OBJECTIVE: To investigate trends in population incidence of HIV-positive hospital admission and risk of in-hospital death among adults living with HIV between 2012 and 2019 in Blantyre, Malawi. DESIGN: Population cohort study using an existing electronic health information system ('SPINE') at Queen Elizabeth Central Hospital and Blantyre census data. METHODS: We used multiple imputation and negative binomial regression to estimate population age-specific and sex-specific admission rates over time. We used a log-binomial model to investigate trends in risk of in-hospital death. RESULTS: Of 32 814 adult medical admissions during Q4 2012--Q3 2019, HIV status was recorded for 75.6%. HIV-positive admissions decreased substantially between 2012 and 2019. After imputation for missing data, HIV-positive admissions were highest in Q3 2013 (173 per 100 000 adult Blantyre residents) and lowest in Q3 2019 (53 per 100 000 residents). An estimated 10 818 fewer than expected people with HIV (PWH) [95% confidence interval (CI) 10 068-11 568] were admitted during 2012-2019 compared with the counterfactual situation where admission rates stayed the same throughout this period. Absolute reductions were greatest for women aged 25-34 years (2264 fewer HIV-positive admissions, 95% CI 2002-2526). In-hospital mortality for PWH was 23.5%, with no significant change over time in any age-sex group, and no association with antiretroviral therapy (ART) use at admission. CONCLUSION: Rates of admission for adult PWH decreased substantially, likely because of large increases in community provision of HIV diagnosis, treatment and care. However, HIV-positive in-hospital deaths remain unacceptably high, despite improvements in ART coverage. A concerted research and implementation agenda is urgently needed to reduce inpatient deaths among PWH.