For many health conditions, medication is usually prescribed in specific quantities at regular intervals to achieve safe and effective use for best health outcomes. Not following evidence-based recommendations can have serious negative consequences for patients, health systems, and the society. Electronic healthcare data are nowadays routinely used to estimate whether patients follow the medication prescribed by their healthcare provider, with delays in obtaining a new medication prescription or dispensation being essentially interpreted as clues that patients took less medication than prescribed. These data represent an increasingly accessible source of information, and databases are usually accompanied by advice on data extraction and preparation. Yet, once the data are ready for analysis, performing the actual computations correctly is not as simple as comparing the quantity prescribed with the quantity obtained on a given interval, as the validity of the computation depends on the extent to which the methods fit the clinical context of the medication studied and adherence theory in general. Moreover, its reliability depends on carefully recording the analysis choices and on providing transparent protocols that other analysts can reproduce. Without valid and reliable methods of computing adherence, there is no way to know if we can trust (and compare) the myriad studies that use these estimates.
But how do we actually calculate adherence in a valid and reliable manner? Alexandra Dima (University Claude Bernard Lyon 1 – HESPER, and University of Amsterdam – ASCoR) was confronted with this problem while reviewing the literature on electronic healthcare data and being surprised by the lack of transparency and replicability in this area, in contrast to the widespread sharing of algorithms and computer code that is becoming the rule in science. If only there were algorithms available to any analyst, to replicate prior studies, record their own decisions, perform sensitivity analyses… and even better, wouldn’t it be great to have interactive visualizations of medication histories that analysts can use in discussions with clinical experts to decide which calculations fit their specific clinical reality better? She decided to develop these algorithms together with Dan Dediu (Max Planck Institute for Psycholinguistics, Nijmegen), a computational linguist, statistician and software developer.
The result of this collaboration is a new R package, AdhereR, freely available for use and further development. The article introducing the software is now published in PLOS ONE (DOI:10.1371/journal.pone.0174426), and describes the methods implemented and giving practical examples of use. The package can be installed from the standard R repository for packages (CRAN). Its source code is available on GitHub, and it contains detailed help and examples, allowing the interactive exploration of medication histories and the efficient computation of several adherence measures in large databases.
Vous pouvez retrouver l'intégralité de l'article au niveau du lien ci dessous :