Patterns and Prediction of Competing Causes of Mortality in Older Adults Diagnosed with Indolent Non-Hodgkin Lymphoma Public Deposited

Downloadable Content

Download PDF
Last Modified
  • March 20, 2019
  • Hester, Laura
    • Affiliation: Gillings School of Global Public Health, Department of Epidemiology
  • Non-Hodgkin lymphoma (NHL) consists of heterogeneous hematological malignancies that are broadly categorized into aggressive or indolent tumor growth groups. In the past two decades, there have been notable increases in the proportion of NHL diagnoses aged >65 and cancer-specific survival with the aging US population and improvements in NHL treatments. These population changes have important implications for non-cancer mortality, particularly for indolent NHL subtypes, which display remitting-relapsing patterns and a slower progression. This dissertation sought to address gaps in knowledge about non-cancer mortality in NHL by providing foundational evidence on: 1) cancer-specific and non-cancer mortality patterns in NHL subtypes and 2) characteristics of indolent NHL patients at greatest risk of non-cancer mortality. We identified adults aged >66 at diagnosis with a first, primary NHL diagnosis from 2004-2011 using a database linking the US Surveillance, Epidemiology, and End Results (SEER) cancer registry with Medicare health insurance claims. Using death certificate data and Fine-Gray competing risks methods, Aim 1 estimated the 5-year cumulative incidence of NHL-specific and non-cancer mortality by prognostic factors (subtype, age, comorbidity level) in 26,809 NHL patients. Among aggressive subtypes, NHL-specific mortality exceeded non-cancer mortality across all ages and comorbidity levels. In indolent subtypes, non-cancer mortality was similar to or exceeded NHL-specific mortality for patients with older ages, higher comorbidity burdens, or specific subtypes. The results support development of tools predicting non-cancer mortality in older indolent NHL patients. In Aim 2, we developed and internally validated risk prediction models for short- and long-term mortality outcomes in 9789 indolent NHL patients. We created 16 elastic net penalized regression models predicting 1- and 5-year all-cause and non-cancer mortality (four models per outcome) in 100 randomly resampled training sets. In 100 validation sets, we compared average performance statistics of the elastic net to those from comorbidity score models. For all outcomes, the elastic net models had a higher discrimination and lower false-positive rate than comorbidity score models. However, differences were not statistically significantly. This project supports development of personalized prediction models integrated into electronic medical records that can inform physicians and patients on non-cancer mortality risk in indolent NHL treatment decision-making.
Date of publication
Resource type
Rights statement
  • In Copyright
  • Lund, Jennifer
  • Brookhart, M. Alan
  • Stürmer, Til
  • Wood, William
  • Park, Steven
  • Doctor of Philosophy
Degree granting institution
  • University of North Carolina at Chapel Hill Graduate School
Graduation year
  • 2017

This work has no parents.