The Choice of Sampling Frequency and Product Acceptance Criteria to Assure Content Uniformity for Continuous Manufacturing Processes

The Choice of Sampling Frequency and Product Acceptance Criteria to Assure Content Uniformity for Continuous Manufacturing Processes

Authors Tim Kramer Sal Garcia Jeff Hofer Xiaoyu Zhang Ian Leavesley Wyatt Roth Leo Manley Ahmad Almaya 15Jan2016 Kramer et al, IFPAC 2016 2

Outline Lilly continuous manufacturing Measure of closeness for content uniformity Implications for acceptance criteria at the feed frame Sampling frequency at the feed frame Sampling frequency and acceptance criteria of core tablets measured by NIR Summary 15Jan2016 Kramer et al, IFPAC 2016 3

Continuous Drug Product Manufacturing Process at Lilly Frequency: continuous Loss-in-weight feeders Frequency: continuous Rapid real-time measure of API content in blend Residence-time distribution (RTD) model Feed frame Frequency: 2~3x per ~3-5 min NIR tablet assay 15Jan2016 Kramer et al, IFPAC 2016 4

Predicting Concentrations Output from loss-in-weight feeders combined with residence time distribution model provides concentration estimates at feed frame May be used to reject tablets having extreme concentrations NIR scans at feed frame provide alternative estimates of concentration May also be used to reject tablets having extreme concentrations 15Jan2016 Kramer et al, IFPAC 2016 5

Assuring Content Uniformity Want majority of individual dosage units to be close to target Use 85% to115% label claim as interval that represents closeness Internal sampling and batch acceptance criteria relate to these intervals Use ASTM 2810 with 50% confidence that 80% of samples will pass USP<905> for routine batches Use ASTM 2810 with 50% confidence that 95% of samples will pass USP<905> for process validation batches 15Jan2016 Kramer et al, IFPAC 2016 6

Example OC Curves when True Mean, µ=100 100 90 80 Percentage of Time Pass Criteria 70 60 50 40 30 20 10 0 100% 99% 98% 97% 96% 95% 94% 93% 92% 91% 90% Percentage of Individual Dosage Units Within 85-115% LC 15Jan2016 Kramer et al, IFPAC 2016

Assuring Dose Uniformity Criteria % of Individual Dosage Units Between 85% and 115% Label Claim (with 95% Confidence) ASTM 2810 50% Confidence of 95% Passing (Process 98% Validation Criteria) ASTM 2810 50% Confidence of 80% Passing (Routine 94% Release Criteria) USP<905> 86% 15Jan2016 Kramer et al, IFPAC 2016 8

Assuring Content Uniformity Acceptance criteria are guided by the desire to ensure product is between 85% and 115% label claim With the understanding That the process is targeting 100% label claim There are process and measurement variabilities That there are potential biases between NIR and HPLC measurements 15Jan2016 Kramer et al, IFPAC 2016 9

Potential Feed Frame Criteria Having some dosage units outside 85% to 115% does not imply that you will fail content uniformity However, having vast majority within 85% to 115% does imply that you will pass content uniformity requirements A proposed option is to set feed frame criteria to ensure that underlying concentration is between 85% and 115% 15Jan2016 Kramer et al, IFPAC 2016 10

Hypothetical Feed Frame Observations 110 True Concentration and NIR Observed (1 Spectra and Average of 5) 105 Concentration (% of Nominal) 100 95 90 0 20 40 60 80 100 120 140 160 180 Observation Blue: True Concentration Red: NIR, Individual Spectra Green: NIR, Average of 5 15Jan2016 Kramer et al, IFPAC 2016 11

How to Assure True Concentration is > 85% Label Claim Assuming NIR independence, no measurement bias and a constant true concentration: 85% + 3 σσ NNNNNN / nn σσ NNNNNN Number of Signals Averaged 1 3 5 7 1 88.0 86.7 86.3 86.1 2 91.0 88.5 87.7 87.3 3 94.0 90.2 89.0 88.4 4 97.0 91.9 90.4 89.5 5 100.0 93.7 91.7 90.7 15Jan2016 Kramer et al, IFPAC 2016 12

Adding Underlying Process Variability To assure underlying concentration is 85% (assuming NIR independence, no measurement bias and an independent concentration process): 85% + 3 (σσ NNNNNN 2 + σσpppppppppppppp 2 )/ nn σσ NNNNNN Number of NIR Signals Averaged 3 5 7 Standard Deviation of Underlying Concentration (% Label Claim) 0 1 2 0 1 2 0 1 2 1 86.7 87.4 88.9 86.3 86.9 88.0 86.1 86.6 87.5 2 88.5 88.9 89.9 87.7 88.0 88.8 87.3 87.5 88.2 3 90.2 90.5 91.2 89.0 89.2 89.8 88.4 88.6 89.1 4 91.9 92.1 92.7 90.4 90.5 91.0 89.5 89.7 90.1 5 93.7 93.8 94.3 91.7 91.8 92.2 90.7 90.8 91.1 15Jan2016 Kramer et al, IFPAC 2016 13

Adding potential bias (NIR relative to HPLC) To assure underlying concentration is 85% (assuming NIR independence, systematic bias and an independent concentration process): 85% + bbbbbbbb + 3 (σσ NNNNNN 2 + σσpppppppppppppp 2 )/ nn 15Jan2016 Kramer et al, IFPAC 2016 14

Observed NIR Averages to Assure True Concentration is 85% Label Claim With 2% assumed bias: σσ NNNNNN Number of NIR Signals Averaged 3 5 7 Standard Deviation of Underlying Concentration (% Label Claim) 0 1 2 0 1 2 0 1 2 1 88.7 89.4 90.9 88.3 88.9 90.0 88.1 88.6 89.5 2 90.5 90.9 91.9 89.7 90.0 90.8 89.3 89.5 90.2 3 92.2 92.5 93.2 91.0 91.2 91.8 90.4 90.6 91.1 4 93.9 94.1 94.7 92.4 92.5 93.0 91.5 91.7 92.1 5 95.7 95.8 96.3 93.7 93.8 94.2 92.7 92.8 93.1 Observed averages of 5 NIR signals between 93.0% and 107.0% assure underlying concentrations are between 85% and 115% for σσ NNNNNN = 4%, σσ PPPPPPPPPPPPPP = 2% and bias = 2%. 15Jan2016 Kramer et al, IFPAC 2016 15

Feed Frame Sampling Frequency Potential goal is not just to monitor the process but to react to deviations Have limited time between feed frame signal and ejection of tablets Want to reject anomalies (transients) that occur downstream of feeders (Feeder issues are handled separately not considered in this talk) 15Jan2016 Kramer et al, IFPAC 2016 16

Hypothetical Transient Seen at the Feed Frame True Concentration and NIR Observed (1 Spectra and Average of 5) 105 100 Concentration (% of Nominal) 95 93 90 85 80 75 70 Excluded from tableting 65 0 20 40 60 80 100 120 140 160 180 Observation 15Jan2016 Kramer et al, IFPAC 2016 17

Hypothetical Downstream Concentrations When Product Is Rejected at Feed Frame True Concentration and NIR Observed (1 Spectra and Average of 5) 105 100 Concentration (% of Nominal) 95 93 90 85 80 75 70 65 0 20 40 60 80 100 120 140 160 180 Observation 15Jan2016 Kramer et al, IFPAC 2016 18

What s the Worst That Could Happen? Picking limits of 100 ± x% with averages of n signals will effectively catch short-term transients of nx% from target ± 7% limits with n=5 will catch short-term transients of 35% from target For example, average of five signals {100, 100, 100, 100, 65} is 93 15Jan2016 Kramer et al, IFPAC 2016 19

Feed Frame Sampling Frequency Sample as fast as system allows Each spectra is available every 1.2 seconds (one spectra is average of 20 scans) Number used in average is limited by reaction time Want to be able to reject all material that comprises average In addition to reaction time, actual number of spectra used in average will depend on process standard deviation, NIR standard deviation, and importance of quick detection 15Jan2016 Kramer et al, IFPAC 2016 20

Core Tablet Sampling Frequency and Acceptance Criteria Purpose is to confirm that upstream processes have delivered as expected and measured Does not need to be any more stringent than for batch processing Use same sampling frequency and acceptance criteria as for HPLC-measured batches 15Jan2016 Kramer et al, IFPAC 2016 21

Acceptance Criteria for NIR Determinations (Core Tablets) For core tablets, treat NIR and HPLC determinations equivalently (once NIR method is qualified) Potentially use ASTM 2810 with 50% confidence that 80% of samples will pass USP<905> for routine batches Potentially use ASTM 2810 with 50% confidence that 95% of samples will pass USP<905> for process validation batches 15Jan2016 Kramer et al, IFPAC 2016 22

Potential Routine Release Sampling Frequency and Acceptance Criteria Tier 1 Sample 10 locations throughout the batch with 1 sample per location (10x1 plan) Test against ASTM E2810 criteria for 50% confidence of passing the USP <905> test 80% of the time using Sample Plan 1 for the 10x1 plan Tier 2 Sample 20 additional locations throughout the batch with 1 sample per location for a total of 30 locations (30x1 plan) Test against ASTM E2810 criteria for 50% confidence of passing the USP <905> test 80% of the time using Sample Plan 1 for the 30x1 plan 15Jan2016 Kramer et al, IFPAC 2016 23

Potential Process Validation Sampling Frequency and Acceptance Criteria Tier 1 Sample 20 locations throughout the batch with 2 samples per location (20x2 plan) Test against ASTM E2810 criteria for 50% confidence of passing the USP <905> test 95% of the time using Sample Plan 2 for the 20x2 plan Tier 2 Sample 20 additional locations throughout the batch with 2 samples per location for a total of 40 locations (40x2 plan) Test against ASTM E2810 criteria for 50% confidence of passing the USP <905> test 95% of the time using Sample Plan 2 for the 40x2 plan 15Jan2016 Kramer et al, IFPAC 2016 24

Summary Feed frame acceptance criteria can be guided by goal of ensuring concentrations between 85% and 115% Number of spectra used at feed frame should consider reaction time, process standard deviation, NIR standard deviation, potential bias and resulting rejection limits Acceptance criteria and sampling frequency of core tablets measured by NIR can be identical to batch processes measured by HPLC 15Jan2016 Kramer et al, IFPAC 2016 25

Thank you for your attention! Questions??? 15Jan2016 Kramer et al, IFPAC 2016 26