Improved CPMs Needed for Guiding Methotrexate Use in RA

The study analyzed 20 clinical prediction models (CPMs) for rheumatoid arthritis (RA), all of which were deemed to have a high risk of bias, leading the researchers to urge against their use to guide decision-making without addressing their limitations.

There is a need for clinical prediction models (CPMs) that can successfully guide providers in using methotrexate for the treatment of rheumatoid arthritis, found a systemic review and meta-analysis.

The study analyzed 20 CPMs from 13 studies and 4 validation studies, all of which were deemed as having a high risk of bias (ROB), leading the researchers to urge against their use to guide decision making without addressing their limitations. The group largely attributed this to small sample sizes, lack of validation, insufficient handling of missing data, inconsistent reporting, and failure to account for discontinuation rates due to adverse events (AEs).

Their findings were published in Seminars in Arthritis & Rheumatism.

“Only one study predicted discontinuation due to AEs and model performance was poor. This may be due to the small outcome prevalence compared to predicting treatment response as defined through the DAS28 [health assessment questionnaire],” explained the researchers. “A relatively small outcome prevalence may result in low statistical power, leading to risk models making inaccurate predictions. As various outcome definitions are currently used, further input from patients and prescribers on which outcomes are clinically important may be valuable.”

In addition to AEs, outcomes analyzed in the study included state of disease activity and European Alliance of Associations for Rheumatology response criteria. According to the researchers, most models were able to predict state of disease activity, which they note may be due to treat-to-target recommendations that promote remission or low disease activity as quickly and safely as possible.

Just one model accounted for potential competing risks, an approach that could offer more realistic and clinically relevant predictions, the group wrote. Forty-rive percent of models were internally validated using recommended resampling techniques in the development dataset to correct for optimism.

Across the CPMs, the researchers found a lack of data reporting on certain patient characteristics, including comorbidities, which play an important role in methotrexate prescribing and response. Forty-rive percent of models included data on concomitant antirheumatic treatment.

There were 8 (40%) models that used multiple imputation to account for missing data, producing various plausible datasets and averaging results across each, which the researchers noted is considered a robust approach to account for missing data. The other models used complete case analysis, which is limited to patients with full baseline data available, resulting in smaller sample size for analysis.

“There are, however, some encouraging results identified by our review, as ROB was universally rated as low in the participants, predictors, and outcome domains, and the meta-analysis showed overall good performance of the 2 CPMs with multiple external validations,” described the researchers. “Therefore, once the methodological limitations outlined above are addressed, there is scope to develop an accurate CPM of methotrexate outcomes to guide treatment decisions in rheumatoid arthritis (upon undertaking appropriate impact assessment studies). However, even if patients at high risk of poor response of methotrexate were successfully identified, there is currently no clear indication that they will respond better to alternative therapies, such as biological disease-modifying antirheumatic drugs.”

Reference

Gehringer C, Martin G, Hyrich K, Verstappen S, Sergeant J. Clinical prediction models for methotrexate treatment outcomes in patients with rheumatoid arthritis: a systematic review and meta-analysis. Semin Arthritis Rheum. Published online July 31, 2022. doi:10.1016/j.semarthrit.2022.152076