GRABADO

assumed that the mean particle size and shape for both the drug and the excipient is the same (cubes of height/width/depth approximately 50 pm) and that 1 mg of a blend contains approximately 8,000 particles, then for a 200 pg product strength in 12.5 mg of blend (1.6% w/w) there should be an average of 128 drug particles within 1 mg.

Assuming that the drug particle distribution follows a Poisson distribution then the theoretical relative standard deviation (RSD%) may be predicted^^^ as:

Standard deviation = >/l28 = 11.3 particles/mg

Then RSD% = (11.313/128) x 100 = 8.8%

A value of 9% for RSD% does not offer much confidence in the degree of mixing. Further still, for the lowest concentration of blend (0.4% w/w) this value would be even worse. To obtain an RSD% equivalent to that from the current HPLC-UV method it will be necessary to measure and average multiple N IR spectra. For example, 12 NIR measurements of a “1 mg” sample size should be taken to approximate the real sample size of a 12.5 mg blister, or 25 measurements to compare to the sample size taken for HPLC-UV assay.

5.4.

Conclusions

In conclusion, it was not possible to practically measure a precise sample volume measured by N IR using a static sample. However, from experimental data and

assumptions, a value was obtained of approximately 1 mg. Since the effective sample size measured by N IR is much smaller than an individual blister (of 12.5 mg) and a therapeutic dose (25 mg) then an N IR method w ill only be able to be used to measure homogeneity if multiple measurements are made and co-added to produce a mean. These conclusions are applied to Chapter 7.

experimental measurement shows good agreement with the research by Bemtsson et alP^ for their studies. Confirmation of this measurement for this specific work has raised concerns about the detrimental effect that fouling of the optical window of the probe may have for on-line measurements. In the next chapter, the position of the probe within the blender is investigated. Thus, emphasis on the potential for probe fouling w ill also be studied.

CHAPTER SIX: OPTIMISATION OF NIR PROBE POSITION

6.1.

Introduction

In this chapter the position of the NIR probe within the blender is considered. The NIR probe intended for this application was purchased to satisfy a dimensional requirement - a probe with an outside diameter of approximately 12.5 mm would fit through an existing port-hole in the blender wall which was originally engineered to accept a temperature sensor, see Figure 6.1.

Figure 6.1 Aerial view of inside of the high shear Fielder TRV 8 blender

showing position of NIR probe

4

The scientific literature contains many examples where NIR has been used on-line to monitor blending, as discussed in section 4.2.5, page 163. However, this specific application has not been demonstrated (i.e. low quantities of drug). Whilst it would be interesting to position the probe at different locations within the blender, only one port­ hole was available for this research. For future investigations, engineering of the blender to include more access holes for probes may be worthy of investigation.

Engineering costs for this would perhaps be easier to justify with encouraging on-line measurements from this research. Thus in this chapter, the depth of the probe into the blender could be varied from flush against the inside walls of the blender (0.00 mm) to a maximum depth of approximately 53 mm. At this point the probe was in danger of obstructing the top impeller.

O f particular concern in on-line measurements is the possible fouling of the optical window of the probe. The conclusions from the previous chapter only serve to highlight this. Finding the optimum probe position is therefore very important.

The overall goal of this research was to develop a N IR method to measure the

concentration of active within the powder during the blending process. In Chapter 7 an approach to developing a model is presented whereby the calibration model developed from off-line data w ill be applied to data collected on-line. To achieve this it is

desirable to have minimal spectral differences between the off-line and on-line

measurements. An important part of this chapter is therefore, to assess and understand what (if any) spectral differences existed between spectra recorded for static and

moving samples and ways to minimise any effects so as to aid transfer of the calibration model.

Many NIR spectrometer parameters w ill be pre-determined by the more demanding on­ line data acquisition (i.e. number of averaged scans per spectrum). However, the probe depth was one parameter that could be assessed to determine whether or not different positions within the blender were better or worse than others.

For these experiments lactose monohydrate only w ill again be used as justified in the previous chapter.

6.2.

Experimental details

6.2.1. Materials

See section 5.2.1, page 169.

6.2.2. NiR instrumentation, parameters and spectral data handling

6.2.2.1. NiR spectrophotometer and probe See 5.2.3.1, page 170.

6.2.2 2. NIR parameters

For on-line measurements, NIR spectra were recorded in kinetic mode (a feature of the Bomem Grams software that automatically and continuously recorded and saved spectra). For these investigations 2 scans* were recorded, averaged and then saved as one “sub-file” (or spectrum) within a single multi-file. The time between each sub-file (subsequent spectra) was 1.4303 seconds.

6.2 2.3. NIR data handling - exporting routines

See 5.2.3.3, page 170. Further data processing is discussed in the results section.