Study Design
Treatment assignment in observational analyses – a practical application in the Swedish Primary Care Cardiovascular Database Anna Humphreys* Anna Humphreys Anthony Matthews Anita Berglund Jessica Young Bertil Lindahl Björn Wettermark Thomas Kahan Miguel Hernán
Introduction
In randomized trials, the intention-to-treat effect is the effect of assignment to a treatment strategy. When using observational data to emulate a target trial, estimating the intention-to-treat effect requires treatment strategy assignment to be defined. While the prescription of a treatment is a natural analog of treatment assignment, many databases only contain treatment dispensation data.
Objectives
We aimed to assess the implications of using prescription versus dispensation data to define treatment assignment in observational analyses by emulating two target trials that estimate the effect of Angiotensin Converting Enzyme Inhibitors (ACEI) and Angiotensin Receptor Blockers (ARB) versus Calcium Channel Blockers (CCB) on the five-year risk of cancer.
Methods
We used the Swedish Primary Care Cardiovascular Database (SPCCD), a database composed of electronic health records linked with national registers for more than 75,000 primary care patients with hypertension. SPCCD uniquely contains both prescription and dispensation data, and thus the first emulation used prescription as an analog for assignment, and the second used dispensation. The confounding structure differed between emulations, with the dispensing emulation also adjusting for education, birth country, income and medication history.
Results
In both the prescribing emulation and dispensing emulation, no difference was found between ACEI/ARB and CCB on the five-year risk of cancer (RR 0.92 (95% CI 0.78 – 1.10), and RR 1.00 (95%CI 0.80-1.21) respectively).
Conclusion
The use of prescription or dispensation data to define treatment assignment must be considered when interpreting effect estimates from observational analyses, as the effect of a treatment strategy may vary depending on how treatment is assigned and whether the differing adjustment variables are sufficient. While similar effect estimates were seen, there may be situations (eg treatments with acute effects) with greater implications.