Time-dependent mortality effects of dynamic antiretroviral therapy regimens

INTRODUCTION: To compare the mortality effects of dynamic ART regimens among people living with HIV (PLWH), while accounting for time-varying CD4+ T cell counts and HIV viral loads. METHODS: Leveraging data from 32,713 ART-naïve individuals (2015-2024) in Henan, China, we compared three approaches: (1) baseline, time-fixed Cox model that classified participants by ART class at initiation, (2) time-varying Cox model that updated ART class over follow-up; and (3) marginal structural model (MSM) to address time-varying confounding. ART was classified as non-nucleoside reverse transcriptase inhibitor (NNRTI)-, integrase strand transfer inhibitor (INSTI)-, or protease inhibitor (PI)-based regimens (reference: NNRTI). Outcomes included all-cause mortality and AIDS-related mortality. RESULTS: Effect estimates shifted from the baseline time-fixed model toward stronger protective associations after incorporating time-updated ART exposure and further adjusting for time-varying confounding. In the MSM, compared with NNRTI-based regimens, INSTI- and PI-based regimens were associated with lower all-cause mortality (INSTI: HR=0.38, 95% CI: 0.29-0.49; PI: HR=0.69, 95% CI: 0.52-0.91) and lower AIDS-related mortality (INSTI: HR=0.31, 95% CI: 0.17-0.57; PI: HR=0.47, 95% CI: 0.25-0.89). Secondary exploratory cumulative exposure analyses yielded patterns broadly consistent with lower mortality for INSTI exposure, whereas estimates for PI exposure were closer to the null and should be interpreted cautiously. CONCLUSION: Accounting for time-updated ART exposure and time-varying confounding, INSTI-based regimens were associated with lower all-cause and AIDS-related mortality than NNRTI-based regimens. These findings underscore the need to account for dynamic ART exposure and time-varying confounding in observational comparative effectiveness studies of ART regimens.