Most methods of sample size estimation for survival analysis address the problem of detecting difference in survival between two of more groups using (e.g.) the log-rank test. I think that you have a different problem. You want to estimate survival in a single group. The sample size required will depend on what you want to estimate. You may (e.g.) want to estimate (with a specified precision) the survival probability at 6 months or you may want to estimate median survival time. For the problem of estimating a survival probability you could work out expected standard errors using Greenwood's formula for variance of a survival probability and find a suitable sample size by a gradient descent method (i.e. start with a test sample size and calculate expected precision and increase (or reduce) this and recalculate until you find a sample size that gives you the desired precision).You can take a similar approach for median survival time (in this case you would use the method of Brookmeyer and Crowley). This sort of approach (like all sample size calculations) requires you to have some idea of what the survival curve will look like. A simple approach is to use a rule-of-thumb based on experience. For a similar problem (i.e. breastfeeding survival) using a cluster sample, I have found that n = 192 works quite well. I hope this is of some use. Please keep us informed of how this works out.