Principles of dose finding studies in cancer: a comparison of trial designs.  
Jump to Full Text  
MedLine Citation:

PMID: 23299793 Owner: NLM Status: MEDLINE 
Abstract/OtherAbstract:

PURPOSE: One key aim of Phase I cancer studies is to identify the dose of a treatment to be further evaluated in Phase II. We describe, in nonstatistical language, three classes of doseescalation trial design and compare their properties. METHODS: We review three classes of doseescalation design suitable for Phase I cancer trials: algorithmic approaches (including the popular 3 + 3 design), Bayesian modelbased designs and Bayesian curvefree methods. We describe an example from each class and summarize the advantages and disadvantages of the design classes. RESULTS: The main benefit of algorithmic approaches is the simplicity with which they may be communicated: it may be for this reason alone that they are still employed in the vast majority of Phase I trials. Modelbased and curvefree Bayesian approaches are preferable to algorithmic methods due to their superior ability to identify the dose with the desired toxicity rate and their allocation of a greater proportion of patients to doses at, or close to, that dose. CONCLUSIONS: For statistical and practical reasons, algorithmic methods cannot be recommended. The choice between a Bayesian modelbased or curvefree approach depends on the previous information available about the compound under investigation. If this provides assurance about a particular model form, the modelbased approach would be appropriate; if not, the curvefree method would be preferable. 
Authors:

Thomas Jaki; Sally Clive; Christopher J Weir 
Related Documents
:

22154543  Linear mapping of numbers onto space requires attention. 23263563  On folivory, competition, and intelligence: generalisms, overgeneralizations, and model... 8855483  The use of the 'binormal' model for parametric roc analysis of quantitative diagnostic ... 
Publication Detail:

Type: Comparative Study; Journal Article; Research Support, NonU.S. Gov't; Review Date: 20130109 
Journal Detail:

Title: Cancer chemotherapy and pharmacology Volume: 71 ISSN: 14320843 ISO Abbreviation: Cancer Chemother. Pharmacol. Publication Date: 2013 May 
Date Detail:

Created Date: 20130426 Completed Date: 20130620 Revised Date: 20140220 
Medline Journal Info:

Nlm Unique ID: 7806519 Medline TA: Cancer Chemother Pharmacol Country: Germany 
Other Details:

Languages: eng Pagination: 110714 Citation Subset: IM 
Export Citation:

APA/MLA Format Download EndNote Download BibTex 
MeSH Terms  
Descriptor/Qualifier:

Algorithms Antineoplastic Agents / administration & dosage* Bayes Theorem Clinical Trials, Phase I as Topic / methods* DoseResponse Relationship, Drug Humans Neoplasms / drug therapy Research Design* 
Grant Support  
ID/Acronym/Agency:

G0800792//Medical Research Council; //Medical Research Council 
Chemical  
Reg. No./Substance:

0/Antineoplastic Agents 
Comments/Corrections 
Full Text  
Journal Information Journal ID (nlmta): Cancer Chemother Pharmacol Journal ID (isoabbrev): Cancer Chemother. Pharmacol ISSN: 03445704 ISSN: 14320843 Publisher: SpringerVerlag, Berlin/Heidelberg 
Article Information Download PDF © The Author(s) 2013 OpenAccess: Received Day: 25 Month: 10 Year: 2012 Accepted Day: 12 Month: 12 Year: 2012 Electronic publication date: Day: 9 Month: 1 Year: 2013 pmcrelease publication date: Day: 9 Month: 1 Year: 2013 Print publication date: Month: 5 Year: 2013 Volume: 71 Issue: 5 First Page: 1107 Last Page: 1114 PubMed Id: 23299793 ID: 3636432 Publisher Id: 2059 DOI: 10.1007/s0028001220598 
Principles of dose finding studies in cancer: a comparison of trial designs  
Thomas JakiAff1  
Sally CliveAff2  
Christopher J. WeirAff3 
Address: +441316503230 +441316503224 Christopher.Weir@ed.ac.uk 
Medical and Pharmaceutical Statistics Research Unit, Lancaster University, Lancaster, UK 

Edinburgh Cancer Centre, Western General Hospital, Edinburgh, UK 

Medical Research Council Hub for Trials Methodology Research, Centre for Population Health Sciences, University of Edinburgh Medical School, Teviot Place, Edinburgh, EH8 9AG UK 
Ethical considerations [^{1}] require the use of efficient trial designs in order to optimize the balance of risk versus benefit for participants. In Phase I cancer studies, this would include minimizing the numbers of patients allocated to ineffective or excessively toxic doses, while addressing the principal aim of identifying the best dose of a treatment to recommend for further evaluation in a Phase II trial. In this article, we focus on the identification of the maximum tolerated dose (MTD), which remains a key factor in this decision. In practice, additional aspects such as the biological level of anticancer activity of a dose would also be considered. The 3 + 3 trial design [^{2}, ^{3}], which has been widely implemented, has substantial limitations: we describe two alternative classes of doseescalation strategy, using specific examples [^{4}–^{6}] drawn from the many designs available to illustrate their superiority to the 3 + 3 design on several relevant study design criteria.
Phase I studies play an extremely important role in the development of a cancer treatment as the agent is often given to humans for the first time. As a result, conservative approaches that tend to start at doses much lower than the anticipated highest safe dose and slowly approach the dose of interest from below are usually employed. The population studied is not necessarily the population to be treated and the questions are many despite the subjects being few. Consequently much uncertainty about dose, safety and efficacy will remain afterward. These studies are, however, critical to successful drug development as the decision on whether to continue and the design of any subsequent trials depend on the outcome of this first study.
Phase I doseescalation studies in cancer traditionally enroll late stage patients for whom other therapies have failed. Due to the narrow therapeutic index of cytotoxic drugs and the variable toxicities that may occur at the therapeutic dose of newer targeted anticancer drugs, the MTD has been considered a reasonable basis on which to determine an appropriate dose for further clinical use. Hence, in addition to investigating the pharmacokinetics, toxicities and biological activity of a drug, Phase I cancer studies aim to find the MTD, the highest dose that can safely be administered. More precisely, one seeks the dose that has an acceptable risk of doselimiting toxicity (DLT). In this context, a DLT is a serious adverse event that impairs usual activities and requires therapeutic intervention [^{7}–^{9}]. Late toxicities or lower graded, cumulative or additive toxicities that do not individually meet the DLT criteria also contain information on the safety and activity of a drug; however, the designs used to date do not take account of such events [^{10}]. Advances in this area would clearly have potential to enhance decision making at the end of Phase I.
Commonly the dose at which the probability of a DLT is π (for 0 < π < 1) is called the TD100π. For example, the TD25 would refer to the dose associated with a 25 % toxicity risk. The rationale for seeking the TD100π is to ensure a safe dose is identified for further study and is based on the common assumption, specifically for cytotoxic compounds, that efficacy increases with toxicity. The latter implies that finding the highest tolerable dose will ensure that the most efficacious dose is investigated subsequently. First, cycle toxicity is a critical and immediate measure that can guide the decision on whether the dose may be escalated safely, and the use of this rather than later efficacy outcomes ensures minimal trial duration.
A key ingredient of any doseescalation study is the set of doses to be explored. Figure 1 illustrates four different dose schedules for a given, hypothetical, dose–toxicity relationship. Each graph depicts dose versus toxicity risk and assumes an increasing relationship between them. The arrows indicate the doses available and it is assumed that a toxicity risk of 25 % is acceptable. Figure 1a has very low risk of toxicity for the lower two doses while the upper two doses have a toxicity risk close to 100 %. Such a dose schedule therefore only allows the conclusion that the TD25 lies somewhere between the middle two doses; neither of the doses available is, however, close to the desired toxicity risk. In Fig. 1b, all the doses lie above the desired toxicity level. In this situation, not only does none of the doses have an acceptable toxicity risk, but we cannot even be sure that any dose of the compound is safe. Similarly, if all doses were below the desired TD25, we would not know if any dose corresponded to the TD25 and hence whether a biologically active dose had been attained.
Figure 1c shows the advantage of allowing many different doses. It is clear that such a dose schedule allows much more accurate TD25 estimation as the interval between doses is much smaller. While adding doses to allow for more accurate estimation is appealing, doing so requires a different doseescalation strategy to the commonly employed methods in order to limit the sample size and avoid many patients receiving doses well below those expected to have either toxicity or efficacy. Figure 1d represents an ideal situation where many doses are available, but excessively toxic or ineffective doses represented by lighter shaded arrows are explored less often or possibly even skipped altogether.
Traditionally, dose escalation starts at a low dose defined preclinically and slowly escalates dose in decreasing increments according to a predetermined strategy (the modified Fibonacci sequence, in which the dose increments for succeeding dose levels are 100, 67, 50, 40 % then 33 % for all subsequent levels [^{11}]). Alternative escalation strategies [^{12}] include accelerated titration designs [^{3}], pharmacokinetically guided dose escalation [^{13}] and continual reassessment or Bayesian methods [^{4}, ^{14}–^{16}] which have been successfully incorporated into Phase I trial design [^{17}, ^{18}].
From here on, we assume that a sensible selection of doses is available but stress that inclusion of more potential doses within the range of interest may allow more accurate estimation of the highest tolerable dose. To avoid “wasting” patients at doses that are no longer of interest based on the data gathered so far in the trial, skipping doses [^{6}] or the use of single patients at early dose levels [^{3}] could be considered.
There exists a vast literature on doseescalation methods for Phase I trials in cancer, yet very few of them have been used in practice. By far the most popular approach to dose escalation is the 3 + 3 design [^{2}] which, according to Le Tourneau et al. [^{12}], was used in 96.7 % of Phase I trials published in 2007 and 2008. A similar finding was made by Rogatko et al. [^{19}] when they investigated the use of Bayesian designs in Phase I cancer trials. Only 20 (1.6 %) of 1,235 trials followed a Bayesian design despite the existence of around 100 publications demonstrating the statistical properties of such designs.
There are three general approaches to dose finding studies in cancer. We will introduce their main features and compare their advantages and disadvantages. In particular, we will show that the most popular design in practice has severe shortcomings when it comes to identifying the dose of interest.
Rulebased doseescalation methods include the Storer upanddown design [^{3}] and the frequently applied 3 + 3 design (Fig. 2). Although several variants of this longestablished approach have been published, the earliest source is a book chapter arising from the proceedings of a clinical pharmacology course held in Brussels in May 1972 [^{2}]. The main principle of algorithmic designs is that a small group of patients is treated at a given dose and, dependent on the observed number of toxicities, a decision is made on to whether to study a further group of patients at the next dose up the scale, to study more patients at the same dose or to stop the trial.
The 3 + 3 design, for example, enters three patients into the trial one at a time at intervals of seven days or more and treats them at the chosen starting dose. The core features of this design, as outlined by Storer [^{3}], are:
 If no doselimiting toxicity (DLT) occurs in the first cycle of any of the initial group of 3 patients treated at a dose, the dose should be escalated for the next group of patients.
 If two or more of the 3 patients treated at a given dose experience a DLT, the trial should stop.
 If one patient of the 3 treated at a given dose level experiences a DLT, a further 3 patients are treated at the same dose level. If a DLT has occurred in exactly one of these 6 patients, escalation may continue as in (a); otherwise, the trial should stop.
A frequent variation allows the dose to be reduced if more than one DLT is observed within the first three patients studied at a particular dose.
Following completion of the trial according to this design, the MTD is defined as either the actual dose at which the trial was stopped or the next lower dose, possibly depending on the frequency and severity of toxicity observed in the patients evaluated in the final group [^{3}]. A number of adaptations of this design attempt to reduce the number of patients treated at the lowest dose levels [^{20}]. Such accelerated titration designs [^{18}, ^{21}] include singlepatient dose levels and more rapid dose escalation (e.g., dose doubling) until the first DLT is observed, after which the doseescalation scheme would then revert to the modified Fibonacci sequence.
Bayesian modeling combines prior knowledge of the drug with the observed data from the current trial to provide updated information about the distribution of the trial outcome of interest. Bayesian methods have been incorporated into a number of early phase clinical trial designs, the best known being the continual reassessment method (CRM) [^{4}, ^{5}]. In contrast to the algorithmic approach of the 3 + 3 design, the continual reassessment method aims to identify the dose at which the proportion of patients experiencing a DLT reaches a specific target level (e.g., 25 %). Furthermore, the investigator performs repeated analyses on all of the data gathered to date—hence the term continual reassessment—rather than simply observing the data recorded at the current dose, as is the case for the 3 + 3 design.
The CRM uses a oneparameter Bayesian model which assumes that the probability of toxic response increases with dose: as we move from one dose up to the next higher dose, the toxicity risk at the higher dose is greater than or equal to that at the previous dose. Before the trial has commenced, Bayesian modeling requires a prior distribution for the dose–toxicity curve parameters to be specified [^{22}], representing the expected shape of the dose–toxicity relationship. In the original version of the CRM, a simple oneparameter prior distribution is recommended; subsequent developments allow the prior to be informed by data (such as information from preclinical studies) or expert opinion (see for example [^{23}] for alternatives). The model may be updated as soon as new data become available on previously included patients. Following this the next patient, or group of patients, is treated at the dose, based on the evidence to date, that has an estimated probability of toxicity closest to the target level.
Once all patients have been treated and followed up, the MTD is taken to be the dose at which the estimated toxicity probability is closest to the target level. Although the originally proposed CRM method allows for the initial dose to be above the lowest available dose level, in practice, a modified version that enforces dose escalation to start from the lowest dose [^{24}] is usually employed. Stopping rules have been developed for the CRM [^{25}] to allow early discontinuation of the trial in the situation where all doses are found to be excessively toxic.
Alternative Bayesian modelbased approaches include escalation with overdose control (EWOC) designs [^{16}] and the TITECRM extension of the CRM which allows the trial to be completed more quickly by incorporating time to toxicity event data [^{26}].
This class of escalation strategies includes work by Gasparini and Eisele [^{27}] and Whitehead et al. [^{6}] and aims to minimize the number of assumptions within the dose–toxicity modeling. No assumption is made about the form of the relationship between dose and toxicity except that, as in the CRM, the probability of toxic response increases with dose. The risk of toxicity is modeled directly, resulting in an easy to interpret table of probabilities for each risk level.
The possible levels of toxicity risk at a dose are described qualitatively as very safe, safe, ideal, risky or toxic. The numerical probability value that corresponds to each of these descriptors depends on the target toxicity level. The “ideal” risk category has exactly the target toxicity probability, while the “safe” and “very safe” categories have progressively lower risks of toxicity and the “risky” and “toxic” categories have progressively higher risks. For example, in a study aiming to identify the dose which has toxicity rate 25 %, the descriptors very safe, safe, ideal, risky and toxic might be assigned the probability values 0.05, 0.15, 0.25, 0.4 and 0.65, respectively.
The prior distribution summarizing previous knowledge of the expected risk at each dose level, required under the Bayesian framework, is informed by investigator opinion. As data accumulate during the trial, the updated probabilities of each risk level being associated with each dose are calculated. The dose to be allocated to the next patient enrolled in the trial is selected to avoid doses that have anything other than a small chance of being “toxic” and to target the dose that has the greatest chance of being “ideal”. Table 1 illustrates the probabilities of each level of toxicity during a hypothetical trial. Here, 25 % is the target toxicity rate, and dose level 7 is the one with the highest probability of having that toxicity rate.
At the end of the trial, one of three approaches may be used to determine which dose to take forward for further study. The first method bases the decision on the final table of updated risk level probabilities. If the dose with a toxicity risk of 25 % was being sought, then the dose with the highest probability of having a toxicity risk of 25 % would be recommended. In the second, perhaps more realistic, strategy, the table of updated risk level probabilities would form just part of the information being considered by investigators when deciding what dose to recommend. The complete study data set will contain far more information than DLT occurrences: pharmacokinetic and clinical data and the expert opinion of investigators could also inform the recommendation. The third approach would be to take the data set and apply any appropriate method of statistical analysis, independently of the dose allocation method used in the trial.
The merits and shortcomings of the three classes of doseescalation procedures introduced above will now be discussed. A broad range of important criteria, including statistical and practical aspects, to consider when evaluating a doseescalation procedure will be considered under five headings. The weighting for each of the criteria will depend on the specific trial and consequently it is unlikely that each point will carry equal weight across all trials. We have summarized this in Table 2 where we have graded the performance of the doseescalation procedure for each item under the five headings qualitatively as poor, intermediate, good or not applicable.
We expect a good statistical procedure to estimate a specific measure of toxicity and that the precision of this estimate can be gauged. Moreover, we want additional information (i.e., additional patients) to increase the precision in the estimate. Both the modelbased and curvefree approach aim to estimate a clear cut measure of toxicity: the dose at which a certain proportion of patients is expected to experience a toxic event. Both methods also quantify the precision of the estimated dose and this precision increases as the sample size increases: they have good statistical properties. Furthermore, were complete information to be available on toxicity of every dose for each patient, the CRM modelbased design has performance very close to that of the best design that could exist theoretically [^{28}].
The algorithmic approach, in contrast, tries to find the MTD which is difficult to quantify. As a consequence of this loose definition, it is also impossible to obtain a measure of precision for the MTD. Furthermore, the algorithmic nature of the method means that additional data do not feature in the estimated MTD and the decision about whether a dose is the MTD only depends on what has been observed at this dose level: no learning from the doses above or below is possible. Consequently, algorithmic approaches have very poor statistical properties: the 3 + 3 design correctly identifies the dose with toxicity rate closest to the target level less frequently than modelbased designs [^{29}] and over the course of the trial exposes many additional patients to doses above or below that optimal dose. Only 35 % of patients are treated at the optimal dose with the 3 + 3 design compared to 55 % for Bayesian adaptive designs [^{19}].
We gain insight into this poor performance of the algorithmic design by quantifying the evidence when one out of a group of three patients has experienced a DLT: the 95 % confidence interval (CI) around the best estimate of 33 % for the toxicity rate ranges from 0.8 to 90.6 %, illustrating that little has been learned about the true rate. There is not substantially more evidence even when two of six patients studied at a dose experience a DLT: the corresponding 95 % CI for the toxicity rate is (4.3, 77.7 %).
The ideal method would be easy to describe to clinicians and be straightforward statistically. The latter criterion does not apply to the algorithmic approach as it is not a statistically derived method. It is, however, the simplest to explain to nonstatisticians and may be used without the involvement of a statistician. The Bayesian modelbased approach is of moderate statistical complexity, but is easily implemented in practice, in part due to software being readily available (e.g., the R package CRM [^{30}]). A particular challenge that does remain is specifying the prior distributions, which we believe is best done in consultation with a statistician. Finally, the curvefree Bayesian method is relatively simple to explain to nonstatisticians (though much more complex than the 3 + 3 design) as it directly estimates the risk of each dose rather than using a specific model. The price for this conceptual simplicity is a much more involved statistical process.
The contrast between the modelbased and the curvefree approaches is that the former requires the dose–toxicity model to be specified in advance of the study through discussion between the clinicians and a statistician. Nevertheless, it can still consistently identify the dose that has the desired toxicity rate, even when the model has been misspecified [^{4}, ^{31}]. The curvefree approach only assumes nondecreasing risk of toxicity as dose increases. Although this assumption is often reasonable, the curvefree approach depends on it while a modelbased approach could allow for decreases in toxicity risk provided that this was specified in the model.
The optimal doseescalation method would recommend doses for subsequent patients or for Phase II trials with which an experienced investigator would be comfortable. In general, all three approaches perform well although it does depend on the exact implementation: the originally proposed CRM [^{4}] does sometimes give a counterintuitive dose recommendation [^{23}] but subsequent modifications have been developed [^{24}] to overcome this.
In particular, we are interested whether any target toxicity rate can be used, how robust the method is to variations in the true dose–toxicity relationship and whether the method can skip doses. The algorithmic approach fairs poorly on all three counts. In its design, the MTD rather than a certain toxicity rate is sought, while dose skipping is prohibited. The dose recommendation only depends on the current dose and hence the underlying dose–toxicity relationship is not exploited. The modelbased approach on the other hand caters for any toxicity level and, depending on the exact implementation, can allow dose skipping. In addition, any model can be used although it is clear that specifying the correct model before starting the study will sometimes be difficult. As an alternative to modeling the dose–toxicity relationship, the modelbased approach may also be used in a continual reassessment of pharmacokinetic data [^{13}, ^{18}], pharmacodynamic data [^{32}, ^{33}] or toxicity and efficacy data in combination [^{34}]. The Bayesian curvefree approach may also be applied to combined toxicity and efficacy data [^{35}] (and indeed pharmacodynamic data) if these are binary. As well as allowing any target toxicity rate and dose skipping, the biggest advantage of the curvefree approach is that it is suitable for any underlying dose–toxicity relationship provided toxicity increases with dose. This implies that, unlike the modelbased approach, the form of the relationship does not have to be specified in advance. Both the modelbased and Bayesian curvefree approaches permit singlepatient dose cohorts, which can dramatically improve the trial design efficiency.
As illustrated in Fig. 1c, a large number of doses allow more precise estimation of the dose of interest although it potentially comes at a price of many patients being treated at suboptimal doses. With the algorithmic approach, the decision about a dose only depends on what has been observed at this dose level: it does not make use of the information from any of the previous dose levels. In consequence, a large number of doses increase the chance of selecting too safe a dose as even a very safe dose may result in two out of six patients having a DLT [as noted above, the toxicity rate confidence interval in that scenario would be (4.3, 77.7)]. The situation is exacerbated by disallowing skipping of doses and thereby enforcing assessment of every dose incrementally. In contrast, both the modelbased and curvefree methods allow dose skipping and use data from all patients in estimating risk of toxicity at each dose level, and so perform well when many doses are available.
Overall the main benefit of the algorithmic approach is the simplicity of communicating the method and incorporating it into trial protocols and it may be for this reason alone that it is still employed in the vast majority of Phase I trial designs. On the grounds of statistical and other practical considerations, however, it cannot be recommended. The modelbased and curvefree approaches have similar merits and are preferable to algorithmic methods due to their superior ability to identify the dose with the desired toxicity rate. Although it is desirable to involve a statistician in the planning of a Phase I oncology trial, Bayesian modelbased methods can readily be implemented by a numerate scientist or clinician. The decision whether a modelbased approach is to be used will largely depend on the previous information available about the compound under investigation. Substantial evidence from preclinical studies or studies in different indications regarding the shape of the dose–response curve would motivate use of the modelbased approach over the curvefree method. If there is sufficient evidence of high enough quality from previous studies, the modelbased approach will be slightly superior to the curvefree one, but if not, resulting in the wrong model, it will perform less well.
Notes
Thomas Jaki, Sally Clive, Christopher J. Weir made equal contributions to the development of this manuscript.
The views expressed in this publication are those of the authors and not necessarily those of the UK National Health Service, the UK National Institute for Health Research or the UK Department of Health.
This work was supported by a United Kingdom (UK) National Institute for Health Research Career Development Fellowship [NIHRCDF20100332 to T.J.]. C.J.W. was supported in part by the UK Medical Research Council Edinburgh Hub for Trials Methodology Research. The authors thank Professor David Cameron, Professor Duncan Jodrell and two independent reviewers for their constructive suggestions on this manuscript.
The authors have declared no conflicts of interest.
References
1..  Joffe S,Miller FG. Rethinking riskbenefit assessment for phase I clinical trialsJ Clin OncolYear: 2006242987299010.1200/JCO.2005.04.929616809725 
2..  Carter SK. Staquet MJStudy design principles for the clinical evaluation of new drugs as developed by the chemotherapy programme of the National Cancer InstituteThe design of clinical trials in cancer therapyYear: 1973Mount KiscoFutura Publishing Co242289 
3..  Storer BE. Design and analysis of phase I clinical trialsBiometricsYear: 19894592593710.2307/25316932790129 
4..  O’Quigley J,Pepe M,Fisher L. Continual reassessment method: a practical design for phase I clinical trials in cancerBiometricsYear: 199046334810.2307/25316282350571 
5..  O’Quigley J,Chevret S. Methods for dose finding studies in cancer clinical trials: a review and results of a Monte Carlo studyStat MedYear: 1991101647166410.1002/sim.47801011041792460 
6..  Whitehead J,Thygesen H,Whitehead A. A Bayesian dosefinding procedure for phase I clinical trials based only on the assumption of monotonicityStat MedYear: 2010291808182410.1002/sim.396320658549 
7..  Geller NL. Design of phase I and II clinical trials in cancer: a statistician’s viewCancer InvestYear: 1984248349110.3109/073579084090485226509359 
8..  Eastern Cooperative Oncology Group (2012) http://www.ecog.org/general/ctc.pdf. Accessed 10 Dec 2012 
9..  National Cancer Institute (2012) Common terminology criteria for adverse events v4.0. http://ctep.cancer.gov/reporting/ctc.html. Accessed 10 Dec 2012 
10..  PostelVinay S,GomezRoca C,Molife LR,et al. Phase I trials of molecularly targeted agents: should we pay more attention to late toxicities?J Clin OncolYear: 2011291728173510.1200/JCO.2010.31.923621444876 
11..  Hansen HH,Selawry OS,Muggia FM,et al. Clinical studies with 1(2chloroethyl)3cyclohexyl1nitrosourea (NSC 79037)Cancer ResYear: 1971312232274926242 
12..  Le Tourneau C,Lee JJ,Siu LL. Dose escalation methods in phase I cancer trialsJ Natl Cancer InstYear: 200910170872010.1093/jnci/djp07919436029 
13..  Collins JM,Zaharko DS,Dedrick RL,et al. Potential roles for preclinical pharmacology in phase I clinical trialsCancer Treat RepYear: 19867073803753662 
14..  Tighiouart M,Rogatko A,Babb JS. Flexible Bayesian methods for cancer phase I clinical trials: dose escalation with overdose controlStat MedYear: 2005242183219610.1002/sim.210615909291 
15..  Rinaldi DA,Kuhn JG,Burris HA,et al. A phase I evaluation of multitargeted antifolate (MTA, LY231514), administered every 21 days, utilizing the modified continual reassessment method for dose escalationCancer Chemother PharmacolYear: 19994437238010.1007/s00280005099210501910 
16..  Babb J,Rogatko A,Zacks S. Cancer phase I clinical trials: efficient dose escalation with overdose controlStat MedYear: 1998171103112010.1002/(SICI)10970258(19980530)17:10<1103::AIDSIM793>3.0.CO;299618772 
17..  Dees EC,Whitfield LR,Grove WR,et al. A Phase I and pharmacologic evaluation of the DNA intercalator CI958 in patients with advanced solid tumorsClin Cancer ResYear: 200063885389411051234 
18..  Clive S,Webb DJ,MacLellan J,et al. Forearm blood flow and local responses to peptide vasodilators: a NOVEL pharmacodynamic measure in the phase I trial of Antagonist G, a neuropeptide growth factor antagonistClin Cancer ResYear: 200173071307811595697 
19..  Rogatko A,Schoeneck D,Jonas W,et al. Translation of innovative designs into phase I trialsJ Clin OncolYear: 2007254982498610.1200/JCO.2007.12.101217971597 
20..  Simon RM,Freidlin B,Rubinstein LV,et al. Accelerated titration designs for phase I clinical trials in oncologyJ Natl Cancer InstYear: 1997891138114710.1093/jnci/89.15.11389262252 
21..  Gadgeel SM,Wozniak A,Boinpally RR,et al. Phase I clinical trial of BMS247550, a derivative of epothilone B, using accelerated titration 2B designClin Cancer ResYear: 2005116233623910.1158/10780432.CCR05012716144926 
22..  Tighiouart M,Rogatko A. Gail M,Krickeberg K,Samet J,Tsiatis A,Wong WDose finding in oncology—parametric methodsDose finding in drug developmentYear: 2006New YorkSpringer5970 
23..  Neuenschwander B,Branson M,Gsponer T. Critical aspects of the Bayesian approach to phase I cancer trialsStat MedYear: 2008272420243910.1002/sim.323018344187 
24..  Goodman SN,Zahurak ML,Piantadosi S. Some practical improvements in the continual reassessment method for phase I studiesStat MedYear: 1995141149116110.1002/sim.47801411027667557 
25..  Zohar S,Chevret S. The continual reassessment method: comparison of Bayesian stopping rules for doseranging studiesStat MedYear: 2001202827284310.1002/sim.92011568943 
26..  Cheung YK,Chappell R. Sequential designs for phase I clinical trials with lateonset toxicitiesBiometricsYear: 2000561177118210.1111/j.0006341X.2000.01177.x11129476 
27..  Gasparini M,Eisele J. A curvefree methods for phase I clinical trialsBiometricsYear: 20005660961510.1111/j.0006341X.2000.00609.x10877324 
28..  O’Quigley J,Paoletti X,Maccario J. Nonparametric optimal design in dose finding studiesBiostatisticsYear: 20023515610.1093/biostatistics/3.1.5112933623 
29..  Iasonos A,Wilton AS,Riedel ER,et al. A comprehensive comparison of the continual reassessment method to the standard 3 + 3 dose escalation scheme in phase I dosefinding studiesClin TrialsYear: 2008546547710.1177/174077450809647418827039 
30..  Mo Q (2012) http://cran.rproject.org/web/packages/CRM/. Accessed 10 Dec 2012 
31..  Shen LZ,O’Quigley J. Consistency of continual reassessment method under model misspecificationBiometrikaYear: 19968339540510.1093/biomet/83.2.395 
32..  Calvert AH,Plummer R. The development of Phase I cancer trial methodologies: the use of pharmacokinetic and pharmacodynamic end points sets the scene for Phase 0 cancer clinical trialsClin Cancer ResYear: 2008143664366910.1158/10780432.CCR07455918559580 
33..  Mandrekar SJ,Cui Y,Sargent DJ. An adaptive phase I design for identifying a biologically optimal dose for dual agent drug combinationsStat MedYear: 2007262317233010.1002/sim.270717016867 
34..  Zhang W,Sargent DJ,Mandrekar S. An adaptive dosefinding design incorporating both toxicity and efficacyStat MedYear: 2006252365238310.1002/sim.232516220478 
35..  Whitehead J,Thygesen H,Whitehead A. Bayesian procedures for phase I/II clinical trials investigating the safety and efficacy of drug combinationsStat MedYear: 2011301952197010.1002/sim.426721590794 
Figures
Tables
Example of risk level probabilities generated by the Bayesian curvefree approach
Dose level  Risk of toxicity  

5 % “very safe”  15 % “safe”  25 % “ideal”  40 % “risky”  65 % “toxic”  
1  0.48  0.44  0.08  0.00  0.00 
2  0.40  0.48  0.12  0.00  0.00 
3  0.33  0.52  0.15  0.00  0.00 
4  0.25  0.55  0.19  0.01  0.00 
5  0.17  0.54  0.28  0.01  0.00 
6  0.08  0.49  0.40  0.03  0.00 
7  0.01  0.28  0.61  0.10  0.00 
8  0.01  0.14  0.40  0.33  0.12 
9  0.01  0.05  0.19  0.40  0.35 
Bold value indicates the dose level with the highest probability of having the “ideal” toxicity rate
Characteristics of doseescalation strategies
Issues to consider  Algorithmic  Model based  Curve free 

1. Statistical properties  
1a Does method provide estimate of a relevant parameter and allow the precision of the estimate to be quantified?  Poor  Good  Good 
1b Does precision increase with sample size?  Poor  Good  Good 
1c Reliably arrives at correct decision on dose with target toxicity risk  Poor  Good  Good 
2. Simplicity  
2a Nontechnical explanation of method  Good  Poor  Intermediate 
2b Statistical complexity  Not applicable  Intermediate  Poor 
3. Intuitive dose recommendations  Good  Intermediate  Good 
4. Flexibility  
4a Target toxicity rate  Poor  Good  Good 
4b Accommodating underlying shape of dose–response  Poor  Good  Good 
4c Dose skipping  Poor  Good  Good 
5. Impact of number of doses in schedule  Poor  Good  Good 
Article Categories:
Keywords: Keywords 3 + 3 design, Bayesian method, Clinical trial, Phase I, Continual reassessment method, CRM, Curve free. 
Previous Document: Population pharmacokinetics of cabazitaxel in patients with advanced solid tumors.
Next Document: XRCC1 and GSTP1 polymorphisms and prognosis of oxaliplatinbased chemotherapy in colorectal cancer: ...