Non-randomized routine care data offers many opportunities to study the effectiveness of therapeutic interventions in less controlled environments. However, beyond the well-known concerns about exposure-related bias, the analysis of non-randomized routine care data may be prone to bias due to informative missingness of relevant patient outcomes. This situation can, for instance, arise when outcomes are assessed at irregular visit schedules.
In this talk, I will describe and illustrate (a selection of) common methods for estimating comparative treatment effects from routine care registry data when outcomes are assessed at irregular visit schedules. These methods transform the patient visits into equally spaced observations and adopt (simple) imputation methods to account for the presence of missing outcome data. Subsequently, I discuss how multilevel models can be used to estimate the individual patients' disease trajectories, and to alleviate the need for generating equally spaced observations.
Each method is illustrated using real world data from a multi-center registry of patients that suffer from multiple sclerosis (MS). MS is a chronic progressive disorder that affects approximately 2.3 million people worldwide. Although the efficacy of disease-modifying therapies has previously been studied in multiple randomize trials, real-world evidence can provide valuable insight into the effectiveness in routine medical practice, outside the structured clinical trial settings.
Subsequently, all methods are evaluated in an extensive simulation study, where we mimic routine care data from patients suffering from multiple sclerosis (MS). Individual disease trajectories were generated for MS patients from multiple practices. Patient outcomes were expressed as Expanded Disability Status Scale (EDSS), a standard reference scale to assess progression of MS disease, and generated for distinct months using multilevel normal distributions correlated over time. To mimic the irregular visit times in clinical practice, patient outcomes were censored according to an informative missingness procedure. For each simulated dataset, treatment effect estimates were estimated in terms of time to confirmed EDSS progression at 6 months, as defined in clinical practice.
This work was supported by Biogen (Cambridge, MA, USA).