Co-Intervention Bias: When the Treatment Arm Gets More Than the Treatment
Anas H. Alzahrani, MD PhD MPH
Department of Preventive Medicine and Public Health
Faculty of Medicine, King Abdulaziz University
A randomized heart-failure trial compares a new remote-monitoring program with usual care. Patients in the program arm receive more nurse calls, earlier diuretic changes, faster clinic appointments, and more home visits. Hospitalizations fall. Did the monitoring technology work—or did the treatment arm receive an entire care system that the control arm never had?
Co-intervention bias becomes a concern when additional treatment, monitoring, rescue care, or behavior differs between trial groups and can affect the outcome. Randomization balances prognosis at baseline. It does not guarantee equal post-randomization care when clinicians and participants know the assignment.
Randomization Balances the Starting Line, Not the Journey
Co-interventions include concomitant drugs, procedures, rehabilitation, counseling, surveillance, referrals, rescue medication, and other care received alongside the assigned intervention. Some are clinically routine. Some are triggered by side effects or lack of response. Others appear because an open-label trial changes how people behave.
The clean metaphor
Randomization deals two fair hands. It does not stop one table from receiving extra cards after the deal. Reviewers still need to see what each group was given during play.
The methodological trap is to label every imbalance “bias.” If the intervention is deliberately a care strategy—remote monitoring plus protocolized escalation, for example—those downstream care changes may be part of what assignment is intended to cause. The total effect of that strategy includes them.
Interactive co-intervention audit
Is the extra care bias, part of the strategy, or just background noise?
Classify one concomitant treatment, procedure, monitoring practice, or rescue intervention at a time. The same imbalance can mean different things depending on the trial question and why it occurred.
High risk, weakly auditable
Outcome-relevant care differs between groups, is not part of the intended strategy, and was not measured well enough to explain. Randomization does not equalize post-randomization care created by treatment knowledge or the trial context.
Review move: request group-specific type, timing, intensity, reasons, and site patterns before accepting the treatment-only story.
Teaching aid only. A full risk-of-bias judgment also considers blinding, adherence, outcome measurement, missing data, the effect of assignment versus adherence estimand, and whether deviations arose because of the trial context.
Three Questions, Three Different Conclusions
| Question | Role of extra care | Interpretation |
|---|---|---|
| Effect of assignment to the care strategy? | Expected downstream pathway | Usually remains inside the treatment-policy effect. |
| Effect of a product apart from unequal background care? | Competing explanation | The trial may not isolate the product effect. |
| Effect if participants adhered without rescue? | Intercurrent event | Requires a prespecified estimand and assumptions beyond naive adjustment. |
That distinction is why “adjust for concomitant medication” is not a universal remedy. Post-randomization care can be a mediator of assignment, a response to evolving prognosis, or both. Conditioning on it may remove part of the effect the trial was designed to estimate, open a biased path, or answer a hypothetical question the protocol never defined.
A Clinical Example: Rescue Therapy Is Not Just a Nuisance Variable
Scenario
An analgesic trial allows rescue medication. More control patients need rescue because pain remains severe. The primary pain score is lower in the experimental arm.
What not to say
“Rescue medication was imbalanced, so the randomized comparison is biased.” Rescue use may be a consequence of treatment effectiveness and part of the real clinical strategy.
What to ask instead
Does the estimand compare strategies regardless of rescue, count rescue as an unfavorable outcome, or ask what pain would have been without rescue? Each question needs different data and analysis.
Now change the story: investigators, convinced the experimental drug works, call those participants more often and prescribe unplanned physical therapy before outcome assessment. That differential care is not an intended component and can create a serious competing explanation.
Where Co-Intervention Problems Hide
Open-label enthusiasm
Clinicians monitor, counsel, refer, or rescue one group more intensively because they expect benefit or harm.
“Usual care” without content
The comparator is a label, not a measured bundle of medications, visits, expertise, and escalation rules.
Ever-versus-never reporting
Equal percentages conceal earlier starts, higher doses, longer exposure, or more intensive delivery in one arm.
Site-level bundles
Treatment expertise travels with better staffing, referral access, or follow-up systems, especially when sites favor one strategy.
What a Defensible Trial Reports
- Define permitted and prohibited care prospectively. Include rescue rules, modifications, procedures, and what “usual care” contains.
- Measure what could affect the outcome. Record type, timing, dose, intensity, duration, reason, and provider—not only whether it ever occurred.
- Report by randomized group. Overall totals can hide exactly the imbalance readers need to assess.
- Separate protocol from delivery. Describe what was intended, what participants received, adherence, provider fidelity, and deviations caused by the trial context.
- Align analysis with the estimand. Lead with the randomized treatment-policy effect when that is the target. Treat product-isolation or no-rescue questions as distinct estimands with explicit assumptions.
A Reviewer Red-Flag Checklist
- Could additional care plausibly change the endpoint or its measurement?
- Were clinicians, participants, or care coordinators aware of assignment?
- Does the paper report concomitant care separately by group, including timing and intensity?
- Is unequal care part of the protocol-defined strategy or an unintended trial-context deviation?
- Does the causal interpretation match the effect of assignment, adherence, or hypothetical no-rescue estimand actually analyzed?
- Did the authors adjust for post-randomization care without drawing its causal role first?
Reviewer red flag
The paper attributes the full group difference to the named treatment, calls the comparator “usual care,” and gives no group-specific account of monitoring, rescue treatment, concomitant medications, or provider contact.
Why This Matters for Aqrab
Co-intervention bias is a document-level critique problem. The result table cannot tell you what the treatment contrast contained. You have to connect the protocol, actual delivery, blinding, adherence, concomitant care, intercurrent events, and estimand before deciding what the trial identified.
Use Aqrab Try to pressure-test whether a trial's treatment label matches the care its groups actually received. The strongest critique does not merely spot imbalance. It explains whether that imbalance is bias, mechanism, or a different clinical question.
Methods Anchors
SPIRIT 2025 item 15d asks protocols to define relevant permitted and prohibited concomitant care and plans to record it. CONSORT 2025 item 24b asks reports to show outcome-relevant concomitant care by group, including cumulative or average exposure where appropriate. Item 24a distinguishes the planned intervention from what was actually delivered, including adherence and provider fidelity. The ICH E9(R1) estimand framework illustrates why rescue medication can be handled as part of a treatment-policy, composite, or hypothetical strategy rather than as a generic covariate.
Keep reading
Don't stop at one method.
Good methods judgment comes from contrast. Read the neighboring guides, see where the assumptions diverge, and avoid treating every observational problem like it needs the same hammer.
Hierarchical Testing in Clinical Trials: When a Significant Secondary Endpoint Is Still Descriptive
A practical hierarchical testing guide for clinical researchers. Reconstruct the prespecified testing path before treating a small p-value on a secondary endpoint as confirmatory evidence.
Randomized Withdrawal Trials: Why a Relapse-Prevention Win Is Not a New-Patient Effect
A practical randomized withdrawal trial guide for clinical researchers. Audit the run-in, responder enrichment, withdrawal contrast, safety, and target population before generalizing a maintenance-effect claim.
Allocation Concealment: The Randomized-Trial Safeguard That Works Before Assignment
A practical allocation concealment guide for clinical researchers and peer reviewers. Separate sequence generation, concealment, implementation, and blinding before trusting the word randomized.