Miscible but supersaturated: what Flory–Huggins theory says about nifedipine and felodipine in PVP

Background and motivation

An amorphous drug dissolves faster and reaches higher apparent solubility than its crystal, but it can crystallize during storage, and the authors open by noting that crystallization "may" reduce apparent solubility, dissolution rate and bioavailability and compromise the dosage form (Abstract). C1 For low-Tg amorphous drugs, adding a polymer as a crystallization inhibitor can prevent crystallization over pharmaceutically relevant times, yet in 2006 there was "much debate" about how the polymer does this, with no consensus beyond a general sense that drug and polymer should be miscible (refs 1–13, including Law 2001, Taylor and Zografi 1997, Crowley and Zografi 2003). C2

The authors give two reasons why miscibility should matter (Introduction). First, only molecular-level mixing changes the drug's local environment; if the two components do not mix, the pure amorphous drug "will largely dominate" crystallization and the polymer can do little (Law 2001; Six 2002). Second, mixing lowers the drug's chemical potential, and with it the thermodynamic driving force for crystallization. C5 They also stress that a dispersion made by solvent removal or melt extrusion is perturbed during processing and must re-equilibrate afterwards: it "may" stay one phase or become metastable or unstable, and in the latter case slow dynamics may still keep it usable. C4

Lattice solution models such as Flory–Huggins theory were well established for polymer–solvent mixtures, but had seen "little progress" in pharmaceutical use, "in part" because the split of the mixing free energy between enthalpy and entropy was uncertain and reliable thermodynamic data were hard to obtain. C6 The key unknown is the Flory–Huggins interaction parameter χ. The usual polymer–solvent methods for χ (vapour pressure reduction, inverse gas chromatography, osmotic pressure) do not suit viscous, non-volatile drug–polymer blends. C21 Two routes remained: an a priori estimate from solubility parameters, which had been advocated for pharmaceutical miscibility (Greenhalgh 1999; Hancock 1997; Forster 2001), and melting point depression, which had given χ for polymer blends (Nishi and Wang 1975) and polymer–solvent systems but had not been applied to drug–polymer pairs. C22·C26

The paper therefore sets out to test whether lattice-based solution models apply to drug–polymer systems, to estimate the relevant thermodynamic parameters, to link miscibility of the amorphous drug with the solubility of the crystalline drug in the polymer, and to use these to understand stabilization. The model systems are two dihydropyridine calcium channel blockers, nifedipine and felodipine, with PVP; a companion paper in the same issue (Marsac, Konno and Taylor 2006, ref 28) had reported that both drugs are miscible with PVP over all concentrations and that nifedipine forms extensive hydrogen bonds with it. A1·A5·R13·R14

Background concepts

Miscibility and solubility are different questions. In this paper miscibility means that the amorphous (supercooled liquid or glassy) drug and the amorphous polymer form one phase by liquid–liquid mixing; the authors call this "clearly" an oversimplification below Tg, where the system is not at equilibrium. C3 Solubility means how much crystalline drug the polymer, acting as a "solvent", can dissolve at equilibrium. A dispersion can therefore be fully miscible and still contain more drug than the polymer can dissolve; crystals appearing in a dispersion "does not necessarily imply" immiscibility, because the solubility limit may simply have been exceeded. C43 Knowing that solubility would give the degree of supersaturation, which the authors call a measure of the driving force for crystallization. C44

Flory–Huggins lattice theory. The theory places molecules on a hypothetical lattice. A polymer chain occupies many adjacent sites, so it has far fewer ways to arrange itself than the same mass of small molecules, and the free energy of mixing is best written in volume fractions rather than mole fractions (Hildebrand and Scott 1950; Flory 1953). C7 Despite its known limits it is "still considered" an excellent starting point for polymer thermodynamics. C8 Treating the amorphous drug as a solvent and taking one lattice site as the volume of one drug molecule, the authors write the free energy of mixing as Eq. 1: C9·E1

Here $\Delta G_M$ is the free energy of mixing, $n$ the number of moles of drug or polymer, $\Phi$ the volume fraction, $R$ the gas constant, $T$ the absolute temperature and $\chi$ the interaction parameter. In plain terms: the first two terms are the entropy gained by mixing, always favourable but small when one partner is a long chain; the third term is the energy change on mixing, favourable when χ is negative and unfavourable when χ is positive. E1

The interaction parameter χ. χ expresses the balance of adhesive interactions (drug–polymer) against cohesive interactions (drug–drug and polymer–polymer). Negative values describe "strong and numerous" adhesive interactions that favour mixing; positive values describe stronger cohesive interactions. C16 χ also has an entropic part and can change with temperature and composition (Flory 1953; Newing 1950; Bawn 1950). C15 Whether a mixture is miscible over the whole composition range depends on χ being below a critical value, which itself depends on how the lattice site is defined. C18

Solubility parameters. The Hildebrand solubility parameter δ is the square root of the cohesive energy density. Here it was estimated by the Fedors group-contribution method: the contributions of molecular fragments to the energy of vaporization are summed and divided by the summed fragment molar volumes, and δ is the square root of that ratio. P4·E9 From δ, χ follows as Eq. 2: E2

where $V_{\mathrm{s}\mathrm{i}\mathrm{t}\mathrm{e}}$ is the volume of a lattice site. In plain terms: the larger the difference between the two solubility parameters, the more unfavourable the mixing. Because the difference is squared, χ from Eq. 2 can never be negative; the relation assumes that the interaction between unlike molecules is the geometric mean of the interactions between like molecules (Hildebrand and Scott 1950). E2·C24

Melting point depression. A crystal mixed with a liquid in which it dissolves melts at a lower temperature, because the drug in the mixture has a lower chemical potential and reaches equality with the crystal at a lower temperature. The authors call melting point depression "a manifestation of the reduced thermodynamic activity of the drug in the presence of the polymer". C39 For a crystalline solvent in an amorphous polymer, lattice theory gives the size of the depression (Hoei et al. 1992); the authors extend it to a crystalline drug in an amorphous polymer as Eq. 3. C27·E3

Driving force for crystallization. The free energy released when an amorphous (supercooled liquid) drug crystallizes, written here $\Delta G_v$, grows as the temperature falls below the melting point. The Hoffman approximation estimates it from the heat and temperature of fusion alone (Hoffman 1958), and the companion paper found that it "adequately" predicts this quantity for felodipine and nifedipine. C37·R15

Methods in detail

The paper combines a small experimental programme with a chain of calculations. Only three kinds of measurement were made: true densities by helium pycnometry, melting onsets of drug–PVP K12 physical mixtures by DSC, and the solubility of the crystalline drugs in 1-methyl-2-pyrrolidone (methyl pyrrolidone), the small-molecule analogue of PVP. P5·P1·P3 Everything else is computed from Flory–Huggins theory and the Hoffman equation with the inputs of Table I. P6·P8·P9

Inputs (Table I). Nifedipine has MW 346.34 g/mol, density 1.20 g/cm³, molecular volume 288.62 cm³/mol, heat of fusion 39.89 kJ/mol, melting temperature 445.25 K and solubility parameter 22.9 J^1/2/cm^3/2. Felodipine has 384.26 g/mol, 1.28 g/cm³, 300.20 cm³/mol, 30.83 kJ/mol, 414.75 K and 25.0. PVP K12 has MW 2,500, density 1.11 g/cm³, molecular volume 2,252.25 cm³/mol and solubility parameter 27.4. M1·M2·M3 Densities were measured by helium pycnometry on the amorphous materials and molecular volumes are MW divided by density, which the AI writer recomputed and found exact. P5·E7 The text gives melting points of 171 °C (nifedipine) and 142 °C (felodipine), and Tg values of 100 °C for PVP K12 (Taylor et al. 2001) and 168 °C for PVP K29/30. M1·M2·M3·M4 Note by the AI writer: 445.25 K is 172.1 °C, about one degree above the 171 °C of the text and the measured pure onset in Figures 3b and 4b. X6

Contribution 1: how much entropy a polymer brings to mixing

Problem. If the entropy of mixing a drug with a polymer depended strongly on the polymer's molecular weight, a formulator could tune miscibility by choosing a grade. The authors test this with Eq. 1. C14

What they did. With χ = 0, Eq. 1 gives the entropy of mixing across composition. It was computed for nifedipine with methyl pyrrolidone and with PVP K12 (MW 2,500), PVP K29/32 and PVP K90 (MW 1,250,000), with one lattice site equal to the nifedipine molecular volume (Figure 1). P6·R1 Eq. 1 was then evaluated for nifedipine–PVP K29/32 with χ = −4.2, −3.8, 0.5, 1.0 and 2.0 to see how the shape of the free-energy curve changes with χ (Figure 2). R2 The temperature of these simulations is not stated. P6

Why it works. Both figures are computed shapes rather than data, and their vertical axes carry no scale, so they show only relative behaviour. R1·R2·X7

Contribution 2: two ways to obtain χ

Problem. Eq. 1 predicts miscibility only if χ is known. The cheap route is Eq. 2; the paper's new route is melting point depression. C22·C26

The solubility-parameter route. Eq. 2 with the Table I values gave χ = 0.5 for felodipine–PVP and 2.0 for nifedipine–PVP. R4 Note by the AI writer: these two printed values cannot be reproduced from Table I at a single temperature; Eq. 2 gives 2.36 and 0.70 at 298.15 K, or 1.58 and 0.50 at each drug's melting temperature, and the paper does not state which temperature it used. Both values are positive in every case, so the conclusion drawn from their sign is unaffected. E2·X1

The melting-point route. Extending equations for crystalline solvents in amorphous polymers (Hoei et al. 1992), the authors relate the melting temperature of the drug in the mixture to χ: E3

Here $T_M^{\mathrm{m}\mathrm{i}\mathrm{x}}$ and $T_M^{\mathrm{p}\mathrm{u}\mathrm{r}\mathrm{e}}$ are the drug's melting temperature with and without polymer, $\Delta H_{\mathrm{f}\mathrm{u}\mathrm{s}}$ is the heat of fusion of the pure drug and $m$ is the volume of a polymer chain divided by the lattice-site (drug) volume. In plain terms: the bracket is the logarithm of the drug's activity in the melt; the more negative it is, the lower the melting point. The first two terms come from mixing entropy, the last from the interaction. E3 The method requires that drug and polymer are chemically stable over the temperature range and interact enough for the depression to show. C27

Physical mixtures were made by geometric mixing of particles sieved to 45–75 µm and dried over phosphorus pentoxide for at least one week: nifedipine with 5, 10, 15, 20 and 25 wt% PVP K12, and felodipine with 5, 10, 15, 17.5, 20, 22.5 and 25 wt%, two to three samples per concentration. P1 Melting was measured in a TA 2920 modulated DSC at 1 °C/min under 45 ml/min nitrogen, with indium and benzophenone for temperature calibration and indium for enthalpy, and taken as the extrapolated onset of the bulk melting endotherm. P1 Volume fractions were obtained from weight fractions and the true densities. E8

The method needed development. The first attempts, with PVP K29/30 at 10 °C/min, showed "essentially no" depression (data not shown). R11 The authors attribute this "most likely" to the polymer's high Tg (168 °C): the polymer is not liquid-like at the drug melting points (171 and 142 °C), so drug and polymer mix too slowly, an artefact also known from polymer blends (Nishi and Wang 1975). C29 Slowing the scan to 1 °C/min, controlling particle size and switching to PVP K12 (Tg 100 °C) made a "substantial" depression visible. C30

To extract χ, Eq. 3 was rearranged so that the left side, plotted against ${\Phi }_{\mathrm{p}\mathrm{o}\mathrm{l}\mathrm{y}\mathrm{m}\mathrm{e}\mathrm{r}}^2$, gives a straight line of slope χ: E10

In plain terms: subtracting the entropy terms from the measured depression leaves only the interaction term, which grows with the square of the polymer fraction. Only data from 0 to 0.25 were used, and the authors note that the resulting χ refers to temperatures close to the drug's melting point. P7·C32

How it differs from earlier work. Melting point depression of drugs had been documented mainly in polyethylene glycol, which melts near 60 °C and so is liquid when the drug melts, making equilibration easy (Kaur 1980; Bartsch 2004; Anastasiadou 1983). C28 With a high-Tg polymer like PVP the experiment only works when the polymer is mobile enough at the drug's melting point, which is why the low-molecular-weight grade and slow scans were needed. C29·C30

Contribution 3: how much mixing lowers the driving force

Problem. If mixing with PVP strongly lowered the drug's chemical potential, thermodynamics alone could explain why dispersions crystallize slowly. C38

What they did. For the pure amorphous drug the Hoffman equation gives: E4

In plain terms: the free energy released on crystallization is zero at the melting point and grows roughly in proportion to the undercooling; the approximation assumes that the enthalpy difference between crystal and supercooled liquid varies linearly with temperature. E4·C37 For nifedipine mixed with 30% and 40% PVP, a "modified version" of the Hoffman equation (Hoffman 1958) was used with the melting point depression data; the modified form is not printed. P8·E11 Figure 5 plots the result from 0 to 180 °C. R12

Contribution 4: a model for the solubility of a crystalline drug in a polymer

Problem. The solubility of drugs had been measured in a low-Tg polymer (Jasti et al. 2004), but not in high-Tg polymers such as PVP, and one practice estimated it from a small analogue of the polymer (Breitenbach 2002). C44·C51

What they did. The solubility of a crystal in a solvent is given by Eq. 5, and the authors assume that the polymer can play the role of the solvent: E5·C45

Here ${\gamma }_{\mathrm{d}\mathrm{r}\mathrm{u}\mathrm{g}}$ is the activity coefficient, $x_{\mathrm{d}\mathrm{r}\mathrm{u}\mathrm{g}}$ the mole fraction of dissolved drug, $T_m$ the melting temperature and $\Delta C_P^{\mathrm{c}\mathrm{o}\mathrm{n}\mathrm{f}\mathrm{i}\mathrm{g}}$ the configurational heat capacity difference between supercooled liquid and crystal. In plain terms: the right side is the ideal solubility of the crystal, which falls with a higher heat and temperature of fusion; the activity coefficient corrects it for non-ideal mixing. E5 Flory–Huggins theory supplies the activity coefficient: E6

In plain terms: the first two terms account for the size difference between drug and polymer, and the last for the interaction; a negative χ lowers γ and so raises the solubility. E6 Eqs. 5 and 6 were solved together with the melting-point χ for PVP K12 and PVP K29/32. P9

Validation through methyl pyrrolidone. The solubility of a crystal in a polymer cannot easily be measured, so the authors tested the method on methyl pyrrolidone. They assumed that χ for the drug with methyl pyrrolidone equals the χ measured with PVP K12, and that mixing with this small molecule has ideal entropy, so that the first two terms of Eq. 6 vanish. C46·C47 The solubility of both crystalline drugs in methyl pyrrolidone was measured at 25 °C: excess crystals in a jacketed vessel for at least 24 h with vigorous stirring and protected from light, then decanting, centrifugation, filtration through a 0.22 µm filter, dilution with ethanol and UV–Vis against a standard curve. P3 The number of replicates is not stated. P3

Note by the AI writer (not discussed in the paper): the paper states neither the temperature of the predictions nor the heat-capacity values used in Eq. 5. Setting the heat-capacity terms to zero and taking 25 °C, the AI writer reproduced the PVP K12 predictions (0.07 g/g nifedipine, 0.31 g/g felodipine) and the nifedipine value in PVP K29/32 (0.06 g/g), but obtained about 0.29 g/g instead of the printed 0.25 for felodipine in PVP K29/32. The methyl pyrrolidone predictions could not be checked, because its molar volume is not printed. E5·E6·X11

Experimental analysis

A polymer adds little mixing entropy, and the grade hardly matters

In Figure 1 the entropy curve for nifedipine with methyl pyrrolidone is far deeper (minimum near a volume fraction of about 0.44) than the curves for any PVP grade (minima near about 0.6); PVP K12 lies slightly below PVP K29/32 and K90, which the caption calls "virtually identical" at every composition. R1 The authors conclude that mixing with a polymer is "much less favorable" entropically than with a small molecule, because chain connectivity reduces the polymer's configurational entropy, while the effect of the grade is "much less marked". C10·C11 From this they generalize that for "typical" drugs (MW 200–600) and polymers (MW 10,000–1,500,000) the entropy term is "predicted to be relatively constant", which "leads to the inference" that the balance of adhesive and cohesive interactions "will essentially determine" miscibility; a chemically different polymer is therefore "more likely" to bring miscibility than a lower-molecular-weight grade. C12·C13·C14 Note by the AI writer: only nifedipine was simulated, PVP K12 (MW 2,500) lies below the authors' own "typical" range, and the plot has no vertical scale, so "relatively constant" is a reading of curve shapes. X8·X7

Solubility parameters predict the wrong answer for nifedipine

Figure 2 shows how χ changes the free-energy curve of nifedipine–PVP K29/32: at χ = 2.0 the free energy of mixing is positive at all compositions (immiscible); at 0.5 and 1.0 it is only slightly negative; at −3.8 and −4.2 it is deeply negative with a single minimum near a volume fraction of about 0.54. R2 From these simulations the authors estimate that χ must be "approximately 0.6 or less" for miscibility over the whole composition range, and infer that some unfavourable interaction can be tolerated because of the mixing entropy, though less than for mixtures of small molecules. R3·C19

The solubility-parameter values, 2.0 for nifedipine–PVP and 0.5 for felodipine–PVP, predict endothermic mixing and, for nifedipine, immiscibility. R4·C23 This contradicts the companion study, in which both drugs were miscible with PVP over all concentrations. R13 The authors' explanation is that the discrepancies "most likely arise" from using solubility parameters for systems with specific interactions: nifedipine hydrogen-bonds extensively to PVP (companion study), and the geometric-mean assumption behind Eq. 2 is reasonable for van der Waals interactions but "not" for specific directional ones. C24·R14 Because many drugs and pharmaceutical polymers form such interactions (Taylor and Zografi 1997; Tang et al. 2002), they judge that solubility-parameter estimates of χ "are likely to be inaccurate for such systems". C25 Hydrogen bonding itself was not measured in this paper. C24

Melting point depression in PVP K12

Figure 3a shows the melting endotherms of nifedipine with 0–25 wt% PVP K12: one sharp, deep endotherm with its minimum near about 173 °C and shallower, broader endotherms that start at lower temperatures; the curves are not labelled by composition. R5 Figure 3b gives the extrapolated onsets, summarized in Table A. R6

Table A. Extrapolated melting onsets of drug–PVP K12 physical mixtures at 1 °C/min, read from Fig. 3b (approximate); "—" = no point plotted. R6

The Methods describe a linear decrease of the melting point over 0–25 wt% PVP K12. R7 The depression is clear for both drugs: nifedipine falls by about 15 °C and felodipine by about 19 °C at a volume fraction of about 0.28, and the authors find the extent "similar" to that seen for crystalline polymers in solvents (Mandelkern 1964). R6·C30 Note by the AI writer: three details of Figure 3b need care. The figure plots more compositions than the Methods list (up to a volume fraction of about 0.38, which corresponds to about 35 wt%). X3 The nifedipine and felodipine points sit at the same volume fractions, although the drugs' densities differ, and these positions match the conversion with felodipine's density (1.28 g/cm³); with nifedipine's own density, 25 wt% PVP would be at 0.265 rather than about 0.278. X2·E8 And the felodipine onsets are not monotonic beyond about 0.2 (about 122, 126 and 122 °C at 0.224, 0.250 and 0.277). X4

A negative χ, and it reproduces the melting curve

Plotted in the rearranged form of Eq. 3, the nifedipine data fall on a straight line at low polymer content (Figure 4a); r² was 0.99 for nifedipine and 0.92 for felodipine over 0–0.25, and linearity was lost at higher concentrations. R8 The slopes give χ = −3.8 for nifedipine–PVP and −4.2 for felodipine–PVP (Table B). R9 The authors attribute the loss of linearity "most likely" to the composition dependence of χ and to slower mixing as the depressed melting point approaches the polymer Tg. C31 Note by the AI writer: the range "0–0.25" is called a concentration range on p. 2421 and a volume fraction on p. 2422, while Figure 4a includes points up to a volume fraction of about 0.28 (25 wt%); the dashed fit line has a small positive intercept (about +0.016, read) although Eq. 3 implies none; and the felodipine plot is not shown. X4

Table B. The two estimates of the interaction parameter (pp. 2420–2422). Full miscibility of nifedipine–PVP needs χ of about 0.6 or less. R4·R9·R8·R3

Figure 4b puts χ back into Eq. 3. All three predicted curves start at about 172 °C. With χ = −3.8 the curve falls to about 170 °C at a volume fraction of 0.10, about 164 °C at 0.20 and about 155 °C at 0.30, and the measured onsets lie within about 1 °C of it (about 156.5 °C measured against about 157 °C predicted at about 0.28). R10 With χ = 0, the depression comes only from mixing entropy and is small (about 168.4 °C at 0.30); with χ = 2.0, the solubility-parameter value, the curve rises to about 175.8 °C at 0.30, which the authors call "not physically meaningful". R10·C34 Recomputing Eq. 3 with the printed inputs gives 157.1 °C at a volume fraction of 0.278, matching the plotted curve. E3 The authors conclude that the observed depression is "certainly consistent with" a negative χ and exothermic mixing (Abstract: "indicating that mixing was exothermic"), that χ = −3.8 predicts miscibility of nifedipine–PVP (Figure 2), "consistent with experimental observations", and that "our results suggest" the melting point depression method can give reasonable values of χ. C35·C33·C36

The driving force for crystallization falls only a little

Figure 5 gives the free energy change on crystallization of amorphous nifedipine with and without PVP (Table C). R12 Recomputed from Eq. 4, the pure-drug value at 0 °C is −9.46 kJ/mol, as plotted. E4

Table C. Free energy change on crystallization of amorphous nifedipine, Hoffman estimate, read from Fig. 5 (approximate). R12

Mixing with 40% PVP shifts the curve by only about 1 kJ/mol at room temperature, and the authors write, in italics, that the predictions "do not forecast a large reduction in the driving force", particularly at lower temperatures. The difference between pure nifedipine and the 40% PVP dispersion is smaller than the difference between nifedipine and felodipine reported in the companion study. C41 They add that the reduction is "most likely underestimated", because equilibrium melting point depression is hard to reach at high polymer content. C40 Since pure nifedipine and felodipine nucleate much faster than dispersions with even small amounts of polymer (companion study), the reduced driving force "seems unlikely" to explain the slower crystallization, "suggesting that kinetic and/or interfacial energy factors are of great importance" ("as demonstrated here quantitatively for the first time"). R16·C42

Note by the AI writer: the 30% PVP curve reaches zero at about 148 °C, the melting onset measured at a volume fraction of about 0.33, but the 40% PVP curve reaches zero at about 128 °C, below every nifedipine onset in Figure 3b (the lowest is about 144.5 °C), although the text says the curves use the Figure 3 data. Replacing the melting temperature in Eq. 4 by 148 and 128 °C reproduces both curves at 0 °C (−9.09 and −8.67 kJ/mol), so the "modified" equation appears to be Eq. 4 with the depressed melting point, but the source of the 40% value is not reported. X5

The crystalline drugs dissolve poorly in PVP

Table D. Solubility of the crystalline drugs, g drug/g "solvent", transcribed from Table II. R17

For methyl pyrrolidone the predictions (0.15 and 0.48 g/g) are close to the measured values (0.21 and 0.45 g/g), and the authors take the agreement as showing that the melting-point χ is "a reasonable estimate". R17·C48 With χ = 0.5, nifedipine solubility in methyl pyrrolidone drops to 0.004 g/g, "more than an order of magnitude" below the measured value, which shows how sensitive solubility is to χ. R17·C49 Note by the AI writer: the felodipine prediction lies inside the measured range, but the nifedipine prediction is about 30% below the measured value and outside its stated spread, so "good agreement" holds better for felodipine than for nifedipine; the check also tests two assumptions at once, the transfer of χ from PVP to methyl pyrrolidone and the ideal mixing entropy. X10

Solubility in PVP is lower than in methyl pyrrolidone (nifedipine 0.06–0.07 against 0.15 g/g), which the authors trace to the polymer's smaller entropic contribution; estimating a drug's "solubility" in a polymer from its solubility in a small analogue therefore needs correcting ("need to be corrected") to avoid an overestimate. R17·C50·C51 The lower-melting felodipine, with its lower heat of fusion, is predicted to be more soluble than nifedipine (0.25–0.31 against 0.06–0.07 g/g), and the authors generalize that drugs with a high heat or temperature of fusion "will therefore be less soluble in a polymer", while a positive mixing enthalpy lowers and a negative one raises the solubility. R17·C52·C53 Note by the AI writer: the two drugs also differ in χ (−4.2 against −3.8), so this comparison is not a pure test of the melting properties. X13

For nifedipine the formulation would need more than 90% polymer for no drug to crystallize on a thermodynamic basis, feasible only for very low doses. R18 The authors conclude that the results "also suggest" that drug solubility in polymers is low enough for the drug to be supersaturated in most practical formulations, and that the physical stabilization of such dispersions is "largely due to altering the kinetics". C55·C56

Discussion

The authors' chain of argument. The paper's conclusion is built in steps, each resting on the one before. The diagram follows the authors' own reasoning; the split at the bottom is the case distinction they draw for formulations. C12·C33·C41·C55·C56

The step from 3 to the bad branch is where the paper's main message lies, and it is reached by elimination: thermodynamics leaves a large driving force, the pure drugs nucleate faster than the dispersions (companion study), so something kinetic must be at work. C42·C56·R16 The authors' wording is stronger in the Conclusions ("can therefore be attributed predominantly") than in the Abstract ("suggesting that kinetic rather than thermodynamic stabilization plays a significant role"). C56 Their final sentence keeps miscibility important, but for a different reason than the debate they started from: molecular-level mixing matters because it changes the drug's environment, "which leads to altered thermodynamics and more importantly dynamics". C59

Where the approach applies. The melting point depression method needs a crystalline drug that is chemically stable at its melting point and a polymer that is mobile there; with PVP this meant the low-molecular-weight grade K12 and slow heating. C27·C29 The χ it delivers belongs to temperatures near the drug's melting point (here roughly 120–172 °C), not to storage conditions, and χ is known to depend on temperature and composition. C32·C15·R6 The solubility model then assumes the polymer behaves as a solvent and that the same χ holds at the (unstated) prediction temperature. C45·X11

What it suggests for formulation (authors' statements). If a drug does not mix with a polymer, a chemically different polymer is "more likely" to help than another grade of the same polymer. C14 Solubility-parameter estimates of χ are "likely to be inaccurate" for drugs that hydrogen-bond to the polymer. C25 Solubility measured in a small analogue of the polymer overestimates solubility in the polymer. C51 High-melting drugs "will most likely be harder to stabilize since they will be more supersaturated". C54

Note by the AI writer (not discussed in the paper). The practical reading for a formulator is that a negative χ answers "will it mix?", not "will it stay amorphous?". Unless the dose is very low, a PVP dispersion of these drugs is predicted to be supersaturated with respect to the crystal, so its stability depends on keeping the drug immobile, for example by storage well below Tg and by limiting moisture. The paper itself does not measure mobility or crystallization, so this is the consequence of its argument, not a tested result. X12

Limitations

Limitations the authors state. Miscibility as defined is "clearly" an oversimplification below Tg. C3 The critical value of χ depends on how the lattice site is defined. C18 Eq. 3 gives χ at temperatures close to the drug's melting point. C32 Linearity of the χ plot is lost at higher polymer content, "most likely" through composition-dependent χ and slow kinetics. C31 The reduction in driving force is "most likely underestimated" because equilibrium melting points are hard to reach at high polymer content. C40

The AI writer's own judgement (1): the measured core is solid, the extrapolations are not. The melting point depression of both drugs in PVP K12 is clear, and for nifedipine the fitted χ reproduces the measured curve within about 1 °C; the sign of χ is robust, because neither χ = 0 nor a positive value can produce the observed depression. R6·R10·C34 The method comparison is also clear-cut: positive solubility-parameter values contradict the reported miscibility. R4·R13 What is weaker is everything built on χ: the solubility predictions carry no stated temperature or heat-capacity values, one of them (felodipine in PVP K29/32) does not reproduce, and the only check, in methyl pyrrolidone, is about 30% low for nifedipine. X11·X10

The AI writer's own judgement (2): inconsistencies in the source. The solubility-parameter values (2.0 and 0.5) cannot be recomputed from Table I at one temperature. X1 The nifedipine volume fractions in Figures 3b and 4 appear to use felodipine's density, and Figure 3b shows compositions that the Methods do not list. X2·X3 The fit range is stated in two ways and the felodipine fit is not shown. X4 The 40% PVP curve of Figure 5 rests on a melting temperature that does not appear in Figure 3. X5 The PVP grade is printed both as K29/30 and K29/32, and the Figure 1 axis label (ΔS/RT, no scale) does not match the plotted negative curves. X7

The AI writer's own judgement (3): the generalizations outrun the data. The claim that the mixing entropy is "relatively constant" for typical drugs and polymers rests on one drug and three PVP grades. X8 The miscibility of felodipine with PVP was not computed with its own volumes; its χ appears only as a curve on the nifedipine system. X9 The kinetic conclusion is inferred by elimination, with no kinetic, nucleation or mobility data in the paper. X12 And the comparison of the two drugs' solubilities mixes the effect of melting properties with a difference in χ. X13

Conclusion

The paper made three contributions. It showed, with nifedipine and felodipine in PVP, that melting point depression can be turned into a drug–polymer Flory–Huggins interaction parameter, provided the polymer is mobile at the drug's melting point; it gave χ = −3.8 and −4.2, while solubility parameters gave positive values that wrongly predicted immiscibility for nifedipine. R9·R4·C36 It separated miscibility of the amorphous drug from solubility of the crystalline drug in the polymer, and built a Flory–Huggins model for the latter. C43·E5·E6 And it used these numbers to argue that the solubility of crystalline drugs in PVP is low (0.06–0.07 g/g nifedipine, 0.25–0.31 g/g felodipine), that the driving force for crystallization drops only slightly on mixing, and that crystallization inhibition in these dispersions is therefore "predominantly" kinetic. R17·R12·C56

The measured part, the melting point depression and the negative χ, is well supported; the solubility values and the kinetic conclusion rest on calculations with unstated conditions and on cited nucleation data, and remain to be confirmed by direct measurements of drug solubility in the polymer and of crystallization kinetics. R6·R10·X11·X12