Loading field map
For decades, the pharmaceutical industry has faced a brutal arithmetic: fewer than one in ten drug candidates that enter clinical trials ever reach patients. The gap between promising preclinical results and disappointing human outcomes is the central problem that translational pharmacology exists to close. The subfield asks a deceptively simple question: how can we use data from laboratory models—cells, animals, computer simulations—to predict whether a drug will work safely in people? The answers have shifted dramatically over the past sixty years, producing four distinct frameworks that remain in active tension today.
From the 1960s through the end of the twentieth century, the dominant approach to translating preclinical findings into human dosing was the Empirical Dose-Finding Paradigm. Its logic was straightforward: start with a conservative dose in humans, observe for toxicity, and escalate cautiously until a maximum tolerated dose is identified. The most widely used method was the 3+3 design, in which cohorts of three patients received a given dose; if no dose-limiting toxicity occurred, the next cohort received a higher dose. If one of three patients experienced toxicity, three more were added at the same dose. This rule-based algorithm required no mathematical model, no biomarker, and no mechanistic understanding of the drug's action.
The paradigm's strength was its simplicity and its perceived safety: it prioritized avoiding harm over rapid dose escalation. But its weaknesses were severe. The 3+3 design and its variants used only a small fraction of the accumulating data—typically just the presence or absence of toxicity at each dose level—and ignored the rich information about drug concentration, individual variability, and biological effect that could be measured. As a result, many drugs advanced to later-stage trials with suboptimal doses, contributing to the high failure rates that plagued drug development. The empirical paradigm treated each new drug as a black box, relying on trial and error rather than quantitative prediction.
By the early 1990s, the limitations of empirical dose-finding had become impossible to ignore. Two new frameworks emerged in parallel, each offering a different path forward. They shared a common goal—better prediction of human response—but diverged fundamentally in their sources of confidence.
Model-Informed Drug Development (MIDD) replaced the empirical paradigm by placing mathematical modeling at the center of decision-making. Instead of escalating doses by fixed rules, MIDD uses pharmacokinetic-pharmacodynamic (PK/PD) models to describe how drug concentration changes over time and how that concentration relates to both efficacy and toxicity. These models are built from preclinical and early clinical data and then used to simulate outcomes for untested doses, schedules, and patient populations. The U.S. Food and Drug Administration formally embraced MIDD in the 1990s and 2000s, recognizing that quantitative models could reduce the number of failed trials and accelerate approval for drugs that showed clear benefit.
What made MIDD a genuine replacement rather than a refinement was its shift in epistemology: confidence came from inference, not from direct observation. The empirical paradigm trusted only what had been observed in a small number of patients; MIDD trusted what could be predicted from a model fitted to all available data. This allowed developers to explore dose ranges that would be unethical or impractical to test empirically. However, MIDD models were largely phenomenological—they described the relationship between dose and effect without necessarily explaining the underlying biology. A PK/PD model might predict that doubling the dose doubles the effect, but it could not say why.
At roughly the same time, the Translational Biomarker Paradigm took a different approach. Instead of relying on mathematical inference, it sought to measure biological signals that could serve as direct bridges between preclinical models and humans. A biomarker—whether a protein level, an imaging signal, or a genetic variant—is a quantifiable indicator of a biological process or drug effect. The paradigm's core commitment is to validate biomarkers in preclinical systems and then use them as surrogate endpoints in clinical trials. If a biomarker reliably predicts clinical outcome, it can replace the need for long-term efficacy studies.
This framework drew heavily on advances in molecular pharmacology and pharmacogenomics. For example, a drug that inhibits a specific kinase might be evaluated by measuring phosphorylation of its target in tumor biopsies from both mice and patients. The biomarker paradigm offered a measurement-driven alternative to the model-driven approach of MIDD. Its strength was directness: a well-validated biomarker provides evidence that the drug is hitting its intended target in humans. Its weakness was the difficulty of validation—many biomarkers that worked in animals failed to predict human outcomes, and the process of proving a biomarker's clinical utility was itself slow and expensive.
MIDD and the biomarker paradigm coexisted uneasily. Proponents of modeling argued that biomarkers were just another data type to be incorporated into quantitative models, while biomarker advocates countered that models were only as good as their assumptions and that direct biological measurement was more trustworthy. In practice, the two frameworks began to merge: MIDD models increasingly incorporated biomarker data, and biomarker studies increasingly used modeling to interpret their results.
By the early 2000s, a third response emerged that built directly on MIDD but pushed it in a more ambitious direction. Quantitative Systems Pharmacology (QSP) aims to construct mechanistic, multiscale models that represent the network of biological pathways underlying drug action. Where MIDD models treat the body as a black box with input-output relationships, QSP models attempt to open the box, representing receptors, signaling cascades, gene regulation, and cell-cell interactions in mathematical form.
QSP was formalized through workshops at the National Institutes of Health in 2008 and 2010, which called for a discipline that merged systems biology with pharmacology. The distinctive commitment of QSP is its insistence on mechanism: a QSP model does not just fit a curve to data; it encodes hypotheses about how the system works. This allows it to predict emergent behaviors that a phenomenological model would miss, such as drug resistance arising from network rewiring or paradoxical effects at different doses. However, QSP models are far more data-hungry and computationally intensive than traditional PK/PD models, and they require expertise in both biology and mathematics that is still scarce.
QSP did not replace MIDD; rather, it expanded the scope of what modeling could achieve. Today, the two coexist, with MIDD used for routine dose selection and regulatory submissions, and QSP reserved for complex questions about mechanism, combination therapies, and disease progression.
No single framework has won the day. The Empirical Dose-Finding Paradigm has largely been abandoned for first-in-human studies, but its legacy persists in some oncology trials where rapid dose escalation is still practiced. MIDD is now standard in regulatory submissions for new drugs, especially in areas like oncology, infectious disease, and rare disorders. The Translational Biomarker Paradigm has become essential for targeted therapies, where biomarkers are used to select patients and monitor response. QSP is growing rapidly in academic and industry research, particularly for complex diseases like cancer immunotherapy and neurodegenerative disorders.
What the leading frameworks agree on is that prediction requires data—lots of it—and that no single source of evidence is sufficient. They disagree on the optimal unit of analysis: MIDD and QSP trust mathematical models to integrate and extrapolate data, while the biomarker paradigm trusts direct biological measurement as the gold standard. This disagreement is not merely academic; it shapes how companies allocate resources, how regulators evaluate evidence, and how quickly new therapies reach patients.
In practice, translational pharmacology today is a hybrid discipline. A typical drug development program might use MIDD to design initial doses, biomarker assays to confirm target engagement in early trials, and QSP to explore resistance mechanisms and combination strategies. The frameworks are no longer seen as competitors but as complementary tools, each best suited to a different stage of the translation pipeline. The central tension between empiricism and modeling, and between measurement and inference, remains unresolved—and that tension is precisely what drives the field forward.