The aim of this meta-analysis study was to evaluate the relative risk of death or AIDS-defining events associated to CD4+ guided treatment interruption in patients with chronic HIV infection. A search was conducted using PubMed and Cochrane Library; key words for PubMed were: "antiretroviral therapy and interrupt*" in the full papers from January 1, 2000 up to and including December 31, 2007. To limit the publication bias, clinical trials performed on the topic of the meta-analysis were searched also on http://www.clinicaltrial.gov. Inclusion criteria of studies were: starting a CD4+ guided interruption of HAART in HIV chronically infected patients with CD4+ cell count > 350 cells/mm3, age > 13 years old, and absence of concomitant use of immunomodulatory drugs. Using a conservative approach, to be included in the meta-analysis, studies had to have a follow up period > 100 person years to minimize the bias of a too short observation time. The studies were classified into two categories: randomized clinical trial (one arm stops therapy and other arms continues HAART) and cohort studies. For each study measures of effect (hazard ratio or incidence rate ratio) were reported, when available, uncorrected and corrected for potential confounders. Publication bias was assessed graphically through funnel plot. Pooled relative risk and pooled risk difference were calculated by use of a random effects model following the DerSimonian-Laird method. Observational studies were considered separately and the incidence of primary endpoint was evaluated in each study and the cumulative incidence was calculated. Of the 555 full papers found, all abstracts were screened and 58 full text articles for potential inclusion were retrieved and 18 were retained (seven randomized clinical trials and 11 observational studies). In randomized clinical trials, the meta-analysis showed that the pooled relative risk of AIDS-defining event or mortality was 2.50 (95% CI: 1.87-3.34; p < 0.001); the pooled risk difference of AIDS-defining event or mortality was 0.02 (95% CI: capital ER, Cyrillic0.01-0.05; p = 0.168). The respective values corrected for latest CD4+ value were 1.77 (95% CI: 1.29-2.42; p < 0.001) and 0.01 (95% CI: capital ER, Cyrillic0.01-0.02; p = 0.37). The pooled relative risk of death was 1.8 (95% CI: 1.18-2.77; p = 0.007), and the corresponding pooled risk difference was 0.01 (95% CI: 0.001-0.012; p = 0.03). The risk of death resulted to have increased in patients that interrupted treatment; the corresponding value of risk difference was significant, although it was low (one extra death per 100 person years). Considering that a separate analysis corrected for the latest CD4+ value was not feasible for this endpoint, and that mortality rates in HIV-infected patients are inversely correlated with the CD4+ count, the value reported is extremely conservative. In cohort studies, the cumulative incidence of deaths or AIDS-defining events in the five studies with follow-up > 100 person years, was 0.77 (95% CI: 0.37-1.42 events per 100 person years), ranging in different studies from 0 to 3.2 events per 100 person years. This meta-analysis suggests that in patients undergoing a treatment interruption, there is an increased risk of developing AIDS or death, and that this risk is decreased if a relatively high CD4+ threshold is chosen to reinitiate the treatment, while the risk difference does not reach statistical significance.