Plasma drug concentrations are a cornerstone of pharmacokinetics and are widely used as proxies for systemic drug exposure. This use is rooted in the assumption that drug concentrations in the blood reflect the bioavailable fraction reaching the site of action. In most classical PK/PD models, drug effects are modelled as functions of the concentration at the effect site, which is typically approximated by plasma concentration (Meibohm and Derendorf, 1997). Plasma levels also facilitate parameter estimation for key PK variables (clearance, volume of distribution, bioavailability), which are then used for dose adjustment and exposure prediction.

NONMEM and Monolix are population PK/PD modeling tools which allow estimating individual PK profiles using sparse plasma concentration data implementing a bayesian approach. This enables personalized dose prediction, especially when paired with clinical outcome data in a model-informed precision dosing (MIPD) paradigm (Mould and Upton, 2012).

One of the core assumptions of classical PK/PD models is that plasma drug concentrations reflect the exposure at the pharmacological target site. However, this assumption is often violated in cases where the site of action is compartmentalized or poorly perfused. For example, inhaled corticosteroids such as budesonide or fluticasone are intended to act locally in the lungs. Systemic absorption is low and variable, and plasma levels correlate poorly with anti-inflammatory efficacy in the airways (Derendorf et al., 2006).

In a study evaluating inhaled budesonide, Derendorf and colleagues showed that changes in plasma concentration did not predict changes in exhaled nitric oxide or eosinophilic counts in sputum, which are more relevant markers of airway inflammation. This highlights that systemic concentration is a poor surrogate for local lung exposure.

Topical corticosteroids, ophthalmic drops, and intra-articular injections also exemplify situations where therapeutic effects are achieved with minimal or undetectable plasma concentrations. For instance, dexamethasone eye drops effectively reduce ocular inflammation with negligible systemic absorption. Measuring plasma levels in these cases adds no clinical value and may misrepresent exposure entirely (Patel et al., 2013).

Similarly, intra-articular corticosteroids used in rheumatologic conditions achieve high local tissue levels with minimal systemic spillover. Here again, the absence of systemic concentrations does not imply pharmacological inactivity.

Pharmacological agents with slow onset, long half-lives, or indirect mechanisms of action may show hysteresis in the concentration-effect relationship. Selective serotonin reuptake inhibitors (SSRIs) are a prime example. Although plasma concentrations stabilize within days, clinical antidepressant effects emerge only after several weeks, reflecting delayed neuronal plasticity and transcriptional changes (Richelson, 2001).

This phenomenon produces counterclockwise hysteresis loops on PK/PD plots, meaning the same concentration can be associated with different levels of clinical response depending on time since initiation. This undermines the utility of plasma monitoring, especially when the pharmacodynamic response is not immediate.

Some drugs exert their clinical effect not through the parent compound but via active metabolites. For instance, codeine requires biotransformation via CYP2D6 to morphine to exert its analgesic effect. Plasma concentrations of codeine do not correlate with analgesic efficacy in CYP2D6 ultra-rapid metabolizers or poor metabolizers (Gasche et al., 2004). Here, the concentration of the parent compound may be misleading, and genetic factors override traditional PK assumptions. Similarly, clopidogrel, an antiplatelet agent, is a prodrug requiring hepatic bioactivation. Measuring its plasma concentration does not reflect platelet inhibition, and pharmacodynamic assays (e.g., platelet aggregometry) are more informative (Mega et al., 2009). Although its antiplatelet effect is attributed primarily to an active thiol metabolite, it remains unclear whether the parent compound or other metabolites contribute independently or synergistically to clinical efficacy (Simon et al., 2015). More complex case involve both parent compound and metabolites acting together. Quantifying the relative contribution of each is challenging, especially when they differ in potency, receptor targets or pharmacokinetics.

Two common problematic assumptions persist: Historically, two problematic assumptions have persisted: summing plasma concentrations of parent and metabolite(s) to estimate effect or ignoring metabolites entirely (Bonate, 2013). While such simplifications are less common in modern drug development, they may still appear in translational research or clinical practice, often without robust empirical support.

In these diverse scenarios, relying on systemic levels may lead to inappropriate dose adjustments, false assumptions of inefficacy, or neglect of alternative explanatory mechanisms (e.g., pharmacogenetics, tissue kinetics).

Traditional pharmacokinetic models often focus on single-point concentrations (e.g., Cmax, Ctrough, Cavg) or aggregate metrics such as the area under the curve. However, the rate of change (i.e., how quickly a drug concentration rises or falls) can critically influence the magnitude and quality of the effect.

This is particularly well demonstrated in the case of central nervous system (CNS) active substances. For instance, studies have shown that the rate of increase in plasma and brain concentration of psychostimulants such as cocaine and methylphenidate is tightly associated with their reinforcing (and addictive) potential (Volkow et al., 1997). The same dose administered slowly produces significantly weaker euphoria than when administered rapidly, due to slower brain penetration.

A similar effect is observed with alcohol. In a controlled human study, de Wit et al. (1992) found that a rapid rise in blood alcohol concentration (e.g., from consuming spirits) produced more intense subjective effects than the same dose ingested more slowly (e.g., as beer), despite similar AUCs. This suggests that not only the extent of exposure (AUC), but also the rate of increase in plasma concentration (dC/dt) and the peak concentration (Cmax), play critical roles in determining the intensity of effect.

A rapid intravenous bolus and a slow infusion or transdermal delivery may reach the same Cmax, but the kinetics of onset can affect receptor engagement, downstream signaling, and central nervous system responses. Fentanyl provides a clear example of how route of administration can dramatically affect both the kinetics and the clinical consequences of a given dose. Intravenous bolus injection produces a rapid rise in plasma and brain concentrations, which may lead to abrupt respiratory depression, particularly when administered quickly (Stanley, 2014). In contrast, transdermal fentanyl produces a delayed Tmax (8–16 h) and sustained systemic exposure over 2–3 days, making it suitable for chronic pain management. These differences highlight the importance of considering both the rate and route of delivery in predicting pharmacodynamic effects, beyond peak concentration values alone. Similarly, dopamine administered as a slow infusion supports cardiovascular tone, whereas bolus injections can provoke arrhythmias or excessive vasoconstriction, highlighting that rate-dependent pharmacodynamics are clinically relevant even with identical exposure levels (Goldberg, 1972).

The causal inference framework developed by Judea Pearl offers a rigorous method to test assumptions about causal pathways. In Pearl’s model, a variable C (concentration) is a valid mediator between D (dose) and E (effect) only if.

1. It is affected by the dose.

But the key insight is this:

If changing the dose alters the effect even when the concentration is held constant, then concentration is not a sufficient mediator. Although experimentally maintaining a constant concentration while varying the dose is unrealistic, this counterfactual test helps frame our interpretation. In real-world pharmacology, this implies that other latent variables (e.g., different mode of administration, phenotypes of metabolizing enzymes, target sensitivity) may mediate clinical effects in ways not fully captured by conventional PK metrics such as Cmax, Tmax, or AUC, especially when therapeutic decisions rely on sparse sampling or isolated plasma measurements.

Drugs with delayed pharmacodynamic responses often demonstrate hysteresis when plotting effect versus concentration. For example, levodopa in Parkinson’s disease shows a lag between peak plasma concentration and maximal motor benefit due to complex intracellular signaling and dopamine storage kinetics (Gancher et al., 1987).

Such counterclockwise hysteresis loops suggest that effect compartment kinetics (i.e., the delay between plasma and target site concentrations) or downstream adaptive mechanisms contribute significantly to the observed effect, thereby weakening the centrality of plasma levels as the causal driver, and that the magnitude of this lag may itself vary with disease progression or repeated drug administrations (Valenzuela et al., 2023; Lott et al., 2018).

For a reliable PK/PD modeling, this delay should therefore be considered as a time-varying parameter rather than a fixed constant.

Beyond concentration itself, the temporal profile (rate of rise, Tmax, fluctuation range) may exert more direct causal influence on effect. This has implications for drug formulation and abuse potential.

Lisdexamfetamine, a prodrug of dextroamphetamine used in ADHD, has a slower onset and lower abuse liability than immediate-release amphetamine, despite achieving similar AUCs. Its slow conversion to the active compound results in a blunted peak and gradual increase, attenuating the euphoric effects (Ermer et al., 2016).

These examples underline a crucial point: while comprehensive plasma concentration profiles may inform pharmacodynamic outcomes, individual concentration values, often used in clinical or translational settings, do not necessarily reflect the true causal mediators of drug effect. Pearl’s causal lens offers a rigorous way to ask whether our models truly reflect interventions on the effect, or merely observe associations.

When plasma concentrations may fail to reliably indicate pharmacological activity, more direct measurements of drug-target interaction may be preferred. One such approach is receptor occupancy, often assessed using positron emission tomography (PET) imaging.

In the case of antipsychotics (e.g., risperidone, olanzapine), PET has been used to estimate the percentage of dopamine D2 receptor occupancy in the striatum. Studies show that therapeutic efficacy in schizophrenia is associated with 60%–80% D2 occupancy, while higher occupancy correlates with extrapyramidal side effects (Farde et al., 1986). This receptor-based marker is more predictive of response and tolerability than plasma concentration, which may vary widely due to metabolism and protein binding.

Similarly, naltrexone, an opioid receptor antagonist used in addiction, displays a poor correlation between plasma levels and clinical effect. PET studies demonstrate that effective blockade of μ-opioid receptors is better evaluated by receptor occupancy rather than circulating concentrations (Weerts et al., 2008).

Some effects are better captured using functional biomarkers, such as.

• Glucose levels for insulin action,

In these examples, plasma concentrations may be misleading or redundant. For instance, clopidogrel, a prodrug, requires hepatic activation, and its parent compound has no pharmacological effect. Platelet function tests provide more accurate guidance for dose adjustment or resistance detection than plasma assays (Gurbel et al., 2003).

It is important to emphasize that biomarker monitoring remains clinically appropriate in some settings such as blood glucose self-monitoring in diabetic patients, where the biomarker provides a direct reflection of disease control.

In indirect response models, drug effect is mediated not by direct interaction but by modulation of the production or elimination of a biological mediator (Jusko and Ko, 1994). These models are widely used for.

• Corticosteroids, which suppress cytokine production (e.g., IL-6),

In these systems, plasma drug levels are temporally and causally decoupled from effect. Systems pharmacology models aim to simulate these dynamics using networks of ordinary differential equations, integrating PK, PD, and biological feedback loops (Danhof et al., 2007).

Technological innovations have enabled continuous monitoring of physiological or pharmacodynamic signals via.

• Biosensors for glucose (e.g., continuous glucose monitors),

These tools may offer more real-time, clinically meaningful information than intermittent plasma level assessments, especially when effects are time-sensitive or behaviorally driven (Insel, 2017).

In the age of real-world data, causal inference methods such as propensity score weighting, marginal structural models, and instrumental variable analysis allow researchers to estimate treatment effects without relying exclusively on PK measurements (Hernán and Robins, 2020).

These methods are particularly useful when concentration data are missing, unreliable, or confounded. For example, in assessing the effectiveness of opioid-sparing strategies post-surgery, models can estimate causal impact from Eletronic Medical Record (EMR) data using these techniques.

Plasma concentrations remain valuable, but they are not universally necessary nor sufficient to understand drug action. As pharmacology becomes increasingly multimodal and individualized, alternative tools (imaging, functional assays, systems modeling, biosensors, and causal inference frameworks) offer richer insights into the mechanisms and effects of drug therapy.

Drugs intended to act locally, such as inhaled, topical, intraocular, or intra-articular formulations, often have minimal systemic exposure. In such cases, plasma concentrations provide limited or misleading information.

When the effect-site kinetics are slow or complex, plasma concentrations at a given time point may not correspond to the pharmacodynamic response.

The rate of rise (dC/dt), Tmax, or fluctuation amplitude may have more causal power than static concentrations.

Prodrugs, polymorphic metabolism, or transporter activity can cause dissociation between parent drug levels and clinical response.

Even when imperfect, plasma concentrations are valuable if they enable dose adjustment, toxicity avoidance, or therapeutic confirmation.