Coping strategies, trajectories, and their associations with patient-reported outcomes among women with ovarian cancer

The ovarian cancer prognosis and lifestyle (OPAL) Study is a national cohort study of women with primary invasive ovarian cancer that aims to evaluate the relation between lifestyle and outcomes. It was approved by the Human Research Ethics Committees of QIMR Berghofer Medical Research Institute and all participating hospitals/treatment centers. Data collection for long-term outcomes is still ongoing. A total of 1451 women aged 18–79 years diagnosed between 2012 and May 2015 were identified through 18 major Australian treatment centers. Of these, 219 women were excluded (too sick, unable to complete study documents) and a total of 958 women (78%) consented to participate. Participants completed a baseline questionnaire and then follow-up questionnaires about coping and patient-reported outcomes were mailed to participants at 3, 6, and 9 months after diagnosis and also at 12 months for some patient-reported outcomes. A total of 43 women withdrew or died before completing the 3-month questionnaire, 269 were identified later than 3 months after diagnosis, and 12 did not complete the coping scale at 3 months leaving 634 women who completed the 3-month questionnaire (with coping data) and contributed data to this analysis (Fig. 1). A comparison of participants included (n = 634) and excluded (n = 324) indicated minimal differences in baseline characteristics (Table 1).

Demographic characteristics including age and education as well as self-reported personal medical history were collected in the baseline interview.

Clinical and histopathology data regarding FIGO stage (2009 International Federation of Gynecological Oncologists classification), histological subtype and grade, treatment, and disease recurrence were abstracted from participants’ medical records.

Use of coping strategies was assessed via the Brief-COPE scale [19]. This instrument allows for use of all or a choice of selected subscales. We omitted two subscales (substance use and self-blame) which were deemed less relevant to this sample. The 24 items included assessed 12 pre-defined sub-scales (self-distraction, active coping, denial, use of emotional support, use of instrumental support, behavioral disengagement, venting, positive reframing, planning, humor, acceptance, religion) comprising two items each. Women responded to each item, indicating 0 = “I have not been doing this at all” to 3 = “I have been doing this a lot.”

Anxiety and depression at 12 months after diagnosis were assessed using the 14-item Hospital Anxiety and Depression Scale (HADS) which has well-described psychometric properties (reported Cronbach alpha coefficients for anxiety and depression = 0.93 and 0.90, respectively [20]; both 0.86 in this study). Higher scores indicate higher levels of anxiety and depression. Scoring cut-offs distinguish between “normal” (0–7), “subclinical” (8–10), and “clinical” (11–21) levels.

Quality of life at 12 months was assessed using the Functional Assessment of Cancer Therapy-General (FACT-G) which measures physical, social, emotional, and functional well-being [21]. The full ovarian cancer-specific scale (FACT-O) includes the FACT-G and an additional 11 ovarian cancer-specific items that were only measured at 3, 6, and 9 months after diagnosis. At 12 months, women were asked to complete only two of the additional ovarian cancer-specific items (about control of bowels and ability to get around) as five of the other items (stomach swelling, vomiting, stomach cramps, hair loss, and appetite) were duplicated on a separate instrument, the Measure of Ovarian Cancer Symptoms and Treatment Concerns (MOST) [22]. Scores for the missing 12-month FACT items were imputed from women’s responses to the corresponding MOST questions. The remaining four items (weight loss, appearance of body, feeling like a woman, and interest in sex) did not change appreciably over time and so women’s responses at 9 months were carried forward to 12 months. These data were then used to create an estimated FACT-O (eFACT-O) score. At 9 months, when full FACT-O data were available, the correlation between FACT-O and e-FACT-O scores was 0.99 (full FACT-O mean 121.0; SD 19.5; e-FACT-O mean 121.0; SD 19.4). The FACT scales have well-described psychometric properties (reported Cronbach alpha coefficients for the FACT-G subscales range from 0.69 to 0.82, FACT-G = 0.89 [21], the ovarian cancer-specific subscale = 0.74 and FACT-O = 0.92 [23]; in this study, subscales ranged from 0.72 to 0.90, FACT-G and eFACT-O were both 0.92). Higher scores indicate better outcomes.

First, to determine clusters of coping strategies, an exploratory factor analysis with varimax rotation was conducted using data from the Brief-COPE at the 3-month follow-up (n = 624). A factor or cluster loading cut-off of ≥ 0.5 was used to identify items for retention [24]. Two, three, and four cluster models were examined, and the model that balanced parsimony with proportion of variance explained (the three cluster model) was selected. Identified clusters were assessed for their internal consistency using Cronbach alpha coefficients.

Second, group-based trajectory modeling [25, 26] of the three coping clusters identified in this sample across follow-ups at 3, 6, and 9 months was implemented using SAS proc traj [27] to identify patterns in coping over time. To avoid convergence problems due to missing data and to ensure that the intercepts of the trajectories were estimable, we specified that women included in these models have 3-month data and ≥ 1 observation at 6 or 9 months (n = 577 of 634) (Fig. 1). We chose censored normal models to reflect that our scales were continuous. First, models with quadratic orders were fitted, ranging from a single trajectory group to a maximum of six groups. The final number of trajectory groups was then selected based on the Bayesian information criterion (BIC). We then proceeded systematically to choose the model with maximum BIC using combinations of both linear and quadratic orders to find the best fit and a minimum sample size of ≥ 5% in each group.

Finally, associations between coping group and patient-reported outcomes including anxiety and depression, quality of life (FACT-G and eFACT-O), and FACT wellbeing subscales, at 12 months after diagnosis, were investigated using general linear models. A total of 510 women out of the 577 with trajectory group assignment were included in these analyses; of those women not included, 42 died or withdrew before 12 months and 25 did not complete the 12-month questionnaire (Fig. 1). An additional 45 women dropped out of the models of eFACT-O and ovarian cancer-specific wellbeing as they did not complete the MOST questionnaire that was used to generate these scores. All models were adjusted for baseline personal factors identified in Folkman and Greer’s theoretical model of coping [17] including age, marital status, and anxiety and depression at 3 months after diagnosis, as well as recurrence status at the 12-month follow-up (yes/no) as this was important in the model adapted for people with advanced cancer [18]. The coefficients for education and FIGO stage were not statistically significant in any models and thus these variables were not included.

To assess the value of the group-based trajectory modeling, we assessed the associations between coping scores (in tertiles) at 3 months and patient-reported outcomes at 12 months after diagnosis using general linear models.

On average, the 634 women in our analysis were aged 61 years (range 21–79), most (70%) were married or living with their partner, half (51%) had completed further education after high school, and most (70%) were diagnosed with late-stage disease (FIGO stage III/IV) and were treated with surgery and adjuvant (64%) or neoadjuvant (28%) chemotherapy (Table 1). Characteristics were similar between women included in and excluded from the analysis with the exception of those excluded being less likely to have disease progression within 12 months of diagnosis (Table 1).

In the factor analysis, 15 items achieved a cluster loading of ≥ 0.5 or ≤ − 0.5 (Supplementary Table 1). Cluster 1, which we identified as “taking action and positive framing,” included seven items from four constructs (planning, active coping, positive reframing, and humor). Cluster 2, which we identified as “social and emotional support,” included all four items from the instrumental and emotional support constructs. Cluster 3 included the two denial (positive loading) and two acceptance (negative loading) items; thus, this cluster is identified as “acceptance or denial.” Due to the opposite loading directions, we reversed the scores for the denial items so that lower scores indicated more denial and higher scores indicated less denial (i.e., more acceptance). The Cronbach alpha coefficients for the clusters were 0.81, 0.78, and 0.60 respectively, demonstrating adequate internal consistency.

For the taking action and positive framing cluster of coping strategies, four distinct trajectories of use were identified: low stable (44%), medium stable (32%), medium decreasing (11%), high stable (12%) (Fig. 2a). Four distinct trajectories were identified for the use of social and emotional support coping cluster: low increasing (7%), low decreasing (44%), medium decreasing (40%), high stable (8%) (Fig. 2b). Despite variation in levels of use, most women (89%) had stable trajectories for use of taking action and positive framing while use of social and emotional support mainly (84%) declined over time. Use of acceptance or denial was also largely stable over time. Two distinct groups were identified on the acceptance or denial scale: women who “accepted their reality” (26%) or “used some denial” (74%) (Fig. 2c).

Women with high-stable use of taking action and positive framing or those who “accepted their reality” reported significantly less depression at 12 months, and those who accepted their reality also reported significantly less anxiety at 12 months (Table 2). Among women who accepted their reality, 12% (n = 16) and 7% (n = 9) reported clinical or subclinical levels of anxiety or depression, respectively, at 12 months compared to 26% (n = 96; p < 0.01) and 15% (n = 56; p = 0.02) among those who used “some denial” to cope. Six percent (n = 4) of women who had a high stable use of taking action and positive framing reported clinical or subclinical levels of depression compared with 14% (n = 61) who did not have a high stable use of this coping strategy.

Women with high stable use of social and emotional support or who accepted their reality reported significantly better quality of life at 12 months (Table 2). This association was driven by different wellbeing subscales for the different types of coping, but none was associated with physical wellbeing.

In general, better patient-reported outcomes at 12 months were reported by the 12% of women with high stable use of taking action and positive framing, the 8% of women with high stable use of social and emotional support and the 26% of women who accepted their reality. Overall, 65% (n = 375) of women were not in any of these three groups, while 26% (n = 151) were in one, 6% (35) were in two, and 3% (n = 16) of women were in all three of the positive coping behavior groups. There was a significant linear relationship between increasing number of positive coping behaviors (0/1/2+) and better overall quality of life, and having one or more positive coping behavior versus none was associated with lower levels of anxiety and depression (Table 2).

The only significant association with patient-reported outcomes found when modeling coping level in tertiles for coping clusters 1 and 2 was between higher use of social and emotional support at 3 months and better social wellbeing at 12 months. Whereas for the third coping cluster, higher acceptance of reality modeled as tertiles at 3 months was significantly associated less anxiety and depression and better quality of life and wellbeing (across all 5 subscales) at 12 months (data not shown).