Pharm Med (2017) 31:113–118 DOI 10.1007/s40290-016-0176-0

ORIGINAL RESEARCH ARTICLE

Quantitative Methods for Safety Monitoring of Rare Serious Adverse Events Susan P. Duke1 • Christi Kleoudis2 • Margaret Polinkovsky2 • Dimitri Bennett3 Deanna Hill3 • Eric Lewis1

•

Published online: 25 January 2017  Springer International Publishing Switzerland 2017

Abstract Background Rare but serious adverse events are often reasons for the modification of drug product labelling, termination of further development and even withdrawal of a treatment from the market. Objective Our objective was to use analytical methods to evaluate adverse events of special interest to aid in the determination of whether a particular treatment increases risk above what would be expected in the target population. These methods could be used during drug development and in the post-marketing setting as a supplement to the program safety analysis plan for compounds that have potential for rare serious adverse events.

Electronic supplementary material The online version of this article (doi:10.1007/s40290-016-0176-0) contains supplementary material, which is available to authorized users. & Eric Lewis [email protected] 1

Global Clinical Safety and Pharmacovigilance, GlaxoSmithKline, Research Triangle Park, NC 27709, USA

2

Statistics and Programming, GlaxoSmithKline, Research Triangle Park, NC, USA

3

Worldwide Epidemiology, GlaxoSmithKline, Collegeville, PA, USA

Methods Two well-known statistical methods were used in a new way by a cross-functional internal matrix team of safety physicians, scientists, epidemiologists and statisticians to compare background adverse event rates in a specific disease population with the observed rate in clinical trials to date. If the clinical trial rate is similar to the background rates, one can surmise that the study drug has not increased the risk of the event. Method 1 uses binomial probabilities to calculate the probability of observing the event; method 2 uses incidence rates to assess risk. To illustrate our methods, we evaluated two compounds with immunosuppressive characteristics for cases of progressive multifocal leukoencephalopathy or herpes zoster due to reactivation of the varicella zoster virus. A literature search was used to help determine background rates of these adverse events in the populations of interest. Results For method 1, data are presented in tabular form to show the estimated probability of observing one or more cases of progressive multifocal leukoencephalopathy, assuming a background rate of 0.4% as a result of the disease and 500 subjects exposed to the study medication. For herpes zoster, a background rate of 0.32 per 100 patient-years was predicted from the literature and steps to assess the likelihood of the incidence rates occurring by chance are shown in tabular form. Conclusions These analytical tools may contribute to a better understanding of the association of rare serious adverse events with an approved or experimental compound by helping distinguish rates related to the drug vs. that of the underlying disease or other factors. These methods can aid the drug safety professional by providing a quantitative context of a medicine’s benefit-risk profile during drug development and the post-marketing setting.

114

Key Points Embedding benefit-risk thinking and planning and using a quantitative framework to evaluate a serious adverse event for use by the drug safety professional to more clearly understand whether the drug increases the event rate are critical Similar to how drug efficacy is quantified, the use of quantitative methods, such as the ones presented, aid the drug safety professional by providing statistical clarity in population-based safety concerns, such as the situation where both the disease and drug have a potential association. This additional clarity is valuable for internal decision-making bodies and external filings assessment, in alignment with industry best practices

1 Introduction There are many challenges during drug development where safety-related issues continue to be a major cause of drug attrition in preclinical and clinical phases [1]. There is a need to improve and enhance our understanding of safety issues during the clinical development process as well as the post-approval environment. One of the major challenges in drug safety and drug development overall is gaining clarity on the association between rare serious adverse events (SAEs) and the test treatment being studied. One common question is whether a given adverse event (AE) rate increases with the use of the medication under development (or on the market) or does it remain at the background rate of the underlying disease? When investigating potential causes of a safety concern, some causes can be discerned at the individual case level, e.g. concomitant interventions, medical history and patient choices. Other concerns are evaluated at the population level to understand the role of the study drug and the underlying disease. A quantitative comparison of event rates in a treated population with disease background rates can aid in clarifying whether the addition of the study drug is associated with an increased safety risk. To illustrate our methods, we present two AEs of special interest. First, the SAE of progressive multifocal leukoencephalopathy (PML), which is a rare infection of the central nervous system associated with conditions that suppress the immune system. First described in 1958, PML was reported in patients with hematologic malignancies and then rheumatologic diseases [2, 3]. With the advent of human immunodeficiency virus infection, it also was

S. P. Duke et al.

recognized as a common presenting feature of this acquired immunodeficiency disorder [4]. As more novel immunosuppressive therapies were developed, PML became an AE of significant concern for physicians and patients. PML was first described shortly after the approval of Tysabri (natalizumab) for relapsing–remitting multiple sclerosis in 2005 [5, 6]. It has also been associated with other immunosuppressive agents. Symptoms of PML include cognitive deterioration, confusion, difficulty with walking and coordination, hemiparesis, limb paresis and visual changes, although symptoms can vary depending on the region of the brain involved [3, 7, 8]. As there are no symptoms specific only to PML, it is often difficult to distinguish the disease from multiple sclerosis, neuropsychiatric systemic lupus erythematosus (SLE) or central nervous system lupus, and isolated central nervous system lupus vasculitis or cerebral vasculitis [9]. Additionally, PML can present as a seizure, stroke or headache (i.e. ictally), but develops progressively over a period of weeks to months. Given the high degree of morbidity and mortality associated with PML, it is understandable that regulatory authorities, sponsors and the public have a heightened concern for this SAE. PML serves as an excellent example for the occurrence of rare but significant AEs associated with the development of immunosuppressive treatments. While PML serves as an example for considering analytical approaches to rare events, there are many other applications. Given the unique nature of each drug development program and the generalisability of analytical thinking, it is reasonable to show an example from another disease area, for example, herpes zoster from infectious diseases. Herpes zoster, commonly known as shingles, is caused by the reactivation of the varicella zoster virus with the overall incidence of zoster among immunocompetent subjects between 1.2 and 4.8 per 1000 people per year; it has been noted that the incidence of zoster increases markedly with age [10]. The clinical presentation is a unilateral vesicular dermatomal rash, which is usually accompanied by radicular pain along that dermatome. Patients experience significant pain and distress that may last several months; thus, this pain affects the daily lives of patients, impacts their emotional health and results in poor quality of life. It has been indicated that the most common complication of zoster is post-herpetic neuralgia, which develops in approximately 20% of individuals aged C50 years [10]. Analytic methods aid in the determination of whether a particular treatment increases risk above what would be expected as a background rate in the treated population. Safety methods are standardly described in the program safety analysis plan [11, 12]. We recommend the following analyses for compounds that have potential for rare SAE risks to numerically clarify whether SAE events are

Quantitative Framework to Evaluate Adverse Events

associated with the drug under evaluation, underlying disease or concomitant medications that are part of the standard of care [13]. Therefore, a cross-functional internal matrix team of safety scientists representing the functional areas of physicians, epidemiologists and statisticians collaborated with the objective to describe methods undertaken to contextualise the risk of a rare SAE, PML and a more common and less serious AE, herpes zoster, as examples.

2 Methods As our first initial step in this quantitative research, we conducted a systematic literature review where we assessed the risk associated with SLE and the medications used to treat PML, with particular emphasis on concomitant exposure to immunosuppressant medications [14]. This was followed by a similar search to evaluate the more common and less serious AE, herpes zoster, in an infectious disease. Following this, two complementary statistical methods were used to compare background rates in a specific disease population with the observed rate in clinical trials to date. If the clinical trial rate is similar to the background rates, one can surmise that the study drug has not increased the risk of the event. 2.1 Method 1: Probability of Observing an Event Binomial probabilities can be calculated to determine the probability of observing one or more events given the expected background rate in the disease population of interest. This calculation can be especially useful in singlearm studies or compassionate use programs for which there is no internal control. 2.2 Method 2: Risk Assessment Using Incidence Rates The second method, which calculates incidence rates, is useful when determining whether the incidence observed in a controlled trial is expected given the background rate. It can be particularly helpful when the exposure between the treatment and control groups is not balanced. Ideally, one would want to use an exact Poisson distribution for calculating incidence rates, but when dealing with rare events, the low number of events can create computing difficulties in some software packages. Therefore, the exact binomial distribution (as an approximation to the exact Poisson distribution) was used to calculate the incidence rate [15]. Details of this method are described in the Supplementary Digital Content.

115

3 Results 3.1 Probability of Observing an Event In the comprehensive literature search, we reported that while PML was a very rare disease in SLE patients, it does appear that there is an increased risk of PML associated with SLE compared with the general population, potentially owing to immunosuppression, other contributing factors in their underlying disease, treatments prescribed to manage disease or some combination of these factors. For a drug used to treat SLE, one can determine the probability of observing one or more cases of PML in 500 patients exposed and an estimated background rate from the literature, by calculating the binomial probabilities under several scenarios, as illustrated in Table 1. The background rate, 0.4% [16], was chosen based on the safety physician’s and epidemiologist’s understanding of the literature for SLE. In our case, approximately 500 patients had been exposed. A similar table can be calculated for other levels of patient exposure and/or background rate. Depending upon the dataset, this approach may or may not account for the period of observation. In this example, the period of observation has not been accounted for because of the nature of the available information. Figure 1 shows this probability graphically (vertical axis) for 500 patients exposed for a range of possible numbers of cases (x-axis). The Log scale is used because the risk range graphed is quite large. The x-axis is labelled as the number of cases per 500 patients exposed and this is the information a drug safety professional would need if, with 500 cases, another case arose. Note the 0.4% rate highlighted in the figure is identical to the estimates in Table 1. Table 1 and Fig. 1 can serve as a means for safety and clinical development teams to determine whether the association of SAEs is from the disease (assumed if Table 1 Estimated probability of observing one or more cases of PML, assuming a background rate due to the disease of 0.4% (for SLE background rate) and 500 patients exposed to the study medication Number of cases

Probability (%)

At least 1

&87

At least 2

&59

At least 3

&32

At least 4

&14

At least 5

&5

At least 6

\5

PML progressive multifocal leukoencephalopathy, SLE systemic lupus erythematosus

116

S. P. Duke et al.

Fig. 1 Probability of observing the specified number of cases in 500 subjects exposed. See the main text for details on how the rate of 0.4% was chosen for progressive multifocal leukoencephalopathy (PML). Note there is a [80% chance of seeing at least one PML case in 500 patients exposed at a 0.4% estimated risk rate, and a \5% chance of seeing six cases. This provides context to the safety clinician when additional PML cases arise

Table 2 Steps to assess the likelihood of the incidence rates occurring by chance Step

Assessment

Control N = 703 Exposure = 703.8 patient-years

Test N = 2012 Exposure = 5799 patient-years

Total N = 2715 Exposure = 6502.8 patient-years

1

Observed number of events in clinical study data

0

7

7

2

Expected number of events per 100-patient-years given a background IR of 0.32/100 patient-years

2.25

18.56

20.81

3

Risk of event [IR (95% CI)] based on observed events in clinical study data per 100 patient-years

0 (0–0.524)

0.121 (0.049–0.249)

0.108 (0.043–0.222)

4

Probability of observing 7 events with 0 events in the control group and 7 events in the test group

0.449

5

Expected number of events/100 patient-years based on IR for total clinical study population (0.108)

0.758

6.242

7

CI confidence interval, IR incidence rate

expected and observed rates are very similar) or increased by administration of the study drug. 3.2 Risk Assessment Using Incidence Rates In this example, a drug development program observed seven cases of herpes zoster, of which zero were in the control group (703.8 patient-years of exposure) and seven were observed in the test group (5799 patient-years of exposure). A review of the published literature in the population of interest revealed a background rate of 0.32/ 100 patient-years (Table 2). The observed result falls within the realm of what is expected (Step 5) when the internal rate (0.108) within the

clinical trial is applied, and is lower than what is expected (Step 2) when the background rate (0.32) for the disease is applied. The apparent difference in events (0 vs. 7) appears to be driven by the difference in the exposure between the two arms and not the test treatment.

4 Discussion In our quantitative research, we present a systematic approach to safety monitoring that helps characterise SAEs in the context of background disease risk. This can be applied in both the clinical development and the postmarketing setting. We have found that the practical benefit

Quantitative Framework to Evaluate Adverse Events

of these objective tools is optimised when applied by a cross-functional team of safety clinicians, epidemiologists and statisticians. We encourage others to consider similar collaborative efforts across these disciplines. This approach to the interpretation of rare events is not intended as a method of signal detection. Potential safety signals are received both during drug development and from post-marketing sources. It is not infrequent that a potentially alarming event is reported in the context of a clinical development only to be later determined to be related to underlying disease or pathological processes that are distinct and separate from the initial concern. This recommended approach can assist teams during the conduct of safety review meetings to contextualise potentially significant rare (or common) events and further assist with their analysis and interpretation. Whether an approach such as this will impact go-no-go decisions, the progression of a drug’s development, the filing of an new drug application or market withdrawal will depend upon the strength of the signal of concern. The major limitations to this approach are not statistical but rather relate to the uncertainty of the underlying rate of occurrence for an AE of interest in a population. 4.1 Probability of Observing an Event To provide a sense of risk rates to be expected in the clinical trial setting, Fig. 1 shows risk rates for 500 cases exposed. Important factors to consider in determining what rate is reasonable for this disease and the patient population are: • • •

•

How certain are the estimates for the occurrence of PML in the population of interest? What are the data sources for the estimate of the underlying rate? Given the nature of the disease under study, what would be considered an acceptable increase in the risk of PML (an acceptable change from the background rate) given the projected benefit of treatment? Consider the duration of observation: e.g. the PML rate would be higher in a dataset that encompasses 5 years of observation for a drug under study compared with 0.5 years of observation in the control arm.

While exposure was expressed per unit of time (e.g. patient-years of exposure) in our example, other agents (particularly oncology drugs) express exposure in terms of cycles. For agents that have a durable pharmacodynamic effect, it may be helpful to display the impact on a key biomarker(s). It is important to consider that risk may increase as exposure increases [17]. The methods presented here do not account for increasing risk over time exposed.

117

4.2 Risk Assessment Using Incidence Rates When exposure is unbalanced between treatment arms, such as the case with chronic diseases with an openlabel long-term extension, comparing observed incidence rates to expected rates reveals that the observed number of events within both the test and control arms were as expected using the internal clinical trial rate and lower than expected compared with the background rate for the disease. While this would be obvious if exposure rates between the test and placebo were 1:1, this methodology provides the safety scientist with the same level of discernment when the ratio is unbalanced (in this example, 7:1). Crowe et al. have described for the layperson the inaccuracy of ‘crude pooling’ when multiple studies are involved and the exposure ratio is unbalanced [18]. 4.3 Limitations A limitation of both methods is their reliance on the point estimate of the background rate. Uncertainty is not directly addressed; therefore, a sensitivity analysis using a range of reasonable rates for the disease and AE of interest is recommended. The population under study may be different and/or have an inherently different risk for the AE of interest owing to the clinical history or treatment with certain medications. Further study in the application of quantitative methods to drug safety is warranted. Much more has been done in the quantification of drug efficacy than drug safety. Given the value of these simple methods described for contextualising background rates for AEs of interest, there is potential for much more to be achieved for patient safety from the collaboration of clinical and quantitative scientists.

5 Conclusions Safety monitoring is by its nature numerically oriented, lending itself to quantitative methodologies. As safety monitoring of ongoing clinical trials continues to mature, it is expected that other applications of common statistical methods to safety monitoring will also emerge. The two methods presented here can be valuable tools for understanding whether a drug is associated with an increased incidence of AEs relative to the background rate for the underlying disease. The intent of this article is to apply some well-known statistical methods in a novel way for drug safety. Our multi-disciplinary discussion concluded there has to be

118

something more we can do quantitatively for drug safety professionals in support of their need to contextualise background AE rates. Acknowledgements We thank our co-collaborators on this project, Jonathan Haddad, Sarah Kavanaugh, Barbara Haight, Lorrie Schifano, Marilyn Metcalf and Qiming Liao, for their insights and support. Compliance with Ethical Standards Funding This work was sponsored by GlaxoSmithKline. Conflict of interest Eric Lewis is a current employee of GlaxoSmithKline and Susan P. Duke, Christi Kleoudis, Margaret Polinkovsky, Dimitri Bennett and Deanna Hill were employees of GlaxoSmithKline and completed this research as part of their employment with GlaxoSmithKline. Currently, Susan P. Duke is with Statistical Sciences, AbbVie; Christi Kleoudis and Margaret Polinkovsky are with Biostatistics, PAREXEL; Dimitri Bennett is with R&D Data Science Institute, Informatics, Pharmacoepidemiology, Takeda; and Deanna Hill is with the Center for Observational and Real-World Evidence, Pharmacoepidemiology, Merck & Co.

References 1. Arrowsmith J, Miller P. Trial watch: phase II and phase III attrition rates 2011–2012. Nat Rev Drug Discov. 2013;12(8):569. doi:10.1038/nrd4090. 2. Richardson EP Jr. Progressive multifocal leukoencephalopathy. N Engl J Med. 1961;265:815–23. doi:10.1056/ NEJM196110262651701. 3. Astrom KE, Mancall EL, Richardson EP Jr. Progressive multifocal leuko-encephalopathy; a hitherto unrecognized complication of chronic lymphatic leukaemia and Hodgkin’s disease. Brain. 1958;81(1):93–111. 4. Richardson EP Jr. Progressive multifocal leukoencephalopathy 30 years later. N Engl J Med. 1988;318(5):315–7. doi:10.1056/ NEJM198802043180510. 5. Kleinschmidt-DeMasters BK, Tyler KL. Progressive multifocal leukoencephalopathy complicating treatment with natalizumab and interferon beta-1a for multiple sclerosis. N Engl J Med. 2005;353(4):369–74. doi:10.1056/NEJMoa051782.

S. P. Duke et al. 6. Langer-Gould A, Atlas SW, Green AJ, et al. Progressive multifocal leukoencephalopathy in a patient treated with natalizumab. N Engl J Med. 2005;353(4):375–81. doi:10.1056/ NEJMoa051847. 7. Boothpur R, Brennan DC. Human polyoma viruses and disease with emphasis on clinical BK and JC. J Clin Virol. 2010;47(4):306–12. doi:10.1016/j.jcv.2009.12.006. 8. Lima MA, Drislane FW, Koralnik IJ. Seizures and their outcome in progressive multifocal leukoencephalopathy. Neurology. 2006;66(2):262–4. doi:10.1212/01.wnl.0000194227.16696.11. 9. Calabrese LH, Molloy ES. Progressive multifocal leucoencephalopathy in the rheumatic diseases: assessing the risks of biological immunosuppressive therapies. Ann Rheum Dis. 2008;67 Suppl 3:iii64–5. doi:10.1136/ard.2008.097972. 10. Forbes HJ, Thomas SL, Langan SM. The epidemiology and prevention of herpes zoster. Curr Derm Rep. 2012;1:39. doi:10. 1007/s13671-011-0004-4. 11. Crowe BJ, Xia HA, Berlin JA, et al. Recommendations for safety planning, data collection, evaluation and reporting during drug, biologic and vaccine development: a report of the safety planning, evaluation, and reporting team. Clin Trials. 2009;6(5):430–40. doi:10.1177/1740774509344101. 12. Xia HA, Crowe BJ, Schriver RC, et al. Planning and core analyses for periodic aggregate safety data reviews. Clin Trials. 2011;8(2):175–82. doi:10.1177/1740774510395635. 13. Hill D, Eudy A, Bennett D, et al. Benefit-risk analysis and management plan for rare serious adverse events (SAEs): progressive multifocal leukoencephalopathy (PML) case example (abstract). Pharmacoepidemiol Drug Saf. 2015;24(Suppl. S1):1–587. doi:10.1002/pds.3838. 14. Henegar CE, Eudy AM, Kharat V, et al. Progressive multifocal leukoencephalopathy in patients with systemic lupus erythematosus: a systematic literature review. Lupus. 2016;25(6):617–26. doi:10.1177/0961203315622819. 15. Stokes ME, Davis CS, Koch GG. Categorical data analysis using SAS. 3rd ed. SAS Institute; 2012.Cary, NC. 16. Amend KL, Turnbull B, Foskett N, et al. Incidence of progressive multifocal leukoencephalopathy in patients without HIV. Neurology. 2010;75:1326. doi:10.1212/WNL.0b013e3181f73600. 17. Tysabri (natalizumab) prescribing information. Cambridge: Biogen, Inc.; 2016. https://www.tysabri.com/content/dam/ commercial/multiple-sclerosis/tysabri/pat/en_us/pdfs/tysabri_ prescribing_information.pdf. Accessed 13 Jul 2016. 18. Crowe B, Chuang-Stein C, Lettis S, Brueckner A. Reporting adverse drug reactions in product labels. Ther Innov Regul Sci. 2016;50(4):455–63. doi:10.1177/2168479016628574.