Examining the causal pathways through which the Daraja intervention led to sustained improvement in HIV outcomes, including retention in HIV care, ART adherence, and viral load suppression, could highlight critical mechanisms for improving post-hospital outcomes in people with HIV. This pre-specified analysis of the Daraja trial assessed five mediators identified from the Gelberg-Andersen model (self-efficacy, social support, perceived need for HIV services, perceived stigma, and traditional HIV beliefs) and two additional potential mediators, depression and alcohol use. Mediators were quantified pre- and post-intervention. A causal mediation analysis framework was used to construct independent single mediator models for each hypothesized mediator to understand their effect on HIV-related outcomes. Our intervention-mediator analysis demonstrated that the Daraja intervention improved self-efficacy (1.34 (95% CI (0.2, 2.48)) but no effect on other factors in the Gelberg-Andersen model. In our mediator-outcome analysis, improvements in self-efficacy, social support, and traditional HIV beliefs were associated with higher retention in care at 12 months. Results were similar for other HIV outcomes. Self-efficacy was associated with all three HIV outcomes, according to the intervention-mediator effect, mediator-outcome effect, average causal mediation effect, and average total effect. For these outcomes, 37-42% of the intervention effect was mediated by self-efficacy. Self-efficacy was an important mediator of the effect of the Daraja intervention in promoting retention in HIV care, improving ART adherence and reducing HIV viral load. Targeting self-efficacy appears to be critical for enabling hospitalized people with HIV to make the difficult transition from hospital to primary HIV care and live a healthy life with HIV. Clinical Trial Number ClinicalTrials.gov Identifier NCT03858998.