How many joints in the hands and wrists should be included in a score of radiologic abnormalities used to assess rheumatoid arthritis.

1326
HOW MANY JOINTS IN THE HANDS AND WRISTS
SHOULD BE INCLUDED IN A SCORE OF
RADIOLOGIC ABNORMALITIES USED TO ASSESS
RHEUMATOID ARTHRITIS?
JOHN T. SHARP, DONALD Y. YOUNG, GILBERT B. BLUHM, ANDREW BROOK,
ANNE C. BROWER, MARY CORBETT, JOHN L. DECKER, HARRY K. GENANT, J. PHILIP GOFTON,
NEAL GOODMAN, ARVI LARSEN, MARTIN D. LIDSKY, PEKKA PUSSILA,
AARON S. WEINSTEIN, and BARBARA N. WEISSMAN
Numerous methods for reading abnormalities of
rheumatoid arthritis in hand and wrist radiographs
have been proposed over the past several decades. There
are many differences among these methods, one of the
more striking of which is the variation in the number of
joints that are scored. In this study, we tested the
number of joints that need to be read in order to
represent abnormalities accurately and reproducibly,
using the scores of multiple observers. Thirteen
rheumatologists and radiologists each read a set of 41
Presented at a workshop sponsored by the Joe and Betty
Alpert Arthritis Center, Rose Medical Center, Denver, CO, November 28-29, 1983.
Supported by a grant from the Eli LiHy Company, Indianapolis, IN.
John T. Sharp, MD: Director, Joe and Betty Alpert Arthritis Center, Rose Medical Center, Denver, CO; Donald Y. Young,
PhD: Statistician, ARAMIS Data Bank Network, Stanford University, Palo Alto, CA: Gilbert B. Bluhm, MD: Coordinator, Clinical
Investigation, Rheumatology Research, Henry Ford Hospital, Detroit, MI; Andrew Brook, MB: Rheumatologist, Yarralumla, Australia: Anne C. Brower, MD: Department of Radiology, George
Washington University Medical Center, Washington, DC: Mary
Corbett, MB: Consultant Rheumatologist, Middlesex Hospital,
London, England: John L. Decker, MD: Director, Clinical Center,
NIH, Bethesda, MD; Harry K. Genant, MD: Radiologist, Department of Radiology, University of California Medical Center, San
Francisco, CA; J. Philip Gofton, MD: Rheumatologist, University
of British Columbia, Vancouver, BC: Neal Goodman. MD: Radiologist, Department of Radiology, St. Anthony Hospital, Denver, CO:
Arvi Larsen, MD: Spenshults Reumatikersjukhus, Oskarstrom,
Sweden: Martin D. Lidsky, MD: Chief, Rheumatology Section, VA
Medical Center, Houston, TX; Pekka Pussila, MD: Radiologist,
Department of Radiology, Reumasaation Sairaala, Heinola, Finland; Aaron S. Weinstein, MD: Department of Radiology, University of Cincinnati Medical Center, Cincinnati, OH: Barbara N.
Weissman, MD: Radiologist. Department of Radiology, Brigham
and Women’s Hospital, Boston. MA.
Address reprint requests to John T. Sharp, MD, Alpert
Arthritis Center, 4567 East Ninth Avenue, Denver, CO 80220.
Submitted for publication November 13, 1984; accepted in
revised form May 6, 1985.
Arthritis and Rheumatism, Vol. 28, No. 12 (Decemher 1985)
hand and wrist films from patients with rheumatoid
arthritis. Ten of 13readers scored 27 joints in each hand
and wrist; the other 3 readers scored fewer areas.
Fourteen combinations of joints were selected based on
the frequency of involvement and the technical adequacy of routine films in assessing a given area. After
testing these 14 different combinations, 1 scheme, which
included 17 areas read for erosions and 18 areas read for
joint space narrowing, was tested further. The correlation coefficients for 10 intraobserver scores derived from
this modified scheme compared with the original scores
were between 0.981 and 0.997. Seventy-one of 78
interobserver comparisons were better using the new
scheme than using the original scheme. These data
indicate that the simplified scheme, using a combination
of 17 joints to score erosions and 18 to score joint space
narrowing, more accurately reflects the extent of abnormalities perceived by a panel of experts than does the
original scheme. This abbreviated number of joints
shortens the amount of time required to read a set of
films and simplifies the scoring of films, since a number
of areas that are difficult to read are eliminated from
radiographic assessment.
Destruction of bone and cartilage is a regular
consequence of persistent, active synovitis in patients
with rheumatoid arthritis (RA). Because finger and
wrist joints are frequently involved in this disease, a
number of investigators over the past several decades
have proposed that an assessment of the severity of
erosions and cartilage loss in hand and wrist joints,
logically, would represent an index of the outcome of
this disease process (1-6). More recently, it has been
proposed that individual joints should be scored separately and the scores summed in order to accurately
SCORING JOINT ABNORMALITIES IN RA
represent the degree of changes in the hands and
wrists (7-10).
In November 1983, a group of radiologists and
rheumatologists collaborated in testing several different methods of measuring the radiologic changes in
these joints. There was generally good agreement
among experienced observers, even when using different methods (11). However, that study did not
resolve 2 important questions: 1) Does any single
method of scoring abnormalities of individual joints
give results that are more reproducible than other
methods do? and 2) Which joints in the hands and
wrists should be scored? This report addresses the
latter question.
In determining which joints and how many
joints should be included in the score, we considered:
the frequency of involvement of individual joints, the
time course of development of lesions in individual
joints with greater disease duration, the technical
factors which determine ease of reading definitive
lesions of a given joint or area, and the number of
joints required to sample adequately the abnormalities
found in a population of patients having a broad
spectrum of disease severity.
Frequency analysis was done on radiographic
data in the Alpert Arthritis Center American Rheumatism Association Medical Information System
(ARAMIS) data bank (12) to assess the contribution of
a given joint to the total radiologic score. Differences
in thle frequency of joint involvement, as perceived by
different observers, were determined using the scores
of 13 observers who read the same set of study films in
the project mentioned above (1 1). However, the films
in this study set were not randomly chosen, but were
selected in order t o represent a broad spectrum of
involvement, ranging from those films that were considered as demonstrating no diseases to those films
demonstrating very severe disease. Therefore, the
frequency data presented were based on analysis of
the film records in the Alpert Arthritis Center
ARAMIS data bank.
Cluster analysis was performed on the erosion
and joint space narrowing (JSN) scores, using the data
from the Alpert Arthritis Center to identify groups of
joints that contributed similar proportions to the total
scores. From this analysis, several schemes were
tested on the ARAMIS data. Finally, we chose I
modified scheme which was based on the results of the
cluster analysis, the frequency of joint involvement,
clinical judgment regarding ease and reliability of
reading specific joints, and some limited data on the
1327
long-term progression of radiologic lesions. We tested
this scheme, using the data from the collaborative
study of November 1983 (11). The results indicate that
this modified scheme, which eliminates areas that are
technically difficult to read and other areas that are
infrequently involved, provides an appropriate sample
of the joints in the hands and wrists that will accurately
represent the radiologic abnormalities of patients with
RA .
Figure 1. Original scheme. Twenty-seven areas of the hands and
wrists scored for erosions (large black dots) and joint space narrowing (parallel lines) in assessing the severity of rheumatoid arthritis.
1328
SHARP ET AL
METHODS
Source of data. Radiologic scores on 131 patients
with RA whose clinical, laboratory, and radiologic data had
been entered in the ARAMIS data bank from the Joe and
Betty Alpert Arthritis Center, Denver, CO were analyzed.
Selection of data on these 131 patients was based on whether
hand and wrist films were of sufficient technical quality to
allow a reading of all joints specified in the original scheme
for scoring erosions.
The films were scored according to the method
described by Sharp et a1 (7). Briefly, 27 areas of each hand
and wrist were scored for erosions and JSN (Figure 1).
Erosions were counted when discrete. Surface erosions,
Table 1. Description of schemes for scoring erosions and joint space narrowing (JSN) on radiographs
of the hands and wrists of patients with rheumatoid arthritis*
~
~
n
Original scheme?
Scheme A
Scheme B
Scheme C
Scheme D
Scheme E
Scheme F
Scheme G
Scheme H
Scheme I
Scheme J
Scheme K
Scheme L
Scheme M
Scheme N
Description
~~
~~
~
~
Areas read for JSN
Areas read for erosions
n
Description
54 8 DIPs, 10 PIPs, 10 MCPs. 10
54 8 DIPs, 10 PIPS. 10 MCBs, 10
CMCs, 12 intercarpal. 2 radioMCPs, 12 carpal bonest, 2 radius, 2 ulnar bones
carpal, 2 radioulnar joints
48 Removed from original scheme: R 48 Removed from original scheme: R
5th DIP, L 4th and 5th bIPs, L
5th DIP, L 4th and 5th DIPS, L
5th PIP, R 2nd MCP, R hamate
5th PIP, R 2nd CMC, R hamate-capitate joints
bones
45 Same as A, but also removed: R
45 Same as A, but also removed: R
2nd and 4th DIPs, R radius
2nd and 4th DIPs, R radiocarbones
pal joints
36 Same as B, but also removed: R
36 Same as B, but also removed: R
2nd MCP, R 3rd CMC. R mul2nd MCP, R 3rd MCB, R multangulars, R lunate, R triquetangular-navicular, R lunatetrum, R capitate, L lunate, L
triquetrum, R triquetrum-hacapitate. L hamate bones
mate. R capitate-navicular-lunate, L lunate-triquetrum, L
capitate-navicular-hate, L hamate-capitate joints
46 Same as original, but all 8 DIPS
46 Same as original, but all 8 DIPs
removed
removed
44 Same as original, but all 10 MCBs 44 Same as original, but all 10 CMCs
removed
removed
36 Same as original, but all 8 DIP5
36 Same as original, but all 8 DIPS
and all 10 MCBs removed
and all 10 CMCs removed
34 Same as F, but also removed: R
34 Same as F, but also removed: R
and L capitate bones
and L capitate-navicular-hate
joints
32 Same as G, but also removed: R
32 Same as G , but also removed: R
and L hamate bones
and L hamate-capitate joints
30 Same as H, but also removed: R
30 Same as H , but also removed: R
and L triquetrum bones
and L triquetrum-hamate joints
28 Same as I , but also removed: R
28 Same as I, but also removed: R
and L h a t e bones
and L lunate-triquetrum joints
20 All joints excluded except the 10
20 All joints excluded except the 10
PIPs and 10 MCPs
PIPs and 10 MCPs
27 All joints in L hand excluded;
27 All joints in the L hand excluded
joints and areas of R hand in
original scheme were included
27 All joints in the R hand excluded
27 All joints in R hand excluded
35 Included were: 10 PIPs, 10 MCPs, 36 Included were: 10 PIPs, 10 MCPs,
Rand L 1st MCBs. R and L
R and L 3rd, 4th. 5th CMCs, R
multangulars (as I unit), R and
and L multangular-navicular, R
L navicular, R and L lunate, R
and L lunate-triquetrum, R and
and L triquetrum (and pisiL capitate-navicular-lunate, R
form), R and L radius, R and L
and L radiocarpal, R and L raulnar bones
dioulnar joints
* DIP = distal interphalangeal; PIP = proximal interphalangeal; MCB = metacarpal base; MCP =
metacarpophalangeal; CMC = carpometacarpal; R = right; L = left.
t See Figure 1 for graphic representation.
t The greater and lesser multangulars are usually superimposed and are therefore read as 1 unit; for the
same reason, the triquetrum and pisiform are read together.
SCORING JOINT ABNORMALITIES IN RA
1329
1.1
2nd
PIP
1i iilhl
,tn
A
n
MCP
MCB
L
T
CARPALS
P
~
C
RAD. U L N A
Figure 3. The frequency of erosions in individual joints of the hands
and wrists, as determined by reader I, utilizing data from 131
patients in the Alpert Arthritis Center American Rheumatism Association Medical Information System data bank. R = right (open
bars); L = left (closed bars); DIP = distal interphalangeal; PIP =
proximal interphalangeal; MCB = metacarpal bases; MCP =
metacarpophalangeal; M = multangulars; N = navicular; L =
lunate; ‘TP = triquetrum-pisiform; H = hamate; C = capitate; RAD.
= radius.
Figure 2. Scheme N . Seventeen areas scored for erosion (large
black dots), and 18 areas scored for joint space narrowing (parallel
lines) in assessing the severity of rheumatoid arthritis.
which are rarely discrete, were scored according to the
surface area involved. When the erosive process resulted in
extensive loss of bone from more than one-half of either
articulating bone in a metacarpophalangeal (MCP) or proximal interphalangeal (PIP) joint or a carpal bone, the erosion
was scored as 5. JSN was given a score of 1 if it was focal,
2 if narrowing was 4 0 % of the original joint space, and 3 if
>SO% of the original joint space; ankylosis was scored as 4.
The data for each joint were entered into the Alpert
Arthritis Center ARAMIS data bank (12). Erosion scores,
JSN scores, and total scores, which were the sum of the
erosion and JSN scores, were calculated. The frequency of
involvement of individual joints and correlation coefficients
were obtained utilizing TOD software programs (13).
Since the radiologic data on the films at the Alpert
Arthritis Center represent radiologic scores assigned by a
single reader, the frequency distribution of specific joint
involvement and correlations between scores obtained with
the original scheme and scheme N (which will be defined
later) were compared for 13 different observers, using data
obtained in the collaborative study (11). In that study, 41
films of RA patients were selected to represent a broad
distribution of disease severity, ranging from those films that
were judged by multiple observers to be within normal limits
to films that showed advanced disease. Films were read by
13 different observers using 5 different methods of scoring
radiologic abnormalities. Data from this study were entered
in a Tandem computer at the University of Colorado Health
Sciences Center, Denver, CO.
Statistical analysis. The Pearson correlation coefficient was used to compare radiologic scores obtained by
different schemes. These correlation coefficients were calculated using TOD (13), SPSS (14), or SAS (1s) software
programs. Cluster analysis was done on the Alpert Arthritis
Center data bank, using a BMDP program (16). These cluster
1330
SHARP ET AL
Table 2. Correlation coefficients comparing candidate schemes
with the original scheme for scoring erosions and joint space
narrowing (JSN) seen on radiographs of rheumatoid arthritis
patients*
I-
DIP
I
n
MCB
Candidate
scheme
No. of joints
or areas
Erosions
JSN
A
B
C
D
48
45
36
46
44
36
34
32
30
28
20
27
21
35
0.9971
0.9938
0.9627
0.9899
0.9984
0.9860
0.9844
0.9799
0.9723
0.9595
0.8108
0.9692
0.9736
0.9802
0.9969
0.9939
0.9778
0.9891
0.9927
0.9829
0.9781
0.9714
0.9657
0.9559
0.9028
0.9584
0.9621
0.9816
E
F
G
H
I
J
K
L
M
N
* Data are from the Alpert Arthritis Center American Rheumatism
Association Medical Information System data bank. Correlation
coefficients were calculated between t h e original erosion or JSN
scores and scores derived from the modified schemes.
1.1
2nd
3rd
MCP
4th
6lh
M
N
L
T
P
CARPALS
n
c
RAD
RAD.
CARPAL ULNAR
Figure 4. The frequency of individual joint involvement by joint
space narrowing in the hands and wrists, as determined by reader 1,
utilizing data from 131 patients in the Alpert Arthritis Center
American Rheumatism Association Medical Information System
data bank. M = multangular-navicular joint: N = navicular-hate
joint; L = lunate-triquetrum joint; TP = triquetrum-pisiform-hamate joint; H = hamate-capitate joint; C = capitate-navicular-hate
joint; RAD. CARPAL = radiocarpal joint; RAD. ULNAR =
radioulnar joint. See Figure 3 for other abbreviations.
following 4 areas in the wrist were scored as 1 if abnormal:
I ) The first metacarpal base was examined for erosions, and
the third, fourth, and fifth carpometacarpal (CMC) joints
were examined for narrowing; 2 ) The multangulars,
navicular, lunate, and triquetrum and pisiform were examined for erosions, and t h e multangular-navicular,
lunate-triquetrum, capitate-navicular-hate were examined for narrowing; 3) The radius was examined for erosion,
and the radiocarpal joint was examined for narrowing; 4)The
ulna was examined for erosion, and the radioulnar joint was
examined for narrowing.
RESULTS
analyses were done separately on erosion scores and JSN
scores.
Schemes. Based on the cluster analysis, frequency of
joint involvement, and experience regarding technical factors in reading specific joints, various combinations of joints,
hereafter called schemes, were selected for further testing.
These schemes are given in Table 1 . One group of schemes
was designed by subtracting progressively more joints from
groups identified by cluster analysis. Another group of
schemes was formed by removing groups or combinations of
groups of joints.
Scheme K excluded all joints except the 10 MCPs
and 10 PIPS; scheme L excluded all joints in the left hand;
scheme M excluded all joints in the right hand; scheme N
excluded all distal interphalangeal (DIP) joints and selected
areas in the wrists. Included in scheme N were 17 areas
scored for erosions and 18 areas scored for JSN in each hand
(Figure 2).
The method used for obtaining the count of abnormal
joints defined an abnormal joint as one which showed an
erosion or JSN. Each MCP, each PIP, and each of the
Utilizing data from 131 patients included in the
Alpert Arthritis Center ARAMIS data bank, the frequencies of individual joint involvement by erosions
and JSN were determined; these are depicted in Figures 3 and 4. Abnormalities were most frequently
observed in t h e MCP, PIP, radiocarpal, and
intercarpal joints, and the third, fourth, and fifth CMC
joints. To test whether multiple observers judged the
same areas to be involved, we tabulated the frequency
of involvement by erosions and JSN as assessed by 11
readers who participated in the collaborative study.
The frequency of abnormalities in individual joints,
based on a single reader, was below the median
frequency of involvement derived from the scoring by
I f observers, with the difference from the median
ranging from &29% (data not shown). Thus, it can be
inferred that the frequency data depicted in Figures 3
and 4 represent conservative estimates.
1331
SCORING JOINT ABNORMALITIES IN RA
Table 3. Correlation coeficients and ratios between total scores
derived from scheme N and scores derived from the original scheme
Source of scores
n
AAC ARAMIS data bank* 131
41-film set, reader no.t
I
41
3
41
4
41
5
41
7
41
8
39
9
41
10
12
13
15
41
41
41
41
Correlation
coefficient Ratio
Range
0.9891
0.9228
0-1.0
0.9951
0.9971
0.9923
0.9951
0.9913
0.9937
0.9845
0.9936
0.9920
0.9807
0.9806
0.8778
0.9724
0.8121
0.8469
0.6679
0.9158
0.8044
0.7667
0.8382
0.6557
0.8372
0-1.0
0-1.0
0.5-1.0
0-1.0
0.188-1.0
0.677-1.0
0.444-1 .0
0.364-1.0
0.333-1.0
0.368-0.8201
0.659-1 .0
* Data were derived from 131 films in the Alpert Arthritis Center
American Rheumatism Association Medical Information System
(AAC ARAMIS) data bank, read by a single observer.
t Data were derived from a collaborative project in which 1 1
observers read a standardized set of 41 hand and wrist study films.
Readers 2 and 6 were not included here since the areas they read to
obtain the original scores were almost identical to the areas included
in the modified scores of the other readers. Reader 1 1 was not
inclulded since he did not complete the reading of the set. Reader 14
was not included because he used an electronic device for reading
joint space narrowing.
After the cluster and frequency analyses to
reduce the number of joints to be read were performed, the various schemes selected were tested by
corriparing the total scores derived for each scheme
with the total scores taken from the original scheme.
Correlation coefficients were calculated, and the ratio
of the 2 scores (score from the modified scheme
divided by the score from the original scheme) was
determined using the Alpert Arthritis Center data. The
correlation coefficients are shown in Table 2.
Taking into consideration the data derived from
the cluster analysis and various candidate schemes,
the frequency of involvement, and the ease of reading
a joint, which is primarily related to the projection of
the ,area on standard films, 1 modified scheme (scheme
N) was chosen for further testing (Figure 2). In scheme
N , 17 areas in each hand and wrist were read for
erosions, and 18 areas were read for JSN. A total score
was derived by summing all of the erosion and JSN
scores for these areas.
The suitability of this total score as a representation of the severity of radiologic changes in all of the
joints of the hands and wrists was tested by calculating
a correlation coefficient, comparing the total scores
calculated for scheme N and the total scores obtained
with the original scheme. This correlation coefficient,
calculated on the Alpert Arthritis Center ARAMIS
Table 4. Comparison of joint groups included in and excluded
from scheme N*
Frequency of
involvement, 9?
(range o f .
medians)
Joints excluded
Erosions
DIPs
MCBs
Carpals
Mean all areas
Joint space narrowing
DIPs
CMCs
Intercarpals
Mean all areas
Joints included
Erosions
MCPs
PIPS
Wrists
Mean all areas
Joint space narrowing
MCPs
PIPS
Wrists
Mean all areas
Coefficient of
variation
(SD + mean)
5-13.0
0-1 8.5
11-23.5
14.4
78-145
70-2 17
74-1 I 1
104.8
5-22.0
10-38.5
17-47.0
24.4
80-124
44- 144
44-78
84.6
27.0-66
15.0-39
12.5-53
37.5
22-69
34-77
27-86
51.1
17.0-56
8.5-36
17.0-55
32.1
30-89
56-123
27- 102
59.0
* Data are from scores of 9-1 I observers who read the 41-film study
set of nonrandomly selected films. DIP = distal interphalangeal;
MCB =. metacarpal base; CMC = carpometacarpal; MCP = metacarpophalangeal.
data, was 0.9891 (Table 3). The ratio of the scheme N
total score to the original score was 0.9228. This ratio
indicates that 4%
of the original total score was
contributed by the areas eliminated in scheme N.
Scheme N was tested further, using the scores
of the 41-film study set read by 11 observers. Intraobserver correlation coefficients, comparing the modified scores, based on scheme N , with the scores
derived from the original scheme, were above 0.9800
for 11 readers (Table 3). The ratios of the modified
scores to the original scores varied between 0.6557 and
0.9724 and averaged 0.8177. These ratios indicate that
Table 5. Interobserver correlation coefficients, original scores
compared with scheme N scores
Range
Original total scores
0.736-0.956
Scheme N total scores* 0.752-0.969
~~
*
~
~ 0 . 8 5 0 20.900
50
60
31
38
20.950
4
6
~
In 71 of 78 comparisons, interobserver coefficients were higher
when total scores derived from scheme N were used than when the
original total scores were used.
1332
SHARP ET AL
Table 6. Ratios of total scores derived from scheme N to total scores derived from original scheme,
by patient group stratified by disease severity*
Range
Patients with original
of
scores >O
All patientst
originaI
total
Mean
Mean
Patient g r o w
score
n
ratio
SEM
n
ratio
SEM
Entire group
Greatest severity
Intermediate severity
Least severity
0-151
>21
4-2 1
<4
110
44
40
26
0.9080
0.8756
0.9172
0.9487
0.0161
0.01695
0.0304
0.0401
131
44
40
47
0.9228
0.8756
0.9172
0.9716
0.01385
0.01695
0.03041
0.02228
* Data are from the Alpert Arthritis Center American Rheumatism Association Medical Information
System data bank.
t Technically, 0 divided by 0 is a nonentity. However, for purposes of this calculation, since a 0
modified score found with a 0 original score is an identical finding, the ratio was considered to be 1 for
those cases.
1 observer, on reading the selected films in the study
set, attributed an average of 35% of abnormalities for
the entire set to those areas not included in scheme N,
while another observer attributed <3% of abnormalities to those areas; the mean was 18%.
The variability in interpreting these areas is
further illustrated in Table 4, which compares the
frequency and coefficients of variation of erosion
scores and JSN scores for the areas retained in scheme
N with those areas eliminated. The frequency of
involvement was relatively low for the DIP joints, but
much higher for most of the areas in the wrists.
However, the coefficients of variation were much
greater for the eliminated areas than for the areas
included in scheme N.
Interobserver correlation coefficients were calculated for the scores of all 13 readers on the 41-film
study set, and showed that 71 of 78 correlations were
higher when the total scores based on scheme N were
Table 7. Correlation coefficients of total scores derived from
scheme N compared with total scores derived from original scheme,
by patient group, stratified by disease severity*
Patient group
n
Correlation
coefficient
Entire group
Greatest severity
(total score >21)
Intermediate severity
(total score 4-21)
Least severity
(total score <4)
Patients with disease duration
>I0 years
131
0.9891
44
0.9740
40
0.9450
47
0.9662
30
0.9908
* Data are from the Alpert Arthritis Center American Rheumatism
Association Medical Information System data bank.
used than when the total scores based on the original
scheme were used (Table 5 ) .
Since scheme N might perform well for examination of data from films in the study set, which
represented a broad spectrum of disease severity, yet
might not be an accurate instrument for measuring
abnormalities in patients with more advanced disease,
the data from the Alpert Arthritis Center ARAMIS
data bank were stratified into 3 groups of increasing
disease severity. As expected, in patients with the
most severe disease, the frequency of erosions and
JSN was greater in the joints excluded from the new
scheme; this is shown by the mean ratios of modified
total scores to original total scores. These ratios,
Table 8. Correlation coefficients and ratios between counts of
involved joints for areas in scheme N and original total scores*
Source of scores
n
AAC ARAMIS data bank 131
41-film set, reader no.
I
41
4
41
5
41
6
40
7
41
8
39
9
41
10
41
12
41
13
41
15
41
Correlation
coefficient Ratiot
0.8887
0.5791
0.9392
0.9488
0.9348
0.8735
0.8752
0.8825
0.8912
0.8673
0.9451
0.8866
0.8621
0.4753
0.3192
0.4317
0.3668
0.2697
0.4256
0.3213
0.2943
0.4317
0.2348
0.2957
Range
0-1.0
0-1.0
0.141-1.0
0-1.0
0-0.625
0.125-0.833
0-1.0
0.108-1.0
0.125-0.563
0.154-1.0
0.129-0.452
0.144-0.667
* A joint was counted as involved if either erosion or joint space
narrowing was present. AAC ARAMIS = Alpert Arthritis Center
American Rheumatism Association Medical Information System
(131 films, read by a single observer).
t Ratio was calculated between joint count and original total score
(count + total score).
SCORING JOINT ABNORMALITIES IN RA
calculated for the 3 groups, demonstrated that as
disease severity increased, a decreasing proportion of
the ‘totalscore was represented in the scheme N scores
(Table 6 ) . Nevertheless, the correlation coefficients
were similar for each disease-severity subset (Table 7).
To test whether the scoring method could be
further simplified, a count of abnormal joints was
performed, assigning a value of 1 for each joint or area
that showed either erosion or JSN. The joints and
areas counted were those used in scheme N. Areas in
the wrists were divided into 4 groups so that the total
weight assigned to each wrist, if a minimum of 1 bone
orjoint in each group was involved, was 4. Correlation
coefficients between these counts of involved joints
and the original total scores were calculated on the
intr,aobserver data of 1 1 readers. These correlation
coeficients varied between 0.8621 and 0.9488. The
ratio of counts to total scores ranged between 0.2348
and 0.4753, with a mean of 0.3515 (Table 8).
DISCUSSION
Reading 54 areas separately for erosions and
joint space narrowing, i.e., 27 areas in each hand and
wrist, can be tedious and time consuming. Therefore,
we have explored the possibility of simplifying the
procedure by selecting a smaller number of areas. The
prolblem is primarily one of sampling; namely, do the
areas chosen accurately reflect the extent of changes
in the entire hand and wrist? This is not a simple
question. For example, there are more than 8 million
combinations that can be formed by selecting any 17 of
the 27 total joints, and an even larger number of
combinations can be formed by selecting fewer than 17
joints. Therefore, after preliminary analysis by several
methods, we arbitrarily selected a scheme. Cluster
analysis was performed to identify groups of joints that
contributed a similar proportion of the total scores.
Frequency analysis of joint involvement was performed to determine which joints contributed the most
to the radiologic scores. Based on this information and
on Iclinical judgment regarding the technical problems
involved in reading specific joints, a series of modified
schemes was tested by calculating correlation coefficients between newly derived scores and the original
score.
Finally, scheme N , which eliminated more than
one-third of the original joints scored, was chosen for
further study. This modified scheme was then tested
on 2 data sets, the Alpert Arthritis Center ARAMIS
data bank, representing radiologic scores on films from
1333
131 patients read by a single observer, and another set
of data representing radiologic scores of 13 different
observers on a study set of 41 selected films. Correlation coefficients were used to test how well scheme N
represented the overall score. Although no standards
have been established as to what is a satisfactory level
of reproducibility, intraobserver correlation coefficients between scores derived from scheme N and the
original scheme were above 0.9800, indicating a high
degree of similarity of the scores derived from the 2
schemes.
Furthermore, correlation coefficients between
observers were improved when the modified scheme
was used, suggesting that those areas which were
excluded from scheme N were read with less consistency among the 13 observers than those areas which
remained. This result is not unexpected since the
original scheme included DIPS with the understanding
that the observer would distinguish between erosive
osteoarthritis and the lesions of RA. By eliminating
these joints, the inconsistencies that might arise
among observers’ interpretations were also eliminated.
The same argument applies to several areas in
the wrists that are difficult to read because projection
in standard posteroanterior and oblique films is not
consistent; for example, the navicular-lunate joint and
the hamate-capitate joints are rarely “opened up” in
standard films. That, in fact, there was greater inconsistency in reading the areas of the original scheme
that were not included in scheme N is demonstrated by
the greater coefficients of variation for many of those
areas (Table 4).
Comparison of scores derived by the 2 schemes
for patients who had been stratified into subsets according to the severity of radiologic evidence of involvement, showed that a higher proportion of scores
were derived from those areas eliminated by scheme N
in patients with the most severe disease. This observation suggests 2 possible interpretations. First, the
DIP, selected metacarpal bases, CMC, and intercarpal
joints may not be involved proportionally, compared
with MCP and PIP joints in patients with varying
severity of disease. In this case, the consequence of
eliminating these areas from the new scheme would be
to reduce the sensitivity of the radiologic scores in
deteci ing progression of disease among patients whose
disease was already severe at the baseline observation. Alternatively, observers may be reluctant to
attribute a given erosion or narrowed joint space in the
DIP joints and, perhaps, even in the CMC and some
1334
intercarpal joints, to RA if there is no involvement or
if there is limited abnormality in the MCP and PIP
joints. In this case, the modified scheme should increase the specificity and precision of scoring. It is
noteworthy that Larsen et a1 (8) and Genant (9) do not
include DIP joints and they limit the wrist areas
included in their schemes; Bluhm et a1 have also
eliminated the DIP joints in their scheme (10).
Some have proposed that eliminating further
areas and using an even smaller number of joints
would be desirable. For example, Bluhm et a1 have
suggested reading only 16joints in each hand and wrist
for narrowing, although they proposed reading 23
areas for erosions (10). At present, there are serious
reservations about oversimplifying the scheme. If sensitivity in measuring the progression of disease is of
paramount importance-for example, in a drug trialany further reduction in the number of joints read
might significantly reduce the sensitivity of detecting
change. More data on longitudinal studies of untreated
patients or patients treated with the usual and customary regimens will be required to assess how extensive
this loss in sensitivity might be.
Others have proposed that the wrist should be
scored as a single unit. Based on the number of
articulations in the wrist, this seems inappropriate.
There are 15 bones involved in the articulations in the
wrist joint. Considering that 10 scores are derived
from the 10 bones represented in the articulations of
the 5 MCP joints and 10 scores from the 10 bones of
the 5 PIP joints in each hand, representing the wrist
with 2 scores-a single score for erosions and another
score for narrowing-would be grossly disproportionate. The modified scheme proposed here, which was
derived by selecting areas that are frequently involved
and are usually well projected in standard films, results
in 15 scores for the wrist, which is 1.5 times the
number of scores for the MCPs and for the PIPS. Thus,
scheme N represents a balance that reflects the number of joints in each area.
Use of this modified scheme should reduce the
time required to score hand and wrist films. Although
more than 80% of the original total score was derived
from the areas included in scheme N, elimination of
those areas which are difficult to read because they are
often technically poorly projected, and of other areas
which may require interpretation as to the nature of
the changes, should speed the reading process significantly.
Again, it should be emphasized that selection of
joints to read is a highly arbitrary process, primarily
because of the extremely large number of possible
combinations. Proof that one selection is the “most
suitable” scheme for multiple circumstances is not
within the realm of practical possibility. It should
therefore be recognized that the choice of scheme N ,
which incorporates 70 observations (34 erosion scores
and 36 JSN scores), is a compromise to allow for a
more rapid reading, using a method which demonstrates better interobserver agreement than the original scheme, but still incorporates enough areas to be
sensitive to minor progression of disease. Proof that
eliminating approximately one-third of the areas does
not seriously impair the sensitivity of the method for
detecting disease progression will require field testing
in a therapeutic trial or study of the natural history of
RA .
A more abbreviated scheme, which in essence
reduces the scale and therefore the sensitivity of
detecting change, might be adequate for some purposes, such as describing outcome in general terms for
a large, population-based sample, but might be an
extremely expensive shortcut in a drug trial if the
number of patients studied had to be greatly expanded
or the length of the study had to be extended. For
example, the joint count method described here, although clearly not as sensitive to change as the scoring
method, might be appropriate for the general followup
of patients not in special studies, but it would be
inappropriate for studies that would benefit from the
more precise data acquired with the expanded scale
provided by the scoring system.
Finally, it should be pointed out that this study
does not deal with methods of scoring individual
joints. That is an issue that has not yet been resolved.
Steinbrocker 0, Traeger CH, Batterman RC: Therapeutic criteria in rheumatoid arthritis. JAMA 140:659-665,
1949
Atlas of Standard Radiographs of Arthritis: The Epidemiology of Chronic Rheumatism. Vol. 2. Philadelphia,
FA Davis, 1963
Berens DL, Lin RK: Roentgen Diagnosis of Rheumatoid
Arthritis. Springfield, IL, Charles C. Thomas, 1969
Engel A, Roberts J, Burch TA: Rheumatoid Arthritis in
Adults: United States, 1960-1962, National Health Services Publications, series 1 1 , no. 17. Government Printing Office, 1966
SClORING JOINT ABNORMALITIES IN RA
5. Larsen A: A radiological method for grading the severity
of rheumatoid arthritis (thesis). University of Helsinki,
Finland, 1974
6. Severs K: The rheumatoid factor in definite rheumatoid
arthritis. Acta Rheumatol Scand [Suppl] 9: 1-28, 1965
7. Sharp JT, Lidsky MD, Collins LC, Moreland J: Methods
of scoring the progression of radiologic changes in
rheumatoid arthritis: correlation of radiologic, clinical
and laboratory abnormalities. Arthritis Rheum 14:706720, 1971
8. Larsen A, Dale K , Eck M: Radiographic evaluation of
rheumatoid arthritis and related conditions by standard
reference films. Acta Radio1 [Diagn] (Stockh) 18:481491, 1977
9. Genant HK: Methods of assessing radiographic change
in rheumatoid arthritis. Am J Med 74:35-47,1983
10. Bluhm GB, Smith DW, Mikulashek WM: A radiologic
method of assessment of bone and joint destruction in
rheumatoid arthritis. Henry Ford Hosp Med J 31:
152-161, 1983
1335
11. Sharp JT, Bluhm GB, Brook A, Brower AC, Corbett M,
Decker JL, Genant HK, Gofton JP, Goodman N , Larsen
A, Lidsky MD, Pussila P, Weinstein AS, Weissman BN,
Young DY: Reproducibility of multiple-observer scoring
of radiologic abnormalities in the hands and wrists of
patients with rheumatoid arthritis. Arthritis Rheum
28 16-24, 1985
12. The Computer Committee of the American Rheumatism
Association: A uniform database for rheumatic diseases.
Arthritis Rheum 22:1029-1033, 1979
13. McShane DJ, Harlow A, Kraines RG, Fries JF: TOD: a
software system for the ARAMIS data bank. Computer
12 34-40, 1979
14. Nie HH, Hull CH, Jenkins JG, Steinbrenner K , Bent
DH: Statistical Package for the Social Sciences. First
edition. New York, McGraw-Hill, 1975
IS. Helwig JT, Council KA: SAS Users Guide. Cary, NC,
SAS Institute, Inc., 1979
16. Dixon WJ, Brown M: Biomedical Computer Programs.
Berkeley, University of California Press, 1979