The incidence of localized kidney cancer has been increasing for the past several decades. Much of this increase is likely due to the detection of small tumors found incidentally after ultrasound, CT scans, or MRI scans ordered for non-kidney cancer related reasons. This suggests two possibilities:1) many tumors that would previously have been found at a later stage are simply being found at earlier stages, and 2) tumors that would never have progressed to a symptomatic or lethal stage are now being found. The possibility that biologically inconsequential small renal masses are now being detected has opened a debate concerning the benefits of active surveillance prior to or in lieu of surgical or other interventions for the treatment of small localized renal masses. Many patients who are treated with surgery will die from other diseases within five years of treatment. The survival benefit of therapy for kidney cancer is hence modest for patients who would die within five years regardless of therapy. Further, there is some evidence that certain treatments might actually worsen survival outcomes in some patients with localized kidney cancer. Radical nephrectomy, for example, has been associated with an increased risk of chronic kidney disease (CKD) compared to partial nephrectomy. Given the debate surrounding the appropriateness of treatment for localized kidney cancer, particularly among older patients and those with comorbidities, better prognostic models are needed to identify who might benefit from active surveillance (also called observation). The goal of this work is to improve prognostic modeling by developing models that can classify individuals according to their underlying hazard of death either with or without treatment. We will apply the models using linked SEER-Medicare data. The importance of accounting for heterogeneity of progression or mortality rates has already been noted in the medical decision making literature. However, such methods often assume that individuals are either rapid or slow disease progressors. This research, in contrast, proposes the development of models that can identify four potential survival rate groups in the investigation of clinical effectiveness: 1) those that have long survival with or without treatment who can therefore be observed, 2) those that have short survival without treatment but long survival with treatment who should hence undergo immediate intervention, 3) those that have long survival without treatment but short survival with treatment who should be observed, and 4) those that have short survival with or without treatment who can avoid unnecessary surgery. Further, we assume that relatively long and short survival hazards can vary between treatment arms. We propose using principal stratification and Rubin's causal model as conceptual tools for this investigation. This project will further knowledge concerning heterogeneity in survival rates among those with localized kidney cancer.
This work will improve prognostic modeling of localized kidney cancer outcomes by developing models that can classify individuals according to their underlying hazard of death either with or without treatment. Those whose life expectancy would not change or would worsen with treatment could be spared surgery.
|Austin, Steven R; Wong, Yu-Ning; Uzzo, Robert G et al. (2015) Why Summary Comorbidity Measures Such As the Charlson Comorbidity Index and Elixhauser Score Work. Med Care 53:e65-72|
|Simhan, Jay; Smaldone, Marc C; Egleston, Brian L et al. (2014) Nephron-sparing management vs radical nephroureterectomy for low- or moderate-grade, low-stage upper tract urothelial carcinoma. BJU Int 114:216-20|
|Corcoran, Anthony T; Smaldone, Marc C; Egleston, Brian L et al. (2013) Comparison of prostate cancer diagnosis in patients receiving unrelated urological and non-urological cancer care. BJU Int 112:161-8|
|Egleston, Brian L (2011) Response to Pearl's comments on principal stratification. Int J Biostat 7:24|