Nightmares receded on dronabinol and the wider PTSD signal did not survive correction
- Double-blind, placebo-controlled trial at four clinical sites in Berlin, Hamburg and Mannheim, led by Stefan Roepke of Charité – Universitätsmedizin Berlin and published in Nature Medicine on 5 August 2026 (ClinicalTrials.gov NCT04448808). 171 adults with PTSD and at least two nightmares a week were randomised to dronabinol (Δ9-THC) or to matching placebo, taken each night for 10 weeks; 169 entered the full analysis set, 87 and 82. The investigator-initiated trial was funded by an unrestricted grant from Bionorica SE, which also supplied the study medication.
- On the primary endpoint, the CAPS-IV B2 item that rates nightmare frequency and intensity over the past week on a 0–8 scale, scores fell 3.66 points on dronabinol and 2.17 on placebo by week 10: difference -1.50 (95% CI -2.28 to -0.71), p < 0.001. At week 10, 36.8% of the dronabinol arm reported no nightmares against 14.5% on placebo, odds ratio 3.70 (1.64 to 9.09), nominal p = 0.002.
- The two key psychiatric secondary outcomes at week 10 did not hold after correction for multiple testing. Clinician-rated PTSD severity on the CAPS-5 differed by -3.43 (95% CI -7.33 to 0.47), p = 0.08; depression on the MADRS differed by -2.68 (-5.17 to -0.20), p = 0.035. Under the Hochberg procedure the authors applied to key secondary outcomes, only the patients' global impression of change remained significant, odds ratio 4.66 (2.05 to 11.32).
- Several secondary scales moved on nominal p values that carry no correction. Self-reported PTSD symptoms on the PCL-5 fell 6.03 points further on dronabinol at week 10 (-10.30 to -1.75), p = 0.006, and the ITQ 5.98 points further (-10.30 to -1.66), p = 0.007. Serious adverse events occurred in 7 of 87 participants on dronabinol and in none of 83 on placebo; none was classed as treatment-related.
No participant in this trial was allowed trauma-focused psychotherapy, from four weeks before the start until the final visit. The protocol published with the paper states the rule three times. Other therapy continued at a stable frequency for the 46.2% already in it. With that treatment held back, the design asks what a nightly dose of dronabinol changes in ten weeks. Dronabinol is Δ9-tetrahydrocannabinol, given here as an investigational oral solution, BX-1. In the United States the FDA has approved dronabinol products for other indications: anorexia with weight loss in AIDS, and nausea and vomiting with cancer chemotherapy in patients who have failed to respond adequately to conventional antiemetic treatments.
Stefan Roepke's group randomised 171 adults and analysed 169. Most were women (79.3%), with a mean age of 37.9. For 58.6% the index trauma was sexual assault, and for 41.9% it had happened between the ages of 6 and 18. Entry required PTSD with a CAPS-5 score of 26 or more and at least two nightmares a week. Two in five had current depression.
The dose rose over four weeks from 2.5 mg towards the level at which nightmares stopped, with a ceiling of 15 mg. Two-thirds of the dronabinol arm with dose data, 54 of 81, finished at that ceiling. The primary endpoint was the CAPS-IV B2 item, a clinician's rating of nightmare frequency and intensity over the previous week, scored 0–8. The arms separated from week 3 and stayed apart. At week 10 the difference was -1.50 points, and 36.8% of the dronabinol arm reported no nightmares, against 14.5% on placebo.
In Table 2, clinician-rated PTSD severity on the CAPS-5 fell 3.43 points further on dronabinol at week 10, with a 95% interval from -7.33 to 0.47; MADRS depression scores fell 2.68 points further, p = 0.035. The authors controlled multiplicity across the key secondaries with the Hochberg procedure, and after it only the patients' global impression of change remained significant. Extended Data Table 1, with nominal p values, shows the CAPS-5 separating at week 6 (p = 0.02), the MADRS at week 4, and self-reported PTSD on the PCL-5 at weeks 6 and 10. In the per-protocol set, week-10 CAPS-5 and MADRS gave p = 0.17 and 0.08.
Sleep beyond the nightmare item was recorded only by diary and questionnaire. Diary sleep time ran longer on dronabinol in weeks 3, 5 and 6, not in week 10, and time to fall asleep did not differ in any week.
Blinding was imperfect, and the authors measured by how much: of those who answered, 59 of 73 on dronabinol (80.8%) guessed their allocation correctly, against 31 of 59 on placebo (52.5%). Among those who believed they had received placebo, the nightmare difference was -1.79 (95% CI -3.30 to -0.28). The psychiatric scales lean towards dronabinol without clearing the correction, and the published tables leave two explanations standing side by side: 169 people over ten weeks may be too few to settle a difference of this size, and ratings may have been shaped by participants knowing which arm they were in. On either reading the effect on the disorder stays unresolved; the trial does not show that PTSD or depression stayed where they were. Serious adverse events occurred in seven participants, all in the dronabinol arm, and none was classed as treatment-related.
Ten weeks of a nightly dose moved the nightmare score 1.5 points further than placebo, while the clinician-rated PTSD total and the depression scale did not survive the correction the authors themselves applied.
Trauma-focused psychotherapy was excluded for the whole trial, so the results say nothing about dronabinol given alongside it or compared with it, and there was no active comparator such as prazosin or imagery rehearsal. Treatment lasted ten weeks; long-term efficacy and safety were not studied, and the authors say so. The key psychiatric secondary outcomes did not remain significant after the Hochberg correction, and the other secondary scales are reported with nominal p values. Of those who answered the allocation question, 59 of 73 on dronabinol (80.8%) and 31 of 59 on placebo (52.5%) guessed correctly. Sleep outside the nightmare item was measured by diary and questionnaire only, with no objective recording. The sample was mostly women whose index trauma was most often sexual assault; people with current substance use disorder, psychosis, bipolar disorder, acute suicidality or a current depressive episode with a MADRS score above 29 were excluded. 171 people were randomised and 169 analysed; by the abstract, 84 were assigned to placebo against 82 analysed in that arm, and the open parts of the paper do not give the reason. The trial was funded by Bionorica SE, which supplied the medication.