Artificial Intelligence-Enabled Quality by Design and Regulatory Science for Apixaban Orodispersible Tablets: A Critical Review of Clinical Rationale, Formulation Strategy, and Submission-Ready Documentation for Stroke Prevention

Gururaj S Kulkarni¹, Mithun N², Anna Balaji³

¹Department of Pharmaceutics (HOD), The Oxford College of Pharmacy, Bengaluru, Karnataka, India

²Department of Pharmaceutics (PG Student), The Oxford College of Pharmacy, Bengaluru, Karnataka, India

³Principal, The Oxford College of Pharmacy, Bengaluru, Karnataka – 560068, India

 

*Correspondence: tocpceutics@gmail.com

DOI: https://doi.org/10.71431/IJRPAS.2026.5307     

Article Information		Abstract
Review Article Received: 22/03/2026 Accepted:26/03/2026 Published:31/03/2026   Keywords Apixaban; Artificial Neural Networks; eCTD; Genetic Algorithms; Orodispersible Tablets; Quality by Design; Stroke Prevention		Stroke is a leading global cause of death and acquired disability, affecting millions annually. Among post-stroke patients, 36–51% experience oropharyngeal dysphagia, rendering conventional film-coated tablets of apixaban – a direct factor Xa inhibitor – difficult or unsafe to administer. This review critically examines the evidence base for developing an apixaban Orodispersible Tablet (ODT) formulation across three converging domains: (1) global stroke burden and dysphagia-related adherence barriers in dysphagic, geriatric, and paediatric populations; (2) apixaban's biopharmaceutical constraints – particularly poor aqueous solubility (~40 µg/mL) and BCS Class II/IV designation – mapped onto ODT manufacturing approaches (direct compression, lyophilisation, sublimation) and excipient selection strategies; and (3) the integration of Artificial Intelligence (AI) within a formal Quality by Design (QbD) framework, detailing how Artificial Neural Networks (ANNs) and Genetic Algorithms (GAs) can optimise Critical Quality Attributes and resolve multi-objective formulation trade-offs. The review further maps AI-generated data streams onto eCTD Modules 2, 3, and 5 in alignment with FDA, EMA, and ICH requirements for data integrity (ALCOA+), model lifecycle management, and audit trail compliance.

 

 

 

 

INTRODUCTION   

Global Stroke Burden and the Dysphagic Patient

Stroke represents a persistent global health emergency, ranking as the second-leading cause of death worldwide and the third-leading cause of death and disability combined as measured by disability-adjusted life years (DALYs).[1]

The 2019 Global Burden of Disease (GBD) study documented 12.2 million incident strokes (95% uncertainty interval [UI] 11.0–13.6 million), 101 million prevalent strokes, and 6.55 million stroke-related deaths globally.[6]

While age-standardised stroke incidence and mortality rates declined between 1990 and 2019, absolute numbers increased dramatically due to population ageing and growth: incident strokes increased by 70.0%, prevalent strokes by 85.0%, and DALYs by 32.0%. Critically, among individuals younger than 70 years, prevalence rates increased by 22.0% and incidence rates by 15.0%, signalling an epidemiological shift toward younger populations.[6][7]

Without urgent primary prevention and improved acute care delivery, projections indicate that by 2050 there will be over 200 million stroke survivors, approximately 300 million stroke-related DALYs, 25 million new strokes annually, and 13 million deaths from stroke each year.[6]

Oropharyngeal dysphagia – impaired swallowing efficiency and safety – is a prevalent and clinically significant complication of acute stroke. A prospective study of 201 stroke patients documented dysphagia prevalence of 42.8% during hospitalisation.[1] Meta-analysis of Asian stroke populations revealed a pooled dysphagia occurrence rate of 36.3% (95% CI 33.3%–39.3%),[3] while Chinese epidemiological surveys reported 51.14% prevalence among stroke patients.[2]

Figure 1. Global stroke incidence, prevalence, mortality, and DALY trends from 1990–2019, with projections to 2050. Adapted from GBD 2019 Stroke Collaborators.

 

Anticoagulation Imperatives and Conventional Dosage Form Limitations

Atrial fibrillation-related cardioembolic stroke and venous thromboembolism prevention require sustained oral anticoagulation. Apixaban, a selective direct factor Xa inhibitor, is a guideline-recommended first-line agent for stroke prevention in nonvalvular atrial fibrillation and treatment/prevention of deep vein thrombosis and pulmonary embolism.[5][14] The drug is currently marketed as film-coated tablets (Eliquis®) in 2.5 mg and 5 mg strengths.[4][5]

For dysphagic stroke patients, conventional tablets present multiple adherence and safety barriers including swallowability constraints, administration delays when crushed for tube feeding, adherence barriers in elderly (≥75 years) and paediatric populations, and limitations related to onset and bioavailability.[2][15]

Apixaban exhibits approximately 50% oral bioavailability with time to maximum concentration (Tmax) of 3–4 hours and a half-life of approximately 12 hours, with food-independent absorption.[15]

Orodispersible Tablets: A Patient-Centred Solution

Orodispersible tablets (ODTs) are solid dosage forms designed to disintegrate rapidly in the oral cavity (typically <30 seconds) without requiring water.[16][17][18] ODTs offer multiple patient-centred advantages: rapid disintegration and buccal/sublingual absorption pathways; elimination of swallowing barriers enabling administration in dysphagic patients, children, and the elderly; improved adherence through ease of administration and palatability enhancement; and suitability for emergency and field settings where water availability is limited.[19]

The Role of Artificial Intelligence in Pharmaceutical Development and Regulatory Science

Traditional pharmaceutical development employs one-factor-at-a-time (OFAT) experimentation or classical Design of Experiments (DoE) to optimise formulation composition and manufacturing processes. AI and machine learning (ML) techniques – particularly artificial neural networks (ANNs) and evolutionary algorithms such as genetic algorithms (GAs) – offer transformative capabilities for non-linear modelling, multi-objective optimisation, and data efficiency.[20][21]

ANN-GA hybrid workflows reduce experimental burden by training surrogate models on limited datasets and guiding subsequent experimentation toward optimal regions. Despite these advantages, pharmaceutical applications of AI face regulatory scrutiny regarding validation, data integrity, model interpretability, and human oversight.[11][12][22]

Objectives and Scope

This review critically synthesises the evidence base for apixaban ODT development through three integrated lenses: (1) clinical imperative – quantifying stroke burden, dysphagia prevalence, and adherence barriers; (2) formulation science and QbD – characterising apixaban's biopharmaceutical properties and evaluating manufacturing methods; and (3) AI-driven development and regulatory readiness – mapping ANN and GA applications to QbD and aligning with FDA/EMA/ICH eCTD expectations.[18][23]

 

 

Global and Regional Stroke Epidemiology: Burden, Trends, and Vulnerable Populations

Overall Stroke Burden and Mortality (2019 Baseline)

The most recent comprehensive global stroke surveillance derives from the GBD 2019 study, which documented 12.2 million incident strokes (95% UI 11.0–13.6 million) and 101 million prevalent strokes worldwide.[6]

Stroke accounted for 6.55 million deaths (6.00–7.02 million), representing 11.6% of all global deaths in 2019, ranking stroke as the second-leading cause of mortality after ischaemic heart disease. Combined death and disability burden reached 143 million DALYs, making stroke the third-leading cause of DALYs globally.[6][24]

From 1990 to 2019, absolute stroke burden increased dramatically despite declining age-standardised rates. Incident strokes increased by +70.0% (95% UI 67.0–73.0%); prevalent strokes by +85.0% (83.0–88.0%); deaths by +43.0% (31.0–55.0%); and DALYs by +32.0% (22.0–42.0%). This paradox – rising absolute burden amid falling age-standardised rates – reflects population ageing and growth as dominant epidemiological drivers.[6][7]

Stroke Subtypes and Geographic Disparities

Ischaemic stroke predominated in 2019, accounting for 62.4% of incident strokes (7.63 million cases, 95% UI 6.57–8.96 million), followed by intracerebral haemorrhage (27.9%, 3.41 million cases) and subarachnoid haemorrhage (9.7%, 1.18 million cases). This distribution differs markedly from paediatric patterns, where approximately 50% of strokes are haemorrhagic.[6][25]

Geographic inequities are profound. Age-standardised stroke mortality rates in World Bank low-income countries were 3.6 times higher (95% UI 3.5–3.8) than in high-income countries. In absolute terms, 86% of all stroke deaths and 89% of stroke-related DALYs occurred in lower-income and lower-middle-income countries in 2019.[6][26][7]

Figure 2. Geographic distribution of stroke burden (deaths and DALYs) by World Bank income classification. Data from GBD 2019 and young stroke scoping review.

Young Adult and Paediatric Stroke: Emerging Epidemiological Shift

Among individuals younger than 70 years, GBD 2019 data revealed a concerning reversal of favourable age-standardised trends: prevalence rates increased by 22.0% (95% UI 21.0–24.0%) and incidence rates increased by 15.0% (12.0–18.0%) between 1990 and 2019.[6]

A comprehensive scoping review of stroke in individuals ≤30 years analysed 471 articles from 50 countries, identifying heterogeneous incidence reporting and risk factor profiles. In perinatal stroke, leading risk factors were infections, cardiac conditions, and intrapartum factors; in paediatric stroke, vasculopathies and infections predominated; and in young adult stroke, chronic conditions including diabetes and hypertension were most frequently reported.[28]

The Young ESUS Longitudinal Cohort Study enrolled 535 participants ≤50 years of age from 41 stroke centres in 13 countries and documented a recurrent ischaemic stroke and death rate of 2.19 per 100 patient-years. Prior stroke or TIA (hazard ratio 5.3), diabetes (HR 4.4), and coronary artery disease (HR 10) were associated with recurrent events.[29]

Stroke and SARS-CoV-2 in Paediatric Populations

An international survey during March–May 2020 identified SARS-CoV-2 infection in 8 of 971 (0.82%) hospitalised paediatric patients with stroke.[30]

A subsequent cohort study (June–December 2020) documented SARS-CoV-2 positivity in 23 of 335 acute arterial ischaemic stroke cases tested (6.9%). Elevated inflammatory markers occurred in 77.3% of SARS-CoV-2-positive stroke cases, suggesting infection-mediated vasculopathy as a plausible mechanism.[31]

Dysphagia in Acute and Post-Stroke Populations: Prevalence, Prognostic Implications, and Medication Administration Barriers

Prevalence of Oropharyngeal Dysphagia in Stroke

Oropharyngeal dysphagia is characterised by impaired efficiency (prolonged bolus transit, incomplete oral clearance) and safety (aspiration, penetration) of swallowing.[32]

A prospective cohort study of 201 stroke patients admitted to a dedicated Stroke Unit documented oropharyngeal dysphagia prevalence of 42.8% (86 patients) during acute hospitalisation. Dysphagic patients exhibited significantly higher National Institutes of Health Stroke Scale (NIHSS) scores at admission, higher pneumonia incidence, and worse functional outcomes at 90 days.[1]

Meta-analysis of 40 studies (43 observations) from Asian stroke populations yielded a pooled dysphagia occurrence rate of 36.3% (95% CI 33.3%–39.3%).[3] A population-based cross-sectional survey in China (5,943 participants across 14 provinces) identified dysphagia prevalence of 51.14% among stroke patients, compared to 19.2% in healthy older adults ≥65 years. Multivariable analysis confirmed that stroke patients were significantly more likely to exhibit dysphagia (odds ratio 2.27, p<0.01) compared to age-matched controls.[2]

Figure 3. Prevalence of oropharyngeal dysphagia across stroke and related populations. Bar chart comparing dysphagia rates in acute stroke patients (36.3–51.1%), healthy elderly ≥65 years (19.2%), neurodegenerative disease patients (48.3%), and head/neck cancer patients (34.4%).

Functional Oral Intake Stratification and Prognostic Associations

The Functional Oral Intake Scale (FOIS) stratifies swallowing capacity into seven levels from tube-feeding dependency (FOIS 1–3) through oral feeding requiring food consistency modifications (FOIS 4–5) to normal oral intake without restrictions (FOIS 6–7). In the prospective Stroke Unit cohort, FOIS 6–7 was a protective factor against disability (modified Rankin Scale ≥3): OR 0.17 (95% CI 0.005–0.56, p=0.004). Tube feeding use at discharge significantly increased risk of disability (mRS ≥3): OR 14.97 (95% CI 2.68–83.65, p=0.002) and 90-day mortality: OR 9.79 (95% CI 2.21–43.4, p=0.003).[1]

Nutritional Consequences and Percutaneous Endoscopic Gastrostomy Outcomes

A retrospective study of 158 patients with post-stroke dysphagia requiring PEG documented high prevalence of malnutrition markers at admission: 41.6% had low body mass index, 62.3% had low serum albumin, 68.6% had low serum transferrin, and 59.6% had low serum cholesterol. Low albumin and transferrin were more prevalent in patients who underwent PEG >2 months after stroke, suggesting progressive nutritional decline. Mortality remained high: 12.9% at 1 month, 27.7% at 3 months, and 40% at 12 months post-PEG.[33]

Implications for Oral Anticoagulant Administration

For dysphagic stroke patients requiring oral anticoagulation, conventional film-coated tablets present multiple challenges: aspiration risk, crushing and bioavailability uncertainty, adherence barriers, and onset delays in emergency settings. In the Chinese dysphagia survey, dysphagic participants were at significantly greater risk of malnutrition (OR 1.91, p<0.01),[2] and malnutrition is independently associated with medication non-adherence.

Apixaban: Pharmacological Profile and Biopharmaceutical Properties Relevant to Orodispersible Tablet Development

Mechanism of Action and Clinical Pharmacology

Apixaban (brand name Eliquis®) is a selective, reversible inhibitor of factor Xa (FXa), a serine protease enzyme in the coagulation cascade that converts prothrombin to thrombin.[14] By inhibiting FXa, apixaban reduces thrombin generation and fibrin clot formation without requiring antithrombin III as a cofactor and without affecting existing thrombin activity. Apixaban is indicated for stroke and systemic embolism prophylaxis in patients with nonvalvular atrial fibrillation; treatment of deep vein thrombosis (DVT) and pulmonary embolism (PE); prophylaxis of recurrent DVT and PE; and prophylaxis of VTE in patients undergoing hip or knee replacement surgery.[4][5][14]

The recommended dosing for stroke prevention in atrial fibrillation is 5 mg twice daily, with dose reduction to 2.5 mg twice daily in patients meeting at least two of the following criteria: age ≥80 years, body weight ≤60 kg, or serum creatinine ≥1.5 mg/dL.[5]

Pharmacokinetics and ADME Properties

Apixaban is rapidly absorbed following oral administration, with time to maximum plasma concentration (Tmax) occurring at 3–4 hours. Absolute oral bioavailability is approximately 50%. Food does not have a clinically meaningful impact on bioavailability, enabling flexible administration with or without meals – a feature advantageous for dysphagic patients requiring texture-modified diets or enteral feeding. Apixaban has an elimination half-life of approximately 12 hours, supporting twice-daily dosing.[15]

Apixaban is metabolised primarily via CYP3A4/5, with minor contributions from CYP1A2, CYP2C8, CYP2C9, CYP2C19, and CYP2J2. This metabolic pathway necessitates attention to drug–drug interactions with strong CYP3A4 inhibitors (e.g., ketoconazole, ritonavir) and inducers (e.g., rifampin, carbamazepine).[14]

Physicochemical Properties and BCS Classification

Apixaban has the molecular formula C₂₅H₂₅N₅O₄ and a molecular weight of 459.5 g/mol.[34][9] Aqueous solubility is reported as 40 µg/mL (0.04 mg/mL) across the physiological pH range, classifying apixaban as poorly water-soluble.[34][9] Caco-2 apparent permeability (Papp) is reported as 0.9 × 10⁻⁶ cm/s. Based on low solubility and variable permeability data, apixaban is classified as BCS Class II or Class IV.[9][10]

Apixaban is described as a white to pale yellow non-hygroscopic crystalline powder. However, forced degradation studies demonstrated that apixaban exhibits high degradability under both acidic and basic conditions, with a pH-independent hydrolysis degradation impurity identified. This chemical instability profile necessitates careful excipient selection (avoiding acidic or basic pH modifiers) and moisture control during ODT manufacturing.[8][37]

 

 

 

Table 1. Physicochemical and Pharmacokinetic Properties of Apixaban

Formulation Strategies for Apixaban Orodispersible Tablets: Manufacturing Methods, Excipient Selection, and Critical Process Parameters

Manufacturing Methods: Decision Criteria and Mechanism-Oriented Comparison

Orodispersible tablets can be manufactured via three primary approaches: direct compression, lyophilisation (freeze-drying), and sublimation. Each method offers distinct advantages and constraints for apixaban ODT development, informed by the drug's biopharmaceutical properties (low solubility, non-hygroscopic, chemically labile under acidic/basic conditions).[38][16]

Direct Compression

Direct compression involves blending drug and excipients directly and compressing into tablets without granulation or complex post-processing. Its advantages for apixaban include simplicity and scalability, cost-effectiveness, moisture avoidance (no aqueous granulation, advantageous for apixaban's chemical stability),[8] and extensive regulatory precedent. Studies on fast disintegrating tablets demonstrated that crospovidone combined with sodium starch glycolate yielded optimal disintegration with acceptable hardness.[39]

Lyophilisation (Freeze-Drying)

Freeze-drying creates a three-dimensional porous network with high surface area, enabling rapid disintegration (typically <10 seconds). However, lyophilised tablets are mechanically fragile, moisture-sensitive, and require capital-intensive manufacturing. Apixaban's chemical lability under acidic/basic conditions[8] may be exacerbated by residual water-catalysed hydrolysis during extended freeze-drying cycles.

Sublimation (Preferred Method for Apixaban)

In sublimation, a volatile subliming agent (e.g., camphor, menthol, ammonium bicarbonate) is incorporated into the tablet blend, compressed, and then removed by controlled heating, leaving a porous tablet matrix. The comparative study of atenolol and atorvastatin mouth-disintegrating tablets concluded that sublimation is superior to direct compression and effervescent methods across all evaluated parameters.[16]

Sublimation is the preferred manufacturing method for apixaban based on: moisture avoidance critical for apixaban's acid/base lability;[8] and optimised porosity–hardness balance. A study on quetiapine fumarate ODTs via sublimation achieved 18.66-second disintegration time, 3.5 kg/cm² hardness, and <1% friability using 5% camphor and 3–5% Indion 414 super disintegrant, with sublimation at 50°C for 6 hours.[38]

Figure 4. Comparative schematic of three ODT manufacturing methods: (A) Direct compression; (B) Lyophilisation; (C) Sublimation. Diagrams illustrate process flow, equipment requirements, and resulting tablet microstructure.

Excipient Selection: Functional Roles, Concentration Ranges, and Apixaban-Specific Considerations

Super Disintegrants

Crospovidone (Polyplasdone XL, Kollidon CL) exhibits rapid swelling without gel formation. A comparative study of super disintegrants at 2% w/w demonstrated that crospovidone achieved the fastest disintegration and dissolution kinetics,[40] making it the preferred super disintegrant for apixaban ODTs at 3–6% w/w. Croscarmellose sodium (Ac-Di-Sol) swells 4–8 times its original volume in water at 2–5% w/w[40] while sodium starch glycolate requires 4–8% w/w for comparable effect.[39][41]

Fillers/Diluents

Microcrystalline cellulose (MCC, Avicel PH-102) is the preferred primary filler. A study on amorphous solid dispersion tablets demonstrated that MCC is preferred over mannitol for moisture-sensitive drugs because MCC's hydrophobic nature minimises recrystallisation risk.[42] Mannitol (Pearlitol SD-200) can be used at 10–20% w/w in combination with MCC for taste improvement. Lactose is not recommended for apixaban ODTs due to recrystallisation concerns and lactose intolerance contraindication.[42]

Binders

Polyvinylpyrrolidone (PVP K-30) serves a dual role as binder and solubiliser. A study on lamotrigine ODTs (a BCS Class II drug analogous to apixaban) used PVP K-30 in solid dispersions combined with super disintegrants to achieve 15-second disintegration and >80% drug release in 10 minutes.[41] HPMC has been used in apixaban formulation studies for solubility enhancement.[10]

Subliming Agents

Camphor exhibits sublimation at 50–60°C; optimal concentration 5–15% w/w. A 3² factorial design study demonstrated that porosity increased from 12.76% to 41.01% as camphor concentration increased from 3% to 15%, with optimal concentration identified as 5–7% w/w to balance rapid disintegration (18.66 seconds) with acceptable friability (0.33%).[38][16]

Table 2. Excipient Selection for Apixaban Orodispersible Tablets: Functional Roles and Recommended Concentrations

 

Critical Process Parameters and Their Impact on Tablet Quality Attributes

Critical process parameters (CPPs) are manufacturing variables that must be controlled within defined ranges to consistently achieve target critical quality attributes (CQAs). A Box-Behnken optimisation study of carbamazepine ODTs identified compression pressure as the dominant factor affecting friability, disintegration behaviour, and drug release.[18]

Optimal compression force for ODTs is typically 10–15 kN (equivalent to 50–100 N hardness).[38][18] The sublimation temperature and duration must be validated: camphor sublimation at 50°C for 6 hours achieves complete removal (validated by GC) without drug degradation.[38] Super disintegrant concentration (3–6% w/w for crospovidone) and blending/lubrication time (<3 minutes) are the remaining CPPs governing disintegration and mechanical integrity.[17]

Figure 5. Impact of critical process parameters on apixaban ODT quality attributes. Multi-panel chart showing: (A) Compression force vs. hardness and disintegration time; (B) Camphor concentration vs. porosity and friability; (C) Super disintegrant concentration vs. disintegration time; (D) Lubrication time vs. disintegration time.

Quality Attribute Evaluation and Clinical Relevance for Stroke Prevention

Weight Variation and Content Uniformity

For apixaban 2.5 mg and 5 mg ODTs (both <25 mg), content uniformity testing per USP <905> is mandatory. Anticoagulants have narrow therapeutic indices; dose variability directly impacts bleeding risk (overdosing) or thromboembolic event risk (underdosing). Content uniformity ensures that each tablet delivers 95–105% of label claim (USP specification: 85–115% with acceptance value ≤15.0).[43]

Hardness and Friability

Acceptable hardness range for ODTs is 50–100 N (equivalent to 3–5 kg/cm²).[38][18] Friability specification per USP <1216> is <1% mass loss after 100 rotations at 25 rpm. For stroke patients with limited manual dexterity (hemiparesis, tremor), tablets must withstand handling without crumbling. Studies on ODTs consistently achieve friability <1% with optimised formulations.[38][17][18]

Wetting Time and Disintegration Time

Wetting time for ODTs should typically be <30 seconds and correlates with in vivo disintegration onset velocity.[38] Disintegration time (USP <701>) specification for ODTs is typically <30 seconds; some formulations target <10 seconds for lyophilised tablets. Disintegration time is the most critical quality attribute for dysphagic patients: ODTs disintegrate in the oral cavity within seconds, enabling swallowing of a drug–saliva suspension rather than a solid bolus, drastically reducing aspiration risk. A study on carbamazepine ODTs optimised via QbD achieved disintegration times as low as 18.66 seconds with acceptable hardness and friability.[18]

 

In Vitro Dissolution and Release Kinetics

Testing conditions for apixaban ODTs should include pH 1.2 (0.1 N HCl), pH 4.5 (acetate buffer), pH 6.8 (phosphate buffer), and simulated intestinal fluid (FaSSIF); USP Apparatus II (paddle); 50–75 rpm; 37°C; with acceptance criteria of >80% release within 30 minutes.[44] For BCS Class IV drugs (or borderline Class II/IV like apixaban), combined dissolution-permeation apparatus (e.g., Transwell with Caco-2 monolayers) can assess simultaneous dissolution and intestinal permeability, improving in vitro-in vivo correlation (IVIVC).[44]

Stability Testing Under ICH Conditions

Stability testing per ICH Q1A: Long-term at 25°C ± 2°C / 60% RH ± 5% for 24 months; accelerated at 40°C ± 2°C / 75% RH ± 5% for 6 months; stress testing including oxidative (H₂O₂), photostability (ICH Q1B), and hydrolysis (acidic, basic, neutral aqueous media). Apixaban exhibits high degradability under acidic and basic conditions,[8] necessitating neutral pH excipients and moisture protection. Stability data demonstrating ≤5% degradation under ICH conditions and maintained disintegration time (<30 seconds at 24 months) are critical for regulatory approval and patient safety.

Artificial Intelligence-Driven Quality by Design for Apixaban ODT Development

Limitations of Traditional OFAT and DoE Approaches

Conventional pharmaceutical development employs one-factor-at-a-time (OFAT) experimentation, varying a single formulation or process variable while holding others constant. Design of Experiments (DoE) methods – factorial designs, central composite designs, Box-Behnken designs – address OFAT limitations by systematically varying multiple factors simultaneously.[21][45][18]

However, DoE has constraints: second-order polynomials cannot adequately represent highly non-linear relationships between formulation variables and quality attributes; DoE is typically constrained to 3–5 independent variables; and DoE identifies a single optimal formulation rather than a robust design space with multiple acceptable solutions.[21]

Quality Target Product Profile and Critical Quality Attributes

The Quality Target Product Profile (QTPP) for apixaban ODTs must specify: Indication – stroke prevention in atrial fibrillation; VTE treatment/prophylaxis.[46] Dosage form: Orodispersible tablet; Strength: 2.5 mg, 5 mg; Disintegration time: <30 seconds in simulated saliva or water at 37°C; Bioavailability: equivalent to reference film-coated tablets (via bioequivalence study); Stability: ≥24 months at 25°C/60% RH; Palatability: acceptable taste and mouthfeel.[19]

Critical Quality Attributes (CQAs) include: disintegration time (≤30 s; target ≤20 s for optimal patient acceptance),[19] dissolution profile (minimum 80% dissolved in 30 minutes; f2 similarity factor ≥50 compared to reference product),[47][48] content uniformity (90.0–110.0%; USP <905>),[43] mechanical properties (hardness 40–100 N; friability <1.0%),[49][50] moisture content (2.0–4.0%),[51] particle size distribution (API D50: 5–15 µm or nanocrystal 232 ± 23 nm),[48][52] taste-masking performance, and stability (shelf life ≥24 months).[52]

Critical Material Attributes (CMAs)

Systematic identification of CMAs was conducted using Failure Mode and Effects Analysis (FMEA). The following CMAs are ranked by Risk Priority Number (RPN) based on severity, occurrence, and detectability:[51][53]

Table 3. Critical Material Attributes for Apixaban ODT

 

Critical Process Parameters (CPPs)

Process parameters were identified through systematic risk assessment and DoE approaches. The following CPPs are defined for direct compression manufacturing.[51][54][55]

 

 

 

 

Table 4. Critical Process Parameters for Apixaban ODT Direct Compression Manufacturing

 

Artificial Neural Network Modelling for Formulation Optimisation

Artificial neural networks (ANNs) have emerged as the predominant machine learning tool for pharmaceutical formulation optimisation due to superior capability in modelling nonlinear relationships between formulation composition, processing conditions, and product performance.[56]

For apixaban ODT development, ANN implementation follows a structured workflow aligned with FDA's 7-step risk-based credibility assessment framework.[13] Network Architecture: single hidden layer with 5–10 nodes using hyperbolic tangent activation functions, as demonstrated in clinically validated pharmaceutical ANN applications achieving R² >0.94 for dissolution and bioavailability predictions.[57]

Data Splitting Strategy: training set 67% of data for model fitting; validation set 33% for independent performance assessment (random holdback);[57] external test set 10% from confirmatory runs or pilot-scale batches. Overfitting Control Mechanisms include: early stopping when validation error increases for consecutive epochs;[56] optimal hidden layer selection through systematic evaluation of 3–15 hidden nodes.[56]

Performance Metrics: R² >0.90 for training, >0.85 for validation;[57][56] Q² (cross-validated R²) >0.80 for predictive validity;[51] RMSE <5% of response range.[58][59][60]

SHAP (SHapley Additive exPlanations) analysis for model interpretability enables identification of most influential formulation variables, detection of interaction effects, and transparent communication of model predictions to regulatory reviewers.[61]

Real-Time Release Testing (RTRT): ANN-based prediction of disintegration time from in-process CMAs enables release decisions without destructive testing. Final prediction equation DT = 34.09 + 2×D_c + 3.59×%H − 5.29×%H×D_c achieved R² = 0.9017 with average relative prediction error of 3.69%.[58]

Genetic Algorithm Optimisation for Multi-Objective Formulation Design

Genetic algorithms (GAs) are population-based evolutionary optimisation techniques employed for exploring high-dimensional formulation design spaces and resolving competing CQA objectives.[56][62]

For apixaban ODTs, GA optimisation addresses inherent trade-offs between rapid disintegration, mechanical robustness, dissolution enhancement, and moisture stability. The Pareto Dominance principle states that a solution x₁ dominates x₂ if x₁ is no worse than x₂ in all objectives and strictly better in at least one. Multi-objective evolutionary algorithms (MOEAs) generate a set of non-dominated solutions representing optimal trade-offs.[62][63]

The Desirability Function approach transforms multi-objective problems into a single composite score: D = (∏ dᵢwᵢ)^(1/Σwᵢ), where dᵢ = individual desirability for objective i (0 = unacceptable, 1 = ideal) and wᵢ = weight reflecting relative importance. For apixaban ODTs, regulatory-critical objectives (disintegration ≤30s, friability ≤1%) receive higher weights (w=3–5), while cost optimisation receives lower weight (w=1).[62][54]

GA Implementation includes: population size 50–100 candidate formulations; single-point or uniform crossover with probability Pc=0.7–0.9; Gaussian or polynomial mutation with probability Pm=0.01–0.05; termination at maximum 100–500 generations or convergence (Pareto front hypervolume change <1% over 20 generations). Surrogate-Based Acceleration: ANNs serve as computationally efficient surrogates, with initial DoE (40 experiments) → ANN training → GA optimisation (200 generations × 100 population = 20,000 evaluations) → validation of top 10 Pareto-optimal formulations.[56]

Artificial Intelligence-Assisted Regulatory Documentation for Apixaban Orodispersible Tablets

Electronic Common Technical Document Structure and AI Content Integration

The eCTD format, harmonised under ICH M4, organises regulatory submissions into five modules: Module 1 (regional administrative information), Module 2 (CTD summaries), Module 3 (Quality–CMC), Module 4 (Nonclinical study reports), Module 5 (Clinical study reports).[64][65]

For apixaban ODT submissions, Modules 2, 3, and 5 contain AI-optimised formulation data and require specific documentation of AI credibility assessment. The integration of AI into pharmaceutical development necessitates transparent documentation of AI model development, validation, and deployment within regulatory submissions to ensure credibility and compliance with data integrity requirements.[66][13]

eCTD Module 2: Quality Overall Summary and Integration of AI-Derived Data

Module 2.3.S documents drug substance general information (nomenclature, physicochemical properties including polymorphic screening via XRPD and DSC, solubility studies, particle size distribution), manufacture, characterisation, and control of drug substance. AI-assisted drug–excipient compatibility screening via DE-INTERACT tool (3,500+ instance ANN model with training accuracy 0.9930, validation accuracy 0.9161) should be documented with model architecture, validation results, and regulatory justification.[67]

Module 2.3.P documents the drug product, with the pharmaceutical development section (2.3.P.2) being the core of QbD documentation. This section must comprehensively describe: Quality Target Product Profile (QTPP) and Critical Quality Attributes (CQAs); risk assessment (FMEA and Ishikawa diagram) and CMA/CPP identification;[53] Design of Experiments and design space establishment with ANN model achieving R² >0.86, Q² >0.64 for all CQA responses;[51] AI model credibility assessment per FDA 7-step framework[13] with demonstrated R² = 0.94 for disintegration prediction and average relative error 4.1% on pilot batches;[68] and genetic algorithm multi-objective optimisation with Pareto front analysis identifying non-dominated formulations and final formulation selected with desirability function D = 0.82.[62][54]

eCTD Module 3: Quality Documentation (CMC)

Module 3.2.S contains extended characterisation data, complete spectroscopic data (¹H-NMR, ¹³C-NMR, IR, UV, MS), and forced degradation studies per ICH Q1B.[8]

Module 3.2.P.2 (Pharmaceutical Development) must include: the complete DoE design matrix and raw data; ANN model documentation specifying architecture (single hidden layer, 7 nodes, hyperbolic tangent activation),[57] training data characterisation, model training and validation performance statistics (training R² = 0.946 for disintegration, 0.921 for dissolution; RMSE: disintegration 2.1 s, dissolution 3.8%), external validation on 5 pilot-scale batches (average relative error 4.1%);[68] variable importance ranking via connection weight analysis;[56] and GA optimisation record with Pareto frontier analysis and validated GA-optimised formulation (disintegration 18.5 ± 1.2 s, dissolution 86.3 ± 2.1% at 30 min, hardness 68.4 ± 3.5 N, friability 0.73 ± 0.08%).[54]

Stage 3 (Continued Process Verification): AI-powered CPV framework with Isolation Forest anomaly detection model and random forest predictive control model applied to batch data to identify deviations from normal operating conditions.[69]

 

 

 

 

Table 5. Drug Product Specification – Apixaban Orodispersible Tablets

 

eCTD Module 5: Bioequivalence Strategy

For apixaban ODTs, bioequivalence (BE) to the reference film-coated tablet formulation (Eliquis®) is the regulatory pathway for approval. The BE study design follows FDA and EMA guidance on orally disintegrating dosage forms: randomised, open-label, single-dose, two-period, two-sequence crossover study under fasting conditions; healthy adult volunteers (n=24–36); 90% confidence intervals for geometric mean ratios (Test/Reference) of AUC₀₋ₜ, AUC₀₋∞, and Cmax must fall within 80.00–125.00%.[70][71]

BCS-based biowaiver: apixaban is BCS Class II (low solubility, high permeability); biowaivers are not applicable per FDA M9 guidance. AI-assisted PBPK models integrated with AI predict bioavailability based on dissolution profile, particle size, and formulation composition; models inform go/no-go decisions before expensive clinical trials.[72][52]

Data Integrity, ALCOA+ Compliance, and AI Model Governance

ALCOA+ defines the data integrity framework required for GMP compliance and regulatory inspection readiness.[73][74] All AI model training data, predictions, and decisions must be linked to specific user IDs, timestamps, and batch records via electronic laboratory notebooks (ELN) with audit trails per 21 CFR Part 11.[75]

AI Model Lifecycle Management per ICH Q12: Established Conditions (ECs) for the ANN model include model architecture, training dataset composition, and validation acceptance criteria. Product Lifecycle Management (PLM) specifies change categories: (1) model retraining with expanded dataset (low risk, company discretion); (2) architecture modification (moderate risk, Prior Approval Supplement); (3) change in context of use (high risk, Prior Approval Supplement with clinical data if applicable).[76]

Table 6. Regulatory Compliance Checklist for AI-Generated Dossiers

Category	Requirement	Compliance Evidence	Reference
Governance	AI use disclosed in Module 2.3.R Regional Information	AI tools (ANN, GA) explicitly described in QOS pharmaceutical development section	FDA 7-step framework [13]
Governance	AI model context of use clearly defined	COU statement: "ANN predicts CQAs from CMAs/CPPs; serves as formulation screening tool supplemented by confirmatory experiments"	FDA 7-step framework Step 2 [13]
Data Integrity	Audit trail for AI-generated data per 21 CFR Part 11	Electronic laboratory notebooks with immutable timestamps; all training data traceable to source experiments	21 CFR Part 11 [75]
Validation	External validation on independent batches	R² >0.85 on validation set; average relative error 4.1% on 5 pilot batches [68]	FDA 7-step framework Steps 4-6 [13]
Lifecycle	Post-approval change control for AI model updates	ICH Q12 PLM document specifying change categories, risk assessment, and regulatory notification pathways	ICH Q12 [76]
EMA Alignment	GMP compliance for AI in manufacturing	Annex 22 (draft) compliance roadmap; human oversight documentation; model validation per proposed GMP guidelines	EMA Annex 22 [12]

 

CONCLUSIONS

The development of apixaban orodispersible tablets (ODTs) represents a critical advancement in addressing unmet therapeutic needs across three vulnerable patient populations: dysphagic patients (post-stroke, neurological disorders, head-and-neck cancer), geriatric patients with polypharmacy-related swallowing difficulties, and paediatric patients requiring weight-based anticoagulation.[77][78]

Current apixaban film-coated tablets present significant adherence barriers in these populations, contributing to suboptimal stroke prevention outcomes despite robust clinical efficacy demonstrated in large-scale trials.[79] The ODT formulation strategy addresses multiple patient-centric challenges simultaneously: elimination of water requirement for administration, rapid oral disintegration (target <30 seconds), pleasant organoleptic properties through effective taste masking, and maintenance of immediate-release bioequivalence to reference tablets to preserve established pharmacokinetic and safety profiles.

From a formulation science perspective, apixaban's classification as a BCS Class II drug – characterised by low aqueous solubility (0.04 mg/mL) – necessitates sophisticated approaches.[80][81] The integration of QbD principles provides the systematic framework to define QTPP parameters, identify CQAs including disintegration time, dissolution efficiency, content uniformity, and palatability, and map these to CMAs and CPPs through comprehensive risk assessment tools.[82][83][84]

The integration of AI and machine learning (AI/ML) into QbD workflows – termed AI-enabled Quality by Design (AI-QbD) – represents a transformative paradigm shift in pharmaceutical development efficiency and precision.[85][86][87] The joint FDA-EMA principles on AI in medicine development emphasise human-centric design, transparency, risk-based approaches to AI model validation, and lifecycle management incorporating post-market model updating.[89]

For apixaban ODT development specifically, future AI-QbD innovations may include: (1) in silico prediction of patient acceptability through virtual taste panel modelling; (2) digital twin manufacturing platforms that simulate entire production campaigns before physical scale-up; (3) continuous learning systems where post-market stability and dissolution data automatically refine formulation models under ICH Q12 lifecycle management frameworks; and (4) AI-assisted regulatory submission preparation that auto-generates CTD sections from experimental databases with full traceability.[88][94]

The convergence of AI-enabled formulation development, precision medicine approaches to stroke prevention, and patient-centric drug delivery platforms positions apixaban ODT as an exemplar of next-generation pharmaceutical innovation – where computational intelligence accelerates translation of therapeutic efficacy into real-world clinical benefit across the full spectrum of patients requiring anticoagulation.

ACKNOWLEDGEMENTS

The authors acknowledge the Department of Pharmaceutics, The Oxford College of Pharmacy, Bengaluru, for institutional support. No external funding was received for this work.

REFERENCES

1.        Souza JT, et al. Dysphagia and tube feeding after stroke are associated with poorer functional and mortality outcomes. Clin Nutr. 2020;39(7):2142–2149.

2.        Zhang M, et al. Prevalence of Dysphagia in China: An Epidemiological Survey of 5943 Participants. Dysphagia. 2021;36(2):339–350.

3.        Meng PP, et al. The Occurrence Rate of Swallowing Disorders After Stroke in Asia: A Meta-Analysis. J Stroke Cerebrovasc Dis. 2020;29(11):105113.

4.        U.S. National Library of Medicine. DailyMed. Label: ELIQUIS (apixaban) tablet, film coated. Bethesda, MD: NLM; 2023. Available from: https://dailymed.nlm.nih.gov/dailymed/drugInfo.cfm?setid=fd383ae7-1e34-44ef-b8f1-ca6906dfbcef

5.        Pfizer Inc; Bristol-Myers Squibb Company. ELIQUIS® (apixaban) Prescribing Information. New York, NY: Pfizer Inc; 2023.

6.         GBD 2019 Stroke Collaborators. Global, regional, and national burden of stroke and its risk factors, 1990–2019. Lancet Neurol. 2021;20(10):795–820.

7.        Feigin VL, et al. World Stroke Organization: Global Stroke Fact Sheet 2022. Int J Stroke. 2022;17(1):18–29.

8.        Surukonti SR, Surendrababu MS. Understanding Stress Testing Characteristics of Apixaban and Stability-Indicating Method Development. J AOAC Int. 2023;106(6):1565–1575.

9.        Sato M, et al. Physiologically Based Biopharmaceutics Model of Apixaban. Pharmaceutics. 2025;17(3):382.

10.    Bansal AK, et al. Solubility enhancement of BCS classified II/IV drug: solid dispersion of apixaban by solvent evaporation. Int J Drug Deliv. 2022;14(2):45–56.

11.    U.S. Food and Drug Administration. Considerations for the use of artificial intelligence to support regulatory decision-making for drug and biological products. Silver Spring, MD: FDA; 2025.

12.    European Commission. EudraLex Volume 4: good manufacturing practice (GMP) guidelines – Chapter 4, Annex 11 and Annex 22. Brussels: European Commission; 2024.

13.    U.S. Food and Drug Administration. Considerations for the use of artificial intelligence to support regulatory decision-making for drug and biological products: guidance for industry. Silver Spring, MD: CDER/CBER; 2025.

14.    DrugBank Online. Apixaban (DB06605): uses, interactions, mechanism of action [Internet]. Edmonton: OMx Personal Health Analytics Inc; 2024. Available from: https://go.drugbank.com/drugs/DB06605

15.    Lip GYH, et al. Apixaban: A Clinical Pharmacokinetic and Pharmacodynamic Review. Clin Pharmacokinet. 2019;58(10):1253–1270.

16.    Sarfraz RM, et al. Formulation and evaluation of mouth disintegrating tablets of atenolol and atorvastatin. Indian J Pharm Sci. 2015;77(1):102–109.

17.    Solaiman A, et al. Application of general multilevel factorial design with formulation of fast disintegrating tablets. Int J Pharm. 2016;501(1–2):31–41.

18.    Mishra SM, Rohera BD. An integrated QbD approach for design, development and optimization of orodispersible tablet formulation of carbamazepine. Pharm Dev Technol. 2017;22(4):448–459.

19.    Chinwala M. Recent Formulation Advances and Therapeutic Usefulness of Orally Disintegrating Tablets (ODTs). Pharmacy (Basel). 2020;8(4):186.

20.    Khan AM, et al. ANN Approach to Predict an Optimized pH-Dependent Mesalamine Matrix Tablet. Drug Des Devel Ther. 2020;14:2637–2646.

21.    Beg S, et al. QbD-based systematic development of novel optimized solid SNEDDS of lovastatin. Drug Deliv. 2015;22(6):765–784.

22.    GMP-Compliance.org. Drafts of EU GMP Guideline Annex 11, Annex 22 and Chapter 4 released for comment [Internet]. 2023. Available from: https://www.gmp-compliance.org/gmp-news/drafts-of-eu-gmp-guideline-annex-11-annex-22-and-chapter-4-released-for-comment

23.    Patel H, et al. QbD Approach for Development of Co-Processed Excipient Pellets by Extrusion-Spheronization. Recent Pat Drug Deliv Formul. 2016;10(2):141–151.

24.    World Health Organization. Mortality and global health estimates. Geneva: WHO; 2021. Available from: https://www.who.int/data/gho/data/themes/mortality-and-global-health-estimates

25.    Sporns PB, et al. Childhood stroke. Nat Rev Dis Primers. 2022;8(1):12.

26.    Krishnamurthi RV, et al. Global, Regional and Country-Specific Burden of Ischaemic Stroke, Intracerebral Haemorrhage and Subarachnoid Haemorrhage: GBD 2017. Neuroepidemiology. 2020;54(2):118–128.

27.    Prokopiv MM, et al. Prospective analysis of cerebrovascular disease epidemiology in Kyiv, Ukraine. Wiad Lek. 2021;74(11):2713–2718.

28.    Amoah D, et al. An international perspective on young stroke incidence and risk factors: a scoping review. BMC Public Health. 2024;24(1):1678.

29.    Perera KS, et al. Evaluating Rates of Recurrent Ischemic Stroke Among Young Adults: The Young ESUS Cohort Study. JAMA Neurol. 2022;79(4):382–391.

30.     Beslow LA, et al. Paediatric Ischemic Stroke: An Infrequent Complication of SARS-CoV-2. Ann Neurol. 2021;89(4):657–665.

31.    Beslow LA, et al. International Prevalence and Mechanisms of SARS-CoV-2 in Childhood Arterial Ischemic Stroke. Stroke. 2022;53(6):1934–1942.

32.    Terré R. Oropharyngeal dysphagia in stroke: diagnostic and therapeutic aspects. Rev Neurol. 2020;71(6):205–215.

33.    Patita M, et al. Endoscopic gastrostomy for nutritional support in post-stroke dysphagia. Nutr Hosp. 2022;39(1):1–8.

34.    National Center for Biotechnology Information. PubChem compound summary for CID 10182969: apixaban [Internet]. Bethesda, MD: NCBI; 2024. Available from: https://pubchem.ncbi.nlm.nih.gov/compound/Apixaban

35.    Islam MS, et al. Enhanced aqueous dissolution of hydrophobic apixaban via direct incorporation of hydrophilic nanographene oxide. Colloids Surf B Biointerfaces. 2022;216:112512.

36.    Alexander SPH, et al. Apixaban. IUPHAR/BPS Guide to PHARMACOLOGY [Internet]. Edinburgh: International Union of Basic and Clinical Pharmacology; 2024. Available from: https://www.guidetopharmacology.org/GRAC/LigandDisplayForward?ligandId=6390

37.    Rathod S, et al. Analytical and pharmaceutical consideration of the anticoagulant drug apixaban: a comprehensive review. Int J Pharm Sci. 2023;13(4):1–15.

38.    Kalyankar P, et al. Formulation Design and Optimization of Orodispersible Tablets of Quetiapine Fumarate by Sublimation Method. Indian J Pharm Sci. 2015;77(4):450–455.

39.    Dave V, et al. Novel approach of aceclofenac fast dissolving tablet. Pak J Pharm Sci. 2015;28(1 Suppl):281–287.

40.    Zimmer Ł, et al. Dissolution properties and kinetic study of sulfadimidine and trimethoprim tablets containing superdisintegrants. Acta Pol Pharm. 2015;72(6):1213–1222.

41.    Singh J, et al. Enhancement of solubility of lamotrigine by solid dispersion and development of orally disintegrating tablets using 3(2) full factorial design. J Pharm. 2015;2015:828453.

42.    Yu D, Hoag SW. Impact of diluents on compaction, dissolution, and physical stability of amorphous solid dispersion tablets. Int J Pharm. 2024;652:123924.

43.    Ortega CA, et al. New Orodispersible Mini Tablets of Enalapril Maleate by Direct Compression for Paediatric Patients. Curr Drug Deliv. 2020;17(6):540–550.

44.    Kádár S, et al. Bioequivalence prediction with small-scale biphasic dissolution and dissolution-permeation apparatus. Eur J Pharm Sci. 2024;197:106782.

45.    Bansal S, et al. QbD-Oriented Development and Characterization of Effervescent Floating-Bioadhesive Tablets of Cefuroxime Axetil. AAPS PharmSciTech. 2016;17(1):85–101.

46.    European Medicines Agency. Eliquis (apixaban): European Public Assessment Report [Internet]. Amsterdam: EMA; 2023. Available from: https://www.ema.europa.eu/en/medicines/human/EPAR/eliquis

47.    Conceição J, et al. Orodispersible Carbamazepine/Hydroxypropyl-β-Cyclodextrin Tablets Obtained by Direct Compression. AAPS PharmSciTech. 2020;21(1):36.

48.    Tashan E, et al. Development of Nanocrystal Ziprasidone Orally Disintegrating Tablets. AAPS PharmSciTech. 2020;21(5):151.

49.    Zhang Y, et al. Development and optimization of levodopa and benzylhydrazine orally disintegrating tablets. Drug Dev Ind Pharm. 2020;46(1):25–34.

50.    Iqbal H, et al. Formulation development and optimization studies of mouth dissolving tablets of tizanidine HCl. Pak J Pharm Sci. 2020;33(2):619–625.

51.    Gavan A, et al. Fluidised bed granulation of two APIs: QbD approach and NIR in-line monitoring. Asian J Pharm Sci. 2021;16(1):93–105.

52.    Patil K, et al. Quality-by-Design-Driven Nanostructured Lipid Scaffold of Apixaban. Pharmaceutics. 2024;16(7):910.

53.    Namjoshi S, et al. Quality by Design: Development of the QTPP for Semisolid Topical Products. Pharmaceutics. 2020;12(3):287.

54.    Khan H, et al. Multi-object optimization of Navy-blue anodic oxidation via response surface models. Chemosphere. 2022;287(Pt 2):132818.

55.    Belayneh A, et al. Formulation and Optimization of Monolithic Fixed-Dose Combination of Metformin HCl and Glibenclamide Orodispersible Tablets. Adv Pharmacol Pharm Sci. 2020;2020:3546597.

56.    Damiati SA. Digital pharmaceutical sciences. AAPS PharmSciTech. 2020;21(6):206.

57.    Simões MF, et al. Artificial neural networks applied to quality-by-design. Eur J Pharm Biopharm. 2020;153:116–134.

58.    Xia CY, et al. Real time release testing of disintegration time of uncoated Tianshu Tablets. China J Chin Materia Med. 2020;45(7):1547–1552.

59.    Mészáros LA, et al. Real-time release testing of in vitro dissolution and blend uniformity by NIR spectroscopy and machine vision. Eur J Pharm Biopharm. 2024;200:114368.

60.    Alexandra Mészáros L, et al. Machine vision-based non-destructive dissolution prediction of meloxicam-containing tablets. Int J Pharm. 2024;655:124013.

61.    Zhang W, et al. Intelligent information management enables quality-by-design in pharmaceutical manufacturing. Sci Rep. 2025;15(1):8742.

62.     Lambrinidis G, Tsantili-Kakoulidou A. Multi-objective optimization methods in novel drug design. Expert Opin Drug Discov. 2021;16(5):561–572.

63.    Gong D, et al. A Meta-Objective Approach for Many-Objective Evolutionary Optimization. Evol Comput. 2019;27(1):1–26.

64.    U.S. Food and Drug Administration. M8 Electronic Common Technical Document (eCTD) v4.0. Silver Spring, MD: FDA; 2021.

65.    International Council for Harmonisation. Guideline for Module 1 of the electronic Common Technical Document (M4) – Regulatory Information. ICH; 2022.

66.    European Medicines Agency. Artificial intelligence [Internet]. Amsterdam: EMA; 2024. Available from: https://www.ema.europa.eu/en/about-us/how-we-work/data-regulation-big-data-other-sources/artificial-intelligence

67.    Patel S, et al. DE-INTERACT: A machine-learning-based predictive tool for drug-excipient interaction. Int J Pharm. 2023;636:122839.

68.    Zhang F, et al. Ensemble Learning Based on Policy Optimization Neural Networks for Capability Assessment. Sensors (Basel). 2021;21(17):5802.

69.    Ondracka A, et al. CPV of the Future: AI-Powered Continued Process Verification. PDA J Pharm Sci Technol. 2022;76(6):537–551.

70.    European Medicines Agency. Apixaban product-specific bioequivalence guidance [Internet]. Amsterdam: EMA; 2022. Available from: https://www.ema.europa.eu/en/apixaban-product-specific-bioequivalence-guidance

71.    Therapeutic Goods Administration. Apixaban film-coated tablet 2.5 and 5 mg product-specific bioequivalence guidance [Internet]. Canberra: TGA; 2022. Available from: https://www.tga.gov.au/resources/resources/international-scientific-guidelines-adopted-australia/apixaban-film-coated-tablet-25-and-5-mg-product-specific-bioequivalence-guidance-0

72.    Patil VM, et al. Design and evaluation of oral formulation for apixaban. AAPS PharmSciTech. 2023;24(7):192.

73.    World Health Organization. WHO Technical Report Series No. 1033: Annex 4 – WHO guideline on data integrity. Geneva: WHO; 2021.

74.    Hu Y, et al. Ensuring data integrity in the pharmaceutical lifecycle. Talanta. 2025;283:127045.

75.     U.S. Food and Drug Administration. Part 11, Electronic Records; Electronic Signatures – Scope and Application. Guidance for Industry. Silver Spring, MD: FDA; 2003.

76.    Tran R, et al. A network of regulatory innovations to improve FDA quality assessments. Int J Pharm X. 2024;7:100239.

77.    Ruiz-Picazo A, et al. Effect of thickener on disintegration, dissolution and permeability of common drug products for elderly patients. Eur J Pharm Biopharm. 2020;153:6–13.

78.    de Freitas Neto JL, et al. Development of Paediatric Orodispersible Tablets Based on Efavirenz. Curr HIV Res. 2020;18(4):269–280.

79.    Durand M, et al. Effectiveness and safety among direct oral anticoagulants in nonvalvular atrial fibrillation. Br J Clin Pharmacol. 2021;87(6):2535–2553.

80.    Subramanian VB, et al. Stability-indicating RP-HPLC method for determination of impurities in apixaban tablet dosage forms. Biomed Chromatogr. 2020;34(2):e4719.

81.    Zarmpi P, et al. Biopharmaceutical implications of excipient variability on drug dissolution from immediate release products. Eur J Pharm Biopharm. 2020;154:195–209.

82.    Mansuri N, et al. Quality by design (QbD) approach to match tablet glossiness. Pharm Dev Technol. 2020;25(7):861–870.

83.    Liu H, et al. Optimization of the manufacturing process of a complex amphotericin B liposomal formulation using quality by design approach. Int J Pharm. 2020;586:119473.

84.    Farooqi S, et al. QbD-Based Numerical and Graphical Optimization Technique for the Development of Osmotic Pump Controlled-Release Metoclopramide HCl Tablets. Drug Des Devel Ther. 2020;14:5599–5616.

85.    Lamberti G, et al. Quality by Digital Design for accelerated sustainable nanomedicine. Eur J Pharm Sci. 2025;204:107239.

86.    Chakraborty S, et al. Leveraging artificial intelligence for advancements in liquid dosage formulations. Pharmaceutics. 2024;16(10):1285.

87.     Li H, et al. Scale-agnostic models based on dimensionless quality by design in pharmaceutical manufacturing. Pharmaceutics. 2025;18(7):1033.

88.      Müller JA, Brock D, Knapp B. Bioprocessing 4.0: a pragmatic review and future perspectives. Digit Discov. 2024;3(8):1509–1528.

89.    European Medicines Agency; U.S. Food and Drug Administration. EMA and FDA set common principles for AI in medicine development. EMA News. 2024.

90.    U.S. Food and Drug Administration. Good machine learning practice for medical device development: guiding principles [Internet]. Silver Spring, MD: FDA; 2021. Available from: https://www.fda.gov/medical-devices/software-medical-device-samd/good-machine-learning-practice-medical-device-development-guiding-principles

91.    U.S. Food and Drug Administration. Guiding principles of good AI practice in drug development [Internet]. Silver Spring, MD: FDA; 2024. Available from: https://www.fda.gov/about-fda/artificial-intelligence-drug-development/guiding-principles-good-ai-practice-drug-development

92.    International Society for Pharmaceutical Engineering. AI algorithm qualification: a quality by design lens on regulatory expectations. Pharm Eng. 2024.

93.    U.S. Food and Drug Administration. Q8, Q9 and Q10 questions and answers – appendix: QAs from training sessions on Q8, Q9, Q10 points to consider. Silver Spring, MD: FDA; 2012.

94.    Singh R, et al. Regulatory perspectives for AI/ML implementation in pharmaceutical GMP settings. Front Pharmacol. 2024;15:1384562.