Abstract
Objective. To perform a systematic literature review on the diagnostic and predictive value of conventional radiographs (CR) in patients with undifferentiated arthritis (UA).
Methods. We performed an extended search using Medline, Embase, the Cochrane Library, and abstracts from the 2007 and 2008 meetings of the American College of Rheumatology and the European League Against Rheumatism. Articles were included based on predefined inclusion criteria, and quality was assessed by using validated quality scales.
Results. In total, 25 articles were included from 6003 retrieved references. Five articles described a pure UA population, 20 articles described a mixed population [mostly rheumatoid arthritis (RA) and UA]. In studies on UA, erosions on CR were strong predictors of RA diagnosis [positive likelihood ratio (LR+) 3.5–10.9; odds ratio 7.6 and 8.7). In a more heterogeneous mixed population, 20 studies reporting on 11 cohorts found a relationship between CR findings and subsequent diagnosis of RA. LR+ for erosions and/or bony decalcifications ranged from 1.8 to 9.7, and there was greater prevalence of erosions and higher Sharp-van der Heijde score in the RA group at followup. With regard to prognosis in both UA and mixed populations, an association was found between number of abnormalities on CR and poor outcome.
Conclusion. Several studies, in pure UA and mixed populations, clearly demonstrate that CR are helpful in predicting future diagnosis of RA or worse prognosis. However, absence of abnormalities on CR does not sufficiently exclude RA or other unfavorable outcome.
Undifferentiated arthritis (UA) is an ill-defined disease entity, since it is characterized by the absence of other diseases. Establishing an early diagnosis or prognosis in patients with UA is of major importance to obtain earlier and targeted treatment, leading to better outcomes for these patients1. To understand and learn more about this group of patients, “How to Investigate and Follow up Undifferentiated Peripheral Inflammatory Arthritis (UPIA)” was chosen as the subject for the 2009 3e (evidence, expertise, exchange) Initiative in Rheumatology.
The 3e Initiative promotes evidence-based medicine by formulating recommendations using both data from the literature and expert opinion2. Seventeen countries and almost 700 experts participated in this project. In total, 10 clinical questions selected by clinicians were chosen for a systematic review. The final recommendations based on the 10 different systematic reviews can be found elsewhere3.
We present results of the systematic reviews of one of the 10 clinical questions: “What is the diagnostic and predictive value of X-ray in UPIA? Should it be performed at baseline and repeated at what interval?”.
Conventional radiographs (CR) are commonly used for arthritis patients as additional tests in clinics. CR are relatively safe, inexpensive, and widely available, which makes them a convenient test in current clinical care. Radiographs can help confirm or exclude diagnoses such as rheumatoid arthritis (RA) or psoriatic arthritis. Abnormalities on radiographs might also be valuable in predicting other outcomes in UA, such as structural damage or impaired physical functioning4.
At this time there are no systematic reviews that describe the importance of this test in UA for establishing either diagnosis or prognosis, since most data are collected in patients with early RA.
We assessed the diagnostic and predictive value of CR in patients with UA by reviewing all available literature using an extensive search strategy. As part of the 3e process, an evidence-based recommendation on the use of CR was then formulated by combining results of this review and the opinion of experts.
MATERIALS AND METHODS
The selected clinical question was rephrased according to the PICO method (Patients, Intervention/index test, Comparison, Outcome)5, which is used to translate a clinical question into one with epidemiological terms to make a literature search possible. Population was defined as the patient with UA, and conventional radiographs as Intervention.
For our search there is no relevant control group. For the diagnostic search, outcome was defined as a specific diagnosis such as RA or psoriatic arthritis. Outcome in the search on prognosis was defined very broadly, in principle as every possible unfavorable outcome (e.g., progression of disease, radiological damage, impaired physical functioning), and the final selection was dictated by the available literature.
Three types of studies were considered for inclusion: (1) cohort studies in which patients from a given UA population had CR at baseline and in whom the outcome after a period of followup was recorded; (2) retrospective case-control studies in which patients had CR at baseline and who were known to have had UA when the baseline investigation was performed; and (3) randomized controlled trials of UA patients that implicitly addressed the question of diagnostic or prognostic value, as each arm of a trial can be seen as a separate cohort study.
Medline, Embase, and the Cochrane Library were searched for articles published between 1950 and February 2009. The extensive search strategy (see online appendix available from: www.3eupia.com) was developed in close collaboration with a trained librarian and consisted of 3 parts: target population, intervention, and preferred study type (diagnostic or prognostic). Abstracts presented at the 2007 and 2008 meetings of the American College of Rheumatology (ACR) and the European League Against Rheumatism (EULAR) were searched using the following terms: undifferentiated, undiagnosed, unclassified, early, or probable arthritis. All references of selected articles and relevant reviews were hand-searched for additional articles.
The selection process consisted of 2 phases: (1) All titles and abstracts were checked for relevant articles; and (2) the full articles were reviewed in detail and retained or excluded based on predefined criteria. Criteria included: UA patients ≥ 18 years of age, presence of at least one clinically swollen joint, and use of CR to predict diagnosis or prognosis in these patients. Articles were split into 2 groups: a pure UA group and a mixed population, where only part of the group consisted of patients with UA.
Articles included in the review were assessed for quality using validated scales. Quality assessment of diagnostic studies was performed using a scale based on the Evidence-Based Medicine Working Group Quality6. Assessment of the prognostic studies was done by the Newcastle-Ottawa Quality Scale looking for 3 items: selection, comparability, and outcome. The maximum number of stars that can be awarded to a study is 97. The level of evidence was determined using the scale developed by the Oxford Centre for Evidence Based Medicine8 (see online appendix available from: www.3eupia.com).
Data were extracted using a predefined format and analyzed by 2 researchers. If necessary, corresponding authors were contacted for additional details. Likelihood ratios (LR) and confidence intervals were extracted or calculated when possible; if these data were unavailable, descriptive results were used. The higher the positive LR (LR+) and the lower the negative LR (LR−), the higher the value of the test. LR+ > 5 and LR− < 0.2 represent strong diagnostic or prognostic evidence6.
RESULTS
In total, 6003 references where found via Medline and Embase using the developed search strategy, of which 115 articles where reviewed in detail (see online appendix, available from: www.3eupia.com). In total, 25 articles were included, 5 with a pure UA population and 20 with a mixed population. The Cochrane Library did not retrieve any relevant articles, and 2 abstracts from the 2007 and 2008 EULAR and ACR meetings had already been included in the review.
Of the 5 studies with a pure UA population, 4 were diagnostic studies and one assessed prognosis. Three11,14,15 of these 5 studies were performed using the same cohort (the Leiden Early Arthritis Cohort) but with different numbers of patients included. Inclusion criteria and baseline characteristics are presented in Table 1.
Two articles, van Aken11 and Duer12, found a high LR+ (Table 2) for developing a diagnosis of RA according to ACR criteria13, the first for erosions according to the Sharp-van der Heijde (SvdH) method9, the other for Larsen grade 110. In the study by van der Helm-van Mil14, erosions were found to be a predictor in univariate analysis but not in multivariate analysis. The article by van Gaalen15 demonstrated that in a model with and without anti-cyclic citrullinated peptide antibodies the odds ratios for developing RA were moderate (Table 2). Prognosis was assessed by the study of Jansen16. When differentiating between mild and progressive disease at 1 year, SvdH scores at baseline are significantly different (Table 3).
Of 20 studies with a mixed population that could be included, heterogeneity in inclusion criteria or baseline characteristics was far greater than in the UA group (Table 1). In total, these studies are derived from 11 cohort studies from different countries. Most studies used different features of the radiographs, which makes them difficult to compare.
In general, studies found high LR+ for erosions predicting diagnosis. When using both hand and foot radiographs, LR increased in comparison to using hand radiographs only.
Bony decalcification, on the other hand, was not very informative for diagnosing RA. LR− were too high to be considered clinically important17,18,19,20,21 (Table 2). Results from the remaining diagnostic studies suggest that numbers of erosions or SvdH scores are in general different in the patients developing RA, compared to UA, but not independently predictive22,23,24,25,26,27,28(Table 3).
When predicting prognosis, such as progressive disease, onset of disease modifying antirheumatic drug treatment, or functional ability, several studies with numerous scoring methods found that more severe abnormalities are related to worse prognosis21,26,29,30,31,32,33,34,35,36 (Tables 2 and 3). There was no literature on how often radiographs should be repeated or on other radiographic characteristics that could be helpful in diagnosing or following patients with UA.
DISCUSSION
Our systematic review summarized and evaluated all available evidence on the diagnostic and prognostic value of conventional radiographs. The evidence found in this review together with expert opinion was used to make a clinical recommendation, as part of the 3e Initiative, which promotes evidence-based medicine in rheumatology. The description of the final recommendations can be found elsewhere3.
Several studies in both UA and mixed populations clearly showed that finding erosions on radiographs gives a high probability for developing RA or a worse prognosis, as demonstrated by high LR+. High LR− indicated that diagnostic or prognostic value in the absence of radiographic abnormalities is low11,12,17,18,19,20,21. There was no literature available on whether repeating radiographs during the diagnostic process or performing radiography to determine prognosis are of any value.
Even with our extensive search strategy, the available evidence from the literature for a pure UA population was scarce, certainly when compared to (early) arthritis in general. Many mixed cohorts included patients that fulfilled ACR criteria at baseline and differed in the percentage of UA patients (Table 1). Nevertheless, all studies retrieved, in both UA and mixed populations, pointed in the same direction, with high LR+ for the diagnostic and prognostic value of erosions on CR (Table 2), which increases the generalizibility of the results.
A problem in estimating the performance of CR as a diagnostic test is that there is circular reasoning, since erosions are part of the 1987 ACR criteria for RA. This may lead to an overestimation of the diagnostic value of radiographs. Consequently, the ACR criteria as a measure of outcome are open for debate. Yet this is the outcome that was used by the majority of the studies. It is nevertheless reassuring that the studies that used a different definition (RA as determined by a panel or by persistent disease) yielded the same results.
In conclusion, radiographs can be a valuable test in patients with UA and in mixed populations for predicting diagnosis and prognosis. These findings formed the basis for the final recommendation, given in detail elsewhere3.
Acknowledgments
We acknowledge J.W. Schoones, Walaeus Library, Leiden University Medical Center, The Netherlands, who participated in the development of the systematic search strategy; and all participants of 3e, especially the bibliographic team that participated in developing the search strategy and planning the analyses.
Footnotes
-
Supported by an unrestricted educational grant from Abbott. Abbott had no role in the design, literature search, data collection, data analysis, data interpretation, or writing of this report. Dr. Bombardier holds a Pfizer Chair and Canada Research Chair in Knowledge Transfer for Musculoskeletal Care.