Once the chemical structure of penicillin was established, chemists began work on synthesizing new versions, each with its own distinct properties (Figure 9-1). Penicillin-resistant bacteria produce β-lactamase enzymes that attack the penicillin structure. A simple modification of the structure of penicillin created a bulky chemical chain that blocks β-lactamases from the β-lactam site, resulting in the β-lactamase–resistant drug cloxacillin.

Figure 9-1 A, Penicillin. B, Cloxacillin. C, Amoxicillin.

Preclinical Testing

After a compound has been synthesized, and appears to have efficacy, fine tuning is then done, focusing on the following issues.

Pharmacodynamics

Tests are performed in vitro to determine receptor binding affinity. Investigators at this point are interested in how specific the drug is for its receptor target, as this will provide an indication of the potential for the drug to cause side effects. Generally speaking, the less specific a drug is for its receptor, the greater the incidence of side effects.

Pharmacokinetics

A key step in the development process is to determine the route of metabolism for the drug. If metabolized by CYP450 enzymes, the isoenzymes involved will be identified. Tests will also be performed to determine whether the drug is an inducer or inhibitor (or neither) of metabolizing enzymes, again with a focus on CYP450.

Being an inhibitor or inducer of metabolizing enzymes is generally considered to be a weakness for a new drug, owing to the potential for drug interactions. Early tests can be performed on isolated enzymes in vitro, but in vivo testing will inevitably be performed to determine the effects of a drug in a whole animal.

Pharmacokinetic studies will also provide the elimination half-life of a drug, and this information will be used to determine the administration frequency of a drug. The gold standard is once-daily administration, which provides the most convenient regimen for the patient.

Toxicology

Tests for the toxic effects of a drug are generally performed on live animals, most commonly rats and mice. These tests are considered necessary because of the current limitations with using either in vitro testing or computer modeling to simulate the effects of a drug in a human.

A key feature of these toxicology studies is that increasingly high, supraphysiologic doses of drug are administered. The therapeutic index of a drug is determined using quantal dose-response curves (see pharmacodynamics chapter for further description of quantal dose-response curves). The therapeutic index is the ratio of the toxic (TD) or lethal dose (LD) of a drug to its effective dose (ED) (Figure 9-4).

One of the limitations of these data is that the doses used are so high that they often have minimal clinical relevance, except in overdose situations. It is not uncommon for widely used drugs to have significant toxicities that emerged in animal studies but have not been evident in humans.

Figure 9-4 Quantal dose-response curve.

Chemical and Pharmaceutical Development

Once the manufacturer has fully characterized the pharmacokinetics, pharmacodynamics, and toxicology of the new chemical, the manufacturer must also consider practical issues, such as how the drug will be produced and what form it will take. The chapter on pharmacokinetics provides a complete description of various dosage forms. Typically, the oral route is the preferred route.

At this stage, the manufacturer will also get an idea of the cost of producing the new drug. Small molecules, which are simple chemical structures, are typically very inexpensive to produce, often pennies per tablet. At the other end of the spectrum are biologics, which require a much more sophisticated and expensive production procedure. Although not proportional, this increased cost of production is one of the reasons cited for the high price of biologic agents.

Once all preclinical tests have been performed, the company may apply to the appropriate regulatory agency to undergo clinical testing. In the United States, the agent at this stage is referred to as an investigational new drug (IND). The regulatory agency that is responsible for approving new drugs in the United States is the Food and Drug Administration, whereas the corresponding agency in Europe is the European Medicines Agency (EMA).

Clinical Process

Stages in the Drug Approval Process

Once a new drug application has been filed, manufacturers may begin the process of clinical trials, with the overall goal of proving that the drug is both efficacious and safe. The process is traditionally carried out in phases, with increasingly large numbers of patients in each phase. The manufacturer is responsible for the conduct of these trials, although regulatory agencies may conduct site inspections to confirm that studies are being conducted properly. Table 9-2 lists the key characteristics of each phase of the clinical trial process.

TABLE 9-2 Key Characteristics of Phase 1 to Phase 4 Clinical Trials

Phase 1

Purpose: establish the pharmacokinetic profile of the drug. This includes the area under the curve (AUC) and the half-life (t_1/2), as well as the metabolic routes of the drug.

Until recently, this stage was primarily carried out in healthy (usually) male volunteers, because of (among other considerations) concerns over running experiments on sick individuals who might benefit from proven therapies. Males were preferred owing to liability issues from teratogens. Lately, especially in the case of serious diseases such as cancer, phase I trials have included patients as well, providing these individuals with an earlier (faster) exposure to promising new agents.

Phase 2

In phase 2 we begin to see the drug compared with results in a control group, usually a placebo control. These trials include patients and are of moderate size, usually 100 subjects per group or less. By this time, a range of doses should be established, such that the main purpose is to determine efficacy and safety.

Phase 3

Phase 3 represents the final stage before drug approval and establishes the efficacy of the doses that will be used in the clinic. By this time, obvious safety issues should have been seen, so phase 3 trials are open to a much larger group of patients. However, owing to the size of these trials, less-obvious safety issues may arise, sometimes either preventing the approval of the drug or warranting a strong warning on the labeling.

Once a drug has successfully completed phase 3, the manufacturer can apply for a Notice of Compliance (NOC) from the FDA or equivalent regulatory agency. The FDA will review the efficacy and safety data then make a decision as to whether the drug should be approved for widespread use. Typically around 30 drugs receive an NOC each year in the United States.

Phase IV

Phase 4 is also referred to as postmarketing surveillance. Regulatory agencies may require double-blind controlled trials in this phase, depending on how many unresolved issues remain from phase 3. Phase 4 has not been a regulatory requirement, although recent events with drugs such as rofecoxib have illustrated the importance of postmarketing surveillance.

Limitations of the Drug Approval Process

There are some important limitations to the clinical trial process, which highlight the need for postmarketing surveillance.

Clinical trials are typically not large enough to reveal rare but serious adverse events that may become obvious once the drug is available to the general population.

Clinical trials are typically of too short a duration to reveal serious adverse effects that may take years of continual use to develop.

Note that the commonality between these two limitations is time. Large trials take longer to produce results (owing to logistics), and of course longer trials take more time. Time is an important issue for indications such as cancer and human immunodeficiency virus (HIV) infection, and the fast-tracking of drugs has become routine in these diseases.

The International Conference on Harmonisation of Technical Requirements for Registration of Pharmaceuticals for Human Use (ICH) is an agreement among Europe, the United States, and Japan to improve consistency in the way products are registered.

Studies that follow ICH guidelines provide assurance to countries that are using data from these studies that they follow a set of standards.

The purpose of ICH is to reduce duplication of studies among countries, thus speeding up the approval process and reducing costs.

ICH does not, however, guarantee that all of these studies have been performed in a manner that would minimize the potential for bias. Therefore it is important to understand the key features of a good clinical trial, and this will be reviewed in the following section.

Clinical Trial Design

The gold standard in clinical trial design is the double-blind randomized controlled trial (DBRCT). However, simply having a DBRCT design does not guarantee that the study is of sufficient quality to provide reliable results. The key issues are bias and avoiding any confounding factors that might influence results either in favor of or against the intervention under review. In a DBRCT the goal is to minimize bias and confounding and thus be as confident as possible that the results are solely a function of the interventions under review.

Randomization is a process by which study participants are randomly assigned to one of the interventions in the trial.

One of the goals of randomization is to stratify groups in the trial so that the composition of each intervention group is identical to that of the other. Demographic and relevant baseline characteristics of the participant’s medical condition should be balanced to ensure that comparison groups all begin from the same starting point.

It is also essential that randomization be carried out in a way that ensures that the identity of treatment allocation is not revealed to the investigators (physicians, nurses), the patients, or anyone directly involved in the trial.

Ideally, to maintain allocation concealment, randomization should be carried out by a third party; an automated interactive voice response system (IVRS) is typically used for this purpose. With an IVRS, the patient interacts with a computer (typically over the phone), and the computer randomizes the patient to a treatment group, assigning the patient a number.

Blinding ensures that patients and providers are not influenced by knowledge of their assigned intervention.

This is particularly important for subjective outcomes, such as quality-of-life scales, that rely on a patient’s own assessment of well-being. Patients who know they have been assigned to the placebo group, for example, may be less likely to believe they are benefiting from therapy.

Blinding should again be carried out in a way that ensures allocation concealment. The most important step in ensuring allocation concealment is to make sure all interventions are identical in appearance. Therefore the authors of a study should explicitly state that interventions were identical in appearance.

The next considerations are statistical. One of the most common problems with clinical trials is that they are too small to properly answer the questions under review. A study should identify a primary outcome or outcomes; these are endpoints that are considered to be of most importance to the designers of the trial. The plan for statistical analysis, including sample size, is typically based on this primary outcome. Calculation of sample size is also known as statistical power. The larger a study is, the more power it has to reveal statistical differences between interventions, if they exist. In small or underpowered studies, it is difficult to know whether a finding of no difference was the result of small sample size or the fact that there were no differences in efficacy between interventions.

Another consideration when calculating sample size is the type of statistical comparison being performed. There are three types of comparisons: superiority, noninferiority, and equivalence.

The simplest design is a superiority design, in which one intervention is determined simply to be statistically superior to another. Superiority designs typically involve a comparison with placebo.

In a noninferiority design, the goal is to confirm that a given intervention is not inferior to a comparator. Noninferiority trials are typically conducted between two active comparators.

A margin for noninferiority is defined before the start of the trial.

• For example, in a hypertension trial the margin for noninferiority for an intervention is a reduction in diastolic blood pressure of 10 mm Hg.

• This means that the intervention can reduce diastolic pressure less than its comparator, as long as the difference between the two groups does not exceed 10 mm Hg.

• If the difference does exceed 10 mm Hg, the intervention is deemed to be inferior to its comparator; otherwise it is considered to be noninferior.

Equivalence trials are rare, and the purpose of an equivalence trial is to determine whether one intervention is statistically no different in efficacy than another.

• The challenge with such a design is that not only is it necessary to prove that the intervention is not inferior to its comparator, but the study also must prove simultaneously that the intervention is not superior.

Analyzing Data from Clinical Trials

Efficacy Data

In general, data from clinical trials are expressed in two ways: as either dichotomous or continuous data. Continuous data (e.g., change in blood pressure) are often expressed as a mean difference. Dichotomous data (all or none) are often expressed as a relative risk (RR) or an absolute risk reduction (ARR).

Relative risk is a ratio of probabilities. For example, the probability of getting heart disease is 10% in smokers, and 5% in smokers who take statin therapy; therefore in smokers the relative risk of heart disease with statin therapy is 0.5 (0.05/0.10). In this case, it can be said that statins reduce the relative risk of developing heart disease by 50% (1 − 0.5).

The absolute risk of an event occurring is simply the proportion of patients in whom an event occurs. Continuing with the smoking example, 10% of smokers develop heart disease. The ARR would be the amount by which risk is reduced by an intervention (statins). The risk of heart disease in patients prescribed statin therapy is 5%; therefore statins reduce the absolute risk of heart disease by 5%.

Subsequently, the ARR is expressed as a reciprocal, the number needed to treat (NNT). The NNT is the number of patients who must be treated with a given intervention to prevent an event in one patient.

• In the example, ARR in heart disease with statin therapy is 5%, and the NNT is 20 (1/0.05). Quite literally, the NNT in this case indicates that 20 patients must be treated with statins to prevent one patient from developing heart disease.

• The NNT is a popular tool used by clinicians to quickly provide perspective on the efficacy of a drug.

Note that there can be a huge difference in absolute and relative risk, yet the two terms sound similar, and they are often used interchangeably even though they mean two different things. Statin therapy reduced the absolute risk of heart disease by 5% and the relative risk by 50%. The latter sounds more impressive than the former—and that is why improvements in relative risk are often reported in the literature and (especially) by the media, rather than absolute risk.

Another commonly used calculation is the odds ratio (OR). The OR compares the odds of two events occurring.

In the example, out of every 100 smokers who do not take statins, 10 will develop heart disease and 90 will not. Therefore the odds of developing heart disease in these patients are 10/90, or 0.11.

Of every 100 smokers on statin therapy, 5 will develop heart disease and 95 will not. Therefore the odds of developing heart disease in these patients are 5/95, or 0.053.

The OR is the ratio of these two odds: 0.053/0.11 = 0.47.

Note that the OR and RR often yield similar numbers, and OR will also be reported instead of absolute risk. These two ratios are a preferred means for reporting data in the literature, despite the fact that reductions in absolute risk are far more intuitive and easy to grasp for patients and providers.

Safety Data

The version of the NNT used to compare the rate of adverse effects in a study is the number needed to harm (NNH). The NNH is the number of patients who must be treated with an intervention before one patient is harmed.

In a clinical trial, an adverse event is any harmful event that occurs to a patient during a trial. An adverse event does not have to be related to the drug and therefore is not the same as a side effect. An example of an adverse event is a patient being hit by lightning during the course of the study. Clinical trials also report serious adverse events, typically defined as any events that result in death, disability, or prolonged hospitalization. Adverse events and serious adverse events are sometimes collectively referred to as harms.

Pooling Data from Studies: the Meta-Analysis

Unless single studies have very large sample sizes, data from these studies are not typically considered to be as reliable as data from multiple studies when one is trying to assess the efficacy or safety of a drug. A systematic review is a way of gathering data from separate trials of the same drugs in a scientific manner that promotes reliability and reproducibility.

Authors of a systematic review will design a protocol that defines the populations, interventions, comparisons, and outcomes that are of interest in the review, also known by the acronym PICO. The protocol is analogous to a protocol in scientific experiments, in that if another investigator were to apply the same protocol to the literature at the same time, he or she should find the same results.

A meta-analysis is a pooled analysis of the papers included in a systematic review. By pooling data from several studies, a meta-analysis is a way of overcoming the limitations associated with single, small studies.

The data from a meta-analysis are often presented as a forest plot (Figure 9-5). The plot provides a quick graphical representation of the data, as well as numeric summaries to the left of the plot.

The plot gives a quick summary of the direction of the data. In this case, four of five studies favor the experimental drug over placebo, the only exception being the OUTLIER study.

The plot also provides an indication of the relative size of studies. The PANIC study has the largest point estimate (blue square) and is the largest study. The OUTLIER study is the smallest study.

The horizontal line running through the point estimate gives an indication of the confidence interval, or the degree of variation within the study. Note that larger studies have less variation.

Studies whose point estimate or confidence interval crosses unity (1 on the graph) show no statistical difference between comparison groups. In this case, all comparisons are statistically significant except OUTLIER.

The plot also provides a summary point estimate, the final black diamond at the bottom. Again, if the black diamond does not cross unity, the results of the meta-analysis are statistically significant, in this case favoring the study drug.

Figure 9-5 Forest plot.

The final key piece of information that a forest plot provides is an indication of the heterogeneity between studies. In Figure 9-5, the results of four studies are fairly consistent, with the exception of the OUTLIER study. An important consideration when assessing the reliability of these data is heterogeneity of the included studies.

The meta-analysis in Figure 9-5 also provides a measure of heterogeneity, expressed as either a P value or an I² value. When the test for heterogeneity yields P < .05, significant heterogeneity is considered to exist within the meta-analysis. The higher the I² value, the greater the heterogeneity. In this case, another analysis should be performed to account for this heterogeneity.