Confounding Factors in Clinical Trials
Many diseases tend to wax and wane in severity; some disappear spontaneously, even, on occasion, cancer. A good experimental design takes into account the natural history of the disease by evaluating a large enough population of subjects over a sufficient period of time. Further protection against errors of interpretation caused by disease fluctuations is sometimes provided by using a crossover design, which consists of alternating periods of admin-istration of test drug, placebo preparation (the control), and the standard treatment (positive control), if any, in each subject. These sequences are systematically varied, so that different subsets of patients receive each of the possible sequences of treatment.
Known and unknown diseases and risk factors (including lifestyles of subjects) may influence the results of a clinical study. For example, some diseases alter the pharmacokinetics of drugs. Other drugs and some foods alter the pharma-cokinetics of many drugs. Concentrations of blood or tissue com-ponents being monitored as a measure of the effect of the new agent may be influenced by other diseases or other drugs. Attempts to avoid this hazard usually involve the crossover technique (when feasible) and proper selection and assignment of patients to each of the study groups. This requires obtaining accurate diagnostic tests, medical and pharmacologic histories (including use of recre-ational drugs), and the use of statistically valid methods of ran-domization in assigning subjects to particular study groups. There is growing interest in analyzing genetic variations as part of the trial that may influence whether a person responds to a particular drug. It has been shown that age, gender, and pregnancy influence the pharmacokinetics of some drugs, but these factors have not been adequately studied because of legal restrictions and reluc-tance to expose these populations to unknown risks.
Most patients tend to respond in a positive way to any therapeutic intervention by interested, caring, and enthusiastic medical per-sonnel. The manifestation of this phenomenon in the subject is the placebo response (Latin, “I shall please”) and may involve objective physiologic and biochemical changes as well as changes in subjective complaints associated with the disease. The placebo response is usually quantitated by administration of an inert mate-rial with exactly the same physical appearance, odor, consistency, etc, as the active dosage form. The magnitude of the response var-ies considerably from patient to patient and may also be influ-enced by the duration of the study. In some conditions, a positive response may be noted in as many as 30–40% of subjects given placebo. Placebo adverse effects and “toxicity” also occur but usu-ally involve subjective effects: stomach upset, insomnia, sedation, and so on.Subject bias effects can be quantitated—and minimized rela-tive to the response measured during active therapy—by the single-blind design. This involves use of a placebo as describedabove, administered to the same subjects in a crossover design, if possible, or to a separate control group of well-matched subjects. Observer bias can be taken into account by disguising the identity of the medication being used—placebo or active form—from both the subjects and the personnel evaluating the subjects’ responses (double-blind design). In this design, a third party holds the code identifying each medication packet, and the code is not broken until all the clinical data have been collected.Drug effects seen in clinical trials are obviously affected by the patient taking the drugs at the dose and frequency prescribed. In a recent phase 2 study, one third of the patients who said they were taking the drug were found by blood analysis to have not taken the drug. Confirmation of compliance with protocols (also known as adherence) is a necessary element to consider.The various types of studies and the conclusions that may be drawn from them are described in the accompanying text box. (See Box: Drug Studies—The Types of Evidence.)
Drug Studies—The Types of Evidence
As described, drugs are studied in a variety of ways, from 30-minute test tube experiments with isolated enzymes and receptors to decades-long observations of popula-tions of patients. The conclusions that can be drawn from such different types of studies can be summarized as follows.Basic research is designed to answer specific, usually single,questions under tightly controlled laboratory conditions, eg, does drug x inhibit enzyme y? The basic question may then be extended, eg, if drug x inhibits enzyme y, what is the concentration-response relationship? Such experiments are usually reproducible and often lead to reliable insights into the mechanism of the drug’s action.First-in-human studies include phase 1–3 trials. Once a drugreceives FDA approval for use in humans, case reports and caseseries consist of observations by clinicians of the effects of drug(or other) treatments in one or more patients. These results often reveal unpredictable benefits and toxicities but do not generally test a prespecified hypothesis and cannot prove cause and effect. Analytic epidemiologic studies consist of observations designed to test a specified hypothesis, eg, that thiazolidine-dione antidiabetic drugs are associated with adverse cardiovas-cular events. Cohort epidemiologic studies utilize populations of patients that have (exposed group) and have not (control group) been exposed to the agents under study and ask whether the exposed groups show a higher or lower incidence of the effect. Case control epidemiologic studies utilize populations of patientsthat have displayed the end point under study and ask whether they have been exposed or not exposed to the drugs in question. Such epidemiologic studies add weight to conjectures but can-not control all confounding variables and therefore cannot con-clusively prove cause and effect.Meta-analyses utilize rigorous evaluation and grouping ofsimilar studies to increase the number of subjects studied and hence the statistical power of results obtained in multiple pub-lished studies. While the numbers may be dramatically increased by meta-analysis, the individual studies still suffer from their varying methods and end points and a meta-analysis cannot prove cause and effect. Large randomized controlled trials are designed to answer specific questions about the effects of medi-cations on clinical end points or important surrogate end points, using large enough samples of patients and allocating them to control and experimental treatments using rigorous randomiza-tion methods. Randomization is the best method for distributing all foreseen confounding factors, as well as unknown confound-ers, equally between the experimental and control groups. When properly carried out, such studies are rarely invalidated and can be very convincing.