Serum prealbumin is a clinically relevant indicator of nutritional status and inflammation in patients with acute kidney injury (AKI). This study aimed to determine whether serum prealbumin and its longitudinal changes over a week could improve the prediction of 90-day mortality in AKI patients. This prospective cohort study included 340 adults with AKI between 2014 and 2015. There were 94 (27.6%) patient deaths within 90 days. Serum prealbumin level <10 mg/dL at the time of AKI diagnosis was associated with a 155% increased death risk ratio (adjusted hazard ratio [HR], 2.55; 95% confidence interval [CI], 1.18 to 5.49; P = 0.02). Serum prealbumin fall >4 mg/dL was also associated with 90-day mortality in adjusted Cox regression models (HR, 1.79; 95% CI, 1.06 to 3.03; P = 0.03). Compared to serum albumin, mortality-predictability of serum prealbumin (P = 0.01) and its changes (P = 0.01) were both increased. Adding prealbumin and its changes on the conventional covariates improved the prediction of progression to 90-day mortality (NRI 0.29, P = 0.04; aIDI 0.08; P = 0.03). In conclusion, serum prealbumin, and its changes were independent predictors of worse prognosis in AKI, and could be potential surrogates to better predict 90-day mortality.