模型引导的药物开发（MIDD）是一个紧密结合的生态系统网络，可以无缝定位候选新药，同时最大限度地减少技术和监管成功的不确定性。Indeed, MIDD affords rational scientific hypothesis testing while enabling the clinical development program using strategic pillars. These strategic pillars form a core component of an innovation ecosystem (Figure 1) and are further elaborated below in the context of rare diseases.
Figure 1: The MIDD innovation ecosystem
Biomarkers and surrogate endpoints
The search for meaningful biomarkers and surrogate endpoints continues to be within the scope of drug discovery and development in rare diseases. For example, consider the choice and adaptation of the 6-minute walk distance (6MWD) as an end point for studying Duchenne muscular dystrophy (DMD). McDonald et al [3,4] modified the American Thoracic Society version of the 6-minute walk test (6MWT) for DMD and validated the 6-minute walk distance (6MWD) as a measure of disease progression in ambulatory DMD patients. The progressive loss of muscle function is an excellent candidate for modeling outcomes based on natural history studies in terms of the chronology of this disease progression. McDonald and colleagues assessed changes in the 6MWD in longitudinal studies to support evidence for reliability and validity of the 6MWD. These results supported in the identification of the minimally clinically important difference for the 6MWD endpoint in DMD interventional studies [5]. The choice of early biomarkers to support dose selection should consider the relationship with the surrogate endpoint [6].
Enrichment strategies and choice of historical/external controls
Rare diseases are often defined by a common genetic basis or based on commonality of disease progression. Thus, rare disease clinical trials are ideal for enrichment approaches. Since consistency in treatment effects is desirable, one could gain increased treatment magnitude effects in a higher proportion of study subjects with a smaller sample size for hypothesis testing.
Perrone and coworkers [7] developed a statistical model that linked longitudinal total kidney volume (TKV) measurements with age and estimated glomerular filtration rate (eGFR) to the probability of a 30% decline of eGFR or end-stage renal disease (ESRD). As such, the purpose of this study was to use the model as a drug development tool for trial enrichment in patients with autosomal dominant polycystic kidney disease (ADPKD). TKV is an imaging biomarker for predicting the natural history of patients with ADPKD. Observational data collected over multiple decades for an eGFR decline and end-stage renal disease in patients with ADPKD revealed that predicted TKV at the time of a 30% decline of eGFR and that at the time of end-stage renal disease were both highly significant. These findings revealed that when eGFR is preserved, patients with larger TKV are more likely to progress to a 30% decline of eGFR within the course of a clinical trial. Such approaches can optimize trial designs in patients with ADPKD. Choice of historic controls will permit allocating a limited sample pool to the test treatment. An externally controlled trial as “one in which the control group consists of patients who are not of part of the randomized study as the group receiving the investigational agent i.e., there is no concurrently randomized control group” [8]. There are many examples of using external controls in regulatory decision making (Table 1).
Table 1: Partial list of approved drugs leveraging an external control
| Product | Indication | Type of external control | Reference |
|---|---|---|---|
| Zolgensma | Infantile spinal muscular atrophy | Data from 23 patients that were used as an external control and is a good example of the use of natural history data | 9 |
| Novoeight | Hemophilia A | Historical control data from prior trials were used as external controls | 10 |
| RIXUBIS | Haemophilia B | Historic control group supported regulatory approval | 11 |
| Myozyme | Infantile-onset Pompe disease | Comparison of outcomes for 18 pivotal trial participants together with outcomes for infants followed in a natural history study | 12 |
Statistical methods
Bayesian statistical approaches can be used to analyze data from confirmatory studies. One good example is using such an approach in the US FDA approval of the B-domain deleted factor VIII product moroctocog alfa (AF-CC, XYNTHA) [13]. XYNTHA is a recombinant antihemophilic factor indicated in adults and children with hemophilia A for on-demand treatment and control of bleeding episodes, for perioperative management, and for routine prophylaxis to reduce the frequency of bleeding episodes. Such analyses established an upper threshold for acceptable inhibitor frequency based on the assumption of a non-informative prior distribution from marketed full-length recombinant factor VIII products. Two prior studies with moroctocog alfa (AF-CC) formed the basis of an informative Bayesian prior distribution [13]. Such an approach was used to meet the primary safety endpoint when the Bayesian distribution was updated with data from the pivotal study.
Another helpful statistical innovation tool is the choice of adaptive designs, wherein adaptation is applied to optimize the assessment of the dose-response relationship. They are also principally used to right-size the doses to yield the projected target dose. This approach uses preset levels of clinical thresholds that are based on biomarkers and/or surrogate endpoints. There are many helpful tools of adaptive approaches including, but not limited to, those supporting design elements or analytical elements. In design-based adaptive approaches, for example, a general adaptive dose allocation approach uses Bayesian modeling to identify an appropriate dose for each new trial patient based on partial and complete responses from the previously enrolled patients. D-optimal response-adaptive approach uses the D-optimality criterion to select the patient allocation scheme that provides most information about the dose-response relationship from interim data. In analytical based adaptive approaches, Bayesian methods can be used to produce weights used to combine information obtained.
Optimal sampling
Clinical pharmacology studies are performed to characterize the pharmacokinetics and pharmacodynamics of drug development candidates in healthy subjects and patients with target diseases. A combination of rich sampling and sparse sampling is often prudent in a pivotal study. To account for sparse data and the resulting challenges in parameter identification, non-linear mixed-effects (NLME) models are often used when fitting data in a population. Such models compartmentalize sources of variability through the application of hierarchical statistical models. Non-linear, mixed-effects models (also known as “population PK models”) allows the user to distinguish between residual, unexplained variability and variability due to differences within and between subjects in the study.
A poorly designed population PK experiment can often lead to inaccurate and unreliable parameter estimates. A more sensible alternative when designing optimal sampling strategies in studies of limited sample size is through optimal experimental designs, or D-optimal designs [14-16]. D-optimality seeks to reduce the scalarized covariance of the estimated population parameters by selection of an experimental design. Using such approaches can minimize the sampling burden in patients while ensuring rigor and applicability of the generated population PK/PD models.
Model-based meta-analysis
In simple terms, meta-analyses integrate the results of multiple clinical drug trials to generalize or to lend credibility to the observations [17]. There are three types of meta-analysis approaches used in drug development and these include pairwise meta-analysis, (PMA), network meta-analysis (NMA), and model-based meta-analysis (MBMA). PMA compares treatments in pairs; however, it cannot indirectly compare drugs that have not been studied head-to-head in a clinical trial. NMA combines trials with different treatments and comparators into a single framework, allowing simultaneous accommodation of direct and indirect comparisons. Lastly, MBMA is a meta-analysis tool that incorporates parametric pharmacology dose/duration models through integrating relevant biomarker, surrogate endpoints, and clinical safety and efficacy data of competing treatment options within a certain disease area. The MBMA approach also supports bridging across studies, and common biomarker elements, enabling comparing treatments that may never have tested in the same clinical trial.
Summary
Rare diseases encompass a somewhat restrictive and poorly accessible patient pool. Moreover, inherent in the scarcity of the patient population for clinical trials make data interpretations quite challenging. These limitations can be significantly overcome using a model-informed drug development enabled program, which will reduce uncertainty in both technical and regulatory success. The three areas that need further research includes the choice of more proximal biomarkers of target engagement, their relationship with clinical benefit, and the natural history of the disease state being investigated. These considerations can be augmented using a mathematical framework supporting MIDD. To learn more about best practices in rare disease drug development, please read this white paper:
References
1. Arkin S. The challenges of conducting clinical trials in diseases with small target populations. RSC Drug Discovery Series, No. 38 Orphan Drugs and Rare Diseases. Edited by David C Pryde and Michael J Palmer, pp. 53. 2014.
2. Asbury CH. The Orphan Drug Act: the first 7 years. Journal of the American Medical Association 265(7):893-897, 1991.
3. McDonald CM, Henricson EK, Han JJ, Abresch RT, Nicorici A, Atkinson L, Elfring GL, Reha A, Miller LL. The 6-minute walk test in Duchenne/Becker muscular dystrophy: longitudinal observations. Muscle Nerve, 42, 966–974, 2010.
4. McDonald CM, Henricson EK, Han JJ, Nicorici A, Abresch RT, Atkinson LA, et al. The 6-minute walk test (6MWT) as a clinical trial outcome measure in Duchenne/Becker muscular dystrophy (DMD/BMD). Neuromuscul Disord. 18:739, 2008.
5. McDonald CM, Henricson EK, Abresch RT, Florence J, Eagle M, Gappmaier E, Glanzman AM, Spiegel R, Barth J, Elfring G, Reha A, Peltz SW. The 6 minute walk test and other clinical endpoints in Duchenne muscular dystrophy: reliability, concurrent validity, and minimal clinically important differences from a multicenter study. Muscle Nerve, 2013, 42, 357–368.
6. Fleming TR. Surrogate endpoints and FDA’s accelerated approval process. Health Affairs 24(1):67-78, 2005.
7. Perrone RD, Mouksassi MS, Romero K, Czerwiec FS, Chapman AB, Gitomer BY, Torres VE, Miskulin DC, Broadbent S, Marier JF. A Drug development tool for trial enrichment in patients with autosomal dominant polycystic kidney disease. Kidney Int Rep 2, 451–460, 2017.
8. FDA 2019. 罕见病：natural history studies for drug development guidance for industry 2019. https://www.fda.gov/media/122425/download (Last accessed: 2021 年 11 月 19 日).
9. FDA 2021. Summary basis of approval for onasemnogene abeparvovec-xioi (Zolgensma) https://www.fda.gov/vaccines-blood-biologics/zolgensma (Last accessed: 2022 年 3 月 2 日).
10. FDA 2018. Summary basis of approval for Antihemophilic Factor [Novoeight]. https://www.fda.gov/vaccines-blood-biologics/approved-blood-products/novoeight (Last accessed: 2022 年 3 月 2 日).
11. FDA 2018. Summary basis of approval for Coagulation Factor IX [RIXUBIS]. https://www.fda.gov/vaccines-blood-biologics/approved-blood-products/rixubis (Last accessed: 2022 年 3 月 2 日).
12. FDA 2006. Summary basis of approval for Alglucosideasealfa [Myozyme]. https://www.accessdata.fda.gov/drugsatfda_docs/nda/2006/125141s000_MyozymeTOC.cfm (Last accessed: 2021 年 11 月 19 日).
13. FDA 2020. Summary basis of approval for moroctocog alfa [Xyntha]. https://www.fda.gov/vaccines-blood-biologics/approved-blood-products/xyntha (Last accessed: 2022 年 3 月 2 日).
14. Dodds MG, Hooker AC, Vicini P. Robust population pharmacokinetic experiment design. Journal of Pharmacokinetics and Pharmacodynamics, 32 (1): 33-64, 2005.
15. Duffull SB, Retout S, Mentre´ F. The use of simulated annealing for finding optimal population designs. Comp. Meth. Prog. Biomed. 69:25–35, 2002.
16. Hooker AC, Foracchia M, Dodds MG, Vicini P. An evaluation of population D-optimal designs via pharmacokinetic simulations. Ann. Biomed. Eng. 31:98–111, 2003. 17. Boucher M, Bennetts M. The many flavors of model based meta-analysis: Part I – Introduction and Landmark data. CPT:Pharmacometrics & Systems Pharmacology, 5(2):54-64, 2016.
