Report

Statistical Evaluation of Dissolution for Specification Setting and Stability Studies Fasheng Li Associate Director, Pharmaceutical Statistics Worldwide R&D Pfizer, Inc 37th Annual MBSW Muncie, IN May 20, 2014 Motivation Dissolution routinely tested to provide in vitro drug release information Drug development: prediction of in vivo drug release profiles Quality control: assessment of batch-to-batch consistency Decision making during dissolution method and drug development Data based specification setting for USP <711> dissolution test Dissolution monitoring on stability Statistical assessment integral to decision making process. 2 2 Outline Setting Extended Release Dissolution Specifications Number of time points needed Case 1: Two-point spec Case 2: Three-point spec Evaluation of possible specifications Dissolution on Stability No significant linear trend observed Non-linear trend observed 3 3 Dissolution Specification Setting How many time points are necessary for setting dissolution specifications? Based on “Guidance for Industry: Extended Release Oral Dosage Forms: Development, Evaluation, and Application of In Vitro/In Vivo Correlations”, at least three points (early, middle, late) on the dissolution profile should be used to have specifications Are fewer than three time points sufficient? Are three time points enough? 4 4 Dissolution Spec Setting – Case 1 Mean disso profiles of three typical batches of a sustained release drug product Specs at 30 and 180 minutes Proposed to have specs at two time points (30 and 180 minutes) Team discussed to add a spec at either 15 or 60 minutes Add either 15 or 60 minutes? 5 5 Dissolution Spec Setting – Case 1 , An empirical first-order twoparameter nonlinear regression model fit to the release profiles Goodness-of-fit of the model evaluated by a coefficient of determination R2-type measure R2 = 1 SSE SSTotal Model appropriateness evaluated by the lack-of-fit test Release = A(1-e-bt) is a two-parameter Weibull model 6 6 , Dissolution Spec Setting – Case 1 Dissolution profiles defined well by a two-parameter release model Mathematically, any two points on the profile would be able to sufficiently define the release profile No need to add a third time point for specification Two-point spec Team agreed to set disso specifications without a third point 7 7 Dissolution Spec Setting – Case 2 Mean disso profiles of three typical batches of a extended release drug product Originally specs at 5 time points proposed; should 1 more be added to improve control? Question: How many time points are needed for setting dissolution specifications? 8 8 Dissolution Spec Setting – Case 2 , An empirical three-parameter non-linear regression model (Weibull ) fit to the release profiles Goodness-of-fit of the model evaluated by a coefficient of determination R2-type measure R2 = 1 Three-point spec SSE SSTotal Model appropriateness The three-parameter Weibull model is sufficient and adequate to define dissolution profiles in this case evaluated by the Recommend three-time points for dissolution lack-of-fit test 9 specifications 9 , Brief Summary Three-point specifications are apparently more advantageous than six-point specifications: Cost Savings Save 50% reducing from 6 to 3 time points Quicker Analytical Results Conformity Risk Reduction: Assuming the probability of passing USP <711> dissolution test at each time point is the same (e.g. 98%), the overall probability to pass: 0.983 = 94% with three time points vs. 0.986 = 89% with six time points 10 10 , Evaluation of Dissolution Specifications After Determining Number of Time Points Evaluate proposed dissolution specifications against USP <711> at each time point Recommend new dissolution specifications if necessary Statistical Approach Simulations performed on individual dissolution data at each of the specification time points to check the probabilities of passing different stages (L1, L2, and L3) of USP <711> dissolution test 11 11 USP <711> Dissolution Test Acceptance Criteria for Extended Release Drug Products 12 12 Evaluation of Dissolution Specifications Controlled release product specs: @1 hour: <= 30% @4 hours: 40-60% @24 hours: >= 80% Re-evaluate specs due to method change Current three-point specifications Data: 46 unique dissolution conditions Each have 6 to 96 individual disso profiles A total of 1578 disso profiles collected 50000 simulations performed on 46 dissolution data sets to check probabilities of passing USP <711> stages (L1, L2, and L3) 13 13 Evaluation of Dissolution Specifications % of Disso Tests Acceptance Probability by Stage Lx Specifications Stage Need Stage Lx Pass 100% by Stage Lx 100.0 13.0 Q1: <= 25% L1 87.0 52.2 Q4: 35-55% L2 Q24: >= 80% L3 47.8 80.4 100.0 15.2 Q1: <= 30% L1 84.8 58.7 Q4: 35-55% L2 Q24: >= 80% L3 41.3 87.0 100.0 4.3 Q1: <= 30% L1 95.7 17.4 Q4: 40-60% L2 Q24: >= 85% L3 82.6 47.8 Mean 95.3 98.5 99.5 96.2 99.0 99.8 59.4 87.1 91.0 14 StDev 9.0 2.9 1.4 7.6 2.1 0.7 38.2 19.1 19.9 Comparable Specs Proposed Specs 14 Evaluation of Dissolution Specifications Controlled release product recommended specifications for new dissolution method Individual Dissolution Profiles @1 hour: <= 30% @4 hours: 35-55% @24 hours: >= 80% Revised three-point specifications Mean Dissolution Profiles Revised three-point specifications 15 15 , Brief Summary – Disso Spec Setting The number of time points on dissolution profiles used for specification setting Can be justified by fitting a non-linear release model Based on the number of parameters of the non-linear release model Specifications at each time points Can be evaluated by performing simulations on dissolution data against USP <711> criteria Calculate the probability to pass USP criteria 16 16 Dissolution on Stability Dissolution usually monitored on stability as a numerical quality attributes with numeric specifications e.g. %Release at 6 hours should be between 40-60% Dissolution data may not have a significant linear trend along stability time Linear trends not significant Non-linear trends observed How to evaluate dissolution data on stability? Typical Q1E shelf life analysis not appropriate. 17 17 Dissolution on Stability – No Linear Trend Clear linear trend for the chemical impurity data ICH Q1E Analysis Appropriate ICH Q1E Analysis is not meaningful No overall statistically significant trend in dissolution at 10 hours 18 18 Dissolution on Stability – No Linear Trend Shelf-life predicted based on the major chemical impurities: Apply linear regression analyses following the ICH Q1E guidance The risk of failing dissolution on stability will be quantified Make sure the risk of failing dissolution spec is low Utilize dissolution profiles tested at each of the stability time points 19 19 Dissolution on Stability – No Linear Trend A three-parameter Weibull model: %Release = A(1-exp(-b*tm)) fit to all mean or individual dissolution profiles at each of the stability time points for all registration stability batches The risk of failing dissolution at a future stability test time since time is not relevant can be quantified by 1. Constructing prediction limits with confidence level p% 2. Checking the limits against the spec of (45, 65) 3. If the prediction limits are within the spec limits, the risk of failing a future average dissolution would be no more than 100-p% 20 20 • Model fit mean dissolution profiles of stability times points very well (R2 > 0.99) • The risk of failing dissolution test on stability at a future time is no more than 0.9% Risk of Dissolution on Stability % Release A(1 - e Storage Condition 25°C/60%RH 30°C/75%RH -bt m ) Nonlinear Model Parameters and Fit Statistics 99 % Pred Limits for Dissolution at P-value 10hr 99 % Pred Limits %Chance for a Meet Future Disso Test Spec (45, 65)? to Fail Strength A b m R2 1 2 3 4 5 6 94.5 94.0 94.8 95.7 96.2 94.4 0.0176 0.0147 0.0159 0.0142 0.0140 0.0135 1.711 1.760 1.739 1.777 1.783 1.800 0.9939 0.9952 0.9960 0.9964 0.9960 0.9955 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 49.6, 63.0 47.8, 59.7 49.5, 60.6 49.5, 60.1 49.5, 60.8 48.3, 60.1 Yes Yes Yes Yes Yes Yes 0.045 0.012 0.003 0.003 0.003 0.003 7 93.6 0.0107 1.867 0.9945 0.0000 44.5, 57.4 Not the lower limit 0.865 1 2 3 4 5 6 95.8 94.4 95.7 97.5 96.6 96.1 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 50.9, 65.3 48.8, 60.9 49.7, 63.4 49.6, 62.1 49.2, 62.0 48.9, 61.3 Not the upper limit Yes Yes Yes Yes Yes 0.669 0.003 0.074 0.012 0.007 0.003 7 95.1 0.0108 1.867 0.9955 0.0000 46.2, 57.9 Yes 0.112 0.0180 0.0150 0.0165 0.0149 0.0137 0.0132 1.714 1.764 1.734 1.756 1.796 1.811 0.9932 0.9952 0.9940 0.9951 0.9950 0.9952 21 Dissolution on Stability – No Linear Trend Risk of failing disso on stability is < 0.9% 22 Dissolution on Stability – No Linear Trend Risk of failing disso on stability is < 0.7% 23 Dissolution on Stability – Non-linear Trend Extended release product: with a clear non-linear trend for dissolution data at x hours 24 24 Dissolution on Stability – Non-linear Trend Empirical model of: %Release at x hours = A(1-e-b*(t+t0)) can be fit to dissolution at x hours from manufacturing age for all registration stability batches Shelf life (in terms of manufacture age) can be predicted when the 95% confidence interval intersects with the spec limits Shelf life (in terms of stability storage age) can be determined by subtracting the manufacturing age of the initial stability time point (stability time 0 month) 25 25 Dissolution on Stability – Non-linear Trend Stability program started at 7.4 months of manufacturing age Predicted shelf life is about 54.5 -7.4 = 47.1 months 26 26 , Brief Summary – Dissolution on Stability Stability dissolution data often show no significant linear trends No significant linear or non-linear trend Dissolution profile data can be utilized to remediate the risk of meeting dissolution specifications • More information used versus evaluate at 1 time point on the profile Non-linear trend Empirical non-linear model fit to stability data could help justify the prediction of shelf life 27 27 Summary Dissolution for extended release drug products facing decision makings in areas such as Setting Specifications Number of time points on the profile for spec Specification limits at the time points Dissolution on Stability No significant linear trend Non-linear trend • Statistics will be able to contribute greatly in the above areas to make regulatory appealing decisions Statisticians need to work proactively with team scientists 28 Acknowledgment Kim Vukovinsky, Senior Director, Pharmaceutical Statistics, Worldwide R&D, Pfizer Inc. 29