Abstract:
|
Today's healthcare environment is data-rich, with data arising from healthcare claims, medical records and other sources. At the same time, regulatory agencies require that new medications undergo post-marketing assessments to fully understand safety in routine care, with hypotheses formed from pre-approval studies and spontaneous reports. The vast availability of data invites automation, but blind automation can yield non-causal results. We propose a design-driven automated system that draws on experienced investigators to specify key parameters, such as study design, choice of comparator group and biologically-motivated confounding factors, while using cloud-based computing infrastructure to link data, empirically identify additional confounding factors, perform propensity score matching and multivariate adjustment, and rapidly accumulate evidence about the safety of new medications. Results are reported in individual databases and as meta-analyses. This system leverages investigator insights for crucial design choices and interpretation of results, and applies modern, automated infrastructure for rapid and scalable generation of least-biased information about medication safety.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.