Treatment Interaction Analysis¶
Method — instantiates Interaction Effect Mapping
A method for evaluating whether an intervention's effect changes under different co-treatments, conditions, populations, or moderators.
Treatment Interaction Analysis is the method that asks not does this intervention work? but for whom, and alongside what, does its effect change? Its defining move is to estimate one treatment's effect within strata of a moderator — a co-treatment, a subgroup, a condition — and to test whether those stratum-specific effects genuinely differ. This is effect modification: the treatment's benefit is not a single number but a function of context. Crucially, the moderator is often observed rather than manipulated — patients arrive already differing in biomarker, comorbidity, or co-medication — so the method leans on statistical comparison of subgroup effects rather than on a fully crossed experimental grid. It is the mechanism that catches a treatment that helps on average but harms a subgroup, or that only works in the presence of an enabling co-treatment.
Example¶
A trial has tested a smoking-cessation drug against placebo and found a modest average quit-rate benefit. Treatment Interaction Analysis asks whether that average hides variation. The analysts stratify the effect by a plausible moderator: whether each participant also received behavioral counseling. Within the counseling stratum the drug's benefit is large; within the no-counseling stratum it nearly vanishes. They fit an interaction term (drug × counseling) and test it — the effect difference across strata is unlikely to be chance, so this is a real quantitative interaction: the drug needs the co-treatment to work.
They probe a second moderator, a metabolizer-status subgroup, and find something sharper — a qualitative interaction, where the drug helps fast metabolizers but appears to slightly worsen outcomes in slow metabolizers, a sign reversal rather than a size change. Each finding is tagged with its confidence, and the qualitative one is flagged as tentative pending replication because subgroup reversals are exactly the kind of result that evaporates on retest. The conclusion is not "the drug works" but "it works with counseling and in fast metabolizers" — an attribution the average effect concealed.
How it works¶
- Name the moderators. Specify, ideally in advance, the co-treatments, conditions, or subgroups whose presence might change the treatment's effect.
- Estimate the effect within strata. Compute the treatment effect separately in each moderator stratum, on a chosen outcome measure.
- Test the interaction, not the subgroups. Fit and test an explicit interaction term for the difference between stratum effects — a difference of effects, not a comparison of raw subgroup means.
- Classify and rate. Label the interaction quantitative (size change) or qualitative (direction reversal), and attach a confidence rating that accounts for multiplicity and confounding.
Tuning parameters¶
- Pre-specified vs. exploratory moderators — moderators named before seeing data yield trustworthy tests; those hunted afterward inflate false positives. Pre-specification trades discovery breadth for credibility.
- Multiplicity control — how aggressively p-values or intervals are adjusted for the number of subgroups probed. Tighter control suppresses spurious interactions but can miss real ones.
- Interaction scale — whether effects are compared on an additive (risk-difference) or multiplicative (ratio) scale; an interaction can appear on one scale and vanish on the other, so the scale must match the decision.
- Confounding adjustment — for observed (non-randomized) moderators, how thoroughly co-varying differences are controlled before a stratum difference is called an interaction.
- Replication requirement — how much independent confirmation a subgroup finding needs before it drives a decision, especially for qualitative reversals.
When it helps, and when it misleads¶
Its strength is improved attribution: it dissolves the false "it worked / it failed" verdict by showing that a treatment's effect is conditional, and it is the standard way to detect an enabling co-treatment or a harmed subgroup that an average effect buries.[1] It is indispensable in clinical, behavioral, and policy settings where populations and co-interventions vary.
Its failure mode is spurious interaction from subgroup fishing: slice a dataset enough ways and some stratum will show a "significant" effect difference by chance alone, and observed moderators invite confounding that mimics a real interaction. The classic misuse is the data-dredged subgroup claim — a post-hoc split presented as a discovery — which is why pre-specification, multiplicity control, and replication are the discipline. The guarding move is to test the interaction term rather than eyeballing subgroup effects, to treat unplanned subgroups as hypothesis-generating only, and to hold qualitative reversals to a higher evidentiary bar.
How it implements the components¶
effect_measurement— estimates the treatment's effect on a defined outcome within each moderator stratum.context_stratification— partitioning by co-treatment, condition, or subgroup is the method's core: effect within context, not on average.interaction_type_classification— labels the result quantitative (size change) or qualitative (reversal), the design-relevant distinction.evidence_confidence_rating— rates each interaction for confidence, discounting for multiplicity and confounding.
It does not implement factor_level_set or interaction_matrix — systematic level-crossing and the coverage design belong to Factorial Experiment and Design of Experiments Protocol; this method estimates effect modification across a (frequently observed) moderator rather than manipulating a full crossing.
Related¶
- Instantiates: Interaction Effect Mapping — supplies moderation and effect-modification findings to the map.
- Sibling mechanisms: Factorial Experiment · Pairwise Combination Testing · Interaction Matrix Table · Compatibility Screening · Dependency Interaction Map · Design of Experiments Protocol
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Treatment Interaction Analysis operates as an analytical, modeling, inference, comparison, or optimization procedure that derives insight or a solution because it a method for evaluating whether an intervention's effect changes under different co-treatments, conditions, populations, or moderators.
Independent corroboration: The frozen evidence defines Treatment Interaction Analysis as 'A method for evaluating whether an intervention's effect changes under different co-treatments, conditions, populations, or moderators', so its operative form is Analysis, Modeling & Optimization.
Nearest alternative: Assessment, Review & Assurance — Treatment Interaction Analysis includes features of a bounded evaluation of existing evidence or work that produces a finding or disposition, but its defining operation is an analytical, modeling, inference, comparison, or optimization procedure that derives insight or a solution.
Review outcome: Independent reviewer agreement; medium confidence.
Origin Attribution¶
Primary origin: Medicine & Healthcare
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Specialized
Rationale: Both independent reviews identify medicine healthcare as the historical home of the operation—A method for evaluating whether an intervention's effect changes under different co-treatments, conditions, populations, or moderators.. The retained alternates document formative adjacent traditions; the reach field, not the origin field, carries later applicability.
Related originating lineages:
- Psychology — Experimental, clinical, and behavioral psychology supplies a parallel or contributing lineage for the mechanism's defining operation: a method for evaluating whether an intervention's effect changes under different co-treatments, conditions, populations, or moderators.
- Statistics & Experimental Design — Statistics, experimental design, and measurement theory supplies a parallel or contributing lineage for the mechanism's defining operation: a method for evaluating whether an intervention's effect changes under different co-treatments, conditions, populations, or moderators.
Review resolution: Both blind reviewers independently place the defining operation—A method for evaluating whether an intervention's effect changes under different co-treatments, conditions, populations, or moderators.—in medicine healthcare. Their queued differences are secondary: alternate_origin_disagreement, origin_mode_disagreement, domain_reach_disagreement, encyclopedia_synthesis_disagreement. Reviewer A contributes no unique alternate; reviewer B contributes ['psychology']. I preserve the full evidence-supported union of 2 alternate domain(s), without a numeric cap. origin_mode=cross_disciplinary_synthesis reflects the reviewers' evidence about historical construction, while domain_reach=specialized separately reflects present-day portability. The affirmative encyclopedia-synthesis finding is preserved, and confidence=high uses the more conservative reviewer level.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Reconciled after independent review; high confidence.
References¶
[1] VanderWeele, T. J. Explanation in Causal Inference: Methods for Mediation and Interaction. Oxford University Press (2015). Explains interaction as conditional or heterogeneous treatment effect and describes methods for identifying subgroups for whom treatment is beneficial or harmful. registry ↩