Assessing protocol adherence in a clinical trial with ordered treatment regimens: Quantifying the pragmatic, randomized optimal platelet and plasma ratios (PROPPR) trial experience

  • Hongjian Zhu
    Correspondence
    Corresponding author at: Department of Biostatistics, University of Texas Health Science Center at Houston (UTHEALTH) School of Public Health, 1200 Pressler St. RAS W922, Houston, TX 77030, USA.
    Affiliations
    Department of Biostatistics, University of Texas Health Science Center at Houston School of Public Health, Houston, TX, USA
    Search for articles by this author
  • Erin E. Fox
    Affiliations
    Center for Translational Injury Research and Division of Acute Care Surgery, Department of Surgery, McGovern Medical School, University of Texas Health Science Center at Houston, Houston, TX, USA
    Search for articles by this author
  • Sarah Baraniuk
    Affiliations
    Department of Biostatistics, University of Texas Health Science Center at Houston School of Public Health, Houston, TX, USA
    Search for articles by this author
  • John B. Holcomb
    Affiliations
    Center for Translational Injury Research and Division of Acute Care Surgery, Department of Surgery, McGovern Medical School, University of Texas Health Science Center at Houston, Houston, TX, USA
    Search for articles by this author
  • Charles E. Wade
    Affiliations
    Center for Translational Injury Research and Division of Acute Care Surgery, Department of Surgery, McGovern Medical School, University of Texas Health Science Center at Houston, Houston, TX, USA
    Search for articles by this author
  • Deborah J. Del Junco
    Affiliations
    Center for Translational Injury Research and Division of Acute Care Surgery, Department of Surgery, McGovern Medical School, University of Texas Health Science Center at Houston, Houston, TX, USA
    Search for articles by this author
  • Barbara C. Tilley
    Affiliations
    Department of Biostatistics, University of Texas Health Science Center at Houston School of Public Health, Houston, TX, USA
    Search for articles by this author
  • on behalf ofthe PROPPR Study Group
    Author Footnotes
    1 Collaborators: Clinical Coordinating Center: John B. Holcomb, MD; Charles E. Wade, PhD; Deborah J. del Junco, PhD; Erin E. Fox, PhD; Nena Matijevic, PhD; Jeanette Podbielski, RN; Angela M. Beeler, BS. Data Coordinating Center: Barbara C. Tilley, PhD; Sarah Baraniuk, PhD; Joshua Nixon, MS; Roann Seay, MS; Savitri N. Appana, MS; Hui Yang, MS; Michael O. Gonzalez, MS. Core Laboratory: Lisa Baer, MS; Yao-Wei Willa Wang, MD; Brittany S. Hula, MS; Elena Espino, BS; An Nguyen, BS; Nicholas Pawelczyk, BS; Kisha D. Arora-Nutall, BS; Rishika Sharma, MD; Jessica C. Cardenas, PhD; Elaheh Rahbar, PhD; Tyrone Burnett, Jr., BS; David Clark, BS. Resuscitation Outcomes Consortium: Gerald van Belle, PhD; Susanne May, PhD; Brian Leroux, PhD; David Hoyt, MD; Judy Powell, BSN, RN; Kellie Sheehan, BSN. Systems Biology Committee: Alan Hubbard, PhD; Adam P. Arkin, PhD. Transfusion Committee: John R. Hess, MD; Jeanne Callum, MD. PROPPR Clinical Sites (listed in order of number of patients enrolled): University of Texas Health Science Center at Houston: Bryan A. Cotton, MD, MPH; Laura Vincent, BSN, RN, CCRP; Timothy Welch; Tiffany Poole, DC; Evan G. Pivalizza, MD; Sam D. Gumbert, MD; Yu Bai, MD, PhD; James J. McCarthy, MD; Amy Noland, MD; Rhonda Hobbs, MT(ASCP)SBB. University of Washington: Eileen M. Bulger, MD; Patricia Klotz, RN; Lindsay Cattin, BA; Keir J. Warner, BS; Angela Wilson, BA; David Boman, BA; Nathan White, MD, MS; Andreas Grabinsky, MD; Jennifer A. Daniel-Johnson, MBBS. University of California, San Francisco: Mitchell Jay Cohen, MD; Rachael A. Callcut, MD, MSPH; Mary Nelson, RN, MPA; Brittney Redick, BA; Amanda Conroy, BA; Marc P. Steurer, MD, DESA; Preston C. Maxim, MD; Eberhard Fiebig, MD; Joanne Moore; Eireen Mallari, MT. University of Cincinnati: Peter Muskat, MD; Jay A. Johannigman, MD; Bryce R. H. Robinson, MD; Richard D. Branson, MSc, RRT; Dina Gomaa, BS, RRT; Christopher Barczak, BS, MT(ASCP); Suzanne Bennett, MD; Patricia M. Carey, MD; Christopher N. Miller, MD; Helen Hancock, BS, MT(ASCP); Carolina Rodriguez, BA. University of Southern California: Kenji Inaba, MD; Jay G. Zhu, MD; Monica D. Wong, MS; Michael Menchine, MD, MPH; Kelly Katzberg, MD, FACEP; Sean O. Henderson, MD; Rodney McKeever, MD; Ira A. Shulman, MD; Janice M. Nelson, MD; Christopher W. Tuma, BA, MT(ASCP), SBB; Cheryl Y. Matsushita, BS, MT(ASCP). Shock, Trauma and Anesthesiology Research – Organized Research Center (STAR-ORC), R Adams Cowley Shock Trauma Center, University of Maryland Medical Center: Thomas M. Scalea, MD; Deborah M. Stein, MD, MPH; Cynthia K. Shaffer, MS, MBA; Christine Wade, BA; Anthony V. Herrera, MS; Seeta Kallam, MBBS; Sarah E. Wade, BS; Samuel M. Galvagno, Jr, DO, PhD; Magali J. Fontaine, MD, PhD; Janice M. Hunt, BS, MT(ASCP) SBB; Rhonda K. Cooke, MD. University of Tennessee Health Science Center, Memphis: Timothy C. Fabian, MD; Jordan A. Weinberg, MD; Martin A. Croce, MD; Suzanne Wilson, RN; Stephanie Panzer-Baggett, RN; Lynda Waddle-Smith, BSN; Sherri Flax, MD. Medical College of Wisconsin: Karen J. Brasel, MD, MPH; Pamela Walsh, AS, CCRC; David Milia, MD; Allia Nelson, BS, BA; Olga Kaslow, MD, PhD; Tom P. Aufderheide, MD, MS; Jerome L. Gottschall, MD; Erica Carpenter, MLS(ASCP). University of Arizona: Terence O’Keeffe, MBChB, MSPH; Laurel L. Rokowski, RN, BSN, MKT; Kurt R. Denninghoff, MD; Daniel T. Redford, MD; Deborah J. Novak, MD; Susan Knoll, MS, MT(ASCP) SBB. University of Alabama at Birmingham: Jeffrey D. Kerby, MD, PhD; Jean-Francois Pittet, MD (Anesthesia Chair); Patrick L. Bosarge, MD; Albert T. Pierce, MD; Carolyn R. Williams, RN, BSN, BSME; Shannon W. Stephens, EMTP; Henry E. Wang, MD, MS; Marisa B. Marques, MD. Oregon Health and Science University: Martin A. Schreiber, MD; Jennifer M. Watters, MD; Samantha J. Underwood, MS; Tahnee Groat, MPH; Craig Newgard, MD, MPH; Matthias Merkel, MD, PhD; Richard M. Scanlan, MD; Beth Miller, MT(ASCP)SBB. Sunnybrook Health Sciences Centre: Sandro Rizoli, MD, PhD; Homer Tien, MD; Barto Nascimento, MD, MSc, CTBS; Sandy Trpcic; Skeeta Sobrian-Couroux, RN, CCRP, BHA; Marciano Reis; Adic Pérez, MD; Susan E. Belo, MD, PhD; Lisa Merkley, BA, MLT, CBTS; Connie Colavecchia, BSc, MLT.
  • Author Footnotes
    1 Collaborators: Clinical Coordinating Center: John B. Holcomb, MD; Charles E. Wade, PhD; Deborah J. del Junco, PhD; Erin E. Fox, PhD; Nena Matijevic, PhD; Jeanette Podbielski, RN; Angela M. Beeler, BS. Data Coordinating Center: Barbara C. Tilley, PhD; Sarah Baraniuk, PhD; Joshua Nixon, MS; Roann Seay, MS; Savitri N. Appana, MS; Hui Yang, MS; Michael O. Gonzalez, MS. Core Laboratory: Lisa Baer, MS; Yao-Wei Willa Wang, MD; Brittany S. Hula, MS; Elena Espino, BS; An Nguyen, BS; Nicholas Pawelczyk, BS; Kisha D. Arora-Nutall, BS; Rishika Sharma, MD; Jessica C. Cardenas, PhD; Elaheh Rahbar, PhD; Tyrone Burnett, Jr., BS; David Clark, BS. Resuscitation Outcomes Consortium: Gerald van Belle, PhD; Susanne May, PhD; Brian Leroux, PhD; David Hoyt, MD; Judy Powell, BSN, RN; Kellie Sheehan, BSN. Systems Biology Committee: Alan Hubbard, PhD; Adam P. Arkin, PhD. Transfusion Committee: John R. Hess, MD; Jeanne Callum, MD. PROPPR Clinical Sites (listed in order of number of patients enrolled): University of Texas Health Science Center at Houston: Bryan A. Cotton, MD, MPH; Laura Vincent, BSN, RN, CCRP; Timothy Welch; Tiffany Poole, DC; Evan G. Pivalizza, MD; Sam D. Gumbert, MD; Yu Bai, MD, PhD; James J. McCarthy, MD; Amy Noland, MD; Rhonda Hobbs, MT(ASCP)SBB. University of Washington: Eileen M. Bulger, MD; Patricia Klotz, RN; Lindsay Cattin, BA; Keir J. Warner, BS; Angela Wilson, BA; David Boman, BA; Nathan White, MD, MS; Andreas Grabinsky, MD; Jennifer A. Daniel-Johnson, MBBS. University of California, San Francisco: Mitchell Jay Cohen, MD; Rachael A. Callcut, MD, MSPH; Mary Nelson, RN, MPA; Brittney Redick, BA; Amanda Conroy, BA; Marc P. Steurer, MD, DESA; Preston C. Maxim, MD; Eberhard Fiebig, MD; Joanne Moore; Eireen Mallari, MT. University of Cincinnati: Peter Muskat, MD; Jay A. Johannigman, MD; Bryce R. H. Robinson, MD; Richard D. Branson, MSc, RRT; Dina Gomaa, BS, RRT; Christopher Barczak, BS, MT(ASCP); Suzanne Bennett, MD; Patricia M. Carey, MD; Christopher N. Miller, MD; Helen Hancock, BS, MT(ASCP); Carolina Rodriguez, BA. University of Southern California: Kenji Inaba, MD; Jay G. Zhu, MD; Monica D. Wong, MS; Michael Menchine, MD, MPH; Kelly Katzberg, MD, FACEP; Sean O. Henderson, MD; Rodney McKeever, MD; Ira A. Shulman, MD; Janice M. Nelson, MD; Christopher W. Tuma, BA, MT(ASCP), SBB; Cheryl Y. Matsushita, BS, MT(ASCP). Shock, Trauma and Anesthesiology Research – Organized Research Center (STAR-ORC), R Adams Cowley Shock Trauma Center, University of Maryland Medical Center: Thomas M. Scalea, MD; Deborah M. Stein, MD, MPH; Cynthia K. Shaffer, MS, MBA; Christine Wade, BA; Anthony V. Herrera, MS; Seeta Kallam, MBBS; Sarah E. Wade, BS; Samuel M. Galvagno, Jr, DO, PhD; Magali J. Fontaine, MD, PhD; Janice M. Hunt, BS, MT(ASCP) SBB; Rhonda K. Cooke, MD. University of Tennessee Health Science Center, Memphis: Timothy C. Fabian, MD; Jordan A. Weinberg, MD; Martin A. Croce, MD; Suzanne Wilson, RN; Stephanie Panzer-Baggett, RN; Lynda Waddle-Smith, BSN; Sherri Flax, MD. Medical College of Wisconsin: Karen J. Brasel, MD, MPH; Pamela Walsh, AS, CCRC; David Milia, MD; Allia Nelson, BS, BA; Olga Kaslow, MD, PhD; Tom P. Aufderheide, MD, MS; Jerome L. Gottschall, MD; Erica Carpenter, MLS(ASCP). University of Arizona: Terence O’Keeffe, MBChB, MSPH; Laurel L. Rokowski, RN, BSN, MKT; Kurt R. Denninghoff, MD; Daniel T. Redford, MD; Deborah J. Novak, MD; Susan Knoll, MS, MT(ASCP) SBB. University of Alabama at Birmingham: Jeffrey D. Kerby, MD, PhD; Jean-Francois Pittet, MD (Anesthesia Chair); Patrick L. Bosarge, MD; Albert T. Pierce, MD; Carolyn R. Williams, RN, BSN, BSME; Shannon W. Stephens, EMTP; Henry E. Wang, MD, MS; Marisa B. Marques, MD. Oregon Health and Science University: Martin A. Schreiber, MD; Jennifer M. Watters, MD; Samantha J. Underwood, MS; Tahnee Groat, MPH; Craig Newgard, MD, MPH; Matthias Merkel, MD, PhD; Richard M. Scanlan, MD; Beth Miller, MT(ASCP)SBB. Sunnybrook Health Sciences Centre: Sandro Rizoli, MD, PhD; Homer Tien, MD; Barto Nascimento, MD, MSc, CTBS; Sandy Trpcic; Skeeta Sobrian-Couroux, RN, CCRP, BHA; Marciano Reis; Adic Pérez, MD; Susan E. Belo, MD, PhD; Lisa Merkley, BA, MLT, CBTS; Connie Colavecchia, BSc, MLT.
Open AccessPublished:July 20, 2016DOI:https://doi.org/10.1016/j.injury.2016.07.028

      Abstract

      Background

      Medication dispensing errors are common in clinical trials, and have a significant impact on the quality and validity of a trial. Therefore, the definition, calculation and evaluation of such errors are important for supporting a trial’s conclusions. A variety of medication dispensing errors can occur. In this paper, we focus on errors in trials where the intervention includes multiple therapies that must be given in a pre-specified order that varies across treatment arms and varies in duration.

      Methods

      The Pragmatic, Randomized Optimal Platelet and Plasma Ratios (PROPPR) trial was a Phase III multi-site, randomized trial to compare the effectiveness and safety of 1:1:1 transfusion ratios of plasma and platelets to red blood cells with a 1:1:2 ratio. In this trial, these three types of blood products were to be transfused in a pre-defined order that differed by treatment arm. In this paper, we present approaches from the PROPPR trial that we used to define and calculate the occurrence of out of order blood transfusion errors. We applied the proposed method to calculate protocol adherence to the specified order of transfusion in each treatment arm.

      Results

      Using our proposed method, protocol adherence was greater in the 1:1:1 group than in the 1:1:2 group (96% vs 93%) (p < 0.0001), although out of order transfusion errors in both groups were low. Final transfusion ratios of plasma to platelets to red blood cells for the 1:1:1 ratio group was 0.93:1.32:1, while the transfusion ratio for the 1:1:2 ratio group was 0.48:0.48:1.

      Conclusions

      Overall, PROPPR adherence to blood transfusion order pre-specified in the protocol was high, and the required order of transfusions for the 1:1:2 group was more difficult to achieve. The approaches proposed in this manuscript were useful in evaluating the PROPPR adherence and are potentially useful for other trials where a specific treatment orders with varying durations must be maintained.

      Keywords

      Introduction

      Medication dispensing errors, generally defined as a deviation from the medication order as written in the protocol, are not uncommon in clinical trials. For example, the ARISTOTLE trial successfully proved the efficacy of apixaban in the reduction in stroke, major bleeding and death [
      • Alexander J.H.
      • Levy E.
      • Lawrence J.
      • Hanna M.
      • Waclawski A.P.
      • Wang J.
      • et al.
      Documentation of study medication dispensing in a prospective large randomized clinical trial: experiences from the ARISTOTLE Trial.
      ]. However, the approval of apixaban was delayed because of the reported medication dispensing errors including giving the wrong drug. In trials, it is important to evaluate the extent of adherence to the protocol in terms of medication dispensing. The most common medication dispensing errors reported in the literature are wrong drugs, strength, form or quantity, as well as wrong labels [
      • James K.L.
      • Barlow D.
      • McArtney R.
      • Hiom S.
      • Roberts D.
      • Whittlesea C.
      Incidence, type and causes of dispensing errors: a review of the literature.
      ]. Examples and detailed discussion about these issues can be found in the literature [
      • Spencer M.G.
      • Smith A.P.
      A multicentre study of dispensing errors in British hospitals.
      ,
      • Allan E.
      • Barker K.
      Fundamentals of medication error research.
      ].
      In this paper, we studied a special case of medication dispensing errors illustrated using the Pragmatic, Randomized Optimal Platelet and Plasma Ratios (PROPPR) trial, i.e. deviation of the order of transfusion of blood products from the order specified in the protocol [
      • Holcomb J.B.
      • Tilley B.C.
      • Baraniuk S.
      • Fox E.E.
      • Wade C.E.
      • Podbielski J.M.
      • et al.
      Transfusion of plasma, platelets, and red blood cells in a 1:1:1 vs a 1:1:2 ratio and mortality in patients with severe trauma: the PROPPR randomized clinical trial.
      ,
      • Baraniuk S.
      • Tilley B.C.
      • Del Junco D.J.
      • Fox E.E.
      • Van Belle G.
      • Wade C.E.
      • et al.
      Pragmatic randomized optimal platelet and plasma ratios (PROPPR) trial: design, rationale and implementation.
      ]. We propose a fundamental rationale and specific approaches to define and count the blood transfusion errors. We applied our approach to data from the PROPPR trial, and carried out a comprehensive evaluation to compute adherence to the order of treatment as specified in the protocol. The approaches we developed for quantifying adherence to blood product order can be applied to other clinical trials where order of interventions is complex and may vary over time.

      Materials and methods

      The PROPPR trial was a Phase III multi-site, randomized trial to compare the effectiveness and safety of a 1:1:1 transfusion ratio of plasma to platelets to red blood cells (RBCs) with a 1:1:2 ratio, where the 1:1:1 (1:1:2) ratio means that for each transfused unit of plasma, another one unit of platelets and one (two) units of RBCs are also transfused to the patient. According to the protocol, these three types of blood products were to be transfused to patients in a pre-specified order as shown in Fig. 1. Each dose of platelets in the figure represents a pool of 6 units on average. The blood products were delivered in containers which differed according to treatment assignment. Every container for the 1:1:1 group was the same, while two different containers were delivered sequentially to the patient in the 1:1:2 group. How the units of plasma were made rapidly available has been discussed [
      • Novak D.J.
      • Bai Y.
      • Cooke R.K.
      • Marques M.B.
      • Fontaine M.J.
      • Gottschall J.L.
      • et al.
      Making thawed universal donor plasma available rapidly for massively bleeding trauma patients: experience from the pragmatic, randomized optimal platelets and plasma ratios (PROPPR) trial.
      ], and other trial details can be found in previous publications [
      • Holcomb J.B.
      • Tilley B.C.
      • Baraniuk S.
      • Fox E.E.
      • Wade C.E.
      • Podbielski J.M.
      • et al.
      Transfusion of plasma, platelets, and red blood cells in a 1:1:1 vs a 1:1:2 ratio and mortality in patients with severe trauma: the PROPPR randomized clinical trial.
      ,
      • Baraniuk S.
      • Tilley B.C.
      • Del Junco D.J.
      • Fox E.E.
      • Van Belle G.
      • Wade C.E.
      • et al.
      Pragmatic randomized optimal platelet and plasma ratios (PROPPR) trial: design, rationale and implementation.
      ].
      Fig. 1
      Fig. 1Blood transfusion order in the Protocol of the PROPPR trial.
      Note: The left side is for the 1:1:1 group. The right side is for the 1:1:2 group. One dose of platelets represents a pool of six units and one dose of red blood cells or plasma represents one unit.
      The order of blood products transfused during the PROPPR protocol was decided by multiple factors. For example, Platelets could not be placed on ice in the container with the other blood units, so one dose of platelets was placed into a separate opaque package attached to the outside of the transport container. In order to conceal the contents of the study containers until after patients were randomized and enrolled and promote blinding, a “sham” platelet bag was attached to the odd numbered container in the 1:1:2 treatment group.

       Identification of out of order treatments

      In this paper, protocol adherence was measured by one minus the patient’s average blood transfusion error proportion, and each patient’s blood transfusion error proportion was calculated by dividing the total number of wrongly transfused blood products by the total number of transfused blood products. The fundamental rationale behind the definition of protocol adherence included the following considerations. First, the total number of errors was counted based on the number of blood product units given in the wrong sequence. Because we were measuring adherence to the protocol in terms of the order of the blood products transfused, and not assessing the effect of adherence on treatment outcome, we did not assign weights to different types of errors. Second, the intention was to only penalize an error once. Instead of counting every deviation from the correct order as an error, we determined what specific approaches had been used to fix an error and only penalized for the original error rather than the fix. Third, our goal was to be consistent across different cases and avoid ad hoc approaches that might not be relevant to other projects. Lastly, in this trial the transfusion protocol could be stopped in the middle of a container due to death or hemorrhage control, but the error rate prior to stopping was of interest. The specific methods of defining and calculating blood product transfusion errors are listed in Table 1, Table 2. Next, we will explain these rules which can be generalized to other situations.
      Table 1Error Criteria for the 1:1:1 group.
      Condition (1–9 for single unit, 10–15 for tie)Error
      1. First unit transfused from a container is NOT platelet, and there are no other platelet units transfused from that container.+1
      2. First unit transfused from a container is NOT platelet, and there exists other platelet units transfused from the container.+(number of units of platelet-1)
      3. First unit transfused from a container is platelet, and next unit is NOT RBC+1
      4. First unit transfused from a container is platelet, and there exists other units of platelet transfused from the same container+number of excess units of platelets
      5. RBC transfused which is followed by plasma+0
      6. Plasma transfused which is followed by RBC+0
      7. RBC transfused followed by RBC+1
      8. Plasma transfused followed by plasma+1
      9. Either RBC or plasma is transfused followed by a wrong unit of platelet.+0
      10. The tie is not at the beginning of a container, and there are no platelets in the tie.+Etie1
      11. The tie is not at the beginning of a container, there is at least one platelet in the tie, and platelet has been transfused from the container.+Etie1
      12. The tie is not at the beginning of a container, there is at least one platelet in the tie, and platelet has NOT been transfused from the container+Etie2
      13. The first transfusions from the container are a tie, and there are no platelets in the tie.+Etie1
      14. The first transfusions from the container are a tie, there is at least one platelet in the tie.Assume the first platelet has been given correctly, and the error for the rest are calculated based on rule 11 or rule 12.
      15. Either RBC or plasma is transfused, followed by a tie.+0
      Table 2Error Criteria for the 1:1:2 group.
      2a: Error Criteria for odd numbered container in the 1:1:2 group
      Condition (1–6 for single unit, 7–8 for tie)Error
      1. First unit transfused from a container is NOT RBC+1
      2. The second unit transfused from a container is NOT RBC+1
      3. The unit transfused following 2 or more RBCs is NOT plasma+1
      4. The unit transfused following plasma is NOT RBC+1
      5. Platelets were transfused from a container+number of units of platelet
      6. Either RBC or plasma were transfused following a wrong unit of platelet+0
      7. Any ‘tied’ transfusions from a container+Etie1
      8. Either RBC or plasma trasnfusion is followed by a tie.+0
      2b: Error Criteria for even numbered container in the 1:1:2 group
      Condition (1–7 for single unit, 8–13 for tie)Error
      1. First unit transfused from a container is NOT platelets+1
      2. First unit transfused from a container is NOT platelets, and other platelet units were transfused from the container+(number of units of platelet-1)
      3. First unit transfused from a container is platelets, but the next unit is NOT RBC+1
      4. First unit transfused from a container is platelets, and other units of platelet were transfused from the same container+number of excess units of platelets
      5. The unit transfused following 2 or more RBCs is NOT plasma+1
      6. The unit transfused following plasma is not RBC+1
      7. Either RBC or plasma were transfused following a wrong unit of platelet+0
      8. The tie is not at the beginning of a container, and there are no platelets in the tie.+Etie1
      9. The tie is not at the beginning of a container, there is at least one platelet in the tie, and platelet has been transfused in the container.+Etie1
      10. The tie is not at the beginning of a container, there is at least one platelet in the tie, and platelet has NOT been transfused in the container+Etie2
      11. The first units transfused from the container are a tie, and there are no platelets in the tie.+Etie1
      12. The first units transfused from the container are a tie, there is at least one platelet in the tie.Assume the first platelet has been given correctly, and the error for the rest are calculated based on rule 11 or rule 12.
      13. Either RBC or plasma was transfused, followed by a tie.+0
      The first challenge was how to define an error. One possibility was to compare the transfused blood products with the required order in the protocol one by one (unit-based approach). However, if plasma was the first transfused blood product of the first container in the 1:1:1 group instead of platelets, a nurse could fix the error in at least two ways. The nurse might follow the protocol from the second unit by transfusing RBCs and continuing in compliance with the protocol until the end of this container. The unit-based error-counting method defined previously works well in this case and returns one error. Alternately, a nurse might fix this error by immediately transfusing a platelet unit and then follow the protocol as if starting a new container. In this case, the unit-based strategy of defining errors would return 13 errors. Therefore, the unit-based error calculation approach is not a consistent strategy across multiple situations, so we considered an alternate approach.
      As in Fig. 1, the overall pattern of transfusion for the 1:1:1 group was alternating units of RBCs and plasma interspersed with platelets. For this reason, we can consider a pattern-based approach to error counting within each study container: the first unit was a platelet, every RBC should be followed by plasma, and every plasma unit should be followed by a RBC except for the thirteenth unit (plasma) in the container. For example, if two successive RBCs were given after the initial platelet to a patient in the 1:1:1 group, one error would be counted for the second RBC if the next unit transfused was plasma, and that plasma would be correct, since it followed an RBC. Similarly, the pattern of transfusions from odd-numbered containers for the 1:1:2 group was that every two RBCs should be followed by a plasma unit and every plasma unit should be followed by an RBC. Therefore, any RBC after one or more plasmas was correct, and any plasma after two or more RBCs was correct. For example, if the order of transfusion for the first three units in an odd numbered container for the 1:1:2 group was plasma, plasma, RBC, the RBC was deemed to be correct, even if the third unit in the protocol should be plasma. For all transfusions from even numbered containers for the 1:1:2 group, the pattern was to first transfuse a unit of platelets, then follow every two RBC units with plasma, and every plasma unit transfused should be followed by an RBC. The rationality of the above idea can be intuitively explained by similar activity in real life such as a marching forward of a soldier. The correct order is left leg followed by right leg and right leg followed by left leg. If a soldier took two successive left leg steps, the next correct one is always a right leg step. This idea has been proved to be consistently feasible for all situations in this project, and should work for other projects with similar problems, and complies with the third rationale mentioned above.
      The second major difficulty was how to deal with potential errors involving platelets. Ignoring platelets, RBCs and plasma were supposed to be transfused in a consistent pattern depending on treatment group. An out of order platelet can affect counting errors for this platelet as well as other blood products transfused after it. There were three general cases of out-of-order platelets in the 1:1:1 group and in the even-numbered container of the 1:1:2 group, both of which should have a platelet at the first position: (1) No platelets were transfused from the container at all; (2) The first unit transfused from the container was not platelet, but one or more doses of platelets were reported in another position; (3) The first unit transfused from a container was platelets, but additional doses of platelets were transfused from the same container. For case (1), the first position was counted as an error. We counted errors for case (2) based on the second rationale of the project and only penalized for the original error and not the fixing of the error. According to the protocol, there should be one platelet in the container. If the first one was not platelet, the nurse might try to fix the original error by transfusing one platelet in the middle of this container in order to attain the appropriate ratio. Therefore, this incorrectly positioned platelet used to fix the original error should not be penalized. But additional doses of platelets after the first transfused were counted as errors. In Case (3), all the excess units of platelets should be counted as errors. Finally, in the odd numbered container in the 1:1:2 group which did not contain any platelets based on the protocol, all platelets transfused were counted as errors.
      The third major difficulty followed from the latter error, i.e., how to define errors for the blood units transfused after an out of order platelet. This was the most complicated situation and no strategy was perfect for this case. Note that the out of order platelet has already been counted as an error (for example, rule 1 and 2 in Table 1). So we proposed that both RBC and plasma, which followed a wrong unit of platelet, could be deemed to be correct. Using the 1:1:1 group as an example, the following cases might occur: (i) RBC-platelet-plasma, (ii) RBC-platelet-RBC, (iii) plasma-platelet-plasma, (iv) plasma-platelet-RBC. A plasma should follow a RBC and a RBC should follow a plasma based on the protocol. Since we have counted the platelet as an error, the nurse might want to follow the protocol ignoring the wrong platelet in order not to produce more errors. As a result, both case (i) and (iv) were acceptable under the pattern-based approach. For case (ii), the nurse might be trying to follow the protocol by giving an RBC after the platelet, which was replacing a plasma unit. In this situation, all following units could be delivered following the original protocol and might be the nurse’s way of minimizing the number of errors. Another reasonable explanation for case (ii) was that, if the protocol was followed, the next unit after a platelet would be a unit of RBCs. Similar arguments can be given for case (iii) and for the 1:1:2 group.
      Other complications in measuring protocol adherence arose because some sites administered multiple units of blood products simultaneously, through either a rapid infuser or more than one patient blood vessel. Our analysis treated these multiple simultaneous transfusions as “tie” scenarios. In PROPPR, the exact time of transfusion of blood products was recorded. First, there was no real order in a tie, but we could assign the ties a sequence in order to compare them with the protocol. The problem was that many sequences were possible. Consistent with rationale 2, we assigned the order leading to the fewest blood transfusion errors. Second, ‘tie' scenarios varied according to the number of units in the tie and the beginning position. For example, if the tie was from the second unit transfused from a container for the 1:1:1 group and contained four units, the correct order was RBC, plasma, RBC, plasma. If the tie started from the third unit and contained three units, the correct order was plasma, RBC, plasma. In a small dataset, it might be possible to assign an order to the ties manually, but our goal was creating a consistent computerized algorithm that works universally for all cases.
      Starting with the simplest case where only units of RBC and plasma comprised the ‘tied’ positions in the 1:1:1 group, we compared the number of units of RBC and plasma reported with the corresponding numbers based on the protocol. Regardless of the correct order in the protocol, the number of units that can be matched between the reported data and the protocol (i.e. the maximum number of units without an error) was the minimum of the actual number of RBCs in the tie and number of RBCs in the protocol plus the minimum of actual number of plasma in the tie and number of plasma in the protocol for the tie positions. As a result, the number of errors for the tie was as follows:
      Etie1 = Ntie-min (Rtie, RProtocol)-min (Ptie, PProtocol)


      where Ntie was the total number of units in the tie, Rtie and Ptie were the actual number of RBC and plasma in the tie respectively, and RProtocol and PProtocol were the number of RBC and plasma in the protocol for the tie positions respectively. This formula worked for the following three cases: (1) The tie was not at the beginning of a container, and there were no platelets involved in the tie; (2) The tie was not at the beginning of a container, there was at least one platelet involved in the tie, and another dose of platelets has been transfused from the same container. In this case, all the platelets involved in the tie were errors, so the above formula still worked; (3) The tie was at the beginning of the container, and there were no platelets involved in the tie. In this case, other than the matched RBCs and plasma, all the other units were errors. In summary, for all three cases, we only need to match RBCs and plasma between the data and the protocol, and all unmatched units were errors. The above formula can be applied in other projects where ties are possible.
      In this project, there were two more complicated cases. The first case was if the tie was not at the beginning of a container from the 1:1:1 group, there was at least one dose of platelets in the tie, and no other platelets were given in the container, then the number of errors for this tie was
      Etie2 = Ntie-min (Rtie, RProtocol)-min (Ptie, PProtocol)-1,


      since this platelet in the tie can be thought of as a correct substitute of the missing one at the first position. If the tie was at the beginning of a container for the 1:1:1 group and there was at least one platelet in the tie, we could manually assign one of the platelets to the first position, and calculated the number of errors for the remaining units based on rule 10 or rule 11 in Table 1. The guidelines for dealing with ties in the 1:1:2 group (Table 2) was similar to what we have introduced for the 1:1:1 group.

       Analysis of error rates

      Analysis of errors focused on the unique patient level. We defined error proportion to be the proportion of errors in blood product transfusion for each patient, where the denominator was the total number of blood products for each patient. Note that the blood product transfusion errors could be affected by numerous factors that changed over time such as availability of trained personnel. For this reason, the nonparametric approach, Wilcoxon rank sum test, was used to test the hypothesis that the two treatment groups differed in transfusion error proportions, assuming that the treatment effects and error severity do not vary over time.

      Results

      There were 19659 transfused blood products for 680 patients in the data collected, among which 12091 unique RBCs, plasma, and platelets units were transfused during the randomized phase and were included in this study. There were 7568 observations representing other types of blood products (n = 268), missing data (n = 10) or from prior to or after the randomized period (n = 7290), which were excluded from this analysis. As a result, 11 patients without any remaining transfused blood products were removed, among whom 2 are from the 1:1:1 group and 9 are from the 1:1:2 group. Finally, the analysis included 669 unique patients, 336 patients (6417 observations) in the 1:1:1 group and 333 patients (5674 observations) in the 1:1:2 group.
      Table 3 describes the proportion of the transfused blood products by treatments. In the 1:1:1 group, 2781 units of plasma, 655 doses of platelets, and 2981 units of RBCs were transfused. In the 1:1:2 group, 1733 units of plasma, 293 doses of platelets, and 3648 units of RBCs were transfused. Note that one dose of platelets in the table represented a pool of 6 units of platelets on average. Therefore, overall average transfusion ratios of plasma and platelets to RBCs for the 1:1:1 group were 0.93:1.32:1, while the corresponding transfusion ratios for the 1:1:2 group were 0.48:0.48:1. Looking at the ratios transfused in the trial, adherence to the protocol was high. Note that the transfusion protocol could be stopped in the middle of a container due to death or hemorrhage control or other reasons, and the 6 units of platelets were always given at the beginning of a container, which explained why the platelets were higher than expected in the 1:1:1 group.
      Table 3Number and proportion of the transfused blood products by treatment arm.
      Blood products1:1:1 group

      (n = 336)

      N
      N is the number of doses of corresponding transfused blood products, where one dose of platelets represents a pool of six units and one dose of red blood cells or plasma represents one unit.
      (%
      Column percentage represents the proportion of corresponding transfused blood products within each treatment arm.
      )
      1:1:2 group

      (n = 333)

      N
      N is the number of doses of corresponding transfused blood products, where one dose of platelets represents a pool of six units and one dose of red blood cells or plasma represents one unit.
      (%
      Column percentage represents the proportion of corresponding transfused blood products within each treatment arm.
      )
      1. Platelets655(10.2%)293(5.2%)
      2. Red Blood Cells2981(46.5%)3648(64.3%)
      3. Plasma2781(43.3%)1733(30.5%)
      a N is the number of doses of corresponding transfused blood products, where one dose of platelets represents a pool of six units and one dose of red blood cells or plasma represents one unit.
      b Column percentage represents the proportion of corresponding transfused blood products within each treatment arm.
      The mean error proportion in the 1:1:1 group was 0.04 with standard deviation 0.11, while the mean error proportion in the 1:1:2 group was 0.07 with standard deviation 0.11. The median error proportions for both treatment arms were 0, which meant that more than half of the patients did not have any wrong transfused blood products. Actually, 248 patients out of 669 had wrong blood transfusion, among whom 99 patents were from the 1:1:1 group and 149 patients were from the 1:1:2 group. From the p-value of less than 0.0001, we concluded that the two treatment groups differed significantly in blood product transfusion errors. Fig. 2 is a box plot showing the distribution of error proportion by treatment. The 1:1:2 group showed a higher error proportion than the 1:1:1 group.
      Fig. 2
      Fig. 2Box plot of error proportion by treatment.
      In addition to the overall error proportion, we made further studies on the PROPPR trial and detected other important patterns. First, the relationship between the error proportion and total number of transfused blood products were investigated through Spearman's rank correlation coefficient (Fig. 3). There was a statistically significant positive correlation with a p-value less than 0.0001. But the positive correlation was weak, and the correlation was not apparent from the figure. Second, about half of the patients (n = 334) stopped in the first cooler, among whom 183 patients were from the 1:1:1 group and 151 patients were from the 1:1:2 group. The mean error proportion for these patients in the 1:1:1 group was 0.04 with standard deviation 0.14, while the mean error proportion for these patients in the 1:1:2 group was 0.06 with standard deviation 0.12, which was quite consistent with the whole cohort. Third, we explored the error proportion for patients who died. The mean error proportion for these patients in the 1:1:1 group was 0.08 with standard deviation 0.19, while the mean error proportion for these patients in the 1:1:2 group was 0.09 with standard deviation 0.11. Fourth, we further explored the error proportion for patients who died before the PROPPR protocol was discontinued, among whom 22 patients were from the 1:1:1 group and 37 patients were from the 1:1:2 group. The mean error proportion for these patients in the 1:1:1 group was 0.10 with standard deviation 0.15, while the mean error proportion for these patients in the 1:1:2 group was 0.11 with standard deviation 0.14.
      Fig. 3
      Fig. 3Relationship between error proportion and total number of transfused blood products.

      Discussion

      Medicine dispensing errors are very common in clinical trials. Typically, these trials focus on simple treatment arms delivering only one type of medicine with a fixed dose. In these types of trials, the most common errors are either wrong medicine or wrong dose or not taken. However, the PROPPR trial utilized a much more complicated treatment regime, where multiple blood products were given to patients in a pre-specified order. In this paper, we proposed a series of methods to define and calculate different types of blood transfusion errors that would be generalizable to other studies using ordered intervention regimes. These approaches are potentially useful for future trials with multiple interventions and order issues in each treatment arm where it is of interest to calculate overall adherence.
      This paper focused on the adherence to the protocol in terms of the order of the blood products transfused, and it was measured by the error proportion. We did not assess the effect of adherence on treatment outcome with respect to different types of errors. However, we did do a sensitivity analysis excluding patients who experienced blood transfusion error, and obtained results similar to the main analysis [
      • Holcomb J.B.
      • Tilley B.C.
      • Baraniuk S.
      • Fox E.E.
      • Wade C.E.
      • Podbielski J.M.
      • et al.
      Transfusion of plasma, platelets, and red blood cells in a 1:1:1 vs a 1:1:2 ratio and mortality in patients with severe trauma: the PROPPR randomized clinical trial.
      ]. Also treatment during the trial was quite complicated and evaluating the change of error proportion over time would be difficult. Future research could be carried out to study the pattern of error proportions as the protocol continues.
      It is beyond the scope of this article to identify approaches to reducing errors. The approaches would generally be study specific. For example, in interpreting the observed results of this analysis, Level 1 trauma centers require the presence of a specialist trauma surgeon at the bedside. The presence of such a specialist could have reduced the percentage of errors observed in this analysis. Additionally, in this study, all transfusions for each patient were reviewed for gross errors as the study progressed and feedback was provided to the sites in order to prevent similar future errors. This two could have reduced the percent of errors observed. However, the focus of this paper was not approaches to reducing errors but rather to describe an approach to the calculation of these errors for a complex treatment regimen. As it stands our method addresses the adherence question of interest, i.e, can we define each participant as an adherent or non-adherent.
      Finally, some of the approaches to defining and calculating errors were specific to this trial, and served only as a model to consider for other complex studies of adherence. But it is worth noting that the proposed fundamental rationale and major specific approaches generalizable to other trials with similar situations. By giving trials a tool to assess errors, study specific approaches can be put into place to reduce errors, study specific corrective actions can be taken, and progress in the reduction of errors can be monitored.

      Conclusions

      Overall, PROPPR adherence to blood transfusion order pre-specified in the protocol was high, and the required order of transfusions for the 1:1:2 group was more difficult to achieve. The approaches proposed in this manuscript were useful in evaluating the PROPPR adherence and are potentially useful for other trials where a specific treatment orders with varying durations must be maintained.

      Conflicts of interest

      The authors have no declared conflicts of interest related to this manuscript.

      Acknowledgements

      This work was supported with grant U01HL077863 from the US National Heart, Lung, and Blood Institute and funding from the US Department of Defense , the Defense Research and Development Canada in partnership with the Canadian Institutes of Health Research-Institute of Circulatory and Respiratory Health (grant CRR-120612 ).

      References

        • Alexander J.H.
        • Levy E.
        • Lawrence J.
        • Hanna M.
        • Waclawski A.P.
        • Wang J.
        • et al.
        Documentation of study medication dispensing in a prospective large randomized clinical trial: experiences from the ARISTOTLE Trial.
        Am Heart J. 2013; 166: 559-565
        • James K.L.
        • Barlow D.
        • McArtney R.
        • Hiom S.
        • Roberts D.
        • Whittlesea C.
        Incidence, type and causes of dispensing errors: a review of the literature.
        Int J Pharm Pract. 2009; 17: 9-30
        • Spencer M.G.
        • Smith A.P.
        A multicentre study of dispensing errors in British hospitals.
        Int J Pharm Pract. 1993; 2: 142-146
        • Allan E.
        • Barker K.
        Fundamentals of medication error research.
        Am J Health-Syst Pharm. 1990; 47: 555-571
        • Holcomb J.B.
        • Tilley B.C.
        • Baraniuk S.
        • Fox E.E.
        • Wade C.E.
        • Podbielski J.M.
        • et al.
        Transfusion of plasma, platelets, and red blood cells in a 1:1:1 vs a 1:1:2 ratio and mortality in patients with severe trauma: the PROPPR randomized clinical trial.
        JAMA. 2015; 313: 471-482
        • Baraniuk S.
        • Tilley B.C.
        • Del Junco D.J.
        • Fox E.E.
        • Van Belle G.
        • Wade C.E.
        • et al.
        Pragmatic randomized optimal platelet and plasma ratios (PROPPR) trial: design, rationale and implementation.
        Injury. 2014; 45: 1287-1295
        • Novak D.J.
        • Bai Y.
        • Cooke R.K.
        • Marques M.B.
        • Fontaine M.J.
        • Gottschall J.L.
        • et al.
        Making thawed universal donor plasma available rapidly for massively bleeding trauma patients: experience from the pragmatic, randomized optimal platelets and plasma ratios (PROPPR) trial.
        Transfusion. 2015; 55: 1331-1339