YouTube Facebook LinkedIn Google+ Twitter Xingrss  

Novartis Savors Early Modeling Success


By John Russell
 
March 19, 2009 | Modeling and Simulation (M&S) has a complex history in the life sciences. Big pharma and small biotechs alike have struggled with finding the right recipe. Like a good wine, having the right general ingredients—management buy-in, modeling talent, sufficient resources, suitable problems—is necessary but not enough for success. A certain amount of vintner’s experiment and bottle aging has to occur and more than a few companies have gulped some awful tasting stuff along the way.

Pretty clearly there won’t be a single best recipe, but Novartis has been demonstrating it has found one that works and works quite well. A bit more than three years ago, Donald Stanski, M.D., during his recruitment process convinced Novartis that Modeling and Simulation would never reach its potential embedded in the biostatistics department. It needed to be its own department; organized in close alignment with Novartis’ drug development therapeutic franchises; enjoy the right reporting structure; and have enough resources and time to prove itself.

Modeling scientists would sell themselves to the drug development project teams and sink or swim, based on results, deliverables, and value, argued Stanski.

Of course, it’s more complicated than that. Many key personnel were already in place such as Gabriel Helmlinger, Ph.D., director and global head, M&S-Biology. And senior management was ready to buy in. Trevor Mundel, M.D., Ph.D., then head of Translational Medicine, is now global head of drug development at Novartis.  He’s a physician scientist who did graduate work in the applied mathematics of dynamical systems.

“I’ve always seen modeling and simulation as a key technology that really hasn’t been as widely deployed or as effectively deployed as it could be,” says Mundel. “We’re now having the chance to work with Don’s group and to give it its proper place in the context of development. We are starting to see some real traction.”

Not surprisingly, Mundel is focused on “the phase two problem (attrition, duration, and cost) we talk about in industry and how one de-risks the process with proof of principle efforts. Given that proof of principle gives you a pretty small data set, how can you play that up and extract more information so you don’t have to spend excessive time in phase two. We bring in modeling and simulation as one aspect of it.” 

The Novartis bet on Stanski seems to be paying off. He joined as Vice President and global head of Modeling & Simulation and grew his team from 30 to 50 scientists in the first year. There was a pause in the second year to refine operations and review. This year, he has approval to add 15 staff in what surely must be one of the most difficult environments in memory to pitch growth to senior management. The M&S group has even attracted funding for half a dozen extra FTEs from separate Novartis business units such as Oncology, Sandoz, and Animal Health who like what they see and want more.

Stanski says playfully, “One of the coolest problems we’ve been involved with was with Animal Health. It was around dosing for cows. They pour the medication on the backs of the cows and the cows self-dose by licking each other’s backs! The question was around the dose to the cows, and the resulting kinetics in the animals.”

The scope of Novartis M&S department activities is broad. Biological modeling (pathways, mechanisms, drug-disease), pharmacological modeling (various PK/PD and trial design), biophysical modeling (organ- and tissue-level models such as the spinal cord), and most recently, decision analysis, are all part of the mix. The scope of questions tackled is equally expansive and generally familiar: appropriate biomarker selection and measurement strategy, dosing guidance, candidate and target selection, clinical study design, competitive analysis, etc.

Though impact is difficult to measure, Brian Stoll, senior expert modeler who works extensively with biologics says, “Today it’s unlikely that a new biologic candidate would make it into Novartis’ late stage pipeline without contributions from M&S.” Stanski agrees.

Given the challenges biopharma has encountered adopting M&S activities, Novartis graciously agreed to share some of its M&S experience with Bio-IT World and Predictive Biomedicine. Joining me and Bio-IT World editor-in-chief Kevin Davies for a briefing were Mundel (by phone from Basel), Stanski, Helmlinger, Stoll, James Dunyak, expert modeler, and Kai Wu, senior modeler. The wide-ranging discussion encompassed key organizational principles, which Stanski insists are essential for M&S success, as well as presentation of five examples of concrete M&S projects representative of the department’s work. 

It’s probably worth noting the M&S department’s new digs in Cambridge, Mass., where it is co-located with the Novartis Translational Sciences group, are distinctly high-tech and sleek. No dingy, windowless cubicles for this crowd. The setting for so many mathematical biology groups often feels distinctly broom-closet-like and the group’s persona too often reflects the same.

It also hasn’t hurt that Stanski is one of the thought leaders who has been working to get the modeling approaches right for years. Trained as a physician/anesthesiologist, he is also a researcher and clinical pharmacologist with stints in industry (VP, scientific and medical affairs at Pharsight), academia (Chair of the Department of Anesthesia at Stanford University from 1992 to 1997), and the FDA (advisor for the director, Center for Drug Evaluation and Research). Stumbles are frequent in the implementation of any new technology and often underappreciated in the early stage of roll-out. Having Stanski’s multi-year perspective no doubt helps.

“Most companies create modeling silos by distributing the modeling talent throughout the company,” he says. Biological pathway modeling is separate from PK/PD modeling, which is separate from statistics. That’s usually a crippling mistake, according to Stanski. All of those disciplines need to be brought to bear on problems and the cross-fertilization is essential and opportunistic.

“I had never worked with this [biological pathways modeling] before,” says Stanski. “Gabriel and his colleagues represented learning for me. When I saw what they could do and how you could integrate it, to me that represented one of the most interesting high value opportunities.”

“The Novartis M&S team is organized loosely around ‘three dimensions’,” he explains, modeling skills, therapeutic drug development focus, and software working platforms. Modeling skills are basically what your scientific modeling background is—biological pathways, Bayesian statistics, or trial design, as examples. Next, and perhaps the most formal, the department has organized “clusters” to mirror the Novartis therapeutic franchise drug development approach (Oncology, Cardiovascular, etc). The heads of each M&S modeling cluster work closely with their drug development franchise counterparts to identify opportunities to add value.

“We align ourselves based upon how Trevor has set up the therapeutic drug development focus of the company. We [M&S] call them clusters to be a little different,” says Stanski. “These are the basic therapeutic franchises in Novartis. Oncology is a separate business unit.  But we do the modeling for them and they provide funding for FTEs.”

The organizing principle, Platforms, refers to re-useable modeling tools that address a recurring problem in Novartis M&S jargon. “We have different types of modeling platforms,” explains Helmlinger. “Some are very methodology-driven. For example, look at data mining. There are many algorithms out there, classifiers, search algorithms, inference algorithms. If we get a new question around data mining, we might have to search the whole methodology background of the algorithms to see which one we should try. But we have organized this into a library form so we know which types of algorithms we can use, and which are software-ready to apply to a particular problem.”

Helmlinger indicates, “There are also safety (cardiac safety modeling is one) as well as drug-disease modeling platforms. Take diabetes, we have addressed many diabetes problems through our projects using PK/PD modeling, so there‘s whole range of PK/PD models with different focuses. We capture all of those in a diabetes drug-disease modeling platform. Furthermore we can then integrate these models maybe to gather the molecular pathways that we know of and how we think they can be networked into one, let’s say, higher level model. So that’s the goal.”

Platforms represent both institutional memory and practical tools. Currently the M&S team is the prime user but there is no reason, and every hope, that over time a wider Novartis constituency will use the platforms. The cardiac safety platform has been shared with Novartis basic researchers who evaluate the electrophysiology of all compounds that will be given to humans.

A good deal of IT support is obviously required for Modeling and Simulation. Stanski says the group works closely with the Novartis IT services but needed to also make its own major contributions. “Our IT needs are specific to what we do. It’s been time-consuming, extensive, and painful, but it’s finally starting to work.  We have 2 to 3 FTEs working on IT issues for the past several years. That’s expensive for a group of 50, but you know what, there’s been no choice.”

Winning acceptance from the therapeutic drug development franchises is the key to sustaining Stanski’s vision. As the saying goes, nothing succeeds like success.

“I’m going provide two examples of projects on the Discovery side; I’d classify both of these as mechanistic modeling,” says Stoll. “By mechanistic modeling we mean modeling in which we try to capture the underlying biophysics and biology and for which the parameters have a meaningful physical interpretation.”

The first example involves the selection of antibody drug candidates. The challenge is to rationally select from among a pool of candidates those most likely to meet with success in the clinic. In addition, in this case, there is a competitor that hits a different target in the pathway already on the market.

Leveraging what they know about the disease and some clinical data, “we develop a model of cell signaling pathways in which an ordinary differential equation describes the change in each molecular species with respect to time,” indicates Stoll. “We used the model to identify candidates that would not only be successful but that would also be well positioned against our competitor.”

The uniformity in the pharmacokinetics of certain antibodies along with an understanding of the underlying biology enables “model prediction of relationships such as dose-response curves very early in discovery even before certain pre-clinical studies are conducted.”
The model provides quantitative evidence that their target has potential and helped narrow the pool of candidates to be evaluated further. That’s not all.

“We can further compare the cost of time spent further maturing the antibody affinity and compare it with the savings in cost of goods associated with administering a lower dose because you have a higher affinity. This is an example of tradeoff calculations we can make to facilitate more informed decision-making,” indicates Stoll.

The franchise development project team makes the final decision, using the modeling and simulation information along with the multiple other factors that contribute to drug development decision-making. Stoll presented a second, more complex example, involving a signaling pathway with many non-linear characteristics. Among other the things, the purpose of the modeling was to understand the activation dynamics of biomarkers of response. Both modeling efforts are in silico efforts validated using experimental data.

In many ways, these modeling analyses integrate data and present them in quantitative ways to assist decision making for the drug development team. This is commonplace in other industries in which the underlying phenomena are better understood. But in the pharma/biotech industry getting from the data to the models and to their revealing predictions can be a lengthy, arduous process.

It is perhaps too easy to forget how important macroscopic modeling can be. Gabriel Helmlinger showed Novartis work modeling the spinal column and the flow of cerebrospinal fluid through it. One impressive example involved the application of a biological protein antibody into the cerebrospinal fluid of humans to treat spinal cord injury. The example was complex because of the unique site of administration (intrathecal).

“We had very good efficacy data in animal models but the big question is how you scale it up to humans. In the classical PK/PD world there are many methods, some of them are quite simple to try to scale up properly to humans,” says Helmlinger. “But here, dealing with the spinal cord, we’re in a different geometry, a different fluid dynamics and kinetic sampling site that traditionally is not easily accessed,” says Helmlinger.

“So what we did, like building an airplane on a computer, we built a part of the human body on the computer, the spinal cord and surrounding tissues, using a biophysical modeling platform. [The challenge] from an engineering standpoint, is to describe the geometry of the system and the transport properties that function within that geometry—namely injecting a large molecular weight antibody protein into this a unique, anatomically specific space, the intrathecal region. ”

The key question is if the patient is injected with the biological antibody at a particular point in the spinal canal, will the drug get to the site of action? Variables include the syringe type, the orientation of the syringe, the location, the rate of injection, is the patient upright or lying down. You can put all of this in such a biophysical model.

Coincidentally, Stanski, an anesthesiologist, is very familiar with exactly this problem. “I’ve put lots of small molecules/local anesthetics into people’s spinal fluid for knee or hip surgery, but a monoclonal antibody at a molecular weight of approximately 300,000 Daltons is very different than a small molecule with a molecular weight of 300.”

“The whole platform vision here, depending on the injection conditions, is to know what happens to the antibody. Knowing from preclinical data that the antibody is efficacious, we just want to make sure it gets to the site of injury in patients with a spinal cord injury” Helmlinger says.

Space constraints prevent reviewing all the examples presented. No less impressive was senior modeler Kai Wu’s description of an integrated drug efficacy and safety modeling analysis in which the correct dosing regime was determined. The Novartis M&S team seems bent on leaving no stone unturned with regard to how M&S might be applied to enhance decision making and progress toward corporate goals.

One unusual and somewhat new effort is decision analysis led in Cambridge by Jim Dunyak. His work, literally, involves examining all of Novartis’ drug development practices and examining the decision making processes that take place.

“Our vision of decision analysis, actually, is that it should be used at all levels of the company,” explains Dunyak. “I‘m doing some work in early development on how we treat patients in clinical trials in different countries,” says Dunyak. “There are all kinds of issues associated with manufacturing and formulation and there are almost always unsolved technical issues, not just tactical issues of where are you going to make it and what are the raw ingredients, but there’s usually formulation issues. For many projects I’ve ever been on, the problem has been described as less soluble than brick dust.” 

More prosaically, he presented work on a specific project to model in which the original strategy presented by the franchise development project was flawed and would have introduced delay without achieving substantial benefit; the strategy was subsequently changed because of his analysis.

Stanski says, “Our challenge is to find areas where we can mix decision analysis and economic modeling, to the drug development. I can’t say we’ve got it all figured out. Some of it is opportunistic. You know, Jim has built credibility. Now they’re coming back for more which can create a capacity issue. This is early yet. We’re six months to a year into this. But this isn’t where we started. We started with Trevor on a proof of concept and drug-disease model and did that for 2 to 3 years.”

So far, Stanski’s commitment to keeping modeling scientists together, aligning efforts directly with the therapeutic drug development franchises, and making certain that modeling directly supports decision-making activities rather than serve a broadly descriptive function is paying off. Mundel’s goal, after all, has little to do with smart-looking graphs and charts, and everything to do with speeding up and effectively improving drug development.

The group keeps close tabs on itself. After every project, Stanski says, “We do a specific interaction interview with the key decision makers on what was the value of the modeling efforts. We also have our own software to keep track of projects and outcomes. When a project is done there will be a value statement illustrating the impact. It’s very tangible. We make a proactive effort to capture the value of everything we do,” says Stanski.

What portions of this formula are transferable to other companies isn’t clear. Every company is different in terms of resources, understanding of quantitative decision-making and leadership understanding of modeling and simulation. But Stanski’s hopes for M&S at Novartis are ambitious. It will be interesting to check back on the team’s progress in a couple of years. Right now the Novartis M&S “vinification process” is bubbling nicely.

--

This article first appeared in Bio-IT World’s Predictive Biomedicine newsletter. Click here for a free subscription.

Click here to login and leave a comment.  

0 Comments

Add Comment

Text Only 2000 character limit

Page 1 of 1

For reprints and/or copyright permission, please contact  Jay Mulhern, (781) 972-1359, jmulhern@healthtech.com.