A number of significant things happened this past year. The most interesting, in my humble opinion. was the release (escape?) of the FDA Guidance, “Process Validation: General Principles and Practice” (January 2011). I had heard quite a bit about it from my friend, Ali Afnan, who worked mightily on the Guidance while at FDA. Briefly, it outlines what is needed for a “proper” production of a pharmaceutical product. Many manufacturers have been saying, “Oh sure, we would love to do PAT/QbD, but how?”
Guess the “whistling past the graveyard” is over now. The essence of the Guidance is simply stated (the italics are mine):
“A successful validation program depends upon information and knowledge from product and process development. This knowledge and understanding is the basis for establishing an approach to control of the manufacturing process that results in products with the desired quality attributes. Manufacturers should:
- Understand the sources of variation
- Detect the presence and degree of variation
- Understand the impact of variation on the process and ultimately on product attributes
- Control the variation in a manner commensurate with the risk it represents to the process and product
Each manufacturer should judge whether it has gained sufficient understanding to provide a high degree of assurance in its manufacturing process to justify commercial distribution of the product. Focusing exclusively on qualification efforts without also understanding the manufacturing process and associated variations may not lead to adequate assurance of quality. After establishing and confirming the process, manufacturers must maintain the process in a state of control over the life of the process, even as materials, equipment, production environment, personnel, and manufacturing procedures change.”
Currently, companies without a working PAT (QbD) program are relying on the 1960s approach to product control, namely, testing a 10 or 20 sample set and either selling or discarding the product.
After-the-fact product testing, as has been stated repeatedly by knowledgeable workers in the industry, is ineffective in managing the process. Consequently, little knowledge is gained and no control is exercised. Indeed, the pittance of a sample taken is not really adequate to even make the decision on whether a process is “good or bad.” Numerous people have suggested more samples (cGMP does call for “a statistically significant number,” after all), but the question of how many is “significant” has been up in the air. However, that may have changed.
Recently, I was made aware of a new(ish) ASTM standard, E2790-10, for calculating the proper sample sizes of lots of materials. While I was waiting for a greater mind (Howard Mark) to examine the standard, I considered an older, simpler approach: the √n for a batch of 106 would be 103 and be quite expensive to perform under the current paradigm of “all HPLC, all the time.”
My friend and super-statistician, Howard Mark, explained the application to me and I will attempt to paraphrase:
Three key parameters are needed: the fraction of samples OOS, the fraction of OOS tablets captured, and the confidence level. One million tablets is a good approximation of infinity, allowing assumptions about Normality and other properties (as the number approaches infinity, tolerance and confidence limits become the same). Confidence intervals for the normal distribution for large numbers of degrees of freedom allows an approximation to the much more complicated formulas otherwise needed. It shows that 99.99% of the readings are within 3.7 standard deviations and 99.999% of the samples are within 4.2 standard deviations. Take that number as 0.1%.
That means, in a batch of one million tablets, 1,000 of them would be OOS. To a first approximation, one tablet per thousand would be at or beyond the 0.001 probability point of the Normal distribution, corresponding to 3.15 standard deviations. That corresponds to (0.001 x 1,000,000) = 1,000 tablets.
Measuring 1,000 tablets allows a chance to capture one of the OOS tabs. Since the actual number of tablets that would be beyond the 3.15 std. dev. value are distributed according to a Poisson distribution, measuring only 1,000 tablets gives only a 50% chance that one of the OOS tabs will be found. To increase the probability of including at least one OOS tablet among those measured, more than 1,000 samples need be measured. That means a true statistician would either laugh hysterically or have an embolism if told we currently consider ten or twenty samples as representative of a 1,000,000-5,000,000-unit batch!
Unfortunately, business as usual would dictate using HPLC, as is done now. However, even a mere 1,000 assays would be expensive and time-consuming. Changing from the status quo to meaningful (non-destructive, information-rich) testing will satisfy both the FDA and ASTM. Therefore, the good news is that we now have blueprints for QbD; the bad news is that manufacturers have run out of excuses for not running QbD.