To explore clinical predictive value of short-term dynamic changes in platelet counts (PLT) for prognosis of sepsis patients in intensive care unit (ICU). A retrospective cohort study was conducted. The patients aging 18 to 80 years old who were diagnosed by Sepsis-3 admitted to ICU of South Branch of Shanghai General Hospital from November 2015 to October 2018 were enrolled. According to whether the patients died within 28 days, they were divided into death and survival groups. General information and clinical baseline data [including disease severity score, infection biomarkers, PLT and organ function parameters (cardiac, liver, kidney, coagulation) and inflammatory cytokines] between the two groups were compared. Based on clinical indicators which had statistically significance, receiver operating characteristic (ROC) curve was drawn to predict the prognosis of the patients within 28 days. Then, risk factors of 28-day mortality of sepsis patients in ICU were screened by univariate and multivariate Logistic regression analysis. On the basis of multivariate Logistic regression analysis results, a multiparameter model was built, and the ROC curve was drawn to predict its prognosis within 28 days. A total of 220 sepsis patients were enrolled. Among them, 61 patients died and 159 patients survived within 28 days with a 28-day mortality of 27.7%. Compared with the survival group, the patients in the death group were senior in age, more likely to suffer from chronic cardiovascular, chronic kidney and immune system disease, had higher scores in acute physiologic and chronic health evaluation II (APACHE II) score, sequential organ failure assessment (SOFA) score, disseminated intravascular coagulation (DIC) score and less PLT on the 1st and 7th day, sustained a higher incidence of persistent thrombocytopenia (PLT were all < 100×109/L in the first week after ICU admission) or acquired thrombocytopenia (PLT ≥ 100×109/L on the day of ICU admission, but dropped over 50% during the first week after ICU admission), were subjected to higher procalcitonin (PCT) and interleukin-6 (IL-6) levels and endured worse organ function (cardiac, kidney, coagulation) with statistically significant differences. However, there was no significant difference in gender, disease type, infection sites, pathogens or liver function. The ROC curve analysis for the 28-day prognosis of sepsis illustrated that the three disease severity scores could predict the 28-day prognosis of sepsis in ICU, and the area under ROC curve (AUC) of SOFA score was the highest (AUC = 0.878). The AUC of PLT on the 7th day was higher than that on the 1st day (AUC: 0.862 vs. 0.674), and the AUC of other clinical indicators were all < 0.8. Univariate and multivariate Logistic regression analysis showed that SOFA score [odds ratio (OR) = 1.423, 95% confidence interval (95%CI) was 1.089-1.859, P = 0.010], troponin I (TnI; OR = 2.056, 95%CI was 1.057-3.999, P = 0.034), and persistent or acquired thrombocytopenia (OR = 13.028, 95%CI was 4.033-42.090, P < 0.001) were three independent risk factors for 28-day mortality of the sepsis patients in ICU. Based on the multivariate Logistic regression analysis results, a multiparameter model was built with SOFA score, TnI and persistent or acquired thrombocytopenia, which showed a AUC of 0.926 to predict the 28-day mortality of sepsis patients in ICU. When the optimum cut-off value was 0.398 in the model, the sensitivity was 76.8%, and the specificity was 92.8%. Persistent or acquired thrombocytopenia within the first week of hospitalization proves to have a relatively momentous clinical predictive value for prognosis of sepsis patients in ICU. Clinical intervention focusing on thrombocytopenia may become a new potential therapy for these sepsis patients.