To explain differences in survival in the first three years of combination anti-retroviral therapy (cART) between HIV treatment centres in The Netherlands.We developed a mathematical simulation model, parameterised using data from the ATHENA cohort that describes patients entering care, being monitored and starting cART. Three scenarios were used to represent three treatment centres with widely varying mortality rates on cART that were differentiated by: (i) the distribution of CD4 counts of patients entering care; (ii) the age distribution of patients entering care; (iii) the average rate of monitoring the patients not on cART. At the level of the treatment centre, the fraction of Dutch MSM dying in the first three years of treatment ranged from 0% to 8%. The mathematical model captured the large variation in observed mortality between the three treatment centres. Manipulating the age-distribution of patients or the frequency of monitoring did not affect the model predictions. In contrast, when the same national average distribution of CD4 count at entry was used in all the scenarios, the variation in predicted mortality between all centres was diminished.Patients entering care with low CD4 counts appears to be the main source of variation in the mortality rates between Dutch treatment centres. Recruiting HIV-infected individuals to care earlier could lead to substantial improvements in cART outcomes. For example, if patients were to present with at least 400 CD4 cells/mm(3), as they do already in some centres, then our model predicts that the mortality in the first three years of cART could be reduced by approximately 20%.