Knowledge-accelerated drug development

Abstract

Knowledge-accelerated drug development is a step-by-step process that companies can adopt to take full advantage of state-of-the-art computer-assisted technology. At an operational level within projects, research teams can use knowledge-accelerated drug development to assess different trial designs by accounting for many more variables that affect outcome than traditional methods can accomplish. That allows the team to choose the best design. Moving from current clinical planning methodology to new technology that provides better assessments can occur in 4 sequential steps: 1. computer-based trials management, 2. computer-assisted trial design, 3. computer-assisted clinical program design, and 4. computer-assisted portfolio management.

Headnote

Pharmaceutical companies strive to develop new drugs as quickly and efficiently as possible. A step-by-step process shows how to use computer-assisted technology to reduce development time and improve the yield of useful knowledge derived from individual clinical trials.

Developing innovative products for unmet medical needs in a short time frame is the main goal of R&D in pharmaceutical companies. Uniformly, companies are setting development standards to reduce clinical development times to half of what they were a few years ago. That means that companies will typically strive to formulate a new drug, plan and conduct clinical trials, and bring the drug through FDA approval within a fourto six-year time frame.

To consistently meet this goal, major shifts are necessary in the methods by which companies formulate and execute clinical development plans. Any clinical trial that fails to reach a definitive conclusion will cause unacceptable delays. Companies try to carry out as many clinical trials in parallel as possible without pursuing dangerous shortcuts. Accurate methods to assess the relative net present values of products competing within a company for development dollars are also essential. Although sophisticated economic and financial models are currently available and used by companies, equally sophisticated planning tools for clinical development are less widely used.

Until recently, innovations in computer software used in clinical development have focused on data capture and storage. To perform an optimal assessment of each clinical project's timelines and probability of success, companies need software that goes well beyond traditional data collection, storage, and analysis. New technology is becoming available that improves the yield of useful knowledge derived from individual trials, thus allowing what can be termed "knowledge-accelerated drug development."

Knowledge-accelerated drug development is a step-by-step process that companies can adopt to take full advantage of state-of-the-art computer-assisted technology. This technology allows employees with different expertise and operational and managerial functions to access a rich repository of knowledge. They can use that knowledge to design clinical projects and manage a development portfolio. Executives can then explicitly quantify and analyze decision trees.

The technology also allows companies to test sensitivities of outcomes of clinical trials and programs to assumptions so that they can ascertain the overall probabilities of success. At an operational level within projects, research teams can use knowledge-accelerated drug development to assess different trial designs by accounting for many more variables that affect outcome than traditional methods can accomplish. That allows the team to choose the best design. State-of-the-art information science is now providing tools that elevate clinical planning to a higher level of sophistication.

This article describes how these novel clinical planning tools work and how they contribute to designing a smarter, more informative drug development process.

Moving from current clinical planning methodology to new technology that provides better assessments can occur in four sequential steps-from the current state-ofthe-art, computer-based trials management to the highest level, computer-assisted portfolio management.

STLP

Computer-Based trials Management

In the existing state-of-the-art method, a clinical team distributes information orally and in written documents. Its data analysis of that information focuses on elucidating descriptive facts and testing hypotheses. Statistical planning addresses potential sources of bias and power calculations. Even with that type of analysis, however, clinical teams miss opportunities to more fully use the richness of existing data.

The planning and modification of sequential clinical trials in a complex development program relies on a progressive accumulation of clinical data and a synthesis of conclusions from that data. A project team encompassing a wide array of disciplines, including medical, clinical pharmacology, biostatistics, project management, and regulatory affairs, plans the overall clinical program. The team ensures that each department is ready to perform its assigned tasks so that development proceeds in an integrated fashion. The project director will usually define items on the critical path and manage them. Typically, the project director uses project-management software to facilitate the process.

Power analysis software is the main tool clinicians and statisticians use in designing trials. Power analysis examines estimates on the magnitude of the drug's effect on the disease-state, the difference in effect between treatment arms, and the variability of the measured efficacy variable to calculate the number of subjects to be included in a trial.

However, power analysis per se fails to allow companies to assess simultaneously the relative importance of the individual factors that contribute to the variability of response. Those factors include choice of end point and variability in measures of the end point. They also include sensitivity of the end point to the frequency of measurement, to compliance, to duration of therapy, to interacting medications, and to the timing and rate of patients dropping out of the trial. Although the designers of clinical trials have some sense of those effects, it is very qualitative. Thus, the final design of a clinical trial emerges from the best-educated guess that an experienced, highly trained trialist can make.

Certainly the existing process yields a high number of successful trials and development programs. The current goal of many companies is to design trials and programs that result in the most informative analysis of results using modern techniques of managing complex information. Companies can apply appropriate technology, such as that described below, to facilitate the process of managing information and examining data. The application of such technology results in enhanced clinical judgment because of a more accurate, quantitative understanding of the effects of many variables on the outcome of a trial.

Computer-Assisted Trial Design

A clinical trial has "failed" if it provides an ambiguous answer to the question the trial is trying to address. Trial designers, as described in Step 1, currently rely on power analysis to estimate the number of patients to include in a clinical trial to obtain a statistically significant result. Power analysis bases its assumptions on the variability of response, and the magnitude of the response in the control and experimental groups. The drug, therapeutic indication, and trial design are factors that affect these assumptions.

Computer-assisted trial design (CATD), building on Step 1, allows companies to explore the effects of those factors on the outcome of the trial and to design a trial that is most robust to those factors. CATD allows companies to build a drug- and disease-specific knowledge base and to better estimate how a drug will interact with human pathophysiology. It also allows companies to perform "what if' scenarios to optimize clinical trial properties by exploring a variety of alternative designs and quantifying which design variable will have the biggest influence on the results of a trial. The following examples illustrate potential uses for CATD:

The optimal treatment schedule of a drug. The efficacy of an anti-arrhythmic is well correlated with the pharmacokinetic area-under-the curve. To be adequately effective, the drug must prevent arrhythmias in 90 percent of patients. CATD allows companies to calculate the probability that a particular dose and frequency would accomplish that. Companies can then evaluate a whole range of dose and frequency combinations by CATD and test in early Phase II trials a few regimens that are clinically and commercially acceptable.

The range of clinical responses. Within patients with moderate hypertension, suppose that Phase II studies indicate that 55 percent of patients have excellent control of blood pressure, 25 percent have adequate control, and 20 percent have poor control. Suppose also that with an increment in dose, the percent of patients with poor control drops to 10 percent, but there is an increase in patients experiencing nausea. CATD indicates that if a patient takes a higher dose, a less favorable balance between efficacy and toxicity will result. That is partly because patients are becoming nauseated and failing to take all doses. In addition, CATD predicts the drop-out rate will be too high. Thus, higher doses are excluded from the Phase III studies.

The effect of poor compliance and methods to minimize the effect of poor compliance. Suppose that Phase II trials show that a drug for HIV decreases viral load and increases CD4 counts. There is a strong suggestion that the incidence of candida infections is reduced as a result. However, patients need to take the drug three times a day for optimal efficacy. Thus, a certain degree of noncompliance may occur. CATD allows a company to quantitate how such behavior will affect the outcome of the trial. It informs the team that by increasing the dose by 20 percent, the effect of a skipped dose will be minimized without an undue increase in toxicity. The team decides to include that increased dose in a trial. In fact, it is that dose that is most effective.

The frequency of measurement of a response. CATD determines that measuring serum cholesterol monthly will give results for a new lipid-lowering agent that are the same as measuring cholesterol every two weeks or weekly. In addition, CATD predicts that after two months of therapy, further reductions in cholesterol are very unlikely to occur. Thus, a company plans a Phase III trial as a two-month trial with collection of cholesterol at baseline, month one and month two. It scuttles a plan to measure cholesterol weekly for two months and monthly for six months, and a faster and cheaper trial results.

The effect of inclusion/exclusion criteria on outcome. An antidepressant excreted mainly by the liver has hastened elimination in patients consuming alcohol because of enzyme induction. But that effect is variable and impossible to predict in any individual patient. The targeted patient population for the drug includes a moderate number of alcoholics. If alcoholics are included in the first major Phase II study, there will be more variability in the clinical end point because of the effects of alcohol on the drug's elimination.

The team must decide whether to include alcoholics. By excluding them, a company can conduct a smaller trial because there is less expected variability in outcome between patients. CATD, by simultaneously analyzing the variabilities in pharmacokinetics and clinical effects, determines that, to obtain a definitive result, the size of the trial including alcoholics needs to be 20 percent larger compared with a trial excluding alcoholics. The team includes alcoholics in the trial, and thus it has broader applicability. The team would have opted to exclude alcoholics if it required an increase in sample size of greater than 50 percent. Including alcoholics would increase the duration and cost of that trial, which the company would review to make a go/no-go decision for the program.

CATD allows companies to conduct "virtual clinical trials" on a computer so that they can implement the trial most likely to give a definitive answer. (See "Path to Prediction.") The clinical design team can agree on assumptions of drug, disease, and population models. They can then review the trial structure, treatment schedule, clinical end points, compliance, and dropout assumptions. They can also run multiple trials using a virtual patient population and analyze the "results" graphically or with standard statistical tests to help predict the range of plausible results from a clinical trial. The team can review the results and modify the model assumptions or trial design and repeat the sequence.

CATD incorporates a sophisticated model of a drug's pharmacology, including information on pharmacokinetic-pharmacodynamic relationships, the target disease's pathophysiology and natural history, and the characteristics of the target patient population. CATD software can access a company's existing data base and analysis software. It also provides a user-friendly interface making complex data more accessible and useful to preclinical scientists, clinicians, and senior management.

Using current practices, experienced planners of clinical trials can simultaneously consider only a select number of factors that could affect a trial's outcome. Taking advantage of modern information technology, CATD allows companies to simultaneously consider how multiple factors would affect the outcome of a trial. It also allows them to explore the best doses to use and the choice of end point, which is especially important when surrogate end points are used in Phase II.

Companies can also explore the sensitivity of the end point to a number of other factors, including the frequency of measurement, compliance, placebo effects, the natural history of the disease being treated, duration of therapy, interacting medications, and the timing and rate of patients dropping out of the trial.

Another important application of CATD is in pharmacoeconomic studies. The commercial viability of a drug increasingly depends on demonstrating an economic as well as medical benefit. CATD can incorporate a pharmacoeconomic model and allow companies to adapt Phase III and IV trials to optimally capture informative economic data. Pharmacoeconomic modeling depends heavily on mathematical models, which are developed and analyzed by computer technology.

The integration of such models into CATD is seamless because CATD is also based on using models to predict outcomes. Thus, clinicians could build a pharmacoeconomic model into CATD and integrate it with the pharmacodynamic model.

Researchers can automatically enter efficacy data from the pharmacodynamic predictions produced by CATD into the pharmacoeconomic model. That will produce estimates of pharmacoeconomic outcomes in parallel with the plausible range of clinical outcomes predicted by CATD.

Suppose that the development team wants to predict the pharmacoeconomic benefits of a new anticancer therapy. As the team tests the clinical benefits of the drug, they could simultaneously test the economic benefits based on the clinical benefits.

Computer-Assisted Clinical Program Design STRP

Computer-assisted clinical program design (CACPD) integrates CATD with decision sciences. Although companies use CATD to plan one clinical trial, CACPD incorporates assumptions across many clinical trials in a given drug development program. The goal is to optimize a program's design and provide strong justification for its timelines and overall approach. CATD allows companies to design a series of clinical trials that are likely to lead to clear go/no-go decisions, such as differentiating between a failed trial and a failed drug in the following example:

A drug targeted for treatment of rheumatoid arthritis is only slightly better than placebo in a Phase II study. One assumption used in planning the study, based on preliminary data in humans, was that the duration of therapy needed to show an effect was eight weeks. However, after the trial ended, researchers noted that only one-third of the effect expected was actually observed.

A pharmacokinetic trial carried out in parallel with this trial demonstrated that the area-under-the curve in the first week of therapy were higher than areas-under-the curve after six weeks of chronic therapy. When researchers used CATD to combine the effects of the pharmacokinetic trial, the Phase II efficacy data, and concentrations gathered in the Phase II trial, the results showed that patients would not reach a steady state effect until 12 weeks. Researchers concluded that they should use doses 50 percent higher in the trial. Thus the company can continue the project and repeat the Phase II trial with a longer schedule for treatment and efficacy measurements.

CACPD can predict which trials are necessary to provide information needed for the success of subsequent trials. The insights CACPD provides on the comparative risks associated with different trial designs, the sequence of executing trials, and the need to focus on particular aspects of data collection can help management plan an optimal and efficient clinical development program.

Companies can implement CACPD at any stage in drug development, depending on their needs and interests. However, using CACPD throughout the development process would help find a balance between doing too many trials and too few to complete a development program that gains regulatory approval.

CACPD predicts the probability of success of a clinical program to demonstrate a new product's efficacy and safety. It does that by evaluating the analyses performed by CATD for each trial. Thus, as the clinical team considers changes in the design of each trial that makes up the clinical development plan, the CACPD software can show how a changed design for one trial-with its commensurate change in the information gathered and the data received-would change the likelihood that the program will reach its goals.

That is possible because CACPD can quantitate how a change in one trial will affect the outcome and the timing of the whole program. For example, consider a program for a new analgesic that, in preclinical studies, had a very rapid absorption and onset of action. The planned development program for this drug consists of the following:

two dose tolerance studies, one in normal volunteers and the second in patients

small Phase II proof-of-concept trial to test time to onset of maximal effect of the drug in patients undergoing dental surgery

a Phase II study to demonstrate efficacy adequately to make a go/no-go decision whether to move to Phase III

two Phase III trials in patients undergoing dental surgery

a trial in renally impaired patients

a trial in hepatically impaired patients

a pediatric program consisting of

Phase II and II trials in children. Researchers plan each study using CATD and apply CACPD to the whole program. After analyzing the dose-tolerance studies, the development team uses CATD to test whether enlarging a Phase II study to include both children and adults would shorten the program compared with running the trial first in adults and using the adult data to predict the outcome in children. The first alternative seems attractive because it could shorten the whole program. It turns out that CATD demonstrates that a relatively small amount of data in children would be possible to predict the pharmacokinetic/pharmacodynamic response in children using adult data properly scaled to children. The team then decides to change the design of Phase II so that a major goal is to define the relationship of adult and pediatric dosing and response.

This relationship, in turn, would allow the team to apply efficacy data from Phase III to children as well as adults. Because many more adults undergo dental surgery than children, the data in adults could efficiently lead to adequate data for children as well. Thus, the development team applies CATD to each trial and CACPD to the whole program. By doing that, they discover an efficient path to demonstrate effcacy in children as well as adults more quickly than management had originally surmised.

A company can explore a range of program alternatives when reacting to an unexpected advanced competitor. (See "Exploring Alternatives.") It can also explore how many trials a team will really need to gather enough information to warrant approval and to produce a label that will meet clinical and marketing needs. Using CACPD, a clinical management team can determine how changes in results or assumptions in one trial would affect the outcome in other trials planned in the program. That is helpful in contingency planning and in assuring the team that they can recognize risks to the overall program because of untested assumptions.

The team can compare actual results within a development program to predictions made from computer-assisted trial and clinical program design. By using linked CATD analysis for individual studies, the team can use CACPD to evaluate the probability of success of running some studies in parallel and/or some in sequence. The team can use forecasts of clinical results to better quantify risks and returns based on the likelihood of outcomes of a clinical program.

CACPD that incorporates CATD for individual trials followed by decision analysis has multiple and related applications, including the following hypothetical examples:

Time to onset of a measurable effect: A company is considering developing a drug that has analgesic properties to treat migraine headaches. However, to be commercially competitive, the onset of clinically significant pain relief must occur within 15 minutes of taking the drug. CATD demonstrates that this is very unlikely at tolerable doses. The company puts the study on hold because they deem that another study for a different drug at the same stage of development as the analgesic is more likely to lead to a successful program.

Analyze the importance of specific information to the outcome of a trial: CATD demonstrates that a favorable outcome for a drug to prevent rejection of a renal allograft in the maximum fraction of patients is dependent on the drug inhibiting an enzyme in lymphocytes by a certain percent. It would then be important to know, in humans, how long such an effect lasts after each dose, so that researchers can choose an optimal dosing interval. Thus, it would be prudent to investigate this in a small Phase II trial before embarking on larger trials.

STRP Computer-Assisted Portfolio Management

Computer-assisted portfolio management (CAPM), the highest level of computerassisted technology, builds on CACPD so that the integration of planning clinical development occurs throughout the development organization in all projects. Senior management can readily look at how empirical scientific data and the likelihood of obtaining such data affect the overall outcome of single programs and the whole enterprise.

They can integrate clinical information with market research and consider it in the context of an evolving scientific, economic, and social environment. Companies can then assess the probability that each program will achieve its goals and the effects of different resource allocation on these probabilities. They can also use the probabilities of success to weight calculations of net present values. CAPM allows comparison across programs by using new data to update each product's net present value. Management can quantify the risk/ reward ratios of programs competing for resources and change tactics based on the updated assessments of the ratios.

Consider a hypothetical example of a company's portfolio of drugs in clinical development. (See "Probable Outcomes.") Executives can use CAPM to explore the trade-offs of value and probability of approval based on likely outcomes of clinical trials.

In this example, each project is represented by an application of CATD and CACPD to create an estimate of the clinical development costs and the likelihood that a clinical program will reach a particular conclusion. A company can then evaluate those activities against the probable market value expectations to derive a Net Present Value and probability of approval. Management can use a team approach drawing on the company's proprietary scientific knowledge base to quantitatively assess the probability of approval based on the expected results. Each of the assessments uses explicit assumptions and outcomes of trials based on those assumptions to arrive at the probability of success.

In the hypothetical example, management teams continually assess the probabilities of success of each program in CAPM. The results showed that a drug for Alzheimer's disease has a high value and moderate probability of approval; however, the investment of resources is large. In contrast, a weight-reducing drug has a similar value and lower chance of success but requires less investment. A male contraceptive has low value in the marketplace and a low chance of approval. An antihypertensive drug with a new mechanism of action has a good chance of approval and would be of moderate value.

Finally, an anesthetic has a good chance of approval but little value. Thus, companies can use that knowledge to make investment decisions. Also, the management team can make assumptions about a program and then assess how sensitive the results would be to each assumption.

Suppose the market value of the antihypertensive would increase fivefold if people could use it as sole therapy. CACPD and CATD inform the group that there is an 80 percent chance that this would be the case based on assumptions about how animal data translates into human response. It would behoove the company to test that hypothesis in humans as quickly as possible. On the other hand, if CATD and CACPD indicate that the chances of the drug working as sole treatment of hypertension are 10 percent, the urgency of testing that drug may be less compared with the urgency of other projects under consideration.

State-of-the-art information technology is now available that allows more complex analysis of clinical trial designs, development programs, and portfolio management than is currently in use in the industry. Knowledge-accelerated drug development allows those who design and manage clinical development to consider more explicitly a wide array of factors that affect outcome.

This approach will have the greatest value when applied to all trials in a program and to all programs. Companies can then better quantify and manage uncertainty of clinical results. Accordingly, they can optimize trials and programs to give definitive answers to the questions required to gain regulatory approval of a commercially attractive product. Computer-assisted trial design is the heart of this new

methodology.

Pharmaceutical Executive encourages readers to respond to this article. Write PE at 859 Willamette Street, Eugene, OR 97401-6806, or e-mail us at [email protected].