Adapting models with single time-to-event outcomes to include a competing outcome: an exemplar adjusting risk of recurrence after nephrectomy for clear cell renal cell carcinoma for death from other causes
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
Risk prediction models, in particular prognostic models, are used by clinicians to inform care and communicate risks to patients. However, many time-to-event models typically consider only one disease-specific outcome, which leads to overestimation of risk in populations where other-cause mortality is high. An example of this is the widely used Leibovich model, which models distant metastatic recurrence risk in patients with clear cell renal cell carcinoma (ccRCC, the most common form of kidney cancer) who have been treated surgically with radical nephrectomy.
Methods
In this study, we describe a novel approach for adapting existing risk prediction models retrospectively to include adjustment for a competing outcome, using population level data. We apply this approach to the Leibovich model, using life tables from the OZice of National Statistics, to generate the Leibovich Plus model and then illustrate the impact of increasing age on estimated risk of recurrence using both models.
Results
Comparing the predicted risk from the Leibovich model with the predicted risk of distant metastatic recurrence using the Leibovich Plus model, we show how distant-metastatic recurrence risk is overestimated when competing risks are not considered, particularly in older patients with high-risk tumours when using only a disease-specific outcome. For example, the risk of distant metastatic recurrence in individuals with a high-risk tumour pathology is 84.6% in a 55 year old individual after 10 years, but drops to 52.1% in an 85 year old individual with the same tumour pathology after 10 years.
Conclusions
This work describes an approach for adapting existing time-to-event models with disease-specific outcomes to include a competing outcome without the need for new data and illustrates the impact incorporation of competing risks has on estimated risk, particularly in older populations with high overall mortality risk. Such models, for example, the Leibovich Plus model for RCC, can be used in clinical consultations to provide a risk of recurrence adjusted for the risk of death from other causes.
