Recently, I uttered words no one ever thought I would say: “Just using Near-Infrared (NIR) may not be enough.” I was attending an IFPAC session and the speaker was showing how his company used several NIR instruments to monitor/control an extruder for continuously manufacturing a “ribbon” of polymer/API that was chopped, lubricated, and pressed into tablets. The NIR instruments measured the excipients and API as they blended, melted, mixed, and extruded.
The one point that was ignored was where a vacuum pump pulled off volatiles. From experience with this type of extruded polymer/API mix, I suggested another option. Depending on the dwell time at that stage, the temperature, and level of vacuum, a number of things happen. The desired result is that unwanted volatiles, often monomers from the polymer used, are removed, making the product safer. Another effect is that the porosity of the granulation changes. Usually, the higher the vacuum and temperature, more gases are exhausted, causing the granulation to have smaller and fewer pores.
This can affect the dissolution rate. Since most polymer extrusions are used for continuous/sustained/controlled release tablets, the porosity could be a CQA (Critical Quality Attribute) for a number of products. One control approach might be a vision device. (There are a number of good units on the market.) This type device can correlate the release properties of the product with visual appearance and parameters can be set.
Another application of NIR (mentioned quite often at the meeting) was for the release of tablets. I have seen transmission NIR used on tablets removed from the process line and reflection NIR used on-line; either give more than sufficient results and, as importantly, allow the producer to have many more sample points to aid in process control.
One measurement that has become quite popular is to use NIR (or Raman) and an MVA (multi-variate analysis) such as partial least squares (PLS) to either predict the release at a specific time or the time at which a specific percentage is released. Usually, such an approach is quite fine. The resultant equation uses a few factors and can be tested quite easily. There are situations where the correlation between the physical parameters (as seen in spectral differences, mostly peak shifts) needs a rather large number of factors to produce satisfactory statistics. I have seen 13 PLS factors needed to generate an equation.
Keep in mind that companies seldom do follow-up or on-going IVIVC (in-vivo in-vitro correlation) studies to show that dissolution is more than an indicator of “something” happening to the product over time. This tenuous blood level correlation becomes even more tenuous by further correlating dissolution to a NIR spectrum, using a 13- or 14-factor equation!
If a process is to be controlled in anything near real-time, the test used needs to be fairly rapid. Obviously, performing a classic dissolution test on a time-delayed would take far too long to be of use to the production staff. There is a possible solution, however: a number of instrument manufacturers’ dissolution test instruments that are capable of rapid, nearly continuous readings (through fiber optics). This potentially generates a massive number of data points that can be used to build a predictive equation.
How can this be accomplished? Perform “classic” dissolutions on a number of (passing) production lots, using this equipment. The number will depend on the reproducibility of the product; the better the lot-to-lot agreement, the fewer lots will be needed. The analyses should be run at the maximum data acquisition rate, generating a digitalized release curve. Gather the data and arrange it, depending on the software you plan on using. (A novice with statistics can ask his company statistics people for recommendations.)
Basically, you fit a polynomial equation to the actual data (possibly ignoring the initiation period of a minute or two) and generate an equation. Using this equation, perform the first 10-15 minutes of the dissolution of a different set of tablets and, using the equation, predict the percent released at the required time points. If the predicted numbers are not statistically “good” enough (depending on QA requirements), you either need more data or a stronger equation.
When you are able to generate a (statistically) satisfactory equation, then you will have a dissolution prediction test that will generate numbers fast enough to be considered a process control tool. Yes, it is based on math, just like the NIR method, but the math is based on actual dissolution data, not spectra. That makes me feel more comfortable about using it for a control tool.
