Keywords: Database studies, Selection bias, Epidemiological Methods, Cohort studies, Research design
Studies using Medicaid data can be challenging for researchers, as lapses in coverage are not uncommon. Setting various thresholds for enrollment (e.g. percent of days enrolled during a period of study, or by setting an upper limit on number of days of lapsed coverage) can bias study findings and limit generalizability. In commercial claims data, upon disenrollment a subject will no longer be able to be followed for research studies. However with Medicaid data, enrollment can vary due to income fluctuations, resulting in intermittent periods of enrollment and still allow patients to be followed longitudinally. We study the effects of using a range for the enrollment study inclusion criterion—from constant enrollment to allowing for up to 90-day gaps in enrollment—to ascertain the sensitivity of outcomes to changes in this criterion. To illustrate these differences we will use a cohort of Medicaid pediatric asthma patients that are propensity score matched to a group of non-asthmatic Medicaid patients to compare diverse outcomes, including emergency department utilization rates and cost of care. Focus will be given to how outcomes vary due to changes which affect study enrollment.