Aims. To investigate the association between mean plasma glucose and inhospital death proportion. Methods. We retrospectively collected 162,169 inpatient data in Huashan Hospital from January 2012 to December 2015. Mean plasma glucose was calculated and considered as the average glycemia control during hospitalization. Patients were stratified into six groups according to mean plasma glucose. Nonlinear regression was performed to determine the associations between mean plasma glucose and inhospital death proportion, medical cost, and length of stay. Multivariate logistic regressions were performed to evaluate the relationship of mean plasma glucose and outcomes controlling for confounders including age, gender, and others. Subgroup analyses were performed on basis of whether they were surgical patients, ICU patients, patients with diabetes, or others. Results. Of the 162,169 hospitalized participants, 53.32% were male and 989 died during hospitalization. Nonlinear regression showed there were positive and significant associations between mean plasma glucose and death proportion, medical cost, and length of stay (
P
<
0.001
for all). Multivariate logistic regressions showed that, compared with group B, a statistically significant association between mean plasma glucose and predicted outcome was apparent, with the odds ratios (95% confidence interval) of 5.79 (3.51–9.55), 2.85 (2.40–3.38), 6.29 (5.24–7.54), 9.34 (7.51–11.62), and 23.52 (16.64–33.26), for group A, group C, group D, group E, and group F, respectively. There was a U-shaped association between mean plasma glucose and death proportion. Subgroup analyses showed similar associations between mean plasma glucose and death proportion, medical cost, and length of stay as in the whole sample. Conclusions. There was a U-curve association between mean plasma glucose with inhospital death proportion. Mean plasma glucose was associated positively with medical cost and length of stay.