This is the third post in our four-part taXonomy X blog series (part 2 here),
Today, we’ll focus on another high-impact endpoint in real-world evidence (RWE): mortality. Specifically, we’ll discuss how mortality data are incorporated into taXonomy, why post-mortem claims activity must be carefully addressed, and provide supporting evidence via sensitivity analyses against CDC reported mortality trends to demonstrate the reliability of mortality data in taXonomy.
Mortality is one of the most critical endpoints in RWE. It is clinically meaningful, methodologically sensitive and foundational to generating high-quality insights. Accurate mortality directly shapes cohort eligibility, informs censoring decisions and defines follow-up requirements. Ultimately this influences how researchers interpret time-to-event outcomes, treatment persistence and healthcare resource utilization.
Because of this, mortality measurement must be reliable. Even modest misclassification can bias survival estimates and distort utilization and cost trajectories. All of these are important considerations to ensure teams are producing strong study evidence and the importance becomes heightened when results are reviewed by external clinical, payer, and regulatory stakeholders.
taXonomy’s mortality offering is powered by the Veritas Data Research Fact of Death dataset, a comprehensive U.S. mortality index spanning from 1935 to present and containing 177M+ death records. Veritas compiles mortality from 40,000+ national, regional and local public and private sources, including the Social Security Administration’s Limited Access Death Master File and other publicly available death-related records.
A key strength of how the dataset is assembled is source diversity and cross-validation. Veritas consolidates, de-duplicates and aggregates death attributes into a single curated index, with a significant share of records supported by more than one source. Veritas also applies rigorous data quality and standardization processes, including, but not limited to: date standardization, name and geography validation and scheduled evaluation of records that do not meet quality thresholds.
Together, this multi-source assembly and rigorous standardization framework increases confidence that mortality events are accurate and consistently represented for downstream research use.
Even with a robust underlying mortality index, claims data introduces a distinct real-world challenge: post-mortem activity (PMA). PMA exists when a patient’s claims appear after their recorded death. However, this phenomenon often occurs for legitimate operational reasons, such as adjudication timing, claim adjustments, billing cycles and coordination of benefits.
If post-mortem claims activity is not appropriately controlled for, patients may be misclassified, either censored too early or labeled deceased, despite utilization patterns that are inconsistent with death. This can directly impact feasibility, validity, and reproducibility in longitudinal research and necessitates proper evaluation to bolster study designs.
To mitigate patient misclassification, HealthVerity applies PMA logic specifically designed to reduce the inclusion of invalid mortality events. At a high level, the logic measures post-mortem activity in days relative to the death month, building in the assumption that death occurs on the 15th of the month to establish consistency in margin of error. By default, if medical or pharmacy claims activity is detected more than 60 days after the 15th of the death month, the mortality indicator is removed and the patient’s claim activity is retained. This serves as a pragmatic quality control step, grounded in research findings, to ensure mortality indicators used in research are not contradicted by downstream claims activity.
A key methodological consideration is properly defining a post-mortem activity window. If the window is too short, the logic can over-correct, resulting in removal of valid mortality signals by not properly controlling for routine claims processing lag, for example. If it is too long however, under-correction may occur and allow for invalid mortality events to persist in the data.
To determine an appropriate PMA window, HealthVerity performed sensitivity analyses using 30-, 60-, and 90-day PMA windows to assess the level of agreement between each respective mortality rate with CDC-recorded deaths, widely viewed as the gold standard reference for population mortality statistics
The findings demonstrated that the 30-day window was more likely to over-correct and remove mortality events that were otherwise consistent with expected claims lag, whereas a 60-day window was associated with the best overall performance, achieving strong alignment with the CDCs reported trends. Extending the window beyond 60 days did not materially improve CDC alignment and created more opportunity for under-correction by allowing additional records to remain flagged as deceased even when claims activity persisted well beyond the death month.
Further, when examining the data month-over-month, taXonomy mortality rates mirrored CDC mortality rates with closely aligned trends in both direction and magnitude (Figure 1).
Figure 1: taXonomy mortality rates mirror CDC mortality rates with closely aligned trends in both direction and magnitude. CDC mortality rates month over month were calculated using the CDC’s Wonder Database
Across the months where both measures are available, taXonomy closely tracked CDC month-over-month mortality, with an average absolute difference of 0.0027 percentage points (RMSE 0.0052 percentage points) and a month-to-month correlation of r = 0.91. Even at its largest deviation, the gap was 0.010 percentage points in a single month, supporting that the taXonomy mortality signal is highly consistent with CDC-reported patterns at the population level.
This matters because mortality is foundational to high-quality RWE generation to strengthen survival analyses, improve the defensibility of follow-up and censoring logic and reduce the risk that downstream outcomes, utilization and cost findings are driven by misclassified patients from invalid death events.
taXonomy’s mortality methodology, paired with PMA logic and validation against CDC mortality patterns, supports reliable and reproducible RWE generation for life sciences organizations and the researchers working on their behalf.