Motility-mediated drug interactions with GLP-1 receptor agonists: labeled evidence and a framework for quantitative assessment

Abstract

Glucagon-like peptide-1 receptor agonists (GLP-1 RAs) are widely prescribed, and two additional oral products received US approval in 2025 and 2026. All eleven US labels examined describe delayed gastric emptying as potentially affecting the absorption of co-administered oral medications. Since the exenatide program, no study has evaluated a narrow therapeutic index victim across more than one administration schedule; the original warfarin and digoxin studies each used a single schedule. The omission may reflect an interaction paradigm built around enzymes and transporters. Current frameworks do not routinely trigger motility-specific studies because the standard in vitro screening cascade does not address gastric emptying. Total area under the concentration–time curve (AUC) may also fail to characterize rate-dependent effects in some victim drugs. Once a class warning is established, further study may receive less priority, and interactions managed by routine dose adjustment may go unreported. Of eleven US prescribing-information documents screened, six contributed quantitative data showing that the direction of effect is not uniform: thyroxine, atorvastatin and ramipril exposures have all been reported to increase under specific conditions, so a reduction in exposure should not be assumed. The class is also pharmacologically heterogeneous; ICH M12 places absorption-mediated interactions outside its scope, and the European Medicines Agency has adopted a concept paper proposing guidance covering them. I propose an evaluation framework built on biopharmaceutically stratified probe panels, timing- and titration-explicit designs, selective use of prespecified partial AUC endpoints for rate-sensitive victims, and physiologically based pharmacokinetic models qualified against human gastric emptying data.

INTRODUCTION

The glucagon-like peptide-1 receptor agonist (GLP-1 RA) class now includes twice-daily and once-weekly peptides, a peptide reformulated for oral absorption with a permeation enhancer, and a non-peptide small molecule. Every member shares one pharmacodynamic property with potential pharmacokinetic consequences for other drugs: delayed gastric emptying. It is part of the intended effect rather than an off-target one.

In the United States in 2024, 26.5% of adults with diagnosed diabetes—about 6.9 million people—used an injectable GLP-1 RA, a figure that excludes oral products and use for obesity without diabetes [1]. The number who also take a drug for which a change in absorption could matter clinically has not been quantified, but at this scale substantial overlap with warfarin, levothyroxine, digoxin, immunosuppressants, antiepileptics and oral contraceptives is likely. The labels state the mechanism; the quantitative data are sparse and unevenly informative.

WHAT THE LABELS CONTAIN

Eligible documents were the current US prescribing information for each distinct GLP-1 RA molecule and formulation, together with fixed-dose combinations containing a GLP-1 RA; brand-name variants of the same molecule and formulation were represented once. Eleven documents were screened on 5 August 2026 [2,3,4,5,6,7,8,9,10,11,12]. Inclusion in Table 1 required both a quantitative exposure change for a co-administered oral drug and attribution of that change to delayed gastric emptying. Six met the criterion [2,3,4,5,6,7].

Table 1. Quantitative interaction data attributed in US labeling to delayed gastric emptying.

Perpetrator (dose, phase)	Victim (dose)	Timing	Cmax	AUC	Tmax
Exenatide 10 µg BID (BYETTA)	Acetaminophen 1,000 mg	1 h before	nc	nc	nc
		Concurrent	↓37%	↓21%	0.6 → 0.9 h
		1 h after	↓56%	↓23%	→ 4.2 h
		2 h after	↓54%	↓24%	→ 3.3 h
		4 h after	↓41%	↓14%	→ 1.6 h
	Lovastatin 40 mg, SD	30 min after	↓28%	↓40%	+4 h
	Digoxin 0.25 mg QD, SS	30 min after	↓17%	nc	+2.5 h
	Warfarin	-	nr	nr	delayed ~2 h; INR nc†
Exenatide ER 2 mg weekly, week 14 (BYDUREON BCISE)	Acetaminophen 1,000 mg	Fasting	↓16%	ns	1.0 → 1.4 h
		Fed	↓5%	ns	1.0 → 1.3 h
Lixisenatide 10 µg (ADLYXIN)	Acetaminophen 1,000 mg	1 h before	nc	nc	nc
		1 h after	↓29%	nc	+2.0 h
		4 h after	↓31%	nc	+1.75 h
	EE 0.03/LNG 0.15 mg	1 h after	EE ↓52%, LNG ↓46%	nc	+1–3 h
		4 h after	EE ↓39%, LNG ↓20%	nc	+1–3 h
		1 h before / 11 h after	nc	nc	nc
Lixisenatide 20 µg (ADLYXIN)	Atorvastatin 40 mg	Both AM, 6 d	↓31%	nc	+3.25 h
		ATV PM/LIX AM	↑66%	↑27%	nc
	Ramipril	-	↓63%	↑21%	nr
	Ramiprilat (active metabolite)	-	nc	nc	nr
Tirzepatide 5 mg, 1st dose (MOUNJARO)	Acetaminophen	Concurrent	↓50%	nc (0–24 h)	+1 h
	EE 0.035/norgestimate 0.25 mg	Concurrent	EE ↓59%, NGM ↓66%, NGMN ↓55%	↓20%, ↓21%, ↓23%	nr
Tirzepatide, week 4	Acetaminophen	Concurrent	nc	nc	nc
Oral semaglutide, SS (RYBELSUS)	Levothyroxine 600 µg, SD	Concurrent	nc	↑33% * (90% CI, 1.25–1.42)	nr
Orforglipron 0.8 mg, 1st dose (FOUNDAYO)	Acetaminophen	Concurrent	↓28%	nr	nr
Orforglipron 17.2 mg, repeat	Acetaminophen	Concurrent	Effect diminished (not quantified)†	nr	nr
Orforglipron	Oral contraceptive	-	Not studied	Not studied	Not studied

All entries are from the US prescribing information of the product named in the first column [2,3,4,5,6,7,8,9,10,11,12], accessed 5 August 2026, as presented in the labels.

nc, the label states the parameter was not meaningfully changed—an assertion by the label, not a demonstration of equivalence; ns, the label reports no significant change, a statement about statistical significance rather than about meaningful change; nr, not reported, whether or not measured; not studied, no interaction study was conducted.

C_max, maximum plasma concentration; AUC, area under the concentration–time curve; T_max, time at which C_max occurred; BID, twice daily; SD, single dose; QD, once daily; SS, steady state; ER, extended-release; EE, ethinyl estradiol; LNG, levonorgestrel; NGM, norgestimate; NGMN, norelgestromin; ATV, atorvastatin; LIX, lixisenatide; INR, international normalized ratio; CI, confidence interval.

For T_max, ‘+x h’ is the reported prolongation relative to the comparator arm and ‘→’ the absolute value under co-administration. Subjects were healthy unless the label specifies otherwise. Prandial state is shown only for the exenatide extended-release study, the one program that reported it for both conditions.

^*Given by the label as a geometric mean ratio with its 90% confidence interval. ^†Narrative descriptor rather than a numerical result. All other unmarked C_max and AUC entries are percentage changes for which the label does not state the summary statistic; these should not be assumed to be geometric mean ratios. Bold marks the entries discussed in the text as showing a change in total exposure.

The other five did not, for two reasons. Three report co-administration data, including for narrow therapeutic index victims, but attribute no quantitative change to the mechanism: liraglutide reports a 16% lower digoxin area under the concentration–time curve (AUC) with a 31% lower maximum plasma concentration (C_max) and a 37% higher griseofulvin C_max, while dulaglutide and injectable semaglutide evaluated warfarin, digoxin and atorvastatin, present the results graphically, and conclude that absorption was not affected to a clinically relevant degree [8,9,10]. Two report no data of their own: both fixed-dose combinations with insulin give the interaction information of the component GLP-1 RA [11,12]. Table 1 therefore samples the labels making a mechanistic attribution and does not estimate class-wide risk.

First, the magnitude of the reported effect varied with the timing of administration. In the exenatide–acetaminophen dataset the same perpetrator, dose and subjects showed no reported change when the probe preceded the injection by an hour; when it followed by an hour, C_max fell 56% and time at which C_max occurred (T_max) lengthened sevenfold. In the lixisenatide–atorvastatin dataset the reported direction differs between schedules.

Second, acetaminophen is expected to be insensitive to changes in the extent of absorption, though not to changes in its rate. It is highly permeable, freely soluble and absorbed essentially completely on reaching the duodenum. Its AUC was nevertheless 14–24% lower when it followed immediate-release exenatide, the only program reporting an AUC change under post-dose administration; the lixisenatide program tested comparable schedules and reported none.

Third, the orforglipron label contains limited interaction data. It reports a 28% lower acetaminophen C_max after the first dose, does not report AUC, and states that no oral contraceptive study was conducted [7].

SHORT- AND LONG-ACTING AGENTS DIFFER IN GASTRIC-EMPTYING EFFECTS

The labels leave one distinction implicit. Continuous GLP-1 receptor stimulation produces tachyphylaxis of the emptying effect whereas intermittent stimulation does not, so prolongation of emptying half-time declines over time for long-acting but not short-acting agents [13]. Pooled scintigraphic data place the average delay at about 36 minutes (95% confidence interval, 17–55) [14].

Most timing-resolved data in Table 1 come from short-acting agents. For long-acting agents the labeled data are consistent with attenuation: tirzepatide reduces acetaminophen C_max by 50% at the first dose but has no meaningful effect by week 4 [5,13], and the effect of orforglipron was largest after the first dose and diminished with repeated dosing [7]. The clearest comparison holds the molecule constant. Immediate-release exenatide lowered acetaminophen C_max by 37–56% depending on timing, whereas the extended-release formulation of the same peptide lowered it by 16% fasting and 5% fed after 14 weeks of treatment, with no significant AUC change reported under either condition [2,3]. The largest pharmacokinetic effect may therefore occur during initiation and dose escalation rather than maintenance, a hypothesis that remains untested.

Attenuation may not be complete. Before anesthesia, increased residual gastric content was seen in 56% of GLP-1 RA users versus 19% of non-users, and withholding the drug for seven days did not abolish the difference [15]. Residual content is a marker of gastric function rather than of drug absorption, but the finding argues against assuming full normalization during maintenance.

The studies in Table 1 were predominantly in healthy volunteers, whereas baseline emptying varies widely between individuals and is altered in long-standing diabetes [16], so the same delay may produce different exposure changes.

WHY THE CURRENT PARADIGM DOES NOT TRIGGER THESE STUDIES

Four features of how interaction assessment is organized may explain the limited evidence summarized in Table 1.

The standard screening cascade provides no gastric-emptying trigger. For enzyme- and transporter-mediated interactions, contemporary guidance begins with in vitro characterization (Fig. 1, left). Those assays do not address gastric emptying and so provide no analogous trigger. A sponsor may study the interaction clinically, but the guidance reviewed here sets out no criterion that would routinely prompt one. ICH M12 states that interactions affecting absorption, “e.g., gastric pH change, gastric motility change, formation of chelation or complexation… are not part of this document and may be covered by regional guidances” [17], and FDA guidance on combined oral contraceptives likewise notes that absorption-based mechanisms are not addressed there [18]. Acid-reducing agents are a partial regional precedent, handled separately because they have an in vitro correlate in the solubility–pH profile of the victim [19].

Figure 1. Current (left) and proposed (right) pathways for interaction assessment. The dashed arrow indicates iterative refinement of PBPK predictions using clinical concentration–time data.

CYP, cytochrome P450; PBPK, physiologically based pharmacokinetic; R values, the basic-model indices defined in ICH M12 for assessing enzyme or transporter inhibition or induction potential; cut-off criteria, the thresholds used to determine whether further modeling or clinical evaluation is warranted [17]; AUC, area under the concentration–time curve; C_max, maximum plasma concentration.

Such studies are therefore discretionary. In April 2025 the European Medicines Agency adopted a concept paper proposing a guideline on gastrointestinal drug interactions that includes gastric emptying and motility [20]; no draft has been released.

Total-exposure metrics may not characterize rate effects for some victims. Interaction assessments commonly emphasize AUC ratios, interpreted against prespecified no-effect boundaries; C_max is also evaluated, and for some victims, including combined oral contraceptives, both are recommended [18]. Labeling decisions nonetheless often emphasize total exposure. Yet Table 1 contains nine labeled observations in which AUC was reported as unchanged or not significantly changed despite C_max reductions of 5–52% and T_max delays of 0.3–3.25 hours.

For victims without a rate-dependent exposure–response relationship, a change in absorption rate with preserved total exposure may not be clinically relevant; the concern applies to those whose effect depends on the concentration reached within a defined interval, and the labeled data do not identify which they are.

Class labeling may reduce the perceived urgency for new data. A similar class statement appears across the labels examined, so the mechanism is described and prescribers cautioned. Whether this is why no further study was required cannot be established from the public record, but over the following two decades no purpose-designed, timing-resolved study of a narrow therapeutic index victim was reported beyond the original exenatide warfarin and digoxin studies. The oral semaglutide–levothyroxine study [6] is relevant evidence, but its single-dose, single-timing design neither resolves timing dependence nor isolates the mechanism.

Spontaneous reporting is poorly suited to detecting this interaction. When INR changes after GLP-1 RA initiation the warfarin dose may be adjusted, and when TSH changes the levothyroxine dose may be adjusted; that such adjustments go unreported is an inference rather than a documented finding. A real-world analysis found greater INR variability after GLP-1 RA initiation [21], and a case report described suppressed TSH in a levothyroxine-treated patient [22]. Neither establishes the mechanism, since appetite, vitamin K intake and weight all change at once.

THE DIRECTION OF EFFECT IS NOT UNIFORM

Labels generally emphasize reduced absorption, but three labeled observations show increased AUC under specific conditions (Table 1): thyroxine AUC rose 33% with oral semaglutide; atorvastatin AUC and C_max rose by 27% and 66% under a staggered schedule with lixisenatide, whereas concurrent morning dosing lowered C_max and left AUC unchanged; and ramipril AUC rose 21% while C_max fell 63%. The liraglutide label, excluded from Table 1 because it makes no mechanistic attribution, reports a 37% higher griseofulvin C_max with unchanged AUC and unchanged T_max [9]; an increase in C_max is as hard to predict in advance as a decrease.

Each requires qualification. The levothyroxine study used a supratherapeutic 600 µg single dose against a variable endogenous background and cannot distinguish delayed emptying from other semaglutide-mediated effects. The ramipril observation says little about clinical relevance, since exposure to the active metabolite ramiprilat was unchanged. The atorvastatin finding, a within-study comparison of schedules, is the most interpretable.

One hypothesis is that drugs with a narrow proximal absorption window, or with dissolution- or solubility-limited absorption, may be absorbed more completely when gastric residence is prolonged, while for highly permeable, freely soluble drugs the likelier consequence is a lower and later peak. Both predictions remain untested, so direction should be assessed for the individual victim and schedule rather than assumed.

SMALL-MOLECULE AGONISTS ADD A SECOND MECHANISM

Peptide GLP-1 RAs act mainly as perpetrators rather than victims. They are cleared by proteolysis and are not reported to inhibit or induce cytochrome P450 enzymes or major transporters. Their main established interaction mechanism is therefore altered gastrointestinal transit.

Orforglipron is different. It is a substrate of CYP3A4 and CYP2J2 and of P-glycoprotein, OATP1B1 and OATP1B3, and its victim-side interactions are substantial: AUC rises 3.5-fold with clarithromycin and 2.6-fold with cyclosporine and falls 82% with carbamazepine, so the dose is capped at 9 mg with a strong CYP3A4 inhibitor [7]. It also shows perpetrator effects the label does not explain mechanistically: simvastatin acid AUC rose 2- to 2.5-fold while simvastatin lactone AUC fell, and rosuvastatin and atorvastatin AUC rose 1.7- and 1.5-fold. The simvastatin acid effect persisted despite two-hour staggering and occurred without in vitro OATP1B inhibition or a change in coproporphyrin-I [7]. Delayed absorption alone does not explain these findings.

Evaluation should therefore follow two paths. Peptide agents primarily require assessment as perpetrators of absorption-mediated interactions; small-molecule agents additionally require conventional evaluation as victims and perpetrators of enzyme- and transporter-mediated interactions.

A PROPOSED ASSESSMENT FRAMEWORK

Measure the perpetrator effect first (Fig. 1, right). Quantify the emptying delay by scintigraphy or a validated breath test in the intended population at the first dose, during escalation and during maintenance.

Then select biopharmaceutically stratified victims. A single probe cannot characterize a motility-mediated interaction, because the properties of the probe itself strongly influence the direction and magnitude of the effect (Table 2). No validated probe panel exists; the compounds are chosen because their biopharmaceutical properties should produce distinguishable signals, not because any has been qualified. Tier 2 is not represented in current label studies and is intended to test whether delayed delivery can increase AUC. Tier 5 will rely more on patient-based evidence—therapeutic drug monitoring records, INR time in therapeutic range, registries—because repeated volunteer studies with these drugs are hard to justify.

Table 2. A hypothesis-driven candidate probe panel for motility-mediated interaction assessment.

Tier	Property	Expected signal	Candidate probes
1	High permeability, high solubility	Cmax ↓, Tmax ↑; AUC generally preserved, but reductions possible with some post-dose schedules	Acetaminophen
2	Narrow proximal absorption window	AUC increase possible	Riboflavin, gabapentin
3	Solubility- or dissolution-limited	Direction schedule-dependent	Atorvastatin
4	Administration- and food-sensitive absorption	Direction uncertain; clinically consequential	Levothyroxine
5	Narrow therapeutic index	Consequential shift in either direction	Tacrolimus, warfarin

Tier assignments are a heuristic, not a taxonomy, and the tiers are not mutually exclusive: tacrolimus and warfarin also carry tier 3 and 4 properties, and for tier 5 the consequence of a given shift depends on indication and therapeutic range. Tier 2 is hard to interpret, because both candidate probes have saturable uptake and slower delivery to the absorption site may relieve saturation whether or not gastric residence is prolonged; a tier 2 study would therefore need dose-ranging rather than a single dose. Levothyroxine (tier 4) is constrained by the endogenous baseline, by a week-long half-life that makes adequate terminal sampling impractical, and by the supratherapeutic single dose such studies require. Digoxin was not selected for tier 5 because P-glycoprotein substrate status does not by itself make a drug motility-sensitive, and exenatide left its steady-state AUC unchanged [2].

C_max, maximum plasma concentration; T_max, time at which C_max occurred; AUC, area under the concentration–time curve.

Then use PBPK to simulate the panel and rank the studies. Whole-body PBPK platforms represent gastric emptying, intestinal transit and regional absorption explicitly, and two analyses have applied them in this setting [23,24]. A model can be run on assumed emptying times, but to support decisions it should be calibrated to the product- and phase-specific data from the first step. Qualification requires, at minimum, victim-specific dissolution, solubility, permeability and regional absorption inputs with sensitivity analyses for each; the measured distribution of emptying times rather than a mean, represented in virtual populations [16]; separate evaluation of prandial state and of clinically relevant subgroups; external evaluation against full concentration–time profiles; and a prospectively declared success criterion and intended use. Table 1 can serve as a plausibility check on predicted direction and magnitude but not as a validation set, because summary ratios from labeling lack the individual profiles that qualification requires.

Decision criteria between steps. Table 3 states each transition as a quantity to estimate and a decision rule on its interval, so that the criterion can be checked against data rather than asserted.

Table 3. Provisional criteria for moving between steps of the proposed framework.

Transition	Quantity estimated	Provisional decision rule
Measurement → victim selection and simulation	Ratio of gastric emptying half-time on drug to baseline, by scintigraphy or a validated breath test	The lower bound of the 90% confidence interval exceeds 1.00, and the estimated proportional increase exceeds a repeatability threshold prespecified for the method
Simulation → clinical study, general victim	PBPK-predicted geometric mean ratios for total AUC and Cmax, each with a 90% prediction interval; partial AUC added only where a rate-dependent endpoint has been justified in advance	At least one prediction interval is not contained within the no-effect bounds set for that victim, or the predicted direction reverses across the prespecified sensitivity range
Simulation → clinical study, narrow therapeutic index victim	The same quantities, judged against bounds derived from the therapeutic range	The bounds are set a priori and are narrower than 80–125%; narrow therapeutic index narrows the bounds rather than triggering a study by itself
Clinical study → labeling	Observed geometric mean ratio and 90% confidence interval for each prespecified endpoint	An interval not contained within the bounds supports consideration of dosing or timing advice; an interval contained within them supports a statement that no change is required

Prediction intervals should be generated from the sources of uncertainty declared in the qualification plan—the measured between-subject distribution of emptying times, the confidence intervals of the victim-specific absorption inputs, and the prespecified sensitivity ranges—propagated through the virtual population, and not from between-subject variability alone. None of these bounds has been evaluated against clinical outcomes, and each would have to be set from exposure–response for the victim concerned. No acceptance interval has been established for partial AUC, which is why it enters the table only where a rate-dependent endpoint has been justified in advance. The table is offered so that the framework can be tested as a decision procedure rather than adopted as a standard.

AUC, area under the concentration–time curve; C_max, maximum plasma concentration.

Rate-sensitive endpoints. No generally accepted threshold links a partial AUC change to a clinical consequence in this setting, and the regulatory precedent for this metric lies mainly in bioavailability and bioequivalence assessment [25]. Total AUC and C_max remain the primary endpoints. Partial AUC—truncated at a point prespecified from the observed T_max of the victim—is worth adding as a key secondary endpoint only where the victim has a plausible rate-dependent clinical endpoint or documented exposure–response. Any bound on it should be justified from the therapeutic range and exposure–response relationship of the victim rather than from the variability of the metric, which instead informs sample size; the 80–125% interval should not be applied by default. Where the effect is defined by a threshold concentration, time to reach or time above that threshold may be more interpretable. Tests should be two-sided.

Timing and titration as design factors. In the labeled studies, dosing one hour before the injection was associated with no detectable change and dosing one hour after with the largest effect; the older labels can therefore advise giving some concomitant drugs at least an hour before the injection, whereas the orforglipron label lacks comparable data [2,4,7]. Labels state that the emptying delay is greatest after the first dose and diminishes over time, yet recommend contraceptive protection after each escalation [5,7]. That recommendation raises, without establishing, the hypothesis that escalation transiently restores the emptying effect; for tirzepatide, six approved dose levels would create five such periods. This is directly testable: measure a rate-sensitive probe at the first dose, at steady state, and immediately after the next escalation.

CONCLUSION

GLP-1 receptor agonists delay gastric emptying, and delayed emptying can alter oral drug absorption. What remains missing, on the evidence of the labels examined here, is systematic, timing-resolved quantification across victim drugs and treatment phases.

The work is timely because ICH M12 excludes this mechanism from its scope and the EMA is preparing regional guidance [17,20]; without supporting methods and data, such guidance may remain qualitative. Four directions would be particularly valuable: timing-resolved studies with stratified probes; prospective evaluation of the titration-window hypothesis; narrow therapeutic index data from patients in routine care; and PBPK frameworks able to predict direction of effect.

Population differences deserve separate attention. Baseline gastric emptying, meal composition and habitual dose timing relative to meals may differ between regions, as may the duration of diabetes and the prevalence of autonomic neuropathy in the treated population. Whether such differences are large enough to change the size or the direction of a motility-mediated interaction is untested. Regionally generated data would be informative on both questions: whether the emptying delay itself differs, and whether timing advice transfers between populations.

Use of the class continues to expand, increasing the need for systematic, timing-resolved assessment of these interactions.