The Authors Respond

    loading  Checking for direct PDF access through Ovid


We appreciate points made by Root et al1 in “Re: Biomedical Informatics Approaches to Identifying Drug-Drug Interactions: Application to Insulin Secretagogues.”1 Because we used the self-controlled case series design2 to screen a large number of potential drug–drug interactions rather than to examine individual etiologic hypotheses, we needed to simplify the procedure to maximize logistical efficiency. Some of those simplifications may indeed have violated the underpinnings of the self-controlled case series design.3 When this design and its extensions are used to inform individual etiologic hypotheses, we agree that investigators should quantify and address potential violations of the underlying assumptions.4 We appreciate the opportunity to clarify the methodologic decisions that we made when using the self-controlled case series design in a screening mode.
A self-controlled case series study requires great rigor in operationally defining the observation period for each subject.5 There is considerable guidance in the literature for doing so for self-controlled case series studies of vaccines,3,5–9 but less for studies of individual drugs,3,7,10 and none for studies of drug–drug interactions—absent a few applied examples.11–13 Given that we screened for interactions causing an exaggerated pharmacologic effect of insulin secretagogues (i.e., serious hypoglycemia), it seemed intuitive to define each subject’s observation time by a continuous period of exposure to the secretagogue of interest (i.e., object drug in a drug–drug interaction14). Only during such periods would concomitant exposure to a potentially interacting (i.e., precipitant14) drug and serious hypoglycemia caused by that secretagogue be possible. Yet, we agree that a serious hypoglycemic episode may precipitate an immediate discontinuation of the study secretagogue, functionally leading to event-dependent censoring—an occurrence that may produce unpredictable bias.5 Etiologic self-controlled case series studies to investigate our method’s signals would indeed benefit from Farrington et al’s self-controlled case series extension to handle event-dependent censoring.10
Finally, we appreciate Root et al’s concern that prescriber awareness of a potential drug–drug interaction may curtail or perturb postevent exposure (i.e., concomitant exposure to the object and precipitant drugs).7 In our use of the self-controlled case series design for screening purposes, this was of secondary concern because (1) prescriber knowledge of drug–drug interactions is generally poor;15 and (2) if such interactions were “known” to prescribers, falsely amplified signals generated for these drug pairs would be unlikely to trigger an etiologic study. Nevertheless, we agree that investigators using the self-controlled case series design should consider the impact of interferent events and, if necessary, employ Farrington et al’s approach to impose counterfactuals to render unobserved exposure histories determinate.
    loading  Loading Related Articles