A Network Meta-Analysis of Two Doses of Recombinant Human Thrombopoietin for Treating Sepsis-Related Thrombocytopenia

Previous studies suggest that sepsis remains a common critical illness with a global incidence of 31.5 million. The aim of this study was to evaluate the comparative therapeutic value of recombinant human thrombopoietin (rhTPO) in treating sepsis patients with thrombocytopenia. We conducted a comprehensive electronic search of PubMed, EMBASE, the Cochrane Library, and CNKI from its inception through December 31, 2021. Thirteen randomized controlled trials (RCTs) involving 963 patients were included. Network meta-analyses showed that rhTPO 300 U/kg/day and rhTPO 15000 U/day significantly increased the platelet (PLT) levels on the 7th day and decreased the requirement of transfusion of red blood cells (RBCs), plasma, and PLT compared with IVIG and NAT. SUCRA showed that rhTPO 300 U/kg/day ranked first in terms of 28-day mortality (85.5%) and transfusion, including RBC (88.7%), plasma (89.6%), and PLT (95.2%), while rhTPO 15000 U/day ranked first for the length of the intensive care unit (ICU) stay (95.9%) and PLT level at day 7 (91.6%). rhTPO 300 U/kg/day may be the optimal dose to reduce 28-day mortality and transfusion requirements. However, rhTPO 15000 U/day may be the optimal dose for shortening the ICU stay and increasing the PLT level on the 7th day. However, additional studies to further validate our findings are needed.


Introduction
Sepsis remains a common critical illness with a global incidence of 31.5 million [1] and the leading cause of death in the intensive care unit (ICU) [2], with an annual mortality rate of 16.8% [3]. Trombocytopenia [4], defned as sepsisassociated thrombocytopenia, is frequently seen in patients with sepsis, and has been reported to occur in 35%-59% of the patients [5].
Patients with sepsis diagnosed with thrombocytopenia may develop multiple organ dysfunction and have a higher mortality rate [6][7][8]. Specifcally, thrombocytopenia accounts for 13%-83% of the mortality rate in patients with sepsis [9,10]. Sepsis-associated thrombocytopenia was also found to be strongly associated with longer ICU stays, and the length of hospital stay was a prognostic indicator in patients with sepsis [4]. A previous meta-analysis showed that sepsis-associated thrombocytopenia signifcantly increases the risk of complications such as shock and acute kidney injury [11]. Terefore, there is an urgent need to develop safe and efective treatment strategies to restore platelet (PLT) levels in septic patients with thrombocytopenia [12].
A series of treatments, such as anti-infective therapy, transfusion of PLT, intravenous injection of recombinant human interleukin (rhIL) including rhIL-6 and rhIL-11, intravenous immunoglobulin (IVIG), and administration of platelet-elevating drugs, are currently available for sepsisassociated thrombocytopenia [13][14][15]. Because of the scarcity of resources, transfusion-related complications, and PLT antibody production, the clinical application of PLT transfusion is strictly limited to patients with sepsis [16,17]. Te clinical use of intravenous rhIL is associated with mild thrombopoietic activity and unacceptable adverse efects; therefore, the procedure is limited and needs more caution in clinical application [17]. So, the clinical use of IVIG is not recommended for the treatment of sepsis-associated thrombocytopenia [18].
As a full-length glycosylated TPO, recombinant human thrombopoietin (rhTPO) has biological functions similar to endogenous TPO [19]. Studies have shown that rhTPO efectively increases peripheral blood PLT levels in patients with immune-or chemotherapy-related thrombocytopenia and reduces adverse efects [20,21]. Terefore, rhTPO may be a rescue therapy for septic patients with thrombocytopenia. In addition, a recent meta-analysis demonstrated that in patients with sepsis-associated thrombocytopenia, PLT levels were signifcantly elevated on the 7 th day after administration of rhTPO, and blood product transfusion volumes were reduced [15]. Notably, two diferent dosing regimens of rhTPO, including rhTPO 300 U/kg/day and rhTPO 15000 U/day, were available for treating septic patients with thrombocytopenia, but which dosing regimen might be better remains unclear [22]. Terefore, the present network meta-analysis aimed to compare the therapeutic values of two dosing regimens of rhTPO.

Study Design.
We performed this study according to the preferred reporting items for systematic reviews and metaanalyses (PRISMA) extension statement for reporting network meta-analyses [23]. Ethical approval and informed consent were not required as this was a network metaanalysis of published studies. Moreover, we did not register a formal protocol for this network meta-analysis.

Eligibility Criteria.
We designed eligibility criteria based on the PICOS acronym, and studies that met the following criteria were included in this network meta-analysis: (a) participant (P): adult patient diagnosed with sepsis-related thrombocytopenia [24]; (b) intervention (I): rhTPO was prescribed for patients in the study group; (c) comparison (C): patients in the control group were not prescribed additional therapy (NAT) or IVIG in addition to conventional antibiotic therapy (CAT); (d) outcomes (O): reported at least one of the following: 28-day mortality, the length of ICU stay, platelet level on the 7 th day, and transfusion of blood products including red blood cells (RBCs), plasma, and platelets; and (e) study design (S): only randomized controlled trials (RCTs) with full texts published in English and Chinese.
Studies that met the following criteria were excluded from this study: (a) ineligible study designs, such as case reports and conference abstracts; (b) replicate studies published by the same author or project; and (c) essential data for synthesis was not available.

Study Selection.
After the removal of duplicate records, all titles and abstracts of the remaining studies were independently screened by two researchers (Dandan Chen and Chaochao Wei) for the initial eligibility assessment. Ten, the full texts of the remaining studies were retrieved for the fnal eligibility assessment. With the help of a third researcher, any conficts were resolved.

Data Extraction. Basic information was independently extracted by two researchers (Dandan Chen and Chaochao
Wei) from included studies, including the frst author, publication year, sample size, percentage of male participants, age of participants, baseline PLT level, baseline acute physiology, age, chronic health evaluation II/III (APACHE II/III) score, and outcome data. We emailed the leading author to obtain related data when essential data were not available in the original study. Any conficts were resolved with the help of a third researcher (Xingjun Cai).

Outcomes of Interest.
We regarded the 28-day mortality and the length of ICU stay as the primary outcomes, and PLT levels on the 7 th day posttreatment and transfusion of blood products including RBC, plasma, and PLT, as the secondary outcome.

2.7.
Geometry of the Evidence Network. Te evidence structure for each outcome was displayed using a network plot. In the network plot, the size of the node is weighted by the accumulated sample size, marked as white numerical values, and the width of the solid line is weighted by the number of direct comparisons, marked as a black numerical value close to the solid line [25]. Furthermore, the dotted line indicates the lack of direct comparison between the two interventions.
2.8. Risk of Bias Assessment. Two independent researchers (Dandan Chen and Yu Hou) assessed the risks of bias of each study from the following seven items, according to the Cochrane risk of bias assessment tool [26]: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcomes assessment, incomplete outcome data, selective reporting, and other bias. Based on the evaluation criteria, each item was rated as "low," "unclear," or "high" risk. Any conficts were resolved with the help of a third researcher.

Statistical Analysis.
Te odds ratio (OR) with a 95% confdence interval (CI) was used to express the pooled result for 28-day mortality, and the mean diference (MD) with 95%CI was used to express the diferences in the length of ICU stay, PLT levels on the 7 th day posttreatment, and transfusion of RBC, plasma, and PLT. Te transitivity of included studies was assessed based on clinical and methodological characteristics [27,28]. Consistency between direct and indirect efects was assessed based on the global consistency model test [29] and the local consistency model test [30]. Meanwhile, the node-splitting method was used to check whether there was an inconsistency in the closed loop [31,32]. A random efect model was used to calculate the relative efcacy of diferent doses [33], and a forest plot was used to show the diferences between interventions [34]. Te surface under the cumulative ranking (SUCRA) plot was used to show the ranking of diferent interventions in the same outcome [35]. Publication bias was checked based on the comparison-adjusted funnel plot [36]. All analyses were performed using STATA 14.0 (StataCorp LP, College Station, Texas, USA) with the "network" command [37]. p < 0.05 was considered to be a statistical diference.

Transitivity Assessment.
We conducted a transitivity assessment between comparisons based on fve main characteristics, including sample size, the proportion of males, mean age, baseline PLT levels, and APACHE scores. As shown in Table S2, transitivity was determined for most of the comparisons, except for rhTPO 15000 U/kg/day vs. NAT (p � 0.034) and rhTPO 300 U/day vs. NAT (p � 0.042) for male proportion and IVIG vs. NAT (p � 0.003) for disease severity.

28-Day
Mortality. For 28-day mortality, a network plot of the evidence structure is shown in Figure 2. Global and local consistency model tests showed no inconsistency ( Figure S2), and the consistency model was used for network meta-analysis. No signifcant diference was found between treatment strategies (Figure 3(a)). Te results of SUCRA showed that rhTPO 300 U/day had the highest probability of being the best (85.5%), followed by rhTPO 15000 U/kg/day (46.4%) (Figure 3(b)).

Te Length of ICU Stay.
For the length of the ICU stay, a network plot of the evidence structure is shown in Figure 4. Global and local consistency model tests showed no inconsistency ( Figure S3), so the consistency model was chosen. No signifcant diference was detected between treatment strategies ( Figure 5(a)). Te results of SUCRA showed that rhTPO 1500 U/kg/day had the highest probability of being the best (95.9%), followed by rhTPO 300 U/ day (64.3%) (Figure 3(b)).

PLT Level on the 7 th Day.
For the PLT level on the 7 th day posttreatment, a network plot of the evidence structure is shown in Figure S4a. Global and local consistency model tests showed no inconsistency ( Figure S5a), so the consistency model was chosen. Te pooled results showed that rhTPO 300 U/day and rhTPO 15000 U/kg/day signifcantly increased PLT levels on the 7 th day posttreatment compared with IVIG and NAT ( Figure 6(a)). Te results of SUCRA showed that rhTPO 1500 U/kg/day had the highest probability of being the best (91.6%), followed by rhTPO 300 U/ day (74.4%).

Transfusion of Blood Products.
For the transfusion of RBCs, a network plot of the evidence structure is shown in Figure S4a. Global and local consistency model tests showed no inconsistency ( Figure S5b), so the consistency model was selected. Pooled results showed that rhTPO 300 U/day and rhTPO 15000 U/kg/day were associated with lower RBC transfusions compared with IVIG and NAT, respectively ( Figure 6(b)). Te results of SUCRA showed that rhTPO 300 U/day had the highest probability of being the best (88.7%), followed by rhTPO 15000 U/kg/day (76.1%).
For the transfusion of plasma, a network plot of the evidence structure is shown in Figure S4c. Global and local consistency model tests showed no inconsistency (Figure S5c), and a consistency model was selected. No signifcant diference was detected between treatment International Journal of Clinical Practice 3 strategies ( Figure 6(c)). Te results of SUCRA showed that rhTPO 300 U/day had the highest probability of being the best (89.6%), followed by rhTPO 15000 U/kg/day (52.5%).
For the transfusion of PLT, a network plot of the evidence structure is shown in Figure S4d. Global and local consistency model tests indicated no inconsistency ( Figure S5d), and the consistency model was selected. Te pooled results showed that rhTPO 300 U/day and rhTPO 15000 U/kg/day were associated with lower PLT transfusions compared with IVIG and NAT, respectively (Figure 6(d)). Te results of SUCRA indicated that rhTPO 300 U/day had the highest probability of being the best (95.2%), followed by rhTPO 15000 U/kg/day (70.1%).

Closed-Loop Inconsistency and Publication Bias.
For each outcome, the closed-loop inconsistency was also evaluated based on the node-splitting method. As shown in Figure S6, no closed-loop inconsistency was found, indicating the robustness of all pooled results. In addition, the publication bias of the primary outcomes was further examined. As shown in Figure S7, the symmetric outline of the comparison-adjusted funnel plots indicated that there was no publication bias.

Discussion
Evidence suggests that patients with sepsis-related thrombocytopenia have longer ICU stays [4] and a worse prognosis [8,10]. As a novel rescue therapy, rhTPO has been shown to be efective in increasing peripheral PLT levels [19], as confrmed by a recent pairwise meta-analysis [15]. Unfortunately, which doses of rhTPO might be optimal for septic patients with thrombocytopenia remains unclear as a direct comparison is absent. Terefore, in order to draw frm conclusions, this study indirectly investigated the comparative therapeutic values of two available doses of rhTPO by introducing a network meta-analysis. Based on the results of this network meta-analysis, rhTPO 300 U/day may be the best option for reducing 28-day mortality and blood transfusion requirements. However, rhTPO 15000 U/ kg/day may be the best option for shortening the ICU stay and increasing peripheral PLT levels on the 7th day posttreatment.

International Journal of Clinical Practice
Notably, a pairwise meta-analysis [15] determined whether rhTPO is a benefcial strategy in septic patients with thrombocytopenia. Based on pooled results from 10 eligible RCTs, rhTPO was associated with increased PLT levels on the 7 th day posttreatment and decreased blood product transfusions during hospitalization. Unfortunately, the optimal dose of rhTPO was not determined in this metaanalysis, which greatly confounds clinical decision-making. Furthermore, this meta-analysis missed an eligible study [38] that investigated the therapeutic values between rhTPO 300 U/day and IVIG. Furthermore, since the publication of this meta-analysis, 2 additional eligible studies have been provided. In contrast to the previous meta-analysis, the present study included all available studies to determine the therapeutic values of 2 diferent doses of rhTPO by introducing a network meta-analysis technique. Terefore, the optimal dose for each outcome was determined based on more robust and reliable results.
Tis network meta-analysis yielded some robust fndings due to 3 methodological strengths: (a) creative use of the network meta-analysis to determine the comparative therapeutic values between two diferent doses of rhTPO that were not directly compared in the original study; (b) SUCRA plots based on ranking probabilities were used to determine the optimal dose for each clinical outcome; and (c) this network meta-analysis included both control strategies including no additional treatment and IVIG, to increase statistical power.   Figure 3: Network meta-analysis of the relative efcacy (a) and the rank probabilities (b) among diferent treatment strategies in terms of 28-day mortality. NAT, no additional treatment; IVIG, intravenous immunoglobulin; rhTPO 300, 300 U/kg/d recombinant human thrombopoietin; rhTPO 15000, 15000 U/d recombinant human thrombopoietin. 6 International Journal of Clinical Practice Certainly, some limitations may have negatively impacted our fndings: (a) the sample size is insufcient, because although 13 RCTs were included in this network meta-analysis, only 963 participants were accumulated; (b) all 13 RCTs did not explicitly describe whether participants, personnel, and outcomes assessment were blinded, which could lead to biased implementation; (c) all 13 studies did not report protocol registration and confict of interest, which could be a source of bias; (d) all 13 studies were conducted in China, so the results should be cautiously used with caution in diferent clinical settings; (e) although we conducted this network meta-analysis strictly in accordance with the methodological framework recommended by the Cochrane handbook, a formal public protocol was not available for the current network metaanalysis, which will inevitably negatively afect the transparency of this network meta-analysis; (f ) due to limited data, we did not assess other outcomes such as the length of activated partial thromboplastin time (APTT) and prothrombin time (PT) on day 7, which may be negative for the comprehensiveness of our fndings; (g) one study used APACHE III for severity assessment, and it difered from other eligible studies which used APACHE II for severity assessment, which might be the source of bias; and (h) we confrmed transitivity assumption among most of the available comparisons; however, 3 of these comparisons   Figure 5: Network meta-analysis of the relative efcacy (a) and the rank probabilities (b) among diferent treatment strategies in terms of the length of ICU stay. NAT, no additional treatment; IVIG, intravenous immunoglobulin; rhTPO 300, 300 U/kg/d recombinant human thrombopoietin; rhTPO 15000, 15000 U/d recombinant human thrombopoietin. difered signifcantly in male proportion and disease severity, which may inevitably compromise the reliability of our fndings because subgroup analysis cannot be performed due to limited studies.

Conclusion
Tis network meta-analysis suggests that rhTPO 300 U/day may be the best option for improving 28-day mortality and transfusion of blood products for the treatment of septic patients with thrombocytopenia. However, rhTPO 15000 U/ kg/day may be the best option for shortening the ICU stay and increasing PLT levels on the 7 th day posttreatment. However, given the limitations, we recommend more studies to further validate our fndings.

Data Availability
All data generated or analyzed during this study are included in this published article/as supplementary information fles.

Conflicts of Interest
Te authors declare that they have no conficts of interest.

Authors' Contributions
Xingjun Cai conceptualized the study; Xingjun Cai and Dandan Chen developed the methodology and used the software. Xingjun Cai and Dandan Chen validated the data. Dandan Chen conducted the formal analysis and investigated the data. Dandan Chen and Yu Hou collected the resource materials. Xingjun Cai and Dandan Chen curated the data. Dandan Chen wrote of the original manuscript. Chaochao Wei and Yu Hou wrote, reviewed, and edited the data. Chaochao Wei visualized the data, and Xingjun Cai supervised the study. Xingjun Cai performed project administration. All authors read and approved the fnal manuscript.

Supplementary Materials
Search strategy of pubmed, search strategy of Embase, and search strategy of Cochrane library are given. Table S1. Results of individual studies included in this network metaanalysis. Table S2. Transitivity between diferent comparisons based on major characteristics. Figure S1. Risk of bias summary (a) and graph (b). Red (−), yellow (?), and green (+) color indicates high, unclear, and low risk of bias, respectively. Figure S2. Consistency model test of 28-day mortality. NAT, no additional treatment; IVIG, intravenous immunoglobulin; rhTPO 300, 300 U/kg/ d recombinant human thrombopoietin; rhTPO 15000, 15000 U/d recombinant human thrombopoietin. Figure S3. Consistency model test of the length of ICU stay. NAT, no additional treatment; IVIG, intravenous immunoglobulin; rhTPO 300, 300 U/kg/d recombinant human thrombopoietin; rhTPO 15000, 15000 U/d recombinant human thrombopoietin. Figure S4. Evidence network of secondary outcomes including the level of platelet on the 7 th day (a), transfusion of RBC (b), transfusion of plasma (c), and transfusion of platelet (d    Figure S5. Te consistency model test of the secondary outcomes including the level of platelet on the 7 th day (a), transfusion of RBC (b), transfusion of plasma (c), and transfusion of platelet (d). NAT, no additional treatment; IVIG, intravenous immunoglobulin; rhTPO 300, 300 U/kg/ d recombinant human thrombopoietin; rhTPO 15000, 15000 U/d recombinant human thrombopoietin. Figure S6.