We investigated features of major adherence lapses in antiretroviral therapy (ART) at public Emusanda Health Centre in rural Kakamega County, Kenya using medical records from 2008 to 2015 for all 306 eligible patients receiving ART. Data were modelled using survival analysis. Patients were more likely to lapse if they received stavudine (hazard ratio (HR) 2.54, 95% confidence interval (95%CI):1.44-4.47) or zidovudine (HR 1.64, 95%CI:1.02-2.63) relative to tenofovir. Each day a patient slept hungry per month increased risk of major adherence lapse by 3% (95%CI:0-7%). Isolated home visits by community health workers (CHWs) were more effective to assist patients to return to the health centre than isolated phone calls (HR 2.52, 95%CI:1.02-6.20).