Abstract
Abstract: Establishing an acceptable safety and efficacy profile for hemoglobin based oxygen carriers, HBOCs, is essential. Understanding the data that describes an HBOC “safety profile” requires consideration beyond counting “events” typically associated with intent to treat analysis. The imputation of causation from numerical counts alone, without clinical context, is an incomplete description of a complex situation. Clinical contextualization provides a greater understanding of the clinico-pathological processes involved. Generating alternative hypotheses for the origin of the events, apart from the drug, patient co-morbidities, situational and disease demands, could include protocol design, failure to provide mitigation strategies, patient management issues, or inadequate education of investigators. How clinical contextualization can provide insight for the interpretation of significant safety profile events is discussed in the context of a large phase III clinical trial where an appreciation of factors underlying the differences between intent to treat analysis and other approaches is discussed.
Not everything that can be counted counts, and not everything that counts can be counted. A. Einstein
INTRODUCTION
The quotation from Albert Einstein has particular resonance when evaluating the adverse event profiles obtained in clinical trials of hemoglobin based oxygen carriers (HBOC). There is more to this evaluation than the accounting of events. The integration of clinical contextualization into the analysis of adverse events, as they reflect “safety,” is proposed as a necessary, additional, vital part of post-hoc analysis important and critical to ascribing causality. This construct would include in the safety analysis the totality of clinical care including the expected morbidity and mortality for the underlying disease and mode of treatment as well the impact of underlying co-morbidities on which is superimposed the new treatment or therapeutic agent. The objective of ascribing a causal relationship using clinical contextualization, root cause, and subset analysis to explain the results of an intent-to-treat (ITT) analysis is the prevention of misinterpretation of the results.
The issue of causality and the emergence of adverse safety signals in HBOC trials evoke responses ranging from provocative to contentious, depending on the perspective. HBOCs, more properly called oxygen therapeutics, are recognized as a potentially useful class of drugs with the capacity to deliver oxygen to tissue to maintain metabolism or reverse ischemia. Before being dismissed as “toxic” or “unsafe,” these terms are not synonymous, a thorough multidimensional evaluation is essential to establishing a safety profile. Toxicity relates to drug properties while safety is assumed associated with treatment related properties.
The evaluation of HBOCs in the treatment of acute anemia requires an appreciation for the universe of factors underlying the outcomes; it requires recognition and understanding of all the dimensions of the study, including the protocol design and analytical methodology to which a clinical context must be added. It is insufficient to rely only on the study design and analytical methodology to explain the results; the former may be flawed and the latter incomplete. The objective is to assure clinicians and regulatory agencies that these solutions are safe and effective in clinical use. Whether there is a causal relationship between the infusion of HBOC and the emergence of adverse events depends not only on the properties and intrinsic toxicity of the HBOC but also on the many aspects and dimensions of clinical care, the properties associated with treatment variations.
UNDERSTANDING SAFETY & CAUSATION IN CLINICAL CONTEXT
Demonstrating the efficacy and safety of HBOCs is essential for licensure and clinical use. Exactly how those end points, particularly safety, meet the regulatory criteria and standards for approval requires comment and discussion. The generally recognized “gold standard” for evaluating new therapies, especially drugs, is the double-blind randomized clinical trial with efficacy and “safety profile” evaluation based solely on an “intent to treat” (ITT) analysis. While this established, scientifically sound, ideal model and experimental design is desirable, its rigid application in all situations is problematic. For a variety of well known reasons, maintaining a double blind is complicated and resource intense when evaluating HBOCs.
It is argued that a negative ITT is insufficient of itself to stop drug development; an analysis of why and how critical adverse events emerge is required to understand them. The need to establish a correct data-supported safety profile is not argued; the content and form of that profile is the major issue. A well-grounded safety profile, based on trial emergent adverse events and safety signals, is required to support proposed indications and proper product labeling. It is an unfortunate fact that, based on counting numbers of adverse events, safety profiles are constructed without clinical contextualization to determine whether the HBOC will receive regulatory approval. Alternative explanations for the observations, rational and logically sound, formulated within the context of clinical medicine, must become a component of this critical evaluation. To assist in the evaluation the known properties observed in HBOC clinical trials, transient side effects of minimal clinical impact, should be an element of sub-group analysis, excluded in the between group analysis; residual differences in the subgroups are then assumed to reflect intrinsic HBOC toxicity. It is how and why these differences in the sub-groups differ from the ITT that needs exploration. Discrepancies between an ITT analysis and sub-set analyses should be resolved with all available analytical tools and approaches to confirm or reject hypotheses regarding safety and efficacy.
To appreciate the differences requires clinical contextualization. One tool often disparaged in the analysis of clinical trial data is root cause analysis. This runs counter to the currently invoked and stressed data analysis approaches proposed and used by the safety scientists in areas of quality assurance and drug development where a full and complete understanding of all factors contributing to the safety observations are sought and highly valued. Rejection of post-hoc analyses, especially with respect to including clinical contextualization and root cause analysis, given the complexity of modern medicine, is inappropriate. It limits the potential for identification of critical clinical alternative causality explanations. Clinical contextualization is in many ways a model of root cause analysis as it gives context to answer the question “why” beyond counting numbers.
Root cause analysis provides additional categories and options for explaining the adverse events. This includes more detailed descriptions of demographics, patient co-morbidities, the impact of the current disease on the patient's general health status and factors related to treatment and management issues. Examples of these are introduced below. Furthermore, these same factors play a role in assessing “dose-response” issues, also noted below.
Assessment of whether the adverse events are the results of HBOC toxicity or other clinically identified causes is both critical and essential to understanding the safety profile. The inclusion of clinical context helps focus and complements the analysis, preventing misinterpretation of results.
There are a number of problems in defining the safety profile through the use of an integrated safety summary (ISS), an approach subject to the criticism implicit in the Einstein quotation. Construction of the ISS includes data and observations acquired during prosecution of all clinical studies with the HBOC from all clinical phases of the development process. These studies are inherently heterogeneous on many dimensions with different dosing, protocol designs, controls and comparators, in different clinical settings with differing populations and different outcome measures. Combining data with a high degree of heterogeneity maximizes the identification and quantification of “safety signals” without considering the clinical context. Only in retrospect, using post-hoc analysis, can the key and essential elements clinical context be identified and applied in the analysis. The de facto imposition of a predefined statistical analysis plan with the requirement that it be the primary, if not only, analytical model for defining safety is the procrustean bed of the ISS approach. The ISS is useful in generating testable hypotheses. It does not, however, permit translation of inferior safety arising from treatment to greater toxicity of the drug. These are not comparable variables.
Moreover, it must be emphasized that in HBOC studies where packed red blood cells (PRBC) are the comparator, achieving superiority would be impossible; attaining equipoise in safety and efficacy would, however, be possible when the comparator is a crystalloid or colloid solution. When crystalloid or colloid solution is the comparator, HBOC has the inherent advantage of oxygen carrying capacity and thus a capacity to minimize, prevent, or modulate the ischemic injury concomitant with hemorrhagic shock or low perfusion state. It must be recognized that, in terms of relative efficacy, PRBCs have a distinct advantage over all current HBOCs in development.
The desired analytical requirement for data homogeneity is lost in a broadly constructed ISS. Doing studies in surgery patients only does not assure homogeneity. Combining the incidence of various adverse events into “sets” thought to represent toxicity, physiologic compromise or organ dysfunction without consideration of the intrinsic co-dependence of complex clinical variables lacks face validity and exaggerates, overstating, the safety people.
Establishment of the best possible risk assessment used to calculate an accurate benefit-risk ratio requires an accurate safety profile. In attributing all trial emergent adverse events to the HBOC, using a purely numerical approach without considering known clinical uncertainties, represents a significant bias. To the extent that medicine remains very much an “art” and not yet a hard scientifically evidenced practiced discipline, this approach poses a concern.
While a correlation between HBOC administration and the emergence of adverse events exists, it is proffered that this relationship does not, by itself, reflect causation. It is when clinical contextualization is added that the question of causation is directly and clearly addressed. Outcomes are impacted by many variables, some predictable and others not. Clinical trials impose onto this complexity an additional complicating and confounding variable: the experimental treatment. An assumption of the ideal trial design, with sufficient power to show differences in variables of interest, is that randomization creates equal comparable populations of patients except for the use of test material.
Unfortunately, many of these clinically associated variables are idiosyncratic, not readily predictable and cannot be assumed to “randomize out” even when known. The clinical context in which the adverse events emerge can be neither denied nor excluded in the analysis when establishing the HBOC safety profile.
AN EXAMPLE FROM AN HBOC CLINICAL TRIAL
These issues come into sharp focus in the data analysis of a Phase III pivotal trial with Hemopure®, in elective orthopedic surgery when applying the aforementioned concepts of clinical contextualization to this trial, with cases for emphasis as an example. This randomized, single blind trial conducted at 47 sites on 3 continents involved 688 patients. HBOC-201(Hemopure) or packed red blood cells (PRBC) were administered at the first clinical transfusion decision. The primary efficacy end-point of the HBOC arm was PRBC avoidance within 6 weeks and was attained in 60% of patients, a significant achievement with implications for blood conservation and blood banking in general. Cross over to PRBC within the HBOC-201 arm was permitted, and 40% of the patients did so within hours to days after initial randomization. There was no comparable population in the PRBC study arm, complicating the safety analysis. Developing an understanding of how a concentration of adverse events emerged in this subset of patients demonstrates how clinical contextualization can address the “correlation or causation” question, given that exposure to both test agents confounds any assessment. An unequivocal bias introduced by the asymmetrical design compromised assessment of the intrinsic safety of HBOC-201 in this study.
The HBOC-201 crossover group differed from those avoiding PRBC in a number of different, statistically significant variables. Selected examples from a list of nearly 20 were: total fluid (crystalloid/colloid) administered; higher estimated blood loss; longer anesthesia time; longer operating time; lower baseline hemoglobin concentration at first treatment (32% < 8 gm/dL); greater incidence of history of cardiac disease; first treatment in the OR; more cell saver blood. Interestingly, the volume of PRBCs administered was greater than that given PRBC patients. In surgery, it is difficult to predict which patients and for which surgeons these differences will occur; moreover, it is impossible at randomization to assure equal distribution of these variables in a relatively homogeneous trial let alone a multicenter multinational trial.
Recognizing the incidence of both surgical procedure related morbidity and mortality and the risks surgical intervention superimposes on underlying co-morbidities, a method for integrating appropriate values or their carefully designed surrogates into the protocol and statistical plan would add a significant and necessary frame of reference for comparison not now available. Surgical procedures have complications whose incidence varies from site to site and surgeon to surgeon; similarly, practice patterns and “standards of care” vary from country to country and continent to continent. Attribution of causation is complicated and confounded in these high entropy situations. Randomization in clinical HBOC trials is assumed to accomplish this, yet it may not yield the desired results in large part because the trials are not large enough to achieve that goal.
CLINICAL CONTEXTUALIZATION: CASE EXAMPLES
Underlying analysis of HBOC trial data is the assumption that adverse events and hence the safety profile reflect HBOC toxicity, terms not synonymous. An equally plausible alternative assumption is that the safety profile, composed of the amassed adverse events, very much reflects patient and treatment based variables. The following case examples demonstrate how the addition of clinical context confirms the principles outlined, permitting a more appropriate assessment of safety and challenges HBOC toxicity as the causative factor.
79 year old male, total hip replacement, received 3 units of HBOC-201 in the recovery area. 14 hours later, bleeding from the operative site, hypotensive with low hemoglobin experienced a cerebral vascular accident. He had received 7 additional units of HBOC-201 (maximum allowable dose in the study) and 7 units of PRBC. Persistent hypovolemia from ongoing bleeding and decreased tissue perfusion contributed significantly to the adverse event.
69 year old female, spine surgery, received 2 units of HBOC-201 in the OR and 16L of other fluids. Not given diuretics over the next 3 days. On post-op day 3 experienced pulmonary edema. Fluid overload contributed significantly to the emergence of the adverse cardiovascular event.
62 year old female, total hip replacement, received 4 units of HBOC-201 in OR and recovery area. Experienced a post-operative bleed with Hb to 4.5gm/dL that evening. Received 2 units of PRBC. Sustained period of hypoperfusion. 3 days hence had myocardial infarction and cerebral vascular accident. Delay in treating bleeding a significant factor.
86 year old male, total hip replacement, received 4 units of HBOC-201, 500 ml of cell saver infusion and 500 ml of albumin. On first post-op day Hb was 6.4gm/dL and patient reported chest pain followed by a cardiac arrest. 5 units of PRBCs given. Patient had cardiac catheterization with stent placement. Myocardial infarction is a known complication in orthopedic surgery, especially in elderly patients.
87 year old female, total knee replacement, received 2 units HBOC-201 on post-operative day one and 2 units of PRBC on post-operative day 2. Became hemodynamically unstable on post-operative day 3 and experienced a myocardial infarction, pulmonary edema, with a papillary muscle rupture. Post operative cardiac complication in an elderly patient.
An example from the PRBC arm of the study:
65 year old male, redo-total hip replacement. Long surgery with estimated blood loss of 5000 ml; received 8 L crystalloid, 7.4 L colloid and 14 units of PRBC. Experienced hypotension for 48 hours while on pressors. Cerebral vascular accident on post-op day 3.
Each of these patients had received various doses of HBOC and experienced adverse events. When clinical context is considered for each there is an alternative, plausible, clinically based explanation for the emergence of the adverse events. Similar events were observed in the PRBC arm of the study.
There were 37 significant CNS-cardiovascular adverse events in the cross-over group of the HBOC-201 study arm and 13 in the PRBC arm considered to be ischemia associated. Of note, these events are of particular regulatory concern. If patient oxygen delivery and tissue perfusion needs are not met, ischemic events or the inability to attain physiologic compromise ensues with an emergence of these adverse events. Thought to represent an intrinsic HBOC toxicity, the clinical contextualization of these adverse events actually provides an alternate explanation. Expert clinical review by cardiologists, surgeons and anesthesiologists identified five clinically relevant root causes that could account for the emergence of these adverse events. For each patient, one or more of the following, clinically related root causes could be assigned as causative: (a) under-treatment/resuscitation; (b) delay in adequate treatment; (c) volume overload; (d) patient need exceeded product capability; and (e) patient age. Under-treatment could be related to a lack of efficacy to be noted below. These factors accounted for the greatest differences in safety analysis in the ITT analysis. They were concentrated in the unique asymmetrical HBOC-201 group for which there was no comparable “control” in the PRBC group. It is the excess of these events in this sub-set that represents the difference in safety. All other adverse events in these groups were “balanced.”
Transfusion in the PRBC arm maintained Hgb concentrations above the protocol defined “transfusion trigger.” In the HBOC-201 arm, 60% of patients were maintained and managed safely without PRBC transfusion with Hgb concentrations maintained just below the trigger. For 40% of HBOC-201 patients, the crossovers, sufficient levels of Hgb were not maintained and thus emerged an increased incidence of ischemia-related adverse events.
Clinical contextualization must become an evaluation element, part of the assessment of benefit-risk and risk-tolerance considerations. Clinical contextualization identifies factors that could account for the emergence of the adverse events. Clinical interventions, or their absence, could be responsible for the observations. In evaluating a safety profile, it is essential to identify all factors that could contribute to the unexpected outcomes.
THE DOSE RESPONSE RELATIONSHIP IN HBOC CLINICAL TRIALS
The “classic” dose-response relationship is not a valid representation of events in HBOC trials. Assuming the incidence of adverse events represents “toxicity,” in reality worse safety, then “dose” in reality must represent patient “need.” The “need” for additional oxygen carrying capacity for treating acute anemia, a transfusion, most often reflects a physician's clinical judgment and decision—based on a perception and perhaps an experiential heuristic, conditioned by the clinical situation at hand—of a requirement to meet a patient's need at the moment. “Dose” becomes a surrogate for clinical need, recasting the classical dose-response model in which decreasing doses result in fewer adverse events. However, when replacing dose with its surrogate, “need,” decreasing the amount of oxygen carrier deemed needed, whether PRBC or HBOC, would be the antithesis of intent likely to incur some additional degree of ischemic insult. Paradoxically, there would be an increase in ischemic related adverse events arising from under-treatment, a clinically unacceptable outcome.
The cross-over group represents a population with “high needs” for additional oxygen carrier infusion. Their need could not be met by the HBOC-201 and, critically, the majority of the safety imbalance between the arms of the study resides in this population. Clinical contextualization provides insight, the clinically relevant explanation and reasons underlying this difference. Another possible explanation is a lack of HBC-201 efficacy contributing to the emergence of adverse events and not HBOC toxicity. It is possible that patient needs could not be met with a solution whose hemoglobin concentration is significantly less than that of PRBCs.
The magnitude of the relationship between dose (“need”) and adverse events/patient—a measure of “safety”—can be defined using identifiable clinical elements, each contributing to the emergence of adverse events. They may not be the only ones; however, they are of sufficient clinical importance to be considered when seeking causation within a correlation. In no particular order these elements are: (a) patient co-morbidities; (b) age; (c) acute or chronic disease state; (d) protocol design; (e) local and regional patient management preferences; (f) inadequate education of investigators; and (g) failure to provide mitigation strategies.
A simple correlation model based on the counting events cannot address the causation issue; including “clinical contextualization” is essential to understanding the observations. Specification of causation requires a more complex multidimensional approach that permits dissection and attribution of causative elements from correlative ones.
SUMMARY AND CONCLUSION
The relationship between dose, reflecting “need,” and the emergence of adverse events, measures of safety and not measures of HBOC toxicity, is one of great complexity. Clinical contextualization of the adverse events observed, e.g. under-treatment, delays in treatment, fluid overload, provides non-numerical insight, an alternative hypothesis, to explain their origin and emergence. An essential, predictable lack of HBOC efficacy of the HBOC to meet “high needs” when compared to PRBCs, clarifies the picture. Any outcome difference between a subset analysis and the ITT analysis requires explanation. Clinical contextualization is proposed as a necessary, additional approach, an applicable methodology for providing a rational and plausible explanation. Many of the significant adverse events observed in trials with HBOCs reflect the patient's state of health, the disease state, medical or surgical interventions, and treatment and clinical management variations onto which HBOC treatment has been imposed. An appreciation of the clinical aspects and elements underlying the observations is essential even at the cost of introducing another layer of analytical complexity. However, adding clinical context is necessary to establishing a causal relationship between HBOC use and the incidence of adverse events comprising the safety profile.
Clinical contextualization, understanding the clinical elements at play associated with the emergence of the significant adverse events, must become an essential component of trial design, prosecution and data analysis for HBOCs. These unique solutions hold enormous potential to do well in many clinical situations, especially when blood is not an option or readily available and in applications involving ischemic rescue. There has been for decades a recognized need for an HBOC in trauma, elective surgery and in underdeveloped nations where the blood supply is either tainted or just not available. HBOCs, based on analysis of their clinical data, may have relatively benign, clinically acceptable, safety profiles once there is an appreciation of the multitude of clinical factors that contribute to the emergence of adverse events.
A simple correlation model based on the counting of events cannot address the causation issue; including “clinical contextualization” is essential to understanding the observations. Specification of causation requires a more complex multidimensional approach that permits dissection and attribution of causative elements from correlative ones.
“Not everything that counts can be counted.”