Tailoring Drug Therapy: Bridging Data Science and Clinical Reality

The clinical pharmacology landscape is shifting from a “one-size-fits-all” approach toward a high-resolution, individualized framework. As the discipline continues to evolve, the harmonious marriage of data science and mathematical modeling is providing clinicians with unprecedented tools to “tailor drug therapy,” an aspect which is exemplified in several articles in this CPT issue (Figure 1). The approach is poised to significantly advance medicine by enhancing predictive diagnostics, personalizing treatment plans, and optimizing public health strategies.1 Furthermore, the advent of machine learning (ML) has generated much excitement and is expected to revolutionize health care, but it has yet to realize its full potential.2 Model-informed precision dosing (MIPD) typically uses pharmacodynamic/pharmacokinetic (PK/PD) models to optimize therapy for drugs with narrow therapeutic windows, such as infliximab. However, the bespoke nature of MIPD often requires specialized expertise that limits its scalability. Irie et al.3 describe the application of reinforcement learning (RL), specifically a Deep Q-Network (DQN), to personalize infliximab dosing for pediatric patients with Crohn's disease, which was subsequently validated using real-world data. Although this study demonstrates the potential of RL-guided MIPD as an automated and scalable approach for biologic therapy, the big-ticket item relates to integration of DQN frameworks into clinical dashboards for real-time, patient-specific dosing recommendations. While reinforcement learning can be used to optimize the decision, the quality of that decision depends on the data. Prior knowledge of the glomerular filtration rate (GFR) is essential when dosing drugs that are renally eliminated, such as vancomycin. In clinical practice, various biomarkers (creatinine and cystatin C) and equations are used to estimate GFR (eGFR); results can be highly variable depending on the population. Indeed, in this CPT issue, Wansing et al.4 reported that cystatin C-based eGFR better predicts renal vancomycin clearance than creatinine-based eGFR in patients with allogeneic hematopoietic stem cell transplantation, a finding that is most pronounced in patients with reduced muscle mass (sarcopenia) or those on glucocorticoids. These results demonstrate that patient care can be significantly improved by addressing overdosing (increased side effects) and underdosing (treatment failure) of vancomycin using the robust data and an appropriate biomarker-informed equation. Another drug whose exposure is susceptible to changes in renal clearance across patients is the anticoagulant edoxaban. Older patients with atrial fibrillation (AF) taking oral anticoagulants are at high risk of bleeding for numerous reasons, including chronic kidney disease. Thus, clinicians may consider prescribing suboptimal doses of edoxaban (15 mg daily vs. 60 or 30 mg), potentially putting these patients at higher risk of ischemic events. Based on a post hoc analysis of 2966 patients from across the globe with AF ≥ 80 years, it was previously reported that major bleeding events were lower in patients receiving 30 mg daily compared with those taking 60 mg daily, without increasing the risk of ischemic stroke and systemic thromboembolism (SSE).5 In a multicenter, randomized, double-blind, placebo-controlled study involving 984 Japanese patients with AF ≥ 80 years, very low-dose edoxaban (15 mg daily) was found to be superior to placebo in preventing stroke or systemic embolism and did not result in a significantly higher incidence of major bleeding than placebo.6 Interestingly, herein, Tang et al.7 add to the debate by comparing real-world edoxaban concentrations and outcomes across 15 and 30 mg daily regimens in 402 patients recruited into the Elder Care Atrial Fibrillation (ELDERCARE-AF) trial. Not surprisingly, the lower-dose regimen was associated with low edoxaban concentrations, particularly in those with creatine clearance > 50 mL/min. Importantly, the incidence of SSE was numerically higher in the ELDER-15 mg group than in the ELDER-30 mg group, while major bleeding rates were similar. As we improve our ability to dose individual drugs, we must also confront the “total pill burden” especially with the increasing prevalence of polypharmacy and multimorbidity.8 Polypharmacy is recognized as an issue in diabetes care, but its prevalence and clinical relevance in individuals with type 1 diabetes remain relatively unexplored. Ali et al.9 report that polypharmacy (more than five medications) affects 36.2% of these patients and is associated with individuals who are more likely to be female, older with a longer diabetes duration, higher BMI, higher HbA1c, and have more diabetes-related complications. Notably, polypharmacy is also associated with a higher prevalence of impaired awareness of hypoglycaemia (IAH) as well as increased fear of hypoglycemia. These findings indicate that future research and indeed clinical practice should focus on developing personalized interventions aimed at reducing medication burden and improving psychological well-being in addition to attaining metabolic control in this high-risk population. The necessity of such personalized care is further underscored when considering common comorbidities that increase pill burden, such as hypertension. Specifically, in the context of cognitive health, the management of blood pressure represents a critical yet complex variable. While hypertension is a known risk for Alzheimer's disease and related dementia (ADRD), Jafari et al.10 used electronic health record-based data to identify apparent treatment-resistant hypertension (aTRH) and high social deprivation as primary drivers of ADRD risk. The findings indicate that tailored interventions—identifying and aggressively managing patients with aTRH, while simultaneously addressing the patient's socioeconomic barriers to health—are required to protect cognitive health in these high-risk phenotypes. The transition toward high-resolution clinical pharmacology requires a dual focus: the precision of the individual dose and the holistic management of the patient's total therapeutic well-being.11 To achieve this, robust data, specialized phenotypes, and an understanding of social determinants of health are required. While advanced computational frameworks can help, we must be mindful of the fact that “a feedback loop between medical science and population-level evidence” is essential for successful implementation of dosing strategies by clinicians for effective treatment of the patient.12 No funding was received for this work. The author declared no competing interests for this work.