The Time is Now: Model-Based Dosing to Optimize Drug Therapy

Keywords::

• model-based
• personalized medicine
• pharmacokinetics
• pharmacometrics

Historical perspective

The underpinnings for model-based dosing to support clinical decisions were laid in the late 1960s [1,2]. A seminal paper in 1977 further laid down the principles and methods required to implement individualized dosing into routine patient care [3]. Although these concepts have been utilized with significant impact in advancing model-informed drug development over the last 40 years, little progress has been made in advancing clinical therapeutics on an individual patient level. However, the tide is changing and many of the intellectual and technological barriers are no longer impeding the progress toward a more data-driven and model-informed individualized dosing paradigm for routine clinical care.

The challenge for model-based dosing lies in the idea that drug development programs are often incentivized to devise dosing recommendations in the product label primarily for ease of prescribing. This extends to the dosing guidelines for special populations where the label specifies dosing based on certain cut-points for prognostic factors such as age, weight, kidney function or disease status. Historically, the primary reason for such simplified label recommendations was to facilitate approval for the average patient. This, however, is no longer the case as advances in quantitative clinical sciences (e.g., pharmacometrics) and technology enable an individualized approach to drug therapy. For much therapeutics this signifies an important paradigm shift from a predefined dose to a more tailored and personalized dose aimed to increase efficacy and reduce toxicity. Focus should now turn to multidisciplinary, interprofessional education for future trainees (both professionals and graduates) with modern tools designed to support clinical decision-making for individualized therapy.

The utility of model-based dosing & example of busulfan

Patient, disease and drug characteristics define the need and utility of model-based dosing for drugs [4]. Factors for which model-based dosing can be particularly impactful include complex patient populations such as pediatrics, drugs with a narrow therapeutic index and drugs with significant association with morbidity and mortality [5–8]. Examples of model-based dosing to address clinically relevant challenges currently exist in different therapeutic areas. However their large-scale implementation into routine clinical practice has yet to be achieved.

The overarching goal of model-based dosing is to effectively treat diseases without acute toxicity and to prevent long-term side effects of drug therapy. For example, in pediatrics it is well-recognized that the pharmacokinetics (PK) and pharmacodynamics of drugs in infants can differ widely between children and adults [9]. Within the first year of life, age-related developmental changes in physiologic and metabolic processes can lead to significantly altered drug disposition [10]. Additionally, the relationship between dose, plasma concentration and pharmacodynamics effect may be highly variable across different age groups and disease states. An example of how model-based dosing can be applied to patient care is the use of busulfan, an alkylating agent, commonly used in the setting of pediatric autologous and allogeneic hematopoietic cell transplantation (HCT).

Therapeutic drug monitoring (TDM) is routinely performed in children receiving busulfan as a part of high-dose chemotherapy prior to HCT to optimize systemic exposure. Several strategies for monitoring drug levels exist, including the estimation of steady-state concentration (Css) or cumulative area-under-the-curve (cAUC) over the entire course of therapy. Achievement of optimal busulfan exposure is necessary to promote engraftment and limit drug-related toxicity, including severe mucositis and sinusoidal obstruction syndrome. In the setting of pediatric HCT, a cAUC of 60–90 mg*h/l has been shown to be optimal for a variety of both malignant and non-malignant disorders [11–13]. However, the therapeutic target may differ between individuals based on several factors including stem cell source, degree of donor match, disease (malignant vs non-malignant) and other immunosuppressive agents included in conditioning such as serotherapy [11,14,15].

A population PK model that relates systemic clearance of busulfan with patient-specific factors and clinical covariates allows this fine-tuning of exposure via TDM. Such models with patient-specific body size and age-related metabolic maturation have shown to improve targeted therapy in infants and children when compared with the stratified weight or age-based regimens alone. However, strategies for the TDM of busulfan differ among treatment centers, as some use Css and others use cAUC as the exposure metric. The frequency of dosing is also highly variable (every 6h, 12 h or once daily).

The conventional dosing for busulfan (administered every 6 h) provided by the manufacturer recommends an initial dose of 1.1 mg/kg for patients weighing ≤12 kg, and 0.8 mg/kg/dose for patients weighing >12 kg, regardless of age. Unfortunately, these guidelines do not allow for different dosing frequencies or targeted exposures. In addition, it has been estimated that following the first dose of busulfan only 60% of children will fall within the predefined, goal range for exposure when the conventional dose algorithm is applied [16]. These percentages are highly variable among children of different ages, particularly in those less than 2 years of age.

Offering a significant advantage over the conventional guidelines, model-based dosing algorithm can be used to individualize and target therapy. Given the short duration of busulfan therapy, achievement of individualized drug exposure early on in treatment, preferably with the first dose, is crucial and failure to do so may lead to suboptimal therapy or toxicity [17]. Additionally, the ability for most centers to measure busulfan levels on site is not routinely available therefore limiting the ability to perform early dose-modifications. With recent advancements in technology and modeling tools, especially Bayesian and sampling-based methods, model-based dosing can be easily implemented into clinical protocols and used to individualize therapy irrespective of the therapeutic target or dose interval [13]. This is a significant shift in busulfan dosing from traditional weight-based dosing to a more tailored approach to therapy. It is important to note, however, that even with the application of model-based dosing a proportion of individuals will still fail to achieve optimal exposure with the first dose of busulfan, reinforcing the need for repeat blood collections and re-estimation of drug exposure in some patients.

At our institution the transition to a Bayesian platform for the determination of initial doses of busulfan and TDM was made approximately 2 years ago. The platform uses an established busulfan population PK model on the backend that receives patient-specific factors and clinical covariates entered by the clinical team into a user-friendly, web-based graphical user interface [18]. This allows for quick and easy initial dose estimation for busulfan, irrespective of the therapeutic target or dose interval and subsequent dose modifications. The primary advantage of this approach, in addition to individualized dosing, is that the drug model may be rapidly updated and refined as new data becomes available, thus, improving the model predictability. Using this approach, the model estimates were updated to better describe busulfan exposure in both neonates and children, resulting in a significantly improved fit as compared with previous population PK models supported by the software. This has helped to successfully implement novel clinical trials utilizing busulfan as part of HCT conditioning regimens such as ‘low-dose’ busulfan and gene therapy for infants diagnosed early in life with primary immune deficiencies.

Steps needed to move forward

Significant barriers must be overcome for large-scale adoption of model-based dosing. Individual patient-specific data is at the core of model-based dosing and often key elements of the data required for the model are missing, either due to improper collection or lack of standards within and across institutes. Unbiased, standardized data sharing across institutions and medical centers is the need of the hour. The second barrier is the lack of open-source modeling and simulation tools. Most of the commercial software are cost prohibitive to use in a large-scale setting and are not amenable to modern web and mobile application development. A few open source simulation and evaluation software are being developed and used in R language. It is time for the quantitative clinical community to embrace and adopt open science and open source tools.

Moving forward, the trend toward integration and utilization of model-based dosing into routine patient care will require a change in mindset and education for clinicians, pharmacists and quantitative clinical pharmacologists (pharmacometricians). Structured training in areas such as pediatric pharmacology and pharmacometrics with a focus on optimizing drug therapy should be a part of the required coursework in PharmD curriculums. This will produce progressive leaders in clinical pharmacy that can translate and implement such quantitative methods for optimal use of therapeutics in children and other special populations to improve outcomes.

Final thoughts

Model-based dosing is an integral part of personalized medicine that provides an important shift from a one-size-fits-all dosing regimen to a more personalized model-based dosing. The cumulative knowledge in quantitative clinical pharmacology gained over the last 40 years provides a strong foundation to implement model-based methods for precision dosing. Implementation and usage has already begun in select academic centers but large-scale growth requires overcoming philosophical barriers associated with training, collaborative adoption of open science and open source tools.

Financial & competing interests disclosure

J Long-Boyle currently serves as a scientific consultant to InsightRX. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

No writing assistance was utilized in the production of this manuscript.

Additional information

Funding

J Long-Boyle currently serves as a scientific consultant to InsightRX. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed. No writing assistance was utilized in the production of this manuscript.