This study was approved by the ethical committee at Osnabrück University. All participants provided electronic consent prior to participation in the study. The datasets analyzed during the study are available at the Open Science Framework (OSF).

The study was designed as a randomized waitlist control trial. Participants were assigned to one of two groups using LimeSurvey. Group allocation was conducted via simple randomization (1:1) using a built-in randomization function implemented in LimeSurvey. No block randomization or stratification procedures were applied.

Participants were allocated to one of two conditions:

Intervention group, which received IRT-based counselling by the web app DreamMend for 8 weeks.

Waitlist control group, which did not receive the intervention but did receive access to the web app after 8 weeks.

Eligibility criteria and recruitment

Participants were eligible if they were at least 18 years old, were experiencing nightmares at least every 2 weeks or reported suffering from nightmares, had access to an internet-enabled computer or smartphone to use the web application, and were able to read and understand German. Participants were excluded if they had been diagnosed with PTSD; used sleep medication, psychotropic drugs, or other sedative medications; were under medical or therapeutic treatment for sleep disorders; or abused alcohol or drugs. As compensation, students of the Cognitive Science and Psychology study program at Osnabrück University received course credit. All participants were invited to an exchange meeting after the study.

Procedures

The whole study was conducted online. After eligibility had been confirmed, participants were informed about the study purpose and filled out the first questionnaires. Subsequently, the intervention group received instructions for using the web app and information about the study procedure. An optional initial personal consultation as well as technical and content-related support was offered both in person and online to clarify any questions.

At week 0, participants of the intervention group registered via an assigned login name in the web app and separately completed the Nightmare Distress Questionnaire (NDQ) and the Pittsburgh Sleep Quality Index (PSQI). They were instructed to use the app as often as they wished and mentally rehearse the stored modified version of their dream for 5 min each day for 14 days.

Measures and materials

The measuring tools used in this study include the following questionnaires: baseline questionnaire, NDQ, PSQI, post-study questionnaire, and one additional question about nightmare frequency [4, 5].

Participants completed the NDQ and the PSQI every 2 weeks (see supplementary table S3). After completion of the 8‑week study phase, a self-designed post-study questionnaire was administered.

Primary outcomes

Nightmare frequency and nightmare distress (NDQ total score) were specified as co-primary outcomes. To control the family-wise error rate across the two co-primary endpoints, p-values were adjusted using the Holm–Bonferroni procedure (α = 0.05, two-sided).

Nightmare frequency

Nightmare frequency was assessed retrospectively at each timepoint with the question “How often do you currently have nightmares?” on a 0–7 ordinal scale (0 = never, 7 = several times per week).

Nightmare distress

Nightmare distress was measured using the German version of the NDQ [4]. The NDQ is a 13-item self-report inventory used to assess the degree of distress associated with experiencing nightmares. The response items are rated on a 5-point Likert scale ranging from 1 to 5. All item scores are summed up to form a total distress score (range: 13–65), with higher scores indicating higher nightmare distress.

Secondary outcome

Sleep quality was assessed using the PSQI [5], and global scores were calculated by summing components (C1–C7), as in the original paper.

Composite outcome—the nightmare burden index

In our sample, nightmare distress (NDQ total score) and nightmare frequency were moderately to strongly correlated across all timepoints (r = 0.37–0.84), indicating that participants with more frequent nightmares generally reported higher distress; however, the two measures were not identical. To create a single composite metric, both variables were standardized to z‑scores to place them on a comparable scale and then averaged to a holistic measure of nightmare impact, the nightmare burden index (NBI):

$$\begin&\textit\,\textit=\\&\frac})+z(NF)}\end$$

Secondary and exploratory outcomes were interpreted descriptively, without multiplicity adjustment.

Data preparation and sample size

Participant identifiers were normalized, questionnaire responses converted to numeric scales, and duplicate submissions removed. Analyses included all available observations without listwise deletion. An a priori power analysis indicated that 54 participants were required to detect a medium effect (d ≈ 0.6) with 80% power at α = 0.05. Recruitment challenges reduced the final analyzed sample to 34 participants (16 intervention, 18 control).

Statistical analysis

All analyses were conducted in Python 3.12.3 (pandas 2.3.0, NumPy 2.3.0, SciPy 1.16.1, statsmodels 0.14.5). Statistical significance was set at p < 0.05 (two-tailed).

Trajectory analyses (primary analyses)

To examine temporal dynamics, LMMs with timepoint (t0–t4) as a categorical factor were fitted, including fixed effects for group, timepoint, and the group × timepoint interaction. Participant-specific random intercepts accounted for repeated measures. Jackknife resampling evaluated the robustness of the estimated effects. For the two co-primary outcomes (nightmare frequency and nightmare distress), Holm–Bonferroni adjustments were applied to control the family-wise error rate at α = 0.05. At each timepoint (t3 and t4), p-values were ranked and compared to adjusted thresholds: the smallest p-value was tested against α/2 = 0.025 and the second p-value against α = 0.05. This approach ensures that the overall probability of falsely rejecting at least one null hypothesis across the two co-primary outcomes at each timepoint is controlled at 5% (family-wise error rate).

Complementary analyses—pre–post analyses

Because questionnaire completion varied across timepoints, some participants contributed data at only a subset of assessments, and some only provided a single follow-up measurement. To maximize use of available data while maintaining statistical stability, post-baseline measurements (t1–t4) were collapsed into a single post-phase for complementary pre–post analyses. Linear mixed models were fitted with fixed effects for group (intervention vs. control), phase (baseline vs. post-phase), and their interaction, with participant-specific random intercepts. This approach allows inclusion of participants with incomplete longitudinal data without requiring listwise deletion and is particularly appropriate for small clinical samples. The interaction term tests whether pre–post changes differ between intervention and control groups.

Descriptive and supplementary analyses

Descriptive statistics (means ± standard deviation [SD]) were computed for each outcome at all timepoints. Longitudinal group means with 68% confidence intervals were visualized using line plots.

Sleep quality (secondary analysis)

Global PSQI scores and individual component scores (C1–C7) were analyzed using LMMs with fixed effects of group, timepoint, and their interaction, and random intercepts for participants. These analyses were conducted to explore potential secondary effects of the intervention on sleep quality.