**3. Employment of DoE for the development of novel MLT modified-release formulations**

In the vast majority of experimental procedures in all scientific fields, the optimal conditions are reached by modifying the levels of one factor at a time (OFAT) while keeping all the rest that seem to affect the response constant.

*Melatonin - The Hormone of Darkness and Its Therapeutic Potential and Perspectives*

problems [10].

problems [12].

ing an 8 h period [14].

sustained delivery systems [16].

their release profiles. The results indicated that both formulations (liposomal and solid matrix tablets) could be suitable alternatives for treating sleep-onset/maintenance

Calcium alginate beads were also prepared to investigate the MLT modified release. Excipients utilized in their preparation included calcium alginate, polyvinylpyrrolidone (M.W.: 10.000 and 55.000), hydroxypropyl methylcellulose (M.W.: 15.000 and 100.000), lactose monohydrate and, as a surfactant, sodium lauryl sulphate. The in vitro release of melatonin was investigated at two different pHs (acidic pH 1.2 and basic pH 6.8), and the results concluded that the hormone's release from the beads was reversibly proportional to the extent of their expansion, which depends on the molecular weight/viscosity of the biopolymers present in the beads; the higher the molecular weight/viscosity of the hydrogels, the greater the

Another group of researchers prepared a slow-release tablet of MLT with varying quantities of hydroxypropyl methylcellulose K15 M and Carbopol 971P, as well as other excipients (microcrystalline cellulose, maize starch, magnesium stearate and purified talc). The formulations developed showed a slow release of MLT dur-

To the same end, matrix tablets were formulated using hydroxypropyl methylcellulose and tested in vitro in relation to drug release, as a function of polymer viscosity, drug loading, type and amount of disintegrant, lubricant and glidant and aqueous polymeric coating level, and further compared with two commercial products. The release studies showed that as the polymer viscosity increased, the release decreased, and as the coating level increased, an increased lag time was observed [15]. Other researchers have examined the in vivo sustained release of MLT that was incorporated in solid lipid nanoparticles. The results indicated that solid lipid nanoparticles may act as a reservoir, permitting a constant and prolonged MLT release, after oral administration, which may indicate new possibilities for

In another research project, controlled-release matrix tablets of MLT were developed by the use of a computer programme, D-optimal experimental design, aiming at affecting its modified release at simulated gastrointestinal media. The careful selection of the excipients (polyvinylpyrrolidone (M.W.: 10.000 and 55.000), hydroxypropyl methylcellulose K15 M and lactose monohydrate) at their appropriate quantity resulted to the optimal solution and the controlled release of melatonin with the minimal number of experiments [17]. Moreover, in another research, polymer-reinforced and polymer-coated alginate beads with various concentrations of polymer (Eudragit® RSI00) and plasticizer (aluminum tristearate)

beads swelling and the less the MLT's release [13].

Once more, aiming at the modified release of MLT, another group of researchers studied the MLT release from monolayered and three-layered tablets, incorporating nanofibrous mats composed of cellulose acetate and polyvinylpyrrolidone. The in vitro dissolution release studies of the MLT formulations in simulated gastrointestinal fluids revealed tableting pressure and pH dependence. Comparing the MLT release from the physical mixture tablets and from the nanofibre-based tablets, it was concluded that the release profile was generally slower than the latter, rendering the formulation suitable for both sleep-onset and maintenance dysfunctions [11]. The same group of researchers produced electrospun-MLT loaded nanofibres (with cellulose acetate, polyvinylpyrrolidone and hydroxypropyl methylcellulose, as excipients) and used them to fill hard gelatin and delayed-release (DRcaps™) capsules. The in vitro dissolution results revealed a modified-release profile of MLT from the fabricated matrices in gastrointestinal-like fluids and suggested that the MLT-loaded nanofibrous mats could exhibit a promising profile for treating sleep

**84**

This classical strategy usually requires a large number of experimental runs and subsequent working hours. However, it ignores any potential interactions among the factors, and this is a major drawback, as it may result in an eventually ineffective procedure. On the contrary, a more organized way of conducting experiments, based on statistics, could be cost-effective and time-saving and also enable reaching the real optimal conditions. There are various chemometric approaches, but the most suitable for optimization of a procedure is the design of experiments.

DoE has been applied in many fields, including pharmaceutical product development [24–28]. It is gaining an increasing interest among pharmaceutical researchers, as more and more are becoming familiar with this approach, due to the relatively recent requirement for quality-by-design (QbD) principles. Furthermore, DoE has proven its usefulness in a variety of pharmaceutical applications in this field. Its major advantage has to do with obtaining the optimal conditions among factors for the desired values of responses by conducting a small number of experiments. That way DoE can resolve problems in complex systems, which cannot be easily managed by the trial-and-error approach.

Among the various types of designs like (fractional) factorial, Box–Behnken, central composite design (CCD), etc., the D-optimal design has been established as a robust design strategy. It enables the assessment of both numerical and categorical factors [29], and regarding numerical factors, the latter are examined at many different levels (design matrix), and not at 3–5, as the more classical designs. These levels are generated automatically by computer algorithms from relevant softwares, in order to satisfy the D-optimality criterion, aiming to minimize the generalized variance of the estimated regression coefficients without increasing the total number of experimental runs.

Such design was employed in the study presented by Vlachou et al. [17] regarding MLT controlled-release matrix formulations. One categorical factor, namely, the M.W. of polyvinylpyrrolidone (PVP), was chosen (M.W.: 10.000, low, and 55.000, high), and two numerical factors, namely, the mass (mg) of PVP and hydroxypropyl methylcellulose K15 M, were selected. When a modified release is the aim, as in the current study, setting the right responses is very critical. Herein, the need for a fully release melatonin in a controlled manner within 8 h was the reason for setting as responses the time for 50% drug dissolution at pH = 1.2 and the diffusional exponent (*n*) at pH values 1.2 and 7.4. Initially, a quick melatonin's release is needed for treating sleep-onset problems, while its subsequent slow release is needed to improve sleep quality and/or to assist maintain sleep. Therefore, T50% (pH: 1.2) should be ≤150 min, so that an initial dose will be released to aid the sleep onset of patients, and n (pH: 1.2) = 0.89, in order to achieve zero-order release kinetics and Case II diffusion, and n (pH: 7.4) = 0.80 for first-order release kinetics and anomalous diffusion.

The experiments were conducted as suggested by the experimental plan of Design-Expert software, and then suitable quadratic models were obtained for all (3) responses, satisfying all statistical criteria (ANOVA test, lack-of-fit test, *R<sup>2</sup>* , *adj. R2* and *pred. R2* values). The next step was to estimate the overall optimal conditions, and thus Derringer's desirability function was employed [30], taking into account the necessity for simultaneous optimization of the aforementioned objectives/responses. Desirability function is a tool that is usually included in experimental design softwares and therefore very useful for projects in pharmaceutical development. Each predicted response *Ŷi* and experimentally obtained response *Yi* can be transformed to a desirability function *di*. The latter can have a value from 0 to 1, where *di* = 0 represents completely undesirable response and *di* = 1 represents completely desirable or ideal response. The individual desirability scores *di* can then

**87**

Per Os *Administered Modified-Release Solid Formulations of Melatonin: A Review of the Latest…*

be combined on a single overall (global) desirability *D*, which is optimized to find

In order to reach to optimal solution, the importance of responses should be set by adjusting the importance coefficients. T50% (pH, 1.2) was set as the most important response for consideration, while the rest two were of equal importance. Furthermore, weights (which denote the desired trend of the response within itself) and the range of responses could be changed, according to defined objectives. The optimal solution was reached with a value of global desirability of 0.907, which can be considered as very satisfactory. The suggested solution was performed, and the obtained results were in agreement with the goals defined for the responses and the predictions of the software. Consequently, with just 17 experiments defined by Design-Expert software and few preliminary in order to set the limits of the factors for the experimental plan, a novel MLT modified-release oral

To the best of our knowledge, the previous study was the only attempt to develop novel and improved MLT formulations by utilizing DoE. There has been a previous study [31] in which a different chemometric tool, artificial neural networks (ANN), was utilized. In that study, researchers prepared 27 different tablet formulations with different amounts of hydroxypropyl methylcellulose, xanthan gum and Carbopol® 974P NF. These formulations were subjected to drug release studies, using dissolution test data as inputs for ANN. The authors suggest that ANN with nine neurons in the hidden layer had the best results, meaning that it could predict, after training, dissolution data. In other words, this was a completely different strategy, based on training of the network and prediction of response values for novel (but not very different from training data) excipient mixtures. The optimal solution may not be reached, as the new mixtures are suggested by the user (trial and error) and then tested by the software. On the contrary, DoE is a tool for optimization of a procedure and not prediction of response values, and therefore it

This analysis aims at the review of the latest advances of MLT modified-release oral solid dosage forms including the design of experiments approach. Many scientists have focused on the different ways in manufacturing modified-release oral solid formulations by using various excipients, dosage forms (multilayer or bilayer, coated or uncoated tablets), liposomes, alginate beads, nanofibre mats and nano−/ microparticles, or by employing a variety of techniques (i.e. dry coating, electrospinning, experimental design, etc.) in order to gain knowledge for the production

\_1

*<sup>n</sup>* (1)

*DOI: http://dx.doi.org/10.5772/intechopen.91158*

*D* = (*d*<sup>1</sup> × *d*<sup>2</sup> × …× *dn*)

with *n* denoting the number of responses.

is recommended for application in such projects.

The authors declare no conflict of interest.

**4. Conclusions**

of such dosage forms.

**Conflict of interest**

the optimum set of input variables:

solid dosage form was developed.

be combined on a single overall (global) desirability *D*, which is optimized to find the optimum set of input variables:

$$D = \left(d\_1 \times d\_2 \times \ldots \times d\_n\right)^{\frac{1}{n}}\tag{1}$$

with *n* denoting the number of responses.

*Melatonin - The Hormone of Darkness and Its Therapeutic Potential and Perspectives*

This classical strategy usually requires a large number of experimental runs and subsequent working hours. However, it ignores any potential interactions among the factors, and this is a major drawback, as it may result in an eventually ineffective procedure. On the contrary, a more organized way of conducting experiments, based on statistics, could be cost-effective and time-saving and also enable reaching the real optimal conditions. There are various chemometric approaches, but the most suitable for optimization of a procedure is the design of

DoE has been applied in many fields, including pharmaceutical product development [24–28]. It is gaining an increasing interest among pharmaceutical researchers, as more and more are becoming familiar with this approach, due to the relatively recent requirement for quality-by-design (QbD) principles. Furthermore, DoE has proven its usefulness in a variety of pharmaceutical applications in this field. Its major advantage has to do with obtaining the optimal conditions among factors for the desired values of responses by conducting a small number of experiments. That way DoE can resolve problems in complex systems, which cannot be

Among the various types of designs like (fractional) factorial, Box–Behnken, central composite design (CCD), etc., the D-optimal design has been established as a robust design strategy. It enables the assessment of both numerical and categorical factors [29], and regarding numerical factors, the latter are examined at many different levels (design matrix), and not at 3–5, as the more classical designs. These levels are generated automatically by computer algorithms from relevant softwares, in order to satisfy the D-optimality criterion, aiming to minimize the generalized variance of the estimated regression coefficients without increasing the total

Such design was employed in the study presented by Vlachou et al. [17] regarding MLT controlled-release matrix formulations. One categorical factor, namely, the M.W. of polyvinylpyrrolidone (PVP), was chosen (M.W.: 10.000, low, and 55.000, high), and two numerical factors, namely, the mass (mg) of PVP and hydroxypropyl methylcellulose K15 M, were selected. When a modified release is the aim, as in the current study, setting the right responses is very critical. Herein, the need for a fully release melatonin in a controlled manner within 8 h was the reason for setting as responses the time for 50% drug dissolution at pH = 1.2 and the diffusional exponent (*n*) at pH values 1.2 and 7.4. Initially, a quick melatonin's release is needed for treating sleep-onset problems, while its subsequent slow release is needed to improve sleep quality and/or to assist maintain sleep. Therefore, T50% (pH: 1.2) should be ≤150 min, so that an initial dose will be released to aid the sleep onset of patients, and n (pH: 1.2) = 0.89, in order to achieve zero-order release kinetics and Case II diffusion, and n (pH: 7.4) = 0.80 for first-order release kinetics and anoma-

The experiments were conducted as suggested by the experimental plan of Design-Expert software, and then suitable quadratic models were obtained for all (3) responses, satisfying all statistical criteria (ANOVA test, lack-of-fit test, *R<sup>2</sup>*

tions, and thus Derringer's desirability function was employed [30], taking into account the necessity for simultaneous optimization of the aforementioned objectives/responses. Desirability function is a tool that is usually included in experimental design softwares and therefore very useful for projects in pharmaceutical development. Each predicted response *Ŷi* and experimentally obtained response *Yi* can be transformed to a desirability function *di*. The latter can have a value from 0 to 1, where *di* = 0 represents completely undesirable response and *di* = 1 represents completely desirable or ideal response. The individual desirability scores *di* can then

values). The next step was to estimate the overall optimal condi-

, *adj.* 

easily managed by the trial-and-error approach.

number of experimental runs.

**86**

*R2*

lous diffusion.

and *pred. R2*

experiments.

In order to reach to optimal solution, the importance of responses should be set by adjusting the importance coefficients. T50% (pH, 1.2) was set as the most important response for consideration, while the rest two were of equal importance. Furthermore, weights (which denote the desired trend of the response within itself) and the range of responses could be changed, according to defined objectives. The optimal solution was reached with a value of global desirability of 0.907, which can be considered as very satisfactory. The suggested solution was performed, and the obtained results were in agreement with the goals defined for the responses and the predictions of the software. Consequently, with just 17 experiments defined by Design-Expert software and few preliminary in order to set the limits of the factors for the experimental plan, a novel MLT modified-release oral solid dosage form was developed.

To the best of our knowledge, the previous study was the only attempt to develop novel and improved MLT formulations by utilizing DoE. There has been a previous study [31] in which a different chemometric tool, artificial neural networks (ANN), was utilized. In that study, researchers prepared 27 different tablet formulations with different amounts of hydroxypropyl methylcellulose, xanthan gum and Carbopol® 974P NF. These formulations were subjected to drug release studies, using dissolution test data as inputs for ANN. The authors suggest that ANN with nine neurons in the hidden layer had the best results, meaning that it could predict, after training, dissolution data. In other words, this was a completely different strategy, based on training of the network and prediction of response values for novel (but not very different from training data) excipient mixtures. The optimal solution may not be reached, as the new mixtures are suggested by the user (trial and error) and then tested by the software. On the contrary, DoE is a tool for optimization of a procedure and not prediction of response values, and therefore it is recommended for application in such projects.
