Background: Mortality is an important outcome for post-market drug safety and other longitudinal studies. Health plans typically capture medically attended deaths, but not those occurring out of hospital or outside plan enrollment. Linkage of health plans to state death registries can support mortality studies by integrating data on death status, date and cause, but the validity of linkages is often not documented.
Objectives: To examine the validity of a probabilistic linkage between Tennessee (TN) Medicaid (TennCare) data and TN death certificates using National Death Index (NDI) linked TennCare records as reference.
Methods: Beginning with 2001 enrollment, records of TennCare patients undergo annual probabilistic linkage to TN death certificates. We submitted records for a sample of 4753 patients to NDI for national-level death certificate linkage. These patients either linked to a TN death certificate (n = 3505) or were presumed alive and had disenrolled from TennCare as of 12/31/2016 (n = 1248). NDI classified records as “presumed dead” (reflecting a high-quality match between record and death certificate), or “presumed alive” (a low-quality match or non-matching record). We examined the sensitivity (SEN), specificity (SPE), positive (PPV) and negative predictive values (NPV) of death status according to TN death certificate linked records using NDI-determined deaths status as reference. We compared date of death and underlying cause of death codes between TN and NDI-linked death certificates.
Results: NDI identified possible matches for 4714 of 4753 patients, 3591 (76%) of which were high-quality matches, i.e., deaths. All 3505 patients that linked to TN death certificates were also identified by NDI as deaths (PPV = 100%); 36 (1%) of these were TN resident deaths that occurred outside TN and 25 (<1%) were non-TN residents at time of death but died in TN. The TN linkage missed 86 deaths identified by NDI (SEN = 98%). Having both residence and state of death outside TN was a primary reason for discordance (65 of 86 deaths [76%]). The SPE and NPV of the TN death certificate linkage were 100% and 93%, respectively. Of deaths identified in both sources, death date and cause of death codes were identical for 3505 (100%) and 3502 (>99%), respectively.
Conclusions: Our probabilistic state-based record linkage performs well as compared to NDI, but does not capture all deaths, particularly those occurring among TennCare enrollees who moved and died outside of TN. Additional investigation is needed to understand other factors contributing to potentially missed linkages.
