Investigating the Long-term Effect of Pregnancy on the Course of Multiple Sclerosis Using Causal Inference
Abstract
Background and objectives: The question of the long-term safety of pregnancy is a major concern in multiple sclerosis (MS) patients, but its study is biased by reverse causation (women with higher disability are less likely to experience pregnancy). Using a causal inference approach, we aimed to estimate the unbiased long-term effects of pregnancy on disability and relapse risk in MS patients, and secondarily the short-term effects (during the per-partum and post-partum years) and delayed effects (occurring beyond one year after delivery). Methods: We conducted an observational cohort study with data from MS patients followed in the OFSEP registry between 1990 and 2020. We included MS female patients aged 18-45 years at MS onset, clinically followed-up for more than 2 years and with ≥3 Expanded Disease Status Scale (EDSS) measurements. Outcomes were the mean EDSS at the end of follow-up and the annual probability of relapse during follow-up. Counterfactual outcomes were predicted using the longitudinal targeted maximum likelihood estimator in the entire study population. The patients exposed to at least one pregnancy during their follow-up were compared with the counterfactual situation in which, contrary to what was observed, they would not have been exposed to any pregnancy. Short-term and delayed effects were analyzed from the first pregnancy of early-exposed patients (who experienced it during their first three years of follow-up). Results: We included 9,100 patients, with a median follow-up duration of 7.8 years, of whom 2,125 (23.4%) patients were exposed to at least one pregnancy. Pregnancy had no significant long-term causal effect on the mean EDSS at 9 years (causal mean difference [95% CI] = 0.00 [-0.16; 0.15]), nor on the annual probability of relapse (causal risk ratio [95% CI] = 0.95 [0.93; 1.38]). For the 1,253 early-exposed patients, pregnancy significantly decreased the probability of relapse during the per-partum year and significantly increased it during the post-partum year, but no significant delayed effect was found on EDSS and relapse rate. Discussion: Using a causal inference approach, we found no evidence of significantly deleterious or beneficial long-term effects of pregnancy on disability. The beneficial effects found in other studies were probably related to a reverse causation bias.