Actionable Insights Newsletter

Translating research to power whole health

Issue 5    |    April 2023

A comparison of death data from multiple sources and the National Death Index

Death data from a combination of sources including Elevance Health may prove to be a reliable source for advancing our research and understanding of treatments and mortality.

Enterprise Analytics Core domain(s): oncology, quality of care

Summary

Background
Although identifying patient death dates is critical for many studies, these data are not systematically recorded in health claims or electronic medical records. The National Death Index (NDI) is considered a “gold standard” source of mortality data in the US. However, using NDI data often involves considerable time and resources. While alternative sources may be used, their sensitivity and accuracy may not be consistent over time.

Objective
This study investigated the consistency of alternative sources of death date data available within Elevance Health with the NDI over time.

Methods
This study utilized data from US advanced cancer patients who were already linked to the NDI between January 2010 and December 2018*. Alternative death date sources, including inpatient discharge status, health plan disenrollment, Social Security Death Master File (DMF), and online obituary data, were compared to the NDI. For each year of the study, we evaluated how accurately the alternative source matched the gold standard (NDI) data by calculating the following:

• The percentage of members with an NDI death date who had a matching date from the alternative source (i.e. sensitivity or Se)
• The percentage of members without an NDI death date and without a death date from the alternative source (i.e. specificity or Sp)
• The percentage of members with a death date from an alternative source with a matching death date in the NDI data (i.e. positive predictive value or PPV)
• The percentage of members without a death date from the alternative source and without an NDI death date (i.e. negative predictive value or NPV)

Results
We identified death dates from any source for 27,396 of 40,692 advanced cancer patients from 2010 to 2018. In 2010, the DMF had a matching date for 77% (95% CI: 76%, 79%) of death dates identified in the NDI data (i.e. moderate sensitivity), but this declined to 13% (95% CI: 11%, 14%) in 2018. For the obituary data, these figures rose from 46% (95% CI: 45%, 48%) in 2010 to 71% (95% CI: 68%, 73%) in 2018. The presence of death date from an alternative source accurately reflected the presence of a death date in the NDI data (high PPV) throughout the study period, and a high percentage of members lacking a death date from one soure (alternative or NDI) lacked a date from the other (high specificity and NPV).

Figure 1. Annual validation metrics of death date data sources available in the HIRD 2010-2018*

*NDI data for 2018 were incomplete.

Key takeaways

• Death data available from the alternate sources varied in the proportion of total deaths defined by the NDI data they captured (sensitivity) over time.
• As we demonstrated in the November 2022 Actionable Insights article, combining multiple sources of death data available within Carelon Research offers a sensitive and accurate alternative when use of NDI data is not possible.