Introduction

The Alzheimer's field has had its share of clinical trial flops. Now there's a push to learn from past failures. Computational scientists in pharmaceutical companies are guiding clinical trial design by first putting drug candidates through their paces in simulations. Those are proprietary, but C-Path’s Coalition Against Major Diseases has built some open-access simulation tools that clinicians everywhere can use to design better trials for mild to moderate AD. With the CAMD model, researchers can use a large amount of placebo data collected in prior trials to predict how biomarkers and patient demographics might affect trial outcomes.

On Wednesday, May 13, Richard Mohs, Klaus Romero, Lon Schneider, and Julie Stone discussed why trial simulations are important, what kind of questions they can answer, and how they can lead to better trial design.

Media

Slides

Background

No new drug has emerged for Alzheimer's disease since the FDA approved memantine in 2003, despite decades of research and hundreds of clinical trials. How do trialists avoid adding to this scrap heap? As part of their push for better trials, both industry and academic researchers have developed trial simulation tools based on the very data that doomed many a promising lead. Researchers are using summary and patient-level data from placebo and intervention arms of prior trials to better design news ones. Simulations can predict whom to treat, for how long, and with what dose. Researchers are running such simulations to test how parameters such as age, genotype, demographics and—importantly—dose might affect trial outcomes.  

Most of the large pharma companies run their own simulations on in-house models. Julie Stone is a bioengineer and pharmacokinetics expert at Merck Research Laboratories in Kenilworth, New Jersey. She works together with experimentalists in different diseases to integrate data from different sources and build the algorithms that undergird disease models. She uses systems pharmacology to quantitatively model interactions between the drug and its intended target. This helps predict, for example, how a given dose of Merck's BACE inhibitor MK-8931 perturbs the biology of the disease. The idea is to ensure that the doses being tested in Phase 3 have a meaningful impact on the system being targeted. “You have to hit your target hard if you want it to be effective,” is how Stone put it when speaking about quantitative pharmacology modeling at the 2015 NIH Alzheimer’s disease summit. These types of simulation were key in Merck's design of its Phase 3 trials. Though many amyloid-based interventions have not passed muster in clinical trials, Stone questions whether the amyloid hypothesis has truly been tested. She contends that quantitative modeling can help researchers query the level of perturbation their drug regimen introduces into the system. 

The Coalition Against Major Diseases (CAMD), a program of the Critical Path Institute in Tucson, Arizona, received endorsement from the FDA and the European Medicines Agency for its clinical trial simulation tool (CTS) (see Jul 2013 news and Rogers et al., 2012). Klaus Romero and colleagues at CAMD developed the tool in collaboration with industry leaders including Richard Mohs at Eli Lilly. The CTS is freely available. It comprises three models (see Romero et al., 2015). One estimates disease progression based on parameters such as gender, age, ApoE genotype, and disease severity at baseline; it is based on longitudinal data from the Alzheimer's Disease Neuroimaging Initiative (ADNI). Because AD progresses at different rates in different people, this simulator can be a powerful tool for setting up inclusion criteria, Romero told Alzforum. A second model simulates placebo effects and dropouts based on CAMD's clinical online data repository. CODR contains standardized data, mostly placebo, from 24 prior clinical trials covering 6,500 volunteers (see Dec 2010 conference news and Neville et al., 2015). A third CTS model simulates drug effects based on summary-level data from 73 published trials on nearly 20,000 patients. The CTS models both the standard placebo/treatment parallel group design and more sophisticated cross-over and delayed start designs. The latter may be better suited to detecting changes in AD progression and uncovering signs of disease modification.

Models can also inform post-hoc analyses about what might have happened had a trial used different selection criteria or patients. That then drives decisions for future trials. Lon Schneider and colleagues at the University of Southern California, Los Angeles, have used published data, including the CAMD and ADNI databases, to test some of these hypotheses. Running thousands of trial simulations by resampling the published data, they found, for example, that selection based on an ApoE4 genotype would make trials no more efficient but could make them longer because fewer patients would be eligible (see Schneider et al., 2010; Kennedy et al., 2013). On the flip side, they found that the age of enrolled patients greatly affected how fast their cognition changed, masking differences between placebo and approved cholinesterase inhibitors if the groups were poorly matched for age (see Schneider et al., 2015). Modeling adaptive designs, they found that recalculating sample sizes six months in can boost the power of a trial by a quarter. This depends on how many people are in the trial: Too few make the sample size re-estimation subject to error; too many eliminate the advantage of running the adaptive design.

Confused? Intrigued? Need a demo on how such models work? Join the Webinar and bring your questions.—Tom Fagan

Q&A

Q: What are the biggest obstacle(s) to deploying modeling?
Klaus Romero: Communication barriers between disciplines; lack of specific NIH funding streams; barriers to data-sharing and integration.

Lon Schneider: The biggest obstacles for clinical trials modeling are (1) obtaining the necessary clinical data from the many, many trials, (2) integrating the databases into one meta-database, (3) having the wherewithal to then go about developing the many models and simulations that are needed.    

Julie Stone:  Building the needed cross-discipline support in a forward-looking manner to be intentional about what/how data are obtained and how this will be integrated.

Q: What new modeling tools, if any, are needed?
Klaus Romero: NONMEM, R, Shiny(RStudio), C++, WinBUGS.

Q: Can the speakers introduce us to some useful databases, either publicly  available or commercially available, for computational modeling purpose?

Klaus Romero: CAMD database; CAMD Clinical tool. 

Q: Is there a model for CDR-SB as the primary endpoint?
Klaus Romero: Models based on ADNI data (disease progression only). There are no regulatory-endorsed models, as yet (see Samtani et al., 2014, and Delor et al., 2013).

Q: Does our panel have a consensus on best panel of bio-markers?
Klaus Romero: From our perspective, EMA has qualified the use of magnetic resonance imaging (MRI) as a tool to select patients with early stage cognitive impairment for Alzheimer’s disease clinical trials.

Q: Is the Washington University/C2N stable isotope technique a potentially useful tool?
Julie Stone:  Kinetic tracer data seems to provide complementary data to standard ELISA for the CSF biomarkers.  Models that integrate both types of data can support a more complex interpretation.

Q: Speaking of control arms, why do Alzheimer's studies involving mice have essentially zero to do with how treatments work with human subjects?
Klaus Romero: Translatability and predictive accuracy are a key limiting factor for animal models. I would refer to other areas where the predictive accuracy of non-clinical platforms has been adequately quantified (in the Webinar, Julie’s point about infectious diseases, and Richard’s point about oncology, as well as cardiovascular and metabolic indications).

Julie Stone:  When using animal data, it is important to consider what aspects of human disease are represented in the animal model and which not. For AD, I think there are many open questions in this regard.

Q:  How does the panel rate the need for certain types of pre-clinical animal studies?
Klaus Romero: Non-clinical platforms are an essential part of the drug-development process, but their predictive accuracy and specific applicability need to be accurately understood and quantified.

Julie Stone:  Animal models and experimental systems can provide valuable insight to help us move more quickly or gather information not readily available from clinical work, but it is important to consider the applicability of the animal model (see above).

Q: We have published, back in 2011, algorithms that allow us to forecast disease progression in aMCI subjects with high accuracy. These algorithms were originally based on ADNI data and recently confirmed using data from the NACC data base. We are surprised that these AD progression models, which are known under the acronym of PGSA (placebo group simulation approach) and should be useful for planning secondary preventative trials in AD, are not mentioned today. We have discussed this approach with Lon Schneider earlier and would be interested to hear his comments on this specific work in the current context.
Lon Schneider: The work being referred to is an analysis of change in clinical measures taken in the ADNI-1 cohort and the NACC cohort.  It describes how some participants who volunteered for ADNI or at the approximately 30 ADRCs and ADCs fared over a variable periods of time about a decade ago.  It is an interesting cohort. It does not involve placebos.  

Comments

1. • Hugo Geerts
     Certara
   • Posted: 28 May 2015

   There were some references in this Webinar to the need for more mechanism-based modeling. Someone asked whether modeling and simulation (ever) had been able to identify possible hypotheses for unexpected outcomes in CNS diseases. With the mechanism-based quantitative systems pharmacology approach that we optimized over the last 12 years we had three examples where modeling led to new insights and hypotheses.

   The first example was of a blinded, Phase 1 proof-of-concept study of a novel symptomatic AD drug, PF-04995274, a 5HT4 partial agonist, in scopolamine-treated volunteers (Nicholas et al., 2013). Our modeling approach correctly predicted a completely unexpected worsening of cognition. It did so because we modeled for the differential pharmacology of the drug in people versus rodents and correctly identified the serotonin level in the human cortex from earlier imaging studies.

   In another example, we investigated why memantine, an NMDA antagonist that weakened cognition in most preclinical animal models, would benefit moderate to severe AD patients (Roberts et al., 2012). In our ADAS-Cog calibrated cognitive model, we calculated for preclinical data showing that memantine binds the NMDA-NR2C receptor (on excitatory-inhibitory synapses) with a higher affinity than the NMDA-NR2A/B receptor (on excitatory-excitatory Glu synapses). Memantine thereby reduced glutamatergic drive on the inhibitory interneurons, restoring the excitation-inhibition balance that was created due to the greater impact of AD pathology on pyramidal cells as compared to GABA interneurons.

   A final example was understanding the inverse U-shaped dose response of glycine modulation on the negative symptoms of schizophrenia (Spiros et al., 2014). We modeled basic physiology and biochemistry, such as the differential interaction (in terms of EC50 and Hill slope) between glycine and the agonist site on each of the three NMDA-NR2 subunits, in addition to the nature of the glycine co-transporter (exchanging glycine for Na and Cl). When applying these processes to a neuronal cortical network that drives negative symptoms, this naturally leads to increased glutamatergic strength at lower glycine levels. Our model showed that at higher glycine levels the activation shifts from the excitatory-excitatory glutamatergic synapse to the excitatory-inhibitory glutamatergic synapse, thereby dampening the cortical strength to avoid over-excitation of the network and resulting in an inverted U-shaped dose response.

   References:

   Nicholas T, Sridhar Duvvuri S, Leurent C, Raunig D, Rapp T, Iredale P, Rowinski C, Carr R, Patrick R, Spiros A, Geerts H. Systems pharmacology modeling in neuroscience: Prediction and outcome of PF-04995274, a 5-HT4 partial agonist, in a clinical scopolamine impairment trial. Adv Alzheimer Dis. 2013 Sep;2(3):83-98.

   Roberts PD, Spiros A, Geerts H. Simulations of symptomatic treatments for Alzheimer's disease: computational analysis of pathology and mechanisms of drug action. Alzheimers Res Ther. 2012 Nov 26;4(6):50. PubMed.

   Spiros A, Roberts P, Geerts H. A computer-based quantitative systems pharmacology model of negative symptoms in schizophrenia: exploring glycine modulation of excitation-inhibition balance. Front Pharmacol. 2014;5:229. Epub 2014 Oct 21 PubMed.

   View all comments by Hugo Geerts

Make a Comment

To make a comment you must login or register.

References

Therapeutics Citations

1. Verubecestat

News Citations

1. AD Trial Simulation Tool Receives Regulators’ Blessings 11 Jul 2013
2. DC: Shared Pain Is Lessened—Open-Trial Data Gain AD Model 22 Dec 2010

Paper Citations

1. Rogers JA, Polhamus D, Gillespie WR, Ito K, Romero K, Qiu R, Stephenson D, Gastonguay MR, Corrigan B. Combining patient-level and summary-level data for Alzheimer's disease modeling and simulation: a beta regression meta-analysis. J Pharmacokinet Pharmacodyn. 2012 Oct;39(5):479-98. PubMed.
2. Romero K, Ito K, Rogers JA, Polhamus D, Qiu R, Stephenson D, Mohs R, Lalonde R, Sinha V, Wang Y, Brown D, Isaac M, Vamvakas S, Hemmings R, Pani L, Bain LJ, Corrigan B, Alzheimer's Disease Neuroimaging Initiative, Coalition Against Major Diseases. The future is now: model-based clinical trial design for Alzheimer's disease. Clin Pharmacol Ther. 2015 Mar;97(3):210-4. Epub 2014 Dec 27 PubMed.
3. Neville J, Kopko S, Broadbent S, Avilés E, Stafford R, Solinsky CM, Bain LJ, Cisneroz M, Romero K, Stephenson D, Coalition Against Major Diseases. Development of a unified clinical trial database for Alzheimer's disease. Alzheimers Dement. 2015 Feb 9; PubMed.
4. Schneider LS, Kennedy RE, Cutter GR, . Requiring an amyloid-beta1-42 biomarker for prodromal Alzheimer's disease or mild cognitive impairment does not lead to more efficient clinical trials. Alzheimers Dement. 2010 Sep;6(5):367-77. PubMed.
5. Kennedy RE, Cutter GR, Schneider LS. Effect of APOE genotype status on targeted clinical trials outcomes and efficiency in dementia and mild cognitive impairment resulting from Alzheimer's disease. Alzheimers Dement. 2014 May;10(3):349-59. Epub 2013 May 25 PubMed.
6. Schneider LS, Kennedy RE, Wang G, Cutter GR. Differences in Alzheimer disease clinical trial outcomes based on age of the participants. Neurology. 2015 Mar 17;84(11):1121-7. Epub 2015 Feb 13 PubMed.
7. Samtani MN, Raghavan N, Novak G, Nandy P, Narayan VA. Disease progression model for Clinical Dementia Rating-Sum of Boxes in mild cognitive impairment and Alzheimer's subjects from the Alzheimer's Disease Neuroimaging Initiative. Neuropsychiatr Dis Treat. 2014;10:929-52. Epub 2014 May 24 PubMed.
8. Delor I, Charoin JE, Gieschke R, Retout S, Jacqmin P. Modeling Alzheimer's Disease Progression Using Disease Onset Time and Disease Trajectory Concepts Applied to CDR-SOB Scores From ADNI. CPT Pharmacometrics Syst Pharmacol. 2013;2:e78. PubMed.

External Citations

1. Phase 3 trials
2. CAMD database
3. CAMD Clinical tool

Further Reading