What it does

Fits a dissolution/release profile (% released versus time) to 16 kinetic models, ranks the models by AIC and reports T25…T90 from the coefficients of the selected model. The aim is not merely to find the best-fitting curve but to reach a defensible interpretation of the release mechanism. For the theoretical background, the Turkish naming conventions and the references see the detailed guide (currently in Turkish).

Data entry

• The Time column must be positive; the unit is free (min, h) but must be the same in every row. The point t = 0, F = 0 is not entered; it is assumed.
• Release is entered as a percentage (0–100). If fractions (0–1) are entered a warning is shown; models that approach saturation assume a 100 % ceiling, so the result would be meaningless.
• If replicates (Rep 1–6) are entered, the mean and standard deviation are computed automatically and the analysis uses the mean. If you only have means, type them directly into the Mean column.
• If the profile decreases with time the data are treated as "drug remaining" and converted to release as 100 − F; this is reported as a warning.
• Values above 100 % are not discarded; they are used as entered and a warning is given.
• Copy-and-paste from Excel or another spreadsheet works; the decimal separator may be a comma or a point.

Options

• Geometry is used in two places: in Hopfenberg the exponent n (slab 1, cylinder 2, sphere 3) is not fitted but taken from here; in Korsmeyer–Peppas the geometry-specific thresholds of Costa & Sousa Lobo, Table 1 (slab 0.50/1.00 · cylinder 0.45/0.89 · sphere 0.43/0.85) are applied when interpreting n. Hopfenberg is excluded from the ranking when slab or sphere is selected: for a slab it is the same curve as Zero-order, for a sphere the same as Hixson–Crowell; ranking one curve under two names distorts the Akaike weights. Half sphere (n = 1.5) and triangle (n = 4) follow Karasulu, Ertan & Köse (2000), who adapted the Katzhendler equation to HPMC theophylline tablets; since Costa & Sousa Lobo give no Korsmeyer–Peppas thresholds for these geometries, the slab thresholds are used.
• Tlag: lag time, can be added to every model; F = 0 for t < Tlag. A negative value is allowed and usually means an "intercept".
• F0: initial burst; only in models without a ceiling (Zero-order, Higuchi, Korsmeyer–Peppas). In saturating models Fmax is used instead of F0.
• Fmax: incomplete-release plateau; only in First-order and Weibull. Logistic (Fmax) and Gompertz (Fmax) already carry Fmax in their equations.
• KP: all points: by default Korsmeyer–Peppas is fitted only to the F ≤ 60 % region (Costa & Sousa Lobo). If this box is ticked the whole profile is used; the interpretation of n may then be unreliable.

Reading the results

• The profile summary (AUC, DE, MDT) does not depend on any model; use it to compare two formulations with a single number.
• Ranking is by AIC (lower is better); raw R² does not penalise the number of parameters and therefore unduly favours complex models. The weight is the relative probability that the model is the best one; similar weights mean "the data cannot tell these models apart".
• Korsmeyer–Peppas in the mechanism analysis table is not part of the AIC ranking because it uses a different data set (F ≤ 60 %). Its question is different: not "which model describes the profile best?" but "what is the mechanism in the early phase?"
• Clicking a row plots that model's curve; the coefficients, their units and T25…T90 appear in the panel below. Non Calc means the curve never reaches the target (e.g. Fmax < 75 %, or the Makoid–Banakar peak stays below the target).
• Warning flags must be taken seriously: Fmax > 100 %, Fmax far above the highest observed release (no plateau observed), F0 out of range, n ≈ 0 or m ≈ 0 (degenerate power law), degrees of freedom ≤ 1 and "convergence failed" all say that the fit is unreliable despite a low SS.

Prerequisites

• At least 5–6 time points; with fewer, 2–3-parameter models "memorise" the data. No analysis is performed with fewer than 3 points.
• The points should cover both the rising part and the plateau; plateau points alone cannot determine the rate constant.
• Replicates must come from independent vessels sampled at the same time point; re-reading the same vessel is not a replicate.
• The release percentage must be calculated against the same label amount; otherwise the plateau deviates from 100 % and the Fmax variant misleads.

Limitations

• Not for comparing two profiles. For reference–test similarity use the f1/f2 tool.
• Extrapolation is unreliable. Times such as T90 that lie beyond the last measurement are extensions of the curve; they make a claim about an unmeasured region.
• The spread of the replicates does not enter the model. The analysis uses the mean; differences in variance between time points (heteroscedasticity) are ignored (weighting is not exposed in the interface).
• Formulation-dependent. Weibull may win in one series and Higuchi in another; imposing one model across series needs a justification.
• Not a regulatory acceptance criterion. It does not replace pharmacopoeial or guideline tests (e.g. f2 ≥ 50); it is a tool for formulation development and mechanistic interpretation.