. Philadelphia, FA Davis, Plato CC, Norris AH: Osteoarthritis of the hand: agespecific joint-digit prevalence rates. Am J Epidemiol 109: , Acheson RM, Chan YK, Clemett AR: New Haven survey of joint diseases. XII. Distribution and symptoms of osteoarthrosis in the hands with reference to handedness. Ann Rheum Dis 29: , Roberts J, Burch TA: Prevalence of osteoarthritis in adults. Vital Health Statistics. National Center for Health Statistics, series 11, no. 15. US Dept. of Health, Education, and Welfare, Mikkelsen WM, Duff IF, Dodge HJ: Age-sex specific prevalence of radiographic abnormalities of the joints of

8 RADIOGRAPHIC GRADING SCALES FOR OA 1591 the hands, wrists and cervical spine of adult residents of the Tecumseh, Michigan, community health study area, s. J Chronic Dis 23: , Plato CC, Norris AH: Osteoarthritis of the hand: longitudinal studies. Am J Epidemiol 110:74&746, Busby J, Tobin JT, Ettinger W, Plato CC: Progression of osteoarthritis: significance of starting level, age, and length of follow-up (abstract). Gerontologist 26 (suppl): 141A, Altman RD, Fries JF, Bloch DA, Carstens J, Cooke TD, Genant H, Gofton P, Groth H, McShane DJ, Murphy WA, Sharp JT, Spitz P, Williams CA, Wolfe F: Radiographic assessment of progression in osteoarthritis. Arthritis Rheum 30: , Shock NW, Gruelich RC, Andres RA, Arenberg D, Costa PT, Lakatta EG, Tobin JD: Normal Human Aging: The Baltimore Longitudinal Study of Aging. Washington, DC, US Government Printing Office, Cohen J: Weighted Kappa: nominal scale agreement with provision for scaled disagreement or partial credit. Psychol Bull 70: , Fleiss JL, Cohen J: The equivalence of weighted Kappa and the intraclass correlation coefficient as measures of reliability. Educ Psych Measure 33: , Bartko JJ: The intraclass correlation coefficient as a measure of reliability. Psychol Rep 19: 1-1 1, SAS Institute Inc: Proc Lifetest, SAS User s Guide. Version 5. Cary, NC, SAS Institute, Lawless JF: Statistical Models and Methods for Lifetime Data. New York, John Wiley & Sons, 1982