There was no call for immediate implementation of “Tetris” in clinical practice: Response to the commentary by Halvorsen et al. (2024)

We thank Halvorsen and colleagues for their commentary [1] on the Swiss TrAumatic biRth Trial (START) [2], a multicenter, double-blind, randomized controlled trial (RCT) with an active control group.START tested the efficacy of a single-session intervention, carried out within six hours following an unplanned cesarean section, to prevent maternal symptoms of childbirth-related posttraumatic stress disorder (CB-PTSD).It was a next step in our research program, in which a previous proof-of-principle RCT showed that women receiving this intervention had fewer intrusive childbirth-related memories during the first postpartum week than those receiving routine care [3].At the end of our publication [2], we concluded that the next step would be an implementation study.However, in stark contrast to what Halvorsen et al. claim, this was not a call for immediate implementation of the intervention in clinical practice.Halvorsen et al.'s title claim is thus incorrect and misleading.Halvorsen et al. rightly note that we did not operationalize the primary outcomes, as preregistered, by analyzing means, but rather by counting the symptoms present (scored 2, in accordance with validated ratings [4,5]).Primary outcomes were group differences in the presence and severity of maternal CB-PTSD symptoms at six weeks postpartum on PCL-5 1 and CAPS-5 2 subscale and total scores.These primary outcome measures remain the same in the trial pre-registration (NCT03576586), study protocol [6] and the paper [2].The way we analyzed them as symptoms counts is detailed in the paper, and we agree we should also have explicitly written this differed from the preregistration.There was a strong statistical argument for using this different type of analysis.We analyzed PCL-5 and CAPS-5 (ordinal scales) using dedicated categorical models from Item Response Theory (IRT) [7].Among IRT models, only Rasch models, when validated using suitable fit statistics [8], result in a global measure of severity from nominal presence of symptoms, or an ordinal assessment of their severity [e.g., 9, 10], that is reducible to a summed score.In START, routine IRT analyses showed that none of these models had an