Robert C. Wojcik Setting Dissolution Specifications

1 Setting Dissolution Specifications Introduction Setting a s pecification for di ssolution testing is an important part o f the development of a new pharmaceutical. All drug products a rc required to remain wit hin specificatio ns registered with the regulatory agencies of th e cOllntries in whic h these products arc sold, and they must remain so throughout their she lf life. The use of stability studies to predict potency over time is a well-establis hed approach for Setting th e shelf life of a product, but a high pe rce ntage of product recalls a lso in volve failure to re main within limits for di ssolution. Usin g a ratio nal, data- driven approach to Setting the Dissolution specificatio n can reduce uncertainties about product quality and s he lf life.

2 A di ssolution specificatio n ca n be set usin g tole rance lim its applied to data from batc hes of drug product, provided these batches are drawn from clinical or smbility studies used to support the new drug application. We show that statistical toleraJlce and confidence intervals can be used to set control limits for unit and poo led sa mple di ssolution testing o n ba tc h data. Through simulation, we can also generate unit and pooled sample acceptance probability curves to directly set the control limits by ins pection. Guidelines For immediate re lease produc ts, a di ssolution method s hould confo rm to o ne of t he several methods curre ntly specified for the dissolutio n requirement in USP 23.

3 Certa in guidelines sho uld be followed in preparing the method a nd Setting the specification ( I). The specification "s ho uld be stated in terms of t he minimum quantity of drug substance di ssolved within a standa rd t ime interval;" typical Specifications sho uld nmge from 70% to 85% ~lt Dissolution times between 30 and 60 minutes; Specifications in excess of 85% a re inappropriate sin ce allowance must be made for assay a nd conten t uni fo rmity of til e formulatio n . T 'imes earlier than 30 minutes t~lll below typical disintegration times and s hould only be chosen on the ba sis of a Dirso/lllioIl7ecim%gies/AUGUST 1997 Robert C. Wojcik ''''Yetb-AyentI Ledede Ulbomt01'ies Prese1lfed at tbe AA PS Dissolution Sbort Course ;n Seattle, t-flf/sbil1gtOI1 011 October 27, 1996 particular need ; times later man 60 minutes imply that the formulation is no lo nger of an immediate re lease nature.

4 The s pecifica ti o n should be practical to apply a nd efficient in terms of laboratory resource reqldre me ntsj it s hould be an indicator of the quality o f product with regard to its release prope rties. T hree types o f Specifications a re distinguished when a drug article is tested: expiry, re lease, and contro l limit (2). T h e d issolution specification generated by our methodo logy is analogous to the control limit, and should fall inside the release specification . If it fell o utside the release specificatio n , th e re would be no clear distinction between results that a re merely unusua l, and th ose t hat are clearly unsatisfactory. "I n such cases, a manufacnlrer [would] have to reject, re test, and rework batches simply because o f the inhe rent variability of an in-control process (3).

5 " I n what follows, we present a set of simulated di ssolutio n data, generate control limits fo r this data, and illustrate approaches for selecting a s pecification. Demonstration Dissolution Data '*' Dissolv ed """ "''''' ''''' ''''' ''''' ""+------~-~--+ Om." 10",," lO",," 30",," 40mln Samln &0 min Dissolution Tune Figure J. DemollstTfltioll Dissolution Dntll. A Inximll7JI ObSfl1 Jed Ifl1it l'esld, profile [NltI.:tji 1 IIil1;'",01l OhSf11 Jef/lIl1it result profile IAlill]; and average IIlIit 1"eSldt profile for ,be entin' populatioll {!Hcrlll QUI/llfiry Dissolved} (IS II fUllction of Dissolution time. Figure I s hows Dissolution data derived from t he simulated observation of 250 unit results (fro m in d ividual di ssolution vessels) at ssolution t imes of 10, 20, 30, 4 5 and 60 minutes.)

6 A profile o f th e m ea n quantity re leased fo r th e entire po pulatio n o f unit results with error bars representing the standard deviation of these units is sho wn. Also s ho wn are the maximum a nd mjllimull1 o bserved unit resuJts fo r dissolutio n at each o f dlC time points in th e profile. ~ ~ 150 :c .-~ ~ 100 2~ .. min min Dissolution Time min % Dissolved min SO min Figrn-e 2. Demomt1"lltioll Dissolution Dlltn bnving (f Non-Normal Distribution. Pigllre 2 sh ows t h e ac tual o bse rved di stributio ns of th ese unit results at each time point whe re a slllall numbe r o f slow releasin g unirs can be seen be lo w th e main body of the distribut io ns. Altho ug h th ese units have little effect o n the m ea n o r standard deviation , they strong ly affect the unit sa mple sta ge o ne 10,0 min min % Dissolved min Dissolution Time min 50 min Figure 3.)

7 De711onsn-atioll Dissolution Data baving a Nomlill Disn illlltioll. acceptance pro ba bilities. Figflre 3 shows th e ex pected distributio ns of unit results if they were no rmally distributed with m ean and standard d eviation exactl y the sam c as the obse r ved distributio ns sho wn in Figure 2. Data simulated fro m th ese two typ es o f di stributio ns afC re ferred to as th e observed and norma l data sets. Tn gene ra l, the distribution of o bserved unjt result values is no n- no rmal and skewed towards t he lower range at early dissolutio n t ime points, but approaches no rmality at late r time points. Unit Sample Specifications A to le rance limit on unit results is lIsed to derive th e stage o ne controllimh fo r unit sample di ssolutio n tes ting: "A statistica l tol erance limit furnishes a limit between, abo ve , o r be lo w which we confide ntly expect to find a prescribed propo rtio n o f th e individual item s in a po pulatio n [of o bse rved da ta]:'(3) Associa ted wit h any tole rance liln it is an expressio n of our level o f confiden ce in its accuracy.

8 T he confidence level represents ho w often our m eth o d fo r selecting th e to lerance limit (and the re fo re our controllil11_it) will be accurate. A unit result is de fined as th e re lease value obtained whe n testing a drug article in a sin gle di ssolutio n vessel. The acceptance probability fo r a w1it result is the prescribed pro portio n o f unit results that lie abo ve or at th e stated specifica ti o n. Acceptance probabilities are also de fined fo r th e stages o f th e di ssolu t io n test : A Disso/lltio7lTecim%gies/ AUGUST 1997 Setting Dissolution cont stage one accepta nce probability is the prescribed proportion of di ssolution test results for a product that satisfy the stated USP stage o ne acceptance criteria, llnd this stage o ne acceptance pro bability can be expressed in terms of the unit acceptance probability fo r stage one (Equation I).

9 Similarly, acceptan ce probabilities can be determined for stages two and three using Monte Carlo simulatio n . To set a contro l limit for this data usin g a toleran ce limit approach, we o nl y consid e r stage one, where six w1its a re tested and each unit must not be less than five p e rcent above the specification (Q). The probability of stage one acceptance (P) is the product of the six uilit acceptance probabilities (p) since each unit result is an in depe ndent observation (4): P = p6 ( I ) To set a specification that will res ult in the acceptance of stage one of the test 95% o f the t ime for a given clata set, the unit acceptance probability must be : p = VP = ~ = (2) ii >.

10 Is ~ ""''' .. "'" "" ",. ~~~~~---+~~~~---0""" tOm," 20m., 30""" 40min SO""" 60""" Dissolution TIme --Mean ,(,ty Olnol d --1~!r~~:lc) Telerance Lom,1 hljfll7'e 4. Parametric C07lfTollimi/ for Nonllalf)1 Dim'ibll/ed Data. Average lllli/1'esull profile j01' /be ellth'e poplllfltion [Mean Qllantity Dissolved} and mntTollimit profile based 011 fI parametric tolerance inter val r(Q) Parnmetrif 70!ernncc Limit}. Pigm e 4 and Table 1 (parametric column) show the control limits obtained usin g the e quatio n for each time point in the Dissolution profile. Setting a specification above th ese limjts would res ult in a n unacceptable rate of batch rejection due to th e inherent variability in the and testing of the product.]