Cancer Survival Analysis Software (CanSurv)
CanSurv is a statistical software to analyze population-based survival data. For grouped survival data, it can fit
both the standard survival models and the mixture cure survival models and provides various
graphs for model diagnosis. It can also fit parametric (cure) survival models to individually-listed data. Methods
& Software for Population-based Cancer Statistics shows the relationship of the CanSurv program to SEER*Stat
and other statistical methods and tools.
Currently, CanSurv uses population-based survival data extracted from SEER*Stat
survival session. Other survival data file format (e.g., ASCII data) will be allowed in the future version. The
measure for excess mortality (net survival) can be either relative
survival or cause-specific survival and the survival data can be either stratified with grouped survival times
or individually listed with continuous survival times. The software can fit different survival models to the data,
e.g., standard parametric models and Cox proportional hazards model, parametric mixture cure models and mixture cure
models with power functions. The prognostic factors and demographic variables that are related to survival of cancer
patients can be used as covariates in the survival models in CanSurv. The parameters and associated standard errors
are estimated by Newton-Raphson method. For grouped survival data, the plots of actuarial and estimated survival functions,
k-year survival probabilities and deviance residuals can be generated for model diagnosis. The example graph shows
the plot of actuarial and estimated survival curves for colorectal cancer patients from SEER 9 registries.
Caution
Although the mixture cure models provide simultaneous estimates of the cure fraction, i.e., the proportion of the
patients cured from disease and the survival function for uncured patients (latency), a crucial issue with the mixture
cure models is the identifiability of the cure fraction and parameters of latency distribution. Cure fraction estimates
can be quite sensitive to the choice of latency distribution and the length of follow-up time.
A key factor for the instability of the cure fraction estimates from different latency distributions is the median
survival time for the uncured patients in relation to the length of follow-up. A longer median survival time than
the follow-up time suggests that the survivors include a substantial fraction of uncured patients and a significant
number of deaths will occur after the end of follow-up. To improve the accuracy of cure fraction estimates, we should
have either a larger population size which would increase the power of detecting smaller difference in the survival
functions or have a longer follow-up time which would allow the survival function to level off. See Yu et al. (2004)
for an example.
Getting Help
If you are a new CanSurv user, see Sample CanSurv Analysis.
Other technical support sources:
|