u
u What
Counts as Equipoise?
We appreciate the valuable remarks of the commentators on our article “Were There ‘Additional Foreseeable Risks’ in the SUPPORT Study?” (January-February 2015), especially those regarding the issue of equipoise in the Surfactant, Positive Pressure, and Oxygenation Randomized Trial (SUPPORT). We believe, however, that their remarks (in the same issue of the Report) expose some misunderstanding of our position and of the federal regulations regarding research involving children. First, in “SUPPORT: Risks, Harms, and Equipoise,” Robert Nelson states that we claimed that the study as designed “was not in sufficient equipoise,” as there was “insufficient uncertainty . . . about the relative effectiveness of targeting the higher or lower SpO2 range.” In “The Controversy over SUPPORT Continues and the Hyperbole Increases,” Alan Fleischman states that “[a]ll agreed that there was clinical equipoise for the trial.” We also agree that equipoise existed between the study arms, and we made no statement to the contrary. However, this variant of equipoise, referred to here as “equipoise1,” is neither the one we had discussed, nor is it the equipoise required by 46.406 of the subpart D regulations, here “equipoise2.” The nuance of this regulation includes two requirements. First, the risk of harm associated with an intervention in any arm of a study must be justified by the prospect of direct benefits to the child. We agree that there was a prospect of direct benefits from both study interventions. But having just the prospect of direct benefits is insufficient, as the second part of the 46.406 requires that the balance of the anticipated benefits and the risk of harm of the study arms’ May-June 2015
exchange
t
interventions be as favorable as that presented by available alternative treatments outside the trial. The underlying ethical reason for considering available alternative approaches is that, as the Institute of Medicine has stated, “a child’s health should not be placed in jeopardy by the decision to enter the child into a research protocol.” Another way to explain equipoise2 is that the net risks of research involving children need to be zero. They are greater than zero when the risk-benefit ratio of any of the study interventions is not favorable to available alternative treatments. In research involving adults, net risks can exceed zero, but this needs to be justified by the scientific knowledge to be gained. In research involving children, any positive net risks (above
Having just the prospect of direct benefits to the child was insufficient in SUPPORT. a minor increment over minimal risk) should not be justified by potential scientific knowledge, unless approved by a 407 panel. John Lantos and Chris Feudtner, authors of “SUPPORT and the Ethics of Study Implementation” (January-February 2015 also), would probably counter by opining that equipoise2 did exist in SUPPORT, as “treatment by protocol was and is the standard of care in all NICUs [neonatal intensive care units],” and, hence, “concerns about treatment by protocol are particularly irrelevant to the SUPPORT controversy.” However, the issue is not whether treatment by a protocol is the standard of care but whether a particular research protocol satisfied the equipoise2 requirements. We have argued that the risk-benefit
balance in the SUPPORT protocol was not as favorable as that in standard practice. Fleischman argues that there was clinical equipoise2 between SUPPORT and the available alternative standard therapies, as there was a potential to reduce the incidence of severe retinopathy “for at least some of the subjects” (presumably, in the lower oxygen range). It is doubtful whether the second requirement (equipoise2) in 46.406 regarding risks and benefits could have been satisfied when only “some of the subjects” were not disadvantaged by being in the research compared with what they could have achieved outside the trial. Finally, Nelson claims that we stated that SUPPORT should have included a usual-care group, but we suggested only that such a group might be considered in order to minimize risks. We recognize the complexities of involving a usual-care group in such trials; however, as our article pointed out, recent trials including a usual-care group have shown the effectiveness of such a group, effectiveness that would have remained hidden had these groups been excluded from the trial designs. Unfortunately, a posteriori data like that cited by Nelson to support his contention that a usualcare group would not have been informative will not be available when the ethics of such trials is assessed a priori, which brings us to our final point: it is not surprising that a trial as complex as SUPPORT, which presented differences in the types, severity, and probability of the risk between the study arms and also compared with usual-care practices, has engendered much discussion. A 407 panel, which is warranted by the federal regulations for any trial similar to SUPPORT, might be more desirable so that these discussions can occur at a trial’s
H AS TI N GS C E N TE R RE P O RT
3
inception, which is when the ethical assessment should occur. • Henry J. Silverman University of Maryland School of Medicine • Didier Dreyfuss Paris-Diderot University DOI: 10.1002/hast.437
the author replies tThe need for Henry J. Silverman and Didier Dreyfuss to superscript “equipoise” reminds me why I generally do not use “equipoise” when discussing the ethics of clinical trials, given its various meanings. Rather, I use “genuine uncertainty” to avoid the debate about whether a physician’s therapeutic obligation underlies the ethics of clinical research. Although “net risks” was developed by David Wendler and Franklin G. Miller to avoid the entanglement of “equipoise” with Charles Weijer’s “component analysis,” I also avoid this language as unnecessary and misleading. My main objection to using “net risks” is the resulting misimpression that risk assessment is purely quantitative, rather than a more complex qualitative (and quantitative) judgment about the relative value (and probability) of harms. The distinction between risk and harm is central to my views on the Surfactant, Positive Pressure, and Oxygenation Randomized Trial (SUPPORT). Identifying “reasonably foreseeable risk” requires knowing the probability of harm. Even if the potential harm following assignment to either treatment group was different (in other words, retinopathy or death), absent knowledge of the probability of those harms, we did not have any “reasonably foreseeable risks.” To be clear, parents should have been informed about the different harms to which infants may be exposed through randomization and been able to exercise an informed choice about exposing their infants to them. However, prior to SUPPORT, we did not know whether maintaining an infant’s oxygen saturation between 85 and 89 percent, between 91 and 95 percent, or anywhere in between based on protocoldriven “usual care” was better or worse. 4 HASTIN G S C E N T E R R E P ORT
Therefore, the risks and potential direct benefits of being in either arm of SUPPORT were comparable to the alternative of not being in the trial. I did not misunderstand the use of “equipoise2”; I simply disagree. There were no additional “reasonably foreseeable risks” to being in SUPPORT. When the ethics of SUPPORT are evaluated a priori, infants who were enrolled in the trial were not disadvantaged or placed in jeopardy relative to those not enrolled. SUPPORT was approvable under 45 CFR 46.405 and did not require a referral for federal review under 45 CFR 46.407. Finally, understanding how to design a clinical trial using a continuous variable such as oxygen saturation supports a priori the need for adequate separation between groups to observe any differences in the hypothesized outcomes.
Infants in the trial were not placed in jeopardy relative to those not enrolled. This separation would not have been achieved with a usual-care group. I was not making an a posteriori argument, although the presence of supporting data may have given this impression. The 2014 trial comparing protocol-driven care to usual care in the treatment of early sepsis is simply beside the point. In that trial, the primary hypothesis compared early goal-directed treatment with usual (non–protocol-driven) care. However, a comparison with usual care was not part of the SUPPORT hypotheses, and usual care in managing oxygen in neonatal intensive care units is protocol driven. It is unclear how a usualcare group in SUPPORT would have minimized risks, rather than randomly distributing the various harms of higher or lower oxygen saturations within one arm of the protocol. The opinions expressed in this commentary are those of the author and do not represent the policies of the Food and Drug Administration. • Robert M. Nelson U.S. Food and Drug Administration DOI: 10.1002/hast.438
u Muddled
Measures of Risks and Misremembered Reasons
While there are doubtless lessons to be learned from the Surfactant, Positive Pressure, and Oxygenation Randomized Trial itself, we are struck by two critically important lessons emerging from the post-SUPPORT analyses of the controversy, including those that appeared with our article “SUPPORT and the Ethics of Study Implementation”: “Were There ‘Additional Foreseeable Risks’ in the SUPPORT Study?,” by Henry J. Silverman and Didier Dreyfuss, “SUPPORT: Risks, Harms, and Equipoise,” by Robert M. Nelson, “The Controversy over SUPPORT Continues and the Hyperbole Increases,” by Alan R. Fleischman, and “SUPPORT and Comparative Effectiveness Trials: What’s at Stake?,” by Lois Shepherd (all in January-February 2015). First, even after the study is complete and we know the results, experts cannot agree about how we should describe and think about the various types of potential benefits and harms. Nobody can agree on the proper algorithm to account for both the magnitude and the likelihood of harms and benefits or on ways to cogently describe those that might be related to differences between being inside or outside of the trial compared to the differences in outcomes between the trial’s two arms. The ensuing Tower of Babel debates do not bode well for our efforts to communicate risks to potential research participants. Case in point: an essential assumption in most critiques of the study design and consent process for SUPPORT is that babies were harmed by participating in the study. There is no evidence that this is so. Instead, all the evidence shows that babies in the study had better outcomes than comparable babies outside it. Indeed, one key finding of the study—the difference in mortality between the two arms—arose not because the babies in the low oxygen arm had a higher risk of death than expected. May-June 2015
Copyright of Hastings Center Report is the property of Wiley-Blackwell and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
u What
Counts as Equipoise?
We appreciate the valuable remarks of the commentators on our article “Were There ‘Additional Foreseeable Risks’ in the SUPPORT Study?” (January-February 2015), especially those regarding the issue of equipoise in the Surfactant, Positive Pressure, and Oxygenation Randomized Trial (SUPPORT). We believe, however, that their remarks (in the same issue of the Report) expose some misunderstanding of our position and of the federal regulations regarding research involving children. First, in “SUPPORT: Risks, Harms, and Equipoise,” Robert Nelson states that we claimed that the study as designed “was not in sufficient equipoise,” as there was “insufficient uncertainty . . . about the relative effectiveness of targeting the higher or lower SpO2 range.” In “The Controversy over SUPPORT Continues and the Hyperbole Increases,” Alan Fleischman states that “[a]ll agreed that there was clinical equipoise for the trial.” We also agree that equipoise existed between the study arms, and we made no statement to the contrary. However, this variant of equipoise, referred to here as “equipoise1,” is neither the one we had discussed, nor is it the equipoise required by 46.406 of the subpart D regulations, here “equipoise2.” The nuance of this regulation includes two requirements. First, the risk of harm associated with an intervention in any arm of a study must be justified by the prospect of direct benefits to the child. We agree that there was a prospect of direct benefits from both study interventions. But having just the prospect of direct benefits is insufficient, as the second part of the 46.406 requires that the balance of the anticipated benefits and the risk of harm of the study arms’ May-June 2015
exchange
t
interventions be as favorable as that presented by available alternative treatments outside the trial. The underlying ethical reason for considering available alternative approaches is that, as the Institute of Medicine has stated, “a child’s health should not be placed in jeopardy by the decision to enter the child into a research protocol.” Another way to explain equipoise2 is that the net risks of research involving children need to be zero. They are greater than zero when the risk-benefit ratio of any of the study interventions is not favorable to available alternative treatments. In research involving adults, net risks can exceed zero, but this needs to be justified by the scientific knowledge to be gained. In research involving children, any positive net risks (above
Having just the prospect of direct benefits to the child was insufficient in SUPPORT. a minor increment over minimal risk) should not be justified by potential scientific knowledge, unless approved by a 407 panel. John Lantos and Chris Feudtner, authors of “SUPPORT and the Ethics of Study Implementation” (January-February 2015 also), would probably counter by opining that equipoise2 did exist in SUPPORT, as “treatment by protocol was and is the standard of care in all NICUs [neonatal intensive care units],” and, hence, “concerns about treatment by protocol are particularly irrelevant to the SUPPORT controversy.” However, the issue is not whether treatment by a protocol is the standard of care but whether a particular research protocol satisfied the equipoise2 requirements. We have argued that the risk-benefit
balance in the SUPPORT protocol was not as favorable as that in standard practice. Fleischman argues that there was clinical equipoise2 between SUPPORT and the available alternative standard therapies, as there was a potential to reduce the incidence of severe retinopathy “for at least some of the subjects” (presumably, in the lower oxygen range). It is doubtful whether the second requirement (equipoise2) in 46.406 regarding risks and benefits could have been satisfied when only “some of the subjects” were not disadvantaged by being in the research compared with what they could have achieved outside the trial. Finally, Nelson claims that we stated that SUPPORT should have included a usual-care group, but we suggested only that such a group might be considered in order to minimize risks. We recognize the complexities of involving a usual-care group in such trials; however, as our article pointed out, recent trials including a usual-care group have shown the effectiveness of such a group, effectiveness that would have remained hidden had these groups been excluded from the trial designs. Unfortunately, a posteriori data like that cited by Nelson to support his contention that a usualcare group would not have been informative will not be available when the ethics of such trials is assessed a priori, which brings us to our final point: it is not surprising that a trial as complex as SUPPORT, which presented differences in the types, severity, and probability of the risk between the study arms and also compared with usual-care practices, has engendered much discussion. A 407 panel, which is warranted by the federal regulations for any trial similar to SUPPORT, might be more desirable so that these discussions can occur at a trial’s
H AS TI N GS C E N TE R RE P O RT
3
inception, which is when the ethical assessment should occur. • Henry J. Silverman University of Maryland School of Medicine • Didier Dreyfuss Paris-Diderot University DOI: 10.1002/hast.437
the author replies tThe need for Henry J. Silverman and Didier Dreyfuss to superscript “equipoise” reminds me why I generally do not use “equipoise” when discussing the ethics of clinical trials, given its various meanings. Rather, I use “genuine uncertainty” to avoid the debate about whether a physician’s therapeutic obligation underlies the ethics of clinical research. Although “net risks” was developed by David Wendler and Franklin G. Miller to avoid the entanglement of “equipoise” with Charles Weijer’s “component analysis,” I also avoid this language as unnecessary and misleading. My main objection to using “net risks” is the resulting misimpression that risk assessment is purely quantitative, rather than a more complex qualitative (and quantitative) judgment about the relative value (and probability) of harms. The distinction between risk and harm is central to my views on the Surfactant, Positive Pressure, and Oxygenation Randomized Trial (SUPPORT). Identifying “reasonably foreseeable risk” requires knowing the probability of harm. Even if the potential harm following assignment to either treatment group was different (in other words, retinopathy or death), absent knowledge of the probability of those harms, we did not have any “reasonably foreseeable risks.” To be clear, parents should have been informed about the different harms to which infants may be exposed through randomization and been able to exercise an informed choice about exposing their infants to them. However, prior to SUPPORT, we did not know whether maintaining an infant’s oxygen saturation between 85 and 89 percent, between 91 and 95 percent, or anywhere in between based on protocoldriven “usual care” was better or worse. 4 HASTIN G S C E N T E R R E P ORT
Therefore, the risks and potential direct benefits of being in either arm of SUPPORT were comparable to the alternative of not being in the trial. I did not misunderstand the use of “equipoise2”; I simply disagree. There were no additional “reasonably foreseeable risks” to being in SUPPORT. When the ethics of SUPPORT are evaluated a priori, infants who were enrolled in the trial were not disadvantaged or placed in jeopardy relative to those not enrolled. SUPPORT was approvable under 45 CFR 46.405 and did not require a referral for federal review under 45 CFR 46.407. Finally, understanding how to design a clinical trial using a continuous variable such as oxygen saturation supports a priori the need for adequate separation between groups to observe any differences in the hypothesized outcomes.
Infants in the trial were not placed in jeopardy relative to those not enrolled. This separation would not have been achieved with a usual-care group. I was not making an a posteriori argument, although the presence of supporting data may have given this impression. The 2014 trial comparing protocol-driven care to usual care in the treatment of early sepsis is simply beside the point. In that trial, the primary hypothesis compared early goal-directed treatment with usual (non–protocol-driven) care. However, a comparison with usual care was not part of the SUPPORT hypotheses, and usual care in managing oxygen in neonatal intensive care units is protocol driven. It is unclear how a usualcare group in SUPPORT would have minimized risks, rather than randomly distributing the various harms of higher or lower oxygen saturations within one arm of the protocol. The opinions expressed in this commentary are those of the author and do not represent the policies of the Food and Drug Administration. • Robert M. Nelson U.S. Food and Drug Administration DOI: 10.1002/hast.438
u Muddled
Measures of Risks and Misremembered Reasons
While there are doubtless lessons to be learned from the Surfactant, Positive Pressure, and Oxygenation Randomized Trial itself, we are struck by two critically important lessons emerging from the post-SUPPORT analyses of the controversy, including those that appeared with our article “SUPPORT and the Ethics of Study Implementation”: “Were There ‘Additional Foreseeable Risks’ in the SUPPORT Study?,” by Henry J. Silverman and Didier Dreyfuss, “SUPPORT: Risks, Harms, and Equipoise,” by Robert M. Nelson, “The Controversy over SUPPORT Continues and the Hyperbole Increases,” by Alan R. Fleischman, and “SUPPORT and Comparative Effectiveness Trials: What’s at Stake?,” by Lois Shepherd (all in January-February 2015). First, even after the study is complete and we know the results, experts cannot agree about how we should describe and think about the various types of potential benefits and harms. Nobody can agree on the proper algorithm to account for both the magnitude and the likelihood of harms and benefits or on ways to cogently describe those that might be related to differences between being inside or outside of the trial compared to the differences in outcomes between the trial’s two arms. The ensuing Tower of Babel debates do not bode well for our efforts to communicate risks to potential research participants. Case in point: an essential assumption in most critiques of the study design and consent process for SUPPORT is that babies were harmed by participating in the study. There is no evidence that this is so. Instead, all the evidence shows that babies in the study had better outcomes than comparable babies outside it. Indeed, one key finding of the study—the difference in mortality between the two arms—arose not because the babies in the low oxygen arm had a higher risk of death than expected. May-June 2015
Copyright of Hastings Center Report is the property of Wiley-Blackwell and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
