I am Not Dead Yet: Identification of False-Positive Matches to Death Master File

Abstract

Patient death is an important clinical outcome. It is typically ascertained by matching database records with external death indices. Accuracy of the matching algorithms is imperfect.We have investigated whether clinical records made > 1 month after the date of death accurately identify false positive matches to the Death Master File. Positive predictive value (PPV) varied from 74.7% (notes) to 95.9% (labs) and sensitivity from 57.4% (adverse medication reactions) to 94.9% (notes). Presence of any two out of four (billing data, labs, vital signs and medications) data elements had sensitivity of 83.0% and PPV of 98.3%. Area under the ROC curve for a multivariable logistic model that included the number of these four data elements recorded > 1 month after death was 0.987.Clinical data recorded after the date of death can help identify false positive matches to death indices and could be utilized to improve existing record linkage algorithms.

Figure 1

Linkage Algorithm Solid lines indicate current implementation. Dotted lines indicate components of the algorithm evaluated in this study.

Figure 2

ROC Curve for Identification of False Positive Matches to Death Master File