From Precision to Personalized: Catalyzing AI‐Enabled Innovation in Drug Development

ABSTRACT

In 2015, President Obama announced the Precision Medicine Initiative in his nationally televised State of the Union address. The vision was bold: to enable a new era of medicine through research, technology, and policies that empower patients, researchers, and providers to work together toward the development of individualized care. Now, more than 10 years later, how have the treatments available to patients changed?

1. Current Landscape

Research in precision medicine has been primarily driven by advances in genomics and the discovery of genetic biomarkers linked to drug response [1]. This focus has led to the successful development of targeted therapies in settings where a strong mechanistic relationship exists, such as HER2‐targeted therapies for breast cancer [2] EGFR inhibitors for non‐small cell lung cancer [3], and more. Yet precision medicine remains out of reach for most patients. In oncology, fewer than one‐third of solid tumors harbor a genomic biomarker predictive of response to an approved therapy [4]. Moreover, even among biomarker‐selected populations, response rates remain well below 100%. For example, the pivotal trastuzumab (Herceptin) trial reported a 50% objective response rate in HER2‐positive breast cancer [2].

This gap highlights the distinction between precision medicine (i.e., stratifying patients into subgroups) and truly personalized medicine, where each patient can be told which treatment will work best for them. We define personalized medicine as using patient‐specific data to guide therapeutic decisions for an individual patient rather than a subgroup. In practice, this includes treatment selection, dose individualization, response‐adaptive treatment changes, and the integration of (possibly multimodal) data to predict individual benefit and risk. This progression from precision to AI‐enabled personalized medicine is summarized in Figure 1.

FIGURE 1.

Overview of the path from precision to personalized medicine. The three stages depict the evolution of clinical evidence generation in drug development. Today, precision medicine stratifies patients into biomarker‐defined subgroups and estimates subgroup average treatment effects rather than individual level effects. Tomorrow, enabling tools, such as adaptive enrichment trials and AI‐enabled dosing and response prediction, will allow us to learn who benefits, how, and when. In the future, AI‐enabled personalized medicine will support individual treatment selection, digital twins and foundation models, and response‐adaptive multimodal care, addressing what works best for each patient. The continuum at bottom reflects the increasing data richness and analytical complexity required to move from population‐level to individual‐level evidence.

While the ability to characterize patient‐level biological variability has advanced substantially, most drug development programs are still designed to estimate an average treatment effect. As a result, clinically meaningful heterogeneity in response remains underexplored. If current drug development practices are not designed to answer the question of which treatment will work best for each individual patient, we may need new approaches and new incentives to drive pharmaceutical innovation toward this goal.

2. Opportunities for Innovation

The pharmaceutical industry has historically driven key advances in precision medicine. In 1998, the FDA simultaneously approved trastuzumab and HercepTest, the first companion diagnostic, establishing the paradigm of co‐developing a drug with its predictive biomarker assay. This paradigm has expanded, with more than half of all companion diagnostics approved since 2016 [5]. As the field moves toward truly personalized medicine, an important question arises: Can emerging technologies, particularly Artificial Intelligence (AI), support a similar industry‐led transformation?

One natural extension of the companion diagnostic model could be “drug‐specific” Software as a Medical Device (SaMD): AI‐enabled algorithms approved alongside a drug to optimize its use. Imagine a dosing algorithm that adjusts drug exposure based on individual patient characteristics, integrating covariates like patient demographics, renal function, or genotype. Unlike empirical trial‐and‐error at the bedside, such algorithms could be developed and validated during pivotal trials, then incorporated directly into the drug label. FDA's Predetermined Change Control Plan framework could further support this model by enabling sponsors to prospectively define how algorithms will be updated as postmarket data accumulates.

Beyond dosing, computational approaches like digital twins and foundation models could support the realization of personalized medicine in the clinic. Digital twins, defined as virtual representations of patients that integrate biological and clinical data, could allow clinicians to simulate treatment response prior to prescribing. Recent work has shown that virtual pacing in a cardiac digital twin can predict which heart failure patients will respond to device therapy [6]. AI foundation models, which are pretrained on large datasets and adaptable to multiple downstream tasks, offer another avenue. For example, GluFormer, a foundation model trained on continuous glucose monitoring data from over 10,000 individuals, can predict glycemic outcomes years in advance and forecast individual responses to dietary and pharmacologic interventions [7]. Such approaches could eventually identify, before treatment begins, which patients are most likely to benefit.

However, while these approaches are promising, their translation into routine clinical use remains constrained by practical and methodological challenges. AI‐enabled personalization depends on the availability of high‐quality, representative longitudinal data, and models trained on trial populations may not generalize to real‐world settings without careful external validation. Model performance may also degrade over time due to dataset shift or evolving standards of care, necessitating ongoing monitoring and governance.

Realizing this potential will also require rethinking how trials are designed. Real‐world evidence (RWE) studies are increasingly used to support regulatory decisions, yet they are often conducted postmarket, sometimes years after initial approval. Meanwhile, pharmaceutical companies already collect rich longitudinal data during pivotal trials, such as patient‐reported outcomes, digital biomarkers from wearables, or detailed phenotypic characterizations, but current trial designs rarely prioritize this. Emerging approaches offer practical pathways forward if implemented with appropriate methodological rigor. Adaptive enrichment trials, supported by Bayesian decision frameworks, can refine patient selection mid‐study based on accumulating evidence, but require prespecification of adaptation rules and careful control of multiplicity to avoid inflated false positive findings. SMART designs provide a structured way to evaluate dynamic treatment regimes tailored to individual response trajectories, enabling estimation of individualized decision rules while mitigating overfitting through embedded randomization. AI‐enabled tools used to support personalization, such as response prediction or dosing algorithms, would need to be locked prior to confirmatory testing, prospectively validated within Phase II/III programs, and evaluated for transportability using external or real‐world datasets. Prospective incorporation of these designs could enable earlier identification of patients with differential benefit, reduce post‐approval uncertainty, and support more informative labeling, creating an opportunity that industry is uniquely positioned to lead.

Despite these opportunities, meaningful implementation of personalized medicine within industry faces practical constraints. Development organizations are often structured around indication‐specific programs rather than longitudinal learning, and risk aversion may limit investment in trial designs perceived as complex or unfamiliar. Incorporating adaptive designs, AI‐enabled tools, or rich phenotyping can increase upfront costs and development timelines, while regulatory and reimbursement pathways for individualized evidence remain uncertain. Addressing these barriers will require clearer regulatory expectations, incentives aligned with individualized value generation, and organizational shifts that prioritize learning across development stages.

3. Incentivizing Innovation Through New Policy

Policy is an important catalyst for pharmaceutical innovation. Consider the precedent set by the Orphan Drug Act of 1983: before its passage, only 38 drugs had been approved for rare diseases; today, more than half of new FDA approvals are for orphan indications [8]. Tax credits, market exclusivity, and regulatory flexibility transformed rare diseases from commercial afterthoughts into one of the industry's most lucrative sectors. A similar policy framework could accelerate investment into personalized medicine. On the regulatory side, incentivizing individualized treatment development would be aligned with the FDA's longstanding regulatory framework. The FDA already recognizes that population averages are insufficient to ensure safe and effective use across diverse patients, as reflected by requirements to conduct pediatric studies, organ impairment studies, drug–drug interaction studies, and subgroup analyses during development. Each of these regulatory evolutions acknowledged clinically meaningful heterogeneity and created incentives for sponsors to study it systematically. Building on this precedent, the FDA could create a dedicated “personalized medicine” section within the drug label to capture individualized evidence generated during development. Such a section could include biomarker‐guided dosing recommendations or subpopulation‐specific response rates. This framework would create space for sponsors to differentiate their products and foster competition around personalization. Precision dosing, where pharmacokinetic and pharmacodynamic principles are already well established, represents a particularly accessible starting point.

On the reimbursement side, payors could incentivize drugs with precision labels through preferential formulary placement or enhanced reimbursement rates. The recently announced TEMPO pilot offers a template for coordinated FDA‐CMS initiatives: under this program, digital health devices for chronic disease management can be deployed under FDA enforcement discretion while collecting real‐world performance data, with CMS providing outcome‐aligned payments tied to measurable health improvements [9]. Financial incentives like these are likely to be an important driver for making personalized medicine a key focus of drug development. Although this discussion focuses on U.S. regulatory and reimbursement frameworks, related initiatives are emerging internationally. In Europe, efforts such as DARWIN EU aim to leverage real‐world data and iterative evidence generation to support regulatory decision‐making, and ICH guidelines on adaptive designs suggest growing international alignment around more flexible development paradigms. Absent such incentives, the burden of individualized treatment discovery implicitly falls on healthcare providers. Hospitals lack the infrastructure, resources, and economic incentives required to conduct systematic pharmacologic optimization studies. Reliance on trial‐and‐error prescribing risks fragmenting evidence generation and leaves personalization reliant on circumstance rather than science. By contrast, sponsors possess both technical capabilities and commercial pathways to develop and sustain personalized treatment strategies if appropriately incentivized.

4. Conclusion

Regulatory agencies like the FDA are synonymous with ensuring safety and efficacy. But FDA's mission also includes “advancing the public health by helping to speed innovations that make medical products more effective, safer, and more affordable” [10]. Realizing truly personalized medicine, where a patient can ask their physician, “What treatment will work best for me?” and receive an evidence‐based answer, will require fulfilling this mandate.

The scientific building blocks are increasingly in place. AI‐enabled tools now demonstrate potential to simulate treatment response in individual patients, predict disease trajectories from longitudinal data, and optimize dosing in real time. Coordinated FDA‐CMS pilots are testing new models for aligning evidence generation with reimbursement. Decades of regulatory precedent demonstrate that well‐designed incentives can redirect industry investment toward important and underserved areas of medicine.

What remains is a shift in expectation. Pharmaceutical companies will need to design trials that capture the data required for personalization, not just approval. Regulators will have to create labeling frameworks that reward individualization. Payors must incentivize treatments that demonstrate differential benefit. And patients will need clear evidence that personalized tools are trustworthy and actionable, with clear communication around how individualized recommendations are generated and used. Ensuring equitable access to the data, diagnostics, and digital tools that enable personalization will be essential if these approaches are to improve outcomes broadly rather than selectively. As with previous advances in drug development, the path from precision medicine to personalized medicine will be achievable with stakeholders working together to make treatments that benefit all.

Funding

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.