One patient opens a pain diary on the last day of a trial and writes a four. A month earlier, the entry was a seven. Someone at another clinic in the same trial makes exactly the same entries. Later the treatment codes are opened. The first patient received a cannabinoid medicine, the second a placebo. Both diaries can be entirely accurate. The question facing the investigators is how much of the change they can attribute to the medicine.
Those are invented patients, but the puzzle is common. A review of twenty randomized trials gathered pain outcomes from studies involving 1,459 people. Across these trials, people assigned to placebo reported a substantial reduction in pain from where they began. A reader who looked only at the first and last entries in the active-treatment diaries would miss an important part of the story.
Two subtractions, two questions
Consider an imaginary trial in which the average pain score starts at seven in each group. By the final week, it is five in the cannabinoid group and five and a half in the placebo group. One subtraction gives a two-point improvement for participants assigned the drug. Another gives a difference of half a point in improvement between the groups. Both figures belong in the account. Only the latter uses the comparison that random assignment was designed to provide.
A half-point average also leaves plenty unsaid. One person may have improved by three points, another by none, while someone else felt worse. Patients may want to know how many people improved enough to sleep through the night or get through an ordinary working day. A study can collect those outcomes, but they cannot be reconstructed from the group average alone. An uncertainty interval around the difference also matters, particularly in a small trial.
A placebo group has a history of its own
A person with fluctuating pain may be particularly willing to enroll during a bad stretch. Even without a change in treatment, a later week may be closer to that person’s usual level. Attention from a clinical team, expectations about a new medicine and changes in daily routines may also accompany a trial. A decline in the placebo group includes all these possibilities. It does not tell us what proportion of relief came specifically from believing that a treatment would help.
There is a further wrinkle when a study medicine produces recognizable sensations. A participant who feels an unfamiliar psychoactive effect may form a strong opinion about the assignment. The same can happen to someone who expected such an effect and felt nothing. A review of blinding in cannabinoid analgesia trials explores the difficulty this poses. Asking people to guess can uncover problems, although an accurate guess may itself be based on relief already experienced.
Reading the result back to a patient
The twenty-trial review pooled evidence about changes under placebo across different studies. A separate AHRQ assessment of cannabis products for chronic pain asked about the benefits and harms of particular product types. It found small short-term improvements for some preparations, with much of the evidence coming from people with neuropathic pain. The identity of the product, the type of pain and the duration of follow-up narrow the population to which an estimate applies.
For the patient who wrote a four, the most important fact may be that dinner was easier to cook that evening. The diary is the place to preserve that experience. A clinician deciding what to expect for the next patient needs the diaries from both groups, information about side effects and some idea of how long any benefit lasted. The single entry and the randomized comparison can then do the different jobs for which each is well suited.