close

Вход

Забыли?

вход по аккаунту

?

art.33397

код для вставкиСкачать
LETTERS
321
believe that it is an epiphenomenon of the processes involved
in the development of RA . . . there was no question that the
criteria developed in 1987, based on patients with established
RA, required updating.”
We believe the term “rheumatoid factor,” which was
coined in 1940, also requires updating. Since rheumatoid factor
identifies antibodies that bind Fc fragments of the host’s IgG
molecule, why not call it antiimmunoglobulin factor, anti-Fc
factor, or immunoglobulin antiimmunoglobulin activity?
Piero Stratta, MD
Cristina Bozzola, MD
Marco Quaglia, MD
Amedeo Avogadro University
Maggiore Hospital
Novara, Italy
1. Liao KP, Weinblatt ME, Solomon DH. The end of rheumatoid
factor as we know it? [editorial]. Arthritis Rheum 2011;63:1170–2.
2. Waaler E. On the occurrence of a factor in human serum
activating the specific agglutination of sheep blood corpuscles.
Acta Pathol Microbiol Scand 1940;17:172–88.
3. Rose HM, Ragan E, Pearce E, Lipman MO. Differential agglutination of normal and sensitized sheep erythrocytes by sera of
patients with rheumatoid arthritis. Proc Soc Exp Biol Med 1948;
68:1–6.
4. Westwood OM, Nelson PN, Hay FC. Rheumatoid factors: what’s
new? Rheumatology (Oxford) 2006;45:379–85.
5. Charles ED, Dustin LB, Hepatitis C virus-induced cryoglobulinemia. Kidney Int 2009;76:818–24.
6. Meltzer M, Franklin EC, Elias K, et al. Cryoglobulinemia: clinical
and laboratory study. II. Cryoglobulins with rheumatoid factor
activity. Am J Med1966;40:837–56.
DOI 10.1002/art.33397
The additive value of magnetic resonance imaging for
bone edema in predicting rheumatoid arthritis
development in early undifferentiated arthritis:
comment on the article by Duer-Jensen et al
To the Editor:
We read with great interest the report by Duer-Jensen
et al on the value of magnetic resonance imaging (MRI) in
predicting the development of rheumatoid arthritis (RA) in
patients with undifferentiated arthritis (UA) (1). The authors
performed MRI in patients without a specific rheumatologic
diagnosis who had ⱖ2 tender joints or ⱖ2 swollen joints. After
12 months, a final diagnosis for each subject was determined,
based on the American College of Rheumatology 1987 criteria
for RA (2) or accepted criteria for other rheumatic diseases.
Duer-Jensen and colleagues observed that bone marrow
edema in particular was most predictive of the development of
RA. In addition, they added clinical variables to the MRI
variables and observed that this improved the model. The
present study is relevant, because it indicates that sensitive
imaging techniques such as MRI are of additive value to
known clinical or serologic values in predicting the development of RA.
Duer-Jensen et al also compared the performance of
their prediction model with that of the model we described
previously and concluded that their model performs better. In
our view, the data provided in their report do not permit a fair
comparison. First, there are concerns with regard to the
predictive performance of the model of Duer-Jensen et al. In
addition, there are concerns regarding the manner in which the
2 models were compared.
The first concern is that the data from the model
described by Duer-Jensen et al may provide a view on the
predictive ability of the model that is too encouraging, because
of overfitting of the data. The model is based on 116 patients.
In addition, data are missing; patients for whom data are
lacking for one or more of the variables included in the logistic
regression analyses are not taken into account. This means that
the model is based on fewer than 116 patients; the exact
number is not mentioned. Deriving a model based on a rather
small number of patients increases the risk of overfitting. This
can be addressed by external validation. If that is not directly
possible, correction, to some extent, can be accomplished by
internal validation (e.g., by cross-validation or bootstrapping).
None of these procedures was applied. Consequently, the
results presented are presumably too optimistic.
A second issue is the comparison with our previously
derived model, which was validated, both internally and externally, in several independent cohorts. A common way to
compare predictive tools is by comparing areas of the receiver
operating curve (AUC). This comparison cannot be done here,
because the AUC of the model used by Duer-Jensen et al was
not provided.
Based on the information that is provided in the study
by Duer-Jensen et al, the accuracy of the van der Helm model
in the Danish data set is 60.2%. This information cannot be
interpreted and compared, for several reasons. First, the
authors do not mention which model was used (the version in
which the severity of morning stiffness is measured on a visual
analog scale, or the version in which morning stiffness is
expressed in minutes; both models perform slightly differently)
(3,4). Second, the definition of UA used in the present study is
a bit different from the definition that is frequently used.
Patients in Denmark who had tender joints were also classified
as having UA, despite the absence of swollen joints at the time
of examination. The van der Helm model is based only on
patients with arthritis. Applying a model to a population with
a different background will influence the performance. For
comparison, the anti–cyclic citrullinated peptide test is a very
powerful predictive factor of the development of RA in
patients with UA but is less useful as a predictive tool in the
general population. This is caused by the differences of prior
risk on RA in UA patients compared with the general population. A third relevant issue is that the van der Helm model
does not have a single cut-off point but rather is a continuous
scale, such that every prediction score has its own positive
predictive value and negative predictive value. Duer-Jensen et
al did not mention to what value the prediction score of 60.2%
is related. In other words, this 60.2% belonging to one cut-off
value cannot be compared with the whole Duer-Jensen model.
As mentioned, comparison of AUCs would overcome this. It
would be interesting to see the results of a comparison of AUC
values obtained after internal validation.
In conclusion, we congratulate Dr. Duer-Jensen et al
with their promising study, which is one of the first to indicate
that MRI may be of additive benefit in prognostication for
patients with early arthritis. However, internal and external
322
LETTERS
validation are required for correct interpretation of the predictive ability of the model.
Annette H. M. van der Helm-van Mil, MD, PhD
Tom W. J. Huizinga, MD, PhD
Saskia le Cessie, PhD
Leiden University Medical Center
Leiden, The Netherlands
1. Duer-Jensen A, Horslev-Petersen K, Hetland ML, Bak L, Ejbjerg
BJ, Hansen MS, et al. Bone edema on magnetic resonance imaging
is an independent predictor of rheumatoid arthritis development in
patients with early undifferentiated arthritis. Arthritis Rheum 2011;
63:2192–202.
2. Arnett FC, Edworthy SM, Bloch DA, McShane DJ, Fries JF,
Cooper NS, et al. The American Rheumatism Association 1987
revised criteria for the classification of rheumatoid arthritis. Arthritis Rheum 1988;31:315–24.
3. Van der Helm-van Mil AH, le Cessie S, van Dongen H, Breedveld
FC, Toes RE, Huizinga TW. A prediction rule for disease outcome
in patients with recent-onset undifferentiated arthritis: how to guide
individual treatment decisions. Arthritis Rheum 2007;56:433–40.
4. Van der Helm-van Mil AH, Detert J, le Cessie S, Filer A, Bastian
H, Burmester GR, et al. Validation of a prediction rule for disease
outcome in patients with recent-onset undifferentiated arthritis:
moving toward individualized treatment decision-making. Arthritis
Rheum 2008;58:2241–7.
DOI 10.1002/art.33401
Reply
To the Editor:
We thank Dr. van der Helm-van Mil and colleagues for
their interest in our work and their thoughtful comments on
our study, in which we demonstrated that MRI evidence of
bone edema in the metatarsophalangeal (MTP) and wrist
joints is an independent predictor of future RA in patients with
early UA, based on prospective followup study of 116 patients
with early UA. We also introduced a prediction model that
included as explanatory variables the presence of hand arthritis, morning stiffness, rheumatoid factor positivity, and the
MRI score for bone edema in the MTP and wrist joints, which
correctly identified the outcome of RA or non-RA in 82% of
the patients with UA.
As a secondary analysis, we investigated the performance of the well-established prediction model developed by
the van der Helm-van Mil group (1,2) and observed that in
our cohort of patients with UA (inclusion criteria: ⱖ2 tender
and/or swollen joints among the metacarpophalangeal, proximal interphalangeal, wrist, or MTP joints; no verified specific
rheumatologic disease after routine clinical, biochemical, and
radiologic examination; and a symptom duration of ⬎6 weeks
but ⬍24 months), the van der Helm model (2) predicted the
outcome of RA or non-RA in 60.2% of the population. In their
letter, Dr. van der Helm-van Mil et al raise some issues, which
we will address below.
We fully agree with Dr. van der Helm-van Mil and
associates that our cohort is not identical to the populations
assessed by their group (e.g., it is smaller, and the inclusion
criteria were not identical), and that this difference may
influence the performance of the models. We acknowledged
Figure 1. Receiver operating curve for prediction model.
this in our study and are happy to emphasize this point again.
We also acknowledge, as explained thoroughly in the legends
to Tables 1 and 2 and in the statistical analysis section, that
some data are missing. The van der Helm model is a wellexplored model that is founded on solid statistical work. Our
model is also well-founded on thorough statistical work, providing encouraging results, but the model has not yet been
tested in other cohorts, in contrast to the van der Helm model.
This is due to the fact that, unfortunately, cohorts of patients
with early UA for whom MRI results are available are still very
rare. However, our report describes one of the largest cohorts
of UA patients investigated with MRI, and we believe that our
study provides important new knowledge to the field. We agree
that external validation of our model is important, and we hope
this will be possible in future research.
In their letter, van der Helm-van Mil et al propose
analysis of the ROC of our model. This was actually done (see
Table 4). The ROC value itself was not presented in the
manuscript, but it is now included here. As shown in Figure 1,
the area under the curve was 0.88.
As they mentioned in their letter, Dr. van der Helmvan Mil and colleagues have used several versions of their
prediction model in their studies. We used the model that they
proposed and validated in 2008 (2). This model includes
morning stiffness expressed in minutes, as mentioned in the
introduction to our study, where we listed the variables applied. In that study, the negative predictive values for a
prediction score of ⱕ6 were ⱖ83% in all 3 cohorts, and the
positive predictive values for a prediction score of ⬎8 were
93–100% (2). In our cohort, we used the same cut-off values
are those used by van der Helm-van Mil et al.
Finally, we would like to emphasize that the purpose of
our study was not to develop a model competing with the
well-validated van der Helm models but rather to investigate
whether MRI can significantly contribute to the process of
diagnosing RA. The fact that the presence of bone edema as
Документ
Категория
Без категории
Просмотров
4
Размер файла
52 Кб
Теги
33397, art
1/--страниц
Пожаловаться на содержимое документа