AI Dosing Algorithms: How They Save Specialty Drugs

PK/PD Models: Time-Tested, Data-Driven

PK/PD models use established mathematical formulas to simulate how a drug is absorbed, distributed, metabolized, and eliminated (PK), and how it interacts with biological targets to produce therapeutic effects (PD). These models often rely on population data and are especially valuable for medications with well-characterized dose–response relationships.

In clinical practice, PK/PD models help clinicians estimate how long a drug stays active in the bloodstream, how it behaves across different patient profiles (e.g., pediatric vs. geriatric), and what dosing regimens are likely to be safe and effective. For example, in oncology or transplant care-where therapeutic windows are narrow-PK/PD modeling is essential for minimizing toxicity while maintaining efficacy.

However, traditional PK/PD approaches can struggle with the high variability and complexity found in real-world patients. They are also less adaptive when new data becomes available during a treatment cycle.

ML-Based Predictive Models: Adaptive, Scalable, and Personalized

Machine learning brings a more flexible and adaptive layer to precision dosing. Rather than relying on pre-set equations, ML models learn patterns from vast datasets, including EMRs, lab values, genetic information, and patient-reported outcomes.

By training on historical cases, ML algorithms can predict how an individual might respond to a particular dose of a drug-even without a deep understanding of the underlying pharmacokinetics. These models continuously improve as more data becomes available, making them particularly useful for complex cases, such as patients with comorbidities or rare genetic variants.

One key advantage is that ML models can be integrated directly into digital health platforms. For instance, telepharmacy AI systems can use real-time input (e.g., weight changes, renal function updates) to suggest dose adjustments without requiring an in-person consult. This is especially valuable in remote care settings, where clinical pharmacists can monitor and optimize dosing protocols virtually.

Combining the Two: Hybrid Approaches

Increasingly, healthcare systems are turning to hybrid models that combine PK/PD frameworks with machine learning. These systems offer the best of both worlds: the transparency and interpretability of classical pharmacology and the personalization and adaptability of AI.

Hybrid engines-such as CURATE.AI and other “model-informed precision-dosing” (MIPD) platforms-now adjust recommended doses in near-real-time as new lab or sensor data flow in.

In March 2025, the prospective PRECISE CURATE.AI study (Nature Cancer) reported 97% clinician acceptance of algorithm-guided dosing and an average ~20% reduction in capecitabine exposure without loss of efficacy, underscoring the clinical viability of these hybrid engines.

Such combined models are being explored in areas like chemotherapy optimization, where AI tools can refine baseline PK predictions using live clinical data to continuously calibrate dosing.

Oncology: Optimizing Costly Therapies, Minimizing Toxicity

Cancer treatment often involves specialty drugs such as monoclonal antibodies, targeted therapies, and immunotherapies-all of which come with high price tags and narrow therapeutic margins. Incorrect dosing, whether too high or too low, can result in suboptimal outcomes, unnecessary side effects, and massive drug waste.

AI-powered dosing algorithms help oncologists tailor drug regimens based on a variety of patient-specific inputs: liver and kidney function, tumor markers, genetic mutations, and even immune profiles. For example, machine learning models can detect early signs of toxicity or resistance, enabling clinicians to fine-tune doses or switch regimens proactively-long before traditional methods would flag a problem.

In multiple single-center pilots, ML-guided chemotherapy protocols have lowered drug use by approximately 5–15% per patient without compromising response rates or survival; multi-center randomized trials are still pending.

Hemophilia: Precision at the Molecular Level

In hemophilia care, the stakes are equally high. Patients require regular infusions of clotting factor concentrates-some of the most expensive specialty drugs on the market. The challenge lies in maintaining just the right level of clotting factors in the bloodstream. Too little increases bleeding risk; too much results in waste and potential complications.

AI algorithms, particularly those incorporating PK/PD models enhanced by real-time data, can predict how quickly a patient metabolizes clotting factor, allowing for highly customized dosing schedules. Instead of a fixed number of units per kilogram of body weight, the algorithm adjusts based on observed usage patterns, activity levels, and metabolic feedback.

Recent prospective studies using PK-guided tools such as myPKFiT® report factor-concentrate savings of roughly 5–10%, with equal or improved bleeding control.

Telepharmacy Integration: Extending Reach, Lowering Costs

The combination of AI dosing with telepharmacy workflows further amplifies the value. For example, remote clinical pharmacists using AI-driven dashboards can monitor patient adherence, lab results, and response data in real time. These systems can recommend dosage adjustments or flag anomalies without requiring in-person visits.

Integrated telepharmacy AI platforms—like those envisioned in Telepharmacy 2.0 End-to-end workflow from eRx to delivery—are especially useful in rural or underserved areas, where access to specialists is limited. By embedding dosing intelligence into remote care platforms, healthcare providers can reduce both travel costs and unnecessary hospitalizations due to dosing errors.

Impact of Gene Therapy on Dosing Economics

The landscape in hemophilia is shifting again after FDA approvals of one-time gene therapies such as Hemgenix® (2022) and Beqvez™ (2024). These ultra-expensive but potentially curative infusions can eliminate-or greatly reduce-the need for lifelong clotting-factor dosing, altering the cost-saving calculus for AI-guided factor replacement. Precision-dosing algorithms will likely pivot toward identifying residual-therapy candidates and post-gene-therapy monitoring.

In addition, the FDA approved Roctavian® (valoctocogene roxaparvovec) for hemophilia A on 30 June 2023, further expanding one-and-done gene-therapy options and reshaping long-term dosing economics.

Data Gaps and Algorithmic Reliability

AI dosing algorithms are only as good as the data they rely on. In many healthcare systems, electronic medical records (EMRs) are incomplete, inconsistent, or lack standardization. Missing data on lab values, genetic markers, or medication adherence can significantly skew model outputs, leading to inaccurate dose recommendations.

Additionally, some machine learning models operate as “black boxes,” providing little transparency into how they arrive at dosing decisions. This opacity makes it difficult for clinicians to trust the recommendations, especially in high-stakes cases like oncology or pediatrics.

To build confidence, it’s crucial that models undergo rigorous validation across diverse patient populations and are regularly updated with real-world data. Regulatory agencies such as the FDA and EMA have begun developing frameworks for evaluating AI in clinical decision support-but comprehensive, universally accepted guidelines are still evolving.

Responsibility and Clinical Oversight

AI dosing introduces a critical question: Who is responsible if something goes wrong? If a patient suffers an adverse event due to a model-generated dosing recommendation, accountability can become murky. Is it the software developer, the clinician who followed the suggestion, or the health system that deployed the tool?

To mitigate risk, AI dosing tools should be positioned as decision support systems, not replacements for clinical judgment. Human oversight remains essential, particularly for interpreting AI outputs in context with a patient’s broader medical history.

Ethical Use of AI in Vulnerable Populations

Another concern involves equity. If AI models are trained primarily on data from well-funded hospitals or specific demographics, they may underperform in underrepresented populations-such as racial minorities, rural patients, or those with rare conditions. This could inadvertently widen existing healthcare disparities.

Furthermore, cost-saving potential must be balanced against patient safety. There’s a risk that healthcare providers or insurers might prioritize cost reduction pharma over clinical outcomes if AI dosing appears to save money on paper, even when individual cases require more nuanced care.

Ethical use of AI dosing means placing patient benefit above budgetary concerns and ensuring that algorithms are inclusive, explainable, and continuously monitored for bias or drift.

Regulatory Frameworks

• On 9 January 2025, the FDA released its draft guidance “Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management,” outlining dossier expectations, pre-determined change-control plans, and post-market performance monitoring for AI-driven dosing software.
• In the European Union, the AI Act (Regulation (EU) 2024/1689) entered into force on 12 July 2024. High-risk medical-AI systems—including dosing decision-support software—must now meet new risk-management, transparency, and human-oversight requirements (some provisions apply from February 2025; full compliance is required by August 2026).

Let’s Make It Actionable

• Get in touch with our healthcare innovation team.
• Subscribe to our insights newsletter for the latest updates on ML precision dosing and telepharmacy AI.