While many adverse drug reactions (ADRs) are identified during drug development, others remain elusive until after widespread exposure in clinical populations. Identifying novel ADRs involves a two-step process: (1) signal detection and (2) signal refinement. Signal detection is a hypothesis generating phase using data mining methods to suggest potential ADRs. Signal refinement is a hypothesis testing phase to confirm or reject the causal link through more rigorous methodology. Signal detection has historically relied on analysis of spontaneously reported ADRs (e.g. FDA MedWatch). However, administrative healthcare databases are now being explored as a complementary data source and novel analysis methods are needed.
The underlying objective is to determine whether a core pharmacoepidemiology study design used for signal refinement, prescription sequence symmetry analysis (PSSA), can be successfully adapted for signal detection using national VHA administrative data. The specific aims are to: (1) determine whether signals can be identified using the adapted PSSA method, among a selected set of medications where novel adverse drug reactions have been discovered post-marketing; (2) examine the impact of varying key parameters of the modified PSSA method and determine whether an optimal set of values can be identified; (3) assess the influence of key PSSA weaknesses, including surveillance bias and detection bias.
We will adapt the PSSA method for signal detection by automating the procedure beyond examining a single, a priori hypothesized ADR, to instead generate measures of association for all events that can be identified using pharmacy and medical encounter data. We have selected 7 medications where novel ADRs were identified post-marketing (e.g. rofecoxib and myocardial infarction/stroke). In Aim 1, we will apply our modified PSSA method to build ADR signal profiles for these medications, using national VHA pharmacy and medical encounter data to identify events, which will generate association measures (sequence symmetry rate ratios) for all possible incident drug and diagnostic events. We will examine the profile for each drug (e.g. rofecoxib) to determine whether a signal for the event of interest (e.g. myocardial infarction/stroke) was detected. We will further assess whether this signal stood out within the overall ADR profile as an obvious target for subsequent refinement studies. In Aim 2, we will conduct a series of sensitivity analyses, each varying one of several key design parameters. These findings will be used to assess which parameters have the greatest impact on the resulting ADR signal profiles, and determine if consistent, optimal parameters can be identified. In Aim 3, we will examine the influence of detection bias through sensitivity analyses varying the censoring window, a key PSSA methods parameter. We will further study the impact of surveillance bias by calculating sequence symmetry ratios for several variables related to the quantity of healthcare services received (e.g. count of outpatient refills, count of outpatient encounters, etc.). These findings will be used to determine whether a statistical adjustment based on these variables could be developed and applied to outcome event signals to minimize the potential impact of these biases.
There are no findings at this time.
Innovations of this work include the application of administrative healthcare databases for ADR signal detection and most importantly, we are the first to propose this adaptation of PSSA methodology as a basis for ADR signal detection. If this approach proves effective, it could greatly improve the efficiency of existing ADR detection programs, identify novel ADRs earlier, and thereby prevent unnecessary morbidity and mortality in Veterans receiving newly marketed medications.
None at this time.
Technology Development and Assessment, TRL - Applied/Translational, Treatment - Observational
Healthcare Algorithms, Medication Management, Surveillance