Literature on earthquake impact on hospital admissions is lacking, particularly in low-resource settings. Our aim was to study the pattern of admissions before and after the 2015 earthquake in a tertiary hospital in Nepal. We used routine hospital data from 9,596 admissions, and defined four periods: pre-earthquake (pre-EQ), acute (EQ1), post-acute (EQ2), and post-earthquake (post-EQ). We compared length of hospital stay (LOS) across the study periods using negative binomial regressions. We used logistic regressions to study changes in probability of admission for diagnostic categories, and Generalized Additive Models to model the difference in number of admissions compared to pre-EQ baseline. LOS was longer in EQ1 than during pre-EQ, in particular for injury-related admissions. In EQ1, the odds of injury admissions increased, while they decreased for the majority of other diagnoses, with the odds of pregnancy-related admissions remaining low until post-EQ. The number of admissions dropped in EQ1 and EQ2, and returned to pre-EQ trends in post-EQ, accumulating 381 admissions lost (CI: 206-556). Our findings suggest that hospital disaster plans must not only foresee injury management after earthquakes, but also ensure accessibility, in particular for pregnant women, and promote a quick return to normality to prevent additional negative health outcomes.