Eur J Clin Pharmacol (2002) 57: 835–845
`DOI 10.1007/s00228-001-0405-6
`
`R EV IE W A RT I C L E
`
`Bruno G. Reigner Æ Karen Smith Blesch
`Estimating the starting dose for entry into humans:
`principles and practice
`
`Received: 27 August 2001 / Accepted in revised form: 5 November 2001 / Published online: 29 January 2002
`Ó Springer-Verlag 2002
`
`Abstract Background: Selection of the starting dose for
`the entry into humans (EIH) study is an essential first
`step in clinical drug development.
`Objectives: This paper is a review of different approaches
`that may be used to calculate the starting dose, presents
`the results of a current practice survey that reflect
`practice patterns at a large pharmaceutical company,
`and discusses selected topics related to the calculation of
`the starting dose.
`Results: The methods used in the field of oncology for
`cytotoxic compounds are usually derived from a dose
`associated with some toxicity in animals multiplied by a
`safety factor. In therapeutic areas other than oncology,
`the methods may be
`classified as
`four different
`approaches: (1) dose by factor methods that utilize the
`no observable adverse effect level (NOAEL) from pre-
`clinical toxicology studies multiplied by a safety factor;
`(2) the similar drug approach that may be used when
`clinical data are available for another compound of the
`same chemical class as the investigational drug; (3) the
`pharmacokinetically guided approach that uses systemic
`exposure rather than dose for the extrapolation from
`animal to man; and (4) the comparative approach that
`consists of utilizing two or more methods to estimate a
`starting dose and then critically comparing the results to
`arrive at the optimal starting dose. A ‘‘real-life’’ example
`illustrates the use of each method. Advantages, limita-
`tions, and underlying assumptions of each of
`the
`
`B.G. Reigner (&)
`Pharma Development, Clinical Science,
`Clinical Pharmacology Science (52/1010),
`F. Hoffmann-La Roche Ltd, Grenzacherstrasse 124,
`4070 Basel, Switzerland
`E-mail: bruno.reigner@roche.com
`Tel.: +41-61-6884507
`Fax: +41-61-6881434
`
`K.S. Blesch
`Pharma Development,
`F. Hoffmann-La Roche Inc.,
`Nutley, USA
`
`methods are discussed. The results of the survey showed
`that the pharmacokinetically guided approach is the
`most commonly used method, followed by dose by
`factor methods.
`Conclusion: The task of estimating the starting dose is
`moving beyond empirical methods to those that are in-
`creasingly more systematic and theory based.
`
`Keywords Starting dose Æ Entry into human study Æ
`Pharmacokinetics
`
`Introduction
`
`The entry into humans (EIH) study is the first step in the
`clinical development of any molecule that has shown
`therapeutic promise in preclinical evaluations. An
`essential element of the EIH (also known as ‘‘entry into
`man, EIM’’ or ‘‘first time in man, FTIM’’) study is the
`calculation of the starting dose. Estimating the starting
`dose is a very common and important task, yet there is
`little uniformity or
`standardization of approaches.
`Starting-dose calculations are performed in many
`different ways, very often using empirical methods.
`The approach used often depends on the training and
`experience of the scientists involved, and/or the indus-
`trial or academic setting. Individual scientists may have
`their own rules and methods. Occasionally there is
`some consistency within a pharmaceutical company or
`academic setting, but the methods used vary consider-
`ably across these institutions.
`Estimating the optimal starting dose is complicated
`and presents new challenges each time it
`is done.
`Extrapolation of doses from animals to humans is
`based on multiple assumptions about the compound’s
`behavior across species. Different methods may yield
`widely varying results, and an approach that has
`worked well for one compound may not be appropriate
`for another compound. It is important to find a starting
`dose that is low enough to be safe in humans, but not
`so conservative that excessive costly and time-consum-
`
`CFAD Exhibit 1044
`
`

`
`836
`
`ing dose escalations are needed. These challenges make
`it difficult to establish standard rules for this under-
`taking.
`In 1981 it was noted that
`considerable
`uncertainty and controversy surround the choice of the
`initial human dose of a drug [1], a circumstance that
`was echoed in 1990 [2]. More recently, Boxenbaum and
`DiLea [3] argued that, while there were some basic rules
`that could be followed to ensure the safety of patients
`and healthy volunteers, the present state of knowledge
`did not allow development of realistic or reasonable
`standardized procedures for determining optimal start-
`ing doses for entry into humans.
`These sentiments seem to be confirmed by the wide
`variability in approaches used for starting-dose estima-
`tion and the fact that there currently are no guidance
`documents available from regulatory health authorities
`on this topic. A literature search identified a few papers
`that broadly address the issue of starting dose, most
`from the 1970s and 1980s [1, 2, 4, 5, 6] in addition to the
`1995 paper [3] cited above. The number of references is
`small
`considering that a starting-dose
`calculation
`is needed for each molecule early in drug development –
`a very large number. This is possibly a reflection of the
`complexity of the task and the complicated but incom-
`plete knowledge base that underlies it. The objectives of
`this article are to review the different approaches used to
`calculate the starting dose, to illustrate the approaches
`with a ‘‘real-life’’ example, to present the results of a
`current practice survey that reflect actual practice
`patterns at a large pharmaceutical company, and to
`discuss selected topics related to the calculation of the
`starting dose.
`
`General considerations
`
`For non-cytotoxic compounds, the initial EIH study is
`usually a single ascending dose (SAD) study in healthy
`volunteers. The main purpose of the study is to assess
`the tolerability of the new compound after administra-
`tion of a single dose and to gain some information about
`the pharmacokinetics and pharmacodynamics (if possi-
`ble) of the compound in human subjects. Dose estima-
`tion is based on a dose found to be safe in preclinical
`studies and then adjusted for human use, using various
`correction factors to ensure human safety. The optimal
`EIH starting dose is one that is safe, and not pharma-
`codynamically active, but is close to a dose with some
`minimal pharmacodynamic effect in humans. A starting
`dose that is too low results in the expenditure of addi-
`tional time and resources in reaching potential infor-
`mative and therapeutic dose levels, while a starting dose
`that is too high compromises subject safety and may
`overlook important clinical considerations for lower
`doses.
`The approach just described is generally applicable
`to compounds
`in all
`therapeutic areas with the
`exception of cytotoxic compounds intended to treat
`malignant disease. EIH studies for cytotoxic agents are
`
`conducted in patients with treatment-refractory cancer
`instead of healthy volunteers. In EIH studies for
`antineoplastics, there is always hope for a therapeutic
`benefit and a desire to minimize patient exposure to
`sub-therapeutic doses [7, 8], therefore the EIH study
`for these compounds is usually a multiple ascending
`dose (MAD) rather than a SAD study. In general,
`cytotoxic compounds have a very low therapeutic in-
`dex and a steep concentration–response curve for
`safety; however, in oncology there is considerably more
`acceptance of toxicity to achieve therapeutic benefit.
`The starting-dose calculation for antineoplastics
`is
`generally based on a dose and dose schedule that have
`elicited some toxicity in animals rather than on a dose
`that has been identified as safe in animals. Starting
`doses of anti-cancer agents have traditionally been
`established with the goal of escalating quickly to a
`maximum tolerated dose (MTD) on a given dosing
`schedule.
`There are multiple preclinical doses that can serve
`as the basis for estimating the starting dose. They
`have nearly identical meanings and are occasionally
`used interchangeably, although some subtle differences
`exist. For
`example,
`the highest non-toxic dose
`(HNTD) originated in cancer research and is defined
`as
`the highest dose at which no hematological,
`chemical,
`clinical, or morphological drug-induced
`alterations occur, and doubling this dose produces the
`aforementioned alterations
`[9]. The no observable
`adverse effect level (NOAEL) is defined as the highest
`dose at which no statistically significant and/or
`biologically relevant adverse effect is observed [10].
`For the purposes of this paper we will use the term
`NOAEL when discussing doses from preclinical toxi-
`cology studies.
`Figure 1 describes a simple decision tree for esti-
`mating a starting dose and serves as an organizing
`framework for this paper. Methods for estimating the
`starting dose are different depending on whether or not
`the drug is a cytotoxic intended for antineoplastic
`purposes; we have used this basic principle to divide
`the paper into its two main sections. If a drug is not a
`cytotoxic intended for antineoplastic purposes then,
`after estimating a starting dose on the basis of the
`NOAEL, it is reasonable to consider whether or not
`this dose is expected to be pharmacodynamically active
`and needs to be adjusted downward for entry into
`man.
`
`Methods used for cytotoxic compounds
`
`The literature related to estimating a starting dose for
`antineoplastics
`suggests an organized, knowledge-
`building effort by scientists focused on the task, and a
`clear evolution in methods is seen [11, 12, 13, 14, 15, 16],
`although considerable uncertainty continues to exist in
`this therapeutic area. Historically, starting doses of
`antineoplastics for humans were first extrapolated from
`
`

`
`Fig. 1. Decision tree for start-
`ing-dose calculation
`
`837
`
`toxic doses determined in large animal species (e.g., dog
`and monkey). In 1979, based on a retrospective analysis
`of 12 anti-tumor agents, Penta and colleagues [14]
`demonstrated that mouse data could be effectively used
`in determining safe starting doses. In 1981, mouse data
`were further validated against traditional large-animal
`methods with 21 antineoplastic agents and were also
`found to produce safe starting doses [15]. Methods
`based on large animal species and on mice are both in
`use today.
`
`One-third of the toxic dose low in a large animal species
`
`Using this approach, the starting dose is calculated
`as one-third of the toxic dose low (TDL; expressed as
`mg/m2) in a large animal species (either dog or monkey).
`This method was initially introduced by Freireich and
`colleagues [11] and remains widely in use. TDL is
`defined as the lowest dose that produces drug-induced
`pathological alterations
`in hematological, chemical,
`clinical, or morphological parameters and which, when
`doubled, produces no lethality [9]. The TDL is deter-
`mined on two basic schedules, single dose and daily for
`5 days [14].
`
`One-tenth of the lethal dose in mice
`
`Here the starting dose for EIH is calculated as one-
`tenth of a dose (expressed in mg/m2) that is lethal to
`10% of non-tumor bearing mice (LD10) during a
`specified period of observation [14]. The LD10 is de-
`termined on two basic schedules (single dose and daily
`for 5 days) with groups of ten mice at each dose level
`[14]. Collins et al.
`[17]
`investigated the use of 1/10
`LD10 and noted that, for eight drugs, the area under
`the plasma concentration–time curve (AUC) of the
`compound observed in mice after administration of
`the LD10 was similar to that of the compound pro-
`duced by the MTD in humans. The use of a new
`approach to dose escalation called pharmacologically
`guided dose escalation (PGDE) was advocated [18].
`PGDE is now a well-accepted element of oncology
`phase-1 study design. Most papers that address issues
`related to PGDE utilize 1/10 LD10 in mice as the
`method of reference for estimating the starting dose
`[18, 19, 20].
`If
`there is a significant discrepancy
`between 1/10 LD10 in mice and 1/3 TDL in large
`species, other authors have suggested that the lower
`of
`the two doses be used as the starting dose in
`conjunction with PGDE.
`
`

`
`838
`
`Example
`
`For each method described in this paper, a ‘‘real-life’’
`example of a starting-dose calculation is given. For the
`examples, we use the compound mofarotene. Mofaro-
`tene (Ro 40-8757) is a retinoid with cytostatic properties
`that was used in phase-I clinical trials for a potential
`antineoplastic indication. Mofarotene was chosen for
`the example because several methods for estimating the
`starting dose were used and compared before EIH,
`including methods that are usually restricted to cyto-
`toxic drugs. Preclinical studies had indicated a NOAEL
`of 2 mg/kg/day in dogs and 50 mg/kg/day in rats. The
`(95 mg/m2/day).
`TDL in dogs was
`5 mg/kg/day
`The LD10 in mice was not available. Using the 1/3 TDL
`in large animal species, the estimated starting dose for
`mofarotene was:
`1=3  95 mg=m2  1:8 m2 ¼ 57 mg
`ð1Þ
`where 1.8 is the average body surface area of a human in
`m2. The use of a fixed single dose in this example is an
`exception for an oncology phase-I trial. Generally, for
`cytotoxics, the starting dose for EIH studies in cancer
`patients is individualized based on patient body surface
`area rather than using a fixed dose for all patients [21].
`Because mofarotene is cytostatic, with considerably less
`toxicity than a cytotoxic drug, it was initially adminis-
`tered as a single dose to healthy volunteers.
`
`Critical assessment of the method
`
`Multiple variations of the two basic approaches have
`been used by individual investigators and groups for
`various drugs (e.g., 1/50 safe dose in mouse for faz-
`arabine, 1/3 dog TDL for elsamitrucin and docetaxel,
`1/20 lethal dose in rat for gemcitabine, and less than
`1/10 mouse lethal dose for trimetrexate, [22]). Despite
`the substantial work that has been done with starting
`doses for cytotoxics, there is still considerable vari-
`ability, the basic methods haven’t changed, and there
`is no ‘‘gold standard’’ for estimating the starting dose
`for oncology phase-I clinical trials. There has been
`wide concern that the basic methods provide doses
`that are too conservative and miss opportunities for
`therapeutic benefit in oncology phase-I and -II clinical
`trials [8, 16]. PGDE is considered to be an essential
`element of effective phase-I study design in this thera-
`peutic area.
`
`Methods used for non-cytotoxic compounds
`
`There are four basic approaches to estimating the
`starting dose for EIH studies for non-cytotoxic com-
`pounds. For the purposes of this paper they are called:
`dose by factor, similar drug, pharmacokinetically guid-
`ed, and comparative. The comparative approach uses
`data obtained from the other three approaches to criti-
`
`cally evaluate and determine which starting dose is
`optimal. We describe each approach,
`including its
`strengths and weaknesses, and variations that have
`evolved from the original.
`
`The dose by factor approach
`
`This method consists of identifying a dose (usually
`expressed in mg/kg/day) associated with a specific effect
`in preclinical toxicology studies and then multiplying it
`by one or more factors to estimate a safe human starting
`dose. A commonly used approach is based on the
`highest dose of the compound found to have no toxic
`effect in the most sensitive species tested in 4-week to 13-
`week preclinical toxicology studies. This mg/kg/day dose
`is then reduced by a ‘‘sensitivity’’ factor that adjusts for
`anticipated differences in sensitivity to the drug between
`each animal species tested and man. The sensitivity
`factor is derived from estimated interspecies differences
`in sensitivity to drug toxicity published by the Associa-
`tion of Food and Drug Officials (AFDO) of the United
`States in 1959 [23]. According to this ‘‘modified AFDO’’
`scheme, the maximum starting dose for the EIH study is
`the smallest of the following three doses: 1/10 of the
`highest no-effect dose in rodents, 1/6 of the highest
`no-effect dose in dogs, or 1/3 of the highest no-effect
`dose in monkeys [3, 24]. The smallest of the three doses
`is utilized because it reflects which animal species is most
`sensitive to the drug from the toxicology studies. The
`sensitivity factor (i.e., 1/10, 1/6, 1/3) reflects anticipated
`differences in drug sensitivity in the various animal
`species relative to humans [23]. If, for some reason, there
`is concern about the safety of the starting dose derived
`this way, the dose can be further reduced using an
`arbitrary safety factor.
`
`Example
`
`For mofarotene, the dog was the most sensitive species,
`with a NOAEL of 2 mg/kg/day. The starting dose was
`estimated using the ‘‘modified AFDO’’ approach as:
`ð2Þ
`1=6  2mg=kg  70kg ¼ 23mg
`where the NOAEL in the dog was multiplied by 1/6 and
`then 70 kg (average body weight of a human) to estimate
`a starting dose of 23 mg. Because the toxicologist was
`concerned about skin toxicity appearing several weeks
`after the start of administration, the 23-mg dose estimate
`was multiplied by a safety factor of 1/10 to give a final
`starting dose of 2.3 mg.
`
`Critical assessment of the method
`
`There are multiple variations of this approach that
`allow for more or less conservative results. Different
`sensitivity or
`safety factors may be applied. For
`
`

`
`example, a sensitivity factor of 1/2 rather than 1/3 has
`been suggested when extrapolating from primates [25].
`Kuhlman [26] noted that generally the starting dose is
`about 1/50 to 1/100 or lower of the no-effect dose from
`toxicology studies.
`This is a classic approach that has been widely used
`and generally produces safe starting doses for EIH
`studies. Dose by factor methods have been criticized
`because they ignore preclinical pharmacokinetic data [3,
`24]. The approach is somewhat simplistic and empirical,
`and easily lends itself to variations that may be consid-
`ered to be rather arbitrary in nature. It has been noted
`that this type of extrapolation from animals to humans
`is truly appropriate only if both show similar absorp-
`tion, bioavailability, biotransformation, and sensitivity
`to toxic effects by the drug or its biotransformation
`products [25]. While there is considerable flexibility and
`good safety with results, dose by factor methods may be
`criticized for estimating starting doses that are too
`conservative, requiring excessive dose escalations to
`reach a pharmacodynamically active or maximum tol-
`erated dose.
`
`Similar drug approach
`
`The similar drug approach is used when human safety
`data are available for a drug similar to the one under
`investigation and can serve as a reference point for
`estimating the starting dose [1, 5]. The ‘‘similar drug’’
`is usually of the same chemical class, with similar or
`related chemical
`structure. This
`situation is not
`uncommon in industry where one or more predecessor
`compounds in the same chemical class and with sim-
`ilar toxicological profiles may have been clinically
`investigated prior to the current drug under investi-
`gation. The ‘‘similar drug’’
`is one that
`is already
`marketed or that has clinical safety data available
`when the compound under investigation is a follow-up
`compound.
`This approach is based on the ratio of an optimal
`starting dose of the similar drug to its NOAEL. This
`optimal starting dose is one that has been identified as
`producing no drug-related adverse events or laboratory
`abnormalities after a single dose in humans and with no
`pharmacodynamic activity. The method assumes that
`this ratio is equal to the ratio of the starting dose for the
`compound under investigation to its NOAEL. The
`assumption can be expressed as:
`SDs=NOAELs ¼ SDi=NOAELi
`where:
`• SDs is the optimal starting dose of the similar drug.
`• NOAEL is the no-observable adverse effect level for
`drugs ‘‘s’’ and ‘‘i’’, where ‘‘s’’ is the similar drug and
`‘‘i’’ is the investigational drug.
`• SDi is the estimated starting dose for the investiga-
`tional drug.
`
`ð3Þ
`
`839
`
`The ratio SDs/NOAELs can then be applied to the
`NOAELi to estimate a starting dose that is expected to
`be safe, but not too conservative as:
`ð4Þ
`SDi ¼ ðSDs=NOAELsÞ  NOAELi
`The dose estimate obtained this way is usually multiplied
`by an arbitrary safety factor to accommodate uncer-
`tainty about safety in the estimate of the starting dose.
`
`Example
`
`For mofarotene, the similar drug was etretinate, a
`closely related retinoid with a similar toxicity profile in
`animals; the dose that would be an optimal starting dose
`of etretinate in humans was 10 mg. The NOAEL of
`etretinate in rats was 2 mg/kg/day and that of mofaro-
`tene in rats was 50 mg/kg/day. The NOAEL of etreti-
`nate in dogs was unknown. Applying Eq. 4:
`SDi¼ð10mg=2mg=kg70kgÞ50mg=kg70kg¼250mg:
`ð5Þ
`This dose was then multiplied by a safety factor of 1/4 to
`give a final human starting dose of 63 mg. In this case, as
`with the dose by factor method, an arbitrary safety
`factor was applied to ensure the safety of the healthy
`volunteers.
`
`Critical assessment of the method
`
`This ‘‘similar drug’’ method is not new, makes intuitive
`sense, and is known to provide safe starting doses. The
`main limitation is that applying a cross-species dosing
`ratio for one drug to another drug assumes that phar-
`macokinetic and pharmacodynamic differences between
`animal and man are the same for both compounds. The
`validity of this assumption should always be considered
`and may be tested by calculating the ratio SDs/NOAELs
`for another similar drug of the same chemical class to
`verify that the ratio remains reasonably constant. If this
`second ratio is not similar to the first, this approach
`should not be used.
`
`Pharmacokinetically guided approach
`
`The pharmacokinetically guided approach is increas-
`ingly being used in many pharmaceutical companies and
`institutions [24]. It uses systemic exposure instead of
`dose for the extrapolation from animal to man, a con-
`cept that was originally proposed more than three
`decades ago [26]. A desired systemic exposure (e.g.,
`AUC) for humans is defined as the systemic exposure
`corresponding to the NOAEL. If a NOAEL and its
`corresponding AUC are available from more than one
`animal species, the animal species with the lowest AUC
`is used. The clearance of the drug in humans (CLh) is
`
`

`
`840
`
`predicted using allometric scaling [27, 28, 29]. The
`starting dose, with AUC as the measure of systemic
`exposure, is then calculated as:
`SD ¼ AUCp  CLh
`ð6Þ
`where SD is the starting dose, AUCp is the AUC
`obtained at the NOAEL in preclinical toxicology stud-
`ies, and CLh is the clearance in humans predicted by
`allometric scaling.
`The correlation between pharmacokinetics and tox-
`icity in preclinical toxicology studies may be better with
`maximum concentration (Cmax) than AUC. In this case,
`Cmax rather than AUC should be used as the measure of
`systemic exposure. A detailed description of methods for
`and results of interspecies scaling as it applies to dose
`selection for humans is available for a new glycine
`antagonist [30].
`
`Example
`
`For mofarotene, the AUC in dogs was 17.3 lg h/ml at
`the NOAEL. Using allometric scaling and in vivo
`pharmacokinetic data from three animal species (mouse,
`rat, and dog), as well as intrinsic clearance data from
`mouse, rat, dog, and human microsomes, the predicted
`CLh was 16.0 l/h [31]. Applying Eq. 4, the calculated
`starting dose was:
`ð7Þ
`17:3lg  h=mL  16:0L=h ¼ 277mg
`This dose was then multiplied by a safety factor of 1/3,
`which gave a final starting-dose estimate of 92 mg.
`
`Critical assessment of the method
`
`This method also provides safe starting doses [24]. The
`main limitations concern assumptions about interspecies
`differences in concentration–effect relationships and
`metabolism. Defining a systemic exposure for humans
`on the basis of preclinical data assumes that the con-
`centration–effect relationship for the drug is the same in
`animal and man. This assumption should always
`be kept in mind because interspecies differences in
`concentration–effect relationships do exist
`for some
`pharmacological classes such as cardiac glycosides [32].
`If there are concerns about differences in concentra-
`tion–effect relationships between animal and man, it is
`important to quantify them whenever possible. With
`certain compounds, this may be accomplished with in
`vitro testing. For example, for a platelet aggregation
`inhibitor, human and animal platelets can be tested with
`the drug to determine the concentration–response curve
`and evaluate any pharmacodynamic differences between
`animal and man.
`If differences in the concentration–effect relationships
`can be quantified, then this information can be incor-
`porated into the calculations. As an example, assume
`
`that in-vitro testing determines that the maximum effect
`of a compound is the same in animals and humans and
`that the ratio EC50 human/EC50 animal =0.1, where
`EC50 is the drug concentration that elicits 50% of the
`maximum effect. In this case, the drug is ten times more
`potent in man than in animal, and the starting dose for
`man should then be reduced to 1/10 of that originally
`estimated. The method then becomes a pharmacoki-
`netically and pharmacodynamically guided calculation
`of the starting dose.
`this pharmacokinetically
`Another shortcoming of
`guided approach is uncertainty related to the prediction
`of CLh using allometric scaling. Allometric scaling is a
`challenging scientific area and there is always uncer-
`tainty surrounding the prediction of clearance in
`humans [24, 29]. Standard methods are known to work
`well for drugs that are eliminated primarily by renal
`excretion [33, 34]. Allometric scaling methods are less
`reliable for drugs that are primarily eliminated via
`hepatic metabolism. However, recent advances in this
`field suggest that clearance in animals can be corrected
`using intrinsic clearances obtained in vitro with human
`and animal hepatocytes [27]. Using allometric scaling
`combining in vitro and in vivo data from three animal
`species and in vitro data from humans, a retrospective
`analysis of a large number of drugs showed that the
`prediction of CLh was successful in 82% of the cases;
`that is, predictions were within a twofold factor of actual
`human clearance values [27]. For mofarotene, the mea-
`sured CLh was 31.5 l/h [31], slightly less than a twofold
`difference between the predicted and measured CLh. The
`uncertainty associated with allometric scaling should be
`kept in mind when using the pharmacokinetically guided
`approach. As with the other methods described, an
`arbitrary safety factor may be applied to adjust the
`starting dose on the basis of the technique used for
`allometric scaling and its degree of reliability.
`The pharmacokinetically guided approach carries
`additional complexities that must be considered. The
`first concerns which plasma drug concentration is
`appropriate to use in the calculations. When deriving
`measures of systemic exposure from plasma concentra-
`tion data for the purposes described here, it is important
`to consider plasma concentrations of unbound drug
`(Cu). Cu is responsible for its pharmacodynamic effects,
`and Cu in plasma is expected to be more closely related
`to Cu at the site of action than the total plasma
`concentration. It is well established that the unbound
`fraction (fu where fu=Cu/total plasma concentration)
`can vary between species [32, 35]. Therefore, using Cu in
`the calculations of systemic exposure (AUC or Cmax) is
`highly recommended, especially for drugs that are highly
`protein bound in plasma. A striking example of inter-
`species differences in plasma protein binding is etopo-
`side, with unbound fractions of 0.63, 0.52, and 0.048 in
`dog, rat, and man, respectively [32]. In this case, relying
`on total plasma concentration rather than Cu for
`calculating the desired systemic exposure could result in
`an unnecessarily conservative starting dose.
`
`

`
`Other complexities of the pharmacokinetically guided
`approach concern the dose linearity of the drug’s phar-
`macokinetics and its metabolism. We describe this
`approach in the context of linear pharmacokinetics and
`with the parent drug being the sole active moiety.
`Reality, of course can be more complex, for example,
`with non-linear pharmacokinetics and/or the presence of
`active metabolites that must be considered in the start-
`ing-dose calculation. Another issue may involve the use
`of different formulations in animal and human studies
`(e.g., drinking solution in animals and tablet in man).
`When experimental data are available for systemic
`exposure to active metabolites or the relative bioavail-
`ability of the formulations, this information should also
`be included in the starting-dose calculation. It is beyond
`the scope of this paper to describe the integration of this
`information into the starting-dose estimate because it
`will differ for each compound; however, the general
`principles
`regarding pharmacokinetics,
`interspecies
`scaling, and pharmacodynamics will apply.
`
`The comparative approach
`
`This approach consists of estimating the starting dose
`using at least two but preferably three or four methods
`and then comparing the results and interpreting the
`differences to arrive at an optimal starting dose. To
`illustrate this approach, we utilize the starting doses
`estimated for mofarotene in the examples (Table 1).
`For mofarotene,
`three different approaches (1/3
`TDL, similar drug, pharmacokinetically guided) gave
`approximately similar results (57–92 mg), while the dose
`by factor approach with a safety factor of 1/10 gave a
`considerably lower estimate (2.3 mg). Such results are
`common. Methods based on a dose by factor approach
`usually provide a much lower starting-dose estimate
`than other methods. With mofarotene, a starting dose of
`50 mg was selected for the EIH study in healthy vol-
`unteers. This dose was selected because it was felt that
`the 2.3-mg starting dose from the ‘‘dose by factor’’
`approach was too conservative, and 50 mg was at
`the conservative end of the dosing range indicated by the
`other three methods. Results of the EIH study showed
`that the starting dose and a dose of 150 mg were safe (no
`drug-related adverse event or laboratory abnormalities)
`and that the MTD was 500 mg. This ratio of 10 between
`the MTD and starting dose was satisfactory in terms of
`safety and efficiency with a total of four dose levels
`investigated in this study.
`
`Critical assessment of the method
`
`The comparative approach is not commonly used and,
`up to the present time, has not been addressed in the
`literature. The differences in the results obtained with the
`different methods would be expected to trigger discussion
`among the scientists
`involved in the starting-dose
`
`841
`
`Table 1. Starting doses for mofarotene as estimated from four
`different methods. A starting dose of 50 mg was selected for the
`‘‘entry into humans’’ (EIH) study in healthy volunteers
`
`Method
`
`Estimated
`dose (mg)
`
`1/3 TDL
`Dose by factor
`Similar drug
`Pharmacokinetically
`guided
`
`57
`23
`250
`277
`
`Safety
`factor
`
`None
`1/10
`1/4
`1/3
`
`Final dose
`(mg)
`
`57
`2.3
`63
`92
`
`estimate regarding the principles and assumptions be-
`hind each approach. This would lead to a better under-
`standing of the science behind the starting-dose estimate.
`Getting similar results with two or three methods such as
`we obtained for mofarotene is reassuring. This method
`could be criticized for being time consuming, but this
`criticism is not considered to be valid by the authors in
`view of the importance of estimating an optimal starting
`dose that is both safe and efficient. It is possible that the
`analysis of different results obtained from various
`methods could be confusing for non-specialists, leading
`to a rather low confidence in the starting-dose estimate.
`However, the alternative approach of relying on a dose
`estimated using a single method also should be ques-
`tioned.
`
`Is the starting dose expected
`to be pharmacodynamically active?
`
`For some compounds, particularly those with a high
`therapeutic index, a starting-dose estimate based on a
`toxicology ‘‘no-effect’’ dose may not be satisfactory.
`This is because, in such compounds, a ‘‘no-effect’’ dose
`in toxicology studies may still have some pharmacody-
`namic activity. In most industrial and academic settings,
`the starting dose is intended to be pharmacodynamically
`inactive, with the second dose having some activity. To
`avoid having a pharmacodynamically active starting
`dose in compounds with a high therapeutic index, the
`starting dose may need to be lower than that estimated
`from toxicology studies.
`The dose that
`is predicted to be the highest
`non-pharmacodynamically active dose in humans can be
`estimated using variations of the dose by factor, similar
`drug, and pharmacokinetically guided approaches that
`utilize data from preclinical pharmacology studies
`instead of toxicology studies. Many variations of the
`dose by factor approach have been described in the lit-
`erature. Posvar and Sedman [6] suggest that the starting
`dose is generally based on a fraction (1/10 to 1/1000) of