Helping the World Be Well through Innovative Manufacturing Process Analytical Technology (PAT) for Blend Uniformity Challenges and Lessons Learned Jim Donato1, K. Ganeshwar Prasad1, Fernando Rodriguez-Ares2, Shaira Davila1, Brandye Smith-Goettler3, Manoharan Ramasamy3, Peter Brush3 1 Pharmaceutical Commercialization Technology, MSD Intl GmbH (PR Branch), Las Piedras, PR 2 Integrated Project Team, MSD Intl GmbH (PR Branch), Las Piedras, PR 3 ACDS-PAT, Merck, West Point Helping the World Be Well through Innovative Manufacturing

Drug Product Manufacture API Assay by PAT Blend Uniformity by PAT Blend Uniformity by PAT

Why Blend The uniformity of the blend is critical in the establishment of the uniformity of the dosage units Powder blending of active/excipient is an important and common unit operation for solid oral dosage systems Direct compression, roller compaction, wet granulation, hot melt extrusion For most formulations, it is important to ensure the uniformity of the excipients as well as the API Inhomogeneity may lead to: Poor flow, sticking/picking/capping/delamination, improper disintegration and/or dissolution, sub- and/or super- potent tablets

Disadvantages of Traditional Blend Sampling Conventional procedure: Blend Uniformity samples are collected using a powder thief and analyzed using HPLC, UV-Vis, LIBS or other off-line laboratory technique Labor intensive for both the operator and lab personnel (IBC’s moved to different areas to sample/PPE, typically 4 hrs) Difficult to reproduce sampling technique (and thus blend result) Blender must be stopped to sample blend Long turn-around times for sample analysis Typically not performed once validated in production Potential for exposure of operators to high potency API’s Act of sampling, in itself, may produce sample non-uniformity If anyone in the room has ever had to thief sample an 1800 L bin – they know this is hard work The results - It is formulation – product – blend dependent

NIR as Alternative BU Method Real-time monitoring Monitor API, excipients and lubricant Non-invasive Minimizes inter-operator variability and sample thief failures Collects blend sample at each rotation thus providing process understanding Used for IBC (bin) blenders from 5L (dev) to 1800L+ commercial scale Reduces sample analysis burden on laboratory Improved safety profile as product exposure eliminated

NIR BU Method Considerations-Instrumentation # of Sample/Reference scans 4/128 Effective sample size 200-800mg* Analyzer/bin interface Sapphire window lid with clamp Spectral resolution 8cm-1 Spectral coverage range 5550-7400cm-1 Data acquisition trigger Internal at 150-160° Spectral collection time 1.5s Instrumentation Performance attributes Functional attributes Data collection parameters Spectroscopy API, placebo excipient Got specificity? Sampling interface Non-contact Sapphire window Tri-clamp API Placebo Raw Absorbance spectra

NIR BU Method Considerations – Blend Metrics Qualitative Algorithm Mathematical transformation to connect spectral variability to blend endpoint – moving block average of spectral variation – either in terms of SD or %RSD With specificity Peak Maxima/Minima Max or Min in range Peak area, peak average, root mean square Without specificity Average spectrum Standard deviation spectrum PCA other Connect Spectral Variability to Blend Endpoint

Blend Control Strategy - PAT end-point control via spectral %RSD Risk Mitigated Variations in the lubrication process impact tablet hardness/tensile strength and in turn disintegration and dissolution due to over-lubrication of the blend The blend NIRS dataset compiled with data from: Two manufacturing sites Four scales (150L, 300L, 600L and 800L) Two different spectrometers Fill volumes (27 – 85%)

Scale Up Challenge Scale up: 1400L IBC Different Pivot Point Axis of rotation Pre-blending Blending Scale up: 1400L IBC Different Pivot Point Axis of Rotation Shifted up Bi-directional rotation API blend batches as part of Process Validation All batches pass BU specifications Batch #6 stands out as atypical

NIR Model and Batch Variation As per thief sampling and NIR protocol, all batches complied with BU specifications Without NIR, and using only standard BU protocol, no further action needed With NIR, blend endpoint for Batch #6 was ~2x revolutions compared to others Without the NIR blend profile, process variability is not visible Initiated investigation into why blend is different: Possible causes include: raw material/API variability (particle size, density, fill volume), equipment failures (speed, directional change, blend lid), NIR instrument/sampling/algorithm concerns

Conclusions NIR for online monitoring of BU provides several advantages over powder thief sampling Multiple components can be monitored simultaneously API, lubricant, critical excipient Full blend profiles are obtained rather than single point of data Slug thief sampling is preferred method of sampling for validation of NIR method NIR was used to identify several possible processing issues Sample thief problems and BU failures Powder blend/lubricant sticking during large scale blending Overall endpoint variability in commercial scale blending operations Identified an Out-of-Trend blend profile during process validation

Acknowledgements West Point Las Piedras Las Piedras SLT Mike Gentzler Steve Conway Andre Hermans Jessica Miller Colleen Neu Brandon Ricart Las Piedras Glenda Rodriguez Luz Rodriguez Ana Diaz …and others Las Piedras SLT Patrick Breen Humberto Diaz Luis O. Garcia Andrew Wirths