The relative risk was further muddied by the potential for toxic pulmonary effects associated with all DMARDs. “Essentially, every single medicine that is approved for the treatment of rheumatoid arthritis has been implicated in causing a hypersensitivity pneumonitis, which is basically an allergic-type reaction to the drug that happens within the lungs,” Dr. England says. “It’s just that for RA-ILD, there are a couple medicines out there that people were particularly concerned about.”
On the basis of the limited preliminary evidence, Dr. England and his colleagues initially hypothesized that treatment with non-TNF inhibitors would be more effective and associated with improved survival and fewer respiratory hospitalizations than treatment with TNF inhibitors. Their results to the contrary, however, left him only “modestly” surprised, he says. “The careful design often can lead to a result that is different than what a less stringently defined observational study produces.”
To minimize the potential for bias that often plagues observational studies, the research team used a target trial emulation framework, which Dr. England calls an important advance in pharmaco-epidemiology. “The principle is that when we do an observational study, we should approach it with the rigor and the strategy that we would if we designed this as a clinical trial,” he says.
Among the critical questions are which patient populations should be included in a trial, which participants should receive which therapies and how, and what the best outcome measures would be. The target trial emulation framework then specifies that the same approaches taken for a clinical trial should be applied to observational data. “Ultimately, you can’t pull it off perfectly because in a trial you have randomization: You can flip the coin and give one drug to one person, the other drug to another,” he concedes. Instead, observational studies can use statistical techniques like propensity score-based methods to reduce bias by estimating the probability that a particular participant would have received a particular treatment.
Questions of Interpretation
The study results yielded additional questions that will likely require more follow-up to fully resolve, like how to interpret the null results when comparing the two patient groups. “One of the questions is, is it null because for every single person in that study it didn’t matter which drug you gave them?” Dr. England says. Alternatively, a subset of patients in each group may have done better if they had received the other drug, effectively canceling each other out in the statistical analysis.
