Statistical Considerations for Defining Cut Points and Titers in Anti-Drug Antibody (ADA) Assays Ken Goldberg, Non-Clinical Statistics Johnson & Johnson Pharmaceutical Research & Development, LLC, Chesterbrook, PA Midwest Biopharmaceutical Statistics Workshop Muncie IN, May 24-26, 2010 Outline • Introduction – Why are ADA and IR assays important? • Two case studies 1. RIA: How to define %binding? 2. ECL: How to define titer cut point? 3. Both use a Huber 3-parameter nonlinear logistic regression Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 2 Immune Response (IR) Assay • • • • Primary question: ADA, Yes or No? Every biologic must be evaluated. Safety and Efficacy concerns. Too much IR can kill a compound. Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 3 Biological Drug Products are Different than Traditional Small Molecule Drugs • Made by cells not chemists • Complicated manufacturing process • Small & simple vs large & complex chemical structures Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 4 Reference: Genentech, Inc. http://www.gene.com/gene/about/views/followon-biologics.html Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 5 Adverse Clinical Sequelae • Hypersensitivity & autoimmunity • Altered PK – Drug neutralization – Abnormal biodistribution – Enhanced clearance rate Regulatory bodies require ADA evaluation for all biologics Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 6 Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 7 Immune Response (IR) Assay Challenges • Cut Point for confidence that screening bioassay response (eg, ECL, OD, RLU, CPM) reflects immunogenicity • Statistical issues of variance components, distributions, outliers, … Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 8 Screening Cut Point Flags 5% of Naïve Samples as False Positive • Use Mean + 1.645 x SD with caution – Only for normally independently distributed data without outliers – Usually requires at least a transformation like logs • Nonparametric often easier – Simply use 95th percentile – Caution if unbalanced design Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 9 ELISA Activity Positive Negative Patient Control Control A 1.689 0.153 0.055 Patient B 0.412 Patient Assay C Control 1.999 0.123 Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 10 ELISA Cut Point Example Histogram of -1/OD^.75 Normal Distribution Overlaid -1.61 25 Mean StDev N Frequency 20 15 10 5 0 -24.5 -21.0 -17.5 -14.0 -10.5 -1/OD^.75 -7.0 -3.5 0.0 Mean and Standard Deviation based on mixed effects analysis of 117 non-outliers. Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 11 -3.872 1.381 118 Analysis of an RIA Cut Point Assay Validation Experiment • • • • • • 6 Assay controls 2 Analysts with 3 assays each 2 Populations (Normal and Diabetes) 75 Naïve Human Serum samples Nonnormal data Unequal variances Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 12 RIA Histogram of 450 Naïve Sample Results Transformed %Binding = ln(35+%Binding). Parametric Cut Point = 10.757 ± 3.524. Transformed Cut Point = 3.823 based on adjusted mean = 3.402 and total standard deviation = 0.256. Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 13 RIA Normal Probability Plot of 450 Naïve Sample Results Transformed %Binding = ln(35+%Binding). Parametric Cut Point = 10.757 ± 3.524. Transformed Cut Point = 3.823 based on adjusted mean = 3.402 and total standard deviation = 0.256. Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 14 SAS Code proc mixed; * For Cut Point; class sample run analyst; model t35Pct0_100= / ddfm=sat; random sample; random sample / type=sp(exp)(tube) subject=analyst*run; repeated / group=analyst*run; proc mixed; * For Example Hypothesis Test; class sample run analyst; model t35Pct0_100 = Analyst Tube / ddfm=sat; random sample; random intercept tube / type=fa0(2) subject=analyst*run; repeated / group=analyst*run; Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 15 My RIA Notation MinCPM = Minimum of the 2 Sample CPMs MaxCPM= Maximum of the 2 Sample CPMs AvgCPM = Average of the 2 Sample CPMs CV = Coefficient of Variation of the 2 Sample CPMs B0 B100 B250 B1000 NSB TC = Average of all 6 “Validation sample 0 ng/mL” CPMs = Average of all 6 “Validation sample 100 ng/mL” CPMs = Average of all 6 “Validation sample 250 ng/mL” CPMs = Average of all 6 “Validation sample 1000 ng/mL” CPMs = Average of all 2-6 “NSB” (Non-Specific Binding) CPMs = Average of all 2-6 “TC” (Total Count) CPMs Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 16 Some RIA %Binding Definitions Response %CV Sample Sample Sample Sample 1 Limit N Mean SD Addend %CV1 (MinCPM-B0)/(B100-B0)*100 450 -3.490 7.968 65 13.0 420 -1.173 8.373 85 10.0 (MinCPM-NSB)/(TC-NSB)*100 450 1.249 0.841 4.4 14.9 MinCPM/NSB 450 1.321 0.218 -0.7 35.0 403 5.459 1.119 3 13.2 MinCPM-B0 450 -59.339 151.356 1000 16.1 MinCPM/sqrt(B100*B0) 450 0 15.3 (AvgCPM-B0)/(B100-B0)*100 AvgCPM/(TC-NSB)*100 25 20 0.549 1CV 0.084 of (Response + Addend) = Standard Deviation / (Mean + Addend) x 100%. Addend chosen so that CV is not related to control concentration. Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 17 How to Choose the RIA %Binding Definition? Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 18 New versus Old RIA %Binding Definitions • New: (MinCPM – B0) / (B100 – B0) – Repeat if CV > 25% and (MaxCPM – B0) / (B100 – B0) > 12.0% (the Cut Point) • Old: (AvgCPM – NSB) / TC – Repeat if CV > 20% Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 19 Attributes of Selected RIA %Binding Definitions %CV Cut LOD 0 ng/mL Response Limit Point (ng/mL) %Pos. N (MinCPM-B0)/(B100-B0) .120 23.5 0.04 450 (AvgCPM-B0)/(B100-B0) 25 .149 25.5 1.29 420 (AvgCPM-B0)/(B100-B0) 20 .153 25.0 0.10 403 (AvgCPM-NSB)/TC 20 3.380 31.7 0.112 403 Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 20 RIA Validation Control Curve with Lower 1-sided 95% Prediction Limit 65 + %Binding = A+B·ConcentrationC Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 21 A Logistic Curve with an Infinite Plateau is Linear wrt X C + R XH / ( MH + XH) = Substitute α = C, = H, and R/β = MH α + R X / (R/β + X) = Multiply second term by β/β α + β R X / ( R + βX) Apply L’Hopital’s rule Lim[ α + R β X / (R + β X) ] = α + β X (R) Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 22 RIA Naïve Sample %Binding vs Test Tube Order by Population Scatterplot of MinPct0_100 vs Tubepair 40 Population Diabetes Normal 30 MinPct0_100 20 10 0 -10 -20 -30 0 20 40 60 80 100 Tubepair 120 140 Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 23 160 180 RIA Naïve Sample %Binding vs Test Tube Order by Analyst Scatterplot of MinPct0_100 vs Tubepair 40 A naly st 1 2 30 MinPct0_100 20 12.05 10 0 -10 -20 -30 0 20 40 60 80 100 Tubepair 120 140 160 Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 24 180 RIA Naïve Sample %Binding vs Test Tube Order by Analyst and Run Scatterplot of ln(35+Pct0_100)*100 vs Tubepair 50 1, 1 100 150 1, 2 1, 3 400 ln(35+Pct0_100)*100 385.1 300 2, 1 2, 2 200 2, 3 400 385.1 300 200 50 100 150 Tubepair Panel variables: Analyst, Run Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 25 50 100 150 RIA Naïve Sample Means vs Test Tube Order by Population, Analyst and Run Scatterplot of ln(MeanPct+35)*100 vs Tubepair 50 1, 1 100 150 1, 2 1, 3 450 ln(MeanPct+35)*100 400 350 300 250 2, 1 450 2, 2 2, 3 400 350 300 250 50 100 150 50 100 Tubepair Panel variables: Analyst, Run Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 26 150 Population Diabetes Normal RIA Naïve Sample Mean %Binding vs CV by Analyst and Run Scatterplot of ln(MeanPct+35)*100 vs lnCV -4 1, 1 0 4 1, 2 1, 3 450 ln(MeanPct+35)*100 400 350 300 250 2, 1 450 2, 2 2, 3 400 350 300 250 -4 0 4 lnCV -4 Panel variables: Analyst, Run Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 27 0 4 RIA Naïve Sample Minimum %Binding vs CV by Analyst and Run Scatterplot of ln(35+Pct0_100)*100 vs lnCV -4 1, 1 0 4 1, 2 1, 3 400 ln(35+Pct0_100)*100 385.1 300 2, 1 2, 2 200 2, 3 400 385.1 300 200 -4 0 4 lnCV -4 Panel variables: Analyst, Run Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 28 0 4 RIA Naïve Sample CPM CV vs Mean by Analyst Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 29 RIA Naïve Sample CPM CV vs Mean by Population and Control Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 30 ProbabilityPlots Plotofofln(35+%Binding)•100 ln(35+Pct0_100)*100 RIA Probability by Analyst 385.1 99.9 Analyst 1 2 99 Mean StDev N AD P 344.0 24.90 225 2.052 <0.005 339.9 24.84 225 3.863 <0.005 95 Percent 90 80 70 60 50 40 30 20 10 5 1 0.1 250 275 300 325 350 375 ln(35+Pct0_100)*100 400 425 Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 31 Probability Plot of ln(35+Pct0_100)*100 RIA Probability Plots of ln(35+%Binding)•100 by Population Normal 99.9 Population Diabetes Normal 99 Mean StDev N AD P 341.8 19.87 150 0.542 0.162 342.0 27.14 300 1.573 <0.005 95 Percent 90 80 70 60 50 40 30 20 10 5 1 0.1 250 300 350 400 ln(35+Pct0_100)*100 450 Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 32 Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 33 Electrochemiluminescence (ECL) BioVeris Assay • New way to determine screening cut point (Data = naïve samples) • New way to determine titer cut point (not equal to screening cut point) (Data = positive samples’ Titration series) • Estimator of Titer within-assay CV Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 34 Screening Cut Point Determination ECL of Naïve Sample vs Diluent Alone with Cutoffs by Diluent ECL Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 35 Titer Definition • Smallest distinct dilution in a titration series with a negative response – Response is Sample ECL mean / Diluent Control ECL mean in this case study Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 36 Plot where Sample/Diluent Control ECL Ratio < 4 for 1 Selected Plate out of 24 Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 37 Potential Problems with a Common Screening and Titer Cut Point • Highly diluted samples tend to be positive! – The opposite would not be a problem • Titration curve too flat at cut point – Makes the titer highly variable – Common Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 38 Titer Cut Point Defined • The continuous titer inverse predicted from it has CV ≤ 30.0% with 95% confidence – 30.0% makes best case CV = worst case CV in ideal assay – Continuous titer is exact dilution giving cut point (only as a theoretical concept) Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 39 Asymptotic CV • CV Standard deviation of natural log ratio or titer • CV of dilution@ratio CV of ratio / slope of titration curve@ratio • CV of dilution decreases as ratio and slope increase • These CVs are within-plate CVs Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 40 Four Theoretical Titer Distributions CV = 34.7% = ln(F)/2 30% CV of Continuous Titer => Discrete Titer CV = 37.5% 50 Percent Percent 50% at X and 50% at X*F. CV=ln(F)/2 50 50 50 25 0 2 49 25 0 4 49 Discrete Titer 1 1 1 2 4 8 Discrete Titer CV = 34.7% = ln(F)/2 30% CV of Continuous Titer 75% at X, 12.5% each at X/F and X*F => Discrete Titer CV = 34.7% 75 50 25 0 12.5 2 12.5 4 Discrete Titer 75.16 75 8 Percent Percent 75 50 25 0 0.03 1 12.39 2 12.39 4 8 Discrete Titer Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 41 0.03 16 Titer Cut Point Defined • A continuous (interpolated) titer inverse predicted from it has CV<30.0% with 95% confidence – Exact dilution giving cut point (eg, 1.357 ratio) is the continuous titer – Continuous titer used here only as a theoretical concept – Our cut-point 5 SD above diluent mean so false-positives of noncensored titers unlikely Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 42 Summary • All biologics need ADA evaluation • Use controls to adjust for plate-toplate variance and minimize the LOD • Define titer cut point so best case CV = worst case CV in ideal assay Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 43 Acknowledgements: • Sheng Dai • Allen Schantz • Pam Cawood • Gopi Shankar • Bill Pikounis Reference: Shankar, G. et al, (2008). Recommendations for the validation of immunoassays used for detection of host antibodies against biotechnology products. Journal of Pharmaceutical and Biomedical Analysis. 48:1267–1281. Statistical Considerations for Defining Cut Points and Titers in ADA Assays. Ken Goldberg. Midwest Biopharmaceutical Statistics Workshop, May 24-26, 2010. Slide 44