Searches were made in the MEDLINE and EMBASE databases for articles published before November 2011. The following search terms were used as free text and MeSH terms: "magnetic resonance imaging," "spondyloarthritis," "ankylosing spondylitis," "sacroiliitis," "psoriasis arthritis," "reactive arthritis," "arthritis and inflammatory bowel disease," and "inflammatory back pain." Different forms of spelling and synonyms for each term were also used.

Sensitivity and specificity were retrieved from the article where possible and calculated from raw data when not reported in the article. The Wilson score method without continuity correction was used to calculate the 95% CI, if it were not reported in the article. Meta-analysis was planned to be performed on homogeneous high-quality studies, but was not performed because of insufficient data.

In all, 1,336 articles were found in MEDLINE, and 2,359, in EMBASE. After elimination of duplicates, 2,395 articles were reviewed by title and abstract, and 76 articles were identified for assessment in full text. After full-text review, nine cross-sectional articles met the inclusion criteria 10 11 12 13 14 15 16 17 18 (Figure 1). No longitudinal studies met the inclusion criteria. Reasons for exclusion of articles reviewed in full text are presented in Table 2 for articles on cross-sectional studies and in Table 3 for articles on longitudinal studies.

The nine articles meeting the inclusion criteria were based on eight study samples and were published between 1995 and 2010 10 11 12 13 14 15 16 17 18 . Two studies were conducted on the same study population 17 18 . Seven of the nine articles were designed to address the diagnostic utility of MRI in the diagnosis of either AS, SpA, or IBP. One article compared ultrasound with MRI in the diagnosis of IBP 11 , and one article evaluated referral recommendations 12 . Five studies reported results for the SIJ 10 11 12 13 14 , and four, the results for the spine; one of the lumbar spine 15 and three of the whole spine 16 17 18 . The sample sizes varied from 85 to 187 participants. The diagnoses under evaluation were AS 10 12 13 14 15 16 , SpA 10 12 14 17 18 , and IBP 11 13 16 . The groups of noncases were no LBP 11 13 16 17 18 and/or noninflammatory low-back pain, such as NSLBP, degenerative arthritis, cancer, and LBP of unknown origin 10 12 13 14 15 17 18 . For further details of the included studies, see Table 4.

All studies used an MRI scanner with a field strength of 1.5 Tesla. Four of the included articles had no description of the field strength, and the corresponding authors were therefore contacted for information of the field strength. Eight of the nine articles reported which MRI protocol was used 10 11 13 14 15 16 17 18 . For the SIJ, the most common sequences were T₁-weighted spin-echo in combination with either a gadolinium sequence 10 11 14 or a Short Tau Inversion Recovery (STIR) sequence 13 . The slice orientation was either semicoronal or semiaxial or a combination of both. In the spine, the MRI protocol consisted of fewer sequences. Three articles reported the use of sagittal T₁-weighted spin-echo and STIR 15 17 18 , and one used only sagittal STIR 16 . None of the included studies evaluating the spine used contrast enhancement in the MRI protocol. For more details of the MRI protocols, see Table 5.

The quality score of the nine studies ranged from 1 to 7 points, with a mean score of 3.7. Two articles were rated more than 5 points and were considered of high quality 13 16 , and three articles were rated less than 3 points 10 11 12 (Table 1).

In relation to the description of the study samples, five articles did not fully describe both cases and noncases with respect to age, gender, and diagnosis 11 14 15 17 18 , three articles had no description of the sampling method used for the control groups 11 17 18 , and for two of these 17 18 , it was not clear whether these groups were included in the data analyses. Regarding the sampling methods used, one article reported the use of consecutive sampling 16 . The remaining eight either failed to report on this aspect or used convenience samples. Matching was performed adequately in relation to age and sex in four 10 13 15 16 of the five case-control studies. One study had a case group with a mean age that was more than 20 years older than that of the control group 11 .

In relation to the diagnosis of the target condition, none of the nine articles tested the reproducibility of the clinical diagnosis. Diagnostic criteria were used as a reference standard in five articles 10 11 13 15 16 , expert opinion in three articles 14 17 18 , and both criteria and expert opinion in one article 12 . In four articles describing retrospective studies 14 15 17 18 , the independence of MRI from the clinical diagnosis was unclear. In one prospective study 12 , which aimed to test referral recommendations, MRI was recognized as part of the diagnosis for the group of preradiographic SpA, thereby confounding the association observed. In seven articles 10 11 12 14 15 17 18 , inadequate or no reporting was made of blinding of the MRI results at the time of the clinical diagnosis.

In relation to the MRI evaluation, most studies had acceptable definitions of the MRI findings 13 14 15 16 17 18 . Five articles performed reproducibility tests of the MRI evaluation 13 14 16 17 18 , and five studies blinded the evaluation for the diagnosis 13 14 15 16 17 . None of the nine articles reported the time interval between MRI and diagnosis.

In relation to the statistical analysis, one article failed to report fully the CI, P values, and/or raw data 10 , and one article used an inappropriate statistical method for the types of data available 14 . Five articles had shortcomings in the way the results were presented 10 12 14 17 18 , such as uncertainties in relation to which groups were included in the calculation of the specificity 17 18 and data primarily presented as graphics and per SIJ joint instead of per person 10 .

Because of a substantial heterogeneity in the included studies regarding the methodologic quality, the MRI findings, the MRI technique used, and the regions under evaluation, it was not possible to perform a meta-analysis, and we decided not to report data systematically. Instead, a descriptive assessment of the results based on the extracted data was performed. One article of high quality was found that evaluated the SIJ 13 , and another article of high quality was found that evaluated the spine 16 , both of which were from the same research team. A summary of the extracted data from these two studies is presented in Tables 6 and 7.

The results of the review show that a positive association exists between certain MRI findings in the axial skeleton and the diagnosis of both AS and IBP. However, these results are based on only two high-quality studies, and the amount of relevant data was insufficient to give a reliable quantification of this association.

On the basis of the identified methodologic shortcomings of the included articles, this review highlights important issues that should be taken into consideration when conducting studies with the aim of evaluating the diagnostic utility of MRI in the diagnosis of SpA. The following issues will be discussed in the subsequent text: Size of the study sample, selecting the proper study population, sampling method of the study cohort, independence of MRI from the reference standard, reproducibility testing of both the reference standard and the MRI evaluation, and standardization of the MRI findings.

In this review, the inclusion criterion for the size of study samples was set to a minimum of 20 subjects per group. This was because a small sample size often gives imprecise estimates of diagnostic accuracy. For example, when the number of patients with SpA is 20, the two-sided 95% CI of a sensitivity of 80% is 58% to 92%. This is, however, still a very wide CI, and because wide CIs make it difficult to determine how informative the results of the study are, we would recommend that future studies use larger sample size. For example, if the number of patients with SpA is 200, the two-sided 95% CI of a sensitivity of 80% is 74% to 85%.

When choosing the study sample, it is crucial that an effort be made to make it as comparable to the population seen in a clinical setting as possible. Not surprisingly, the results from the two high-quality studies show that the association was stronger between patients with AS and asymptomatic people than the association between IBP patients and patients with NSLBP 13 16 . However, for the clinician, the challenge will often be differentiating between IBP and NSLBP. When investigating the diagnostic utility of MRI in SpA, it is therefore important that the study samples be representative of "real" patient populations, instead of comparing patients with long-lasting disease and asymptomatic persons. Furthermore, one of the high-quality studies reports inflammatory MRI findings in the SIJ to be relatively common in patients with NSLBP 13 . Because the specificity is estimated entirely from results from the group of noncases or controls, this, additionally, underlines the need for further evaluation of the presence of MRI findings also in patients with NSLBP.

The sampling methods used in the included studies were either insufficient or not fully described in the majority of the studies, which makes it difficult to preclude selection bias. It is mandatory when conducting studies on diagnostic accuracy that both the sampling method and the study sample are well described, that the selection of study participants is performed systematically and unselectively, and, for case-control studies, that the two groups are comparable except for the target condition. Furthermore, in some of the articles identified in this review, the importance of the controls seemed to be underestimated, even though the sampling and description of the controls are just as important as the sampling and description of the cases.

That no objective gold standard exists for the diagnosis of SpA, which is instead based on a composite clinical presentation, constrains the possibilities for the evaluation of the diagnostic accuracy of MRI. However, it is indisputable that when evaluating a diagnostic test, independent evaluation of the index test and the reference standard is essential. When MRI is used in the decision making of the clinical diagnosis, as seen in more than half of the included studies, incorporation bias is highly likely. This was predominantly a problem in the included retrospective studies. The fact that MRI is widely used in the diagnosis of SpA restricts the possibilities of conducting meaningful retrospective studies on the diagnostic accuracy of MRI.

One way to strengthen the reliability of the diagnosis is to test the reproducibility of the selected reference standard. However, none of the nine included articles tested the reproducibility of the clinical diagnosis. Another way to strengthen the reliability of the diagnosis would be to include a longitudinal aspect and to follow the development of more-advanced disease stages to confirm the diagnosis. Even though several longitudinal studies were identified in this review, none met the inclusion criteria, usually because either no or too few controls were used.

Several activity scores for inflammatory and structural MRI findings have been developed; however, no international consensus exists about the definition of diagnostic MRI findings, which was a factor in our review. In general, the included studies contained definitions of the MRI findings. However, little homogeneity existed in these definitions, and only five of the nine articles reported testing of the reproducibility of the MRI evaluation. Before testing the diagnostic utility of an MRI finding, it is necessary to define the MRI finding and further to present acceptably reproducible test results. Furthermore, an experienced musculoskeletal radiologist should be part of the conduct and testing of the evaluation protocol. An international agreement on definitions of different MRI findings and a threshold minimal reproducibility would facilitate both the translation from research to clinical practice and a comparison with future research results.

In summary, the optimal design for future studies that investigate the utility of MRI in the early diagnosis of SpA is a large longitudinal clinical cohort study that recruits participants consecutively and includes patients with NSLBP. Furthermore, a well-defined clinical diagnosis based on consensus criteria and blinded for MRI should be used as a reference standard. Similarly, a validated MRI evaluation with clear definitions blinded to the clinical information should be used for the MRI evaluation.

The literature search associated with the current review revealed no other systematic critical review on the utility of MRI in the diagnosis of SpA. However, a large number of commentaries and narrative reviews with or without transparent search strategies have been published, in which it seems that the majority report a positive view of the value of MRI in the diagnosis of SpA. The obvious need for optimization of the diagnostic process of SpA may have contributed to the positive perception that has developed regarding MRI as a tool in the SpA diagnosis. The fact that MRI of the SIJ is included in the new diagnostic criteria developed by ASAS 8 has led to further optimism in relation to the diagnostic process of SpA. However, on the basis of the results from the current review, the limited number of high-quality studies on the diagnostic utility of MRI in SpA should preclude MRI being seen as the new gold standard in the diagnosis of SpA. If inflammatory and structural MRI findings are interpreted as being more specific for SpA than the research can support, a real risk exists of SpA being overdiagnosed.

As mentioned earlier, to be able to make a reasonable translation of research results to clinical practice, the representativeness of the study sample to clinical practice and the standardization and validation of the MRI evaluation is important. However, it also is essential that the MRI protocols used in the studies are relatively easy to implement and constitute the least health risk for the patients. Three of the studies included in this review used gadolinium contrast-enhanced sequences. However, for future studies, the necessity of gadolinium should be carefully considered, because other techniques, such as STIR, have shown a similar capability of visualizing BMO 22 23 , and the use of contrast is more invasive, expensive, and time consuming than is a noncontrast method. Furthermore, a small but serious risk of contrast medium-induced nephrotoxicity exists 24 , which also must be taken into consideration.

The strength of the current review is the systematic search method and the rigorous assessment of the methodologic quality of the included studies. To our knowledge, this review is the first to give an overview of the level of evidence of the diagnostic utility of MRI, including a quality assessment.

The weaknesses of the review are first, the possibility of relevant articles not being included because of the language limitations, and second, the limited number of databases used in the search. Furthermore, it might be argued that the threshold for a minimal study sample size was either too low or too high. However, it seems reasonable to reduce the width of the 95% CI by not setting the cut-point too low; in contrast, increasing the cut-point for sample size further would have greatly decreased the number of relevant articles that could be included.