When treating patients, healthcare providers (e.g., doctors, nurses, pharmacists) need to weigh the potential health benefits and risks of the medications they are prescribing. Often times, healthcare providers must decide whether to continue patients on a medication they are already taking versus switching them to a new medication. To answer this question, researchers developed the Prevalent New User1 study design (PNUD). This study design has several advantages. One important advantage is that this study design allows researchers to include a larger patient population. This design also allows researchers to compare patients who continue a medication to those who switch from that medication to a new treatment. This design has important implications broadening the scope of research possible in pharmacoepidemiology; however, applying this design in practice can be difficult, requires substantial computing resources. To address this problem, our study will examine several alternative methods that may overcome the technical difficulties of applying the original study design. The overall goal of this study is to compare these alternative methods to the original PNUD study design and to each other. This will help us understand if there is a more efficient way to implement the PNUD design. This study is intended to serve as a teaching example and will include detailed information on its methods to make this type of research more accessible for the research community.
It is often useful to compare outcomes among patients continuing therapy with a given agent to those of patients that initiate therapy with another, potentially newer, agent. Switching to another therapeutic agent maybe associated with disease progression (or lack of therapeutic effect/side effects of the prior agent) and can be influenced by treatment guidelines and formulary design. This scenario can be handled using the prevalent-new-user study design (PNUD). Whereas the traditional new-user design is restricted to true new users, the PNUD includes patients switching treatment, thereby increasing study generalizability. This design may also better mimic the population seen in clinical practice. However, implementing this design can be hindered by computational difficulties stemming from the confounding control approach. The PNUD was originally implemented using a resource-intensive matching approach. Alternative matching approaches and other confounding control methods may overcome the technical problems faced with the original matching approach. This study will build on a recent simulation study comparing multiple propensity score matching approaches to each other as well as to time-stratified standardized morbidity ratio weighting and outcome modeling using the G-formula. Using CPRD data, this study will replicate the original PNUD matching approach and compare a subset of the confounding control methods considered in the simulation study. This study will contribute an empirical application comparing multiple confounding control approaches, thereby advancing our understanding of implementation of the PNUD using real world data. Our empirical application will be the same as in the original PNUD study: we will compare exposure to two anti-diabetic drugs (GLP-1 analogs and sulfonylureas) in relation to the hazard of hospitalization for heart failure. This important clinical question can help clinicians decide whether it would be better for their patients to switch to GLP-1 analogs or remain on sulfonylureas.
Outcomes to be measured
For the primary objective, the two outcomes of interest are: (1) indicators of matching success, and (2)
computation time. Indicators of matching success will include (a) proportion of patients where matches are found, and (b) balance metrics. Balance metrics will be assessed using standardized mean differences (SMDs) for the confounders and time since starting a GLP-1 analog, contrasting the cohort of switchers to GLP-1 with the cohort of sulfonylureas continuers. Computation time will be assessed on a remote computer running the program with no other programs running in the background or by running programs in parallel.
For the secondary objective, the primary outcome of interest will be hospitalization for heart failure. This outcome will be identified using inpatient diagnostic codes through the HES linkage. In CPRD, heart failure will be defined using International Classification of Diseases, 10th Revision codes (ICD-10 [I50.x]). For patients with no history of heart failure, the event definition will require a diagnosis in the primary or secondary position for the hospital stay. For patients with a history of heart failure, the event definition will exclude heart failure as a secondary diagnosis.
For the secondary objective, the secondary outcome of interest will be heart failure, assessed using outpatient diagnostic codes through the HES linkage. In CPRD, heart failure will be defined using the ICD-10 code I50.x or any of the following Read codes (G1yz100, G58..00, G580.00, G580000, G580100, G580.11, G580.12, G580.13, G580.14, G580200, G50300, G580400, G581.00, G581000, G58..11, G581.11, G581.12, G581.13, G582.00, G583.00, G583.11, G583.12, G584.00, G58z.00, G58z.11, G58z.12).4 For patients with no history of heart failure, the event definition will require a diagnosis in the primary or secondary position for the hospital stay. For patients with a history of heart failure, the event definition will exclude heart failure as a secondary diagnosis.
Lastly, for the exploratory objective, the outcome of interest will be the same as for the secondary objective but the analysis will be conducted using an intention-to-treat (ITT) approach instead of an As-Treated approach.
Panagiotis Mavros - Chief Investigator - Janssen Scientific Affairs, LLC
Shahar Shmuel - Corresponding Applicant - University of North Carolina at Chapel Hill
Cynthia Girman - Collaborator - CERobs Consulting
Elizabeth Garry - Collaborator - Aetion, Inc
Jerrod Nelms - Collaborator - Janssen Scientific Affairs, LLC
Jessica Young - Collaborator - University of North Carolina at Chapel Hill
Michael Webster-Clark - Collaborator - University of North Carolina at Chapel Hill
Til Stürmer - Collaborator - University of North Carolina at Chapel Hill
Jerrod Nelms - Collaborator - Lucyna Solutions
HES Accident and Emergency;HES Admitted Patient Care;HES Diagnostic Imaging Dataset;HES Outpatient