Newly marketed medications present unique challenges for nonrandomized comparative effectiveness analyses

Abstract

The vast majority of medicines that come onto the market have met US FDA requirements for efficacy and safety. Of those, only half may have randomized evi dence comparing the new agent to existing drugs, and of that group, only a small fraction may have fully evaluated a new drug’s relative effectiveness [1]. Clinicians, insurers, policy-makers and, ultimately, patients require timely results from postmarketing research to ascertain whether newly marketed drugs offer benefits in effectiveness and safety over existing standards of care. For this, nonrandomized (observational) studies using secondary healthcare data are key for assessment of drugs as they are used in routine care. The general methodological challenges of performing comparative effectiveness research with nonrandomized data are magnified during this early marketing phase [2]. Challenges arise from one or more of the following sources: bias in initial treatment effects due to the differential channeling of sicker patients who may have waited for the arrival of the new drug in hope of improved effectiveness or safety; bias and treatment effect heterogeneity due to changing patient characteristics of the early users of a new drug compared with the broadening user base over the longer term; and the consequences of the potentially slow uptake of a new drug, leading to small numbers of users and thus imprecise effect estimates [3]. Perhaps the most serious of these challenges is the bias due to the differential nature of the early users of a new drug compared with the users of an existing standard of care. Early users of a new drug are likely to be a mix of two types of patients: those with an existing diagnosis that failed an earlier therapy or suffered from intolerable side effects, and those whose insurers and physicians are willing to initiate a newly diagnosed patient on a novel therapy. With the first group, the channeling of sicker patients – those with more severe disease and a poorer prognosis – toward the new drug will introduce confounding bias similar to that found in usual pharmacoepidemiology studies [4], but probably with even greater magnitude. With the second group, biases may result from physician- and systemlevel factors that go beyond the patient’s condition and characteristics. Those patients with generous insurance plans or doctors attuned to new develop ments in the drug market may be differential with respect to socioeconomic status, quality of care and other factors known to affect patients’ risk of a variety of outcomes. On another level, this heterogeneity of the patient population may lead to a heterogeneity of treatment effects; those patients that have failed earlier therapies may be less likely to be treated effectively with the new medication (or any second-line therapy, for that matter). Although profound, these biases, in many instances,