Randomized dose-finding studies: how many dose levels should be used?

Abstract

These days, people live longer. A fact generally accredited to higher standards of living, improved self-awareness of a healthy life-style and also, importantly, better healthcare. For example, life expectancy in the USA has increased by more than 30 years in the last century, 25 years of this gain being attributible to public health benefits [1]. There is an enormous amount of medicine available for patients, ranging from cough medicine to complex chemotherapy drugs for cancer patients. The number of products, ‘distinctive medicines, described by brand or generic name, or both’ [2], listed in the physicians’ compendia in 1981 were 2100 in the UK and 6000 in the USA [3]. Currently, the pharmaceutical industry in the EU, Japan and the USA follow the International Conference on Harmonization of Technical Requirements for Registration of Pharmaceuticals for Human Use guidelines [101] to conduct tests on medicines in preclinical trials and clinical trials before they can be granted a license by regulatory authorities. In preclinical trials, a new compound is tested on animals (in vivo) or cells (in vitro). Its toxic responses are monitored along with pharmacological aspects: pharmaco kinetics – drug concentrations in the body (cell), and pharmacodynamics – drug reactions in the body (cell), targeting certain molecules or receptors and the potentially efficacious response of the drug. Once the compound has been proven safe and has potential efficacy, it will be administered into humans to test the safety and efficacy further. Phase I clinical trials are used to establish the safety of the drug when it is first administered into humans. An identified safe dose from preclinical studies is scaled for human subjects, either by body weight or body surface areas [4]. This will form the starting dose (×) for a Phase I trial. A fixed dose range will be used. Usually it is based on a doubling scale (×, 2×, ...) or a modified Fibonacci sequence (×, 2×, 3.3×, 4.95×, ...) [5]. Each dose of this fixed range is to be tested for its safety. In cancer trials a typical study design is a parallel group, randomized and conducted on a small number of subjects, for example, 20–80. Subjects are randomly allocated into two groups: one is the treatment group (to receive the new active compound) and the other one is the control group (either to receive a placebo or an existing compound). Dose-escalation procedures either follow a predefined rule or are based on clinicians’ judgment. An example of a predefined rule is the ‘3 + 3’ up-and-down rule [6]: each treatment group contains three subjects. If no toxicity is observed within three patients, then the next group of subjects will get the next higher dose. If one out of three patients had toxicity, then the same dose will be repeated for the next group. If more than two out of three patients had toxicity, then the trial will