1C-06 - Increased spontaneous abortion occurrence among treated and untreated groups: is this the Will Rogers phenomenon?

Background: Migraine prevalence in the general population is reported between 4% and 17%. A claims- or electronic medical record–based algorithm that identifies a clinically recognized/treated subset of cases (that tend to be more severe) will result in a lower prevalence. In pregnancy, migraine has been associated with an increased risk for spontaneous abortion (SAb).

Objectives: To explore SAb prevalence in groups determined by migraine diagnosis and migraine treatment using a migraine-identifying algorithm that preferentially identifies severe cases.

Methods: We simulated a cohort of 5,000 pregnant women, 10% of whom had migraine (3%, severe). SAb was randomly distributed with probability 16% in women without migraine, and a 3 and 1.5 times higher probability in women with severe and mild migraine, respectively. Drug A (new) was used by 30% of women with severe migraine and 10% of women with mild migraine. The older active comparator B was used by 50% of women with severe migraine and 30% with mild migraine. Women used either A or B, but not both; neither drug increased the risk of SAb.
Within the cohort, we simulated a migraine-identifying algorithm that resulted in a migraine prevalence of 4%, 6%, 8% or 10%, including 75% or 100% of severe cases (8 scenarios). The algorithm had no false positives.
We used these parameters to simulate 1,000 different cohorts and replicated the analyses removing the effect of migraine on SAb. We present prevalence and percentile-based 95% CIs for SAb by drug use and migraine.

Results: The true (simulated) SAb prevalence in patients treated with drug A was 37.4% (95% CI, 26.9%-48.3%); with drug B, 34.0% (27.0%-41.2%), and 16.0% (14.8%-17.1%) in non-migraineurs (all untreated).
Among algorithm-identified migraineurs, the prevalence point estimates for the different scenarios were 35.7%-45.6% (drug A), 32.4%-43.9% (drug B), 16.0%-16.6% (some algorithm-identified non-migraineurs were treated), and 16.0%-16.3% (restricted to untreated algorithm-identified non-migraineurs). The highest SAb prevalences occurred when the algorithm-identified migraine prevalence was 4%. CIs were wider than in the simulated truth.
Without an effect of migraine on SAb, the true (simulated) SAb prevalence was 15.9% for migraineurs treated with A or B and 16.0% for non-migraineurs. Among algorithm-identified migraine patients, SAb prevalence was 15.8%-16.0% in all groups and scenarios.

Conclusions: When migraine caused SAb, the observed SAb prevalence was increased in most scenarios and indication/exposure groups because the algorithm classified mild migraineurs as non-migraineurs, and mild migraineurs had intermediate SAb risk. This situation is not specific to pregnancy outcomes and resembles the Will Rogers phenomenon.