The process was driving everybody nuts. One batch would be beautiful - high yield excellent purity, no problems. Then the next batch would be awful - half the yield of the batch before it and lower quality. The operators swore that they prepared every batch exactly the way the manufacturing batch records required. And since each batch was worth about $300,000, process failures were costing a lot of money and keeping the plant manager and his supervisors up at night. 

Each time a batch failed, they'd review the executed manufacturing records to try to understand what was "different." They knew that something must have changed, but they couldn't find it. Sometimes, they thought that they had figured it out and they'd tighten up a process control or retrain the operators, but still the maddening variability would return.

So they learned to live with it. And spend the money. And plan for batch failures.

But they didn't have to. 

With unexplained process variability - or "out of control" processes - it can be impossible to discover the underlying causes by just comparing batch records with each other or even looking at trending data and process control charts. 

Why?

Because process variability can be caused by complex interactions between multiple process parameters that don't individually impact the process. For example, I found that process variability in one active pharmaceutical ingredient (API) manufacturing process was caused by a combination of temperature, residual water, and concentration during a certain process step. The effect was not seen if any one or two of the parameters were in the ranges that caused the variability - only if all three were in those ranges together!

So how do you go about finding these relationships?

You can maximize your chances of discovering the root causes of unwanted process variability and find out how to get it under control by following these six steps. These steps integrate quantitative statistical analysis and data mining with qualitative process expertise and systematic evaluation to identify non-obvious and subtle cause-and-effect relationships that drive variability. 

1. Identify the product or process variability that needs to be addressed.
2. Create a database of historical process parameters from a variety of sources such as batch records, raw material test results, and incident reports.
3. Create a comprehensive "Robustness Assessment Report" that contains in-depth analysis of each process step and critical process parameters using available documentation and discussions with knowledgeable personnel.
4. Perform a statistical analysis of the database using data-mining software such as JMP to look for drivers of variation and interactions relative to the process variability.
5. Generate hypotheses with process experts around identified drivers of variation to determine if they are truly causal or merely correlated.
6. Develop a predictive model and identify optimal or close to optimal operating parameters to reduce process variability. 

At the end of this type of process robustness study, you will usually discover three major categories of findings:

1.     Actions that can be taken now: These are process changes that can be made within the current regulatory filing to improve the process now. Often these changes can be made within the current manufacturing batch record parameters or they may require a process re-validation.

2.     Parameters that require further study: These are usually gaps in the manufacturing batch record and analytical data that could be useful in both controlling the process and providing additional insight into process variability.

3.     Future actions: These are process changes that may be outside of the regulatory filings or items such as recommendations for new equipment or procedures, laboratory studies, and Process Analytical Technology (PAT) opportunities.

This systematic blending of quantitative statistical analysis and qualitative empirical scientific analysis is a powerful approach to identifying drivers of variation and improving or optimizing process performance, so it is vital to have both quantitative statistical and qualitative process expertise on your process robustness teams. The statisticians can both search for and identify sources of process variation, and the process experts can help to guide the search as well as help determine if a statistical correlation is truly causal.

Applying a process robustness program to problematic API or drug product processes is one of the most cost-effective and fast ways to dramatically improve pharmaceutical process performance, yield, and quality. Ideally, you'd create a process robustness team that would use project portfolio management to prioritize each process based on value to your company and then systematically evaluate each product process using the steps described above.