Most studies involving cure fraction models in survival analysis relate the probability of cure and possible prognostic factors considering a logistic link function. As is well known, the logistic density function is symmetric, preventing the possibility of asymmetry in the relationship between the cure probability and the covariates. The main purpose of this paper is to predict the chance (or probability) of cure for future patients, considering other links between the covariates with the cure fraction. Some links extensively considered in the literature for modeling the cure probabilities are the probit and the complementary loglog links (c-loglog). We evaluate the use of asymmetric links when the main goal is the estimation of survival probabilities. In addition, the patients are considered to be grouped in clinics and, for a given clinic, there may be a non-null association among the survival times, thus affecting the survival probabilities. An EM-type algorithm conjugated with REML estimation procedure in the M-step is considered for parameter estimation. Model comparisons are implemented using the cAIC and cBIC scores. A simulation study suggests that the proposed estimation procedure behaves well for moderate and large sample sizes. A real data illustration reveals also the good performance of the proposed model and estimation approach.