Bleeding assessment and bleeding severity in thrombocytopenic patients undergoing invasive procedures.

Abstract

Patients with a low platelet count (thrombocytopenia) have an increased risk of both spontaneous and post-procedural bleeding.1, 2 Platelet transfusions are therefore recommended in various guidelines,3-5 either when the platelet count drops below a certain threshold or prior to invasive procedures. The clinical studies forming the basis of these guidelines are known to be of low quality,3-6 essentially reducing the value of transfusion guidelines to the quality level of expert opinion. Most studies designed to assess the optimal platelet transfusion trigger frequently include a clinical assessment of bleeding as outcome measure. A review of studies evaluating platelet transfusion triggers in patients with leukemia reported a spontaneous bleeding incidence that varied between 12 and 66%.7 The authors concluded that this wide variance was more likely a reflection of different methods of bleeding assessment than an actual difference in the occurrence of bleeding. A recent review on coagulopathy prior to central venous catheter (CVC) placement by our group, also found a large variance in the incidence of bleeding.1 Several bleeding scales have been developed to help clinicians and researchers assess bleeding. The most widely used of these is the World Health Organization (WHO) bleeding scale,8 which was created to standardize toxicity reporting in cancer treatment. The Society of Interventional Radiology (SIR) has developed standards for reporting post-procedural complications that includes a bleeding scale.9 These bleeding scales are ordinal in nature. An ordinal scale assigns grades to bleeding of increasing severity, whereas a singular definition gives criteria of bleeding to which the answer is either yes or no. In principle, an ordinal bleeding scale renders more details on bleeding complications than a singular definition, provided it is clear enough to allow unambiguous usage. The WHO bleeding scale in particular, is hampered by subjectivity and while none of the frequently used bleeding scales have ever been formally tested for reproducibility,10 a study on adjudication of the WHO scale revealed high inter-observer variability.11 Another problem with designing adequate bleeding scales is their clinical relevance. Historically, many studies have used WHO Grade 2-4 bleeding complications as an outcome, while Grade 2 bleeding (“mild blood loss”) is widely regarded as clinically irrelevant. Nonetheless, researchers often include grade 2 bleeding in order to capture enough endpoints. The incidence of grade 2 bleeding usually outweighs the incidence of grade 3-4 bleeding. Therefore, while such studies pretend to report clinically relevant bleeding, they mostly report “mild blood loss”, in this case a surrogate outcome.12 In this systematic review, we expect to find different bleeding incidences depending on the assessment methods and bleeding definitions used, but also depending on the study design. Retrospective studies have been shown to be less accurate than prospective studies and heavily depend on chart review. Minor bleeding in particular is not regularly recorded in clinical practice, and may therefore be underreported.7, 13 The primary objective of our study was to systematically review the methods and definitions used to assess bleeding severity in clinical research on invasive procedures. The secondary objective was to investigate the role of the study design in the variability in bleeding incidence. We included clinical studies (randomized controlled trials [RCTs] and cohort studies), both prospective and retrospective, on the following invasive procedures: CVC placement, liver biopsy (LB), renal biopsy (RB), bone marrow biopsy (BMB), or lumbar puncture (LP). Included studies needed to have bleeding complications as their primary or secondary endpoint and had to include at least one thrombocytopenic (<150 × 109/L) patient. An overview of thrombocytopenia and coagulopathy in each included study can be found in Appendix 1. Animal studies and case reports or series were excluded. Additionally, we excluded studies that were unavailable in English or Dutch. We conducted a MEDLINE search in May 2019, for which we used the search strategy that was previously described by the AABB, for the development of platelet transfusion guidelines.3 The search was not limited in time. Two authors independently reviewed citations for eligibility (EvdW & FvB); if any disagreement occurred a third author adjudicated (BB). We manually checked platelet transfusion guidelines to identify missing articles.3-5 The complete MEDLINE search terms are described in Appendix 2. For RCTs, the Cochrane Collaboration tool for the assessment of the risk of bias was used.14 For observational studies, the Newcastle-Ottawa Scale was used.15 Overall study quality was assessed by the Grading of Recommendations Assessment, Development and Evaluation (GRADE) method.16 The quality assessment is provided in Appendix 3. Continuous data was described as mean (SD) if normally distributed or as median (IQR) if not normally distributed. Categorical data was described as number (%). Non-normally distributed data was analyzed with Mann–Whitney U-tests, confidence intervals of bleeding incidences were calculated with the Wilson method17 and all statistical analyses were performed using R-Studio (version 1.1.453). Our MEDLINE search yielded a total of 2692 articles (1190 BMB, 211 CVC insertion, 1247 LB & RB and 44 LP), and the manual search of transfusion guidelines yielded another 472 articles. After removal of duplicates 3018 articles were left, of which 30 met the predefined inclusion and exclusion criteria (Fig. 1). All studies were cohort studies, seven of which were prospective and 23 were retrospective. All studies had bleeding complications as their primary endpoint. There was reasonable variation in study types and populations studied (Table 1). Overall, 11 studies used an ordinal bleeding scale, 13 used a singular bleeding definition and 6 reported no bleeding definition at all. Of the 24 studies with a bleeding definition, five used an existing ordinal bleeding scale (2) or incorporated elements of an existing ordinal bleeding scale in their singular definition (3). Nineteen studies used a bleeding definition (ordinal scale or singular definition) of the researchersʼ own design (Table 2). When investigators designed their own ordinal scale, it was always a two-point scale (major and minor bleeding). The existing scales used in these studies included the SIR Technology Assessment Committee reporting standards9 and the National Cancer Instituteʼs Common Terminology Criteria for Adverse Events (CTCAE)18 (Table 3). A detailed overview of bleeding definitions for all included studies can be found in Table 4. A: no therapy, no consequence; B: requiring nominal therapy, no consequence, including overnight admission for observation; C: requiring therapy, minor hospitalization <48 hours; D: requiring major therapy, unplanned increase in level of care, prolonged hospitalization >48 hours; E: permanent adverse sequelae; F: death 1: mild symptoms not requiring invasive intervention; 2: mild symptoms requiring minimally invasive interventions or aspiration; 3: event indicating transfusion, radiological or surgical procedure; 4: life-threatening consequences necessitating major urgent intervention; 5: death US or CT verified bleeding requiring blood transfusions, angiographic embolizations or surgical interventions. The criteria used to define bleeding could be categorized into three distinct categories: symptoms, interventions, and laboratory results, which were all sometimes limited in time and/or size (Fig. 2). General symptoms included oozing, subcutaneous hematoma, and changes in hemodynamic function. Naturally, some symptoms differed between invasive procedures. Studies on CVC placement included hemothorax and mediastinal hematoma. Studies on LB included hemobilia, subcapsular liver bleeding, and hemoperitoneum. Studies on RB included (subcapsular) perirenal hematoma and hematuria. Studies on LP included spinal, subdural, subarachnoid, and epidural hematoma. The study on BMB did not include specific symptoms. Common interventional criteria included erythrocyte (RBC) transfusion, surgical and/or radiological intervention to stop bleeding, which, together, often determined major bleeding, if such a distinction was made. Others included need for vasopressor or fluid therapy, extension of hospital stay, placement of suture ligaments, compression bandage, or manual pressure. Studies on CVC placement also included catheter removal, while one of the RB studies explicitly included angiographic embolization as a rescue intervention. Laboratory results used to define bleeding were a decrease in either hemoglobin (Hb) or hematocrit (Ht). Some studies put size- or time-limitations on one or more of the prior criteria. Limitations in time were the most common, where the bleeding had to occur within a specified timeframe, varying between 24 hours and 3 months after the intervention. In other studies, symptoms and/or interventions needed a minimum duration, for instance manual compression for >15-20 minutes or oozing of >24 hours. Pertaining to size, one study defined major bleeding as hematomas increasing in size. In five studies there was no mention of routine clinical post-procedural care. Routine care included post-procedural imaging, laboratory and clinical examinations, (overnight) admission, or observation. In 14 of 30 studies at least some data on bleeding assessment were described, in varying details, including chart review without further details on the procedure. Only one study used blinded bleeding assessors, although no details on the blinding procedure were given. Only one study used multiple trained bleeding assessors with an independent arbitrator. No other studies used trained bleeding assessors and/or arbitrators. Although we restricted our study to five predefined invasive procedures, there was little overlap between studies, due to different subtypes of procedures and different study populations. We could identify 23 different combinations of patient populations and procedures, of which only five were represented by at least two studies. Bleeding incidences varied widely between groups (Table 5), but even within groups we found non-overlapping 95% confidence intervals (Fig. 3). A significant difference in median bleeding incidence was observed between prospective studies (12.2% [8.1%-23.0%]) and retrospective studies (0.8% [0.0%-4.3], p = 0.02). We performed a post-hoc analysis on the ratio of major bleeding/minor bleeding for 10 studies that reported separate major and minor bleeding incidences. The median ratio was 0.1 (0.06-0.14) in prospective studies (n = 2), meaning that for every major bleeding there were 10 minor bleeding episodes, and 0.4 (0.2-1.2) in retrospective studies (n = 8), meaning five minor bleeding episodes for every two major episodes. This difference was not significant at p = 0.5. In this study, we reviewed all studies on five frequently performed invasive procedures. We found a large variance in bleeding complications, even between studies assessing the same invasive procedure, mostly due to differences in the way clinical bleeding is assessed and defined, as suggested previously.10, 19 The large proportion (19/30) of studies using a bleeding definition of investigatorsʼ own design forms a major problem, the impact of which is illustrated in the following example: a LB complicated by subcapsular bleeding requiring embolization and causing a 1 g/dL drop in Hb. This would be classified as major bleeding in one study (Kitchin et al20), but would not even be classified as minor bleeding in another study (McVay et al21). This illustrates that the difference in bleeding definitions should be taken into account when interpreting these results. Moreover, in the six studies without bleeding definition it is impossible to interpret the results. Five studies fully or partly used an existing bleeding scale, which seems to increase the validity of these studies. However, even these scales suffer from subjective criteria and have never been tested for inter-observer variability. One of these bleeding scales was used in a different context than its intended use. The CTCAE scale was designed for toxicity reporting in cancer patients and it is therefore questionable to apply it in patients undergoing an invasive procedure. Moreover, the CTCAE scale has no predefined cut-off between minor and major bleeding. Since researchers mostly report minor and major bleeding as separate entities, a clear distinction is needed. Besides the two bleeding scales encountered in this review, many other bleeding scales have been published previously. Koreth et al22 have already analyzed the majority of these scales, all of which are used in settings other than invasive procedures. Interestingly, the HEME bleeding assessment by Arnold et al,23 which was specifically designed for critically ill patients, uses some objective criteria, like hemodynamic measures and specific bleeding sites, but retains subjectivity in defining major bleeding as bleeding requiring major therapeutic intervention. Another limitation of these interventional bleeding scales is the difference in the use of therapeutic interventions according to local clinical practice, as reported by Koreth et al.22 Methods of bleeding assessment varied also. Fourteen out of 30 reported their methods, which were mostly based on review of medical records, resulting in less accurate results than prospectively gathered data.13 The amount of studies mentioning bleeding assessors was especially low (2/30), and none scored full marks with multiple trained, blinded bleeding assessors using independent adjudication. A systematic review on blinded versus non-blinded outcome assessors in RCTs showed that subjective binary endpoints suffer from bias when non-blinded assessors are used.24 Furthermore, disagreement between two independent adjudicators using the WHO bleeding scale was as high as 31.2%.11 The necessity of adjudicating results has not been demonstrated in all situations. For instance, multicenter research seems to have more benefit than single center research, and vague, subjective endpoints need more adjudication than well-defined, objective endpoints.25-28 Not all measures allow for adjudication: a trial on thromboprophylaxis in intensive care patients showed that attribution of bleeding to anticoagulant use was too hard for an arbitrating committee, when so many different causes of bleeding co-existed.29 Chart review is the predominant assessment method in retrospective studies. Our results show a significantly lower reported bleeding incidence in retrospective studies compared to prospective studies. This difference could be explained by the fact that in retrospective studies subtle positive outcomes (i.e., minor bleedings) are missed easily, since the assessment and documentation of minor bleeding is often not performed properly in general clinical practice.7, 13 The higher proportion of major bleeding that we found in retrospective studies further underlines this mechanism. However, due to the small number of prospective studies reporting minor and major bleeding, we were unable to demonstrate a statistically significant difference. Our study is limited by heterogeneity of included studies (including the rate of thrombocytopenic patients), which is due to the broad range of patient populations undergoing different invasive procedures (as addressed in Fig. 3). Although this is a well-known limitation in transfusion medicine research, current guidelines completely rely on these studies, so including them in this review is absolutely relevant. Our results support the hypothesis that reported bleeding incidence depends more on methods of assessment and bleeding definition than on actual bleeding tendency. This is in line with earlier results concerning both SAE reporting and clinical bleeding.1, 7, 30 Also, we have shown that the way of reporting bleeding assessment is often limited. The lack of this essential information reduces the validity and hampers the reproducibility of these studies. A major concern is that these studies form the basis of both current clinical guidelines and sample size calculations for future studies. Clinicians and researchers should be aware of the importance of outcome assessment and bleeding definition. Future research should focus on developing such a uniform, objective, and practical bleeding definition. Through detailing current practices and common criteria in bleeding definitions, the results of this study could form the basis of such a uniform definition. We suggest a definition that is specific to each intervention, proposed by specialists in each field, and perhaps with the help of patient-advocates.12 A specific definition could entail specific symptoms without relying on interventions or on subjective words like “significant morbidity” and “minimal intervention.” We demonstrate a high variability in definition and assessment of bleeding complications in studies on interventions in patients with thrombocytopenia. Hereby, interpretation and comparison of different study results is hampered. This has consequences for clinical practice (uncertainty about transfusion thresholds in guideline development) and clinical research (imprecise sample-size calculations and hampered comparison of studies). There is a dire need of a consensus procedure-related bleeding definition in the field of transfusion medicine, in patients undergoing invasive procedures. The authors declare that they have no conflicts of interest relevant to the manuscript submitted to Transfusion. This study is funded by ZonMW (Zorgonderzoek Medische Wetenschappen; part of the NWO [Nederlandse Organisatie voor Wetenschappelijk Onderzoek; the Dutch Organization for Scientific Research], Den Haag, The Netherlands), project number 843002625. The sponsors of this work were not involved in the study design, the collection, analysis, and interpretation of data, the writing of the report, or the decision to submit the manuscript for publication. <20: N = 33; 20-50: N = 187; >50: N = 761 Isolated <20: N = 11; 20-50: N = 30; 50-100: N = 22 Isolated 1.2-1.5 × ULoN: N = 12 >1.5 × ULoN: N = 6 Isolated 1.2-1.5 × ULoN: N = 4; >1.5 × ULoN: N=3 Combined PT & aPTT >1.5 × ULoN: N=3; PLT & coagulation abnormal: N = 13 Median (IQR; range): Subclavian (n = 352) 81 (51-133; 9-1088) Internal Jugular (n = 306) 83 (53-133; 10-425) Median (IQR; range): Subclavian (n = 352) 2.4 (1.7-3.9; 1-16) Internal Jugular (n = 306) 2,7 (1.8-4.7; 1-17) <80 (n = 122) Mean (range): 47 (8-79) <40% (n = 122) Mean (range): 29% (39%-10%) >77 (n = 3) Mean (range): 92 (78-100) Normal coagulation: N = 57; 1) abnormal parameter: N = 160; 2) abnormal parameters: N = 40 3) abnormal parameters: N = 2 In 88 coagulopathic patients: Median (range): 95 (12-330) In 88 coagulopathic patients: Median (range): 1.8 (1.2-3.5) In 88 coagulopathic patients: Median (range): 54s (22-100) Mean: 48; <20: N = 14; 20-29: N = 48; 30-39: N = 56; 40-49: N = 52; 50-99: N = 140; >100: N = 272 Median (range): Overall (n = 57) 26 (3-128) Transfused (n = 14) 50 (11-100) Not transfused (n = 43) 25 (3-128) <20: N = 16; 20-50: N = 55; 50-75: N = 100; 75-100: N = 146 >3,0: N = 97; 2.0-3.0: N = 139; 1.5-2.0: N = 239; 1.3-1.5: N = 293 >50: N = 17; 35-50: N = 55 1) abnormal parameters: N = 732; 2) abnormal parameters: N = 187; 3) abnormal parameters: N = 17 Isolated 3-19: N = 14; 20-24: N = 26; 25-29: N = 45; 30-34: N = 54; 34-39: N = 65; 40-44: N = 49; 45-49: N = 47 Isolated 1.5-1.6: N = 151; 1.7-1.8: N = 67; 1.9-2.0: N = 34; 2.1-2.2: N = 20; 2.3-3.8: N = 10 Isolated <50: N = 12 Isolated <50%: N = 32 Combined PLT < 50 & PT < 50%: N = 7 <50: N = 25 Mean (range): 251 (7-834) <50: N = 2; 50-99: N = 18; ≥100: N = 157 13.6-15.7: N = 11; 11.6-13.5: N = 65; <11.5: N = 100 43.6: N = 14; 38.0-43.5: N = 23; 34.1-37.9: N = 37; <34: N = 103 30-60: N = 13; 60-90: N = 16; 90-120: N = 21; 120-150: N = 13; 150-180: N = 6; >180: N = 18 Mean: 205 In 7 transfused patients Range: 35-96 Mean: 1.17 In 11 transfused patients Range: 1.25-1.79 Isolated <50: N = 36 Mean(range): 39.5 (18-49) Isolated <50%: N =3 0 Mean(range): 44.3% 28%-49%) PLT > 50 & PT < 50%: N = 19 Mean (range) PLT: 39.2 (22-49) Mean (range) PT: 42.6% (29%-49%) <50: N = 21 50-100: N = 110 >100: N = 1715 Mean (range): 219 (24-751) >1.5: N = 40 1.0-1.5: N=755 <1.0: N=1051 Mean (range): 1.08 (0.8-2.7) In 27 patients with coagulopathy Mean (range): 53 (19-153) In 27 patients with coagulopathy Mean (range): 16.3s (11.4-20.3) BMT group (n = 183) Mean (sd; range): 88 (71; 5-336) Non-BMT group (n = 1417) Mean (sd; range): 174 (107; 8-1507) BMT group (n = 183) Mean (sd): 1.2 (0.5) Non-BMT group (n=1417) Mean (sd): 1.2 (0.4) Mean (sd; range): 260 (85; 107-442) Mean (sd; range): 11.1s (1.2; 9.3-13.4) Mean (sd; range): 26.5 (3.2; 21.7-37.1) Amyloidosis group (n = 101) Median (range): 282 (54-824) Control group (n = 188) Median (range): 265 (35-844) Amyloidosis group (n = 101) Median (range): 0.9 (0.8-1.4) Control group (n = 188) Median (range): 0.9 (0.8-1.4) Amyloidosis group (n = 101) Median (range): 26s (17-54) Control group (n = 188) Median (range): 26s (20-47) Mean: 248 <100: N = 6 (range: 75-94); 100-150: N = at least 5 Median (IQR): 226 (184-273) Median (IQR): 10.1s (9.6s-10.7s) Median (IQR): 31.3s (28.7-33.8) Median (IQR): 236 (182-297); <100: N=97; ≥100: N=1881 1-25: N = 40; 26-75: N = 236; 76-99: N = 111; ≥100: N = 1321 <10: N = 25; 10-20: N = 67; 20-30: N = 88; 30-40: N = 92; 40-50: N = 107; 50-100: N = 729;>100: N = 7980 Mean (sd; range): Bleeding group (n = 223) 12 (0.7; 9.8-13) Non-bleeding group (n = 777) 12.0 (1.1; 8.9-15.5) Mean (sd; range): Bleeding group (n = 223) 29 (2.9; 22-37) Non-bleeding group (n = 777) 31 (8.4; 22-79) 11-20: N = 3; 21-50: N = 17; 51-100: N = 40; 101-150: N = 52; >150: N = 242