Duong et al., 2022
Clinical predictors of response to methotrexate in patients with rheumatoid arthritis: a machine learning approach using clinical trial data
- Objectives
- To identify clinical predictors of response to MTX in patients with RA using ML methods.
- Related articles
- Other articles predicting MTX treatment response are Sergeant et al., 2018, Castrejón et al., 2016, Myasoedova et al., 2021.
- Link
- DOI: https://doi.org/10.1186/s13075-022-02851-5
Background
- MTX is the popular first-line DMARD treatment in RA
- There are no clinically useful tools to predict response to MTX treatment in patients with RA
Methods
- DMARD-naïve patients with RA from RCTs were accessed through a database
- Required DAS28-ESR at baseline and 12 and 24 weeks
- Latent class modelling of MTX response
- Lasso and random forests were used to identify predictors of response
- Model performance was assessed using AUC
Results
- 775 patients from 4 RCTs were included
- Two classes of patients were identified based on DAS28-ESR change over 24 weeks: good vs poor responders
- Top predictors were baseline DAS28, ACPA, HAQ – highest likelihood of response in DAS28 < 7.4, ACPA positive, HAQ < 2
- Isn’t this a bit of a no brainer? If you’re doing better at the start, you’re more likely to achieve a certain low threshold after a short amount of time? See also Capelusnik & Aletaha, 2021.
Conclusions
- Developed and externally validated a prediction model for response to MTX within 24 weeks in DMARD-naïve patients with RA
- One of the first studies to use ML methods to identify latent trajectories of DAS28-ESR over 24 weeks
The vast heterogeneity in response to MTX among individual patients with RA is insufficiently addressed in the current treatment guidelines, and systematic patient-tailored tools to personalize early RA management are lacking (for more information on treatment guidelines, see Smolen et al., 2020, Fraenkel et al., 2021).
- Lower baseline disease activity and better functional status are predictive of good responders to MTX is in line with previous studies ([[Capelusnik & Aletaha, 2021]], [[Sergeant et al., 2018]], [[Castrejón et al., 2016]])
- Authors claim that people below certain cutoffs (see [[#Results]]) have an 80% probability of responding to MTX treatment, but it is unclear to me how much uncertainty there is on an individual level – after all, this prediction should only matter on an individual level
The authors go into further detail about - Relevance of different time windows - Relevance of ACPA positivity - Relevance of individual DAS28 components for prediction and point to another study from the Netherlands where this was similar (see here, yet to be summarised) - Sociodemographic characteristics - External validation of results - Limitations and strengths of their study