Critical Mass for Critical Path?

Innovation in science is no quick trick, and neither is collaboration. Back in March 2004, FDA sounded a now-famous alarm: Despite the drug industry's 250 percent jump in spending on R&D, drug-development productivity had plunged by 50 percent over the previous decade. The report, which became known as the Critical Path Initiative (CPI), last year yielded an industry-wide call to arms regarding 76 action items (aka, the "Opportunities List") in six key areas: biomarker development, the streamlining of trials, the harnessing of informatics, improving drug manufacturing, public health initiatives against infections and bioterrorism, and special programs for adolescents, children, and other at-risk populations.

Qualify my BIOMARKER

Whether or not Critical Path succeeds depends, to a surprising degree, less on new science than on new ways of doing science. Some of these innovations are based on a model of collaboration by which companies share certain kinds of knowledge with one another and with FDA. Not having to create new science should make things easier, right? "

Maybe. A few early CPI collaborations are encouraging, but in general, sharing—predictably enough—creates special challenges for the industry. For one thing, many aspects of the FDA–pharma relationship are inherently adversarial, and decades of practice cannot be overturned overnight. For another, self-interest is encoded in every drug firm's DNA. Companies are duty bound to protect their valuable intellectual property and loath to share it with competitors. Ultimately, voluntary collaborations are doomed to fail unless each partner can find a satisfactory answer to "What's in it for me?"

SHARE the wealth

Collaboration requires an investment of resources and processes. New organizations must be defined, funded, and staffed. Further, when sharing technical data, legal permissions must be obtained, standard definitions and data structures hammered out, and a trusted intermediary assigned to convert it all to an agreed-upon format. And where full data sharing is not possible, there must be explicit disclosure of what is missing, so that users may judge reliability and limitations. This thicket of immediate drawbacks tends to obscure the bright-but-delayed promises of working in new ways.

Drug Development: By the Numbers

Change is painful. Reasons and excuses not to innovate accumulate every step of the way. It helps to keep in focus the serious problem that Critical Path is intended to fix. The cost of bringing a new drug from discovery to market keeps rising, currently consuming $1 billion and 15 years. The latest statistics indicate that Phase III attrition now approaches 50 percent, and as the recent loss of 10,000-plus jobs at Pfizer has shown, the impact of large, late-stage failures such as torcetrapib can prove devastating.

Let's do the math. Real R&D spending, adjusted for inflation, is up 42 percent over the past five years. Where is the money going? The good news is that slightly over half funds a rising number of projects—the pipeline of active R&D projects (counting preclinical and clinical) has grown 24.5 percent since 2001. The bad news is that the rest covers the higher cost-per-project, a 14.2 percent rise over the same period. Most of that bump comes from expensive new tools, such as transgenic animals, robotics, microchips for fixing DNA samples, new intensive cardiac-safety trials, and information technology.

The modle solution

More projects and higher costs would be easier to stomach if they were producing more innovative drugs. But, as Wall Street keeps making painfully clear, it just ain't so. The number of new drugs that the FDA regards (on a good day) as significant advances has remained flat over the past decade. Business-as-usual is no longer a feasible strategy.

Data, Markers, Models: Grin and Share It

So, given that the bar on drug-development innovation keeps rising, how can "data sharing" reverse the trend?

Every drug maker has to painstakingly gather its own high-quality, high-cost data in tightly supervised preclinical and clinical trials (see "Share the Wealth"). The more data there are, the more powerful the information is—and that leads to faster, better decisions and less late-stage attrition. By almost any measure, sharing data, analyses, and models is a win-win.

Proponents of collaboration also argue not only that no individual pharma alone can exploit all medically and commercially attractive opportunities, but that collaboration itself will result in new innovation. With HIV research, the introduction of CD4 and viral load as biomarkers for efficacy led to the approval of an entire class of lifesaving drugs in three years or so. Along the way, there were remarkable examples of collaboration. For instance, Merck published the X-ray crystallography showing the structure of protease, allowing other companies to cut time and money from their protease-inhibitor development efforts. HIV is certainly a dramatic case of critical patient need and extraordinary political focus, but such treatment progress illustrates what can be accomplished when pharma keeps its eye on the prize.

For all these reasons, FDA is pushing collaboration. A number of initiatives have been proposed, with three notable consortia already underway:

• Industry/C-Path Predictive Safety Testing Consortium The goal is to encourage companies to share information about the lab tests and other methods each has developed to screen drugs and identify potential side effects. Companies will agree to test and confirm one another's methods, helping FDA create new drug-development guidelines using state-of-the-art safety tests.

• PhRMA Biomarker Consortium Government and industry scientists are working to identify and validate new biomarkers for use in the prevention and detection of disease. Companies share early, nonproprietary information on biomarkers. (For more on biomarkers, see "Qualify My Biomarker".)

• Serious Adverse Event Consortium This big-daddy consortium of more than a dozen pharmas, academics, and government agencies focuses on discovering DNA variations that help predict a patient's reaction to a drug. Findings are available to the public.

FDA is also a ramping up its model-based drug-development (MBDD) capabilities. MBDD is an approach to drug development that uses model-based methods to ensure that all decisions, such as go/no-go, are grounded in quantitative inputs. (For more on model-based drug development, see "The Model Solution".) In addition to assessing the relationship between a compound's dose, concentration, and response, models can be used to conduct faster, cheaper clinical trials by factoring in a range of critical data, including the following:

• Disease progression When modeling drug effects, it's important to factor in how a disease may progress in patients who are untreated. This offers a baseline against which to evaluate the benefits of treatment. For example, data on patients who are untreated or on placebo can be shared to describe the development of viral resistance over time.

• Patient compliance This describes how failing to follow a prescribed regimen affects trial results and patient health.

• Placebo response and gold-standard therapy This enables researchers to design trials that can succeed despite either a placebo effect or an effect from the gold-standard therapy. For example, some patients with high blood pressure respond (at least for a short time) to a placebo. As a result, one possible design for the study of an anti-hypertensive might be to first put all patients on a placebo and then enroll only those unresponsive to placebo in the treatment phase.

• Dropout data This leads to better understanding of the effect on statistics of missing trial data due to patients dropping out and helps refine trial designs to accommodate likely dropouts.

• Standard safety analyses To develop a drug's toxicity profile, it's standard procedure to record adverse events, clinical chemistry and hemotology values, and any drug-related changes on the QT interval, which measures cardiovascular complications.

A Collaboration Lab: FDA Fits and Starts

Successful collaborations generally feature both tight scientific focus and professional management. The three consortia, for instance, are highly targeted, pursuing very specific scientific questions, such as the screening of molecules to predict liver or kidney toxicity, advanced imaging techniques to assess drug response in non-Hodgkin's lymphoma, and the analysis of highly structured side-effects data to see how patient genetics affects response to drugs. The narrow focus minimizes scientific risk.

But major progress will be difficult without major investment. Many proposed biomarkers will no doubt eventually prove unreliable or unqualifiable for FDA purposes. In addition to the inherent scientific hurdles, there are an immense number of practical considerations—starting with the fact that FDA has yet to issue a framework defining evidence for regulatory-quality biomarker validation.

Other nettlesome practicalities can bog down a collaboration. For instance, drug firms only rarely start research programs at the same time. Some companies accumulate data and expertise much faster than others. And it's unlikely that any two companies will enter a new collaboration with equal amounts to contribute. In the noncollaborative past, firms developed their own patient datasets and used the information to develop models or qualify biomarkers. No one gave information to a competitor—or got any back.

This ratio of information given to received can make or break any proposed collaboration. Imagine a scenario in which four companies have, over several years, prepared and submitted to FDA four datasets. The firm that submitted first has the largest dataset at 10,000 patients, while two have 3,000 each, and the smallest has 2,000. In the name of scientific collaboration, FDA or another intermediary might have the job of making the data (or models or analyses) available to the later submitters. Clearly, the later-contributing, smaller-dataset firms have an advantage many times their contribution. At the same time, the first firm is grossly disadvantaged—and deserves special compensation for such a lopsided arrangement.

According to a recent survey by Don Nichols, clinical pharmacology site head at Pfizer at Sandwich, most companies report that they are willing to share summary-level placebo or active-control data, but they remain close-fisted about detailed information on individual patients taking the investigational drug (see "Share the Wealth"). The detailed statistics for patients on placebo or a control drug can be used to build drug and disease models, and the benefit comes in reducing the number of new patients needed for trials and creating cheaper, faster designs than would otherwise be possible.

FDA's pharmacometrics group is busy building and sharing models from aggregated sponsor submissions of patient data. For a variety of legal reasons, the data used by FDA must be held confidential. By sharing the model designs—but hiding the data—FDA can finesse sensitive patient-confidentiality issues and make available a number of models to promote progress in diseases such as HIV, Parkinson's, obesity, non-small-cell lung cancer, type II diabetes, osteoarthritis, and Alzheimer's.

Still, there is no budget or infrastructure for data sharing, and an organization serving as data custodian would require scientific leadership and serious funding to get off the ground. In fact, there is only $6 million in the FDA's 2007 budget for all 76 CPI items. And barring some radical upheaval on Capitol Hill, the funding and resources for serious collaboration will have to come, inevitably, from pharma.

Companies are beginning to show some enthusiasm for adapting model-based drug development. For example, Novartis recently elevated modeling and simulation to the status of a 35-scientist department; GSK, Pfizer, BMS, and other big firms have established organizations with similar stature and presence. Meantime, FDA doubled the size of its pharmacometrics group to 10 scientists, and the group reports that it was involved with 12 percent of total submissions in a recent 15-month period.

Industry, FDA, and modeling vendors are taking other steps to hone their model-building skills. For one, new PK data-management systems will up both the quantity and the quality of clinical pharmacology data. These repositories represent a major investment but promise to accelerate retrieving, manipulating, and validating data references. With this technology, datasets can also be shared more easily.

FDA has proposed a new collaborative regulatory process, the End of Phase IIa Meeting, to discuss modeling, trial design, and other MBDD issues before Phase III program decisions are set. The agency's aim is to promote MBDD techniques for the design of trials, not just for analysis leading to the exact wording of labels and FDA regulatory decisions. A pilot consisting of 10 to 12 meetings in various therapeutic areas was completed last September, and the feedback was positive. But FDA is far from having sufficient staff to do End of Phase IIa meetings for every drug candidate.

On the Path to Pharma's Future

The intense pressures on drug companies are forcing new ways of approaching drug development. Successful collaboration in HIV lit the way, and some firms are repeating its lessons in other areas. Tightly focused, professionally managed collaborations, such as the biomarker and drug-safety initiatives, are showing promise. More than that, they are persuading members to share nonproprietary data and overcome ingrained tendencies to keep their intellectual property under lock and key. More and deeper collaborations lie ahead. Meantime, relationships between sponsors and FDA are becoming less adversarial through the End of Phase IIa meetings and other processes, creating new channels for additional communication and collaboration on program designs.

But all of this is only a beginning. The obstacles in many cases are immediate and easy to quantify, while the benefits—however huge in theory—are hard to put numbers to and not exactly on a next-day-delivery schedule. Companies don't come into collaborations with the capability to contribute equally. They must find incentives to overcome potentially crippling "free rider" problems. Funding and staff remain high hurdles, especially at FDA. Legal permissions needed for sharing take time to put in place. But any long-view risk/reward calculus leads to only one conclusion: Collaboration must not only continue but increase. The biggest risk for industry is not taking a risk. We have no choice but to walk the Critical Path together.

Dan Weiner is a senior vice president and the chief technology officer at Pharsight dweiner@pharsight.comMark Hovde is the senior vice president of marketing mhovde@pharsight.com