What a control group is for and why before-and-after lies

A control group exists to answer one question: how much of the difference the intervention produced. You build it by deliberately leaving a share of the orders unworked, under the same selection rule and inside the same time window.

Why does a before-and-after overstate the result?

For two reasons acting at once. Health demand is seasonal, so two different months are not comparable by default. And while a campaign is running the provider's own team works the list harder than in an ordinary month, so part of the improvement is produced by that team and would show up in the untouched arm as well. The preference for a parallel arm is written into US drug regulation: FDA's rule on adequate and well-controlled studies treats historical controls as less well assessed than concurrent controls and reserves them for special circumstances.

The cost of running a control group is real and worth saying out loud: those are patients who go unworked during the measurement period. That cost is bounded by a short window and by offering the same outreach to the control arm as soon as measurement closes, which is how it is described in how we measure.

• Randomise at patient level: one patient split across both arms contaminates both.
• Run the arms in parallel, never in different months.
• Use the same definition of denominator in both arms.
• Fix the size of each arm before seeing the first result.
• Report both arms in full, including the control's baseline rate.