`
`ISSN: 1054-3406 (Print) 1520-5711 (Online) Journal homepage: http://www.tandfonline.com/loi/lbps20
`
`Dose response studies I. some design
`considerations
`
`Stephen J. Ruberg Ph.D.
`
`To cite this article: Stephen J. Ruberg Ph.D. (1995) Dose response studies I.
`some design considerations, Journal of Biopharmaceutical Statistics, 5:1, 1-14, DOI:
`10.1080/10543409508835096
`To link to this article: https://doi.org/10.1080/10543409508835096
`
`Published online: 29 Mar 2007.
`
`Submit your article to this journal
`
`Article views: 135
`
`View related articles
`
`Citing articles: 71 View citing articles
`
`Full Terms & Conditions of access and use can be found at
`http://www.tandfonline.com/action/journalInformation?journalCode=lbps20
`
`ATI 1018-0001
`
`ATI v. ICOS
`IPR2018-01183
`
`
`
`journal ot Biopharnraceutrcal Sata\tm, 5 i j , i - i i , l r r S ,
`
`Ke:,, void.^. Dmg development; Factorial studies; Minimum effec-
`tive dose; One-sided tests; Randomized concentration controlled trials
`
`A critical aspect of biomedical research is the characterimtion of
`response reiationship el' a c~rnpmind. This is iriiii iii lab-
`the
`oratory experiments and clinical triais and pertains to efficacy, safety,
`-. , . .
`resuifing brnrili,/r~~$. ratill. fiesen:i,ii, here is Pzrt I (;[
`;hi:;
`2nd
`article, which deals with some ciinical trial design issues surrc)unci-
`ing dose response studies. Some additional comments are made about
`trials for identifying the minimum effective dose, randomized con-
`centration controlled trials, and the use of one-sided hypotheses in
`6esigr.i~g suck! trials. Part I1 is a separate paper reviewing some
`zn~lysis srr&egie.: f ~ r <.!out- i-esponw sti.ii:ies.
`
`Understanding the dose response relationship of a compound is a fundamental
`aspect of research; indeed, it may bc the central issue. This is true whether
`studying a new drug, assessing the effect of environn~ental toxins, or eval-
`
`Copjr~ght C 1995 by Marcel Dehker. Inc
`
`ATI 1018-0002
`
`
`
`2. Some Experimental Designs
`
`2.1 Parallel Designs
`
`The most ronmon and straightforward design is the placebo-controlled, ran-
`domized, parailel dvse response study. In this siiliiji design, patients are ran-
`domly ailocated to one of several active dose groups or placebo. This design
`is nost p ~ p u j a r since the ~ n i y diffcr~ncx ictwccn xrcrrrment groups i:;
`the
`dose of the experimental compound, aiiowing for straightforward interpre-
`tation of the results of such a trial. it is also imporiant that the study inciude
`a placebo group since a significant trend in response with iqcreasing dose in
`the absence of placebo is not necessarily evidence of a drug effect (Fig. 1).
`
`ATI 1018-0003
`
`
`
`Dose Response Smdies, i
`
`c-.-
`
`Us-
`
`In Figure 1 , if the placebo response were absent, onc might conclude there
`.
`. .
`dose effect because respcmse is increasing with dose. How-
`Is a .;lonif~canf
`-IblLa,;,,g
`ever, when the "trendB in psponsc .#ith irn.--
`is taken ir, light of
`$OX
`the placebo response, it may be doubtful that there is any drug effect.
`There are some exceptions to this principle of needing a placebo group
`to assess the significance of dose respozse. Ir, s ~ m e cases, the historical rL nia-
`cebo response is nil or nearly nil (e.g., spontaneous cures of serious infections
`. .
`such 2s efidecar-ilis, the absence of nailsea and vc-,mi:ing fs!!owir,g highly
`emetogenic chemotherapy, or chronic asthma), and therefore, a concurrent
`placebo group is not necessary. Furthermore, ti-om a safety staidpoint, there
`may be instances where a single serious adverse event is evidence of a toxic
`drug effect in the absence of placebo.
`.
`.
`Ill
`instances,
`like to use dose t;trat;on to assess the dFdG b
`effect. This is appealing early in drug development programs since patient
`safety is generally of greater concern. in these trials patients can 'bc started
`at low doses, and depending on their response, doses can be increased grad-
`uailj: tc: achieve a suitable dnse f ~ r
`the patient. The al;ses;men[ :,f ai: in&-
`vidua! p&ient's dose response is satisfying ciinicaiiy and may require fewer
`patients to assess the drug effect since within-patient variability is used.
`The disadvantage of titration studies is that dose and time effects cannot be
`completely separated. Since efficacy and adverse events may have a time-
`
`ATI 1018-0004
`
`
`
`m
`
`The disadvantage of a parallel dose response trial is that the precision of the
`ii:fsrence is driver, by between-subject variabi!ity, which usaa!ly requires greater
`sample sizes to increase the precision of the estimates for drug effect. To
`overcome this ilifficulty, crossover trials can be ~atilizcd so that within-subject
`variability, which is most after, sma!!es
`than between-snl?ject variability, car?
`5e used in the infereccc. Thc statistical criteria for when 2 crossover design
`is better than a parallel design, as well as practical considerations (e.g., the
`stab;iiry =f :he disease state c.x,rer time) are well knuwr! (2). lhrrp I r e several
`variations of the crossover study, but two fundamental designs are the com-
`- , , - A ~ - ; , ~ A
`r l n r i tho. AACP n r ~ ~ i ~ t
`i n n
`C m c c n x r a v -
`- l - t - l . ,
`I - V A P O ~ T T O V
`~IIGLLIY 1a~luuIIliLLu L I V J O V V L I
`U L I U ~ 1 1 b UVJC.
`LOLUIULIWII
`~ I W ~ O W V L I
`.
`
`Typically, Latin squares are used for completely randomized crossover
`studies. When a very broad dose range is of interest, the use of incompieee
`crossover studies (Youden squares or balanced incomplete block designs) can
`be employed effectively. In many clinica! settings it is not possible to evaluate
`clifiical endpoints in a short period of time or ;vithout se-e carric\x:p,r
`J - . " . effect.
`However, in c!inical pharmacology studies where surrogate endpoints of ef-
`-LLy f Lirc , used in the decision-making process ol drug deveioprnenr,
`C- licacy or
`such conditions may hoici. Furthermore, pharmacokineiic dose proportivri-
`" - < .--- L; ,-?-,., :1:.1-:1'
`: O I : ~ I U I ' L I V ~ ~ X : C C I I ; I I ~ u . ~ a v i l i : a u l i ~ i ~ studies ofieri have many
`aii-iy ;ti?c&eh o r ' .--.-I.-':
`treatments or doxc groups. Such studies are weii suited to the use of Youden
`squares ur variaiicins on balanced incompkte block designs. 'if there is a con-
`trol group or reference formulation that serves as a control, optimal blocks
`designs for comparisons with control have been developed (3).
`
`ATI 1018-0005
`
`
`
`cludie~ arc !ncrewing ifi popularity 2s resr,arcfiers
`Factorial d r . ~ e Icsmnw
`I -
`recognize the importance of studying two factors simultaneously to find op-
`timai dosage regimens. There are several different factorial dose rzsponse
`studies that are very useful in drug development.
`- A of study considers the magnitude of the dose and the frequency
`One type
`of dvse as the two factors of inkrest (Design 2). As seen in that design, there
`is need for only one placebo regimen, which is usua!!y covered by r'hc zero
`dose for the most frequent regimen of interest, in this case BID dosing. This
`design is most usehill in mid-to-late phase I1 ciinica! development ir? order to
`find an optimal regimen for phase 111 studies. This type of design might also
`
`Period
`
`Design I . Dose escalation crossover study
`
`ATI 1018-0006
`
`
`
`l.,
`LgL
`
`-
`
`. .
`
`LiLcr ,, 33 .. ..,.,-t- vluc-zlal.
`tsial since it c ~ i i l d s&sfiji &c Food afid D:un xAL:&r.:?r.-
`..nnil
`,
`--"- .
`-
`.
`.
`.
`
`trial, However. this
` &
`c
`2
`i & ...+ -.--- & - - "
`i, ...., -..
`. .
`
`.
`-
`.
`.
`yq,7iii,, --- ..-- 3 : - - -;-,--
`,. ,,?,,
`JCLlllvlb 3 i L C Z . I~ : I i L L l a ' r : l u I I > are a cac;.nzern or
`iiiwuiiiaiay
`-
`IbyLrlLb
`if pairwise ~ i i i i i p ~ i s ~ i i s are required, auuiiionai patienrs could be piaced in
`
`the placebo group and on the highest dose-by-fi.equency regimen to get a
`more powerful test of the drug effect, Of course, this assumes that the greatest
`dose amount produces the largest difference in response from placebo.
`Another variation on this type of design is with intravenous !i.v.) drugs
`where nne f a c t ~ r is the i.v. bolus dose that may be given as initia! them PI T r
`and the ~ t h e r factor is the i.v. infusien rate used for maintenance therapy
`(Design 3). In this case one can study a placebo for both bolus and infusion
`rates, thereby giving ar, independent assessment of pure dose response h r
`
`1
`I
`
`I.? lnfusion Rate
`
`I
`I
`
`Design 3. Optimizing i.v dose regimen. Generally. use equal sample sizes in each cell
`
`ATI 1018-0007
`
`
`
`- .
`
`.
`
`- 3
`
`3
`
`*
`
`3
`
`i
`
`t
`
`
`
`L S , < . , ' X < ? , % & " ,
`
`L l z . , L
`
`0
`
`. .
`,..
`o[ adminisiraiion, Fur ci-inpie, -",- u,r, C;--,.+ > r
`L L I L u ~ ~ > L + .
`each
`r n fi3;~7r
`i u w O ! * * Y O
`~ i o ~ r ~ ,
`. . i > = r .
`k it LLZO b : ? l ~ ~
`d~)..ii-. ~ r c v i d e s the dose r-r.spon<e rei:l~iui?ship for
`w!ilcrr illLib
`L
`.,,L',-L
` .
`..
`.
`,.. .
`-
`the int'gsiofi r a w >miiari)/ . ii\c tirsi ~iiliimri 2: ~ h c cicsifp.
`:or v;tl;cn inere
`i s a zero infusion rate. yiveh ii piire dii;e rcsponsc: rclatim&ip Pi;.: the boius
`... - :>?iir'r. -
`_ -_.., i!g
`l-ic:-in tij idefitifi- .,;y.hich ci;m-
`, \-..,~... . ;;:
`-
`. - - - t ~ e i ; ~ f ~
`I*-
`rhc
`,*,.:
`ihr*
`r ~ j
`j ~ ,
`
`,,
`.-i.,L:i # . f i - r i . :
`.,- >
`.
`. .
`.
`;r> :;;a>, pL: <%g;!(Tl;ll~
`,- 7
`Cj1
`-, + * , . . ? , I
`2 .trk
`L z i L , ~ -,L,
`:71..<
`.
`ci-,~ in -*fiiCh ;UIi;~i;;i;ti:f;; ;her-
` norh he: p p l ? i a r I'accoI
`expti-ifieiii
`apy is desirable. Such concerns arise in a number of therapeutic arcas, most
`=pJ . d. - - a - . L , & A -
`theyiFv. 1 he
`-nt;'nini;,- iherun\,, ynCj zl-,ti)ivnerii,.nqiVe
`-L-----+L,,+--,-,,
`- - ' - I - ' - .
`$l%JL:L!di y
`ri
`--
`dcsigri is GEC ic ~ h i c h ~ ' C T ~ C L ~ S d ~ s e !eye!j cf each & - ~ o
`D ure
`,zri.ics-(:l;ls~:fi&
`the creytcc': clr!d\i invnlving i v rnerapy
`t uw.3.b.. 2: ., . Thi-
`t::
`- ..-- .. i: ~
`~
`c
`r
`~
`~
`~
`
`.:::::j!::r
`/?.A;.;.;
`.
`-
`in ihar a :;cc;";rstb: 3sscss:ne:?L of d ~ ~ r r .
`each drug cai: bt: :naae Dy
`respilinsc
`.., .,-,-. ..' ; .;. - ;hc i.; ,-2: - - - * \ I
`. .
`-
`5rs: ci>{~I?;ll
`&%is;: :.-:I-
`.s[
`of
`c!esigE~
`,Ak
`-
`< ! Z
`'- >
`-
`.
`.
`-
`ievici+. 6: :;cm< j c s : ~ ~ ?
`iicd mezh;>~njr:l,rri_'N ~ ~ \ , g r i C ~ j r L , j & ~ ~
`r; <!i. { T i .
`- > --
`. - .
`S:3tir;ticians have lone - recogrll~rd
`the advantages of using fact~riai ex-
`-
`: y : - i ; -.-. -..- ..- ..-.. . .I--. ..- .-:--~. G.;--ii;;r;:
`U " r l i l 1 2 i l ~ ~ , Llicir Zil"
`.
`.
`.
`.
`,linizz;
`:>
`., .dl.-. a-.::
`l L . ; L s r ~ i ! C ~ i ~ 1 . Z % l i L i &
`f2iriy ;ccln[, Thexi ure \.pry ilseful f&r Linsivei-ii;s 173UltjBle dGsr reis;lonsc sues-
`jimii]taiico.i;sly. TI-aditicnaliy, clinicims hive heen reluctant to use iac-
`.
`torial studies becausc h e y require a iargcr saiiipic sizz, o: for a fixed sample
`
`size, there are fewer observations ai each dose regimen. Obi~i~fi~ij.., the ie-
`quirernent for larger sample size in such studies is needed because several
`hypotheses are being evaluated within the same design. 4 s mentioned earlier,
`nne can intentionailv irnbaiance the design and oversanipie the placebo group
`2nd a high-dme group to iiicrcase the c h a x c s of having a statistical!y Q ~ O -
`
`r-.A-L2.i!iiL!,
`
`.
`
`* ~ . .
`
`> A * - L
`
`.
`
`5 - . , % 3 L
`
`A.
`
`c,
`
`.
`
`"'D
`
`Design 4. Optimi~ing combination therapy. Generally, ute equal sample sizes in each cell.
`
`ATI 1018-0008
`
`
`
`~
`
`~
`
`~
`
`<- c>::. -,:-.2j
`i7iYLlui
`
`<
`
`*
`
`2 -
`
`E
`
`1
`
`.
`
` a
`
`.
`. . . . . . .
`. . . . .
`$- - - --- 1s~iigs rcgar&lig il;l< idenGficc-aiion
`i;i:cre i;:)e;l:j;i_
`I h ; & -,c~:i:;n
`~ ; ? g i i
`-;:ith
`- - * - - L > b LLL33L \ * . ' 3 1 7 z -- ,. I : i i p b
`.. %....
`at the rn;njmum
`.......~..... < :. .....
`i-ffp<-'i\~"
`i'i flrr\,: -1-
`V " ..)-
`i * i i t l
`fCiP ii'?".
`.-'.,I?,
`iij'ii3
`the WED is ::f great inkre:;!. There is :he perception that the pharmaeeul;icaj
`.
`. inausrry
`sruales
`too
`.....
`. - .- = t
`iu
`.
`.
`- . .
`3 f c : r,lii;a~y 11; ifielr cilriliaj
`.
`. iiidi3. . L
`can be &yiaTzd significaniiy d:$ierefit froin
`' : : A w3G3
`~
`~
`~
`~
`~
`.-! .-.-
`I - i ~ ~ i i i i L
`-
`cr;aLebo in
`- 4 X --,.-. ,i
`.-,--
`
`LGIIIIS v LIIGII C I I L L L ~ L ~ , L I I G ~ G ~ m y be lower dmes that ai-e effi-
`-f +I-,.:-
`-
`cacious and wouid provide a wider safety margin. In fact, the FDA has made
`a formai assessment of dosage changes and reports that approximately 10%
`of drugs (new molecular entities) that were approved in 1980- iY8Y have had
`dosage changes (mostly decreases in dosej of greater than 33% ((6. There
`has also been some recent controversy sumunding the drug Ha!cjon@, am!
`in one statement, the European Community's Con~inittee for Proprietary Me-
` the lowest effective dose should
`dicinal Products JCPMPj has suggested, " . .
`.
` . " (7). Finally, with the lilcreased crriphasis on switching drucrc 6"
`be used .
`.
`to OTC status, Carl Peck, formerly of the FDA, has suggested that one of
`. -
`the fundamental questiGas that needs t= be ar,sxvr;eied for O'rC s.r~;t-h
`w ILLII ~ a u -
`*.',-
`didates is "what is the MED?" (8). While this principle of determining the
`MED is easily stated, it is very hard to define in practice and has both an
`efficacy and a safety compcnent.
`I would like to propose the following definition of the MED. The MED
`is ;Ize s;;zaj/est dGse pf.oducing
`c/i;;icGi/y inzlpoi-tcxnt re
`ti'lat i?an be
`declared statistically; ,sigi?$ficantly &,fir-r.n<frnrn the plircebo respmse. Therc
`needs to be a clinicaiiy lmgortant response because small, statis~icaily sig-
`nificant responses can be meaningless. There needs to be a statistically sig-
`nificant response since large, clinical responses that are nut clearly distin-
`guishable from placebo response do not provide substantive evidence of dmg
`effectiveness.
`When doing a study to assess the MED, a placebo group is essential for
`the reasons given previously. Since many dose response studies have iden-
`
`ATI 1018-0009
`
`
`
`Dose Response Studies. ?
`
`I
`
`- . . .
`
`.
`
`. ~
`
`-
`
`~
`
`7
`
`.
`
`
`
`,
`
`* C A I . r l i c L L i L . , L i .
`
`- .
`
`3
`
`.
`
`
`
`a
`
`
`
`.
`
`"1;
`
`F,----...c
`
`&
`
`
`
`- ~ --
`
`
`
`e - . . - z \ c o c
`
`l x r r c
`
`: !
`.-.a-
`
`?,,<,,,
`
`-n=wros.7
`
`~
`
`~
`
`f...dC15L
`I
`
`=
`
`-
`
`.
`
`i ~ ~
`
`s
`
`-
`
`.
`
`
`
`~
`
`~
`
`i
`
`~
`
`g
`
`is probably :.\,ise
`;el,,erai objecti;.es,
`MUD as on:)
`rificatiorl
`to chooss doses that an: equa!ly spaced on ar. ordinai scale or a log scdc. i i i
`i;huuic; !i:ci"dc
`Je3igl-!ii-,g the srudy,
`ieasr Q E ~ <ii5i
`iii;i-,ecii;&
`j;;;
`LriLri
`i:,
`-
`. . - -
`.
`. ;.2,, ;lo diitere,?[ ihail piacebo. i i ai! d o ~ c 3 J ~ C 2iri_:i,i:ic.
`be ii;lCi~~~ectEVE,
`rj;i;-;;
`. .
`ihr. dl~sc respiince
`to a;,nl::atz sir,^:: the r.at%re I.>!
`+c "ijf D i n a j be diffi2:;l:
`and
`- - decjared cffCcli~?ic.
`bct~v\~2C~;
`pj;iceb.l
`. . . - . . iq
`xx l;l;-i;
`. > - . > ; . ;-.*-' .i,-[!
`tiaz ian.2si ;iiikc.
`....
`~
`f
`i
`~
`~
`~
`~
`
`iii&F Ejj; be t. .-!! <."
`The choice oi sample size in dose response srudies is a difflcdi issue.
`.-.
`..
`.
`.
`I ; i ~ ~ l i i x i iii2~0rla11~ ques;!~:; !s hox&
`i:,istribule the <:;Epic Size ::~IOSX :f:r
`dnsc grmrps. Of course. this is cioseiy rieci to the anaiysis strategy for ir-'-'-,-
`'ranulg
`.
`.
`
`diise , p r ~ ~ p j and Pi&22bo, zOmc =.:-?r,,r,.!c:, L,-u .. A " w pL:L+JGiL.7,
`rnmpariyonh
`.
`.
`such as Eur;nc;;'3 ti-:;( (9, ep[!:;?;zca
`paiirl>is in rbi.
`-
`n:?:ii.ins~
`rrl<,r-;:
`. .
`. A!lhcmgh H 3 1 ~ may be il good idea for piikwise comparise;.ns,
`cimtsrd - g,r~?~?p.
`,.
`..
`ilc;fiti~;ist tests ( 1 ~ jc discc:,hej jiil_:!e ;g rsgt 11;
`; ~ ~ o r ~ ~ ~ r
`:_:.:_.r:_:>3 :-:-:an?.
`,;,\. - 5ius
`i ... ... .. , . , I L I I
`.- .
`,...
`. ,-
`' l i - , - . i i i i i ,,,- ,,,..
`,,,-- ,-&-;( '- >q. '-.,-..>.-.I!:-
`i:li :i2xpie size In
`iiuL &,
`.r!;c&h:>t;n,. +L.-
`_
`---! ik-.
`;* ii - - *
`, - - - - - - - - -
`- , --:J
`c z L u i u L , ;
`.. e..,'q,.:".
`'-.'-' '
`-.
`.
`_ _
`. > * - : p y ~& ? ; ' ~ Fjq2;iV ::c::;j;rc
`r,,
`. ~ . ,
`: <.p,,.,, ,.,, ,; -A.!!!.L'h
`-
`.
`- .
`i l I
`. .
`,-. i
`i:;:ereni=e fgr ~
`~
`~
`~
`an :3x:eraj! d 3 2 ~
`&&srLLc4;
`;!ii (37J, j,s
`for
`
`- y ~ ~ ~ ?
`. ~ ~ ~ ~ i
`..c.>::-+;.n.7!
`-
`6 9 s ~ ranee being
`effect. wuuiJ be to
`mgrc &:a
`extrcmeq of
`Piiii-j~ ,,,,; numbers of patients wor-~lil be a!lncated to the
`* .
`siuuibci, 1
`.n-i v,,,,,,,,,,
`.-L-..+lc.2
`~
`~
`-
`
`a < $ L
` while intermediate-dose groups or
`placebo group and the high-dnse group,
`regimens would aii have snialler but equal numbers of patients a!!ccated.
`There is no simple answer to the sample size al!ocatlon problem. One
`must carefully define thc objectives of the study and the primary test of in-
`terest
`then coinpuie p o ~ e i - f o ~ various al&-nativc hypotheses sr,d sar*l,p!e
`size configurations in order to obtain the bcst study design. In some instances
`simulations may be required when power calculations cannot be easily done
`for more complicated analysis strategies. An analysis strategy rhac maximizes
`the minimum power against a variety of likely or suspected alternative hy-
`potheses is generally most desirable.
`An alternative design to the randomized parallel dose response study for
`assessing the MED is the following. One couid randomize patients in doubie-
`blind fashion to drug or placebo. Patients are given a supply of medication
`with instructions to take, for example, " 1-2 tablets, once or twice a day."
`-
`number of
`ratients wouid they! tilratt; themselves by taking the
`tzbiers over rbe course of the day to cmtrol their symptoms. One could es-
`timate the MEU by assessing the average daily dose consiimed by the patients
`during the study. Furthermore, if the response to drug was significantiy dif-
`ferznt from placebo, then .me coilld claim xhar this estimated MED is j i g -
`nificantly different than placebo, thereby satisfying the criteria for an ade-
`qmte, well-controlled trial. Of course, this study design is usefui oniy when
`the disease or symptoms can be evaluated by the patient (e.g., pain, allergy).
`It also is appealing in that it mimics real life situations since patients often
`
`ATI 1018-0010
`
`
`
`*
`
`.
`
`,
`
`
`
`,'i;c;a
`
`t > L l L L i n L L L
`
`~
`
`"
`-
`
`~
`
`~
`
`~
`
`1
`
`L
`
`t,K
`
`T7.-.*-77
`
`
`~ W L
`W U L & U I ~ % O
`1
`
`
`, . n
`an altcmii[ibe ;;: :'afidc:m;rcg7 psrai]ei
`rrms, s a i m h s n a i ~
`,,,.,,
`+">c;>.\n~,>
`----- ,.- ,J -:,- -3 ---.,
`. .. . .. r .;--:. : z ,3fififixij7p<f . - i ~ ~ r ~ ~ f ~ ; ~ $ ; f i ~
`...... T';CK
`.; i v j
`. . 2 n- . I .
`c c ~ t ~ > ! ! s . A
`+-;-I-
`/ )
`. -
`L i i G - i 3
`& > L u p " L z . 3 L L u
`L.-,Li
`L . d L I L i - L * L i . - L A
`L
`iDPP'rr TL-00 tv;,,lc *>vo x r n m r rirniiar tn A n r o r o c n n n c o t r i , ~ l r o v r o n t n q t i o n t c
`\S%.x-%-
`1 J .
`
`l J l b 3 & . ,
`~~.
`~~.
`& z A u a L > U L
`U a - d V b 3 V
`.-, -
`L X T U:/ob L b L ~ y ~ ~ ~ . 3 U
`aujes, p.&tiea~j
`.
`are raiiciiimli?.' -.- - .. , .. .- -.--- '-;?-'"
`.-- .'-- --. .- ...
`iiijii;ilii 01 f i ~ c s
`.,..,..
`!/i()(l,A ii,ii( c , l . i i i i i ; i j f i
`,u
`-
`U:cGsdU I L V L t J . --- -
`n7:>;n* ,.><ce(; .-- ?t>Q<g, ,-.-.--.-.**--%-<>-:.-.%*
`.
`.
`.
`%..-.*.-.=c L : -.-. -.-;+.-*;=-
`,--. 2 ? .7y;-
`,::,=.
`?-;
`?:.
`a
`,
`.
`
`..>*--
`~
`c u b L I I U I L I L I I I I I Q _ I \ L i i : b ~ n u c i u uSJ::.V~::LlaL:'V.:l
`I::&>
`I.J X*J::lL.JII::ij
`the trial and adjusting thi dose as necessary to maintain the cijncciitraiion
`ing
`range. This is motivated by the fact that individual pharmacokineeics may
`play a large roie in assessing the drug effect. That is, individuais randomized
`to a fixed dose may exhibit widely differing responses due to widely differing
`pharmacokinetics. Those who respond may absorb the drug more completely
`or eliminate it more slowly, thereby producing higher blood levels of the
`drug. Those who do not respond may 'naive lower 'niooci ievels Gom rapid
`elimination or incomplete absorption. This has intuitive appeal since dose
`response trials are based on how much drug is ingested (i.e., gets into the
`gut), whereas concentration controlled trials are based on how much drug is
`circulating ifi the b!=o&trea,vk. &es~r;,ab',~, provides a better meta~,eter
`for drug efficacy since what is circulating in the bloodstream is likely to be
`closer to the site of action than what is ingested into the gut.
`Such studies may be useful for drugs that have a direct or immediate
`effect on an easily measured endpoint (e.g., blood pressure). They could be
`c.-.* -L.---:-
`+L<.t L<2?rc Ice'= ch;e,,&x,-
`V G I y UIIIILULL to I I I I ~ L L I I I ~ I I L I V I L i i i V i i i L U I U ~ U L ~ ~ O C L ~ L I L LIUIU
`.A;cccccc
`.-l:G-:.-::1+
`~ b c l o ~ V J U L C ~ V L
`p i n t (e. g . , congestive heart failure, Alzheimer's disease). Another diffi-
`culty in randomized concentration controlled trials is that using the drug con-
`centration in plasma as a metameter for assessing drug effect may induce
`more variability into the analysis than using dose. If concentrations are mon-
`itmed periodically throughout a trial and doses adjusted accordingiy, there is
`a tremendous reliance on that single blood level as an accurate measure of
`the patient's exposure to the drug. In fact, that concentration may be driven
`by the very recent dosing history of the patient (i.e., the time of the last dose
`
`ATI 1018-0011
`
`
`
`Dose Response Studies. 1
`
`a c s %
`
`ed Versus Two-Sided Hypothesis Testing
`
`-C n n a c;riori x r e r r i i c t \ x i n - ~ i A ~ T i h\innihPYiY i r \ i ; r l u ~e
`__I-___a --,elxlS i(T1 De One
`TL- t--;-
`l l l C LUplL U l ullb-aluvu v v ~ v u u r r , u uA u v u yV---il-l
`of those statistical issues that draw strong opinions from both sides. While
`the issue is relevant for aii phases of drug deveiopmeni, it seems of panicular
`relevance in designing and analyzing dose response trials. The topic is in-
`cluded here since the issue really starts with the design of the trial and the
`Ir,suriltlr " j;iitiplb jiii; i i ; y t i l i i L , . v i i L J The analysis then f n l l o ~ s nzt~rl!!y from
`-:-- ..l\-...lrYmPnt"
`- - - - - - I ,
`!.';..
`the design of the trial and the scientific hypothesis of interest. A series of
`articles from this Journd by distinguished authors give a comprehensive re-
`vie\;r/ of this issue ( l l- 15). There is very little to add to their arguments, bui
`the salient points wi!I be summarized here.
`With the exception of Dubey, all authors favor the use of one-sided
`hypotheses for comparing experimental therapies versus placebo. As a gen-
`eral rule, the authors who favor one-tailed hypotheses justifiably base their
`argument on the fact that the most important error to control is Pr (drug
`
`ATI 1018-0012
`
`
`
`an incredibly conservative control of the false positive error rate. This con-
`scwatism comes at quite a price. By switching to me-tailed hypctheses at e
`= .05 instead of two-tailed hypotheses, approximately 25% fewer patients
`could be utilized. Since much of the clinical development costs are driven
`by the number of patients (number ~f case report, farms; z m o g ~ t sf
`o
`suppiics; number of invesf gators; number of personnel to nmnitor trials, build,
`clean, analyze, and report on the data), and clinical development costs are
`typically tens of millions of dollars jscmerimes hundreds ~f miiiions of dol-
`lars!), the potential savings are enormous! There is also a tremendous poten-
`A . 1 "-- '
`~ i a l baviiigs in the development time f ~ r ne\i; drugs, ctmcntly 7-!G jicars. I t
`should be noted that the cost and time savings are not the ends that justify
`the means, but rather a beneficial outcome of taking the appropriate scientific
`approach.
`Finally, several of the authors discuss the significance level of hypoth-
`esis tests that is appropriate when &siEning studies. Next; drugs that are fidr
`-
`-
`-
`non-life-threatening diseases or that represent an incremental improvement
`over existing therapies require two trials demonstrating a significant dnig ef-
`fect at the a = .05 ievei from one-railed tests. New drugs for Me-fhreatening
`diseases or serious diseases for which no good therapies exist generally need
`only one trial demonstrating drug effectiveness. Fisher (13) notes that in these
`circumstances, perhaps a = .a25 for a one-taiied test is appropriate. if the
`drug has been approved for other similar indications or is in a class of drugs
`for which the biological understanding is firm, then Fisher suggests using a
`
`ATI 1018-0013
`
`
`
`References
`
`Rcjddn BE, Tsianco M r . bolognese 1.4, Kerstcn MK: Ci~nicai de~ielupmeni. in: Sio-
`i7hrtrmnce~~tical Stii?isiic.~,for Dncg Deveioptnenl (Peace K E , Ed). Marczl Cckker. N w
`York, 1988.
`Fisher AC, Wa!ienstei~? S: Crossover designs in medical research, In: Stntistics in the
`Pharmaceutical Industry (Buncher CK. Tsay JV, Eds). Marcel Dekker, New York,
`. ,-%%-p
`,
`I Y n l .
`Hedayat AS, Jacroux M, Majurndar D: Optimai designs for comparing test treatments
`with controls, Statist Sci, 3:462-491, 1988.
`Peace KE, Koch GG: Statistical methods for a three-period crossover design in which
`high dose cannot be used first. J Eiophurm Stntisr 3: 103- 1 16, 1993.
`Stablein DM, Novak JW, Peace KE, Laska EM, Meisner MJ: Optimization in clinical
`tria!r: and combination drug development. In: Strrlisriral issurs in Drug i2esearch and
`Development (Peace K E , Ed.). Marcei Dekker, Xew 'ioi,k, 1886.
`Fiiil: Reports, Thc pi& shee::. 53(i8): 14- i5, ?./lay 6 , 1991
`FDC Reports, The pink sheets, 53(42):9-10, Oct 21, 1991.
`F!X Reports, The p ~ n k \hcers. 5315j:T&Gi5-7&Gi6, r e b 4, i991.
`Dwnetr 3CW: .4 rnuitipic comparison procedure for comparing several treatments with
`a conirol. Anz Srutist Assoc J 50: lij96- i i 2 i . 1955.
`Sana~hanar, LP, Peck CC: Thc randomized concentratien-controiied trial An evaiuaiion
`of its sampie site efficieiicj;. Controlled C!in T,rials !2:780-794,
`!991
`Peace KE: One-sided or two-sided p values: Which most appropriately address the ques-
`tion of drug efficacy'? J Biophurm Stccrist 1 : 133-138, 1991.
`
`ATI 1018-0014
`
`
`
`i .-. 3
`
`-L.L,
`
`3 -
`
`
`
`P
`
`.
`
`3
`
`
`
`<
`
`.- v ,
`
`SCi: St;.me thoughts 39 the oIle-sided and [.,:o-:;ded
`h i h e y
`1:13.3-;50, 1.991,
`~ r r cf 21?I-si&rj y=tz iii 2TzC yjn!c: 3.3 29.4 ,\dx;iqE7... Cczr,i?:ez
`r l ~ c e r
`rnrmhrr'j p e ~ c c y t i i ' e . J P.ludza.r:x St~?l.i.f 1 : 15 I - 156. 1 9'3 !
`. . - -. - - -
`,
`14. Overall JE: A comment concerning one-sided tests of significance in new dnig appli-
`<c;iricm:: J Biq>Jicir,vi S r ~ r i s ~ 1 . 157.. 160~ lg9 1
`Sfitfir: i : 61 - -e.
`----- - - 2 -
`'(.,< li r&-3 ^.-- ..:,?-.'
`;.-.- -:'-A
`-
`15
`---4
`i .- -: i_iljr--.iurii -1zi. iiiii-zii~i-u ir>t> a~:u -J.:liui-, 2 f j ~ < ~ r ~ h c l r - t ~
`. - . . . .. - . .
`!%I.
`
`L
`
`,,,,,.
`
`+ , , c - r c
`
`i B;~pi;ha;-ir; Sti~iiji
`
`7
`
`ATI 1018-0015
`
`