Raw Material Properties
In order to design an ultralow dose of melatonin (< 0.5 mg), ODMT formulation with statistical tools, selection of the excipient should be initially carried out. In the preformulation study (31), it was indicated that Granfiller-DTM 215 (GFD) would be the best candidate for the final formulation. In contrast, GalenIQTM 721, which was finally rejected, showed many interesting properties like excellent flowability and beneficial palatability aspects. Both excipients present diverse morphological properties (especially in solidity and circularity parameters). For this reason, it was decided to use both main fillers to evaluate statistical impact on product critical quality attributes (CQAs). Granfiller-DTM 215 is a typical co-processed excipient, a combination of D-mannitol, microcrystalline cellulose, carmellose, and crospovidone. All three components support D-mannitol in acceleration of tablet disintegration process (carmellose works as a wicking agent, crospovidone works as a swelling agent, whereas cellulose creates an insoluble matrix body). GalenIQTM 721 (isomalt) is a combination of 6-O-α-D-glucopyranosyl-D-sorbitol (1,6-GPS) and 1-O-α-D-glucopyranosyl-D-mannitol dihydrate (1,1-GPM). Although it is not a co-processed excipient, according to declaration of manufacturer, GIQ is designed for ODT formulations. It should be noticed that its mechanism of action presents slow dissolution; therefore, an application of isomalt requires an additional disintegrating agent. Both reflect two different approaches to the formulation development. Thus, in this work, main fillers were used in combination with micronized and non-micronized melatonin to conduct a series of formulation and processing experiments with DoE methodology. Since particle size was indicated as the critical material attribute (CMA), the PSD examination was performed with laser diffraction method. The data for each component is presented in Table V. The results show that non-micronized melatonin is a heterogeneous material (the value of the span parameter is the highest in this group) and the median of particle size d0.5 is approx. 30 microns. In opposite, the micronized grade of API is much more uniform and the particle size is below 6 microns. In the case of carriers, the median of particle size is similar; however, the span value of GIQ is lower than GFD. The structure of all materials was also investigated with a scanning electron microscope. The SEM images are depicted in Figure 1. As can be seen, the structure of GIQ particles is spherical, with smooth and regular surface. In contrast, GFD particles have irregular, rather elongated shape with higher surface area.
Table V Particle Size Distribution and span Values of Melatonin and Carriers Determined with Laser Diffraction Method, Mean ± Standard Deviation, n=6 It may be also assumed that such a difference in morphology will have an impact on powder flowability and ability to create homogenous mixtures with active ingredient. Therefore, flow properties were also checked (Table VI).
Table VI Comparison of Main Fillers’ Flowability Properties The results proved that the flowability of GIQ is much better than GFD. The Carr Index (CI) and Hausner ratio (HR) indicators (according to Ph.Eur. 2.9.36) also confirm such observation. Both classify GIQ as a good flowing powder (CI: 11–15%, HR: 1.12–1.18), whereas GFD as fair flowing powder (CI: 16–20%, HR: 1.19–1.25). It should be noticed that the difference in flowability parameter (s/100g) between GIQ and GFD is more than twofold, which means in practice significant contrast.
Plackett-Burman Design
The Plackett-Burman screening design was applied to identify the process parameters and raw material attributes that may cause the greatest product CQAs change. This should enable better understanding of process variability and ensure higher level of its repeatability and predictability in the future. The values of input and obtained output variables are presented in Table VII.
Table VII The Independent Variables and Responses Summary Obtained from Compression Experiments Using 5-Tip Tools, Performed According to Placket-Burman Design Only two variables among the investigated ones may have a potential impact on the powder mixture homogeneity, namely the melatonin and main carrier type. Other variables of the tableting process should be treated as dim variables in this particular case. Thus, the obtained effect values should be treated as background noise, having no effect. It is demonstrated that the blend uniformity depends mainly on the melatonin type (its d0.5 value) that is used to prepare the powder mixture (effect 6.31; p-value <0.05) (Figure 2A, Supplementary material). The positive value of the effect indicates that with the switch of the variable from a lower to a higher level, and in practice, when changing the substance from micronized to non-micronized type, the value of the dependent variable which is blend uniformity expressed as RSD will increase. Increasing the value of the latter means in practice a decrease in homogeneity of the powder mixture. In other words, application of micronized API improves the homogeneity of the powder blend. Thus, micronized melatonin was finally selected for further investigation.
In the case of dissolution at 15 min. as response variable, the main carrier type (effect −54.27; p-value <0.05) and melatonin type (effect −6.18; p-value <0.05) played the major role (Figure 2C, Supplementary material). The use of GIQ as the main filler and non-micronized melatonin resulted in lower release at 15 min. from the dosage form (Table VII, Figure 5, Supplementary material). The use of GFD resulted in significantly faster release at 15 min. in comparison to GIQ regardless of API crystal size. The lowest result for GFD was 87%, whereas the highest result for GIQ was 42.4%. This strong impact is correlated with the different morphological structure of the main fillers and their chemical composition, i.e., the combination of mannitol, carmellose, and crospovidone in the co-processed GFD. It was no surprise that micronized API additionally supports faster dissolution of melatonin from the minitablets, since the smaller crystals size (d0.5 = 2.11 μm) results in expanded surface area, and finally enables faster dissolving. Uniformity of dosage units (expressed by AV value) was mainly affected by melatonin type (effect 10.90; p-value <0.05) (Figure 2G, Supplementary material). The use of micronized API resulted in a lower AV value. It might be explained by its ability to create more homogenous and stable mixtures: thanks to the adhesion to the porous surface of the filler (carrier) particles. The smaller API crystal size, the stronger the bonds between both components. Moreover, this finding confirms that the content uniformity of minitablets is sensitive to API particle size. Especially for low-potency minitablets, micronized APIs are beneficial, as inclusion or exclusion of a single large particle during die filling may have high impact on API absolute content and its uniformity (4). Such an interesting effect might be also explained by the morphology of the GIQ particles, which are characterized by a very smooth and regular surface (having lower surface energy). Thus, it is less prone to sticking of the small API particles on its surface. This is another factor in favor of using a micronized active in further development.
In the case of tablet weight spread and friability variables, no influence of the examined factors was found (Figure 2B, 2F, Supplementary material). On the basis of initial experiments and Plackett-Burman screening test results, a formulation prototype was defined. As the main filler, the GFD was selected and the micronized melatonin as the active ingredient.
Full Factorial Design
In the next step, the influence of pre-selected process variables on product quality attributes was investigated. A three level, full factorial design was employed to establish the impact of two critical process parameters (CPPs), i.e., the main compression force (kN) and press table speed (rpm) on the values of dependent variables. The input values and response results are presented in Table VIII.
Table VIII The Independent Variables and Response Summary Obtained by Applying Full Factorial Design The developed mathematical models that show the influence of CPPs on resistance to crushing parameter are presented below. The former was developed on the basis of data obtained with measurements by compendial method (Eq.1), dominant in the pharmaceutical industry. The latter was created based on data obtained with texture analysis (Eq.2).
$$Resistance\ to\ crushing=-0.014\ast {A}^2+0.558\ast A+0.804\kern4em \left(R^2_{\text{adj}}=0.94\right)$$
(1)
A—main compression force (kN); resistance to crushing is expressed in newtons (N)
$$Resistance\ to\ crushing\ TXT=-0.096\ast {A}^2+2.497\ast A-3.885\kern1.75em \left(R^2_{\text{adj}}=0.90\right)$$
(2)
A—main compression force (kN); resistance to crushing TXT is expressed in newtons (N)
The main factor that influences the tablets’ resistance to crushing is the value of main compression force. The effect and relationship are non-linear(Figure 2, orange line). Not surprisingly, the minitablet crushing strength increases with increasing compression force due to stronger binding of the material (1, 20). Interestingly, at higher compression force values, the response reaches a plateau, which suggests achieving maximum degree of plastic deformation and binding sites by the filler without fragmentation or brittleness which would compromise the ODMT’s mechanical stability. To the authors’ knowledge, such phenomenon has not been described for minitablet case studies, where usually linear increase in tensile strength was observed with increasing compression pressure (1, 18,19,20). This is likely explained by the differences in the used filler, as the cited works do not report the use of co-processed excipient GFD in this context.
Although both methods used to determine resistance to crushing indicate different values, they are correlated with each other (R=0.97). Both compendial method and texture analysis made it possible to develop similar mathematical models. However, the method of resistance to crushing measurement using texture analysis seems to be more recommended for the analysis of ODMTs. The reason is the low hardness values of the analyzed ODMTs when compared to conventional or non-orodispersible minitablets and higher accuracy of the method. While difficulties in applying traditional hardness testers for ODMTs are recognized in the literature and texture analysis has been employed to determine crushing strength in several studies (1, 19, 22, 27), to the authors’ knowledge up to date, no report has compared or modeled the results from the application of two testing methods. As demonstrated, both are correlated with each other and can be used for process characterization of ODMT compression, although texture analysis can be considered superior.
The model developed for disintegration time based on texture analysis is given by the equation (Eq. 3):
$$Disintegration\ time\ TXT=1.206\ast A-0.484\kern1.5em \left(R^2_\text{adj}=0.97\right)$$
(3)
A—main compression force (kN); disintegration time TXT is expressed in seconds (s).
The main factor responsible for the tablets’ disintegration time is the value of compression force used during the compaction process. With an increase in compression force, it was observed that the disintegration time also increased in a linear manner, which is obviously expected (1, 22, 27). Considering this together with resistance to crushing values, it must be stated that despite decreasing ODMT porosity as evident from longer disintegration, the mechanical properties of minitablets developed with GFD as the filler/binder are not affected. Unfortunately, for the data obtained by the compendial method, it was not possible to identify the factors influencing the disintegration time in a statistically significant manner. Consequently, we were unable to create a model. For tablet weight SD parameter, it was not found that any of the analyzed factors had a statistically significant influence on this CQA, which confirms uniform, repeatable filling of 2-mm dies with blend containing micronized API based on GFD excipient. Thus, it was not possible to develop a model. In the case of resistance to crushing SD parameter, no model was achieved that could explain the CQA variability to a satisfactory extent (R2adj >0.5).
The developed models for resistance to crushing and disintegration are characterized by high R2adj, which indicates that over 90% of variability in the response value can be accounted for by the factor incorporated in the model, i.e., the effects of main compression force. Moreover, such models are expected to yield reliable predictions of the output values. To verify this, batches of 250,000 minitablets were compressed at 13.6 and 13.2 kN and the experimentally determined values of their resistance to crushing and disintegration time were compared with theoretical values calculated with model equations. Prediction error (%) was determined as the ratio of difference between the mean observed and predicted value to the observed one.
As can be seen in Table IX, the prediction error values are reasonable. The best agreement between theoretical values and observed ones occurred for the resistance to crushing results determined with texture analysis, which further confirms the superiority of this method. For the values measured with typical tablet hardness tester, the results are underestimated to a higher extent, which might be explained by high RSD values. High variability in minitablet tensile strength has been observed especially with decreasing die sizes. However, the presented comparison here strongly suggests that in the case of our tablets, high RSD of resistance to crushing determined with compendial method is due to unsuitable instrumental accuracy and precision, and not due to processing faults.
Table IX Verification of Predictions Developed with Full Factorial Design Models Fractional Factorial Design
A 3(3-1) fractional factorial design was employed to establish the influence of three critical process parameters (CPPs): main compression force, tableting speed, and amount of multi-tip punches on the values of dependent variables. The input values and response results are presented in Table X.
Table X The Independent Variables and Responses Summary Obtained from Fractional Factorial Design The model developed for resistance to crushing parameter based on texture analysis is given by the equation (Eq. 4):
$$Resistance\ to\ crushing\ TXT=0.655\ast A-0.190\ast B+6.824\kern2.75em \left(R^2_\text{adj}=0.73\right)$$
(4)
A—main compression force (kN);B—table speed (rpm); resistance to crushing TXT is expressed in newtons (N).
The factors that influence the tablets’ resistance to crushing TXT are main compression force and tableting speed. As the compression force increases, a linear increase in the hardness of the tablets is observed. This is indicated by the positive value of the effect (5.24; p-value <0.05). Increasing the table speed causes a linear drop in the resistance to crushing of the obtained tablets. It is assigned to the negative sign of the effect (−1.90; p-value <0.05). This phenomenon may be caused by shorter dwell time with the increasing turret rotation. The shorter time the powder spent under pressure inside the die, the fewer permanent bonds between the compressed powder particles were produced. The absolute value of the tableting speed effect is smaller if we compare it with the strength of the compression force effect. In the publicly available literature on minitablets, the influence of turret speed on tablet tensile strength has not been explored extensively. In the study by Goh et al., this factor was not identified as statistically significant, which was tentatively attributed to the particular formulation’s insensitivity to narrowing compression profile, increased strain rate, and shorter dwell times with increasing turret speed (26).
The R2adj value of the achieved model is lower than in case of the first one built based on data from full factorial design. It means that less variability in tablet resistance to crushing is explained by this model. The difference most likely is related to different datasets being studied and reduced number of points, where relationships are not explored over the whole experimental space and effects cannot be estimated as reliably as with full factorial design. Nevertheless, similar dependence of resistance to crushing on compression force was detected with the same effect sign, and the influence of tableting speed can be considered as of minor importance. Moreover, the results of fractional factorial design experiments confirmed that the number of punches applied in the rotary tablet press was not a significant factor for any of the responses. However, no comparison with literature findings can be made; as to the authors’ knowledge, no study on minitablets has considered this as a variable for investigation.
For dependent variables, resistance to crushing, resistance to crushing SD, disintegration time, weight variation, and friability, no models were developed due to lack of statistically significant impact of independent variables on quality attributes or because simple achieved models did not explain the variability of CQAs to a satisfactory extent (R2adj > 0.5).
Design Space
The design space (DS) was created in order to define the combination of input variables, the use of which in the production process guarantees obtaining a product with the desired quality characteristics. The two CQAs, namely resistance to crushing TXT and disintegration time TXT, were chosen to establish the design space. The DS was graphically represented by means of 2D graphs based on the developed equations. The following constraints were taken into consideration:
-
The average value of resistance to crushing TXT should be more than 9.5 N. From our experience, it is not possible to achieve tablets harder than 14 N. Above this border plateau is observed. In order to set a limit that would ensure that all tablets would meet the requirements, taking into account their natural variance, it was assumed that the standard deviation for this parameter is 0.5 N. Thus, the value of resistance to crushing TXT of single ODMT should be not less than 8.0 N. This constraint is based on the fact that values below 8.0 N were related to unacceptably high friability (> 0.5%).
-
The average disintegration time TXT should be less than 30 s, typically for ODMTs.
Within tested ranges of CPPs, average resistance to crushing TXT exceeds the lower range only if tableting is performed using main compression force lower than 7.55 kN. The requirement that disintegration time TXT should be less than 30 s is fulfilled in the whole range of settings. Thus, by overlapping the requirements for both CPPs, the design space was established. It shows the ODMTs of required quality can be manufactured by using the main compression force higher than 7.55 kN (green area on Figure 3). In conclusion, the created design space enabled the compression process optimization and served to indicate the Proven Acceptable Range of compression force. It will find application during the routine production. Additionally, another DS was alternatively simulated based on the models developed with the use of fractional factorial design (Figure 4), where acceptable lower limit of compression force is marginally dependent on tableting speed. This illustrates the difference in the datasets and statistical treatment used for building model equations according to various matrix designs and the inclusion or exclusion of independent variables of borderline significance, i.e., table speed. Nevertheless, the CPPs ranges established based on the assumptions of both full and fractional factorial design models are similar and mostly overlapping.