EVALUATING THE ROBUSTNESS OF TIME-TO-EVENT ANALYSES IN DE-IDENTIFIED REAL-WORLD DATASET USING A PROXY DATE OF DEATH ALGORITHM: A VALIDATION STUDY IN PATIENTS WITH MULTIPLE MYELOMA
Author(s)
Elisabet Viayna, PhD1, Rahul Rajkumar, MD2, Dan Wheedleton, MPP3, Jiani Zhou, PhD, MSPH4, Jonathan Lilley, BS2, Marley Boyd, MS3, Elisabeth Calderón-Gómez, PharmD, PhD5, Jeffrey S. Brown, PhD3.
1Global HEOR & RWE, Instituto Grifols S.A. Sant Cugat del Vallès, Barcelona, Spain, 2Boston Strategic Partners, Boston, MA, USA, 3TriNetX, LLC., Cambridge, MA, USA, 4Global HEOR & RWE, Grifols SSNA., Durham, NC, USA, 5Global Medical Affairs, Instituto Grifols S.A. Sant Cugat del Vallès, Barcelona, Spain.
1Global HEOR & RWE, Instituto Grifols S.A. Sant Cugat del Vallès, Barcelona, Spain, 2Boston Strategic Partners, Boston, MA, USA, 3TriNetX, LLC., Cambridge, MA, USA, 4Global HEOR & RWE, Grifols SSNA., Durham, NC, USA, 5Global Medical Affairs, Instituto Grifols S.A. Sant Cugat del Vallès, Barcelona, Spain.
OBJECTIVES: Accurate date of death (DoD) ascertainment is critical for time-to-event analyses in real-world studies; however, many databases provide death data in partially de-identified formats (e.g., month/year). This study evaluated the performance of a pragmatic proxy DoD algorithm against full DoD records by comparing severe infection-free survival (IFS) and overall survival (OS) estimates.
METHODS: This retrospective study utilized TriNetX Dataworks-USA to identify adults with multiple myeloma initiating B-cell maturation antigen bispecific antibodies (BsAbs) between Jun-2021 and Jun-2025. Patients initiating immunoglobulin-replacement therapy (IgRT) within 60 days after index BsAbs initiation without severe infection evidence were considered exposed to primary IgRT prophylaxis; remaining patients were classified as unexposed. To mitigate immortal-time bias, IgRT was modeled as a time-dependent exposure. IFS (severe infection or death) and OS were assessed using the Simon-Makuch method with Mantel-Byar comparison, and hazard ratios (HR) were estimated using Cox proportional hazards models. When only month-year of death was available, a proxy DoD was assigned as either the last encounter discharge date within the month of death, or the last day of the month. Proxy-derived survival estimates were compared against analyses using full DoD records.
RESULTS: There was 100% concordance between mortality status and death month-year. The mean difference between actual and proxy DoD was −10 days (median −7; IQR −1-to-0; range −30-to-19). Inference was consistent across approaches. Both Mantel-Byar comparisons remained statistically significant for IFS (0.015 vs. 0.024) and OS (0.0086 vs. 0.0098). Cox models yielded near-identical estimates: HR 0.72 vs. 0.73 for IFS (both p=0.02) and 0.48 vs. 0.46 for OS (both p=0.01).
CONCLUSIONS: In this study of adult patients with multiple myeloma, the proxy DoD algorithm using month-year of death produced survival results practically identical to those derived from full DoD records, confirming the robustness of mortality-related analyses limited to month-year granularity.
METHODS: This retrospective study utilized TriNetX Dataworks-USA to identify adults with multiple myeloma initiating B-cell maturation antigen bispecific antibodies (BsAbs) between Jun-2021 and Jun-2025. Patients initiating immunoglobulin-replacement therapy (IgRT) within 60 days after index BsAbs initiation without severe infection evidence were considered exposed to primary IgRT prophylaxis; remaining patients were classified as unexposed. To mitigate immortal-time bias, IgRT was modeled as a time-dependent exposure. IFS (severe infection or death) and OS were assessed using the Simon-Makuch method with Mantel-Byar comparison, and hazard ratios (HR) were estimated using Cox proportional hazards models. When only month-year of death was available, a proxy DoD was assigned as either the last encounter discharge date within the month of death, or the last day of the month. Proxy-derived survival estimates were compared against analyses using full DoD records.
RESULTS: There was 100% concordance between mortality status and death month-year. The mean difference between actual and proxy DoD was −10 days (median −7; IQR −1-to-0; range −30-to-19). Inference was consistent across approaches. Both Mantel-Byar comparisons remained statistically significant for IFS (0.015 vs. 0.024) and OS (0.0086 vs. 0.0098). Cox models yielded near-identical estimates: HR 0.72 vs. 0.73 for IFS (both p=0.02) and 0.48 vs. 0.46 for OS (both p=0.01).
CONCLUSIONS: In this study of adult patients with multiple myeloma, the proxy DoD algorithm using month-year of death produced survival results practically identical to those derived from full DoD records, confirming the robustness of mortality-related analyses limited to month-year granularity.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
RWD19
Topic
Clinical Outcomes, Real World Data & Information Systems, Study Approaches
Topic Subcategory
Data Protection, Integrity, & Quality Assurance, Reproducibility & Replicability
Disease
Oncology