Abstract:
|
‘Real world’ data analyses (non-randomized administrative and clinical healthcare databases) have become a vital tool, providing key insights for regulators, payers, and other healthcare decision makers. However, conducting causal inference for database studies can be complex. Implementation of a database study can involve a great deal of data manipulation to create appropriately temporally anchored analytic cohorts from data that were not collected for research purposes. The many subtle design and analysis decisions have left some stakeholders concerned about selective reporting of study specifications with favorable results.
Healthcare decision-makers have emphasized a need for greater transparency of research making secondary use of databases. The REPEAT Initiative embarked on projects aimed at improving transparency, reproducibility and validity of database research. These include large scale replication of 150 published database studies, evaluation of robustness of results for 50 studies to alternative design and implementation parameters, and development of structured reporting with design visualization to increase transparency and minimize misinterpretation.
|