Can it be measured? Forty-five minutes is enough to find out.Talk to us about your processLanguageENFR

Predicting tablet dissolution without dissolving the tablet

“Can we know a batch’s dissolution before the lab tests are finished? What in my formulation or my process makes it vary? Can we check far more units than the six or twelve of the test?”

NIR (near infrared) answers all three. On a whole tablet, the spectrum carries both the chemistry (disintegrant, lubricant, water) and the physics (density and porosity, which follow compression force). A model built against your dissolution method predicts the percentage dissolved at each sampling time, and so the whole profile. The application is established on industrial tablets.

The measured tablet stays intact. It can then go through the dissolution test, which compares prediction and measurement on the same unit.

Dissolution profiles and NIR prediction at 120 minutes Dissolution profiles of gastro-resistant tablets at seven coating levels, from 75 to 250 %, over 120 minutes at pH 1.2. For each level, the NIR prediction at 120 minutes overlaps the test value. % dissolved coating minutes, pH 1.2 medium 0 20 40 60 80 100 0 20 40 60 80 100 120 75 % 100 % 125 % 150 % 175 % 200 and 250 %
Worked example: gastro-resistant tablets, seven coating levels. Curves: dissolution test. Diamonds: NIR prediction at 120 minutes on intact tablets, mean of three units; 1.3-point error in cross-validation.

What the model sees, and what it misses

It sees whatever leaves a signature in the illuminated zone: lubricant content and distribution, whose blending time changes wettability; compression force, through the spectral baseline; particle size; coating thickness. In the worked example above, a coating polymer band near 1672 nm carries most of the information.

First derivative of the mean NIR spectra, seven coating levels First derivative of the mean NIR spectra from 1100 to 2100 nm for the seven coating levels. Around 1672 nm the trough deepens as the coating gets thicker. wavelength, nm 1100 1200 1300 1400 1500 1600 1700 1800 1900 2000 2100 1672 nm
The signal that carries the information. First derivative of the mean spectra, from the thinnest coating (light) to the thickest (dark): the trough near 1672 nm, a polymer band, deepens with thickness.

It does not see a mechanism with no optical signature where it measures. In reflectance, the core of a thick tablet remains barely visible, hence the frequent choice of transmission.

What calibration calls for
  • Tablets that vary. Routine batches look too much alike. A design of experiments varies compression force, lubrication, particle size or coating level within the design space.
  • The whole range, bounds included. In the worked example, a first model built without the most heavily coated tablets predicted −24 % dissolved for them. Once those tablets were included in the calibration, the prediction came back to 0 %, like the test.
  • One model per sampling time. Predicting the percentage dissolved directly at each time holds up better than predicting the parameters of a fitted curve. The worked example confirms it, as does the literature.
  • A reference analysis effort. Every calibration unit goes through the dissolution test. It is the heaviest part of the project, and it is costed at scoping.
  • The dispersion of the reference method. It bounds the accuracy the model can reach.
  • A comparison of profiles. Beyond the error at each time point, the f2 similarity factor between predicted and measured profiles.
Prediction at 120 minutes depending on the calibration range For each coating level, the test value and the predictions of two models. The model calibrated without the 200 and 250 % levels predicts −24 % for the 250 % level, where the test gives 0 %. The model calibrated on the full range predicts 0 %. % dissolved at 120 min coating level -20 0 20 40 60 80 100 75 % 100 % 125 % 150 % 175 % 200 % 250 % dissolution test model calibrated without 200 and 250 % model calibrated on the full range
A model predicts well only within the range it has learned. Calibrated without the thickest coatings, it predicts −24 % dissolved for them, an impossible value. Calibrated on the full range, it matches the test. Means of three tablets.

All the way to release

Replacing the release dissolution test falls under real-time release testing as defined in ICH Q8(R2). The method is validated for that use, and the dissolution test remains the reference and the fallback when the measurement is unavailable. What this calls for.

When a second measurement brings what is missing

When dissolution depends mainly on internal structure, TERAHERTZ measures porosity and coating thickness, and complements NIR within one model. Combining measurements.

Further reading
  • Kuny T. et al., Non-destructive dissolution testing correlation, Dissolution Technologies, 2003. Transmission NIR on immediate-release tablets.
  • Baranwal Y. et al., Prediction of dissolution profiles by non-destructive NIR spectroscopy in bilayer tablets, International Journal of Pharmaceutics 565, 2019. Profiles compared with f1 and f2.
  • Galata D. L. et al., Fast, spectroscopy-based prediction of in vitro dissolution profile of extended release tablets using artificial neural networks, Pharmaceutics 11, 2019. NIR and RAMAN, extended release.

Which product keeps you waiting for its dissolution?

Forty-five minutes is enough to tell whether an NIR model is within reach, and what the calibration campaign would call for.

Talk to us about your process