What the Study Did

OnabotulinumtoxinA has been licensed for chronic migraine since 2010, and most of what is known about how well it works comes from the pivotal trials that got it there. Trial populations are selected, monitored and supported in ways that clinic populations are not, so the question of how the treatment performs in ordinary practice keeps coming up.

A new study in Therapeutic Advances in Neurological Disorders reports on 479 patients treated across 12 tertiary headache centres in Poland, all within the national reimbursement programme. It is a retrospective observational cohort rather than a trial, which limits what can be concluded, but the size makes it one of the larger real-world datasets on this treatment.

Results at 9 Months

Monthly headache days fell by a mean of 5.9 days, from 19.1 at baseline. For someone starting at 19 headache days a month, that is a move to around 13.

Disability scores moved further than the day count alone suggests. MIDAS improved by a mean of 52.7 points, with a standard deviation of 48.3. The spread is wide. A standard deviation of 48.3 around a mean of 52.7 means individual experiences differed a great deal, with some patients improving substantially and others barely at all.

Who Counted as a Responder

The headline response figure depends entirely on where the line is drawn, and this study reports two lines rather than one.

At the conventional threshold of a 50% or greater reduction in monthly headache days, 27.3% responded. Adding the partial responders, who achieved a 25 to 49% reduction, brings the total to just over half the cohort at 50.5%.

Medication Overuse

Of the patients who met criteria for medication-overuse headache at baseline, 41.6% no longer met those criteria after treatment.

This matters because medication overuse and chronic migraine tend to reinforce each other, and the traditional approach involved withdrawing the overused drug before or alongside starting a preventive. The 2026 AAN and AHS prevention guideline reports that evidence now favours starting a preventive on its own as being about as effective as adding a withdrawal programme, with the important exception of opioids and barbiturates. This cohort is consistent with that pattern: treating the migraine reduced the medication overuse in a substantial minority without withdrawal being the primary intervention.

The Prediction Model

The research team tried to work out who would respond, using the clinical characteristics available before treatment. Two factors came out statistically significant, both associated with a poorer response.

The third row qualifies the first two.

Both findings can hold at once. Family history and multiple prior failures do shift the odds across a group of 479 people, and knowing them still does not let a clinician predict the outcome for the person in front of them. The authors state this directly, concluding that standard clinical phenotypes alone are insufficient to predict outcomes with this treatment.

Why These Numbers Run Lower

A 27.3% response rate sits below what the pivotal trials reported, and the study is explicit about why that comparison is not like for like.

Entry to the Polish reimbursement programme required documented failure of at least two prior oral preventives, and the treatment was delivered at tertiary headache centres, which see the harder cases. The cohort is therefore more refractory than a typical trial population by construction. The authors describe the setting as a strict programmatic framework and frame their findings as reflecting the challenge of managing highly refractory patients within it.

What It Means in Practice

For someone considering or already receiving this treatment, a few things follow from the data.

The prediction finding has a practical edge for anyone who has been told they are unlikely to respond because of their treatment history. A model built on 479 patients, using exactly that kind of information, performed at close to chance when tested properly. Group-level associations of this strength do not translate into individual forecasts.

Key Takeaways

The study adds a large real-world effectiveness estimate for a treatment whose reputation rests mostly on registration trials, and the estimate is lower than those trials in a population that was selected to be harder to treat.

The prediction analysis is the part that transfers beyond this treatment. Two predictors reached statistical significance across 479 people, and the same model could not forecast the outcome for an individual.