RADIOGRAPHIC ASSESSMENT OF PROGRESSION

Transcription

1 1214 RADIOGRAPHIC ASSESSMENT OF PROGRESSION IN OSTEOARTHRITIS ROY D. ALTMAN, JAMES F. FRIES, DANIEL A. BLOCH, JOHN CARSTENS, T. DEREK COOKE, HARRY GENANT, PHILIP GOFTON, HARRY GROTH, DENNIS J. McSHANE, WILLIAM A. MURPHY, JOHN T. SHARP, PATRICIA SPITZ, CATHERINE A. WILLIAMS, and FREDERICK WOLFE We evaluated methods of grading radiologic progression of osteoarthritis (OA). Sets of radiographs were assessed separately by 8 readers who were blinded to the time sequence. Included were radiographs of patients with OA of the hands (24 pairs), hips (4 pairs), and knees (32 pairs). Most films were taken 124 months apart. The relative contribution of individual joints (such as particular interphalangeal joints), of observations (such as narrowing or spurs), and of a single joint compartment (such as the medial or lateral compartment of the knee) toward evidence of OA progression. From a conference sponsored, in part, by Sandoz Pharmaceuticals, April 25-28, 1985, San Francisco, CA. Supported in part by a grant from Sandoz Pharmaceuticals and by NIH grant AM to the American Rheumatism Association Medical Information System. Roy D. Altman, MD: Department of Medicine, University of Miami and Miami Veterans Administration Medical Center, Miami, Florida; James F. Fries, MD: Department of Medicine, Stanford University School of Medicine, Stanford, California; Daniel A. Bloch, PhD: Department of Medicine, Stanford University School of Medicine, Stanford, California; John Carstens, MD: East Hanover, New Jersey; T. Derek Cooke, MB: Department of Orthopedics, Queen s University, Kingston, Ontario, Canada; Harry Genant, MD: Department of Radiology, University of California Medical Center, San Francisco; Philip Gofton, MD: Department of Medicine, University of British Columbia, Vancouver, British Columbia, Canada; Harry Groth, MD: Portland, Oregon; Dennis J. McShane, MD: Department of Medicine, Stanford University School of Medicine, Stanford, California; William A. Murphy, MD: Mallinckrodt Institute of Radiology, Washington University School of Medicine, St. Louis, Missouri; John T. Sharp, MD: Tifton Medical Clinic, Tifton, Georgia; Patricia Spitz, RN, MS: Department of Medicine, Stanford University School of Medicine, Stanford, California; Catherine A. Williams: Data Bank Network, Stanford University, Palo Alto, California; Frederick Wolfe, MD: The Arthritis Center, Wichita, Kansas. Address reprint requests to Roy D. Altman, MD, University of Miami (D26), PO Box 1696, Miami, FL Submitted for publication September 3, 1986; accepted in revised form April 9, was evaluated, as well as the reliability and concordance of scoring, and the sensitivity in detecting change. In assessing OA of the hand, the greatest sensitivity was achieved by reading a single posteroanterior bilateral hand radiograph for narrowing, spurs, and erosions, and scoring 1 joints (second and third distal interphalangeal, second and third proximal interphalangeal, and trapeziometacarpal joints, bilaterally), using a scale of -3. In OA of the hip, a single anteroposterior radiograph assessed for joint space narrowing and cyst formation yielded the greatest sensitivity. In OA of the knee, an anteroposterior radiograph, with weight-bearing, assessed for narrowing, spurs, and sclerosis in both the medial and lateral compartments yielded the greatest sensitivity. These techniques will be useful to the investigator in designing experimental studies and to the clinician in determining the rate of disease progression in an individual patient. Standards for the radiographic identification of osteoarthritis (OA) were established by Kellgren and Lawrence in 1957 (1) and were accepted by the World Health Organization in 1961 (2). In several recent descriptions of the radiographic appearance of OA, clinical and pathologic correlations have been used to help differentiate OA from normal states and from other disease entities (3-6). However, there are no methods presently available to assess radiographic progression of OA by use of serial radiographs. Radiographs are frequently used in clinical trials of OA to establish inclusion criteria. However, such trials have not used radiographs to assess disease progression. The consistency of OA progression is not known. The reasons for not utilizing radiographs to evaluate progression in long-term trials are, perhaps, Arthritis and Rheumatism, Vol. 3, No. 11 (November 1987)

2 OA PROGRESSION 1215 based on the belief that OA progresses slowly, that medications would not alter the course of the disease, and/or that improvement or slowing of progression is not likely to be measurable. Moreover, standards for disease progression have not been established. With improved understanding of the pathogenic mechanisms in the evolution of OA, it is becoming apparent that OA cartilage damage may not always progress and. under some conditions, may even demonstrate repair (7-12). Thus, methods for judging progression and improvement are needed. Clinical techniques for evaluating patients with OA have been developed but are not sensitive (13,14). This study was designed to improve the accuracy of radiographic assessments and to determine their sensitivity in identifying radiographic evidence of OA progression. The availability of such methods might permit quantitative documentation in trials directed at slowing the progression of OA or inducing healing. The San Francisco Conference Group, the authors of this paper, established the following general objectives relevant to 3 common areas of the body that are involved with OA: the hand, the hip, and the knee. 1. Which radiographic findings of OA are most reproducible among a group of observers (inter-observer agreement, test-retest reliability)? 2. Which radiographic changes of OA best identify radiographic progression of disease? 3. Of those changes of OA that best identify progression, is there a single radiographic finding or a combination of radiographic findings that best evaluates sequential changes? 4. What is the comparative value of a single reader versus 3 readers (i.e., radiographic assessment by 1 reader versus 3 readers whose scores are averaged) in the reliability and sensitivity of the detection of OA progression? 5. Do radiologic scores derived from these readings correlate with selected clinical variables, such as age, sex, height, weight, and duration of disease? Three additional objectives relevant to specific joints were developed. 6. For the hands: What is the relative value of the wrist joints, individual interphalangeal (IP) joints, or combined joints of each digit in the assessment of OA progression? 7. For the knee: What is the comparative value of anteroposterior (AP) weight-bearing films, AP non-weight-bearing films, and lateral films? Is assessment of the medial femorotibial compartment superior to that of the lateral femorotibial and patellofemoral compartments? Is a qualitative measurement of the radiographic joint space as sensitive as a quantitative assessment using a millimeter scale? 8. For the hip: Do the femoral head diameter and height influence measurements of the joint space? Is a qualitative measurement of radiographic joint space as sensitive as a quantitative assessment using a millimeter scale? METHODS Selection of radiographs. Two members of the group (RDA and JC) selected radiographs from among those contributed by 4 rheumatologists. The radiographs were required to be of sufficient quality to allow interpretation of the joint region. The range of time between paired films was from 1 month to 8 years. Radiographs with nonarticular findings that would indicate radiographic film sequence (e.g., hip prosthesis) were excluded. A balance of disease seventy was achieved by the preselection of films exhibiting mild, moderate, severe, and very severe radiographic changes. Similarly, a balance of disease progression from each of those categories was attempted by preliminary examination of the second set of radiographs and selection of those that demonstrated no obvious change, mild worsening, moderate worsening, and severe worsening of disease. The clinical diagnosis of OA was based on symptoms, most often that of pain, associated with radiographic findings suggestive of OA in the absence of other rheumatic diseases. Clinical information regarding sex, age, weight, and duration of disease was also available (Table 1). Readers and training. The 8 readers4 rheumatologists, 2 orthopedists, and 2 radiologists (exclusive of the 2 Table 1. Characteristics of study population (at the time of the first radiograph)* Region examined Hand Knees Hips (n = 24) (n = 32) (n = 4) Age (years) 58.1 f f f 15.2 Sex (% women) Disease duration 6.8 f f f 12.6 (years) Height (inches) 63.8 f f f 3.8 Weight (pounds) f f f 31.6 Time between 4. f f f 1.9 first and second radiographs (years) * Except for the category of sex, values are the mean f SD.

3 1216 ALTMAN ET AL Table 2. Radiographic variables assessed in films of the hand, hip, and knee* Assessment Hand Hip Knee Overall More severe film (1-cm analog scale) Best 3 variables for progressiont Radiographic variable Joint space narrowing Osteophytes (spurs) -3-3 NA Sclerosis Cysts or erosions -3-3 NA Alignment -3 NA -3 MC widening -3 NA NA Buttressing NA -3 NA Bony attrition NA NA -3 Comparisonsf Joint space narrowing, NA Yes Yes qualitative/quantitative Joint-specific comparisons IP/IP Superior/medial Medialhatera1 Digit/digit Femoral/acetabular Femoral/tibial Digit/wrist Joint space, adjust- AP WB/AP non-wb ed for femoral APflateral head diameter * Films were scored using a 1-cm visual analog scale (1-1, where the midpoint indicates no difference between the 2 sets of films), or on a scale of -3 (where = normal, 1 = mild or 1-33% abnormal, 2 = moderate or 3446% abnormal, 3 = severe or 67-1% abnormal). NA = not applicable; MC widening = distance between the bases of the first and second metacarpals; IP = interphalangeal; AP = anteroposterior; WB = weight-bearing; non-wb = non-weight-bearing. t The 3 radiographic variables that the readers believed best demonstrated disease progression. f Qualitative/quantitative changes indicative of disease progression. members who selected the radiographs-ach had at least 12 years of clinical experience. Several had been involved in a similarly designed study evaluating radiologic progression in rheumatoid arthritis (15). Two weeks prior to the study, all readers reviewed a set of training radiographs; scoring sheets were included. Each set contained 3 pairs of radiographs from each of the joint areas studied. Before each reading session of the study, the evaluation technique was reviewed in a training session using the 3 paired training radiographs and 2 additional paired radiographs. Group scoring was used in the training session to familiarize readers with the forms that would be used in the trial and the meaning of the various ratings. Reading sessions. There were 24 pairs of hand films, 4 pairs of hip films, and 32 pairs of knee films. Films were read at 8 reading stations. On day 1, 3 paired hand films in posteroanterior (PA) view were placed at each station. On the morning of day 2, 5 paired AP films of the hip were at each station. In the afternoon of day 2,4 paired non-weightbearing AP and lateral knee films and, by separate randomization, 4 paired weight-bearing AP knee films were placed at each station. Readers remained at each station for 3 minutes. Each reading session lasted for 2.5 hours. Films for each patient were read blindly as pairs, in random sequence order, on double view-boxes. Handheld magnifying lenses were available for use in difficult assessments. High-intensity lights were juxtaposed to each view-box. At the end of each session, each reader reread 3 pairs of randomly assigned films for each of the joint areas so that retest reliability could be evaluated. Immediately after the radiographs were read, the scoring forms were collected and scrutinized for completeness, and all ambiguities were clarified with the reader. The recorded scores were independently tabulated by 2 members of the technical staff (PS and CAW). All disagreements in tabulation were resolved immediately by an arbitrator (DJM). Radiographic scoring. The features scored and the grading methods used are summarized in Table 2. All readers made an overall comparison of the pair of films examined. The extent of the difference between the 2 films was marked on a 1-cm analog scale, for which the center mark indicated no difference. The measurement from the center was graded for 5 cm to the left for the left hand film (graded 1-5) and for 5 cm to the right for the right hand film (graded 1-5). Deviation from the center by a measurement of more than 1 was used to identify the correct time sequence, as described below. For each radiograph, the readers also rated their impression of the most important variable in identifying progression of disease. A detailed form was used to record observations for each joint, with reference to specific abnormalities such as narrowing, spurs, sclerosis, etc. A -3 scale was used for all readings ( = normal, 1 = mild or 1-33% abnormal, 2 = moderate or 3446% abnormal, 3 = severe or 67-1% abnormal). Scoring of hand radiographs. A single PA radiograph of both hands was used for scoring. Twelve radiographic

4 OA PROGRESSION 1217 sites in each hand (total of 24 sites) were scored. These sites included 4 distal interphalangeal (DIP) joints, 4 proximal interphalangeal (PIP) joints, the IP joint of the thumb (first IP), the trapeziometacarpal (first CMC) joint, and the scaphotrapezoid (first carpal) joint at the base of the thumb. The distance between the bases of the first and second metacarpal joints was assessed for widening (MC widening). Each joint area (other than MC widening) was scored on a scale of -3 for joint space narrowing, osteophyte formation (spurs), erosions, and alignment. Scoring of hip radiographs. A single AP supine view of either the right or left hip was assessed. Readings included 11 scores, on a scale of -3, for superior and medial joint space narrowing, superior and inferior femoral osteophytes, superior and inferior acetabular osteophytes, femoral head and acetabular subchondral sclerosis, femoral head and acetabular subchondral cyst formation, and thickening of the medial femoral cortex (buttressing). The superior and medial joint spaces were measured (in millimeters) by a member of the technical staff (PS) who had no knowledge of the film sequences. The femoral head was measured (in millimeters) in a horizontal plane bisecting the fovea and the femoral neck. Scoring of knee radiographs. AP and lateral nonweight-bearing and AP weight-bearing views of either the right or left knee were evaluated. Readings of AP films included 9 scores, on a scale of -3. The medial and lateral compartments were scored for joint space narrowing, spur formation, loss of bone stock (attrition), and subchondral bony sclerosis. Abnormal alignment, by either varus or valgus deformity, was also scored. Readings of the lateral view films included 5 scores (same -3 scale) for patellofemoral joint space narrowing, patellar superior and inferior pole spur formation, sclerosis, and subcondylar erosions. The joint space of the medial and lateral compartments of AP films of the knee was measured (in millimeters) by a member of the technical staff (PS) who had no knowledge of the film sequence. A vertical line was drawn from the midfemoral medial and lateral condyles to the tibia1 plateau, and measurements were derived. Analytic methods. The technical staff calculated a combined score for each reader. This score reflected each radiographic variable for individual joints. Agreement between readers was evaluated by use of the intra-class correlation coefficient (inter-reader agreement), a calculation that utilizes intra-reader versus inter-reader variation and oneway analysis of variance (16). This correlation coefficient was used to assess concordance among readers for first films, second films, and progression scores. Inter-reader agreement was evaluated for both raw data and ranked data for each area scored (16). Inferences drawn from the intra-class correlation coefficient based on raw data depend on the readers scoring the films independently of each other and assume that the readers score with the same precision. Since different readers do not, however, routinely score with the same precision, ranked data were analyzed. By using ranked data, each reader s mean becomes the same, and variances among the 8 readers become equal, allowing one to accept the inter-reader agreement correlations at face value. The analysis sought to determine which scoring techniques would result in high inter-reader agreement for both raw and ranked data. The individual reader s ability to replicate previous reading scores was evaluated by test-retest reliability, using Student s f-test for matched pairs (r value and t value) on first films, second films, and progression scores. The r value is the usual correlation coefficient, as described above. The t value identifies systematic differences on rereading. If the first and second readings are interchangeable, then the t value is close to. If the first and second readings are not interchangeable, then the t value is statistically different from (e.g., <-2. or >+2.), and a learning effect has been demonstrated. The presence of a learning effect suggests that more training in the scoring technique is required prior to formal assessment of such variables. In the ideal situation, the r value will be close to unity and the t value will be close to. Based on radiographic change, a progression score may or may not reflect the correct time sequence of each radiographic finding (15). This provides an opportunity to test each scoring technique against a standard: correct identification of the most recent film as showing the worst disease. In these analyses, correct means that the reader s progression score identified the correct time sequence, tie means that the reader s progression score did not identify a time sequence, and wrong means that the reader s progression score incorrectly identified the actual time sequence. A score for progression was derived for each radiologic variable by combining the results from all readers and adding one-half of the number of film pairs that were tied to the number of film pairs that were correct. The score indicates the degree of sensitivity of a specific abnormality in detecting progression of disease. The effect of averaging 3 readers scores on the reliability and sensitivity of the detection of progression was evaluated with the Spearman-Brown prophecy statistic, which calculates the intra-class correlation coefficient for the mean of 3 readers progression scores (17). Using the 56 possible 3-reader averages for 8 readers on each film, the number of time-sequence identifications that were correct, tied, or wrong were tabulated. These were used to estimate the sensitivity score of 3-reader averages, as compared with the sensitivity of a single reader s score. RESULTS Descriptive information about the 3 study populations is listed in Table 1. No significant correlations between individual variables and radiographic progression could be found, except for the variable of weight. Individuals with OA of the knee were nearly 3 Ibs heavier than those with hand or hip disease. Moreover, excess body weight was associated with more rapid progression of knee disease (data not shown). Patients with hip disease tended to have had symptoms for a longer period of time before radiographic evaluation.

5 1218 ALTMAN ET AL Table 3. Readers assessment of the relative importance of specific findings in determining osteoarthritis progression* Hand films. Single radiographicfindings. Readers rated osteophytes (spurs) as the most important feature Second most Third most in identifying OA progression in the hand (Table 3). The important or important radiographic finding with the highest inter-reader agree- Most important tied for Ortiedfor ment for OA was spurs (.92 for digits, ranked data, or tied for second most third most second radiographs) (Table 4). Osteophyte scores most most important important important often identified the correct time sequence (score 158 for Hand digits) (Table 4). However, neither spurs nor any other Spurs Erosions Narrowing single variable yielded a particularly.~ high level of sensi- Alignment 13 8 MC widening 4 Hip Narrowing Cys1 s Spurs Sclerosis Buttressing Knee (AP WB) Narrowing Spur s Alignment Attrition Sclerosis I * Percentage of films in which a radiographic variable was judged most frequently, second most frequently, or third most frequently among the 8 readers. MC widening = distance between the bases of the first and second metacarpals; AP = anteroposterior; WB = weight-bearing. tivity in identifying OA progression. Erosions were the second most important feature in assessing progression of OA of the hand (Table 3). The best single radiographic variable for determining progression, by inter-reader agreement, was digital erosions (.69 raw data, progression score) (Table 4), with a good test-retest reliability correlation (.68 for digits and.88 for wrists) and an acceptable t value (-.98 for digits). Joint space narrowing was rated third in importance, and these scores performed well in identifying the correct time sequence, but inter-reader agreement was mediocre. Although assessment of digital alignment was not useful for detecting disease progression (.48, Table 4. Analysis of data obtained from scores on hand radiographs Single variables* Combination (progression)? MC wid All 24 Narrowing Spurs Erosions Alignment ening joints joints sites Inter-reader agreement Raw data First film of pairs Last film of pairs Progression Ranked data First film of pairs Last film of pairs Progression Test-retest reliability (progression) Correlation (r value) t value Identification of correct time sequence$ Correct Tie Wrong Score / * Values represent digitslwrists. MC widening = distance between the bases of the first and second metacarpals. t Narrowing plus erosions plus alignment. The lojoints assessed were (bilaterally) the second and third distal interphalangeal, second and third proximal interphalangeal, and the trapeziometacarpal. The 22 joints assessed were (bilaterally) the first interphalangeal, second through fifth distal interphalangeal, second through fifth proximal interphalangeal, trapeziometacarpal, and the scaphotrapezoid. t. Correct = reader s progression score identified the correct time sequence; tie = reader s progression score did not identify a time sequence; wrong = reader s progression score incorrectly identified the actual time sequence; score = degree of sensitivity in detecting osteoarthritis progression (see Methods for details).

6 ~~ ~ OA PROGRESSION 1219 ranked data, progression score), assessment of change or lack of change in alignment was consistent among readers (r value.73). However, the t value of 1.92 indicates that considerable learning was involved. Findings in individual joints or digits. The 3 joints that most often showed OA progression were the left fourth PIP, right third DIP, and left fourth DIP (data not shown). However, the single features which least often identified the correct time sequence were narrowing and spurs at the left fourth DIP or left fifth DIP joint, and erosions at the right fourth DIP. For a single joint, the features that were more often identified correctly as to time sequence were narrowing at the left first CMC (score 122.5), spurs at the left third PIP (score 122.5), and erosions at the left first CMC (score 118.5). In summary, no single feature, isolated IP joint, or wrist joint consistently demonstrated disease progression, and the fourth and fifth DIPS were less indicative than others. The 3 digits that showed the greatest disease progression were the right and left third and left fourth (DIP and PIP joints). The most consistent findings were in the third digit of both hands. The second digit of both hands was second most consistent in showing OA progression, according to the several methods examined. In 4 patients (17%), progression of OA of the hands could only be detected at the base of the thumb. Findings of combined variables. Simple added combinations of radiographic findings markedly increased the sensitivity scores. For example, interreader agreement reached.74 in raw data and.68 in ranked data when all readings were combined (Table 4). Four hundred possible combinations of hand joints and radiographic variables were examined (data not shown). Combined scores created by adding individual scores for narrowing, spurs, and erosions had high inter-reader agreement, good test-retest results, and high sensitivity. Adding alignment and/or MC widening scores did not improve inter-reader agreement or sensitivity. Scores that combined right and left readings were superior to those of either hand alone. The test-retest correlation for scores combining narrowing, spurs, and erosions for all joints was.73. The low t value demonstrated that a learning effect was not a major problem. Although readers thought that spurs were most important in assessing OA progression (also, as a single variable, it best identified time sequence), the combination of narrowing, spurs, and erosions identified the correct time sequence best. Combinations of radiographic variables, which included narrowing, spurs, and erosions, are listed in Table 4. The results from assessing 22 hand joints were Table 5. Analysis of data obtained from scores on hip radiographs Single variables Combination (progression)* Narrowing Spurs Sclerosis Cysts Buttressing N + Sc N + C N + Sp + C N + Sc + C All Inter-reader agreement Raw data First film of pairs Last film of pairs Progression Ranked data First film of pairs Last film of pairs Progression Test-retest reliability (progression) Correlation (r value) I value Identification of correct time sequencet Correct Tie Wrong Score * N = narrowing; Sc = sclerosis; C = cysts; Sp = spurs. t Correct = reader s progression score identified the correct time sequence; tie = reader s progression score did not identify a time sequence; wrong = reader s progression score incorrectly identified the actual time sequence; score = degree of sensitivity in detecting osteoarthritis progression (see Methods for details).

7 ~ 122 ALTMAN ET AL Table 6. Analysis of data obtained from scores on anteroposterior, weight-bearing radiographs of the knee* Inter-reader agreement Raw data First film of pairs Last film of pairs Progression Ranked data First film of pairs Last film of pairs Progression Test-retest reliability (progression) Correlation (r value) t value Identification of correct time sequence Correct Combination (progression) Single variables N + Sp + Sc Narrowing Spurs Attrition Sclerosis Alignment All (Meat) (MedLat) (MedLat) (MedLat) (VarNal) MedLat Med+ Lat MedLat Med+Lat / l l l.57.42l l / Tie Wrong / Score * Med = medial; Lat = lateral; Var = varus; Val = valgus; N = narrowing; Sp = spurs; Sc = sclerosis; Med+Lat = medial plus lateral. See Table 5 for explanation of correct time sequence categories practically the same as those from assessing all 24 sites. A simpler combination, with only 1 joints assessed, performed nearly as well as the more complicated combination in inter-reader agreement (raw data) and identification of correct time sequence, and this combination was actually superior on test-retest reliability (r value). Summary of findings in the hands. Considering all the methods examined, findings from the study set of 24 paired radiographs support confining such an examination to the radiographic assessments of narrowing, spurs, and erosions. Combinations were clearly more effective than were single variables. Using a combination of these 3 variables, the examination of 1 joints (second and third DIPS and PIPS, and first CMC of both hands) was as sensitive as examination of the larger set of 22 joints in identifying the correct time sequence, and was nearly as sensitive as using the scores from all sites examined. Test-retest correlation was actually slightly superior when the scores of fewer joints were used. The advantages of time, convenience, and nearly the same results lead us to suggest the use of the 1-joint assessment method. Hip films. Single radiographic findings. Readers rated joint space narrowing as most important in determining progression of OA of the hip (Table 3). Subchondral cyst formation was second most impor- tant. Osteophytes and subchondral sclerosis tied for third most important. Individual findings of subchondral cyst formation and joint space narrowing produced the highest inter-reader agreement (Table 5). Findings of combined variables. As in the hands, combining scores of radiographic variables was more effective than using the single variable scores. Combining narrowing with sclerosis or narrowing with cyst formation yielded nearly the same results. However, combining all 3 (narrowing, sclerosis, and cysts) did not appreciably improve the results. Qualitative grading was superior to quantitative joint space measurements in identifying the correct time sequence of radiographs. Adjusting for femoral head size did not improve the quantitative results. Summary of findings in the hip. Combined scores of joint space narrowing with sclerosis or joint space narrowing with cyst formation performed best. Combining joint space narrowing, cyst formation, and sclerosis did not improve the scores. Knee films. Readers rated narrowing as most important and spurs as next most important in assessing progression on both weight-bearing (Table 3) and non-weight-bearing AP views (data not shown). Malalignment was considered third most important. Findings on AP weight-bearing views. The best single variable for assessing progression, according to

8 OA PROGRESSION 1221 Table 7. Analysis of data obtained from scores on anteropostenor, non-weight-bearing radiographs of the knee* Single variables Combination Narrowing Spurs Attrition Sclerosis Alignment (progression), all (Meaat) (MedLat) (MedLat) (MedLat) (VarNal) MedLat Med+ Lat Inter-reader agreement Raw data First film of pairs Last film of pairs Progression U.6.75 Ranked data First film of pairs Last film of pairs Progression Test-retest reliability (progression) Correlation (r value) r value Identification of correct time sequence Correct 7811 %I Tie Wrong Score * See Table 5 for explanations of abbreviations and of correct time sequence categories. ranked inter-reader agreement, test-retest correlation, and identification of correct time sequence, were medial compartment narrowing, medial and lateral compartment spurs, varus malalignment of the knee, and subchondral sclerosis of the medial compartment (Table 6). The best scores were those combining narrowing, spurs, and sclerosis or those combining the scores for all variables. Since some patients had disease progression primarily in the lateral compartment, combining medial and lateral compartment data improved all ratings. For simplicity, we recommend using the 3-variable rather than the all-variable technique. There was no apparent value in recording femoral changes separately from tibia1 changes. Table 8. Analysis of data obtained from scores on lateral radiographs of the knee* Single variables Combination (progression) Narrowing Spurs Sclerosis N + Sc Sp + Sc Inter-reader agreement Raw data First film of pairs Last film of pairs Progression Ranked data First film of pairs Last film of pairs Progression Test-retest reliability (progression) Correlation (r value) t value Identification of correct time sequence Correct Tie I7 17 Score * See Table 5 for explanations of abbreviations and of correct time sequence categories.

9 1222 ALTMAN ET AL Table 9. Comparison of reading techniques, single reader s scores versus average of scores from 3 readers, for assessment of osteoarthritis ~romession* Inter-reader agreement Single reader Mean Of 3 read- Scoring techers Site nique Raw Ranked Raw Ranked - Hand 22 joints joints Hip N+C Knee(AP WB) N + Sp + Sc Identification of correct time sequence Single reader Mean of 3 readers Ct T W Sco Ct T W Sco * Ct =: correct (reader s progression score identified the correct time sequence); T = tie (reader s progression score did not identify a time sequence); W = wrong (reader s progression score incorrectly identified the actual time sequence); Sco = score (degree of sensitivity in detect:ing osteoarthritis progression) (see Methods for details). N = narrowing; C = cysts; AP = anteroposterior; WB = weight-bearing; Sp = spurs; Sc = sclerosis. See Table 4 for description of hand joint groups assessed. Findings on AP non-weight-bearing views. Similar to the results on the weight-bearing radiographs, the single best variables were medial compartment narrowing and both medial and lateral compartment osteophyte formation (Table 7). Combining medial and lateral compartments was not unequivocally superior to assessing the medial compartment alone, since lateral disease alone was quite uncommon. Combining 3 or 4 of the measures was almost as accurate as utilizing all measures. No combination of measures without weight-bearing produced results that were as reproducible as those from the 3-variable combination of observations with weight-bearing. Likewise, there was no apparent value in recording femoral changes separately from tibia1 changes. As might be expected, joint space narrowing was the single observation that lost sensitivity for analyzing progression when non-weight-bearing views were compared with weight-bearing views. Although disease abnormalities could be detected in both the weight-bearing and non-weight-bearing radiographs, the detection of progression was superior with the former, Findings on lateral non-weight-bearing films. The highest inter-reader agreement values were those for osteophytes (Table 8). However, the poor identification of time sequence with the use of the lateral radiograph of the knee renders this view an insensitive one for detecting OA progression, according to the measures examined. No combinations of variables improved the analysis of disease progression. Findings of quantitative measures. Qualitative grading was superior to quantitative joint space measurements in identifying the correct time sequence for both weight-bearing and non-weight-bearing views. Summary offindings in the knees. Combining the scores for narrowing, spurs, and sclerosis of both the medial and lateral compartments of the knee on AP weight-bearing films was the most sensitive and reproducible method for detecting OA progression. Non-weight-bearing views were less sensitive. Lateral knee radiographs performed poorly according to the measures examined, and this view was of little value in detecting disease progression. The effect of using multiple readers. Averaging 3 readers scores improved inter-reader agreement and the correct identification of true time sequence (Table 9). Inter-reader agreement increased by , using this method. All sensitivity scores improved, especially those for the hands and knees. DISCUSSION This trial was designed to develop techniques to evaluate the progression of OA as seen on radiographs. Serial radiographs of the hand, hip, and knee were reviewed blindly for findings that would most consistently determine disease progression. The radiographic findings evaluated in this trial were selected for their reported use and convenience in the diagnosis of OA (I,2). Numerous radiographic findings that are useful in helping to define the severity of disease and in differentiating OA from other disease states, as well as from normal states, have been described by other investigators (3-6). The radiographic findings under analysis here were selected for their potential to identify change, based on the clinical judgment of the investigators. It is recognized that radiographic findings may have little relationship to symptoms. Joint space narrowing, osteophytes, and subchondral changes may all occur independently of the clinical syndrome that we

10 OA PROGRESSION 1223 call OA. It is not known whether radiographic progression will cease if symptomatic OA is arrested or if effective repair or protective mechanisms are successfully activated. Despite these unknowns, when considering the techniques that are presently used, the radiograph is probably the best tool for the measurement of OA progression. Radiography is objective, and as demonstrated in this study, changes can be measured with an acceptable degree of reproducibility. The most useful radiographic variables differed according to anatomic site, with the results indicating that there should be different approaches toward the evaluation of OA progression at specific joint regions. Although it is not a conclusion from this trial, the differences at the various radiographic areas assessed may indicate that the pathogenic mechanisms of OA in these areas may be different. Readers were more consistent in detecting OA progression on sets of hand radiographs than on the available radiographs of the hip or knee. Attempts were made to remove redundant or noisy observations so that the scoring techniques would be simplified. Reading 22 joints in the hand provided the most reproducible results. However, use of a 3-reader averaged score showed that both the 22-joint and the IO-joint scoring methods were highly reliable and very sensitive in detecting change. The 1-joint assessment included the second and third DIP, second and third PIP, and the first CMC joints of both hands. The first CMC joint assessment was included because in onethird of the films examined, the first CMC demonstrated OA progression, and in half of that subset, it was the only joint that demonstrated OA progression. Therefore, nearly 2% of the radiographs demonstrated a subset of hand OA in which the base of the thumb was the primary area that demonstrated progression of disease. Test-retest results emphasized the need for proper training prior to the reading sessions. On assessment of hip films, if 3 readers were used and radiologic joint space narrowing was combined with subchondral cyst formation, then there was a high degree of inter-reader agreement and the correct time sequence was best determined. There have been prior reports indicating that spurs of the hip joint space (18) or subchondral cyst formation (19) occurring independently do not, of themselves, necessarily imply the presence of OA or its progression. We know of no dispute with the concept that spurs of the hip joint and subchondral cyst formation in combination are indicative of OA, and that changes in both these findings would indicate OA progression. Joint space narrowing with the presence of sclerosis and joint space narrowing with cyst formation were nearly equal indices for OA in this study. Since narrowing, sclerosis, and cyst formation in combination did not improve the sensitivity, it is our empiric recommendation that assessments of narrowing and sclerosis be used. Ahlback, in a comprehensive monograph (2), described the value of weight-bearing radiographs in the diagnosis of OA of the knee. Subsequently, Leach et a1 (21) reported that weight-bearing views demonstrated narrower medial compartments (of 2 mm or more) in 33% of 13 patients studied, as compared with the findings on non-weight-bearing films. This discrepancy ranged as high as 5 mm and included the identification of narrowing of the joint space on weight-bearing radiographs in the same patients in whom the joint space appeared tq be normal on non-weight-bearing radiographs. Our measurements of joint space did not confirm Leach and coworkers results regarding the value of quantification of joint space narrowing. Actual measurement was not necessary to establish this point and was not as sensitive or reproducible as visual appraisal that considered density, view, etc. However, our trial was smaller than that of Leach and colleagues, and our selection of patients was quite different. What we did confirm was that progression could best be detected on weightbearing radiographs with the use of multiple observations. Progression of OA of the knee was previously examined by Danielsson and Hernborg (22). Between 195 and 1954, they used non-weight-bearing knee radiographs to assess knee OA in 2,195 patients. These cases were reviewed in The authors selected patients who lacked structural change ; however, the authors did not define structural change. Nineteen patients had lateral radiographs, and progressive changes were described in I1 of them (58%). There was a qualitative increase in spur formation in 53 (61%) of 87 patients for whom adequate AP radiographs could be examined. Although they reported that they measured spur size, the actual data were not reported. Our work confirms that spur formation was of value in detecting the progression of knee OA. Subsequently, Hernborg and Nilsson (1 1) reexamined what was apparently the same group of radiographs, for sclerosis and bony attrition. They noted that the predominant area of disease was in the medial compartment (82%) and found that in 25% of the patients, the OA had progressed to involve the lateral

11 1224 ALTMAN ET AL compartment. Those authors thought that the disease progressed more in patients who had initial bicompartmental changes showing bony attrition than in patients without these changes. Solitary lateral compartment changes, when present, showed some progression in women. Attrition was not often identified in our trial, and it did not prove to be a good marker of progression in the overall group. There was about the same degree of disease progression in women as in men. As stated, lateral knee radiographs were of little value in assessing disease progression. Although not all findings in the lateral radiograph were described and assessed, identification of the correct time sequence is a global interpretation that includes nonrecorded findings, and these results were uniformly poor. It is noteworthy that although the presence of osteophytes is of greater value than joint space narrowing in the diagnosis of OA of the knee (23), joint space narrowing is, according to the results of the present sfudy, of greater value in the assessment of the progression of OA of the knee. The present study included assessments of available serial radiographs. Prospectively, the application of special techniques might be of more value. Use of magnification radiography has been suggested, particularly for the hand (24); use of several radiographic views has been proposed for the patella (25-27); use of quantitative assessments has been proposed for weight-bearing radiographs of the knee (28); and use of weight-bearing oblique views has been proposed for examination of the hip (29). Test-retest results emphasize the need to properly train radiograph readers. Training sessions reduced the learning experienced by the readers in this study. Such sessions are therefore highly recommended. As found in a prior study (IS), 3-reader averaged scores proved to be far superior to a single reader s score in this group of radiographs. The value of 3 readers scores is also emphasized by the following points. Improvement of the radiographic findings was not observed in this study. Scores rarely appeared to improve by more than 1 grade from the first radiograph to the followup radiograph. When such improvement was noted by one observer, it was not verified by other observers. Although the relationship of obesity to arthritis is not analyzed in this radiographic study, we found that higher body weight was present in the population with OA of the knee and was associated with more rapid progression of their disease. REFERENCES 1. Kellgren JH, Lawrence JS: Radiological assessment of osteoarthrosis. Ann Rheum Dis 16:494-51, The Epidemiology of Chronic Rheumatism, Atlas of Standard Radiographs. Vol. 2. Oxford, Blackwell Scientific, Cone RO, Resnick D: Radiographic evaluation of articular disorders: degenerative joint disease, Textbook of Rheumatology. Second edition. Edited by WN Kelley, ED Hams Jr, S Ruddy, CB Sledge. Philadelphia, WB Saunders, 1985, pp Norman A: Radiologic diagnosis, Osteoarthritis: Diagnosis and Management. Edited by RW Moskowitz, DS Howell, VM Goldberg, HJ Mankin. Philadelphia, WB Saunders, 1985, pp Resnick D, Niwayama G: Degenerative disease, Diagnosis of Bone and Joint Disorders. Vol. 3. Edited by D Resnick, G Niwayama. Philadelphia, WB Saunders, 1981, pp Forrester DM, Brown JC, Nesson JW: The Radiology of Joint Disease. Second edition. Edited by DM Forrester, JC Brown, JW Nesson. Philadelphia, WB Saunders, 1978, pp Dandy DJ: Abrasion chondroplasty. Arthroscopy , Bland JH: The reversibility of osteoarthritis: a review. Am J Med 6:16-26, Radin EL, Burr DB: Hypothesis: joints can heal. Semin Arthritis Rheum 13:293-32, Howell DS, Moskowitz RW: Introduction: symposium on osteoarthritis. Semin Arthritis Rheum (suppl) l:s96- S13, Hernborg JS, Nilsson BE: The natural course of untreated osteoarthritis of the knee. Clin Orthop 123: , Dieppe P: Osteoarthritis: are we asking the wrong questions? Br J Rheumatol 23: , Altman RD, Hochberg MC: Clinical trials in rheumatic diseases: degenerative joint disease. Clin Rheum Dis 9:681493, Doyle DV, Dieppe PA, Scott J, Huskisson EC: An articular index for the assessment of osteoarthritis. Ann Rheum Dis 4:75-78, Fries JF, Bloch DA, Sharp JT, McShane DJ, Spitz P, Bluhm GB, Forrester D, Genant H, Gofton P, Richman S, Weissman B, Wolfe F: Assessment of radiologic progression in rheumatoid arthritis: a randomized, controlled trial. Arthritis Rheum 29: 1-9, Kendall MC, Stuart A: The Advanced Theory of Statistics. Vol. 2. New York, Hafner Publishing, 1967, pp 32-34

12 OA PROGRESSION Sokal RR: Biometry. San Francisco, WH Freeman, 1969, pp Danielsson L: Incidence of osteoarthritis of the hip (coxarthrosis). Clin Orthop 45:67-72, Resnick D, Niwayama G, Coutts RD: Subchondral cysts (geodes) in arthritic disorders: pathologic and radiographic appearance of the hip joint. AJR 128:799-86, Ahlback S: Osteoarthritis of the knee, a radiographic investigation. Acta Radio1 [Suppll (Stockh) 277:7-72, Leach RE, Greg T, Siber FJ: Weight-bearing radiography in osteoarthritis of the knee. Radiology 97: , Danielsson L, Hernborg J: Clinical and roentgenologic stody of knee joints with osteophytes. Clin Orthop h9:32-312, Altman R, Asch E, Bloch D, Bole G, Borenstein D, Brandt K, Christy W, Cooke TD, Greenwald R, Hochberg M, Howell D, Kaplan D, Koopman W, Longley S 111, Mankin H, McShane DJ, Medsger T Jr, Meenan R, Mikkelsen W, Moskowitz R, Murphy W, Rothschild B, Segal M, Sokoloff L, Wolfe F: Development of criteria for the classification and reporting of osteoarthritis: classification of osteoarthritis of the knee. Arthritis Rheum 29: , Genant HK, Doi K, Mall JL, Sickles EA: Direct radiographic magnification for skeletal radiology: an assessment of image quality and clinical application. Radiology 123:47-55, Merchant AC, Mercer RL, Jacobsen RH, Cool CR: Roentgenographic analysis of patellofemoral congruence. J Bone Joint Surg 56A: , Laurin C, Labelle P: Radiological investigation of normal and abnormal patellae (abstract). J Bone Joint Surg 57B53, Ficat RP, Hungerford DS: Disorders of the Patello- Femoral Joint. Baltimore, Williams & Wilkins, 1977, p Wevers HW, Siu DW, Cooke TDV: A quantitative method of assessing malalignment and joint space loss of the human knee. J Biomed Eng 4: , Lequesne M, Samson M: A functional index for hip diseases: reproducibility value for discriminating drug s efficacy (abstract), Fifteenth International Congress of Rheumatology. Paris, June 21-27, 1981, pp