OBJECTIVES: Our aim was to investigate the durability of different initial regimens in patients starting ART with CD4+ counts 5 log10 copies/mL. METHODS: This was a retrospective study of HIV-infected patients prospectively followed in the ICONA cohort. Those who started ART with boosted protease inhibitors (bPIs), NNRTIs or integrase strand transfer inhibitors (InSTIs), with CD4+ 5 log10 copies/mL, were included. The primary endpoint was treatment failure (TF), a composite endpoint defined as virological failure (VF, first of two consecutive HIV-RNA >50 copies/mL after 6 months of treatment), discontinuation of class of the anchor drug or death. Independent associations were investigated by Poisson regression analysis in a model including age, gender, mode of HIV transmission, CDC stage, HCV and HBV co-infection, pre-treatment HIV-RNA, CD4+ count and CD4+/CD8+ ratio, ongoing opportunistic disease, fibrosis FIB-4 index, estimated glomerular filtration rate, haemoglobin, platelets, neutrophils, calendar year of ART initiation, anchor drug class (treatment group) and nucleos(t)ide backbone. RESULTS: A total of 1195 patients fulfilled the inclusion criteria: 696 started ART with a bPI, 315 with an InSTI and 184 with an NNRTI. During 2759 person-years of follow up, 642 patients experienced TF. Starting ART with bPIs [adjusted incidence rate ratio (aIRR) (95% CI) 1.62 (1.29-2.03) versus starting with NNRTIs; P 5 log10 HIV-RNA copies/mL, the durability of regimens based on InSTIs was longer than that of NNRTI- and bPI-based regimens. © The Author(s) 2019. Published by Oxford University Press on behalf of the British Society for Antimicrobial Chemotherapy. All rights reserved. For permissions, please email: email@example.com.