番茄社区app

Whether patients with ankylosing spondylitis (AS) would improve substantially with tumor necrosis factor (TNF) inhibitor therapy was predicted with moderate to high accuracy based only on a standard lab test and two measures of disease activity, researchers reported from a modeling study.

Taken together, models incorporating the three indicators -- levels of C-reactive protein (CRP), scores on patients' global assessment of their condition, and patient-reported spinal pain -- showed specificity of 82% to 83% and sensitivity of 47% to 54% for predicting a "major" response to TNF inhibitors after 12 weeks of treatment, according to Runsheng Wang, MD, MHS, of Garden State Rheumatology Consultants in Union, New Jersey, and colleagues.

These translated to area under the receiver-operator characteristic curve (AUC) values of 0.65-0.67, the researchers .

Factors predicting the likelihood of no response at 12 weeks included age and Bath AS Functional Index (BASFI) score, with similar overall accuracy (AUCs of 0.65-0.66); specificity was greater (92%-93%) but at the cost of reduced sensitivity (36%-41%). In some applications for both sets of analyses, body mass index (BMI) also aided in prediction.

Wang's group suggested that the models (one using logistic regression, the other a random forest approach) could be useful in routine practice. "Confidence in choosing TNF inhibitor treatment may be enhanced with a high probability score for major response," they wrote; moreover, clinicians may suspect poor adherence in patients showing no response yet predicted to be a good responder.

By the same token, they added, patients for whom no response is likely can have TNF inhibitor stopped quickly if, indeed, they do not improve.

Having a prognostic tool for treatment response in AS is important because around half of patients in randomized trials don't respond to TNF inhibitors. "Limited guidance is available to predict treatment responses in individual patients," Wang and colleagues observed. While studies have identified a number of individual factors associated with treatment response in this population, how they might be combined to improve predictive accuracy in individual AS patients had not been attempted before.

To fill this gap, Wang's group drew on individual patient data collected in 10 previous clinical trials of TNF inhibitors for AS, with a total of 1,899 participants included in the current analysis. About 1,200 were used to develop the models, and the rest served as a testing set for the end products. The primary clinical outcomes at 12 weeks were twofold: "major" response, defined as at least a 2-point decrease in AS Disease Activity Score; and no response, defined as a decrease of less than 1.1 point on this measure.

More than 25 factors were considered for inclusion in the model. Most ended up not contributing significantly to improving predictive accuracy. Those that did were CRP level; patient's global assessment score; response to question 2 on the Bath AS Disease Activity Index (BASDAI) instrument, which asks about spinal pain; BASFI score; age; and BMI.

Probability of achieving major response increased with just the first three factors; however, the probability of not achieving a major response was best predicted by BASFI and BMI. In particular, CRP levels, BASDAI question 2 score, and patient's global assessment score were all higher in patients most likely to enjoy a major treatment response. BMI values and BASFI score were higher in patients having no response.

Limitations to the study included lack of data on patients' smoking status, meaning it could not be incorporated into the model (previous studies have linked smoking to poorer TNF inhibitor response). Also, participants had not previously used these drugs, and thus the results can't be generalized to patients with previous TNF inhibitor exposure. And, perhaps most importantly, the models weren't tested on regular patients in routine practice, nor in those with axial spondyloarthritis.

Comment