Pitfalls in Drug Development: 5 Wrong Choices That Can Tank a Program

These common mistakes can derail the long and arduous journey to market for a drug candidate

We’ve all heard the numbers?- ten years, a billion dollars, and a 90% failure rate. Why is it so slow, expensive, and risky to take a drug candidate through the approval process? In some ways, it can seem very counterintuitive. After all, you have a candidate drug which you know inhibits the molecular target that is responsible for the disease you’re focused on, you’ve tested it, and it seems to work in animals without making them sick. All that remains is to try it out in humans and see what happens. If it works- great, and if it doesn’t, you move on to the next molecule, right?

This “go-big-or-go-home” narrative sounds appealing, but there's more to the story than that. Here’s the thing- when taking a new drug into the clinic (and through the approval process) there are a large number of choices that project teams are tasked with making. These choices (or questions if you prefer) can make the difference between approval and failure. It goes without saying that these are often life-and-death choices for small biotechs and can make a decisive difference to the career of a Project Lead in Big Pharma. So, rolling the dice isn’t a good idea, because the stakes are high and the choices can determine the outcome. On the other hand, knowing what questions need to be answered, and how to answer them quickly and correctly- that can stack the deck in your favor.

So, what are these questions then, and how might one go about answering them?

Choices in Drug Development

The point in a drug R&D program when the candidate drug has been identified marks the transition between discovery and development. At this stage, the project team’s focus is preparing for the first-in-man trial. This is done by putting together an investigational new drug (IND) application (or its European equivalent, the Clinical Trial Application). The first-in-man trial requires a number of choices to be made. Making these choices then sets in motion the operational aspects of the clinical trial.

Let’s take a look at what that means, in practical terms. As an example, we will use a hypothetical small-molecule drug for cancer, although what we will be discussing is equally applicable to almost any other therapeutic modality and indication.

• Pick the dose route

The dose route is usually the first choice that needs to be made, prior to the initiation of animal studies. Dose routes are often selected very early in a program, long before the drug is identified, but there is room for changing one’s mind at this stage as well. Dose routes are often selected for convenience (everyone wants an oral drug!)

• Pick the dose schedule

The dose schedule is typically zeroed in on during the animal studies. Choice of dose route will constrain the schedule somewhat (for example, opting for an intravenous (i.v.) dose route rules out daily dosing, for reasons of patient (in)convenience), but usually there is a wide range of options for schedule that need to be narrowed down.

• Pick the right indication and patient population

The trial should focus on the disease and patient population where the candidate drug’s mechanism of action is most likely to yield a therapeutic benefit. Many Oncology programs have a wide range of options for indication at this stage. However, in a non-Oncology setting, the indication is usually selected much earlier, at the time of program initiation.

• Pick the right starting dose

The starting dose is one of the most crucial choices, and it’s one that the regulators will scrutinize closely. From the regulator’s perspective, it is imperative that the starting dose for a clinical trial be safe, and that the early dose levels in a clinical trial provide the clinician with some forewarning of the expected clinical toxicities before the maximum tolerated dose (MTD) is reached. The traditional approach for selecting the starting dose for humans is to conduct toxicological studies in a couple of different animal species (typically dog and rat), identify the doses associated with a minimal level of toxicity and then use pre-determined formulas to convert to a human dose. However, this approach doesn’t work for a number of drug categories (e.g. biological drugs such as antibodies, immune modulators), and the science around this has evolved in the past decade or so, necessitating different approaches in some cases.

• Pick the right dose escalation scheme

Again, this is a crucial step in the process, and one that will face scrutiny from the regulators. The traditional approach to this is to dose three patients, and then observe them for toxicity. If no toxicities are observed, the dose is increased (usually by anywhere from 50 to 100%) and another three patients are dosed. This 3+3 dose escalation scheme has the advantage of being are easy to explain but performs poorly in terms of getting to the actual MTD without over-dosing patients.

Of course, these are by no means the only pitfalls capable of derailing a drug R&D program. By the time a project team is tasked with taking a candidate drug into the clinic, a number of choices have already been made. The molecular target of their drug candidate is already locked in, as is the chemical matter. A chunk of the difference between success and failure lies in the choice of target. If your candidate anticancer drug is a farnesyltransferase inhibitor or an angiogenesis blocker, for example, it’s all uphill from here! Similarly, (many) billions of dollars have been spent on the beta amyloid hypothesis for Alzheimer’s disease, which has failed to yield therapeutic benefit after three decades of research. For a first-in-class molecule, the choice of target represents an irreducible nub of risk that cannot be addressed except after the Phase II data comes in. Thus, in that sense, the clinical trial process is really a test of the biological hypothesis: “inhibiting my target protein will provide therapeutic benefit for this disease.” Making the right choices during development provides the best chance of testing the biological hypothesis underpinning the program.

Development Choices are Really Easy to Make!

Here’s the thing though- the five choices that we discussed in the previous section can be made by a project team in minutes! How is that possible? Let’s walk through it with our example of a small-molecule cancer drug (e.g. an Oncogene-targeting drug or an immune checkpoint modulator).

• Dose route? Let’s go with oral- oral drugs have the greatest?market potential.
• Dose schedule? How about daily, that’s a convenient frequency.
• Indication and patient population? If it’s an Oncogene targeting drug, let’s look up the literature and find the cancers that most commonly overexpress that oncogene (“addicted” tumors), and then we can select the overexpressing population with a personalized biomarker too! (If it’s an I/O target, we can go with ‘hot tumors’- melanoma, NSCLC or bladder cancer for example. We can look up the literature again to see if there are genotypic markers designating hot tumors for the specific I/O target).
• Starting dose? Run toxicology studies in dogs and rats- pick the starting dose according to the formula specified in the FDA’s (ICH S9) guidance document.
• Dose escalation? That’s easy, let’s go with a 3+3 escalation scheme!

See how simple that was? Five minutes and we’re ready to go. A plan like this will typically pass muster with most people unless they are deep diving into the weeds of the program. From a regulator’s perspective, following the ICH S9 guidance is a reasonable and conservative choice, and the 3+3 dose escalation scheme is extremely well precedented. It’s a good, safe plan- one that is unlikely to raise eyebrows.

Beware the Dartboard

The life story of a typical program advanced on a plan like this is pretty predictable. The first several cohorts of patients will see no effects at all. Then there will be patients for whom some toxicity is observed. A few cohorts later (if all goes well), the MTD will be reached. There may be zero, one or two responses observed. Some stable disease as well, and then everyone can debate whether that stable disease was due to the drug. This is what the vast majority of Oncology Phase I trials look like. Fifteen to forty patients (and several million dollars later), the project team is left reading the tea leaves in a best-case scenario.

Here’s why this happens: beyond the safety data, a first-in-man Oncology trial like this is actually asking two questions. “At the clinical MTD, are we inhibiting the target strongly enough to shut down the pathway?" and "Are the tumors that we think are sensitive to the pathway being shut down actually sensitive?"

But, having made our choices by throwing darts at a board, essentially, there’s a lot that we don’t know when we reach MTD: is the biological hypothesis ('oncogene addiction'/ 'hot tumors') valid? Did we actually choose the best dose route and schedule? Was there enough drug on board in the tumor in patients to be able to make a yea-or-nay assessment of the validity of the biological hypothesis? Is the designated MTD actually the highest dose that could have been achieved, or did we miss the true MTD because of random luck?

Many, if not most, Oncology programs die at this stage, because – without some luck- there is no clear signal suggesting a path forward. So even a trial that doesn’t exactly fail outright – in many cases- will leave the project team with insufficient reason for enthusiasm (crucial for attracting further investment) and insufficient guidance on what to do differently in hopes of a better outcome. So yes, the choices were easy to make, but they led smoothly to a No-Go decision at the end of Phase I.

Let’s examine each of the choices again, and see what could go wrong, and how to avoid it.

Choosing the wrong dose route

Few choices are as consequential for the ability to inhibit the target as dose route. Yet, project teams tend to select oral dosing in a knee-jerk way, based on the premise that patients will find it more convenient (which it is). The oral dose route can lead to poor control over the Cmax (peak concentration) of a drug in a patient population and will lead to variable amounts of total drug as well (Area under the Curve, or AUC). Continuing with Oncology as our example, there are many drugs for which AUC is the pharmacokinetic (PK) parameter that best predicts efficacy (“the PK driver of efficacy”), and Cmax (or something similar) is the PK driver of toxicity. In this situation, orally dosed drugs, with a variable dose-to-Cmax and dose-to-AUC ratio, will lead to poor control over the achieved level of toxicity and efficacy for each patient. For drugs with a narrow therapeutic window (the gap between efficacy and toxicity), this can mean that for a given dose, some patients are being underdosed, others are being dosed in the sweet spot, and others are being overdosed. While convenience is a major driver of commercial success, having an unfavorable therapeutic window will lead to outright failure during the development process. (Put differently, a convenient drug will have a larger market share, but sacrificing probability of success for convenience is not a good tradeoff, as the market share of a failed drug is always 0%). Thus, for the example of Oncology drugs with a narrow therapeutic index, oral dosing may not always be the best choice. On the other hand, there are many therapeutic areas where oral dosing is the only rational option (i.v. claritin, anyone?). It’s important to understand the anticipated PK and the pharmacodynamics (PD) of the candidate drug in humans before committing to the choice of dose route.

Choosing the wrong dose schedule

Dose schedules are also often selected for convenience, with the same pitfall- optimizing for market share for a drug before it has reached the market assumes a low risk of failure in the drug development process. Which is not how it works. Often (particularly in Oncology), the schedule used in the animal studies will be selected for the clinic. Unfortunately, because of species differences in PK, this can lead to a failure to correctly project the time-on-target behavior of the drug in humans. As discussed above with dose route, projecting the expected PK/PD behavior of the drug in humans is important for the rational selection of dose schedule. The right schedule will optimize for maximizing the PK (PD) driver of efficacy, while minimizing the PK (PD) driver of toxicity. So, for example, choosing an every-other-week i.v. bolus for a drug with Cmax-driven toxicity may lead to a smaller therapeutic window than can be achieved with a schedule that uses an i.v. infusion, or doses more frequently.

Choosing the wrong indication or patient population

A paradox of drug discovery and development is that often the link between biological mechanism of action and patient population selection is quite tenuous. This is particularly true, for example, in Oncology. For Oncology, much of the primary literature – particularly biochemistry and genetics- is poorly reproducible. In addition, there are fundamental differences between cellular behavior in tissue culture and in in vivo models (and patients). Project teams often make the mistake of locking into a patient population or indication based on a particular mechanistic rationale (for example, oncogene addiction). It’s always possible to generate in vivo data to support such a rationale. For example, say that a particular marker (K-Ras) is thought to be required for tumor response to the inhibition of your target. Say your candidate drug gets a response in about 50% of the tumor models you run it in. If K-Ras is perfectly non-predictive, you can expect 25% of the in vivo models that you run to be both K-Ras positive and responsive to your drug. Run enough tumor models, and you’ll have the ‘killer slide’ that you need. The problem with this approach is obvious, though- it sets the clinical program up for failure! It is in a project team’s interest to keep an open mind, casting the net wide for responsive patient populations. Designing trials to be able to squeeze out as much information as possible in the Phase I/II setting is one way to avoid this pitfall. This can be done, for example, by using disease progress models to estimate drug effect directly from patient data in the clinic. (We have white papers on this topic, so do ask me if you want to know more!)

Choosing the wrong starting dose

The wrong starting dose can derail a program in two ways- only one of which the regulators will help you avoid. A too-high dose can be a disaster in the clinic, but the FDA (or EMA) regulators will be more conservative than you are. On the other hand, a too-low starting dose can also derail a program. If the starting dose is off by a large amount (say a couple of orders of magnitude), that would result in seven additional cohorts, extending a clinical trial by up to several years, depending on the other challenges in the program. In addition to increasing cost, this can reduce investigator enthusiasm for your candidate drug. Thus, choosing a safe, scientifically rational starting dose is crucial for program success. (Check in with me if you want our white papers on this).

Choosing the wrong dose escalation scheme

Dose escalation schemes are a crucial part of trial design and have been an area of active innovation in the past decade. The traditional 3+3 design is still common, but its weakness is that it misses the ‘true’ MTD by a gap that is half the size of the dose escalation interval. Smaller dose escalation intervals allow for a more accurate assessment of the MTD, but can slow a program down, and a variable dose escalation interval can be difficult to implement in practice. Newer methods like the Continuous Reassessment Method (CRM) use powerful Bayesian statistical techniques to improve the efficiency of estimating the MTD, but these methods are often misapplied in practice, resulting in outcomes that are often no better than a traditional 3+3. (Again, we have white papers on this topic, so ping me if you want to know more!).

A better path forward

By now you’ve probably figured out that the apparent simplicity of each of the five choices we discussed in the beginning is a trap. It’s easy to make the choices by simply throwing a dart at a(n appropriately labeled) dartboard. However, doing so will turn the choice into a pitfall that can sink your drug development program.

The reality is this: each of those five choices for a drug development program should be thought of in terms of a hypothesis. Each one can be addressed by a lean, focused analysis that (typically) combines in vivo or translational pharmacology with mathematical modeling and statistics. For example, dose route and schedule require a set of experiments to identify correctly. There are techniques such as MABEL (Minimum Anticipated Biological Effect Level) that provide a powerful and scientific way of defining a starting dose, but again, these require a focused pharmacology campaign, coupled with careful modeling. Providing science-based justifications for each of the five choices leads to a strong IND application package.

The time and effort taken to put together a well-thought out and rigorous IND package can pay for itself many times over during the clinical program. A hallmark of a robust development program is that success and failure criteria are mapped out clearly, and the trial is designed to gather the information required to make those decisions in a rigorous way. While luck still plays a role in any drug development program, paying close attention to the key choices on the path to the clinic can minimize the time and money wasted in clinical development and maximize the chances of success.

Model-based drug development techniques tie preclinical and clinical pharmacology together tightly to map out a clean and quick path to go/no-go decisions at every step along the critical path for drug R&D. In my next blog post, we will dive into the details of how to pull together a tight IND package using a model-based approach.

In the meantime, if you want to learn more about this, we have a number of additional materials and publications that touch on the topics mentioned here. If you go over to our webpage: https://fractaltx.com, you will find white papers that provide more detail on how to use model-based drug development techniques effectively in the clinic. Alternatively, you can just ping me on LinkedIn!