1 TREATMENT EFFICACY FOR ADULT CHRONIC IMMUNE THROMBOCYTOPENIA: A SYSTEMATIC REVIEW AND NETWORK META-ANALYSIS TEERAYA PUAVILAI A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE (MEDICAL EPIDEMIOLOGY) FACULTY OF GRADUATE STUDIES MAHIDOL UNIVERSITY 2017 COPYRIGHT OF MAHIDOL UNIVERSITY

2 Thesis entitled TREATMENT EFFICACY FOR ADULT CHRONIC IMMUNE THROMBOCYTOPENIA: A SYSTEMATIC REVIEW AND NETWORK META-ANALYSIS... Miss Teeraya Puavilai Candidate... Assoc. Prof. Atiporn Ingsathit, Ph.D. (Clinical Epidemiology) Major advisor Asst. Prof. Chusak Okaschareon, Asst. Prof. Sasivimol Rattanasiri, M.D., Ph.D. (Clinical Epidemiology) Ph.D. (Statistics) Co-advisor Co-advisor Prof. Patcharee Lertrit, Asst.Prof. Chusak Okaschareon, M.D., Ph.D. (Biochemistry) M.D., Ph.D. (Clinical Epidemiology) Dean Program Director Faculty of Graduate Studies Master of Science Program in Medical Mahidol University Epidemiology Faculty of Medicine, Ramathibodi Hospital Hospital Mahidol University

3 Thesis entitled TREATMENT EFFICACY FOR ADULT CHRONIC IMMUNE THROMBOCYTOPENIA: A SYSTEMATIC REVIEW AND NETWORK META-ANALYSIS was submitted to the Faculty of Graduate Studies, Mahidol University for the degree of Master of Science (Medical Epidemiology) on July 26, Miss Teeraya Puavilai Candidate... Assoc. Prof. Ammarin Thakkinstian, Ph.D. (Clinical Epidemiology & Community Medicine) Chair Prof. Tanin Intragumtornchai, M.D., M.Sc., (Clinical Epidemiology) Member Assoc. Prof. Atiporn Ingsathit, M.D., Ph.D. (Clinical Epidemiology) Member.... Asst. Prof. Sasivimol Rattanasiri, Ph.D. (Statistics) Member Asst.Prof. Chusak Okaschareon, M.D., Ph.D. (Clinical Epidemiology) Member Prof. Patcharee Lertrit, M.D., Ph.D. (Biochemistry) Dean Faculty of Graduate Studies Mahidol University Prof. Piyamitr Sritara, M.D., FRCPT, FACP, FRCP (T) Dean Faculty of Medicine, Ramathibodi Hospital Mahidol University

4 iii ACKNOWLEDGEMENTS I would like to express my gratitude to Assoc. Prof. Ammarin Thakkinstian, Assist. Prof. Dr. Sasivimol Rattanasiri, and Assoc. Prof. Atiporn Ingsathit for valuable suggestion, strong encouragement and support throughout my thesis project. I appreciate Ms. Sukanya Siriyotha for support on some part of data management and statistical analysis and Mr. Stephen Pinder for English editing and presentation training. I also would like to thank Ms. Paneevon Palakawong Na Ayutthaya, Ms. Sudasiri Sriwiang and all staffs of the Section for Clinical Epidemiology and Biostatistics, Ramathibodi Hospital, Mahidol University for their grateful help and technical support about administrative management of my thesis. I would like to take this opportunity to appreciate all professors who have taught and facilitated me in every course of my master s degree program. Last, I would like to express immeasurable appreciation and deepest gratitude to my father, my family, my friends and my classmate for their love, continual encouragement, support and understanding throughout my study. Teeraya Puavilai

5 Fac. of Grad. Studies, Mahidol Univ. Thesis / iv TREATMENT EFFICACY FOR ADULT CHRONIC IMMUNE THROMBOCYTOPENIA : A SYSTEMATIC REVIEW AND NETWORK META-ANALYSIS TEERAYA PUAVILAI RAME/M M.Sc. (MEDICAL EPIDEMIOLOGY) THESIS ADVISORY COMMITTEE: ATIPORN INGSATHIT, M.D., Ph.D., SASIWIMOL RATANASIRI, Ph.D., CHUSAK OKASCHAREON, M.D., Ph.D. ABSTRACT Several drugs have been used as second line treatments for adult chronic immune thrombocytopenia (ITP) but their efficacies were still controversial. Therefore, we conducted a systematic review and network meta-analysis of randomized controlled trials to assess their efficacies and safeties. In addition, probability of being best treatments (i.e., high efficacy and low severe adverse events) was estimated and ranked. Studies were identified from MEDLINE and SCOPUS databases. The outcomes of interest were platelet response, platelet count, any bleeding, and composite of severe adverse events. Data were independently extracted by two reviewers. Treatment effects were indirectly compared using a multivariate network metaanalysis. Twelve studies involving 1,314 patients were included, comprising 8 treatments, i.e., danazol, rituximab, eltrombopag, romiplostim, rhtpo, rhtpo+cyclosporin, rhtpo+danazol, and rhtpo+rituximab. Platelet response for patients receiving romiplostim and eltrombopag had significant higher efficacy than placebo with the pooled RR of 5.84 (95% CI: 1.62, 21.09) and 3.21 (95% CI: 2.28, 4.52). Platelet count for patients receiving eltrombopag and romiplostim were significantly higher than placebo with pooled mean differences of (95% CI: 43.08, 78.02) and (95% CI: 13.69, 95.22) x10 9 /L. Any bleeding for eltrombopag was lower than placebo with pooled RR of 0.84 (95% CI: 0.75, 0.93). Composite serious adverse events of eltrombopag and romiplostim were higher but not significant different when compared to placebo with pooled RR of 1.37 (95% CI: 0.62, 3.00) and pooled RR of 1.64 (95% CI: 1.03, 2.61). Romiplostim and eltrombopag had the highest probability of being the best treatment for platelet response with probability 64.3% (SUCRA 90.5) and probability 35.2% (SUCRA 83.3). These two drugs were also the highest ranks for platelet count increasing with probability 21.8% (SUCRA 66.1) and probability 40.4% (SUCRA 77.4). In addition, these two drugs were the best for being lowest probability of any bleeding with probability 38.2% (SUCRA 63.2) and 22.5% (SUCRA 61.4). Furthermore, the two drugs had the lowest probability of being composite serious adverse events with probability 55.8% (SUCRA 84.0) and 24.9% (SUCRA 65.4). The predictive intervals suggested that eltrombopag and romiplostim might be effective in the future for platelet response and platelet count in adult chronic ITP patients. In conclusion, eltrombopag and romiplostim were the best treatments that had high efficacy in platelet response and platelet counts but lowest in severe adverse events for second line treatments in adult chronic ITP patients. KEY WORDS: ADULT/CHRONIC IMMUNE THROMBOCYTOPENIA/ EFFICACY/ SAFETY/ NETWORK META-ANALYSIS 123 pages

6 Fac. of Grad. Studies, Mahidol Univ. Thesis / v การทบทวนวรรณกรรมอย างเป นระบบและใช เคร อข ายการว เคราะห อภ มานเพ อศ กษาประส ทธ ผลของยาร กษาภาวะเกล ดเล อด ต าเร อร งท เก ดจากการสร างแอนต บอด ต อเกล ดเล อดของต วเองในผ ใหญ TREATMENT EFFICACY FOR ADULT CHRONIC IMMUNE THROMBOCYTOPENIA : A SYSTEMATIC REVIEW AND NETWORK META-ANALYSIS ธ รยา พ วว ไล RAME/M วท.ม. (ว ทยาการระบาดทางการแพทย ) คณะกรรมการท ปร กษาว ทยาน พนธ : อต พร อ งค สาธ ต, M.D., Ph.D., ศศ ว มล ร ตนส ร, Ph.D., ช ศ กด โอกาศเจร ญ, M.D., Ph.D. บทค ดย อ ประส ทธ ผลของยาท ใช ส าหร บร กษาภาวะเกล ดเล อดต าเร อร งท เก ดจากการสร างแอนต บอด ต อเกล ดเล อดของ ต วเองในผ ใหญ ย งไม ม ข อสร ปท ช ดเจน ด งน นผ น พนธ จ งได ท าการทบทวนวรรณกรรมอย างเป นระบบจากงานว จ ยทางคล น ก แบบส มและม กล มควบค ม แล วใช เคร อข ายการว เคราะห อภ มานเพ อประเม นประส ทธ ผล ความปลอดภ ยและการเป นยาท ด ท ส ด ตามล าด บโดยม ประส ทธ ผลส งแต ม ภาวะแทรกซ อนต า ด วยการส บค นหางานว จ ยจากฐานข อม ล MEDLINE และ SCOPUS ผล การศ กษาท สนใจได แก การตอบสนองของเกล ดเล อด ปร มาณเกล ดเล อด ภาวะเล อดออกและภาวะแทรกซ อนร นแรง ผ ว จ ย 2 ท านท างานเป นอ สระต อก นในการเก บข อม ล ผลการร กษาของยาแต ละชน ดได ถ กน ามาเปร ยบเท ยบโดยตรงและโดยอ อมด วย การว เคราะห อภ มานเคร อข าย ผ น พนธ ได รวบรวมงานว จ ยจ านวน 12 ฉบ บซ งท าในผ ป วย 1,314 ราย ใช ยา 8 ชน ด ได แก danazol, rituximabม eltrombopag, rhtpo, romiplostim, rhtpo+cyclosporin, rhtpo+danazol, และ rhtpo+rituximab เข าใน การทบทวนวรรณกรรมอย างเป นระบบฉบ บน ผลการว เคราะห อภ มานเคร อข ายพบว าผ ป วยท ได ร บยา romiplostimและ eltrombopag ม ประส ทธ ผลต อการตอบสนองของเกล ดเล อดท ส งข นอย างม น ยส าค ญทางสถ ต เม อเท ยบก บ placebo ด วย pooled RR 5.84 (95% CI: 1.62, 21.09) และ 3.21 (95% CI: 2.28, 4.52) ประส ทธ ผลต อปร มาณเกล ดเล อดท ส งข นอย างม น ยส าค ญทาง สถ ต เม อเท ยบก บ placebo ด วย pooled mean difference (95% CI: 13.69, 95.22) และ (95% CI: 43.08, 78.02) x10 9 / ล ตร ผ ป วยท ได ร บยา eltrombopag พบภาวะเล อดออกน อยกว ากล ม placebo ด วย pooled RR 0.84 (95% CI: 0.75, 0.93) ภาวะแทรกซ อนร นแรงจากยา eltrombopag และ romiplostim ส งกว าแต ไม แตกต างอย างม น ยส าค ญจากกล ม placebo ด วย pooled RR of 1.37 (95% CI: 0.62, 3.00) และ 1.64 (95% CI: 1.03, 2.61) ยา romiplostim และ eltrombopag ม โอกาสในการเป นยาท ด ท ส ดด านการตอบสนองของเกล ดเล อดท ส งข นอย ท 64.3% (SUCRA 90.5) และ 35.2% (SUCRA 83.3) ด านปร มาณเกล ดเล อดท ส งข นอย ท 21.8% (SUCRA 66.1) และ 40.4% (SUCRA 77.4) ภาวะเล อดออกท น อยกว าอย ท 21.8% (SUCRA 66.1) และ 40.4% (SUCRA 77.4) ด านภาวะแทรกซ อนร นแรงท เก ดน อยกว าอย ท 55.8%(SUCRA 84.0) และ 24.9% (SUCRA 65.4) นอกจากน ยา ท งสองชน ดน าจะม ประส ทธ ผลท ด ท ควรน าไปใช ต อไปในอนาคต โดยสร ป ยา eltrombopag และ romiplostim เป นยาท ม ประส ทธ ผลส งในแง การตอบสนองของเกล ดเล อดหร อปร มาณเกล ดเล อดแต ม ภาวะแทรกซ อนน อยท ส ด เหมาะสมส าหร บการ เล อกใช เป นยาร กษาภาวะเกล ดเล อดต าเร อร งในผ ใหญ 123 หน า

7 vi CONTENTS Page ACKNOWLEDGEMENTS iii ABSTRACTS (ENGLISH) iv ABSTRACTS (THAI) v LIST OF TABLES viii LIST OF FIGURES x CHAPTER I BACKGROUND AND RATIONALE Disease background Rationale Research questions Research objectives 5 CHAPTER II METHODOLOGY Study design Study identification Selection of studies Data extraction Risk of bias assessment Statistical analysis 12 CHAPTER III RESULTS Study selection Characteristics of included studies Risk of bias Pairwise meta-analysis Network meta-analysis 25 CHAPTER IV DISSCUSSION Summary of findings Comparison with previous studies Strengths and limitations 94

8 vii CONTENTS (cont.) Page 4.4 Conclusions 94 REFERENCES 95 APPENDICES 101 Appendix A Search strategy by MEDLINE 102 Appendix B Search strategy by SCOPUS 106 Appendix C Data extraction form 110 Appendix D Risk of bias assessment 112 Appendix E STATA commands for pairwise meta-analysis 114 and network meta-analysis BIOGRAPHY 123

9 viii LIST OF TABLES Table Page 2.1 Search results by MEDLINE Search results by SCOPUS Characteristics of included studies for adult chronic ITP patients Describe treatment regimens Risk of bias assessment of included studies for adult chronic ITP patients Efficacy of second line regimen on platelet response in adult chronic ITP 37 patients: Pairwise meta-analysis 3.5 Efficacy of second line regimen on platelet count in adult chronic ITP 38 patients: Pairwise meta-analysis 3.6 Heterogeneity assessment of mean difference of platelet count between 40 active treatment and placebo 3.7 Efficacy of second line regimen on any bleeding in adult chronic ITP 41 patients: Pairwise meta-analysis 3.8 Number of composite serious adverse events for adult chronic ITP patients Safety of second line regimen on composite serious adverse events in 43 adult chronic ITP patients: Pairwise meta-analysis 3.10 All possible pairwise comparisons of platelet response between active 44 versus active treatments versus placebo: A network meta-analysis 3.11 The probability of being highest for platelet response outcome All possible pairwise comparisons of mean platelet counts between active 46 versus active treatments versus placebo: A network meta-analysis 3.13 The probability of being highest for platelet count outcome All possible pairwise comparisons of any bleeding between active versus 48 active treatments versus placebo: A network meta-analysis 3.15 The probability of being lowest for any bleeding outcome 49

10 ix LIST OF TABLES (cont.) Table Page 3.16 All possible pairwise comparisons of composite serious adverse event 50 between active versus active treatments versus placebo: A network meta- analysis 3.17 The probability of being lowest for composite serious adverse event 51 outcome

11 x LIST OF FIGURES Figure Page 3.1 Flow chart of study selection Summary of risk of bias assessment of included studies for adult chronic 53 ITP patients 3.3 Forest plot of eltrombopag versus placebo for platelet response Forest plot of romiplostim versus placebo for platelet response Funnel plot of eltrombopag versus placebo for platelet response Funnel plot of romiplostim versus placebo for platelet Contour-enhanced funnel plot of eltrombopag versus placebo for platelet 58 response 3.8 Contour-enhanced funnel plot of romiplostim versus placebo for platelet 59 response 3.9 Forest plot of eltrombopag versus placebo for platelet count Forrest plot of romiplostim versus placebo for platelet count Forest plot of eltrombopag versus placebo for any bleeding Funnel plot of eltrombopag versus placebo for any Forest plot of eltrombopag versus placebo for composite serious 64 adverse events 3.14 Forest plot of romiplostim versus placebo for composite serious 65 adverse events 3.15 Funnel plot of eltrombopag versus placebo for composite serious 66 adverse events 3.16 Funnel plot of romiplostim versus placebo for composite serious 67 adverse events 3.17 Network map of platelet response Network contribution plot of platelet response 69

12 xi LIST OF FIGURES (cont.) Figure Page 3.19 Surface under the cumulative ranking curves for platelet response of all 70 treatments 3.20 Predictive interval plot of platelet response Adjusted funnel plot of platelet response Network map of platelet count Network contribution plot of platelet count Surface under the cumulative ranking curves for platelet count of all 75 treatments 3.25 Predictive interval plot of platelet count Adjusted funnel plot of platelet count Network map of any bleeding Network contribution plot of any bleeding Surface under the cumulative ranking curves for any bleeding of all 80 treatments 3.30 Predictive interval plot of any bleeding Adjusted funnel plot of any bleeding Network map of composite serious adverse events including death, 83 thrombosis and infection 3.33 Network contribution plot of composite serious adverse events Surface under the cumulative ranking curves for composite serious 85 adverse events 3.35 Predictive interval plot of composite serious adverse events Adjusted funnel plot of composite serious adverse events Cluster rank plot for platelet response and composite serious adverse 88 events 3.38 Cluster rank plot for platelet count and composite serious adverse events Cluster rank plot for any bleeding and composite serious adverse events 90

13 Fac. of Grad. Studies, Mahidol Univ. M.Sc.(Medical Epidemiology) / 1 CHAPTER I BACKGROUND AND RATIONALE 1.1 Disease background Magnitude of problem Immune thrombocytopenia (ITP), also called idiopathic thrombocytopenic purpura, immune thrombocytopenic purpura or primary immune thrombocytopenia, is B-cell (and in some patients to a CD8+ T-cell) autoimmune reaction directed against circulating platelets and megakaryocytes, resulting in thrombocytopenia, and leading to life-threatening bleeding in some patients 1, 2. Most ITP is the primary whereas about 20% of can be the secondary ITP in which its primary could be from malignancy, systemic autoimmune disease, chronic viral infection (e.g., HIV or hepatitis virus B or C (HBV, HCV)), primary immune deficiency, or drugs 3. The incidence of ITP in adults was about 1.6 to 3.9/100,000 person-years 4 and it was even higher in patients aged 60 or older 3, 5.The prevalence of ITP is 9.5 per 100,000 but not known about the prevalence of severe bleeding at diagnosis Disease burden or impacts Clinical manifestations of ITP present as a bleeding tendency, bruising (purpura), or extravasation of blood from capillaries into skin and mucous membranes (petechiae) 1. Bleeding may occur in up to two-thirds of patients and one-third of patients are asymptomatic 7. Severity of bleeding is vary ranging from severe bleeding (which is quite rare e.g. intracranial hemorrhage, overt gastrointestinal bleeding, and hematuria) to mild petechiae on the extremities. The multicenter ITP registry reported that about 1.2% and 1.8% of patients has of gastrointestinal and central venous system bleeding, respectively 5, and bleeding risk might increase with age Predictors of severe bleeding include the degree of thrombocytopenia (from < /L to < /L, depending on the study), previous minor bleeding, and chronic ITP which is diagnosed

14 Teeraya Puavilai Background and Rationale / 2 at >12 months prior), although individual predictors are not consistent across all studies Treatments and goal of treatments The first line treatment of acute ITP is corticosteroids, e.g., oral prednisolone, intravenous methylprednisolone or high dose dexamethasone. Intravenous immunoglobulin and anti-d are used to rescue and rapidly increase platelet level in severe life threatening bleeding. The goal of ITP treatment is to improve platelet count in order to prevent clinically important bleeding, rather than to normalize platelet count 2, Persistent or chronic ITP patient are diagnosed if patients failed to achieve response to initial treatment during 3 to 12 months or longer. These patients require to use second line treatments if they have higher risk of bleeding due to other comorbidities (e.g., hypertension, renal insufficiency), or requiring antiplatelet, anticoagulant for some serious underlying diseases, athletic activities, careers with high risk for trauma, or intolerant adverse effects of corticosteroids Second line therapies include splenectomy and medications. Balancing the benefits and risks of second line treatment are both physicians and patients concerns. Preference of physician in choosing the second line therapies was varied by their experience 16. For instance, clinicians with 20 years experience or longer might prefer splenectomy rather than medications such as rituximab or thrombopoietin receptor agonists. However, patients might not prefer splenectomy because of perceptions that impact of their illness did not warrant 16. There are some patients who transient response or relapse after splenectomy or fail to respond to splenectomy Splenectomy itself also increased risk from surgical procedure as well as long term risks of increased susceptibility to infection and vascular events 20. A systematic review of case series of splenectomy for ITP showed that complications and death occurred 9.6% and 1% for laparoscopic splenectomy, and 12.9% and 0.2% for open splenectomy 21 ; whereas infection was very rare about 1% to 3% 22. Patients might also risk for venous thromboembolism 23.

15 Fac. of Grad. Studies, Mahidol Univ. M.Sc.(Medical Epidemiology) / 3 Several modalities for second line medical treatments have been prescribed including immunosuppressive agents (i.e., azathioprine, danazol, cyclosporin, cyclophosphamide, vincristine, vinblastine, mycophenolate mofetil, dapsone), monoclonal antibody (i.e., rituximab), thrombopoietin receptor agonists (TRA, i.e., elthrombopag, and romiplostim) or combinations of these treatments, which aim to improve platelet level /L with no bleeding symptom 3, These treatments are appropriate for the patient who continue significant bleeding, platelet count < /L to /L, and platelet count in the range of /L to /L after first line treatment 14, 15. Previous evidences showed that azathioprine orally 1-2mg/kg/day, cyclosporine orally mg/kg of body weight per day and mycophenolate mofetil taken 1 g twice daily for at least 3-4 week could improve complete response rate about 40% Cyclophosphamide, an alkylating agent, can be used either oral form with dosage of 1-2 mg/kg/day for up to 4 months, or intravenous form at g/m 2 for 1-3 doses every 2-4 weeks could achieve platelet response 24% to 85% 27, 28. Danazol, an androgen, could be used orally at 200 mg for 2-4 times daily, which could achieve platelet response about 60% 29. Dapsone, an anti-leprotic drug, could be used orally with mg/day and platelet response rate was about 50% 30. Vincristine and vinblastine can be used at 1-2 mg per infusion weekly with total dose of 6 mg and 10 mg per infusion weekly with total dose of 30 mg, respectively. Platelet response rate of these drugs is transient response and ranges from 10% to 75% 31. Adverse effects of immunosuppressive agents are mostly mild to moderate severity and transient effects 32. Rituximab is a chimeric monoclonal antibody directed against the B cell surface protein CD20, which eliminates B cells by apoptosis, antibody dependent cytotoxicity, and complement mediated lysis. The possible doses of rituximab are 375 mg/m 2 intravenously weekly for 4 consecutive weeks or 100 mg/week for 4 weeks or 1000 mg on day 1 and day 15; these could achieve response rates about 40% Adverse effects of rituximab include infusion reactions and prolonged immunosuppression that result in reactivation of hepatitis B virus infection 36. Progressive multifocal leukoencephalopathy (PML) has been reported in few patients 37.

16 Teeraya Puavilai Background and Rationale / 4 TRA consist of small molecule peptide and non-peptide agents, which stimulate thrombopoietin receptor on megakaryocytes for both proliferation and differentiation and then result in increasing platelet productions. The two TRAs, i.e., romiplostim and eltrombopag, are currently FDA-approved for prescription in adult ITP patients. These drugs are well tolerated but remain regarding risk of thromboembolism and bone marrow fibrosis 38. Romiplostim can be used at a 1 to 10 µg/kg subcutaneous weekly injection. Overall platelet response rate is about 80% 39. Eltrombopag can be used at mg/day orally. Platelet response rate is about 60% Rationale As for complication of splenectomy as mentioned previously, a rate of splenectomy for ITP is reducing worldwide, and many patients prefer to go with medical treatments; therefore this study has focused on medications for second line treatment in patients with chronic ITP. Although there were 3 previous meta-analyses studies in ITP patients 41-43, these meta-analyses still could not answer the question of which treatment regimen is the best, i.e., better efficacy and lower adverse drug reaction as following reasons: The first meta-analysis 41 aimed to indirectly compare treatment response between romiplostin and elthrombopag in adults with chronic ITP by including only 3 RCTs (2 RCTs romiplostim vs placebo; 1 RCT eltrombopag vs placebo) in pooling. They demonstrated that romiplostim significant improvement for overall platelet response rate but no significant improvement for durable platelet response when compared with eltrombopag. The second meta-analysis 42 included 13 studies of chronic ITP including 7 studies of romiplostim (3 RCTs and 1 cohort in adults 3 RCTs in children), and 6 studies of eltrombopag ( 4 RCTs in adults and 2 RCTs in children). This meta-analysis directly pooled effects of TRA by combining eltrombopag and romiplostim, which showed significantly increased platelet response and durable platelet response, and decreased incidence of any or severe bleeding event when compared to placebo.

17 Fac. of Grad. Studies, Mahidol Univ. M.Sc.(Medical Epidemiology) / 5 The third meta-analysis 43 considered only additional effects of rituximab in standard treatments (i.e., dexamethasone) or other drug (i.e., rhtpo) comparing with standard treatments or rituximab itself by including 7 RCTs. The results by direct meta-analysis concluded that rituximab effectively elevated platelet count response in different cut-off for adult ITP patients. The metaanalysis by Wang et al 42 included more numbers of TRA studies but they combined romiplostim and elthrombopag in mixed children and adult TPA patients. Although the meta-analysis by Cooper et al 41 indirectly compared romiplostim with elthrombopag, and Feng et al 43 directly compared rituximab with mixed comparators, other treatment regimens (e.g., regimens of immunosuppressive agents and chimeric monoclonal antibody) for chronic ITP were not considered. None of them had compared between three regimens (i.e., immunosuppressive agents, chimeric monoclonal antibody, and TRA). Therefore, this systematic review and network meta-analysis was conducted to estimate treatment efficacy and safety for all available second line treatments of adult patients with chronic immune thrombocytopenia, and also estimate the probability of being the best treatment. 1.3 Research questions - What is the best efficacy of second line treatments in term of platelet response, severe bleeding and platelet count when compared to placebo or between any pairs of second line treatments in patients with chronic ITP? - What is the best safety of second line treatments in term of composite serious adverse events including death and thrombosis in patients with chronic ITP? 1.4 Research objectives Primary objectives To indirectly compare the risk ratios of platelet response, estimate probability being best treatments and ranking probability for platelet response in adult patients with chronic ITP.

18 Teeraya Puavilai Background and Rationale / Secondary objectives To indirectly compare the risk ratios of severe bleeding, and composite serious adverse events among second line treatments, estimate probability of being the best treatment and ranking probability To indirectly compare the mean difference of platelet count among second line treatments and estimate probability of being best treatments and ranking probability in adult patients with chronic ITP.

19 Fac. of Grad Studies, Mahidol Univ. M.Sc.(Medical Epidemiology) / 7 CHAPTER II METHODOLOGY 2.1 Study design The study design was a systematic review and network meta-analysis, which was performed according to Preferred Reposts of Systematic Review and Meta-Analysis expansion to a network meta-analysis. This study was registered in PROSPERO number CRD Study identification Studies were identified by two reviewers (TP and KT) using Medline (via PubMed) and Scopus (Sci Verse Scopus; Elsevier B.V.) databases up to December 31, Search terms were identified according to PICO as follow: Study patients Chronic ITP patients referred to patients who did not achieve remission (platelet /L and at least 2-fold increase the baseline count with absence of bleeding), or could not maintain their response after stop first line treatments with more than 3 months from diagnosis, and also need second line treatment to minimize the risk of clinically significant bleeding Intervention The second line intervention for chronic ITP consisted of - Monotherapy TRA: Eltrombopag Romiplostim rhtpo

20 Teeraya Puavilai Methodology / 8 Monoclonal antibody: rituximab Immunosuppressive agents: azathioprine, cyclosporine, cyclophosphamide, danazol, dapzone, mycophenolate mofetil, vincristine, or vinblastine Placebo - Combination therapy Any pairs of treatments - Dosage of second line treatments Eltrombopag: mg PO once daily for at least 6 weeks Romiplostim: 1-10 µg/kg SC once a week for at least 12 weeks rhtpo: 1 µg/kg SC once daily for 2 weeks Rituximab: 375 mg/m 2 IV once a week for 4 weeks Immunosuppressive agents: no standard recommendation Outcome of interests Our review focused on 2 outcomes: Primary outcom The primary outcome of interest was platelet response which was defined accordingly to original included studies as platelet level /L or /L at 6 weeks after receiving second line treatments The secondary outcomes were as follows: Platelet count at 6 weeks after receiving second line treatments Any bleeding during second line treatment Severe bleeding included grade 3gross blood loss or grade 4 debilitation blood loss according to the WHO bleeding scale 44 Thrombosis defined as occurrence of arterial or venous occlusion after receiving second line treatments Death: after receiving second line treatments Serious infection referred to grade 3-4 or serious infection 45

21 Fac. of Grad Studies, Mahidol Univ. M.Sc.(Medical Epidemiology) / 9 Composite serious adverse events included death, serious infection, and thrombosis. For those studies did not report this composite outcome but reported individual of the three events, a number of patients with highest frequency among death, thrombosis or serious infection would be used Search terms and strategies For Medline in PubMed based on PICO as follows: P: ( chronic *immune thrombocytopen* or chronic idiopathic thrombocytopen* or chronic ITP or refractory *immune thrombocytopen* or refractory idiopathic thrombocytopen* or refractory ITP ) and I: (vincristine or vinblastine or danazol or azathioprine or cyclophosphamide or cyclosporin* or ( mycophenolate mofetil or MMF) or eltrombopag or romiplostim or rituximab) and O: (bleeding or platelet count or thrombopoietin* or (death or mortality) or (thrombosis or VTE or venous thromboembolism )) and (Filters: clinical trial) For Scopus based on PICO as follows: P: ( chronic immune thrombocytopenia or chronic immune thrombocytopenic purpura or chronic idiopathic thrombocytopenia or chronic idiopathic thrombocytopenic purpura or refractory immune thrombocytopenia or refractory immune thrombocytopenic purpura or refractory idiopathic thrombocytopenia or refractory idiopathic thrombocytopenic purpura ) and I: (vincristine or vinblastine or danazol or azathioprine or cyclophosphamide or ( cyclosporin A or cyclosporine) or ( mycophenolate mofetil or MMF) or eltrombopag or romiplostim or rituximab) and O: (bleeding or platelet count or thrombopoietin or (death or mortality) or (thrombosis or VTE or venous thromboembolism )) and ( clinical trial )

22 Teeraya Puavilai Methodology / 10 Details of the search strategies for PubMed and Scopus databases were described in the Appendix C and Scopus database in the Appendix D, respectively. Where there were multiple publications with the same or overlapped subjects, which was detected by authorship lists, setting, study period and subjects, the most complete and/or recent results were used. The reference lists of the selected studies were also reviewed to identify additional relevant publications. The results of search from PubMed and Scopus were described in Table 2.1 and Table 2.2, respectively. 2.3 Selection of studies Inclusion criteria - Randomized controlled trial - Study in adult patients (aged >18 years) who were diagnosed as chronic immune thrombocytopenia - Study which had compared any pair of the following second line treatments: azathioprine, cyclophosphamide, cyclosporin, danazol, eltrombopag, mycophenolate mofetil, rituximab, romiplostim, vinblastine, and vincristine - Study which had any of the following clinical outcomes: platelet response, platelet count, any bleeding, death, thrombosis, and serious infection times Exclusion criteria - Insufficient data for pooling and no response from contacting authors three - No English language study) Design coding for ineligibility - Not interesting designs (case control study, cross-sectional study, cohort - Non-adult - Animal studies - Non-chronic or non-refractory ITP

23 Fac. of Grad Studies, Mahidol Univ. M.Sc.(Medical Epidemiology) / 11 - Non-second line treatments: IVIG, anti-d, prednisolone, methylprednisolone, dexamethasone - Not interested outcomes e.g. risk factor, prognosis, diagnosis, investigation - No comparison groups - Case reports/case series/letter to editor - Systematic reviews and meta-analyses - Review articles, reviews, recommendations, guidelines - Non-English Initial studies were identified from Medline via PubMed and Scopus and then managed using the EndNote software X7 by removing duplicates. The studies were independently selected based on the title and abstracts by one reviewers (TP), and studies that did not meet the eligibility criteria were eliminated with provided reasons. The full text articles were retrieved and read to make decisions to include in the review. Results of study selection between the two reviewers were validated. Any disagreements were solved by discussion and consensus. 2.4 Data extraction Summary data was extracted independently by 2 reviewers (TP and SR) using a standardized data extraction form, see Appendix A. The data extraction forms consisted of 4 parts including general information, study characteristics, general characteristics of participants, and data for pooling (Appendix A). - The general information of the studies included the author, corresponding author, author s , journal and year of publication. - The study characteristics of selected studies included country, study design and period of study. - The general baseline characteristics of participants included mean age, gender, duration of ITP since diagnosis, platelet count, cut-off for platelet count, thrombopoietin level, history of previous treatment, and history of splenectomy. - The data for pooling of the studies were extracted into dichotomous and continuous outcomes. For dichotomous outcomes, cut-off for platelet response, number

24 Teeraya Puavilai Methodology / 12 of subjects and number of events or risk ratio and 95% confidence interval for each treatment and for each study were extracted. For continuous outcomes, number of subjects, mean or median, standard deviation or range for each treatment and for each study were extracted. 2.5 Risk of bias assessment The quality of studies was independently assessed by 2 reviewers (TP and SR). Disagreement was solved by discussion with senior authors (AT). The risk of bias tool was used to assess the quality of randomized controlled trials including 6 domains: selection bias, performance bias, attrition bias, detection bias, detection bias, reporting bias, and other bias. Seven sources of bias were considered for assessing, which focus on random sequence generation and allocation concealment for selection bias; blinding of participants and personnel for performance bias; blinding of outcome assessment for detection bias; incomplete outcome data for attrition bias; selective outcome reporting for reporting bias; and other sources of bias. Each item was graded as yes and no if there was evidence of low and high risk of bias, respectively. If there was insufficient information to judge, it was classified as unclear. The details of this checklist had been presented in Appendix B. 2.6 Statistical analysis Pairwise meta-analysis Pairwise meta-analysis was performed on 4 outcomes, i.e., platelet response, any bleeding, platelet count, and composite serious adverse events (any of death, thrombosis, and serious infection) in any pair of 2 treatment pairs with having at least 3 studies. For platelet response, any bleeding, and composite serious adverse events, risk ratios (RR) were applied to estimate treatment effects. For platelet count outcome, an unstandardized mean difference was used to estimate the treatment effects.

25 Fac. of Grad Studies, Mahidol Univ. M.Sc.(Medical Epidemiology) / 13 Heterogeneity of treatment effects were assessed using Q test and I 2 statistic. If there was no heterogeneity (p-value of Q test greater than 0.1 and I 2 statistic smaller than 25%), a fixed effect model with inverse variance method were used to pool RR or unstandardized mean difference. If there was heterogeneity, a random effect model with Der Simonian and Laird method was applied to pool RR or unstandardized mean difference. The sources of heterogeneity were explored by applying subgroup analysis according to the study characteristics i.e. number of receiving previous treatments ( 2 treatments vs 3 treatments), percentage of splenectomy at baseline, different cut-off for platelet response, mean platelet count, dosage of treatments, or risk of bias. A characteristic was considered as a source of heterogeneity if the I 2 of the subgroup was decreased. Publication bias was assessed using Egger s test and funnel plot. Publication bias presented, if p-value from Egger test was less than 0.05, or funnel plot showed asymmetry. The asymmetry might be due to selection bias from some studies, poor quality of research methods, heterogeneity of treatment effects or publication bias itself. A contour enhanced funnel plot was applied to distinguish whether the cause of asymmetry was due to publication bias or heterogeneity. The funnel plot was contoured in significant and non-significant areas (e.g., < 0.01, < 0.05, and 0.05). If the funnel plot presented missing studies in non-significant areas, asymmetry might be due to publication bias. If the funnel plot presented missing studies in both significant and nonsignificant areas, the cause might be more likely due to other factors than publication bias Network meta-analysis Network meta-analysis was applied to assess treatment effects between different second line drugs. Treatments were coded as 1, 2, 3, 4, 5, 6, 7, 8, and 9 for placebo, eltrombopag, romiplostim, rituximab, danazol, rhtpo, rhtpo+danazol, rhtpo+cyclosporin, and rhtpo+rituximab, respectively; in which was used as the comparator. Indirect comparisons between active treatments were performed by borrowing information from a common comparator (i.e. placebo). A network of all treatments was mapped in which nodes and edges were weighted by numbers of subjects

26 Teeraya Puavilai Methodology / 14 for that comparison. A contribution plot was developed to display the contribution of each direct comparison to the network meta-analysis estimation. Two stage network meta-analysis was applied to estimate treatment effect of all second line treatments using placebo as common comparator. For dichotomous outcomes of platelet response, severe bleeding and composite serious adverse events, a binary regression was applied to estimate ln(rr) and its variance-covariance of each study. For continuous outcomes of platelet count, a linear regression was applied to estimate mean difference for each individual study. A multivariate random effect metaanalysis was applied to pool relative treatment effects (i.e., ln(rr) and mean difference) across the studies. Treatment effects of each treatment were contrasted by comparing relative treatment effects (i.e., the ln(rr) or mean difference) of each treatment with another. Treatments were ranked according to the probability of being the best treatment with highest RR for platelet response, highest mean difference for platelet count and lowest RR for composite serious adverse events, using rankogram and surface under the cumulative ranking curve (SUCRA). Treatment effect in the future of each treatment regimen was estimated by predictive interval. Consistency assumption for network meta-analysis was assessed by applying the design by treatment interaction model. Inconsistency was suggested when p-value is greater than Loop specific approach was used to identify the treatment arms and studies which contributed to inconsistency. The close loops with the highest inconsistency factors were inspected for study characteristics i.e. number of previous treatments ( 2 treatments vs 3 treatments), percentage of splenectomy at baseline, different cut-off for platelet response, mean platelet count, dosage of treatments, or risk of bias. Sensitivity analysis was performed by excluding the studies with a different level of each characteristic at a time. After each exclusion, design-by-treatment interaction model was used to recheck the inconsistency assumption. The excluded studies were considered source of inconsistency if the exclusion resulted in p-value >0.05. After excluding those studies, network meta-analysis for such outcome was performed again from the first step.

27 Fac. of Grad Studies, Mahidol Univ. M.Sc.(Medical Epidemiology) / 15 Publication bias for network meta-analysis was assessed by comparisonadjusted funnel plot. If signal of asymmetry was present in the plot, sensitivity analysis was performed by excluding the studies with sample size in the first and forth quartiles (i.e., smallest and largest sample sizes, respectively), then re-examining the comparisonadjusted funnel plot. Finally, clustered ranking plot for 2 outcomes was applied to platelet response with severe bleeding, platelet response with composite serious adverse events, platelet count with severe bleeding, platelet count with composite serious adverse events, and severe bleeding with composite serious adverse event according to their SUCRA values to demonstrate the ranks of treatments in the dimensions of benefit and risk consideration. All analyses were performed using STATA version A two-sided P value less than 0.05 was considered statistically significant for all analyses except for heterogeneity test, in which a one-sided P value less than 0.1 was applied instead.

28 Teeraya Puavilai Methodology / 16 Table 2.1 Search results by MEDLINE Item Domains Terms for PUBMED Numbers of studies

1

2

3

4

5

6

7

8

9 P chronic *immune thrombocytopen* chronic idiopathic thrombocytopen* chronic ITP

1 OR

2 OR

3 refractory *immune thrombocytopen* refractory idiopathic thrombocytopen* refractory ITP

5 OR

6 OR

7

4 OR

8 1,690 2, , ,096 3,628

10

11

12

13

14

15

16

17

18

19

20

21 I&C splenectomy vincristine vinblastine danazol azathioprine cyclophosphamide cyclosporin* mycophenolate mofetil OR MMF eltrombopag romiplostim rituximab

10 OR

11 OR

12OR

13 OR

14 OR

15 OR

16 OR

17 OR

18 OR

19 OR

20 27,200 28,179 15,304 2,926 20,654 65,820 53,807 10, , ,871

29 Fac. of Grad Studies, Mahidol Univ. M.Sc.(Medical Epidemiology) / 17 Table 2.1 Search results by MEDLINE (cont.) Item Domains Terms for PUBMED Numbers of studies

22

23

24

25

26

27 O bleeding platelet count thrombopoietin* death OR mortality thrombosis OR VTE OR venous thromboembolism 426,015 31,145 4,395 1,459, ,632 1,981,560

22 OR

23 OR

24 OR

25 OR

26

28

29 PICO

9 AND

21 AND

27

9 AND

21 AND

27 Filters: Clinical Trial

30 Teeraya Puavilai Methodology / 18 Table 2.2 Search results by SCOPUS Item Domains Terms for SCOPUS Numbers of studies

1

2

3

4

5

6

7

8

9

10

11 P chronic immune thrombocytopenia chronic immune thrombocytopenic purpura chronic idiopathic thrombocytopenia chronic idiopathic thrombocytopenic purpura

1 OR

2 OR

3 OR

4 refractory immune thrombocytopenia refractory immune thrombocytopenic purpura refractory idiopathic thrombocytopenia refractory idiopathic thrombocytopenic purpura

6 OR

7 OR

8 OR

9

5 OR

10 1,059 2, ,227 5, ,713 6,538

12

13

14

15

16

17

18

19

20

21

22

23

24

25 I&C splenectomy vincristine vinblastine danazol azathioprine cyclophosphamide cyclosporin A OR cyclosporine mycophenolate mofetil OR MMF eltrombopag romiplostim rituximab

12 OR

13 OR

14 OR

15 OR

16 OR

17 OR

18

19 OR

20 OR

21 OR

22

23 OR

24 52, ,290 63,591 14,161 97, , ,642 38,584 1,635 1,309 67, , , ,916

31 Fac. of Grad Studies, Mahidol Univ. M.Sc.(Medical Epidemiology) / 19 Table 2.2 Search results by SCOPUS (cont.) Item Domains Terms for SCOPUS Numbers of studies

26

27

28

29

30

31 O bleeding platelet count thrombopoietin death OR mortality thrombosis OR VTE OR venous thromboembolism 446,893 53,998 15,946 4,027, ,287 4,845,986

26 OR

27 OR

28 OR

29 OR

30

32

33

34 PICO

11 AND

25 AND

31 clinical trial

32 AND

33 3,059 1,944,

32 Teeraya Puavilai Results / 20 CHAPTER III RESULTS 3.1 Study selection A total of 141 and 997 studies were identified from MEDLINE via PubMed and SCOPUS. After combining both databases, 122 duplicates were removed, and 1,016 studies were screened on titles and abstracts. The reasons for ineligibility were provided in Figure 3.1., including 371 non-relevant study populations, 193 review studies/recommendations, 117 non-adult, 101 not interested outcomes, 86 no comparison group, 59 case reports/case series/ letters to editor, 33 non-second line treatments, 19 non-english, 18 not interested designs, 4 animal studies, and 3 systematic reviews and meta-analyses. Finally, 12 studies 38-40, were included for analysis. However, the study by Arnold et al 49 had insufficient data and the author did not provide additional data after our requests. 3.2 Characteristics of included studies The characteristics of these 12 studies were described in Table 3.1. All studies were RCTs with mainly multi-center settings, except one study 48 was single center setting. Seven 38-40, 46, 47, 49, 53 and 5 48, studies were conducted in Western countries and Asian countries, respectively. The sample sizes of the included RCTs 38, 40, 47, ranged from 23 to 196. All studies were 2 arm comparisons, including 4 studies 50 for eltrombopag versus placebo, 3 studies 39, 46, 48 for romiplostim versus placebo, 2 studies 49, 53 for rituximab versus placebo, 1 study 52 for rhtpo+cyclosporin versus rhtpo, 1 study 54 for rhtpo+rituximab versus rituximab, and 1 study 51 for rhtpo+danazol versus danazol. Median age ranged from 34 to 59 years old. Female was mainly in all studies and ranged from 56% to 73%. Median platelet count at baseline ranged from 10 x10 9 /L to 29 x10 9 /L. Platelet response was defined as platelet 50 x10 9 /L in 9 studies 38-40, 46-51, and 30 x10 9 /L in 3 studies Follow up period ranged from 2

33 Fac. of Grad. Studies, Mahidol Univ. M.Sc.(Medical Epidemiology) / 21 to 78 weeks, in which 11 studies 38-40, reported platelet response at 6 weeks. Only one study 39 of romiplostim versus placebo reported baseline thrombopoietin level. History of receiving treatments 3 regimens or more was reported in 4 studies 38-40, 47. Corticosteroids were the most common previous treatment followed by intravenous immunoglobulin. Eight studies 38-40, 47, 48, 50, 51, 54 reported the percentage of splenectomized patients which ranged from 10.4 to 69.6 at the baseline. Median time from splenectomy to enrollment was reported in only 2 studies 39, 47, which was 6.6 and 8.1 years, respectively. Eight studies 38-40, 47, 48, 50, 51, 54 reported the percentage of concurrent treatments which ranged from 11% to 83%. Treatment protocols were demonstrated (see Table 3.2) indicating variation of dosage and duration for each protocol. For eltrombopag versus placebo, dosage of eltrombopag for Caucasian was mg/day for 6 weeks 38, mg/day for 6 weeks 40 and 24 weeks 47, whereas dosage of eltrombopag for Asian was mg/day for 24 weeks 50. For romiplostim versus placebo, dosage of romiplostim for patients was 1-2 µg/kg SC once a week for 24 weeks 39, 3-10 µg/kg SC once a week for 52 weeks 46 and 12 weeks 48. For rituximab versus placebo, dosage of rituximab for patients in both studies 11, 49 was 375 mg/m 2 IV once a week for 4 weeks. For rhtpo, dosage of rhtpo was 1 µg/kg SC once daily for 2 weeks 51, 52, 54 k. 3.3 Risk of bias The quality of studies was assessed using a risk of bias assessment of Cochrane Collaboration s tool for randomized control trial (Appendix B). The results of risk of bias assessment were described in Table 3.3 and Figure 3.2. Random sequence generation were mainly considered as unclear risk in 7 studies 38, 39, 47, 48, because of insufficient information about the sequence generation process. Allocation concealment was mainly considered at low risk in 7 studies 39, 47-50, 53, 54. Blinding of participants and personnel, blinding of outcome assessment, and incomplete outcome data were mainly considered as low risk in 10 studies 38-40, 46-50, 53, 54. All studies were estimated at low risk in domain of selective reporting. For other sources of bias, 7 of 12 studies 38-40, were

34 Teeraya Puavilai Results / 22 considered as unclear risk because sponsors participated in protocol development, data collection, data analysis, interpretation and writing the manuscript. 3.4 Pairwise meta-analysis Data for pairwise comparisons for 4 outcomes, (i.e., platelet response, platelet count, severe bleeding, and composite serious adverse events) are provided in Table 3.4 to Table 3.6, and Table 3.8. Direct meta-analysis was performed for each comparison and outcome where there were at least 2 studies as follows: Platelet response Two treatment pairs (i.e., eltrombopag vs placebo, and romiplostim vs placebo) were directly compared. There were 4 studies 38, 40, 47, 50 for eltrombopag versus placebo (n = 314 versus 136) and 3 studies 39, 46, 48 for romiplostim versus placebo (n = 262 versus 131), their data were described in Table 3.4. Six studies 38-40, 47, 50 reported the numbers of patients with splenectomy, but 2 of 6 studies 47, 50 did not reported the numbers of patients with platelet response. The percentage of patients with splenectomy at enrollment ranged from 37 to 51 for intervention arm and 24 to 56 for control arm. Relative treatment effects of eltrombopag vs placebo were not much different across studies 38, 40, 47, 50 (Chi-square = 2.01, degrees of freedom (df) = 3, P=0.570, I 2 = 0%) with the pooled RR of 3.21 (95%CI: 2.28, 4.52), see Figure 3.3. The chance of platelet response for eltrombopag group was about 3.21 times higher than placebo group. The relative treatment effects of romiplostim versus placebo were heterogeneous across studies 39, 46, 48 (Chi-square = 9.57, df=2, P=0.008, I 2 = 79.1%), see Figure 3.4. The pooled RR was 5.84 (95%CI: 1.62, 21.09). The chance of platelet response for romiplostim group was about 5.84 times higher than placebo group. Publication bias was estimated by funnel plot and Egger s test. The funnel plot for eltrombopag versus placebo showed a bit asymmetry (Figure 3.5), that corresponded to result from Egger s test with indicated evidence of correlation between effect size and variances (coefficient = 1.33, P=0.029). The funnel plot for romiplostim vs placebo was also asymmetrical (Figure 3.6) that was not relevant to Egger s