Online Program

Overview of Estimators in Survival Analysis for Recurrent Event Data: An Application to Morbidity Outcomes in Veterans with Spinal Cord Injury

*Lauren Bailey, Edward J. Hines, Jr. VA Hospital, Hines, IL; University of Illinois at Chicago 
Stephen Burns, VA Puget Sound Health Care System, Seattle, WA; University of Washington 
Laura Carbone, Veterans Affairs Medical Center, Memphis, TN; University of Tennessee, Memphis, TN  
Amy Chin, Edward J. Hines, Jr. VA Hospital, Hines, IL 
Helen Hoenig, Durham VAMC, Durham, NC 
Jelena Svircev, VA Puget Sound Health Care System, Seattle, WA; University of Washington  
Frances Weaver, Edward J. Hines, Jr. VA Hospital, Hines, IL; Loyola University 

Keywords: survival analysis, recurrent event, spinal cord injury, morbidity

AIMS The impact of lower extremity fracture is assessed in terms of morbidity outcomes following fracture using three different estimators for recurrent events in survival analysis. METHODS In a propensity-matched nested case-control study among Veterans with Spinal Cord Injury (SCI), selected morbid events are compared among Veterans experiencing lower extremity fracture and those not experiencing fracture. Cases (Veterans with fracture) were matched to controls (1,027 cases and 1,027 propensity-matched controls) on demographic (age, race), VA connected service status, comorbidities, severity and scope of SCI using one-to-one Mahalanobis metric matching. Recurrent morbidity outcomes under study include urinary tract infections, pressure ulcer utilization, venous thrombosis events, and respiratory illnesses. Distribution of recurrent outcomes are characterized by event frequency and intensity, then compared between the case and control groups using three different estimators: an IID product limit estimator using a stochastic counting process (IID PLE), Wang-Chang product limit estimator (WC PLE), and frailty model maximum likelihood estimator (GF MLE).

RESULTS Lower extremity fracture is associated with increased pressure ulcer utilization post-fracture (p<0.05). Depending on the estimator, there are either significantly more (IID PLE, p<0.05) or no significant difference (WC PLE, GF MLE) in recurrent urinary tract infections and venous thrombosis events among fracture cases compared to non-fracture controls, highlighting the importance of estimator selection in recurrent event analysis. Lower extremity fracture might provide a small, albeit non-significant, protective effect for respiratory illness. CONCLUSION Lower extremity fracture is associated with differentiated risk levels of adverse health complications post-fracture in SCI patients. Careful attention and targeted care is required for SCI patients following fracture. The choice of estimator in recurrent event survival analysis is extremely important; different estimator choices might provide dissimilar conclusions. Techniques in estimator selection are explored.