white paper

Managing Hyperglycemia in Clinical Trials: Rescue Criteria Can Affect Outcomes ppdi.com

AUTHORS Randy Anderson, PhD, PPD Catherine Morrow, PPD Christopher Perkins, MD, PPD

INTRODUCTION All well-controlled clinical trials of antihyperglycemic agents to treat Type 2 diabetes (T2D) that last 4 weeks or longer need a provision for rescue in study participants who have consistently and unacceptably high hyperglycemia. This type of provision, which usually involves adding an open-label antihyperglycemic agent to the study treatment, is an ethical necessity that, if well managed scientifically, can enhance the demonstration of study drug efficacy. This paper summarizes precedent hyperglycemia rescue criteria, describes the necessary and sufficient conditions required to minimize the biasing effects of rescue on efficacy and safety assessments, discusses the clinical implementation and implications of these criteria, and details exactly how hyperglycemia-related discontinuations can impact clinical trial outcomes.

PRECEDENT HYPERGLYCEMIA RESCUE CRITERIA In the context of a clinical trial, it is ethically imperative to include rescue criteria to avoid prolonged intervals of hyperglycemia that would increase the risk of adverse events related to hyperglycemia. Historically, trials that use high thresholds for hyperglycemia rescue (eg, >400 mg/dL) have had difficulty enrolling because

of investigator concerns for patient safety impact of under-treatment. This issue can be addressed by choosing lower rescue thresholds (eg, >250 mg/dL) and earlier intervention times, but, for products that have a gradual onset of treatment effect, this approach may decrease the duration of on-study treatment and impair the treatment effect estimation. The February 2008 U.S. Food and Drug Administration draft Guidance for Industry, “Diabetes Mellitus: Developing Drugs and Therapeutic Biologics for Treatment and Prevention,” recommends that all Phase II or III trials in diabetes include hyperglycemia rescue thresholds that are systematically lowered (stepped down) over the course of the study.1 FDA gives one illustrative example of appropriate criteria: • Confirmed fasting plasma glucose (FPG) >270 mg/dL from Baseline to Week 6 • Confirmed FPG >240 mg/dL from Week 6 to Week 12 • Confirmed FPG >200 mg/dL or HbA1c >8.0% from Week 12 to Week 24 The agency notes that these criteria can be modified as needed based on how quickly the study drug is expected to produce a clinically meaningful change in FPG or HbA1c, as well as the drug’s mechanism of action (eg, affecting postprandial rather than fasting glucose levels). The FDA guidance also states that participants “meeting glycemic rescue criteria ideally should remain in the study even after receiving the additional or alternative therapy to allow for the assessment of safety of the investigational drug or biologic.” The continued study follow-up is essential for preserving the benefits of the original randomization to minimize bias on the safety assessments. The statistical basis for this assertion is discussed in detail in the next section.

Managing Hyperglycemia in Clinical Trials: Rescue Criteria Can Affect Outcomes

2

NECESSARY AND SUFFICIENT CONDITIONS FOR MINIMIZING BIAS FROM HYPERGLYCEMIA RESCUE CRITERIA The following conditions define an irreducible set that must accompany implementation of hyperglycemic rescue criteria in order to preserve the scientific integrity of study findings that are based on the randomized assignment of study treatments. 1. The hyperglycemia rescue criteria must be consistently defined and implemented for all participants without regard to randomized treatment assignment. For a specific drug in development, it is also helpful to use the same provisions in all protocols to streamline the integrated analyses needed for the marketing application. 2. For glycemic control protocols in which efficacy is typically measured as the change from baseline in HbA1c, the primary efficacy endpoint must be defined a priori as the HbA1c change from baseline to end of study, which is either the time of hyperglycemia rescue or the end of the trial (eg, Week 24). This endpoint represents the best practical and ethically acceptable measurement of the anti-diabetic effect of the randomized treatment. Moreover, when the alternative hypothesis is actually true (the investigational drug is more effective than the control), this strategy optimizes the efficacy differentiation between groups. Rescue is more likely in the less-effective treatment group, and the final HbA1c value collected for rescued participants will, by definition, be high. 3. In participants who meet rescue criteria, safety surveillance, particularly for major adverse cardiac events (MACE), must continue, and treatment double-masking must be maintained until study

end. These provisions assure that any bias in safety comparisons is minimized. (eg, bias arising from introduction of non-randomized rescue treatments). 4. Hyperglycemia rescue medications should be at the investigator’s discretion but should not include any other medication in the same mechanistic class as the test product. Allowing the investigator to make an individualized decision is an important component of defining standard of care. Thus, with rescue criteria, efficacy and safety comparisons may be appropriately described as test product versus standard of care. We believe this idea is particularly cogent for studies used to evaluate the cardiovascular (CV) safety of the test product, as the CV safety guidance specifically addresses comparisons of CV risk reduction versus standard of care.

CLINICAL IMPLICATIONS OF HYPERGLYCEMIA RESCUE CRITERIA Choosing hyperglycemic rescue criteria can be challenging, as there are multiple pharmacodynamic, safety, statistical, and logistic considerations. The FDA guidance is a reasonable place to start, and the agency acknowledges that the example criteria may require modification based on the study drug’s mechanism of action and/or time to onset of effect. In our experience, approximately 50% of the total HbA1c reduction will be seen in the first 3 to 4 weeks on most therapies. If the amount of time needed to attain the pharmacodynamic effect of the investigational product is delayed, then it may be reasonable to extend some of the timeframes for rescue at various hyperglycemic levels. This issue is particularly important in trials using an active comparator that produces results more rapidly than the drug under study.

Managing Hyperglycemia in Clinical Trials: Rescue Criteria Can Affect Outcomes

3

It is also appropriate to have earlier rescue criteria focused on FPG (and confirmed by a second sample) and transition to HbA1c levels later in the trial. Also consider the change from baseline in HbA1c. For participants who enter with very poorly controlled hyperglycemia and achieve a significant drop in HbA1c (eg, greater than 2% by Week 12), it may be reasonable to allow them to continue without rescue even if their HbA1c is above the specified rescue criteria at that time point. Later in the trial (eg, Week 24), all participants should be subject to the same rescue criteria based on absolute HbA1c. The choice of rescue criteria should be accompanied by a thoughtful data analysis plan. The FDA guidance states that participants “meeting glycemic rescue criteria ideally should remain in the study even after receiving the additional or alternative therapy to allow for the assessment of safety of the investigational drug or biologic.” To facilitate unbiased safety comparisons, we believe that rescued patients should remain in the study (with rescue treatment on board) and continue randomized treatment in a double-blind fashion until the end of the study. The statistical basis for this assertion is discussed in detail in the next section. In addition, we recommend defining the primary efficacy endpoint as the HbA1c change from baseline to either (a) the time of hyperglycemia rescue or (b) the end of the trial (eg, Week 24). In the protocol or data analysis plan, we suggest detailing the planned comparisons and specifically including whether each is active drug vs placebo or active drug vs standard of care (as would be true after rescue). Analyses evaluating pre- and post-rescue adverse events and other safety comparisons are important to specify in advance of treatment unmasking. Working with Investigators Even the most thoughtful and well-described hyperglycemia rescue criteria are inadequate if not consistently applied by investigators. We recommend discussing the criteria and assessing the investigator’s comfort level

with both the thresholds and the rescue procedures before including a site in the study. If an investigator is not comfortable with the rescue criteria or procedures and consistently deviates from them, the bias introduced will make it difficult to interpret safety and efficacy results from that site. Of course, the role of detailed site training on the interpretation and implementation of the rescue criteria cannot be overestimated. In addition to safety matters, stress to investigators the importance of collecting a sample for final efficacy assessments before starting rescue medication. To aid in appropriate application of rescue criteria, it may be helpful to have the central lab “flag” results from participants who may require rescue. Additional measures, such as regular lab to data management or IVR system transfers, might be needed to verify that rescue has occurred. Finally, when more than one study is being performed (eg, during a development program), consistency in rescue criteria between the studies will also help sites (and the study team) implement the thresholds and procedures correctly. While the choice of rescue medication can be proscribed by the protocol, we have found that allowing investigators to use their own discretion in choosing rescue medications results in a more “real world” scenario that we find preferable. However, it is appropriate to exclude certain medications (eg, those with a similar mechanism of action or unapproved combinations) to protect the safety of study participants. Obtaining High-Quality Data at the Time of Hyperglycemia Rescue To maximize data precision and accuracy, we recommend using a central laboratory to determine HbA1c at the primary endpoint (either at rescue or at the end of the study). When time is critical, results from local laboratories can be used to confirm diagnoses and make patient care decisions, but pre-rescue samples should still be forwarded to the central lab for inclusion in the trial database.

Managing Hyperglycemia in Clinical Trials: Rescue Criteria Can Affect Outcomes

4

Clearly defining rescue criteria as well as the logistics for sample collection is essential for successfully implementing consistent rescue criteria. In the protocol, we suggest including a full description of the process along with a graphic of the steps and parameters for rescue. This graphic could be printed and laminated for easy study staff reference. In addition, the rescue criteria should be discussed with the physician as a part of site selection to ensure compliance with the protocol. Finally, it’s important to recognize and plan for the possibility that, despite best efforts, a few patients may be rescued outside of the protocol-defined criteria. We recommend discussing up front how to manage the data from these patients and detail your decisions in the study’s statistical analysis plan.

INCOMPLETE FOLLOW UP DUE TO HYPERGLYCEMIA-RELATED DISCONTINUATIONS: IMPACT ON TRIAL OUTCOMES Withdrawal of patients experiencing hyperglycemia undermines the randomized assignment of participants to treatment groups and thereby biases both efficacy and safety comparisons. The impact on efficacy is very direct, since the hyperglycemia that motivates early discontinuation is typically the same measure as the primary endpoint. Discontinuation precludes the collection of data that would have been measured later in the study. Thus, removing patients with high values has the effect of lowering the observed mean glycemia of those that remain. If one treatment group has more patients removed than the other, the mean treatment difference computed from the remaining data is biased toward the null hypothesis

(no difference between treatment groups). If the blood glucose-lowering efficacy of the test treatment leads to less hyperglycemia discontinuation in the test group relative to the control group, then the observed treatment group difference in HbA1c is biased toward zero (ie, no treatment effect).2 The prevalent practice for efficacy analysis with hyperglycemia rescue criteria is to carry forward the last observation (ie, the rescue value) for those who meet rescue criteria before the scheduled end of study treatment. In general, a last-observation-carried-forward (LOCF) method may not minimize bias,2 since the approach cannot reflect any trend of worsening over time. However, with consistent application of rescue criteria 1 to 4 detailed above, every instance of rescue corresponds to a durable state of unacceptably high hyperglycemia (ie, treatment failure) as defined by the rescue timing and thresholds. The LOCF rescue value is high by definition. Thus, the rescue + LOCF approach for efficacy analysis reduces bias relative to any unstructured discontinuation and resulting missing data. The impact of hyperglycemia discontinuation on safety parameters is more subtle, affecting treatment group comparisons of adverse events that are related to hyperglycemia. As an illustration, consider blood lipids and lipoproteins (eg, LDL, HDL, and VLDL cholesterol, apolipoproteins A1, B, and C3, and C-reactive protein) that are strong predictors of MACE risk. These measures tend to shift to higher cardiovascular disease (CVD) risk values with progression of T2D severity: that is, they are highly correlated with glycemic instability as measured by HbA1c or FPG. Thus, selective discontinuation related to hyperglycemia tends to result in patients with higher CVD risk factors leaving the study prematurely. Those patients who remain on study have lower CVD risk. As a result, when test treatment is better than control, early discontinuation due to hyperglycemia means that treatment group comparisons of hyperglycemia-related safety parameters become biased in favor of the control group.

Managing Hyperglycemia in Clinical Trials: Rescue Criteria Can Affect Outcomes

5

Table: Impact of Hyperglycemia Rescue Provisions on Study Outcomes

Rescue Provisions Followed

Alternative Hypothesis Is True

Null Hypothesis Is True

None

• Ethical dilemma of potential or actual under-treatment reduces participation interest and leads to higher discontinuation rate in placebo group. • More incomplete follow up in the control group biases efficacy and safety against the test product.

Rescue provision 1 only

• Rescue happens consistently as needed, removing disincentive for study participation. • Incomplete follow up remains an issue since efficacy is not measured at rescue, and safety assessments do not continue after rescue. • Treatment effect bias for efficacy and safety remain, arising from greater rescue incidence in control group.

• Incomplete follow up remains an issue since efficacy is not measured at rescue, and safety assessments do not continue after rescue, but it is balanced between treatment groups for efficacy or safety (ie, no treatment bias).

Rescue provisions 1 and 2 only

• Efficacy bias is minimized. • Incomplete follow-up for safety is greater in control group due to discontinuation at time of hyperglycemia rescue, and the safety bias remains.

• Efficacy has more complete follow up, with no bias. • Discontinuation at rescue leads to moderately underestimated event rates for adverse events (AEs) associated with hyperglycemia.

• Ethical dilemma of potential or actual under-treatment reduces participation interest and leads to high discontinuation rate. • Both treatment groups are equally affected. Incomplete follow-up is balanced between treatment groups for efficacy or safety (ie, no treatment bias), but overall disease effect estimates are biased toward less severe impact.

(Table continued on page 7)

Managing Hyperglycemia in Clinical Trials: Rescue Criteria Can Affect Outcomes

6

Table: Impact of Hyperglycemia Rescue Provisions on Study Outcomes (continued)

Rescue Provisions Followed

Alternative Hypothesis Is True

Rescue provisions 1, 2, and 3 only

• Efficacy bias is minimized. • Safety bias is minimized, since follow up remains balanced even with different rates of rescue between treatment groups. • Potential for the effect of rescue medications to be confounded with treatment effect if the choice of rescue medication includes other products in the same class or if a specific rescue medication is mandated.

Rescue provisions 1 to 4

• Efficacy bias is minimized. • Safety comparisons interpretable as test product versus standard of care defined by investigators’ choices of rescue medications.

Null Hypothesis Is True

• More complete follow-up of both efficacy and safety endpoints, with equal rates of rescue in each treatment group.

• Expected equal rescue incidence in both treatment groups and investigator preference for rescue medication means that overall estimates reflect standard of-care treatment.

Provision 1 = hyperglycemia rescue criteria consistently defined and implemented Provision 2 = primary efficacy endpoint defined as change from baseline to end of study (either the time of hyperglycemia rescue or the end of the trial) Provision 3 = participants meeting rescue criteria continue in the trial and the treatment remains blinded Provision 4 = choice of rescue medications is at the investigator’s discretion

Managing Hyperglycemia in Clinical Trials: Rescue Criteria Can Affect Outcomes

7

The choice of rescue criteria will impact their effect on study outcomes. Using higher thresholds might appear to limit bias because fewer participants will qualify for rescue, but that decision tends to make the trial less attractive to investigators and participants because of the safety concerns related to long-term hyperglycemia. In studies that do accrue, participants with consistent hyperglycemia may withdraw from the trial (or be withdrawn by the investigator) before meeting the rescue criteria and simply cite a different reason for discontinuation (eg, consent withdrawn, investigation discretion, lack of efficacy). Clearly, this action can create bias in both the efficacy and safety results. Choosing lower thresholds is likely to increase the number of participants qualifying for hyperglycemic rescue, which may undermine the statistical assumptions used in power and sample size calculations. In addition, in trials where the investigational drug is effectively lowering glucose, the higher rates of rescue among patients in the control group will shift the characteristics of the population remaining in the study and skew the results.

SUMMARY Long-term hyperglycemia may increase the risk of adverse events, including damage to the cardiovascular, renal, and neurologic systems. As a result, all well-controlled clinical trials of T2D treatments lasting ≥4 weeks should have protocol-specified criteria for the rescue in study participants who have consistently and unacceptably high hyperglycemia. This type of provision is an ethical necessity that, if not well managed scientifically, can significantly bias the results of the trial. Fortunately, adherence to a set of conditions regarding hyperglycemia rescue can preserve the scientific integrity of study findings that are based on the randomized assignment of study treatments.

Managing Hyperglycemia in Clinical Trials: Rescue Criteria Can Affect Outcomes

8

REFERENCES Food and Drug Administration. Guidance for Industry. Diabetes Mellitus: Developing Drugs and Therapeutic Biologics for Treatment and Prevention. Draft. February 2008.

1

White IR, Bamias C, Hardy P, et al. Randomized clinical trials with added rescue medications: some approaches to their analysis and interpretation. Statist Med. 2001;20:2995-3008. 2

Managing Hyperglycemia in Clinical Trials: Rescue Criteria Can Affect Outcomes

9

For more information, please visit us at www.ppdi.com or call us at +1 877 643 8773.

© 2013 Pharmaceutical Product Development, LLC. All rights reserved.