Why the Control Condition Changes the Question
A result described as “treatment reduced the event rate” is incomplete unless it says what participants receiving the comparison strategy were offered. The control condition helps define the clinical question: adding a treatment to standard care is not the same question as replacing observation with maintenance treatment, or adding a new therapy to the care routinely available in a hospital.
This distinction is especially visible across STAMPEDE, FOCUS4, and RECOVERY. Their controls were not three versions of an inert placebo. STAMPEDE used protocol-defined standard treatment for prostate cancer; FOCUS4 used active monitoring in a relevant maintenance comparison; and RECOVERY compared treatment strategies with usual care in hospitalized patients. Each control therefore gives its treatment effect a different clinical meaning.
Three Controls, Three Clinical Contrasts
In STAMPEDE, standard care was an active, protocol-defined prostate-cancer treatment strategy, centered on androgen-deprivation therapy (ADT). The exact protocol and background treatment depended on the comparison and period. When an experimental regimen was compared with standard care, the result addressed the additional effect of that regimen relative to the control treatment in the trial population. It did not estimate the effect of receiving any treatment versus receiving none.
In FOCUS4, active monitoring served as a control for a relevant maintenance-treatment question among patients with advanced colorectal cancer whose disease had not progressed after initial chemotherapy. Active monitoring means scheduled follow-up and assessment, with further clinical care available as appropriate; it is not abandonment of care. A maintenance-treatment comparison therefore asks whether offering the maintenance strategy improves outcomes relative to monitoring without that planned maintenance treatment.
In RECOVERY, usual care meant the care available to hospitalized patients under the trial’s clinical and calendar-time conditions. It could include other appropriate treatments, and it could change as evidence and practice changed. A randomized comparison with usual care estimates the effect of assigning the trial treatment strategy in addition to, or in the context of, that care. It does not estimate the treatment’s effect against a fixed, universal “no treatment” condition.
| Trial setting | Control condition | Clinical meaning of a randomized contrast |
|---|---|---|
| STAMPEDE | Protocol-defined standard treatment, centered on ADT | Effect of the experimental strategy relative to the specified standard-care strategy |
| FOCUS4 | Active monitoring in a relevant maintenance comparison | Effect of offering maintenance treatment relative to monitoring without planned maintenance treatment |
| RECOVERY | Usual care available in the hospital and period of enrollment | Effect of assigning the trial treatment strategy in the context of contemporaneous care |
These descriptions are not interchangeable. For example, “no planned maintenance treatment” does not mean “no follow-up,” and “usual care” does not necessarily mean the same regimen for every participant. The intervention and control must be described as strategies, including the care each permits, rather than as labels alone.
A Formal View of the Treatment Contrast
Let \(Y^A\) denote the outcome a participant would have under assignment to an experimental strategy \(A\), and \(Y^C\) the outcome under assignment to the trial’s control strategy \(C\). For a binary event, a population risk difference can be written as the risk under \(A\) minus the risk under \(C\). The control strategy is part of this definition: changing \(C\) changes the contrast being estimated.
With valid random assignment, adequate follow-up, and outcome measurement that is not differentially biased between groups, the observed groups can estimate the outcomes under their assigned strategies. This interpretation also depends on the trial’s eligibility criteria and setting. Randomization supports a comparison within that trial; it does not make the trial’s control group equivalent to a differently selected or differently treated group in another study.
A short justification shows why the control matters. Suppose the intervention is the same, but the risk under control differs because the control strategy or population differs. Then the risk difference changes even if the intervention-group risk stays fixed:
Subtracting gives \(RD_1-RD_2=P(Y^{C_2}=1)-P(Y^{C_1}=1)\). Thus, a difference between reported risk differences can arise from different control risks, not only from a different intervention effect. This arithmetic is one reason to interpret effects in their original clinical and trial context.
Worked Example: Adding a Strategy to Standard Care
Hypothetical data: In a randomized comparison modeled on the kind of contrast used in STAMPEDE, 100 participants are assigned to standard care and 100 to standard care plus an experimental treatment. By the prespecified follow-up time, 52 participants in the standard-care group and 40 in the added-treatment group have the event. Assume assignment was randomized, follow-up was comparable, and event ascertainment used the same definition in both groups.
The estimated event risks are \(52/100=0.52\), or 52%, under standard care and \(40/100=0.40\), or 40%, under standard care plus the experimental treatment.
The estimated risk difference is 12 percentage points fewer events in the added-treatment group over this follow-up period. Because the control received standard care, the result concerns adding the experimental strategy to that care; it does not compare treatment with no care. The counts alone do not establish statistical significance or clinical importance, which would require uncertainty estimates and clinical judgment.
Worked Example: Interpreting Active Monitoring
Hypothetical data: In a randomized maintenance comparison, 120 patients are assigned to active monitoring and 120 to a maintenance strategy after initial treatment. By six months, 72 monitored patients and 54 maintenance patients have experienced a defined progression event. Assume participants met the same eligibility criteria, follow-up duration is complete and comparable, and progression was assessed by the same method.
The event risks are \(72/120=0.60\), or 60%, with active monitoring and \(54/120=0.45\), or 45%, with maintenance treatment. Taking the maintenance group’s risk minus the monitoring group’s risk gives:
The estimated risk is 15 percentage points lower with the maintenance strategy over six months. The clinical contrast is maintenance treatment versus active monitoring without planned maintenance treatment, not treatment versus abandonment: monitoring includes scheduled assessment and the possibility of further care. These hypothetical results do not by themselves establish that the difference is statistically reliable or that it applies to patients outside the trial’s eligibility criteria.
Worked Example: A Treatment Compared With Usual Care
Hypothetical data: In a randomized hospital trial, 100 patients are assigned to an experimental treatment strategy and 100 to usual care. By a specified follow-up time, 22 patients in the experimental group and 28 in the usual-care group have died. Suppose the groups were enrolled contemporaneously, received comparable outcome follow-up, and usual care was delivered according to practice available during that enrollment period.
The mortality risks are \(22/100=0.22\), or 22%, in the experimental group and \(28/100=0.28\), or 28%, in usual care. The estimated risk difference is:
The estimated mortality risk is 6 percentage points lower under assignment to the experimental strategy in this hypothetical trial. The interpretation is relative to usual care during the study period. If usual care included other treatments, the result reflects the added or alternative strategy in that context, not a comparison with no medical care. This example does not supply enough information to calculate an interval or p-value.
Reading Results Across Trials
The three control conditions clarify why trial results should not be ranked by comparing control-group event rates alone. Patients in STAMPEDE, FOCUS4, and RECOVERY differ in disease, eligibility, clinical course, follow-up, and care setting. Their control groups are also assigned different strategies. A higher observed event rate in one control group does not show that its control care is inferior to another trial’s care.
A randomized contrast is most directly interpretable within its trial: experimental strategy versus that trial’s control, among eligible participants and under its conduct. When a multi-arm platform uses a shared control, the same principle applies. A comparison is informative when treatment and control participants are comparable in relevant ways, including calendar time and eligibility. In an evolving platform, nonconcurrent comparisons may be affected by changes in patient mix, background treatments, or outcome management, so they should not automatically be read as though all groups had been randomized together.
For RECOVERY, usual care could evolve as clinical evidence accumulated. The earlier material on RECOVERY’s usual-care comparator over time is useful context: a treatment contrast belongs to the period and care environment in which its participants were randomized. A result from one period need not represent the effect that would occur under a later version of usual care.
For a careful interpretation, ask:
- What exactly could participants in the control group receive, and what was planned?
- Was the experimental strategy added to the control care, substituted for part of it, or offered instead?
- What population, endpoint, and follow-up period define the effect?
- Did the control strategy or background care vary across enrollment periods or sites?
- Are the effect estimate and uncertainty presented for the randomized comparison, rather than inferred from raw rates across different trials?
Diagnostics, Limitations, and a Common Pitfall
A control label can conceal clinically important variation. “Standard care” may refer to protocol-defined treatment, whereas “usual care” can allow clinician judgment and local practice. “Active monitoring” should specify the assessment schedule and what happens if disease progresses. To diagnose ambiguity, inspect the protocol or methods for allowed co-interventions, timing, monitoring, crossover, and criteria for subsequent treatment. If these details differ between groups or are poorly reported, the intended treatment contrast may be hard to identify.
Also check that comparisons are concurrent and that the outcome definition and follow-up are comparable. A control event rate may change when the enrolled population, supportive care, diagnostic practice, or calendar period changes. Those changes can alter absolute effects and may limit transportability. Even a well-randomized result does not automatically establish effects in patients excluded from the trial or under a substantially different care pathway.
Common mistake: “The control group in Trial A had more events than the control group in Trial B, so Trial A’s control treatment must be worse.” This reasoning treats the groups as if they were randomized against each other. They were not. Their patients, endpoints, follow-up, and control strategies may differ. The correction is to interpret each randomized treatment contrast within its own trial, then use clinical and design information—not cross-trial control rates alone—when considering whether results apply elsewhere.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions below.
- In one sentence, state the clinical contrast represented by an experimental treatment compared with standard care in STAMPEDE.
- Why is active monitoring not equivalent to receiving no care in a maintenance-treatment comparison?
- A hypothetical trial reports event risks of 18% under a new strategy and 25% under usual care. Calculate the risk difference using new strategy minus usual care, and interpret it.
- Why can’t control-group event rates from RECOVERY and FOCUS4 be treated as a randomized comparison of their control conditions?
- Name two features to check when deciding whether a control strategy is described precisely enough to interpret.
Solutions. (1) It estimates the effect of the experimental strategy relative to the protocol-defined standard-care strategy in the eligible trial population. (2) Active monitoring includes planned assessment and may lead to further care when clinically indicated, although it omits planned maintenance treatment. (3) \(0.18-0.25=-0.07\): the estimated event risk is 7 percentage points lower under the new strategy over the trial’s specified period; uncertainty is not given. (4) The groups were not randomized against each other and may differ in population, endpoint, follow-up, setting, and control strategy. (5) Appropriate features include allowed co-interventions, monitoring schedule, criteria for subsequent treatment, and calendar period of enrollment.