Informing modeling and clinical trial simulation using the real world data: data content, quality and availability assessment
Dimple Patel (1), Zhaoling Meng (1), Jeffrey Ming (1), Christine Veyrat-Follet (1), Lei Ma (1), Jeffrey S. Barrett (1)
(1) Translational Informatics, Sanofi, Bridgewater, NJ, USA
Objectives: Model-informed drug discovery and development approaches are increasingly utilized and recognized in the pharmaceutical industry. Clinical trial simulations are essential to leverage models/knowledge accumulated through early stage data and extrapolate the quantitative assessment and decision making to a wider and targeted patient population. To better reflect targeted patient population variability and diversity/heterogeneity,CTS often utilize historical clinical data. Real world databases have the advantages of containing large number of patients and information content, and they become even more valuable when historical clinical databases are limited by trial specific enrollment criteria selection or transition is needed from the clinical study setting to a real world study setting. We aim to investigate whether real world databases, such as EMR or EHR, provide another valuable source of such information.
Methods: We took into account that the real world data usage could be constrained by the nature of its data reporting. The lack of systematic and comprehensive collection of measurements required can lead to either inconclusive or biased population representation. Also, data quality and lack of cleaning could be of concern. To assess these issues, we investigated whether event data obtained from a real world database would be comparable to data obtained from a clinical trial. Data from a recent outcome study[1] for empagliflozin, an SGLT2 inhibitor, was used in this investigation. The study enrollment criteria were used to identify relevant patients in a GE Quintiles EMR diabetes patient database consisting of approximately 3.8 million patients. Patients in the database with medical records within the clinical study timeframe were considered for the inclusion. Similar follow-up period in the database as the study exposure time was used. Relevant patient event records were reviewed and regrouped to match the publication events using ICD9 and ICD10 codes.
Results: The populations were matched to the extent possible on demographic factors and baseline parameters. Selected patient characteristics and cardiovascular event rates (myocardial infarction, stroke and CV death) and overall death rate were summarized and compared to the empagliflozin publication results. % missing data for relevant enrollment criteria, such as eGFR and HbA1c, were summarized.
Conclusions: The differences were attributed to information uncertainty, incomplete data, and population differences. We need to proceed with caution on the utilization of RW data in CTS.
References:
[1] Empagliflozin, Cardiovascular Outcomes, and Mortality in Type 2 Diabetes. Zinman B, et. Al. N Engl J Med 2015; 373:2117-2128.