Identifying the DEAD: Development and Validation of a Patient-Level Model to Predict Death Status in Population-Level Claims Data

Jenna M. Reps (), Peter R. Rijnbeek and Patrick B. Ryan
Additional contact information
Jenna M. Reps: Janssen Research and Development
Peter R. Rijnbeek: Erasmus MC
Patrick B. Ryan: Janssen Research and Development

Drug Safety, 2019, vol. 42, issue 11, No 9, 1377-1386

Abstract: Abstract Introduction US claims data contain medical data on large heterogeneous populations and are excellent sources for medical research. Some claims data do not contain complete death records, limiting their use for mortality or mortality-related studies. A model to predict whether a patient died at the end of the follow-up time (referred to as the end of observation) is needed to enable mortality-related studies. Objective The objective of this study was to develop a patient-level model to predict whether the end of observation was due to death in US claims data. Methods We used a claims dataset with full death records, Optum© De-Identified Clinformatics® Data-Mart-Database—Date of Death mapped to the Observational Medical Outcome Partnership common data model, to develop a model that classifies the end of observations into death or non-death. A regularized logistic regression was trained using 88,514 predictors (recorded within the prior 365 or 30 days) and externally validated by applying the model to three US claims datasets. Results Approximately 25 in 1000 end of observations in Optum are due to death. The Discriminating End of observation into Alive and Dead (DEAD) model obtained an area under the receiver operating characteristic curve of 0.986. When defining death as a predicted risk of > 0.5, only 2% of the end of observations were predicted to be due to death and the model obtained a sensitivity of 62% and a positive predictive value of 74.8%. The external validation showed the model was transportable, with area under the receiver operating characteristic curves ranging between 0.951 and 0.995 across the US claims databases. Conclusions US claims data often lack complete death records. The DEAD model can be used to impute death at various sensitivity, specificity, or positive predictive values depending on the use of the model. The DEAD model can be readily applied to any observational healthcare database mapped to the Observational Medical Outcome Partnership common data model and is available from https://github.com/OHDSI/StudyProtocolSandbox/tree/master/DeadModel .

Date: 2019
References: View complete reference list from CitEc
Citations:

Downloads: (external link)
http://link.springer.com/10.1007/s40264-019-00827-0 Abstract (text/html)
Access to the full text of the articles in this series is restricted.

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:spr:drugsa:v:42:y:2019:i:11:d:10.1007_s40264-019-00827-0

Ordering information: This journal article can be ordered from
http://www.springer.com/adis/journal/40264

DOI: 10.1007/s40264-019-00827-0

Access Statistics for this article

Drug Safety is currently edited by Nitin Joshi

More articles in Drug Safety from Springer
Bibliographic data for series maintained by Sonal Shukla () and Springer Nature Abstracting and Indexing ().