(M1430-09-62) Strategies for Data Treatment of RTD Curves for Pharmaceutical Applications

Purpose: Continuous manufacturing (CM) is becoming increasingly popular in the pharmaceutical industry due to its benefits, leading to research on developing process dynamics for accurate prediction and implementation. One such technique is residence time distribution (RTD), which has been used to study the dynamics of unit operations and entire manufacturing lines following a change in processing conditions. To obtain an RTD profile of a unit operation in pharmaceutical manufacturing lines, a pulse or step change of the selected tracer is added at the inlet of the operation, and the tracer concentration at the outlet is measured as a function of time using process analytical technology (PAT) sensors. This measured tracer concentration provides insights into the system dynamics and enables predictive modeling of the unit operation. The tracer concentration profile obtained through RTD measurements can be affected by measurement variability and external noise from different sources, including environmental and process conditions. Measurement variability is related to the system's internal fluctuations, which can be attributed to the sensor, equipment, and material variability. To accurately develop predictive models for evaluating system dynamics, it is essential to measure, quantify, and remove both the external noise and measurement variability from the experimentally obtained RTD profiles.

Methods: Our group has recently looked into this problem and explored various data pre-treatment and noise handling strategies aimed towards identifying effective strategies for denoising RTD profiles. Moreover, proper truncation of RTD curves was found to be necessary for removing data with very low tracer concentrations (i.e., tail region), which resulted in no loss of accuracy in the determination of out-of-specification products. In this work, our objective was to present a comprehensive, sequential methodology for processing RTD data.

Results: A systematic approach was developed after thoroughly examining various techniques for denoising and truncating the RTD data. A denoising method was performed using wavelet transform, along with a truncation strategy based on the sensitivity of the PAT instrument. To demonstrate the applicability of this framework on actual data, we conducted a case study on a continuous direct compaction line. The formulation consisted of four ingredients: API, filler, disintegrant, and lubricant. A 17-point DoE was performed to study the impact of process parameters on the RTD. To conduct the RTD pulse input experiments, we selected a tracer with physical properties similar to the API, which was also detectable by the PAT sensors.

Conclusion: The study found that frequency averaging methods, such as wavelet transform and Fourier transform, retain signals better than time averaging methods, as indicated by higher values of performance metrics. Denoising methods did not affect the prediction of out-of-specification product in response to perturbations. The PAT-based truncation approach was recommended as it considered system variability and the impact of truncation on funnel plot and RTD model parameters, enabling informed decision-making. In summary, the study provided an easily understandable guide for handling RTD data, allowing manufacturing scientists to effortlessly apply the methodology. This improves the accuracy of RTD model parameter predictions without compromising the accuracy of determining exclusion intervals.

References: Bhalode, Pooja, Sonia M. Razavi, Andrés Roman-Ospino, James Scicolone, Gerardo Callegari, Geng Tian, Abdollah Koolivand, Scott Krull, Marianthi G. Ierapetritou, and Fernando J. Muzzio. "Optimal quantification of residence time distribution profiles from a quality assurance perspective." International Journal of Pharmaceutics 634 (2023): 122653.

Acknowledgements: The authors would like to acknowledge financial assistance from the U.S. Food and Drug Administration through BAA contract 75F40120C00125.

Methodology for data treatment of RTD curves

Raw data vs. pre-treated data