INTERNSHIP EXPERIENCE IN ANALYZING STABILITY DATA Speaker: Charlene Pu ASA UP-STATConference - 4/6/2013 Source: FDA http://www.fda.gov/BiologicsBloodVaccines/GuidanceComplianceRegulatoryInformation/PostMarketActivities/LotReleases/ucm310644.htm A/California (H1N1) A/Victoria (H3N2) B/Wisconsin Potency VACCINE SHELF LIFE Potency of vaccine degrade over time Registration Limit Time, months FLUZONE POTENCY STABILITY PROJECT Reduce Potency Variability of Fluzone Product Partial Least Squares technique (PLS) Take care of multicollinearity issue Latent Factor (component) -optimally chosen to obtain high R-square and low RMSE Variation in the predicted variable + Variation in the response variable Tutorial Video– Using PLS in JMP http://www.youtube.com/watch?v=TtSYTXk_0y4 Potency DEGRADATION PLOT Determine concentration here To Predict the Potency on Expiry here Formulation Target Registration Limit Expiry Manufacturing Time Point Time, months Potency Stability Data Lot 1 Lot 2 Lot 3 Lot 4 Lot 5 Lot 6 Overall Slope 10% Potency Change, % 0% -10% -20% -30% -40% log Time, days 1 -50% -60% 0 100 200 300 Time , days 400 500 600 REGRESSION THROUGH THE ORIGIN y 1 x LOWER CONFIDENCE INTERVAL ON E(Y|X ) 0 ˆ y| x t / 2,df x0 se ˆ1 2 0 SAS CODES proc glm data=asa_example; model _CHANGE=TIME/NOINT; run; log days 1 Potency % Cˆ hange 12.0871 time OVERALL SLOPE Confidence Level: 95% Registration Limit: 27 ug HA/mL Manufacturing Time Point to Expiration Date: 365 days Average Slope: -12.0871 Standard Error of Average Slope, se(slope): 0.5809 Mean Square Error of Regression Model: 73.1774 Degree of Freedom: 47 Formulation Target= LRL /1 Slope log days 1 t / 2,df log days 12 seSlope 100% 27 / 1 12.0871 log( 365 1) t0.025, 47 log( 365 1) 2 0.5809 100% 27 / 1 33.5% 100% 40.6 Potency Stability Data Lot 1 Lot 2 Lot 3 Lot 4 Lot 5 Lot 6 Overall Slope Lot 3 10% Potency Change, % 0% -10% -20% -30% -40% -50% -60% 0 100 200 300 Time , days 400 500 600 SAS CODES SLOPE FOR WORST LOT data ex; set asa_example; l1=0; l2=0; l3=0; l4=0; l5=0; l6=0; if lot='Lot 1' then l1=1; if lot='Lot 2' then l2=1; if lot='Lot 3' then l3=1; if lot='Lot 4' then l4=1; if lot='Lot 5' then l5=1; if lot='Lot 6' then l6=1; time1=Time*l1; time2=Time*l2; time3=Time*l3; time4=Time*l4; time5=Time*l5; time6=Time*l6; proc reg data=ex; model _CHANGE = time1-time6 / noint; run; quit; SEPARATE SLOPE Confidence Level: 95% Registration Limit: 27 ug HA/mL Manufacturing Time Point to Expiration Date: 365 days Slope for the Worst Case: -15.8185 SE of Slope for the Worst Case: 1.0693 Mean Square Error of Regression Model: 41.8014 Degree of Freedom: 42 Formulation Target = LRL /1 Slope log days 1 t / 2,df log days 12 seSlope 100% 27 / 1 15.8185 log( 365 1) t0.025, 42 log( 365 1) 2 1.0693 100% 27 / 1 45.2% 100% 49.2 CONCLUSION The least squares regression analysis could be employed to model potency decay. Partial Least Squares analysis could be performed to quantify manufacturing factors that affect the vaccine stability CONTACT INFORMATION Hsuen- Li (Charlene) Pu Email: pucharlene1224@gmail.com Phone: (917)847-2323 LinkedIn: www.linkedin.com/pub/hsuen-li-charlenepu/42/760/aa7/ SAS CODES SLOPE FOR WORST LOT proc glm data=asa_example; class LOT; model _CHANGE=TIME TIME*LOT/noint solution; run; quit; +