Lifecycle Modeling and Simulation in Clinical Trials

Computer-based modeling and simulation has advanced numerous industries, from aeronautics and engineering to meteorology and finance. Its potential benefits in drug discovery and development have been recognized for decades, but full realization of modeling and simulation in the health sciences has been limited by the vast complexity of biological systems, lack of understanding of disease, lack of large-population data on real-world health outcomes, and uncertainty regarding regulatory acceptance of modeling and simulation applications in clinical drug evaluation. These barriers are gradually being overcome, and modeling and simulation is now poised to transform the entire drug development lifecycle, from discovery to commercialization.¹

Modeling and simulation (M&S) practice has evolved to solve complex problems that could not be addressed using direct observation and measurement. Models are built using historical observations to describe behaviors observed within systems. Models are commonly used to predict a future outcome and can be either deterministic or probabilistic (stochastic). Simulations use models to test how variability within a system can impact outcomes. Simulations can use more extreme model inputs than have been observed to help characterize the range of potential outcomes. In this way, simulation can be used to better understand risk and identify opportunities to improve outcomes. 

In clinical drug development, M&S is used to quantified problems and test assumptions as a means to improve decision-making and increase predictability. Emerging applications are now being used to predict drug safety and efficacy, to plan individual trials and Phase I to III development programs, and to better manage research portfolios.² These applications offer a glimpse of the biopharmaceutical industry’s modeling and simulation-informed future-more focused communication among development experts, greater efficiencies, and higher success rates. M&S will foster a knowledge-based drug development process that creates more value for patients, payers, and healthcare providers.¹

Benefits: More predictability, better decisions 

Predictability is the fundamental challenge of drug development. In the Food and Drug Administration’s (FDA) 2004 report on Critical Path needs and opportunities, regulators called for an aggressive, collaborative effort to create a new generation of predictive tools that could reduce costly development failures: “As biomedical knowledge increases and bioinformatics capability likewise grows, there is hope that greater predictive power may be obtained from in silico (computer modeling) analyses.”³

Computer-based predictive models are essential tools to increase clinical trial efficiencies and probability of success in what FDA envisions as “model-based drug development.” In model-based development, pharmaco-statistical models of drug efficacy and safety are built from available preclinical and clinical data with expert opinion to determine M&S objectives and inputs for simulation. Models are used to simulate scenarios of predicted relationships between drug exposure, drug response, and patient outcomes. 

M&S informs decision-making by leveraging early data to guide downstream decisions and strategies. When practiced across the full development lifecycle, M&S drives more efficient ways of working based on wider applications of data and on communication across disciplines and operations. It places greater reliance on evidence rather than on assumptions; where assumptions are necessary, M&S predicts their implications across a range of assumptions.

To implement well-informed modeling and simulation, it is necessary to eliminate traditional research silos that inhibit information sharing. M&S practice encourages closer collaboration among development experts from

different disciplines. Their combined knowledge ensures that M&S assumptions are reasonable and results are interpreted correctly. By fostering collaboration, M&S adds value by leveraging knowledge across the full drug innovation lifecycle.

Benefits in PK/PD dose modeling. The most mature practice and best current example of the value of M&S in drug development is pharmacokinetic/pharmacodynamic (PK/PD) dose modeling.⁴ This M&S application, which is heavily dependent on cross-functional collaboration, has dramatically improved dose determination-increasing predictability and reducing time and cost by leveraging data as it accrues, from preclinical to Phase III studies.

By clarifying a drug’s exposure-response relationship, PK/PD models can be used to predict optimal dosing regimens for patient testing, to provide insight on endpoints, and to test a range of assumptions about clinical outcomes. Population PK/PD analyses help identify dose adjustments for special populations such as children, the elderly, ethnic groups, patients with impaired renal/hepatic function, and patients likely to experience drug-drug interactions. They do so by better understanding patterns in the exposure-response relationship and their variability.

Enabling knowledge-based decisions: Trial design simulations. M&S facilitates knowledge-based decision-making by quantifying problems and providing a basis for discussion and assessment in a multifunctional team, particularly when there is uncertainty about the safety or efficacy profile of a therapy. For example, one of the authors modeled a number of different strategies that might be used to gain regulatory approval for a treatment indication pertaining to the central nervous system. Each strategy included interdependent Phase IIa, IIb, and Phase III studies. Success or failure at each stage was simulated for a range of plausible assumptions about safety and efficacy, based on evidence from preclinical and early clinical results. The simulated “what if” scenarios helped experts to evaluate each possible strategy. These simulations gave rise to fruitful interdisciplinary discussions about the assumptions that should be used, both pessimistic and optimistic. In addition, they provided an easily shared basis for clinicians, statisticians, regulatory affairs experts, and health economists to work together to make informed, knowledge-based choices for the development plan. Here, the value of M&S was in fostering informed discussion about the risks and benefits of each strategy among diverse experts, rather than in identifying the “best” design.

Managing complexity: Predicting market performance. Modeling and simulation can improve commercial decision-making by taking into account more of the complex variables inherent in the post-marketing landscape. Effective management of a biopharma company’s development portfolio requires assessment of therapeutic benefit, safety, commercial competitiveness, and cost-effectiveness-assessments that change as each drug candidate advances through development. Unpredictable development timelines and costs, regulatory changes, and evolving market conditions pose additional uncertainties. M&S can provide more reliable predictions of a product’s market potential by addressing complex commercial questions. For example, compared to existing products, how much better do therapeutic outcomes have to be for a drug to gain formulary acceptance?⁵

Creating a statistical model of a portfolio allows developers to simulate various changes to the portfolio and their resulting impact on the likelihood that the portfolio will support business objectives. Simulations can be used to evaluate

portfolio tradeoffs through the probabilistic representation of key performance metrics, such as time in development phase and development costs. M&S applications in portfolio management are now being used to understand the impact of tradeoffs (i.e., whether or not to add a risky development project to an existing portfolio) in the context of total portfolio risk and return. “What if” scenarios can help developers consider the probability that a company will meet a specific revenue threshold in a given year, or launch a target number of new products in a given timeframe. Figure 1 shows an analysis of varying development timelines on commercial impact.⁶

Drivers and barriers

Regulatory endorsement is a major driver of the expanding use of pharmacometric (quantitative pharmacology) models to simulate relationships between drug exposure, drug response, and individual patient characteristics. FDA’s 2009 Guidance for Industry: End-of-Phase 2A Meetings encourages sponsors to seek regulatory meetings to discuss quantitative modeling and trial simulations to improve dose selection.⁷ In cardiovascular safety evaluation, the FDA’s Office of Clinical Pharmacology often recommends concentration-QT (C-QT) modeling as a means of better evaluating drug potential for QT interval disturbance.⁸

Impact on approvals and labeling. A 2011 review conducted by FDA’s three-year-old Division of Pharmacometrics found a dramatic increase in both the number of reviews with pharmacometric analyses and their impact on FDA’s drug approval and labeling decisions.⁹ For 198 applications submitted between 2000 and 2008, pharmacometric analyses contributed to approval decisions in 126 applications. In 50% of these applications, M&S provided either pivotal or supportive insights into effectiveness; in 43%, M&S provided pivotal or supportive insights into safety. Pharmacometric analyses impacted labeling decisions in 133 applications. In 41% of those, M&S informed dosage and administration labeling, and in 14%, M&S contributed to safety labeling.

The review points to additional uses of pharmacometric models. FDA is using M&S to select pediatric dosing regimens and to approve drug dosages not studied in Phase III trials. There is also “an increasing trend toward use of model-based primary endpoints in pivotal trials, such as slope in a dose (exposure)-response model.”⁴ The report notes that five of the 198 new drug application (NDA) submissions just discussed used such model-based endpoints. Two of these submissions were indicated for pediatric epilepsy, and the primary endpoints used were slopes of the dose-response relationship and exposure-response relationship in the reduction of seizure frequency.

Another notable trend is toward the application of M&S as a pathway to regulatory acceptance of a single Phase III trial, plus a causal evidence model for demonstration of drug safety and efficacy. A provision of FDA's Modernization Act of 1997 (section 115a) allows new drug approval based on data from one adequate and well-controlled investigation, plus confirmatory evidence.¹⁰ This model has received increasing attention as a more rational, efficient, and informative approach to clinical development.¹¹

Barriers: Time, cost, expertise. There are a number of barriers to the broader application of lifecycle M&S. Despite the compelling value of M&S in clinical research and regulatory review, newer M&S applications still pose uncertainties and industry has been slow to adopt applications that go beyond PK/PD modeling. Organizations tend to accept widely understood approaches, and M&S is still unfamiliar in many applications. Thorough and careful pre-specification is often required by regulators, especially for game-changing uses of M&S in new drug applications. 

M&S also requires special expertise and new ways of working. There are economic barriers as well. M&S takes time to develop and adds cost to lean research budgets.¹²The bottom line for industry is: Will an investment in M&S pay off in terms of greater predictability? The following discussion provides examples of emerging M&S development applications and their benefits.

Applications in clinical trial design

There is growing acceptance of M&S to inform clinical trial design. Simulated trials are used to “test-run” various designs; results predict likely outcomes for a range of assumptions pertaining to dose, trial size, and operational considerations.

Creating virtual patients. Virtual patients can be created at the individual level using health records, PK/PD data, and historic data. Patient attributes also can be added to virtual patients to describe behaviors, such as adherence to treatment and other study procedures. These virtual patients can then be enrolled based on inclusion/exclusion criteria for clinical trial simulations.

Example: Informing enrollment criteria. Archimedes Inc. recently conducted a simulated trial to help researchers define inclusion/exclusion criteria and gain information for powering a diabetes study.¹³ FDA regulations for evaluation of new type 2 diabetes treatments require that Phase II and III trials include patients at higher cardiovascular risk. The drug developer needed to know the expected cardiovascular event rates in various patient populations being considered for enrollment in the intervention and control arms of the trial. Archimedes modeled 25,000 type 2 diabetes patients, including subpopulations of interest, and simulated a five-year trial that predicted cardiac events in control and intervention study arms. The simulated trial generated actionable information on expected trial outcomes, relative contribution of criteria to expected cardiac event rates, identification of optimal subpopulations, and effects of variations in the trial’s protocol on outcomes.

Modeling disease progression. Models of disease progression describe the untreated effect on patients over time. Once disease progression models are established, they can be used to tailor therapy, to evaluate and compare the action of various treatments, and to make more sophisticated tests of equivalence between treatments.

Example: Disease progression model in biosimilar development. One example comes from a study conducted by Novartis and presented at a joint EMA-industry workshop in London in November 2011.¹⁴ The aim was to test for the equivalence between a biologic treatment for rheumatoid arthritis and their candidate biosimilar. The usual test for equivalence is conducted at a single time point. Instead, the Novartis team used historic data to build a model of disease progression over 24 weeks in patients treated with the biologic. Novartis has proposed a new test of equivalence that, in the final assessment of equivalence at 24 weeks, uses the modeled progression over the entire study. Their simulations suggest that this new test of equivalence has better sensitivity than the traditional test, which results in savings of 40% in sample size for the planned study. 

Clinical trial execution models. Operational models also are important in helping to optimize the drug development process. Statistical models have significantly improved efficiencies with accurate predictions of likely recruitment¹⁵ and risk-based scheduling of the distribution of study treatments. Operational models also help to predict workflow peaks and, in the case of event-driven studies, to predict time to the end of the study.¹⁶

Clinical trial simulation. Testing various trial designs in silico before running the actual study is a very efficient method to improve the likelihood of a

successful study and to reduce risk to patients. Many of the aforementioned models are included in clinical trial simulation to help answer study design questions and investigate various assumptions. Clinical trial simulation provides the means to test multiple scenarios, to predict the potential study outcomes for each, and to quantify the risks and benefits of each design.

Example: Dose determination in trial design. An interesting example of M&S utility in trial design comes from a recent Quintiles project aimed at developing a pain treatment. As various designs were being considered, it became clear that pain tolerability would be a critical factor in whether patients would remain in the study. Unless the experimental drug relieved pain early in the course of treatment, a high dropout rate could make the study unfeasible. Early development PK/PD models were used to design a dosing regimen that included a loading dose to achieve pain relief on day one, as opposed to day seven as the sponsor had anticipated with the traditional dosing. In the actual trial, patients experienced early pain relief consistent with simulation results.

In another example of M&S utility in dose determination, simulations were used to assess dosing regimen adjustments of piperacillin/tazobactam in obese patients with varying degrees of renal function (categorized via creatinine clearance) and antimicrobial susceptibility (categorized via minimum inhibitory concentration). Population PK models, based on data from previous studies in normal patients, were used to simulate clinical trials in the obese population with differing degrees of renal function to evaluate proposed new dosing regimens. Obese patient demographics used in the simulation were sampled, with replacement, out of a healthy obese study dataset to maintain realistic demographic correlations within the simulated patients. Results of the clinical trial simulation, as illustrated in Figure 2, showed that the probability of attaining the target minimum inhibitory concentrations were similar in normal and obese patients, taking

renal function into account, across all regimens (in both 30-minute infusion and extended 4-hour infusion regimens), and across categories of antimicrobial susceptibility (minimum inhibitory concentration from 16 mg/L to 64 mg/L). From the simulations, it was also possible to conclude that no weight-based dosing adjustment was necessary in the obese population. The simulations also confirmed that extended-infusion regimens could be used, and potentially preferred, in both normal-weight and obese individuals.¹⁷

Applications in program design and trial strategy

M&S is now being used to design strategies for full research programs. M&S can help sponsors improve success rates by predicting and comparing the likely consequences of various strategies. These data-driven scenarios are used to guide research choices and go/no-go decisions at critical development stages.

Example: Improving go/no-go decisions. In a recent Quintiles project, a sponsor was considering development of a compound that had shown benefit in protecting animals against damaging biochemical effects resulting from traumatic brain injury, but where the benefit and side effects for humans were uncertain. The question was: Should the sponsor proceed with clinical development, given what was known about drug safety? A model was developed using all available toxicity data. Simulations assumed different levels of risk at different doses. Clinicians and statisticians considered seven scenarios, four in which no dose was viable, and three in which at least one dose was both safe and effective. Studies were simulated to find trial designs that could identify a viable (safe and efficacious) dose for approval, but that could also be stopped early for those scenarios in which no viable dose existed. The sponsor decided the risk was too high and halted development before risking patient safety and scarce research dollars.

Modeling and simulation practice is advancing. A 2010 industry survey of model-based development in 10 biopharma companies found broad application of modeling and simulation in both early- and late-stage development.¹⁸ Dose determination remains the primary focus of M&S in development. Survey responders indicated that M&S is having the most positive impact on the rationale for dose selection, on facilitating the work of scientific and strategic project teams, on making early go/no-go decisions, and on facilitating regulatory interactions. They also cited important emerging applications in study design, disease progression, human PK and PK/PD prediction, comparator models, and decision models. Companies expected increasing use of M&S in nearly all areas of development. 

The biopharmaceutical industry is clearly gaining experience with M&S, and the practice is advancing. This report proposes the adoption of a lifecycle approach in which M&S is included at each step of the development process. Although M&S testing and analysis add work at each developmental step, the resulting improvement in incremental decisions can lead to more efficient allocation of resources and greater likelihood of successful outcomes. According to one estimate, a 10% improvement in predicting failures before clinical trials could save $100 million in development costs per drug: “A mere 10% improvement in accuracy of decisions at any stage would confer disproportionately large benefits.”¹⁹ M&S practice supports knowledge-based approaches that can make clinical research processes more efficient and informative and enhance return on investment for drug developers in a challenging market environment. M&S will continue to gain ground as methodologies advance and new applications demonstrate their value.⁴

Andrew Garrett, PhD, is Vice President, Clinical Analysis and Reporting Services, Quintiles; Michael O’Kelly, PhD, is Senior Director, Center for Statistics in Drug Development, Innovation, Quintiles; Davis Walp, MBA, is Head of Value Based Solutions, Global Commercial Solutions, Quintiles (at the time of research); N. Seth Berry, PharmD, is Director, PK/PD Modeling and Simulation, Innovation, Quintiles. 

References

1. M. O'Kelly, S. Berry, D. Walp, and A. Garrett, "Lifecycle Modeling and Simulation: Current Practice and Future Impact on Healthcare Innovation and Delivery," Therapeutic Innovation & Regulatory Science, published online before print January 15, 2013, http://dij.sagepub.com/content/early/2013/01/14/2168479012471831.full.pdf+html.  

2. R.L. Lalonde, K.G. Kowalski, M.M. Hutmacher, et al., “Model-based Drug Development,” Clinical Pharmacology and Therapeutics, 82 (1) 21-32 (2007).

3. Food and Drug Administration, “Innovation or Stagnation? Challenge and Opportunity on the Critical Path to New Medical Products,” March 2004, http://www.fda.gov/ScienceResearch/SpecialTopics/CriticalPathInitiative/CriticalPathOpportunitiesReports/ucm077262.htm. 

4. P.A. Milligan, M.J. Brown, B. Marchant, et al., “Model-Based Drug Development: A Rational Approach to Efficiently Accelerated Drug Development,” Clinical Pharmacology and Therapeutics, 93 (6) 502-514 (2013).    

5. M. O’Kelly, “Examples of Using Modeling and Simulation in Getting Treatments to Market,” presented at the 2nd International Symposium on Biopharmaceutical Statistics, Berlin, 2011, http://www.isbiostat.org/sp2/Program_20110220.pdf. 

6 D. Walp, “Role of Simulation in Representing Real World Risk in Biopharma R&D Portfolio Optimization,” presented at DIA 2011, Chicago.   

7. Food and Drug Administration, “Guidance for Industry: End-of-Phase-2A Meetings,” September 2009, http://www.fda.gov/downloads/DrugsGuidanceComplianceRegulatoryInformation/Guidances/ucm079690.pdf.  

8. C. Garnett, H. Beasley, V. Bhattaram, et al., “Concentration-QT Relationships Play a Key Role in the Evaluation of Proarrhythmic Risk During Regulatory Review,” Journal of Clinical Pharmacology, 48 (1) 13-18 (2008).

9. J.Y. Lee, C. Garnett, J. Gobburu, et al., “Impact of Pharmacometric Analyses on New Drug Approval and Labelling Decisions,” Clinical Pharmacokinetics, 50 (10) 627-635 (2011).

10. Food and Drug Administration, “Food and Drug Administration Modernization Act of 1997,” http://www.fda.gov/RegulatoryInformation/Legislation/FederalFoodDrugandCosmeticActFDCAct/SignificantAmendmentstotheFDCAct/FDAMA/FullTextofFDAMAlaw/default.htm.

11. C.C. Peck, D.B. Rubin, L.B. Sheiner, “Hypothesis: A Single Clinical Trial Plus Causal Evidence of Effectiveness is Sufficient for Drug Approval,” Clinical Pharmacology and Therapeutics, 73 (6) 481-490 (2003).

12. PricewaterhouseCoopers, “Pharma 2020: Virtual R&D-Which Path Will You Take?” June 2007, http://www.pwc.com/gx/en/pharma-life-sciences/pharma-2020/pharma2020-virtualrd-which-path-will-you-take.jhtml.

13. Archimedes Inc., “Clinical Trial Simulation: Informing Enrollment Criteria,” Case Study, www.archimedesmodel.com.

14. B. Bieth, D. Renard, I. Demin, et al., “Longitudinal Model-based Test as Primary Analysis in Phase III,” EMA EFPIA Workshop Break-out Session No. 4,http://www.ema.europa.eu/docs/en_GB/document_library/Presentation/2011/11/WC500118295.pdf.

15. V. Anisimov and V. Fedorov, “Modeling, Prediction and Adaptive Adjustment of Recruitment in Multicenter Trials, Statistics in Medicine, 26 (27) 4958-4975 (2007).

16. V. Anisimov, “Predictive Event Modeling in Clinical Trials with Waiting Time to Response,”Pharmaceutical Statistics, 10 (6): 517-522 (2011).

17. T. Dumitrescu, R. Kendrick, H. Calvin, S. Berry, “Using Monte Carlo Simulations to Assess Dosing Regimen Adjustments of Piperacillin/Tazobactam in Obese Patients with Varying Renal Functions,” poster presented at American Conference on Pharmacometrics, Fort Lauderdale, May 2013. 

18. J. Stone, C. Banfield, M. Pfister, et al., “Model-based Drug Development Survey FindsPharmacometrics Impacting Decision Making in the Pharmaceutical Industry,” Journal of Clinical Pharmacology, 50 (9 Suppl) 20S-30S (2010).

19. Boston Consulting Group, “A Revolution in R&D: How Genomics and Genetics Will Affect Drug Development Costs and Times,” November 2001, http://www.forskningsradet.no/CSStorage/Flex_attachment/BiotekBCGGenomics.pdf.