The Broken Dose: Why 50% of Patients Pay for Drug They Don't Need

Imagine paying $1,000 a month for a drug when half of it is more than your body needs.

That's the reality for millions of patients today. Most FDA-approved drugs carry a single fixed dose on their label — a dose chosen during clinical development to work in roughly 90% of the trial population. It's a pragmatic choice: one dose, one label, one reimbursement code. Simple for regulators. Simple for insurers. Simple for prescribers.

But simple for the system is not the same as right for the patient.

The math is uncomfortable. If a dose is chosen to cover 90% of patients, that means roughly half the population achieves drug concentrations well above what they need for efficacy — they're overtreated. They experience more side effects, their bodies process more drug than necessary, and they (or their insurer) pay full price for exposure they don't need.

Meanwhile, at the other tail of the distribution, approximately 10–15% of patients don't get enough. Their drug levels are subtherapeutic. They fail treatment — not because the drug doesn't work, but because the dose was wrong for them. And when their physician tries to prescribe a higher dose? Insurance often won't cover it. It's not on the label.

The current dosing system is broken. We have the science to fix it. We're just not using it.

How Doses Get Picked

Drug doses are not chosen based on what's optimal for each patient. They're chosen based on what works for enough patients in a clinical trial to demonstrate statistical significance.

The typical pathway looks like this:

1. Phase 1: Test a range of doses for safety. Find the maximum tolerated dose.
2. Phase 2: Run a dose-ranging study. Pick 2–3 doses. Identify which achieves the desired pharmacodynamic response in the largest fraction of patients.
3. Phase 3: Take the “best” dose forward. Confirm efficacy in thousands of patients. This dose goes on the label.

The critical decision happens in Phase 2. The goal is to find a dose where the majority of patients — typically 80–90% — achieve a target exposure or pharmacodynamic response. This is the dose that has the best chance of showing a statistically significant treatment effect in a Phase 3 trial.

But here's the problem: patients are not identical. How the body handles a drug — and how the body responds to it — varies dramatically. Body weight, genetics, organ function, disease severity, concomitant medications, and immune status all influence not just drug levels, but the pharmacodynamic response: the actual biological effect the drug produces.

When you give a fixed dose to a population with this much variability, the resulting range of clinical responses is wide. The dose that achieves the desired pharmacodynamic effect in the median patient will over-suppress the target in many patients and under-treat others. The right question isn't “what's the blood level?” — it's “did the patient reach the clinical or biomarker endpoint that matters?”

10 Drugs Where Fixed Dosing Fails

The following table presents 10 widely prescribed drugs where published population pharmacokinetic data demonstrate significant inter-individual variability — meaning patients on the same dose can have vastly different drug exposures.

PD endpoints shown are the clinically relevant disease markers or pharmacodynamic responses that vary significantly across patients at the same dose. Sources: Published PopPK and exposure-response analyses cited in references section.

When PD Tells the Story Better Than PK

The real question isn't “what's the drug concentration?” — it's “is the patient responding?” For many drugs, pharmacodynamic endpoints or disease biomarkers correlate far better with clinical outcomes than plasma concentrations alone.

Vutrisiran (Amvuttra) is a striking example. This siRNA therapy for hereditary ATTR polyneuropathy is dosed at a flat 25 mg every 3 months — regardless of body weight. The pharmacodynamic endpoint is TTR (transthyretin) knockdown, which ranges from 65% to over 90% at the same dose. Patients with lower body weight or higher baseline TTR levels achieve deeper knockdown than they may need. At $300,000–$400,000 per year, many patients are achieving more TTR suppression than required for clinical benefit — and paying an extraordinary premium for it.

Inotersen (Tegsedi), another nucleic acid therapy for hATTR, drives the point home from the other end of the cost spectrum. At ~$450,000/year for a flat 300 mg SC weekly dose, lean body mass is the dominant covariate — it scales clearance by ~30%. The clinically relevant benchmark isn't the modest 50% TTR reduction from the NEURO-TTR trial era — it's the 80–90% knockdown that siRNA therapies (patisiran, vutrisiran) now routinely achieve. At the approved dose, a small patient (LBM 35 kg) may reach siRNA-equivalent knockdown, but a large patient (LBM 75 kg) often falls well short. One dose. Same price. A wide distribution of PD outcomes measured against a high bar set by competing therapies.

Tirzepatide (Zepbound) for obesity tells a similar story through a different lens: weight loss itself is the measurable PD outcome. At the same dose, weight loss ranges from 5% to over 25% of body weight. A patient who achieves their target weight loss at 7.5 mg doesn't need 15 mg — yet the titration scheme pushes toward higher doses. At $1,000/month, every unnecessary dose escalation costs the patient and the system.

Risankizumab (Skyrizi) targets IL-23p19 in psoriasis and Crohn's disease. The degree of IL-23 suppression reaches a plateau at drug exposures well below what many patients achieve at the standard 150 mg dose. PASI 90 and PASI 100 responses — near-complete skin clearance — are achieved by the majority, suggesting that for many patients, the same clinical outcome could be reached at a lower dose. Body weight and immunogenicity drive a 2–3× range in exposure.

Pembrolizumab (Keytruda), the world's best-selling drug at $150,000–$190,000/year, is dosed at a flat 200 mg for all patients. The relevant PD measure — PD-1 receptor occupancy — reaches near-saturation at exposures far below what many patients achieve. A 50-kg woman and a 120-kg man receive the same 200 mg infusion. For most, the biological effect plateaued long ago. The extra drug cost is pure waste.

Where Individualized Dosing Already Works

Infliximab (Remicade) provides one of the clearest proofs that measuring response and adjusting dose works. The TAXIT trial demonstrated that therapeutic drug monitoring — targeting trough concentrations of 3–7 µg/mL linked to mucosal healing in IBD — led to better clinical outcomes than fixed dosing. Anti-drug antibodies cause some patients to clear the drug 5× faster, leaving them undertreated at the label dose. Dose individualization based on TNF-α suppression and mucosal healing isn't just better science — it's better economics.

Imatinib (Gleevec) for CML shows what happens when you ignore the PD signal. The flat 400 mg dose produces a 4-fold range in trough concentrations. We know that patients with trough levels below 1,000 ng/mL have significantly worse rates of major molecular response (BCR-ABL < 0.1%). Yet the label says 400 mg for everyone. At $100,000+/year, patients who need higher doses to achieve molecular response can't get insurance coverage, while those who could achieve the same BCR-ABL suppression at 300 mg are overexposed.

Busulfan and vancomycin are the existence proofs that model-based dosing works at scale. Busulfan TDM — targeting a specific AUC window tied to graft success vs. veno-occlusive disease — has been standard of care in transplant for decades. Vancomycin's shift from trough-based to Bayesian AUC/MIC-targeted dosing (Rybak 2020 guidelines) improved target attainment from ~40% to over 80%. The PD target is clear: AUC/MIC 400–600 for bacterial kill. The infrastructure exists. The question is why we haven't extended this to dozens of other drugs where disease endpoints could guide dose.

See It Yourself: Risankizumab (Skyrizi) PD Simulator

Don't take my word for it. Below is an interactive simulator built from published population PK/PD data for risankizumab (Suleiman et al., Clin Pharmacokinet 2019, PMID 30945127). It simulates 1,500 virtual psoriasis patients and shows two clinically relevant outcomes: the distribution of PASI scores after treatment and the degree of IL-23 suppression.

Try this: set the dose to 150 mg (labeled dose) and note the PASI 75 and PASI 90 response rates. Then reduce to 75 mg. Many patients still clear their skin. That's the point — for a significant fraction, the labeled dose delivers more IL-23 suppression than needed for clinical response.

The green histogram shows the distribution of PASI scores after treatment. The vertical dashed lines mark the PASI 75 and PASI 90 thresholds — the benchmarks for clinically meaningful skin clearance. The blue histogram below shows IL-23 suppression, where you can see that most patients at the standard dose achieve near-complete target neutralization.

See It Yourself: Inotersen (Tegsedi) TTR Knockdown

Inotersen is an antisense oligonucleotide approved at a flat 300 mg SC weekly for hereditary transthyretin amyloidosis (hATTR). The pharmacodynamic endpoint — TTR knockdown — is directly measurable and clinically meaningful. While the original NEURO-TTR trial used ≥50% reduction as a threshold, the competitive landscape has shifted: patisiran achieves ~80% knockdown and vutrisiran ~85–90%. The real clinical benchmark is now 80%+ TTR reduction.

The population PK/PD model (Yu RZ et al., NAT 2020, PMID 31730427) reveals the problem: lean body mass scales all PK parameters, so a 35-kg patient achieves dramatically higher concentrations than an 80-kg patient at the same 300 mg dose. Many patients fall short of siRNA-equivalent knockdown, while others exceed it — all paying $450,000+/year for a dose that wasn't optimized for them.

The simulator below shows this directly. Try adjusting lean body mass and dose to see how the TTR knockdown distribution shifts. Pay attention to the fraction of patients below the 50% target (undertreated, in yellow) versus those well above it (potentially overtreated).

This is an ASO that costs nearly half a million dollars per year. The PD endpoint is routinely measured (serum TTR). The population model exists. Yet the label specifies a single flat dose for all patients. A dosing simulator that accounts for lean body mass and measured TTR response could help clinicians decide whether their patient truly needs weekly 300 mg — or whether the same clinical benefit could be achieved at a lower dose or extended interval.

The Nuance: When More Drug Is Not Waste

This argument requires honesty about complexity. Dose-exposure-response relationships are not simple, and I want to acknowledge what the data actually shows.

For many drugs, a higher dose produces deeper PD effects in a significant fraction of the population — without additional safety consequences. A patient on risankizumab who achieves PASI 100 (complete skin clearance) rather than PASI 75 is not “overtreated” — they have a better quality of life. A patient on tirzepatide who loses 22% body weight instead of 15% may achieve greater cardiometabolic benefit. Deeper TTR knockdown with vutrisiran may provide greater protection against disease progression.

More is sometimes genuinely better for the patient, even when the label target has been met.

The question, then, isn't simply “is the patient overtreated?” — it's “does the clinician have the information to make that judgment?” Today, in most cases, the answer is no. Clinicians prescribe the labeled dose and hope for the best. They don't know where their patient falls on the exposure-response curve. They don't know whether their patient achieved the PD target at a lower dose months ago, or whether they're barely responding and need more.

The real problem isn't that doses are too high or too low — it's that clinicians have no tools to know which is true for their patient.

The GLP-1 Lesson: How Vials Align Incentives

Tirzepatide and semaglutide illustrate how drug formulation and pricing can either enable or prevent meaningful dose individualization.

Pre-filled pens are sold at fixed price points — regardless of the dose inside. A 5 mg pen and a 15 mg pen cost the same ~$1,000/month. The dose varies, but the price doesn't. Under this model, there's no economic benefit to finding that a patient responds adequately at a lower dose — they pay the same either way.

Lilly recognized this problem. In late 2025, they launched Zepbound single-dose vials available through LillyDirect at significantly lower prices — as low as $349/month for lower doses. This is exactly the right move: vials priced per milligram mean that a patient who achieves their weight loss target at 5 mg pays less than a patient at 15 mg. The incentives align: individualized dosing saves the patient and the system real money.

This should be the model for every high-cost drug. When pricing reflects the actual dose delivered, dose individualization becomes economically rational — not just scientifically sound. Vials enable the kind of dose adjustment that pens prevent.

The Fix: Model-Based Adaptive Dosing

The solution exists. It has existed for decades. Population PK/PD models, combined with sparse sampling and Bayesian estimation, can individualize doses based on each patient's actual response — not just their blood level, but their disease endpoint.

Here's how it works in practice:

1. Start with the labeled dose. Use the population model to predict what PD response the patient is likely to achieve, based on their weight, disease severity, and other covariates.
2. Measure the response. For some drugs, this means a drug level (sparse PK sampling). For others, it means a biomarker: TTR knockdown for vutrisiran, weight loss trajectory for tirzepatide, BCR-ABL ratio for imatinib, PASI score for risankizumab, AUC/MIC for vancomycin. The right measurement depends on the drug and the disease.
3. Update the model. Using Bayesian estimation, combine the population model's prior with the individual's measured response to estimate their personal parameters. This is what dosing software does — and increasingly, what AI can do.
4. Adjust the dose. Based on the individual's response trajectory, adjust the dose to achieve the target PD endpoint — not more, not less.

This isn't theoretical. Vancomycin Bayesian AUC/MIC-targeted dosing runs in hundreds of hospitals today. Busulfan TDM has been standard in transplant for decades. Infliximab TDM-guided dosing outperformed fixed dosing in randomized trials. The technology works. The math is solved. The disease endpoints are measurable. The question is why we haven't extended this approach.

What's missing is the system — regulatory, reimbursement, and clinical workflow infrastructure — to extend this approach beyond the handful of drugs where it's already practiced.

What the FDA Label Should Look Like

Today, Section 2 (Dosage and Administration) of most FDA labels reads something like:

“The recommended dosage is 200 mg administered as an intravenous infusion over 30 minutes every 3 weeks.”

What it should say:

“The recommended starting dosage is 200 mg IV Q3W. Monitor [PD biomarker/clinical endpoint] at Weeks 6 and 12. An FDA-approved dose simulator is available at [URL] to support dose adjustment decisions based on PD markers, clinical endpoints, patient covariates, and safety data. Based on the overall clinical picture, dose adjustment within the range of [A–B mg] may be considered without additional safety evaluation. Clinicians should consider the totality of evidence — efficacy markers, safety signals, patient preferences, and comorbidities — when making dose decisions.”

The key shift: labels should reference an approved dose simulator with decision guidance — a tool that integrates PD markers, clinical endpoints, and safety data to help clinicians make informed dose adjustments. Not a black-box algorithm that replaces clinical judgment, but a decision-support tool that gives clinicians the pharmacometric context they currently lack.

This isn't radical. This is how we already approach busulfan, vancomycin, and aminoglycosides — measure, model, adjust. We're simply advocating for the same standard to be applied to drugs where the clinical variability is just as real and the financial stakes are far higher.

ICH M15 — the newly harmonized guideline on model-informed drug development — establishes the framework for model credibility and regulatory use. The regulatory door is opening. What we need now is for sponsors to build these dose simulators during development, validate them alongside the drug, and include them as part of the approval package.

The Math: Billions in Waste

Let's do the arithmetic for just one drug.

Now extend this across the ten drugs in our table. Add the GLP-1 receptor agonists, where a 50-kg patient on the maximum dose of tirzepatide may be getting 40% more exposure than needed. At $1,000 per month, that's $4,800 per year in unnecessary drug cost per patient. With millions on GLP-1 agonists, the aggregate waste runs into the billions.

The counterargument is always cost of implementation: “TDM requires blood draws, laboratory assays, software licenses, clinical pharmacist time.” True. But the cost of one trough level ($50–$200) is trivial compared to months of unnecessary drug exposure at $10,000–$15,000 per month. The economics are not even close.

A Call to the Field

To pharmacometricians: We build these models. We know the variability. We publish the papers. But we too often stop at the regulatory submission. The models we validate for label support can be the same models that individualize doses in clinical practice. We need to advocate for that translation — loudly.

To regulators: ICH M15 is a step forward. But labels need to go further. Include dosing algorithms. Reference validated software. Define acceptable dose adjustment ranges. Make model-based dosing part of the approval, not an afterthought.

To payers: Reimburse the blood draw. Reimburse the software. Reimburse the pharmacist's time. The math shows you'll save orders of magnitude more on drug costs. And start covering off-label dose adjustments when they're supported by validated pharmacokinetic models.

To clinicians: Ask about drug levels. Ask about exposure. The tools exist — DoseMeRx, InsightRX, ID-ODS, MwPharm — and they're getting better. A single blood draw can transform empiric dosing into precision dosing.

To patients: You deserve to know that the dose written on your prescription was designed for “most people,” not for you. Ask your doctor whether your drug levels have been checked. Ask whether your dose is right — not just right for the average patient, but right for you.

We have the models. We have the math. We have the technology. What we lack is the will to change a system that prioritizes simplicity over precision, convenience over outcomes, and uniformity over individuality.

The dose is broken. Let's fix it.

References

1. Schneck KB, et al. Population pharmacokinetics of the GIP/GLP receptor agonist tirzepatide. CPT Pharmacometrics Syst Pharmacol. 2024;13(4):645-657. PMID: 38356317
2. Habtemariam BA, et al. Single-Dose Pharmacokinetics and Pharmacodynamics of Transthyretin Targeting N-acetylgalactosamine–Small Interfering Ribonucleic Acid Conjugate, Vutrisiran, in Healthy Subjects. Clin Pharmacol Ther. 2021;109(2):372-382. PMID: 34227100
3. Li H, et al. Evaluation of dosing strategy for pembrolizumab for oncology indications. J Clin Pharmacol. 2017;57(10):1260-1269. PMID: 28515943
4. Overgaard RV, et al. Clinical Pharmacokinetics of Oral Semaglutide. Clin Pharmacokinet. 2021;60(10):1335-1348. PMID: 33150578
5. Suleiman AA, et al. Population Pharmacokinetics of Risankizumab in Healthy Volunteers and Subjects with Moderate to Severe Plaque Psoriasis. Clin Pharmacokinet. 2019;58(6):793-803. PMID: 30945127
6. Vande Casteele N, et al. Trough concentrations of infliximab guide dosing for patients with inflammatory bowel disease (TAXIT). Gastroenterology. 2015;148(7):1320-1329. PMID: 25935633
7. Widmer N, et al. Population pharmacokinetics of imatinib and the role of alpha-acid glycoprotein. Br J Clin Pharmacol. 2006;62(1):97-112. PMID: 16869814
8. Staatz CE, Tett SE. Clinical pharmacokinetics and pharmacodynamics of tacrolimus in solid organ transplantation. Clin Pharmacokinet. 2004;43(10):623-653. PMID: 15217303
9. Rybak MJ, et al. Therapeutic monitoring of vancomycin for serious methicillin-resistant Staphylococcus aureus infections: A revised consensus guideline. Am J Health-Syst Pharm. 2020;77(11):835-864. PMID: 32191811
10. Frost CE, et al. Apixaban, an oral, direct factor Xa inhibitor: single-dose safety, pharmacokinetics, pharmacodynamics, and food effect in healthy subjects. Br J Clin Pharmacol. 2013;75(2):476-487. PMID: 24214379
11. Bruno R, et al. Population pharmacokinetics of trastuzumab in patients with HER2+ metastatic breast cancer. Cancer Chemother Pharmacol. 2005;56(4):361-369. PMID: 15838653
12. Bartelink IH, et al. Body weight-dependent pharmacokinetics of busulfan in paediatric haematopoietic stem cell transplantation patients. Clin Pharmacokinet. 2012;51(5):331-345. PMID: 19650675
13. Lee JY, et al. Impact of pharmacometric analyses on new drug approval and labeling decisions: a review of 198 submissions between 2000 and 2008. Clin Pharmacokinet. 2011;50(10):627-635. PMID: 21895036
14. Yu RZ, et al. Population Pharmacokinetic–Pharmacodynamic Modeling of Inotersen, an Antisense Oligonucleotide for Treatment of Patients with Hereditary Transthyretin Amyloidosis. Nucleic Acid Ther. 2020;30(3):153-163. PMID: 31730427
15. ICH M15. General Principles for Model-Informed Drug Development. International Council for Harmonisation. 2024.
16. Mita MM, et al. Dose Determinations at Drug Approval Reviews: FDA-Approved Drugs in Past 5 Years. Clin Pharmacol Ther. 2025. DOI: 10.1002/cpt.3611