To examine the effects of an expressive writing intervention (EWI) on cancer-related distress, depressive symptoms, and mood in women treated for early stage breast cancer.
To examine the effects of an expressive writing intervention (EWI) on cancer-related distress, depressive symptoms, and mood in women treated for early stage breast cancer.
A nationwide sample of 507 Danish women who had recently completed treatment for primary breast cancer were randomly assigned to three 20-min home-based writing exercises, one week apart, focusing on either emotional disclosure (EWI group) or a non-emotional topic (control group). Cancer-related distress [Impact of Event Scale (IES)], depressive symptoms (Beck Depression Inventory—Short Form), and negative (37-item Profile of Moods State) and positive mood (Passive Positive Mood Scale) were assessed at baseline and at 3 and 9 months post-intervention. Choice of writing topic (cancer versus other), alexithymia (20-item Toronto Alexithymia Scale), and social constraints (Social Constraints Scale) were included as possible moderators.
Significant (p < 0.01) group differences in mood change from before to immediately after each session suggested successful manipulation. Reductions over time in psychological symptoms were seen in both groups (p < 0.05), but no time × group interactions were found. Choice of writing topic moderated effects on IES, with women writing about other themes showing greater reductions in cancer-related avoidance than women writing about their cancer. Fewer depressive symptoms and higher levels of positive mood were seen 3 months post-intervention in women writing about their cancer when compared with the control group. Difficulties describing feelings and externally oriented thinking (20-item Toronto Alexithymia Scale) moderated effects on positive mood and IES-total, while no moderating effects were found of social constraints.
In concordance with the majority of previous results with cancer patients, no main effects of EWI were found for cancer-related distress, depressive symptoms, and mood. Moderator analyses suggested that choice of writing topic and ability to process emotional experiences should be studied further. Copyright © 2012 John Wiley & Sons, Ltd.
Research indicating that the willingness, ability, and opportunity to express cancer-related concerns may influence cancer patients' adjustment to the stressors associated with cancer [11, 12] suggests that encouraging emotional disclosure could promote psychological adjustment. One possible approach is expressive writing intervention (EWI), a brief intervention instructing participants to disclose — in writing — their deepest thoughts and feelings about a stressful life event . A wide range of benefits have been reported in a growing number of controlled trials with both healthy and clinical populations, with the first meta-analysis of 13 studies of EWI published in 1998  reporting a medium overall effect size for healthy participants (Cohen's d = 0.47). Later meta-analyses have found more modest effects in clinical samples (d = 0.19)  and in studies with both healthy and clinical participants (d = 0.15) , the latter meta-analysis including several modes of emotional disclosure.
To date, 11 randomized trials of EWI with cancer patients have been published (in 14 papers) [5-18]. Only four unique randomized studies of cancer patients [5, 11, 12, 16, 17] have included control groups writing factually about everyday topics as suggested by Pennebaker and Beall . Three studies included a control condition involving writing objectively about their cancer [8, 16, 18]. The remaining studies included either non-writing or healthy controls [6, 9, 10, 13, 15]. Overall, the available results concerning the effects of EWI in cancer patients are mixed, with most, but not all [16, 18], studies finding no main effects on psychological symptoms or quality of life.
One reason for the mixed results could be insufficient statistical power due to moderate sample sizes (mean n = 79). Another reason could be that emotional expression may not be beneficial for everyone at all times . This is supported by some of the studies, which found the effects to be moderated by individual differences in levels of social support  and social constraints [5, 12], pointing to the need for further examination of factors determining who are more and who are less likely to benefit from EWI.
Social-cognitive processing theory  suggests that when people experience social constraints on their expression of stress-related emotions and thoughts, for example if they perceive their spouse to be unwilling to listen to their concerns, this may adversely influence their psychological adjustment to stressors . Social constraints could be a possible moderator of EWI [5, 12], as EWI may provide opportunity to compensate for the lack of social support, thereby functioning as a potential buffer against the negative consequences of social constraints.
It has also been suggested that individuals who lack the skills and ability to reflect upon and process their emotional experiences will benefit less from EWI . The available results for non-cancer samples are, however, ambiguous, with some studies showing EWI to be ineffective for alexithymic and emotionally repressive individuals , while others have found EWI to be more beneficial for alexithymic individuals . Alexithymia is believed to be more prevalent in patients with psychosomatic disorders, who often show poorer results of talking therapies , and structured, time-limited writing could therefore be an opportunity for early and safe processing of traumatic material, allowing participants sufficient control over how much they want to disclose.
The mostly negative results found for EWI in cancer patients in contrast to other groups could also stem from methodological differences. In the available studies with cancer patients, the disclosure groups were generally constrained to write about experiences related to their cancer or cancer treatment. In contrast, participants in the majority of the non-cancer studies could select their own writing topics. The cancer could theoretically no longer be distressing or even be the most traumatic event for patients who have completed treatment, and when restricting patients to write about their cancer, some may thus be asked to write about a relatively non-traumatic event. No study has, so far, examined whether effects of EWI differ depending on the participants' choice of writing topic.
The present study tested whether women treated for early stage breast cancer randomized to EWI would experience larger reductions in distress compared with active controls writing about a neutral topic. We also explored possible moderating effects of social constraints, alexithymia, and choice of writing topic. On the basis of previous results [5, 12], we expected that women experiencing high levels of social constraints would benefit more from EWI than women with low levels of social constraints, while we had no a priori hypotheses concerning alexithymia and choice of writing topic.
A large nationwide, randomized controlled trial was conducted in collaboration with the Danish Breast Cancer Cooperative Group (DBCG) and 14 of the 16 surgical departments responsible for treating breast cancer in Denmark. A pilot study with 16 women tested procedure feasibility. All participants had been treated according to the Danish National guidelines [25, 26]. The study was approved by the regional scientific ethical committees and the Danish Data Protection Agency. No compensation was provided for participation.
Eligible participants were female Danish residents, able to read and write Danish, aged 18–70 years, and treated surgically within 3 weeks of their diagnosis (mastectomy or lumpectomy) for invasive breast cancer stage I and II between March and September 2006. In the planned 6-month period, 1156 women in the DBCG registry were included in our database. As shown in Figure 1, non-response and ineligibility resulted in the randomization of 507 women (54% of eligible women) (age: 27–70 years) to EWI and control (CTRL). Non-returned questionnaires and dropout during and after the intervention left 436 and 437 women with complete data at 3 and 9 months post-intervention. No cancer-related or treatment-related differences were found between participants and non-participants. Non-participants were older and more likely to be post-menopausal than participants. When adjusting for the remaining variables, only age remained significant. No baseline differences between completers and dropouts reached statistical significance (data not shown).
Eligible patients were contacted by mail between 8 and 12 weeks after surgery, or if indicated in the treatment protocol, 4 weeks after completion of chemotherapy and/or radiotherapy, and asked to give signed permission to be contacted by phone and receive further information. A baseline questionnaire, an information letter, a consent form, and a prepaid return envelope were mailed to all consenting women. If the questionnaire and signed consent had not been returned within 2 weeks, a single reminder was mailed out.
After having returned the baseline questionnaire, the women were randomized to the intervention (EWI group) (N = 253) or control group (CTRL group) (N = 254). The randomization was conducted independently by the Research Unit for Clinical Cancer Research, Aarhus University Hospital, using a stratified sampling method with four mutually exclusive strata reflecting the four standard adjuvant cancer treatment protocols (chemotherapy, radiotherapy, both, or none), and was concealed until intervention assignment.
After randomization, both groups were instructed to write for 20 min, once a week, over a 3-week period. The home-based intervention followed previously used procedures . On each writing day, research assistants contacted the participants by telephone to initiate the writing session and contacted them again after 20 min to terminate the writing. The research assistants had been trained by two clinical psychologists with experience in phone-based EWI. At 3 and 9 months post-intervention, the participants were mailed follow-up questionnaires and prepaid return envelopes. If the questionnaires had not been returned within 2 weeks, they received a single reminder.
Following the procedure described by Pennebaker and Beall , we asked EWI participants to write about a traumatic or distressing event and to explore their deepest feelings and emotions associated with this experience. They were free to write about their breast cancer as well as non-cancer experiences and to switch topics during the intervention. The CTRL group participants were asked to write as objectively and detailed as possible in an emotionally neutral manner about their daily activities. To avoid uniformity, the instructions varied slightly between sessions. All participants were instructed not to worry about grammar or spelling.
The trial hypotheses were masked by officially naming the project the ‘Stress-Management & Quality-of-Life Brief Writing Exercise Program’. Furthermore, participants were told that they were randomized to one of several groups and that the writing tasks of the other participants would not be revealed.
Data concerning treatment and other clinical variables were obtained from the DBCG registry and included the following: age, menopausal status, type of treatment (surgery, chemotherapy, or radiotherapy), tumor size, lymph node status, estrogen receptor status, and grade of anaplasia. Sociodemographic data included educational background, marital status, employment situation, and household income. The participants were asked to report use of medication, counseling, or psychotherapy.
Cancer-related distress was assessed with the Impact of Event Scale (IES) , measuring intrusive and avoidance symptoms during the last 7 days related to the stressor of having been diagnosed with and treated for breast cancer.
Depressive symptoms during the past 7 days were measured with the short 13-item version of Beck Depression Inventory (BDI-SF) [28, 29].
A Passive Positive Mood Scale (PPMS) was developed for the present study using words reflecting non-active positive mood to supplement the active positive mood items of the POMS Vigor subscale. The PPMS consists of items reflecting passive positive mood in the past 7 days (positive/bright, balanced, glad, peaceful, relaxed, at ease, calm, contented).
The Social Constraints Scale—Cancer (SCS-C)  measures how the individual perceived the social environment's response to her attempts to talk about breast cancer.
The Toronto Alexithymia Scale (TAS-20)  provides three subscales measuring Difficulties Identifying Feelings (DIF), Difficulties Describing Feelings (DDF), External Oriented Thinking (EOT), and a TAS-total score.
Writing topic was registered after each session. If the women reported that they had written ‘about emotions related to my own cancer or treatment’ in one or more sessions, they were considered belonging to the ‘own cancer’ writing group. If they had written ‘about emotions concerning traumatic topics other than my own cancer’ in all three sessions, they were considered belonging to the ‘other topics’ writing group.
As a manipulation check, all participants were asked a few questions about their writing process, and their emotional responses to the writing were assessed by phone immediately after each writing session with a 5-item version of POMS  and three items from the PPMS.
The Danish versions of the IES, BDI, SCS, and TAS-20 had previously demonstrated acceptable internal consistencies (Cronbach's alpha: 0.69–0.90) in a nationwide cohort of 3343 women treated for breast cancer [25, 34]. In the present study, all scales revealed high internal consistencies (alpha: 0.81–0.95) at all three time points. Consistency between an independent rater and self-reported EWI and CTRL writing topics was assessed with kappa statistics for a subsample of 237 essays written by 49 EWI and 30 CTRL participants (20% of complete cases).
The study had been designed to approach 1250 eligible patients over a 6-month period. Estimating a 50% inclusion rate and a 20% dropout rate between baseline and follow-up, we aimed at including 2 × 250 participants with complete data at 3 months follow-up. This would provide a statistical power of 80.1% to yield a statistically significant result (p < 0.05, two-tailed) corresponding to an effect size of d = 0.25. This effect size was similar to the overall weighted effect size (d = 0.22) found in the four comparable EWI studies of cancer patients available at the time [10, 11, 15, 16].
If there were 50% or fewer missing values on a subscale, they were substituted with the mean of the remaining filled items on the subscale for each subject . Otherwise, no total score was calculated. The randomization was evaluated by comparing baseline measures of the two groups with chi-square and Mann Whitney U tests.
Repeated-measures analyses of variance and covariance (RM ANOVAs, RM ANCOVAs) were used to compare mean distress scores at the three time points. Effects were calculated for time and for time × group interactions, the latter reflecting significant differences between EWI and CTRL over time. Variables were inspected for outliers and log-transformed (Ln) as needed .
Possible moderator effects suggested in the literature were explored as recommended  with linear regressions entering the interaction terms. The influence of writing topic was explored by repeating the aforementioned RM ANOVAs combined with planned comparisons (simple contrast) between the CTRL group and each of the two writing topic groups (own cancer and other topics). A nationwide cohort of 3343 Danish women treated for primary breast cancer [25, 34] was used as a non-writing reference group.
There were no statistically significant differences between EWI and CTRL at baseline for any of the demographic, disease, or treatment characteristics. Phone call duration was slightly longer in the EWI group than CTRLs (Table 1).
|EWI group||CTRL group||Non-participants||Dropouts (9 months)|
|N (%)||N (%)||N (%)||EWI||CTRL|
|Mean (SD)||Mean (SD)||Mean (SD)||N (%)||N (%)|
|Participants: N (%)||253 (99)||254 (100)||432 (100)||55 (100)||32 (100)*a|
|Age at study entry, mean (SD)||53.6 (9.0)||53.6 (9.2)||56 (9.0)***b||52.4 (9.8)||50 (10.0)*a|
|Education, N (%)d|
|Lower (7–13 years)||140 (57)||158 (63)||–||32 (59)||16 (50)|
|Higher (>13 years)||108 (43)||94 (37)||–||22 (41)||16 (50)|
|Marital status, N (%)d||P < 0.05a|
|Married/cohabiting||202 (80)||205 (81)||40 (73)||22 (69)|
|Singlef||47 (20)||47 (19)||24 (27)||10 (31)|
|Current employment situation, N (%)d|
|Employed/self-employed||95 (38)||88 (35)||26||13|
|Unemployed/sick leave/retirement||153 (62)||161 (65)||29||19|
|Annual household income (US$), N (%)d||P < 0.05a|
|<79.000||136 (56)||149 (60)||37 (70)||21 (64)|
|>79.000||109 (44)||100 (40)||16 (30)||11 (36)|
|Days since surgery, mean (SD)e||152 (55)||150 (56)||146 (54)|
|DBCG 2004 treatment protocols, N (%)f|
|A||62 (25)||58 (23)||106 (25)||14 (26)||9 (28)|
|B||68 (27)||79 (31)||100 (23)||15 (27)||13 (41)|
|C||68 (27)||73 (29)||144 (33)||17 (31)||2 (6)|
|D||55 (22)||44 (17)||82 (19)||9 (16)||8 (25)|
|Nodal status, N (%)|
|0||149 (59)||141 (56)||233 (54)||32 (58)||21 (66)|
|≥1||104 (41)||113 (45)||199 (46)||23 (42)||11 (34)|
|Tumor grade, N (%)|
|I||72 (28)||77 (30)||126 (29)||17 (31)||10 (31)|
|II||100 (40)||105 (41)||169 (39)||20 (36)||9 (28)|
|III||59 (23)||47 (19)||94 (22)||11 (20)||5 (16)|
|Non-ductal||20 (8)||20 (8)||34 (8)||6 (11)||6 (19)|
|Missing||2 (1)||5 (2)||9 (2)||1 (2)||2 (6)|
|Type of surgery, N (%)|
|Mastectomy||96 (38)||99 (39)||164 (38)||23 (42)||11 (34)|
|Tumorectomy||157 (64)||155 (61)||268 (62)||32 (58)||21 (66)|
|Radiotherapy, N (%)f||195 (77)||197 (78)||41 (75)||25 (78)|
|No radiotherapy||43 (17)||38 (15)||9 (16)||5 (16)|
|Missing||15 (6)||19 (7)||5 (9)||2 (6)|
|Writing session participation: N (%)|
|Session 1||230 (91)||246 (97)|
|Session 2||215 (85)||243 (96)|
|Session 3||215 (85)||241 (95)|
|Writing session characteristics, mean/SD|
|Baseline to 3 months post-intervention||150 (39)||148 (29)|
|Baseline to first writing session||31 (47)||32 (33)|
|Between sessions (first–third)||16 (15)||13 (24)|
|From third writing to 3 months follow-up||107 (29)||104 (12)|
|Duration of writing (min)|
|Session 1||20 (0.86)||20 (0.90)|
|Session 2||20 (0.68)||20 (0.68)|
|Session 3||20 (0.87)||20 (0.65)|
|Total call duration (min)|
|Session 1||38 (5.8)||34 (4.7)***c|
|Session 2||33 (5.0)||29 (3.1)***c|
|Session 3||35 (6.5)||31 (3.9)***c|
Total IES baseline scores in the EWI group were slightly lower than in CTRLs (p < 0.05). There were no overall group differences in baseline BDI-SF scores and age. However, as dropouts in the EWI group were older and had higher BDI-SF scores, these variables were included as covariates in analyses of time and interaction effects. Mean scores and standard deviations for all outcome measures at the three time points are shown in Table 2.
|EWI||243||8.6 (8.4)**||8.9 (8.0)||17.6 (14.5)**||252||4.6 (4.1)||251||7.1 (19.3)||251||18.7 (6.7)|
|CTRL||243||9.9 (8.9)||10.3 (8.7)||20.2 (15.9)||253||5.0 (4.3)||251||9.2 (21.8)||251||18.8 (7.1)|
|EWI (cancer topic)b||107||8.9 (8.5)||8.6 (7.7)||17.5 (14.2)||100||4.4 (3.4)||100||6.1 (17.2)||100||19.5 (6.4)|
|EWI (other topic)c||81||7.9 (8.0)*||9.9 (8.5)||17.8 (14.2)||81||4.2 (3.4)||80||5.8 (18.1)||80||19.0 (6.8)|
|Reference groupd||3318||10.1 (8.9)||10.0 (8.8)||20.1 (15.9)||3324||5.0 (4.5)||–||–|
|Post-intervention (3 months)|
|EWI||199||7.5 (7.4)||8.2 (7.9)||15.6 (13.6)||197||4.0 (4.1)||198||4.5 (18.5)||198||19.4 (7.0)|
|CTRL||222||8.0 (7.6)||9.7 (8.8)||17.7 (14.8)||221||4.5 (4.4)||215||5.3 (20.8)||215||19.8 (7.0)|
|EWI (cancer topic)||99||7.3 (7.2)||7.9 (7.4)||15.2 (12.8)||98||3.5 (3.8)||98||1.6 (16.5)||98||20.8 (6.1)|
|EWI (other topic)||76||7.0 (7.2)||8.1 (8.8)||15.1 (14.7)||75||4.2 (4.1)||76||6.5 (19.1)||76||18.1 (7.4)|
|Follow-up (9 months)|
|EWI||207||7.5 (7.4)||8.1 (8.0)||15.6 (13.7)||209||4.2 (4.3)||206||4.6 (18.8)||206||19.6 (6.8)|
|CTRL||223||7.3 (7.7)||8.8 (8.7)||16.1 (14.5)||226||4.5 (4.6)||225||6.7 (22.1)||225||19.5 (6.9)|
|EWI (cancer topic)||99||7.5 (7.0)||8.5 (7.9)||16.0 (13.2)||101||3.9 (4.2)||101||3.0 (17.9)||101||20.4 (6.4)|
|EWI (other topic)||81||7.0 (7.0)||7.6 (8.2)||14.6 (13.5)||81||4.2 (4.1)||79||5.7 (18.4)||79||18.8 (6.8)|
|Reference groupe||2912||7.8 (7.9)||8.4 (8.7)||16.2 (15.3)||3039||4.3 (4.5)||–||–|
Most EWI participants (n = 108) reported to have written about their breast cancer in at least one of the three sessions. The remaining women (n = 85) wrote about other personal traumas in all three sessions. Women who had received mastectomy, had received chemotherapy, or had used pain medication and professional help were more likely to write about having breast cancer than other traumas (chi2 tests; p = 0.012–0.026). No other group differences were found at baseline (data not shown).
Compared with CTRL, the EWI group showed significant changes in the expected directions in negative [Wilks' lambda (λ) = 0.79, F(5, 442) = 24, p < 0.001] and positive mood [λ = 0.85, F(5, 442) = 15.7, p < 0.001] immediately after the writing sessions (RM ANOVAs) (data not shown). The effect sizes were large [multivariate partial eta squared (hp2): 0.15 to 0.21 ≈ Cohen's d: 0.84–1.04].
Repeated-measures analyses of variance revealed statistically significant reductions in cancer-related distress (IES), depressive symptoms (BDI-SF), and negative mood (POMS) (F: 6.43–21.26; p < 0.001). No effects of time were found for positive mood (PPMS, POMS Vigor). The effect sizes were medium to high with hp2 ranging between 0.03 and 0.09 (≈ d between 0.36 and 0.64). When including age and baseline levels of BDI-SF as covariates, no effects of time reached statistical significance. No time × group interactions reached statistical significance in any of the analyses, with or without covariates entered into the model. Effect sizes for time × group interactions were small, with hp2 ranging between 0.001 and 0.012 (d < 0.20) (A complete table of RM ANOVA statistics, adjusted effect sizes (hp2), and effect sizes based on unadjusted change scores can be obtained from the authors by request).
No statistically significant moderation effects of social constraints (SCS-C) were found in any of the analyses, with or without covariates in the model (data not shown).
Scores on the TAS-20 EOT subscale moderated the effect of EWI on IES-total scores at 3 months (beta: 0.26, p = 0.021), with lower EOT scores in the EWI group (but not CTRL) being associated with greater reductions in IES scores from baseline to 3 months. TAS-DDF moderated changes in positive mood (PPMS) from baseline to 3 months (beta: −0.24, p = 0.049), with higher scores on TAS-DDF being associated with increase in PPMS in the CTRL group, but not in EWI group. No other moderation effects were found for TAS-20 total and subscale scores (data not shown).
Repeated-measures analyses of variance with Writing Topic Group as the independent variable revealed a significant interaction (time × Topic-Group) for the IES-avoidance subscale [effect size (hp2): 0.05 ≈ Cohen's d: 0.46], suggesting greater reductions over time in IES-avoidance when writing about other topics than about one's own cancer. However, none of the contrast analyses revealed significant differences when comparing either of the two writing subgroups to controls (SPSS, K-matrix). No further interaction effects (time × Topic-Group) reached statistical significance, when including all three time points in the model.
With only baseline and 3-month follow-up included in the model, the analyses revealed a near-significant (p = 0.052; d = 0.25) interaction effects for the POMS and statistically significant interactions for BDI-SF (p = 0.013, d = 0.29) and PPMS (p = 0.014, d = 0.30), both suggesting greater changes over time in the expected direction when writing about own cancer relative to writing about other topics. These findings remained statistically significant, when adjusted for age, baseline BDI-SF, mastectomy, chemotherapy, pain medication, and professional help. The number of sessions (0–3 sessions) the participants wrote about their own cancer did not influence any outcomes (data not shown). The inter-rater reliability between an independent rater and the randomly selected subsample of 237 essays was 0.75 (kappa statistic, p < 0.000).
The mean reference group IES and BDI-SF scores at baseline, 3 months after surgery, and at follow-up, 12 months later, are shown in Table 2. The EWI group had significantly lower IES-intrusion and IES-total scores at baseline than the reference group. When writing topic groups were analyzed separately, the Other Writing Topic group had lower IES-intrusion scores than the reference group. When change scores were analyzed (data not shown), the EWI group showed smaller reductions over 9 months in IES-intrusion and IES-total scores than the reference group over 12 months. Likewise, the Own Cancer Writing Topic group showed smaller reductions in IES-total over 9 months than the reference group over 12 months. No other differences were found.
The results of this large, randomized controlled trial with a nationwide sample of women treated for breast cancer did not confirm our hypothesis that EWI participants would experience greater reductions in distress when compared with an active control group. The effect sizes found were generally small, with the largest (Cohen's d = 0.07) indicating a small reduction in avoidance at the 3-month follow-up. Likewise, at 9 months, the effect sizes were small, with several effect sizes in the opposite of the expected direction. Our results are in concordance with results of previous methodologically comparable studies with cancer patients showing no main effects of EWI on various measures of emotional distress [5, 11, 12, 16, 17]. While this could be seen as supporting the suggestion that EWI may not be broadly effective as a psychotherapeutic intervention for this group , other reasons for the negative result should also be considered.
First, the participating women may not have been sufficiently distressed for the intervention to be effective (‘floor effect’). Cancer patients, who agree to participate in intervention studies, may be relatively emotionally well adjusted , and, consistent with this possibility, the pre-intervention IES scores of the EWI group were significantly lower than scores in a national reference group. On the other hand, cancer patients have been found — both in general and in the present study — to be more distressed and depressed than healthy women , and EWI has previously been shown to reduce emotional distress in healthy participants [39-41].
Second, if one does not feel in need of psychosocial treatment, it may be difficult to mobilize sufficient motivation to engage in the writing sessions wholeheartedly, and some participants may have participated with only a minimum of enthusiasm. While we omitted measures of motivation or general expectancy so as to not threaten the blinding of the two conditions, baseline measures of motivation and general expectancy could have yielded important information.
Third, the home-based intervention used in both the present and previous studies may be less effective than laboratory-based EWI. A home-based design was chosen to facilitate a population-based, nationwide trial, and the pilot study results supported the feasibility of this design. We also viewed the home environment as more likely both to provide privacy, a factor found to be important for successful emotional disclosure , and to reduce the risk of confounding influence by nonspecific effects of a laboratory setting. While the home environment may have increased non-adherence, the manipulation check suggested that this was not the case.
On the basis of earlier findings [5, 12], we also hypothesized that women experiencing high levels of social constraints, that is, with less access to a social network of individuals willing to listen to their problems, would benefit more from EWI by providing them with the opportunity to disclose their thoughts and feelings. Taken together, our data did not support this hypothesis.
Possible moderating effects of alexithymia were also explored, as participants scoring high on alexithymia had previously been reported to benefit both more  and less from EWI  than low alexithymic participants. Greater difficulties describing feelings were associated with increased positive mood in controls, but not EWI participants, and lower levels of externally oriented thinking with greater reductions in IES-total scores in EWI participants but not controls, both only at 3 months. The results indicate the need for further investigations of associations between ability to process emotional experiences and the experienced gains from EWI.
We left the EWI group free to write about other traumatic topics, as we considered the possibility that some patients may benefit more from writing about other experiences that have been more stressful for them and that the lack of main effects found in previous EWI trials with cancer patients could be due to that writing about their cancer experiences could be too constraining . That choice of writing topic could influence the results was partly confirmed. Writing about one's own cancer was associated with greater short-term reductions in depressive symptoms (d = 0.29) and greater increases in positive mood (d = 0.30), both compared with controls and writing about other traumas. On the other hand, when compared with controls, women writing about other topics than their cancer reported greater reductions in IES-avoidance at 3 (d = 0.25) and 9 months (d = 0.20) follow-up. As the IES referred to cancer-related intrusive thoughts and avoidance, it may seem puzzling that writing about other topics should reduce these symptoms. One explanation could be that it is beneficial to be distracted by focusing on other distressing experiences. As all traumatic experiences have similarities, for example feelings of helplessness, it is also possible that addressing such feelings underlying one trauma could have an impact on feelings related to another traumatic event.
While our randomized clinical trial has several strengths, including a large, population-based sample providing the ability to adjust for a number of potential confounders, a number of limitations could have influenced the results. First, 44% of the eligible women actively refused or failed to respond to the invitation to participate. Another issue could be the slightly longer duration of telephone calls in the EWI group due to more spontaneous talk initiated by the participants during debriefing, and we cannot rule out this debriefing as a confounder. Finally, there are several issues concerning writing topic. One concern could be that we distinguished between women who wrote about their own cancer experience in any one of three sessions and women who wrote about another traumatic experience in all three sessions. The variation in writing dose, however, did not influence the results. Second, writing topic was categorized according to self-report, rather than an in-depth qualitative analysis of all essays. While the inter-rater reliability between an independent rater and self-reported writing topics appeared acceptable for a subsample of essays, it is unknown whether the writing topic effects would hold up, if writing topics had been determined independently. Third, we do not know the reasons for choosing not to write about cancer. The reason could be that for some women, the cancer was no longer stressful, while for others the cancer was experienced as too stressful to write about. Including a measure of perceived or observer-rated stressfulness of their cancer diagnosis and treatment could have been relevant. A fourth potential confounder could be the recency of the non-cancer trauma. While we assume to be comparing cancer with non-cancer writings, we could in fact be comparing the recency of the trauma. Although self-selected writing topic may increase external validity, lack of randomization may limit the generalizability of differences found between writing topic groups. Finally, while we considered a three-arm trial with a third group randomized to a group restricted to write about their cancer, this was abandoned because due to statistical power considerations. These limitations should be considered in future EWI studies with cancer patients.
The study was supported by the Danish Cancer Society (grant no. PP04034) and the Faculty of Business and Social Sciences, Aarhus University.
The authors have no conflicts of interest to report.