Drug combinations need a four-arm factorial trial to prove synergy

What separates additive, synergistic and antagonistic drug combinations?

Additive, synergistic, and antagonistic describe where a combination's measured effect lands against the predicted sum of each drug's solo effect: on it, above it, or below it. The question is quantitative. Given what compound A does alone and what compound B does alone, an additive reference model predicts the combined effect if the two simply add together. Synergy is the specific, testable claim that the measured combined effect exceeds that prediction.

A stack page that draws two mechanism arrows converging on a pathway and calls it evidence is answering a question nobody asked. Mechanistic overlap between two compounds is a hypothesis about synergy, not a measurement of it. A shared pathway is consistent with all three outcomes and does not point toward any one of them. How two interventions antagonize each other is covered separately.

Trials built specifically to isolate a combination effect exist because researchers cannot assume synergy from mechanism alone; they have to test for it. The four-arm PEACE-1 trial, testing abiraterone added to standard androgen-deprivation therapy and docetaxel, is one such trial ^[6].

What trial design shows whether a drug combination works?

A factorial trial is the only design built to isolate a drug combination's effect. It randomizes participants across every combination of the variables in question, not just to the combined arm. For two agents, that means four arms: agent A alone, agent B alone, both together, and neither. Without those four arms, a combination effect has not been isolated from the effect of either ingredient by itself ^[6]⁠^[7].

PEACE-1 shows the structure at work. The trial randomized men with metastatic, hormone-sensitive prostate cancer 1:1:1:1 to standard of care, standard of care plus radiotherapy, standard of care plus abiraterone, or standard of care plus both, across 77 hospitals in seven countries ^[6]. That four-arm structure is what later let investigators separate radiotherapy's contribution from abiraterone's contribution and from the combination of the two ^[7].

A two-arm trial of one agent alone versus both agents cannot do this. Without the "neither" and "other agent alone" arms, an apparent benefit of the pair could belong entirely to one ingredient. Almost no compound stack circulating in unsourced protocols has ever had the four-arm structure applied to it, for benefit or for harm.

What do researchers do when no combination trial exists?

Researchers fall back on indirect or network comparisons when no head-to-head or factorial trial of a specific combination exists. These methods pool separate trials of each component against a shared reference and model the gap statistically. Fixed-dose combinations of long-acting muscarinic antagonists and long-acting beta-agonists have been compared to one another this way, because none had been tested directly against each other ^[3].

Aclidinium plus formoterol was evaluated against tiotropium the same way, through an indirect treatment comparison built on separate placebo-controlled trials ^[4]. Inhaled corticosteroid plus long-acting beta-agonist combinations were assessed as a class through Bayesian network meta-analysis rather than pairwise testing ^[5]. Radiotherapy regimens paired with different breathing techniques have been compared the same way when direct trials were sparse ^[2].

These are legitimate methods, but they carry explicit uncertainty that a direct trial does not need: credible intervals, and the assumption that the underlying trials are similar enough to pool. They also require a network of individual-component trials to exist in the first place. Most consumer stacks have neither the direct trial nor the network that would feed an indirect one.

Does starting each compound at half dose make a stack safe?

Halving each compound's dose does not make a stack safe; it addresses only one failure mode, overshooting a dose ceiling defined for that compound alone. It does nothing for a shared clearance bottleneck, where halving each dose can still leave a shared enzyme or transporter as saturated as either full dose would. It also does nothing for two agents pushing the same physiological axis in the same direction, where the combined effect is not a simple sum.

Dose-response relationships have to be mapped explicitly to know where a ceiling or threshold sits ^[1]. When that mapping has been done properly for an intervention, as in a network meta-analysis modeling exercise dose against BMI change in adolescents, it takes dozens of pooled trials and a formal statistical model to produce a defensible curve ^[1]. No such curve exists for two arbitrary compounds combined at arbitrary fractions of their solo doses.

Are two individually well-tolerated compounds safe together?

Two individually well-tolerated compounds are not automatically safe together, because solo safety data test each compound in isolation. Those data say nothing about shared clearance enzymes, shared transporters, or additive load on the same organ, which show up only when the two are taken together. A separate analysis covers why multi-ingredient stacks need evidence per component.

Three checkpoints account for most of the trouble two compounds cause together, and none requires a dedicated trial to check, only pharmacology:

• Shared clearance. Both compounds route through the same enzyme system or transporter, which turns combined use into competition rather than two independent processes.
• Opposing or overlapping direction. Both push the same physiological axis, such as blood pressure, blood glucose, heart rate, or a receptor's desensitization curve, so the combined effect is not a simple sum.
• Additive organ load. Even mechanistically distinct compounds can stack pressure on the same liver or kidney pathway.

Formal trials that isolate combination effects build safety monitoring around this kind of overlap. PEACE-1 tracked toxicity by arm, because combining docetaxel, abiraterone, and radiotherapy meant stacking load on overlapping systems that no single-agent profile could predict on its own ^[6]⁠^[7].

Rank a stack by how many of these three checkpoints it fails, not by how many citations back each ingredient individually. A pairing with a thin evidence base but no shared clearance, no opposing direction, and no overlapping organ load is lower risk than two heavily documented compounds that share all three.

Does a lack of bad-interaction reports mean a stack is safe?

A lack of bad-interaction reports is weak reassurance at best, because a combination almost nobody has systematically tracked produces an absence of data, not evidence of a flat safety profile. For the overwhelming majority of stacks discussed outside clinical research, none of the machinery of combination testing has been applied: no factorial trial, and no network of individual trials large enough to support even an indirect comparison ^[2]⁠^[3]⁠^[4]⁠^[5].

Both synergy and antagonism therefore remain open possibilities, not the benign "probably additive, probably fine" default that stack pages imply. Treating an absence of reports as reassurance inverts the evidence hierarchy.

Until a specific pairing clears the three checkpoints as low-risk, and ideally has been examined in a design built to isolate a combination effect, the accurate label for its claimed benefit is untested, not synergistic.