The rebound of platelet count could be a predictor of good prognosis of sepsis in the intensive care unit: a retrospective analysis of the large clinical database MIMIC-III

Fa Huang Sun Yat-sen University First A liated Hospital Jinghong Xu Sun Yat-sen University First A liated Hospital https://orcid.org/0000-0003-0541-3368 Li Tong Sun Yat-sen University First A liated Hospital Xiaoguang Hu Sun Yat-sen University First A liated Hospital Ka Yin Lui Sun Yat-sen University First A liated Hospital Yanping Zhu Sun Yat-sen University First A liated Hospital Shuhe Li Sun Yat-sen University First A liated Hospital Changjie Cai (  caichjie@mail.sysu.edu.cn ) Sun Yat-sen University First A liated Hospital https://orcid.org/0000-0002-0096-1982

zero were categorized as the increased platelet count group (ΔPC > 0 group), while the remaining patients made up the decreased platelet count group (ΔPC = 0 group).
The primary endpoint of the study was 28-day mortality from the date of ICU admission. Secondary outcomes included hospital mortality, ICU and hospital length of stay, the percentage of dialysis and mechanical ventilation, total hours of mechanical ventilation, the total amount of dopamine, and norepinephrine.
We used PostgreSQL (version 9.6; PostgreSQL Global Development Group) for data extraction. We extracted data from the following tables: ADMISSION, ICUSTAYS, PATIENTS, CHARTEVNETS, DIAGNOSIS_ICD and LABITEMS.
Statistical analysis SPSS (version 22.0; IBM, Armonk, NY) and R (version 3.5.2, www.r-project.org) were used for statistical analyses. The normally distributed continuous variables were described as the mean ± standard deviation (SD), and skewed distributed continuous variables were presented as the median and interquartile range (IQR; 25th and 75th percentiles). Student's t test or ANOVA was conducted to compare differences in the normally distributed continuous variables between groups. A Mann-Whitney U test or a Kruskal-Wallis test was used to compare differences in the skewed distributed continuous variables between groups. Categorical variables were described using percentages and frequencies, and they were analyzed by chi-square tests or Fisher's exact test.
Survival analysis was performed using the Kaplan-Meier test and the log-rank test to determine whether ΔPC affected 28-day mortality. Multivariate analysis was performed using the Cox hazard model to determine which variables were important for predicting the prognosis of the patients. Variables with P < 0.1 in the survival analysis were selected for the Cox proportional hazards regression model. A P value of < 0.05 was considered statistically signi cant.

Patient characteristics
In total, 3457 patients were included in the current study (Table 1). A total of 510 patients (14.8%) had platelet counts of < 50 ⋅ 10 9 /L, 498 (14.4%) had counts of 50 ⋅ 10 9 to 99 ⋅ 10 9 /L, 731 (21.1%) had counts of 100 ⋅ 10 9 to 149 ⋅ 10 9 /L, 1548 (44.8%) had counts 150 ⋅ 10 9 to 399 ⋅ 10 9 /L, and 170 (4.9%) had counts above the normal range. There were 1476 female and 1981 male patients, accounting for 42.7% and 57.3% of the total population, respectively. The median age of all patients was 68 years, while the patients with lower platelet counts had a lower age. Most of the patients were admitted with emergency status (95.3%), and most of them came from referrals (45.8%) from their admission location. The most common ICU type was a medical ICU (67.4%). The heart rate of patients with very low platelet counts and thrombocytosis was signi cantly higher (P < 0.001). Patients with very low, intermediate-low, or low platelet counts had higher modi ed SOFA scores and SASP (P < 0.001, respectively). Patients with very low or intermediate-low platelet counts received more transfused blood products, including platelets, plasma and red blood cells. (P < 0.001) As shown by the mortality rate s and the length of hospital and ICU stay in Table 2, the 28-day mortality in patients in the very low (43.1%) and intermediatelow (36.9%) platelet count groups were higher than those in the low (26.8%) and normal (23.2%) platelet count groups and thrombocytosis (18.2%) group (P < 0.001), while there was no signi cant difference among the low platelet count, normal platelet count and thrombocytosis groups. Furthermore, the hospital mortality of patients in the very low platelet count group was 56.7%, which was signi cantly higher than in the other groups (P < 0.001). Platelet transfusion did not improve 28-day mortality (P > 0.05) (Fig. 1a). However, the 28-day mortality of patients who received transfusion of plasma was higher (49.0% vs. 39.0%, P = 0.024, 51.4% vs. 31.4%, P < 0.001, 38.1% vs. 25.1% P = 0.007, 32.8% vs. 22.3%, P = 0.06, respectively) except in the thrombocytosis group (Fig. 1b). Red blood cell transfusion did not improve 28-day mortality (P > 0.05) (Fig. 1c). Meanwhile, in the subgroup (          In the comparison 28-day outcomes by demographic and hospital characteristics (supplementary), we chose to input age, gender, admission location, modi ed SOFA, SOFA, SASP II, platelet transfusion, plasma transfusion, red blood cell transfusion, ΔPC and nadir platelet count into the Cox proportional hazards regression model (Fig. 2a). In the regression model, the risk of death within 28 days in patients in the ΔPC > 0 group was 0.570 times that in patients in the ΔPC = 0 group (HR 0.570, 95% CI 0.498-0.651, P < 0.001). The survival curve showed (see Fig. 2b) that the survival probability in the ΔPC > 0 group was higher than that in the ΔPC = 0 group.

Discussion
Of the 3457 patients who were enrolled in the study, 1739 had a platelet nadir of less than 150,000/mm 3 . The occurrence of thrombocytopenia was 50.3%, which is consistent with previous reports [14,15]. Sepsis-related thrombocytopenia may be affected by a variety of factors, including infection, disseminated intravascular coagulation (DIC), myelosuppression, drug factors, uid replacement, and surgery [16]. Our study shows that patient mortality rate is correlated with the degree of thrombocytopenia. With a reduction in the nadir platelet count, the 28-day mortality drastically increases, peaking at 43.1%. It can be easily seen that the prognosis of septic patients with thrombocytopenia is poor. Platelets, which are nonnuclear cells differentiated from megakaryocytes, play a vital role in the coagulation process and immune response and serve as a bridge between endogenous and acquired immune responses [17]. In sepsis, many in ammatory mediators, cytokines, endotoxins and metabolites can be released to promote platelet activation by direct or indirect action on platelet surface receptors and cause damage to the vascular endothelium system. The activated platelets gather around in ammatory cells and induce endothelial cells, smooth muscle cells and macrophages to secrete a large number of cytokines, regulate neutrophil aggregation and exudation to in ammatory sites, and promote the activation of coagulation factor by means of signal transduction and other mechanisms, resulting in the coagulation system being overactivated and the body exhibiting a high coagulation state. Activated platelets also release many in ammatory chemokines (such as IL-6, IL-8, etc.) and procoagulant substances (ADP, P-selectin, thrombin sensitive protein, brinogen, etc.), further aggravating the in ammatory response. This release leads to an uncontrolled in ammatory response and excessive activation of the coagulation system, eventually leading to multiple organ dysfunctions syndrome (MODS) and directly affecting the patient's prognosis [18].
Furthermore, we noted that a relative improvement in platelet count would lead to a decline in hospital mortality, even for those patients whose platelets are within the normal range. One platelet transfusion generally increases platelet count by 23×10 9 /L, although not for all patients. No matter what treatment is used, such as blood transfusion, anti-infection or others, platelet elevation cannot be guaranteed in every patient with thrombocytopenia. This may be related to the aggravation of infection, bone marrow suppression and other factors [6]. If there is no rise in a patient's platelet count, careful attention should be paid, the patient should be treated carefully, and family members should be informed of the possibility of poor prognosis. In our study, platelet transfusion had no effect on 28-day mortality regardless of the platelet count of the patient. Platelet transfusion does not necessarily improve clinical outcomes, but platelet elevation does. This nding reminds us that there is no obvious direct relationship between the amount of platelet infusion and the prognosis of sepsis patients, but the recovery of platelet count indicates the improvement of the condition. Even after platelet transfusion in many patients, platelet still cannot recover is a thorny problem faced by many clinicians in the course of sepsis treatment. The rebound of platelet count can be considered a predictor of good prognosis. Platelets have been shown to limit the growth and spread of bacteria in experimental sepsis [19]. Platelets affect leukocyte recruitment and function [20], cytokine response [21], the activation of vascular endothelial cells [22] and the coagulation system [22]. In addition, platelets maintain vascular integrity, especially in a strong in ammatory environment [23]. More critically, in septic patients, the number of platelets declines while their function is also damaged. The mean platelet volume (MPV) and the platelet distribution width (PDW) are important indicators re ecting platelet function. In severe in ammatory reactions, bone marrow is stimulated by bone marrow compensatory response or pathogenic factors, causing megakaryocytes to proliferate and resulting in elevated MPV, which produces immature, disparate, and disparate platelets. At this time, the platelet volume dispersion increases, but the newly produced larger platelets contain more active dense particles; the adhesion, aggregation, and release functions of platelets become stronger, and the in ammatory reaction of the body is ampli ed, directly affecting the prognosis of patients [24]. Therefore, it is easy to explain why patients who are not diagnosed with thrombocytopenia still bene t from the increase in platelets.
Many scholars believe that platelets should be regarded as a new therapeutic target [25]. The SCCM guideline in 2016 suggested "prophylactic platelet transfusion when counts are < 10,000/mm 3 (10 × 10 9 /L) in the absence of apparent bleeding and when counts are < 20,000/mm 3 (20 × 10 9 /L) if the patient has a signi cant risk of bleeding." However, this guideline is a weak recommendation, and the quality of evidence supporting this guideline is extremely low because it is based on clinical trials of thrombocytopenia, which is usually caused by leukemia and stem cell transplantation. Our study proves that an increase in platelets in septic patients, regardless of the approaches that have been applied, will improve the prognosis.
We must admit that there are some limitations of our study. Although the study population was large, this study was unfortunately based on a single center, which may limit the extension of these results to other cohorts of patients. In addition, we did not analyze MPV, PDW or other indicators to study changes in platelet function during sepsis. Like many other studies [26][27][28], this study does not include a repeat platelet count to con rm TP, thereby limiting the accuracy of TP diagnosis.

Conclusion
In conclusion, our study shows that there is no obvious direct relationship between the amount of platelet infusion and the prognosis of sepsis patients, but the recovery of platelet count indicates the improvement of the condition.

Abbreviations
Thrombocytopenia; ICU-intensive care unit; MICU-medical intensive care unit; CCU-coronary care unit; SICU-surgical intensive care unit; CSRU-cardiac surgery recovery unit; TSICU-trauma surgical intensive care unit; LOS-length of stay; HR-hazard ratio; CI-con dence interval; SOFA-Sequential Organ Failure Score; SAPS-Simpli ed Acute Physiology Score Declarations