Internet-based Cognitive-behavioral Writing Therapy for Reducing Posttraumatic Stress After Severe Sepsis in Patients and Their Spouses (REPAIR): Results of a Randomized Controlled Trial

Background: Posttraumatic stress disorder (PTSD) is a frequent long-term consequence of intensive care in both, patients and spouses, and there is a specic need of respective tailored intervention approaches. Aim of this trial was to investigate the ecacy, safety and applicability of internet-based cognitive-behavioral writing therapy (iCBT) for posttraumatic stress after intensive care for sepsis in patients and their spouses compared to a waitlist (WL) control group. Methods: In a randomized-controlled, parallel group, superiority trial we included patients after intensive care (>5 days) for sepsis and their spouses, if at least one of them had a presumptive PTSD diagnosis (PTSD Checklist [PCL-5] ≥ 33). Dyads were randomized 1:1 either to the iCBT (including two internet-based writing assignments per week over a 5-week period) or to the WL control group. Primary outcome was pre-post change in PTSD symptom severity score (measured via the PCL-5) in participants with a presumptive PTSD diagnosis from randomization to ~5 weeks later. Results: Twenty-ve dyads were included representing 34 participants with presumptive PTSD diagnosis (patient and spouse: 9 dyads, patient only: 12, spouse only: 4). We did not observe evidence for a difference in the primary outcome for iCBT compared to WL in the conrmatory analysis according to the intention-to-treat principle. The mean difference in score point reductions between iCBT and WL was -0.96 (95% condence interval, -5.88 to 3.97; p = 0.703), which was adjusted for pre-values of the PCL-5 in the generalized estimating equations model. No adverse events were reported. Participants conrmed the applicability of iCBT. We applied an iCBT to reduce PTSD after sepsis and intensive care, the rst time addressing both and their spouses. ICBT was applicable in the given We did not observe evidence the changes in PCL-5 PTSD symptom Due our trial should be regarded as a pilot study and ndings remain preliminary. Nevertheless, our results can guide further research, which is needed to determine if modied approaches to post-intensive care PTSD may be more effective.


Introduction
Experience of intensive care could affect mental health of both patients and their partners. About every fth patient and an equal proportion of spouses develop a posttraumatic stress disorder (PTSD) as a long-term consequence of treatment in the intensive care unit (ICU) [1][2][3][4]. Thus, PTSD has been considered as part of the post-intensive care syndrome (PICS) and PICS-family [5]. Research on post-ICU consequences revealed that mental health of patients and their spouses following ICU experiences are Page 3/19 interrelated and that couples seem to react as a dyadic system to a life-threatening situation [6][7][8][9]. In the context of dyadic coping research, it has been suggested to use the term 'we-disease' to describe that both, the patient and his/her partner, face the illness "as a shared 'we-event' and a 'we-experience' rather than an individual problem of one partner requiring support from the other" (p. 595) [10]. The concept of 'we-disease' also implies that the treatment of mental distress associated with the illness should always include both partners as they both suffer but also have resources and can jointly contribute to the coping process [6,10].
In the past few years, various intervention approaches have been developed to address PTSD in patients or family members that might be classi ed as interventions during ICU care to prevent PTSD or as interventions to treat PTSD in the long-term run. Preventive PTSD interventions usually consider a broad target group of ICU patients or family members at risk for post-ICU PTSD, e.g. [11][12][13][14][15][16]. Contrasting, PTSD treatment interventions are provided several months after ICU discharge and speci cally address patients or family members with clinically signi cant PTSD symptoms and PTSD diagnosis, respectively, e.g. [17].
For the treatment of PTSD, clinical guidelines in general strongly recommend trauma-focused psychotherapy with cognitive-behavioral components of exposure and/or cognitive restructuring [18,19].
Implemented via the internet, cognitive-behavioral therapy can meet the speci c needs of post-ICU patients by providing access for mobility-impaired patients or patients with speech and hearing di culties [20].
The primary objective of the REPAIR (Reducing posttraumatic stress after severe sepsis in patients and their spouses) trial was to investigate the e cacy, safety and applicability of a therapist-guided internetbased cognitive-behavioral writing therapy (iCBT) for posttraumatic stress after intensive care for sepsis in patients and their spouses compared to a waitlist (WL) control group. Second, considering the interrelation of mental health between patients and partners, the study aimed at examining dyadic concordance in treatment effects, i.e. indirect effects of the treatment in the respective spouse of the treated participant.

Study design and setting
The REPAIR trial was a randomized, controlled, open-label, parallel group superiority trial conducted at the Jena University Hospital (JUH), Germany. Dyads of a post-ICU patient and his/her spouse were randomly assigned to either iCBT or to a WL control group (allocation ratio 1:1) with the dyad being the unit of randomization. Allocation was performed centrally using computer-generated random numbers provided by an independent center for clinical trials at the JUH and strati ed by the occurrence of PTSD symptoms within the dyad of the post-ICU patient and the spouse (stratum 1: both post-ICU patient and spouse with PTSD, stratum 2: post-ICU patient with PTSD and spouse without PTSD, stratum 3: post-ICU patient without PTSD and spouse with PTSD). The trial was approved by the ethics committee of the Friedrich-

Participants
We included dyads (each member ≥ 18 years) comprising a former patient, who was treated for sepsis on an ICU for more than ve days and discharged from ICU more than one month ago, and his/her spouse (married or cohabited). A patient-spouse dyad was included if at least one of them presented with a presumptive PTSD diagnosis with regard to the life-threating event (see de nition below). Reasons for exclusion on dyad-level were not having a spouse as well as acute psychosis, suicidal ideation, use of neuroleptics, not being uent in German or ongoing psychotherapeutic treatment elsewhere of at least one dyad member. Participants were screened for eligibility in a telephone interview by using the PTSD checklist for

Intervention
Participants allocated to the treatment condition participated in an iCBT targeted to the traumatic ICU situation. They completed two 50-minute internet-based writing assignments per week over a ve-week period (10 essays in total). The treatment consisted of three modules: (I) resource-oriented biographical reconstruction (three essays), (II) in sensu trauma exposure sessions (four essays), and (III) cognitive reconstruction (three essays). After completion of each assignment, the therapist provided individual feedback and further writing instructions to the participant within one workday. Integrated in the third module, the treated participant received a supportive letter from his/her respective partner. Participants without clinically relevant PTSD symptoms (PCL-5 < 33) only completed the assessments and received psychoeducational information about mental health problems after traumatic events. Participants allocated to the WL control group also received iCBT after ve weeks of waiting (duration of treatment), but without a supportive letter from their spouses. For details, we refer to the published study protocol [20].

Outcomes
Primary outcome was change in PTSD symptom severity score from baseline to the end of treatment/waiting time (about 5 weeks after randomization) measured via the German version of 25] covering the four DSM-5 symptom clusters. A cut-off of 33 was used for a presumptive PTSD diagnosis [26]. A change of 10 points or more is regarded as clinically relevant [21].
Secondary e cacy outcomes were (a) symptoms of psychological distress, (b) relationship satisfaction, (c) health-related quality of life, and (d) remission at the end of treatment/waiting time. Safety endpoints were (i) the number of suicide alerts, (ii) the number of participants with a clinically relevant PCL-5 deterioration, and (iii) the percentage of participants leaving the study early (during treatment phase) due to any reason. An additional secondary endpoint was dyadic concordance in treatment effects (in terms of PCL-5). Psychological distress was measured using the German version of the Brief Symptom Inventory-18 including subscales of anxiety, depression, and somatization (BSI-18) [27,28]. Relationship satisfaction was assessed with the German version of the Relationship Assessment Scale (RAS) [29,30].
The German version of the health questionnaire of the EuroQol group (EQ-5D-5L) [31, 32] was used to measure health-related quality of life on ve dimensions (mobility, self-care, usual activities, pain/discomfort and anxiety/depression). Utility values of EQ-5D-5L were derived according to Ludwig et al. [32]. Remission was only considered for participants diagnosed with PTSD before treatment/waiting period. For those, remission was de ned as being free of PTSD diagnosis after the intervention/waiting period. PTSD was diagnosed by a trained clinical psychologist (RG) as described above (for details, Supplemental Digital Content 1).

Sample size
The sample size calculation was based on Student's t-test for a parametric two-group comparison, even though more complex models that address the clustering would be used for the con rmatory analysis. In accordance with a recent meta-analysis [33], we considered effect sizes quanti ed by Cohen's d of 0.95 as realistic. To detect differences between the treatment groups at a two-sided signi cance level of 0.05 with a comparison-wise power of 0.9, a sample size of 2 × 34, i.e. 68 patient-spouse dyads, is required for the intention-to-treat (ITT) analysis. Assuming a dropout rate of 30%, the total sample size would be 98 dyads. A higher power was chosen to address the fact that a more complex statistical analysis approach would be used.

Statistical methods
We rely on two populations. The dyad population includes randomized participants irrespective of their presumptive PTSD diagnosis. The PTSD population only comprises participants with a presumptive PTSD diagnosis. Primary and secondary e cacy/safety outcomes are analyzed in the PTSD population, while dyadic concordance is assessed in the dyad population.
The handling of missing data was prede ned in the study protocol and/or the statistical analysis plan (SAP). Based on the expected high internal consistency of the scores, we substituted missing items with the mean of the provided items of the respective participant if 10% or fewer items were missing. We apply the intention-to-treat (ITT) and the per-protocol (PP) principle to both populations. In case of the perprotocol (PP) principle, we include randomized participants (with a presumptive PTSD diagnosis) who provided pre-(t0) and post-treatment/waiting (t1) information. In case of the ITT principle, we consider all randomized participants (with a presumptive PTSD diagnosis). Missing score values were replaced, strati ed by intervention group and type of treatment (de ned according to the strata for the randomization), according to best-case/worst-case substitution. We denote this data set as "primary analysis set". As additional sensitivity analysis that was not pre-speci ed in the study protocol, we used multiple imputation by chained equations (MICE) using fully conditional speci cation [34] (for details, Supplemental Digital Content 1).
Participant characteristics (dyad population) and outcomes (PTSD population) are summarized as absolute and relative frequencies for nominal variables or as medians together with the rst and third quartile (Q1, Q3) for ordinal/continuous variables. Rough group comparisons were done by Fisher's exact test or Mann-Whitney-U test. For the primary outcome PCL-5 change, we applied generalized estimating equation (GEE) modelling (independent variables: baseline PCL-5 value, treatment condition; cluster: dyad) in the primary analysis set. We performed several sensitivity analyses (ITT principle with MICE, PP principle, extension of the above de ned GEE model by inclusion of further possible confounders as independent variables). For the secondary e cacy outcomes, we adapted the GEE modelling accordingly. For primary and secondary e cacy outcomes, model coe cients (adjusted mean differences or odds ratios (OR)) with 95% con dence intervals (CIs) and p-values are presented. In addition, we provide the corresponding between-group effect sizes (standardized mean differences, Cohen´s d). For the safety outcomes, we provide absolute and relative frequencies from the PP population. Dyadic concordance in treatment effects (in terms of PCL-5) was assessed with Spearman correlation (together with the corresponding 95% CIs) independently from the treatment condition between post-ICU patients and his/her spouse in the dyad population according to both ITT and PP principle. We applied a two-sided signi cance level of 0.05 to the primary con rmatory analysis and did not correct for multiple testing otherwise as the other analyses were considered exploratory. We used R (version 3.6.0) for statistical analyses (for details, Supplemental Digital Content 1).

Participants
Between February 2017 and January 2019, we received 57 enquiries from either a post-ICU patient or his/her spouse. After screening for eligibility 25 dyads were randomized, 12 to iCBT and 13 to WL (Fig. 1). Median age of the study participants was 55 years (Q1-Q3, 47-62). Thirty-four participants had a presumable PTSD diagnosis (9 dyads with affected post-ICU patient and spouse, 12 with post-ICU patient only, 4 with spouse only). Of those, 25 were initially diagnosed with PTSD in the clinical interview (iCBT: 14; WL: 11). Further characteristics of the participants are shown in  The numbers are based on the dyad population. Overall, there are 25 dyads − 12 dyads in the iCBT group and 13 dyads in the WL control group. Note that each dyad comprises one post-ICU patient and one spouse. Characteristics are summarized as median with rst and third quartile (Q1, Q3) or as absolute (n) and relative frequency (%). P-values are derived from Mann-Whitney-U test and Fisher's exact test, respectively, while excluding patients with missing (including non-speci ed) information on the respective characteristic. Abbreviations: iCBT, internet-based cognitive-behavioral writing therapy; ICU, intensive care unit; PTSD, post-traumatic stress disorder; WL, waitlist.
* percentage refers to dyad; ** percentage refers to former ICU patient; *** percentage refers to partner; **** missing for 9 patients (iCBT group: 6, WL control group: 3 The numbers are based on the dyad population. Overall, there are 25 dyads − 12 dyads in the iCBT group and 13 dyads in the WL control group. Note that each dyad comprises one post-ICU patient and one spouse. Characteristics are summarized as median with rst and third quartile (Q1, Q3) or as absolute (n) and relative frequency (%). P-values are derived from Mann-Whitney-U test and Fisher's exact test, respectively, while excluding patients with missing (including non-speci ed) information on the respective characteristic. Abbreviations: iCBT, internet-based cognitive-behavioral writing therapy; ICU, intensive care unit; PTSD, post-traumatic stress disorder; WL, waitlist.
* percentage refers to dyad; ** percentage refers to former ICU patient; *** percentage refers to partner; **** missing for 9 patients (iCBT group: 6, WL control group: 3 Primary outcome Individual, time-dependent PCL-5 curves are shown in Fig. 2. In the primary analysis set, we did not observe evidence for differences between groups in the primary outcome. The adjusted mean difference in PCL-5 score change was − 0.96 (95% CI, -5.88 to 3.97; p = 0.703; Table 2) when comparing iCBT to WL. Sensitivity analyses also showed no evidence for differences in PCL-5 change between the iCBT and the WL control group (ITT with MICE: 4.01; 95% CI, -1.89 to 9.91; p = 0.181; PP: 2.40; 95% CI, -2.29 to 7.08; p = 0.316; Table 2). The corresponding between-group effect sizes varied between − 0.14 (95% CI, -0.81 to 0.54) and 0.48 (95% CI, -0.21 to 1.16) (Supplemental Table S9). The extended multivariable models revealed similar results (Supplemental Digital Content 2, Supplemental Table S4).  Table S5). For all other secondary e cacy outcomes, we did not observe evidence for an association between score changes and iCBT in the primary analysis set (Supplemental Digital Content 2, Supplemental Tables S6 to S8) with corresponding effect sizes of 0.04 (95% CI, -0.64 to 0.71) for BSI-18 and 0.25 (95% CI, -0.42 to 0.93) for EQ-5D-5L (Supplemental Table S9). Among patients with initial PTSD diagnosis (according to CAPS-5), remission rates were 64% after iCBT and 27% after waiting. Of note, 95% CIs for the iCBT effect for remission are wide in both ITT and, particularly, in the PP analyses; a smaller number of participants was considered in theses analyses as only participants with a presumptive PTSD diagnosis at baseline were included (Supplemental Digital Content 2, Supplemental Table S6).

Safety and applicability
Overall, there were ve suicide alerts. All of them were clari ed in immediate therapeutic contacts by telephone (see [20] for a description of safety management). Three revealed to be false alarms, two were caused by reasons not related to the study and the suicidal ideations subsided quickly. During iCBT/waiting, there were no clinical relevant deteriorations in regard of the PCL-5 score. Seven participants prematurely terminated in the iCBT group and two during waiting time, respectively. All dropouts appeared for reasons other than study or treatment participation (for further details on the safety endpoints, Supplemental Digital Content 2, Supplemental Tables S10 and S11). In-depth interviews with participants after the treatment con rmed the applicability of the intervention. It was positively highlighted that iCBT met the speci c needs of the patients and the spouses. In terms of feasibility, no major technical problems emerged and the internet literacy of the participants was su cient to complete the treatment.

Dyadic concordance in treatment effects
No evidence for a correlation between the PCL-5 changes of post-ICU patients and those of his/her spouse could be observed -neither in case of only one dyad member nor in case both dyad members had a presumptive PTSD diagnosis (Supplemental Digital Content 2, Supplemental Table S12).

Discussion
Posttraumatic stress symptoms are a frequent, well-documented sequelae after intensive care, both in the patient and his/her spouse [1][2][3][4] and with dyadic concordance of mental impairments within the couple [6-9]. Previous studies have targeted PTSD prevention during hospital stay [11][12][13][14][15][16] and post-ICU interventions [17] with variable success. To our knowledge there is no previous study that evaluated an intervention involving both patients and spouses with the goal of reducing these PTSD symptoms. Consequently, we conducted a randomized-controlled trial testing the e cacy, safety and applicability of an iCBT for reducing PTSD symptoms in patients and their spouses.
We included 25 dyads resulting in 34 treated participants which however did not represent the planned sample size. With regard to our primary outcome, we could not observe evidence that iCBT led to a larger reduction of PTSD symptom severity than waiting. With regard to secondary outcomes, we explored a relatively larger increase of relationship satisfaction in the iCBT compared to the WL control group which merits con rmation in future studies. Contrary to our hypothesis, there was no evidence for dyadic concordance in any of the treatment effects.
Beyond e cacy, participating in iCBT was safe, as no adverse events such as suicidality or clinically relevant PTSD symptom deterioration occurred that were therapy-related. Although seven participants prematurely terminated in the iCBT group, all dropouts appeared for reasons other than study or treatment participation. Furthermore, participants con rmed the applicability of iCBT and the feasibility of the implementation. Hence, iCBT can be regarded as an applicable intervention in the particular population of post-ICU patients and their spouses.
The need for ICU follow-up care to diagnose and treat PICS impairments after hospital discharge is apparent. Post-ICU patients show an increased utilization of outpatient specialist services, including psychiatric services, higher medication intake, and impaired quality of life [35]. Specialized post-ICU outpatient clinics could provide the necessary services speci c to ICU survivors' healthcare needs [5], but are however not yet established nationwide in Germany. Internet-based treatment approaches like iCBT in the follow-up of ICU patients can be particularly helpful for physically impaired patients or patients living in considerable distance from the hospital or specialized outpatient care, regardless of whether they are cared for in an ICU follow-up clinic or not.
A key strength of our study is the involvement of both patients and spouses in the treatment of post-ICU posttraumatic stress. In this regard, we implemented writing a supportive letter to the respective spouse as a dyadic treatment component in the iCBT [20]. Including both partners in the study (though not always both were treated) might have led to in-depth communication about their individual traumatic experiences and its consequences to their partnership [36]. Considering the concept of 'we-disease', this might have contributed to a strengthening of dyadic resources and coping efforts [10], nally affecting both partnership and individual well-being.
There are, however, several important limitations that may have affected the results. First, we did not achieve the planned sample size. Despite tremendous efforts and a signi cant extension of the recruitment period, we experienced serious problems in recruiting participants. We can only speculate about the reasons. As internet-delivered psychotherapy is not yet part of routine care in Germany, there might have been concerns and caveats about the practicability and safety of the intervention and the (primarily) elderly patients might be less open for such "new" approaches. Furthermore, there are no specialized post-ICU rehabilitation and outpatient ICU follow-up clinics in Germany, making it di cult to " nd" and contact patients after hospital discharge. The small sample size has resulted in a lack of statistical power. Hence, our results should be regarded as preliminary and further trials are needed to prove the e cacy of iCBT in the context of post-ICU PTSD. Another problem emerged from missing data due to premature termination. To follow the ITT principle, we imputed missing data based on the bestcase/worst-case substitution as the most rigorous method. We further included sensitivity analyses applying multiple imputation and relying on the PP population. Note that there are differences in the assumptions of these approaches re ecting common challenges in dealing with missing data. Hence, our conclusions remain fraught with uncertainty. Finally, generalizability of our results is restricted to sepsis patients and their spouses from Germany.

Conclusions
In sum, we could not prove the e cacy of iCBT in contrast to waiting in patients and spouses after intensive care treated sepsis with a presumptive PTSD diagnosis. Explorations indicate that iCBT may lead to a larger increase of relationship satisfaction in post-ICU patients and spouses than waiting. We also demonstrated that iCBT is safe and applicable for both post-ICU patients and their spouses. The largest limitation of the REPAIR trial was the small sample size so that our trial should be regarded as a pilot study and ndings remain preliminary. Future research could bene t by considering our ndings and experiences in the planning of further tailored randomized-controlled trials.    Observed PTSD symptoms (PCL-5 total score) in participants at trial assessments -strati ed by post-ICU patient and his/her spouse as well as by treatment group (iCBT / WL control group). Note: Per treatment condition, dyad membership is color-coded. Higher PCL-5 scores indicate more severe symptoms. Note that one participant (former ICU patient in iCBT group) dropped-out directly after randomization. Values are provided for several time points (including approximately time speci cations): screening (t0 -4 weeks); t0, start of intervention (iCBT group) / waiting (WL control group); t1 (t0 + 5 weeks), end of intervention (iCBT group) / waiting (WL control group). In the WL control group, the end of intervention is at t1+ (t1 + 5 weeks). Abbreviations: iCBT, internet-based cognitive-behavioral writing therapy; ICU, intensive care unit; PCL-5, PTSD Checklist for DSM-5; PTSD, post-traumatic stress disorder; WL, waitlist.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.