(249ab) Monte Carlo Simulation-Based Method for Process Performance Assessment in Sterile Manufacturing of Biopharmaceuticals

The fast expansion of biopharmaceuticals such as monoclonal antibodies is calling for more extensive research in the manufacturing processes of sterile drug products, such as injectables. In this work, we present a Monte Carlo simulation-based method, which enables the assessment of process performance under operational uncertainty in sterile drug product manufacturing. The study is supported by an industrial case study.

Our method enables the systematic performance assessment of processes in commercial operation with a probability-based approach. It also permits the multi-facet assessment of the operations, such as design of the process, human factor and predictive maintenance, in order to ensure high industrial applicability. The method, which consists of five main steps, supports use of Monte Carlo-based mathematical models, and enables its application in an industrial environment. The five steps are: define key performance indicators (KPIs), collect and adapt data, create process model, assess operational properties and identify sensitive task. In the first step KPIs are defined accordingly to the industrial needs and interests, e.g., increase of capacity, yield or robustness of the process. Subsequently, data related to the KPIs are collected from data management systems and adapted to the purpose, resulting in a hierarchical data set. In the third step the process model is created, which covers all tasks inside the process, and considers the KPIs as well as uncertainties such as failures, repetition or human factors. Next, the operational properties are assessed by taking into account historical information on production stops, duration of tasks, remedy times and correlations. During this step, the creation of fitted process durations and empirical event probabilities lays the basis for the Monte Carlo simulation. Stochastic simulation enables reflection of operational uncertainty, which is a prominent element on level of manufacturing shop-floor, into the simulation results. Finally, significant tasks, which would require improvement, are identified by the performance of sensitivity analysis. Because of the non-deterministic nature of the models, variance-based sensitivity analyses approaches, such as Sobolâ??s indices and partial rank correlation coefficients (PRCC), are applied to rank tasks by their importance. Furthermore, for enhancing the industrial applicability, the combination of the KPIs with feasibility indicators, such as implementation time, is considered as an additional feature. This approach allows incorporation of good manufacturing practice (GMP) in the assessment, an omnipresent and critical factor in the pharmaceutical industry.

The method was applied in an industrial case study, and more specifically, on key processes of the sterile manufacturing of biopharmaceuticals termed clean-in-place (CIP) and sterilize-in-place (SIP). These processes have been rarely the research object because of their supportive functions, but are known as the bottleneck of manufacturing sterile drug products in the industry. In the plant which was the object of the case study, the CIP and SIP processes occupy nearly half of the production timeline. In this case study we applied the above mentioned method in view of a future process optimization of the entire CIP and SIP procedure. Our goal was to identify the tasks, which have the highest impact on the production time, the KPI determined in the case study. Generally CIP and SIP are semi-automated processes, consisting of around 200 tasks with different durations and dealing with fluid dynamic laws, which render their complete modelling unrealistic. Also, the operations environment leads to the presence of failures and time distribution due to technical or human issues. There a quasi-Monte Carlo method was applied to simulate the duration of an entire CIP and SIP procedure, which was supported by Latin hypercube sampling (LHS) of task durations generated by fitted non-parametric distributions. The pattern, describing failures and repetitions of process tasks, was randomly generated at each iteration on the base of empirical probabilities. For the validation of process model, Kolmogorov-Smirnov test was applied assuming the identical distributions between actual and simulated production time as the null hypothesis. The identification of the important tasks, which have the highest influence on the CIP/SIP duration, was performed by PRCC analysis. To enhance the applicability, the result is combined with feasibility evaluation regarding various aspects, e.g., GMP, time duration or resource availability, which is provided by the industrial experts. For the first twenty most important tasks, Sobolâ??s indices were calculated to quantify the taskâ??s contribution to the total variance of CIP and SIP durations.

Our method permits the evaluation of the current process with an intention for redesign, as was demonstrated in the case study. Moreover, it enables considerations on the quality of the operations as well as the effects of failures. In practice, this last point represents a failure-mode-effect analysis with a quantitative criticality assessment, and finds great application in a GMP environment. Since the method was developed in an industrial environment, it shows high compatibility with the process improvement methodologies currently in use, i.e., Lean Six Sigma (LSS). In fact LSS procedures can be used for the interpretation of the simulation results, e.g., Pareto chart and root cause analysis (RCA). In future studies we intend to develop a model based on machine learning to be coupled with RCA, which will enable considerations on predictive maintenance and predictive planning.