Background: Immediate initiation of antiretroviral therapy (ART) in asymptomatic adults with CD4 counts higher than 500 cells per μL, as recommended, might not always be possible in resource-limited settings. We aimed to identify subgroups of individuals who would benefit most from immediate treatment. Methods: The START trial was a randomised controlled trial in asymptomatic, HIV-positive adults previously untreated with ART. Participants with CD4 counts higher than 500 cells per μL were randomly assigned to receive immediate ART or to defer ART until CD4 counts were lower than 350 cells per μL. The primary endpoint of the study was serious AIDS-defining illnesses or death from AIDS and serious non-AIDS illnesses or non-AIDS-related death. In this post-hoc analysis, we estimated event rates and absolute risk reduction with immediate versus deferred ART, overall and by subgroup. Subgroups were prespecified in the study protocol or formed post hoc on the basis of baseline characteristics associated with morbidity and mortality in people with HIV. For continuous characteristics, approximate terciles were chosen as subgroup cutoff points, unless different cutoffs were clinically meaningful (eg, age ≥50 years). We estimated the number needed to treat immediately with ART for 1 year to prevent one primary event. Heterogeneity in the absolute risk reduction between subgroups was assessed with bootstrap tests. The START trial is registered with ClinicalTrials.gov, number NCT00867048. Findings: Between April 15, 2009, and Dec 23, 2013, we enrolled 4684 participants from 35 countries across five continents, of whom 2325 were assigned to immediate ART and 2359 were assigned to deferred ART. The primary endpoint occurred in 42 participants in the immediate ART group (0·58 events per 100 person-years) and 100 participants in the deferred ART group (1·37 events per 100 person-years). The absolute risk reduction was 0·80 (95% CI 0·48–1·13) per 100 person-years with immediate treatment, and the number needed to treat immediately to prevent one event was 126 (95% CI 89–208). Significant heterogeneity in absolute risk reduction with immediate ART was found across subgroups according to age (p=0·0022), CD4 to CD8 ratio (p=0·0007), and plasma HIV RNA viral load (p=0·033) at baseline. The highest absolute risk reductions and the lowest numbers needed to treat were found in participants aged 50 years or older, those with CD4 to CD8 ratios of less than 0·5, and those with plasma HIV RNA viral loads of 50 000 copies per mL or higher. Interpretation: Asymptomatic, ART-naive adults with CD4 counts higher than 500 cells per μL who are older, have a low CD4 to CD8 ratio, or a high plasma HIV RNA viral load benefit most from immediate initiation of ART and should be prioritised for treatment. Funding: US National Institute of Allergy and Infectious Diseases.