Redefining the diagnosis of axial spondyloarthritis: The ASAS–SPARTAN 2025 classification criteria

My name is Walter Maksymowych. I’m a rheumatologist and Professor of Medicine in the Department of Medicine at the University of Alberta. I’m also the Chief Medical Officer of Care Arthritis.

Q. What was the rationale for revising the 2009 ASAS classification criteria for axSpA?

There were concerns expressed regarding the specificity of the criteria, because back pain occurs quite frequently in the population. The lack of specificity in the criteria could lead to the recruitment of false positives; patients without axSpA being incorrectly labelled by criteria that are insufficiently specific. This was particularly raised as a concern by the regulatory authorities, especially in regards to the subset of patients with non-radiographic axial spondyloarthritis.

It was for this reason that the regulatory authorities added additional requirements before patients could be recruited into clinical trials under these criteria; those additional requirements were a positive C-reactive protein (CRP) test or MRI. There have been other issues, such as the lack of certain genetic associations that were observed when more stringent criteria, such as the modified New York criteria, were used to identify cases.

Finally, when people used these criteria to look for patients and for population based studies, a very high frequency of axSpA was noted among back pain populations, and this was a very high frequency that was just simply implausible. Quite a few of those patients probably did not have axSpA and were falsely labelled as having it, because of the lack of specificity of the 2009 criteria.

So, it was a conjoint decision of the ASAS International Society and the Spartan Professional Society to move forward on a project. First of all, to validate the 2009 criteria, and if that did not meet pre-specified performance targets, then revised criteria would be generated.

Q. Could you outline the aims and design of the CLASSIC study? (3:04)

First of all, we recruited patients with undiagnosed back pain, who had chronic back pain for at least 3 months symptom duration, with onset less than 45 years of age. These patients were evaluated according to routine clinical practice, but in an incremental way, so that we could assess the contribution of different aspects of diagnostic information.

After the history and physical exam had been conducted, rheumatologists were asked to make a determination whether patients had axSpA, yes or no, and that was a global assessment. This was then repeated after laboratory investigations, namely CRP and HLA B27 were available, after the pelvic radiograph was made available, after the local MRI assessment, and finally after central readers had evaluated the imaging and provided the results of their assessments to the local rheumatologist. That final assessment at stage 5 was the crucial endpoint, the final diagnostic evaluation against which the performance of the criteria were to be tested.

We pre-specified before the study got underway that the 2009 criteria had to meet sensitivity of at least 75% and specificity of at least 90% when tested against that final stage 5 diagnostic gold standard. We found that the 2009 criteria failed to meet that performance target. The specificity was only 77% percent, and so we proceeded to generate revised criteria.

What methods were used to identify and weight the most relevant SpA variables in the revised criteria?

We conducted analyses over 2 years and I’m going to distill that into about 2 minutes. The primary analytical technique that was used to generate data driven criteria was a technique called least absolute shrinkage and selection operator (LASSO) regression. LASSO regression identifies which clinical laboratory imaging variables are independently associated with the diagnosis, and then provides coefficients for each of these significant independent variables. The size of the coefficient reflects a weighting as to the importance of that variable to the diagnosis, and this reflects clinical decision making.

When a clinician is evaluating a patient diagnostically, they are weighing different information in their mind. Certain variables will be weighted more highly, for example MRI, while others are weighted less, because they may be subjective or lack specificity and there may be various reasons for that. The LASSO analysis was able to extract the relevant variables, the independently associated variables, and provide weightings. The final step was to generate a cut-off, and this would be a cut-off above which a patient would be classified as having axSpA.

So there is a list of variables that have been selected by the LASSO, which all have weights and you sum up the numbers of the weights. For example, in our criteria, if a patient has inflammatory back pain, the weighting is 3, and if the patient has a positive MRI, the weighting is 8, so that gives a sum of 11; if they don’t have any other features, the total sum would be 11. The cut-off for the classification criteria is 11, so that patient would be classified as having axSpA. It is important to know that that cut off was based on the requirement to generate sensitivity of at least 75% and specificity of at least 90%.

When we generated the revised criteria, we were left with a list of variables in a table format with the most important variable ranked at the top, which would be a positive MRI for axSpA (weighting: 8), and then there were additional variables that were selected by the LASSO listed in rank order below the MRI variable. The cut-off was 11 and that gets us to the 7,590 targets in the validation data set that was generated. That is how the revised criteria were generated.

We also looked at aspects of face validity; there were certain variables selected by the LASSO that we felt were not appropriate, for example, dactylitis, and we replaced dactylitis with elevated CRP. The final criteria were generated by a combination of a data-driven process based on the LASSO regression, but also aspects of face validity. Our steering committees from ASAS and SPARTAN included members with many decades of expertise in the field, and it was important to reflect that expertise in the criteria.

Q. What were the key findings regarding the performance of the revised criteria compared to the 2009 version?

The primary improvement is a substantial increase in the specificity. The 2025 criteria met the specificity requirement of at least 90%, quite a bit more than the 77% percent for the 2009 criteria. In fact, if you use the data from central imaging readers, you reach a specificity of about 98–99%.

If you look at clinical trials recruiting patients and there are central readers evaluating the imaging, you can be quite certain that the criteria are selecting appropriate patients. This is such an advance compared to the 2009 criteria where central reader evaluations, with the criteria as they were framed, still only gave us a specificity of 84%.

Q. How do these results impact clinical research and diagnosis in axSpA going forward?

It is going to have a major impact in several ways. First of all, it illustrates the importance of imaging and especially MRI, and that’s only as it should be because MRI is undergoing tremendous rapid advances. MRI is producing images that are very highly discriminatory for the details of the sacroiliac joints and is even capable of informing us not only about the degree of inflammation, but also capable of generating an image that looks like a CT scan, so that we have exquisite detail around structural lesions. So it’s really important that the criteria be forward looking and have the ability to incorporate advances in imaging.

Secondly, we see that the specificity is indeed high and when people use the criteria, they can be more confident that they have really selected the appropriate patient for the conduct of basic research, clinical research and especially clinical trials research. We don’t want to be exposing patients to potentially toxic therapies who have not been very clearly identified as having the disease in question. This is really a crucial issue. It’s really important that for clinical trials we do select patients who really have axSpA and that’s what our criteria do.