Cognitive bias modification training (IVY) countering fatigue in people with breast cancer: A waitlist-control feasibility study

Abstract

Objectives

The feasibility of a fatigue Cognitive Bias Modification training was evaluated in women on treatment for breast cancer in a multi-center waitlist-control design assessing feasibility criteria, such as recruitment, retention, and completion rates, as well as effects on fatigue bias.

Methods

Five hospitals were each asked to recruit 30 patients, who were sequentially divided in active and delayed treatment groups. Fatigue bias and self-reported outcomes (fatigue, vitality, avoidance, and all-or-nothing behavior) were measured in baseline, training, and follow-up phases.

Results

Feasibility results were mixed with recruitment and retention not meeting predetermined criteria, but completion and variability were judged positively. Training effects on fatigue bias compared to control were found.

Conclusions

This study evaluated a promising novel fatigue intervention. Continuation with an RCT is encouraged with recruitment and retention strategies reconsidered.

Trial registry

This study’s protocol (Geerts, et al., 2024) was preregistered at the Open Science Framework on October 31st, 2023.

Fatigue is one of the most prevalent and persistent symptoms of breast cancer and breast cancer treatment with cancer-related fatigue (CRF) experienced in 77% of newly diagnosed patients (Cimprich, 1999), 60–93% of patients undergoing radiation, 80–96% of patients undergoing chemotherapy (Bardwell and Ancoli-Israel, 2008; Schmidt et al., 2012), and 54% of patients in the palliative setting (Peters et al., 2014). CRF is defined as a subjective, persistent and distressing sense of physical, cognitive, and/or emotional tiredness or exhaustion that interferes with usual functioning and is not proportional to recent activity (Berger et al., 2015; Jankowski et al., 2023). Many patients perceive fatigue as the most distressing symptom both during and after cancer treatment compared to other cancer-related symptoms, such as pain, nausea and depression, as it has greater negative impact on patients’ quality of life (Berger et al., 2015; Hofman et al., 2007).

Despite the prevalence and impact of CRF, it remains underreported, underdiagnosed, and undertreated, probably due to its complex and multifactorial nature (Berger et al., 2015; Bower, 2014; Jankowski et al., 2023; Mitchell et al., 2014). Most recommended treatment options to manage CRF are (combinations of) exercise, psychological, and pharmaceutical interventions (Beenhakker et al., 2022; Bower, 2014; Mustian et al., 2017). In a meta-analysis, exercise and psychological interventions, such as cognitive behavioral therapy (CBT), were shown to be effective for reducing CRF and superior to pharmaceutical interventions (Mustian et al., 2017).

These interventions mostly focus on reflective ways to change cognitions and behaviors. However, health behavior is often not solely guided by conscious thought (Sheeran et al., 2013). Indeed, Bootsma et al. (2021) recognize that the formation of habits to cope with CRF involves both unconscious and embodied experiences as well as overt learning processes and conscious reflection. This is in line with dual-process models, e.g., the Reflective-Impulsive Model (Strack and Deutsch, 2004), that recognize two integrated and interactive systems: one conscious, reflective and rational, the other implicit, impulsive and associative (Hagger, 2016; Sheeran et al., 2013).

The implicit processes are often described as cognitive biases, “a broad class of automatically activated processes that may persist even when they conflict with conscious goals” (Wiers and Wiers, 2017: 81). For example, people who survived breast cancer have been shown to automatically pay more attention to cancer related words compared to healthy controls, independent of their self-reported fear of cancer recurrence (Custers et al., 2015). Similarly, people with multiple sclerosis and people with chronic fatigue syndrome (CFS) tend to automatically interpret ambiguous somatic information in a threatening way more than healthy controls (De Gier et al., 2024). Furthermore, a combination of attentional and interpretational biases was also found in people with CFS (Hughes et al., 2016).

Cognitive biases have also been observed with respect to pain, a similarly multifactorial and complex symptom as fatigue (Eccles and Davies, 2021). In this field multiple theories have been proposed about how and why cognitive biases related to disease symptoms arise and develop (e.g., Lenaert et al., 2018). The schema-enmeshment model (Pincus and Morley, 2001) proposes that cognitive biases are the result of overlap between three schemas representing the symptom, illness and the self. When the symptom is frequently or continuously experienced, the symptom schema can become enmeshed with the illness and self-schemas (Pincus and Morley, 2001). For example, more schema-enmeshment in people with fibromyalgia was associated with greater pain severity and interference, catastrophizing, impact of symptoms, and depression (Paschali et al., 2021).

Cognitive bias modification (CBM) interventions aim to change unhelpful cognitive biases by systematically practicing an alternative processing style (Hertel and Mathews, 2011). Meta-analyses and reviews show robust effects of CBM on anxiety and promising effects on depression (Fodor et al., 2020; Jones and Sharpe, 2017; Salemink et al., 2023), while individual studies also show promising results in chronic pain (Schoth et al., 2013; Sharpe et al., 2012), acute pain (Sharpe et al., 2012), and fibromyalgia (Carleton et al., 2011). Furthermore, in comparison to traditional reflective interventions for CRF, CBM requires relatively little effort and cognitive resources (Bowler et al., 2012; Hallion and Ruscio, 2011), making it an easy, accessible and inexpensive intervention that can easily be converted to an online 24/7 context (Bowler et al., 2012; Wolbers et al., 2021; Wolters et al., 2021). These aspects could be especially advantageous for targeting fatigue in breast cancer and make CBM easy to combine with other interventions.

To the authors’ knowledge, the presence of fatigue biases in people with breast cancer has not been researched yet. As CRF is common in newly diagnosed patients and becomes more prevalent with cancer treatment (Bardwell and Ancoli-Israel, 2008), it is an interesting patient group to research when and how fatigue biases develop and influence this process. Therefore, a broad range of patients was studied: people in the curative, (neo)adjuvant setting as well as in the palliative or metastatic setting. Moreover, to the authors’ knowledge, the authors’ research team is the first to introduce CBM targeting fatigue bias. Following iterative human-centered design frameworks (Burns, 2018), the authors recently co-designed and evaluated multiple versions of a fatigue CBM intervention with people with breast cancer (Geerts et al., 2025) and people with chronic kidney disease (CKD; Geerts et al., 2025) and CBM effects were found on fatigue biases in people with chronic kidney disease (Geerts et al., 2025).

In the current study, a CBM training targeting self-identity bias related to fatigue and vitality was developed. The training was inspired by the schema-enmeshment model (Pincus and Morley, 2001) and based on the approach-avoidance paradigm, an often-used and proven effective CBM method, particularly in the context of consumer behavior (Kakoschke et al., 2017). This training was implemented in an eHealth app and investigated with a feasibility wait-list control trial. This study aimed to evaluate the feasibility of continuing with a full trial by assessing (1) recruitment, inclusion, and retention rates, (2) completion rates of the measurement and training sessions, (3) variability in demographic and outcome variables, such as participant characteristics and floor- and ceiling effects to assess the suitability of the materials used, and (4) time-series graphs to explore potential effects of the training on the self-report measures. Additionally, the feasibility was judged sufficient to (5) statistically test the effect of the training on self-identity bias.

Method

This paper was written in accordance with the 2010 Consort statement extension for randomized pilot and feasibility trials ((Eldridge et al., 2016), see the Consort 2010 checklist in Additional file 1).

Design

This study is a semi-randomized multi-center waitlist-control feasibility study conducted with people on treatment for breast cancer, in both the curative (neo)adjuvant and the metastatic setting. The study aims were assessed with mixed methods in an embedded design (Bishop, 2015). Feasibility (aims 1–4) was assessed descriptively and visually. These assessments were supplemented with qualitative data, including reasons for dropping out and feedback from participants (see Additional file 3). Feasibility was judged sufficient only to explore effects on self-identity bias (see Deviations from Protocol in Additional file 2).

Effects on self-identity bias (aim 5) were explored in two ways: with between-group (active treatment vs delayed treatment) and within-group (delayed treatment control group) analyses. In both analyses self-identity bias was the outcome variable and time was the within-subjects factor, added by group and their interaction in the between-group analyses. Before both analyses, baseline correlations between the outcome variables were explored. Predetermined feasibility criteria (Geerts et al., 2024; see Table 1) were an 80% inclusion rate and a 20% drop-out rate (Wood et al., 2004), as well as a medium-sized or larger effect of the training on cognitive bias comparing training to control. Furthermore, feedback from participants including serious concerns for acceptance were considered reasons to revisit the study before continuation to a full trial.

Table 1.

Overview of the predetermined feasibility criteria for recruitment, inclusion and retention.

Feasibility criteria	n (%)
Recruitment	150 (100)
Inclusion	120 (80)
Retention	96 (80) a

^aRetention is based on the number of participants included.

Participants

Five hospitals were each asked to recruit 30 patients with breast cancer (15 (neo)adjuvant, 15 metastatic) for this study. Recruitment was planned between end of June 2022 and end of February 2023 and was extended by 1 month. Inclusion criteria were (1) undergoing curative treatment for breast cancer, i.e., (neo)adjuvant chemo (immuno)therapy, antihormonal treatment or palliative systemic therapy, (2) adequate Dutch speaking and reading skills, (3) having a smartphone and the skills to use it, and (4) having a computer with internet and the skills to use it. Exclusion criteria were (1) only undergoing immunotherapy treatment in the adjuvant setting, (2) only undergoing anti-hormonal treatment in the metastatic setting, (3) insufficient Dutch speaking and reading skills, (4) not having a smartphone, a computer, or the skills to use them.

Eligible patients were approached by their oncology specialist, who introduced the study and provided the Patient Information Form. If a patient expressed interest, the specialist asked for permission to share their contact details with the researchers. Upon receiving this permission, the researchers contacted the patient to answer any questions and obtain informed consent. Initially it was planned to allocate the patient groups (curative and palliative) separately, because of their heterogeneity, however, the researchers did not have the medical information upon inclusion. Therefore, all patients were sequentially allocated to the active treatment and the delayed treatment groups by the researchers. The recruiters were blinded for this allocation.

Materials

Each measurement started in the online survey platform Qualtrics (https://www.qualtrics.com/), which was used for email distributions, where participants were automatically sent to the online platform Inquisit Web (https://www.millisecond.com/) for the self-identity bias measurement, and back to Qualtrics for the questionnaires. The first measurement included video instructions about the measurements and the two platforms. Each measurement ended with an open-ended question inviting participants to leave any additional comments.

Self-identity bias

A modified Implicit Association Task (IAT; Greenwald et al., 1998) was used to measure self-identity bias. In seven blocks (5 practice blocks and 2 measurement blocks) of reaction-time tasks, participants categorized randomized stimuli (see Appendix A) related to ‘I’, ‘other’, ‘fatigue’ and ‘vitality’ by pressing the ‘E’ and ‘I’ keys on the keyboard. In the first two practice blocks (20 trails each), participants practiced the ‘I’ versus ‘other’ and the ‘fatigue’ versus ‘vitality’ categories separately, after which the combination ‘fatigue and I’ versus ‘vitality and other’ were practiced (also 20 trials) and measured (40 trials). Then the ‘I’ and ‘other’ categories switched sides. This switch was first practiced with the ‘I’ versus ‘other’ categories (20 trials), then the new combinations (‘fatigue and other’ vs ‘vitality and I’) were practiced (20 trials) and measured (40 trials). A D-score between −2 and 2 was obtained by comparing congruent (‘fatigue and I’) and incongruent (‘vitality and I’) measurement blocks, calculated as the intra-individual standardized difference in mean reaction-times (Greenwald et al., 2003). Positive scores corresponded with fatigue bias, while negative scores corresponded with vitality bias.

Self-reports

In the first measurement, participants were asked about demographic data, specifically gender, age, height, weight, ethnicity, marital status, education, and comorbidity. These variables were collected to provide a comprehensive overview of the sample and to assess its representativeness compared to the general population. Subsequently, participants answered the behavioral subscales of the shortened version of the Cognitive and Behavioral Responses Questionnaire (CBRSQ; Ryan et al., 2018) with the past week in mind. The avoidance/resting subscale contained eight items (e.g., “I sleep during the day to keep my fatigue under control”) and the all-or-nothing subscale contained five items (e.g., “When it comes to doing things, I’m an ‘all or nothing’ kind of person”) that were answered on five-points frequency scales (0 = never, 4 = always) and added so that high total scores indicated more frequent avoidance/resting or all-or-nothing behavior. The subscales showed acceptable to good internal consistency with a Cronbach’s alpha of .80 for the avoidance/resting subscale and a Cronbach’s alpha of .72 for the all-or-nothing subscale, which is reasonably in line with previous research that found a Cronbach’s alpha of .76 for the avoidance/resting subscale and .85 for the all-or-nothing subscale (Loades et al., 2020).

Following, vitality was measured with the Dutch Vitality measure (Vita-16; Strijk et al., 2015) with a total of 16 items (e.g., “I have enough energy to fulfil my daily tasks”) divided over three subscales measuring energy, motivation, and resilience, answered with the past week in mind. Answers were given on a 7-point Likert scale (1 = never, 7 = always) and averaged with high total scores indicating higher vitality. With a Cronbach’s alpha of .92 the scale showed excellent reliability, which is in line with literature (Strijk et al., 2015).

Next, fatigue was measured with the Checklist Individual Strengths (CIS; Vercoulen et al., 1994) with a total of 20 items (e.g., “I feel fit”) divided over four subscales measuring fatigue severity, activity, concentration and motivation, answered with the past week in mind. Answers were given on a 7-point Likert scale (1 = No, incorrect, 7 = Yes, correct) and added so that higher total scores represented high fatigue and low motivation, concentration, and activity levels. Scores higher than 76 represented problematic fatigue (De Vries, 2003). With a Cronbach’s alpha of .94 the CIS showed excellent reliability, which is in line with literature (Schulte-van Maaren, 2014; Vercoulen et al., 1994).

Intervention

In the IVY (Implicit VitalitY) CBM training administered via the Twente Intervention and Interaction Machine (TIIM) app (Van ‘t Klooster et al., 2024), participants were instructed to categorize stimuli (e.g., ‘active’, ‘exhausted’), that randomly and sequentially appeared in the middle of the screen, to 4 categories, ‘vitality’ and ‘I’ at the top of the screen or ‘fatigue’ and ‘other’ at the bottom of the screen. Participants could swipe the in total 120 stimuli up or down to categorize them to the right category. The positions of the categories were chosen in this way to correspond with an approach (towards me) – avoidance (away from me) principle, which was enhanced by a zoom function that made stimuli larger when swiped down and smaller when swiped up. The goal of the training was to strengthen the association between self-identity and vitality and weaken the association between self-identity and fatigue.

A week before the training phase (for the active treatment group in week 4, for the delayed treatment group in week 12) participants received instructions to download and use the application via email. The researchers called the participants if they were not registered to the app at the appropriate moment. A Frequently Asked Questions (FAQ) page was added to the application, and participants were encouraged to contact the researcher if they had additional questions. To encourage engagement and understanding, three short videos where members of the research team gave further information and tips for the training were added each after 5 training sessions (1 week of training).

All participants did the 4-week training with 5 sessions per week. The training sessions were provided around 8:00 am and a reminder was sent around 7:00 pm (both were provided in the form of an app notification). Other than that, participants were given the freedom to do the sessions at any moment during the day. Three months after the first training phase, the active treatment group also received so-called booster sessions. In the booster phase, participants were provided with three booster sessions on Monday morning but were asked to complete at least two booster sessions at any moment during the week.

Procedure

The active treatment and delayed treatment groups both underwent a baseline, training, and follow-up phase, which for the active treatment group was extended with an extra training (booster) phase (see Table 2). The active treatment group received the IVY training immediately after baseline, whereas the delayed treatment group served as a waitlist control and received the training after a 12-week baseline period (i.e., a 12-week delay). For the active treatment group, the 4-week baseline phase included one measurement per week. For the delayed treatment group, the 12-week baseline phase mirrored the first 4 weeks of the active treatment group and was followed by fortnightly measurements. Following the baseline phase, all participants completed the 4-week training phase with fortnightly measurements. This was followed by a 3-month follow-up period for both groups, with weekly measurements in the first month and fortnightly measurements in the subsequent 2 months. Subsequently, the active treatment group received a 1-month booster phase with fortnightly measurements, followed by a final 1-month follow-up phase with weekly measurements. The bias measurement and training took no more than 5 min. Together with the questionnaires, a full measurement took about 10 min. The measurements became available on Monday 8:00 am and a reminder was sent on Wednesday 8:00 am, but participants were free to do them at a convenient time during the week.

Table 2.

Schematic overview of the study design.

Group	Month 1	Month 2	Month 3	Month 4	Month 5	Month 6	Month 7
Active IVY-treatment measurements trainings	Baseline 4	Training 2	Follow-up 4	Follow-up 2	Follow-up 2	Booster 2	Follow-up 2
		20				8
Delayed IVY control measurements trainings	Baseline 4	Baseline 2	Baseline 2	Training 2	Follow-up 4	Follow-up 2	Follow-up 2
				20

Data analyses

For aim 1 (recruitment, inclusion, and retention rates) recruitment was assessed by the number of patients judged as eligible by the oncology specialists. Inclusion was assessed with the number and percentage of participants enrolled to the study from the people recruited as well as by the recorded reasons for not participating. Retention was assessed with the number and percentage of participants dropping out of the study, reasons for dropping out, and feedback from participants. For aim 2 (Measurement and training completion rates) the number and percentage of recorded measurements and training sessions for each participant were assessed. All aim 1 and 2 outcomes were monitored and assessed with Excel.

For aim 3 (variability in demographic and outcome variables) characteristics of included participants were explored by analyzing demographic data. Moreover, mean scores at baseline were explored to assess fatigue bias and fatigue symptomatology in the two patient groups. Additionally, to assess the suitability of the questionnaires, variability in the data, such as floor- and ceiling effects, were investigated. These assessments were done with descriptive analyses in SPSS (version 28.0.1.0). Aim 4 (time-series graphs on the self-report measures) were conducted with Excel using estimated marginal means obtained from repeated-measures linear mixed model (LMM) analyses in SPSS. The LMM analyses were only used with this purpose as this study did not have enough power to test effects of the training on the self-reported outcomes. In the time-series graphs, the outcome variable was on the vertical y-axis while time (phases) was on the horizontal x-axis.

As the current study did have enough power for testing effects with medium effect size, which was expected for aim 5 (training effect on self-identity bias), significance testing was applied in two ways: between-group (active treatment vs delayed treatment) and within-group (delayed treatment). Specifically, in the between-group analysis the active and delayed treatment groups were compared in the first 3 months of the study with time (4: baseline, training, post, follow-up), group (2: treatment/delayed treatment) and time*group interaction as fixed factors. In the within-group analysis, the delayed treatment group’s pre-to post was compared with time (5: baseline, training, post, follow-up 1, follow-up 2) as fixed factor.

The delayed treatment group was chosen for the within-group analysis as this group had the best design for this analysis (see Table 2) as the follow-up phase in the active treatment group was interrupted by the booster phase. Follow-up 1 (measurement 12 to 14) consisted of the weekly measurements after the post measurement, follow-up 2 (measurement 15 to 18) consisted of the fortnightly measurements in the last 2 months of the study. For both analyses, participants’ measurements were averaged per phase to minimize the influence of missing data on the analyses. Before both analyses, the baseline phases were used to explore Pearson correlations between the outcome variables to verify the theory behind the training (see Additional file 3).

In the between-group analyses, four participants that were allocated to the training group had not done any training, so they were assigned to the delayed treatment control group. In the within-group analysis, 9 of the 31 participants in the delayed treatment group (29%) had not done any training and 4 participants (13%) had done fewer than the 20 training sessions assigned (3, 11, 15, and 18, respectively). The nine participants (6 dropped out before the training phase) were deleted; the 4 participants were included in the analysis, reducing the sample size to 22.

Ethical approval

According to the Committee of Human Research [Commissie Mensgebonden Onderzoek, CMO] this study was not applicable (file number: 2021-13261) to the Dutch Medical Research Involving Human Subjects Act and redirected the ethical approval to local ethical committees. Both the ethical committees of the faculty Behavioral, Management, and Social Sciences of University of Twente (file number: 220004) and the hospital (file number: ZGT22-09) approved this study.

Results

Feasibility outcomes

Recruitment

Two of the five hospitals asked to recruit patients did not recruit any participants, one because the process to get local ethical approval took too long, while the other hospital reported having no dedicated breast oncologist and small numbers of breast cancer patients as reasons. One hospital only recruited four patients, all in the (neo)adjuvant setting. The other hospitals recruited 57 and 24 patients, respectively. Of the in total 85 recruited patients, 64 (75%) agreed to participate in the study (see Figure 1). Thus, only 53% of the anticipated 120 participants were included, not meeting the predetermined feasibility criterium of 80% (see Table 1).

Figure 1.

Flow-chart of recruitment, enrolment, inclusion, and retention in the current study.

Retention

Of the 64 included participants, 38 (59%) completed the study (see Figure 1). Thus, 26 (41%, n_{active treatment} = 15, 58%) participants dropped out of the study, which is higher than the predetermined retention rate of 20% (see Table 1). Most participants that dropped out did so in the first 4 weeks (n = 18, 69%). Seven of the dropouts did so before the informed consent was received, therefore, these participants were deleted from the analysis. Within the included participants, patients in the metastatic phase were clearly underrepresented (only 9 of 38 participants, 24%).

Training and measurement completion rates

Because recruitment and retention were challenging, we aimed to maximize the use of all available data. Therefore, participants who discontinued the study after one or more phases were still included in the analyses for the phases they completed (e.g., baseline or training). As a result, the sample sizes (n) reported in the following reflect the number of participants with valid data per measure and time point, which may differ from the total number of included or completed participants described in the above text and Figure 1.

Training. Forty-eight participants (75%; 25 active group, 23 delayed treatment group) registered to the app, a few some weeks later than expected because of technical or personal reasons. Of these registered participants, 34 (71%; 18 active group, 16 delayed treatment group) completed all 20 training sessions and six participants (12.5%) did half or more of the training sessions. Six participants (24%) in the active treatment group also completed all booster sessions. Four participants (8%) had registered to the app without doing any training sessions (but these people also had not received notifications or reminders) and some participants only completed 1 or 2 sessions. All training sessions that were started were also completed.

Measurements. Of the 38 participants that completed the study, eight participants (21%) completed all 17 IAT and questionnaire measurements. Most participants (58% in the IAT measurements, 47% in the questionnaire measurements) only missed one or two measurements. Only 7 (18%) of participants completed 12 or less measurements. Sometimes participants missed part of the measurement (e.g., only IAT or questionnaires). This happened in 12 (22%) participants.

Variability in demographic and outcome measures

Participants had a mean age of 50 (32–77, SD = 9.9) and a mean Body Mass Index (BMI) of 25.8 (17.9–38.8, SD = 5.0). Other demographic characteristics can be found in Table 1 in Additional file 3. The BMI distribution in this sample is in line with the BMI distribution of women in the Netherlands in 2022 (Statistics Netherlands, 2022). Most participants (n = 21, 45%) started participating in this study one or 2 months after start of the treatment (see Table 2 in Additional file 3), which is in line with the authors’ previous study (Wolbers et al., 2021) that recommended the IVY training after the first treatment wave. All participants that started the study 5 or more months after start of treatment were patients in the metastasis phase, who generally receive treatment for a longer time.

Tables 3-5 depict the between-group, within-group, and between patient groups descriptive outcomes. No floor or ceiling effects were found. Notably, mean bias scores were already negative at baseline indicating a vitality bias (M_total = −0.22, slight vitality bias, SD_total = 0.06). In contrast, mean self-reported fatigue scores (M_total = 73.52) were close to or exceeded (metastasis group: M = 80.57, SD = 23.94) the cut-off for problematic fatigue (De Vries, 2003). Otherwise, scores were very similar between the two patient groups.

Table 3.

Descriptive outcomes of the between-group analyses.

	Self-identity bias			Self-reported fatigue			Self-reported vitality			Self-reported all-or-nothing behavior			Self-reported avoidance behavior
	N	M	SD	N	M	SD	N	M	SD	N	M	SD	N	M	SD
Baseline
Total group	50	−0.22	0.45	49	73.52	22.72	49	4.45	1.07	48	7.63	3.33	48	8.60	4.15
Active group	21	−0.16	0.43	21	73.18	24.49	21	4.63	1.05	20	8.50	3.41	20	8.94	4.40
Delayed group	29	−0.27	0.46	28	73.78	21.76	28	4.31	1.09	28	7.00	3.18	28	8.35	4.03
Training
Total group	41	−0.31	0.38	40	73.18	22.82	40	4.46	1.15	41	7.02	3.55	41	7.55	3.94
Active group	18	−0.40	0.34	17	73.22	22.97	17	4.61	1.04	18	7.86	3.46	18	7.61	3.61
Delayed group	23	−0.25	0.40	23	73.15	23.23	23	4.35	1.23	23	6.37	3.56	23	7.50	4.27
Post
Total group	37	−0.23	0.38	36	78.00	23.22	36	4.31	1.16	36	6.58	3.18	36	7.78	3.80
Active group	16	−0.31	0.36	15	77.20	24.87	15	4.44	1.05	15	6.80	3.08	15	8.33	3.66
Delayed group	21	−0.17	0.40	21	78.57	22.58	21	4.22	1.25	21	6.43	3.33	21	7.38	3.94
Follow-up
Total group	40	−0.30	0.34	39	73.27	20.85	39	4.43	1.06	38	6.63	2.97	38	7.10	3.03
Active group	17	−0.29	0.28	16	73.80	17.70	16	4.45	0.92	15	6.91	2.59	15	7.89	3.23
Delayed group	23	−0.31	0.38	23	72.90	23.17	23	4.42	1.17	23	6.45	3.23	23	6.59	2.84

Table 4.

Descriptive outcomes of the delayed treatment group in the within-group analyses.

	Self-identity bias			Self-reported fatigue			Self-reported vitality			Self-reported all-or-nothing behavior			Self-reported avoidance behavior
	N	M	SD	N	M	SD	N	M	SD	N	M	SD	N	M	SD
Baseline	20	−0.27	0.42	20	73.45	19.26	20	4.24	1.13	20	6.40	3.13	20	7.36	2.62
Training	17	−0.34	0.33	17	69.65	23.04	17	4.30	1.28	17	6.29	3.22	17	5.97	2.52
Post	18	−0.33	0.41	18	70.67	26.71	18	4.36	1.35	18	6.78	4.37	18	6.39	2.33
Follow-up 1	19	−0.23	0.32	19	73.88	30.01	19	4.25	1.32	19	6.37	3.70	19	6.32	2.63
Follow-up 2	17	−0.25	0.28	17	66.90	27.93	17	4.50	1.29	17	5.63	3.05	17	6.34	3.48

Table 5.

Descriptive outcomes of the (Neo)adjuvant and metastasis participants at baseline.

Baseline	(Neo)adjuvant			Metastasis
	N	M	SD	N	M	SD
Self-identity bias (range = -2–2)	40	−0.26	0.38	9	−0.14	0.29
Self-reported fatigue (range = 20–140)	40	74.60	21.19	9	80.57	23.94
Self-reported vitality (range = 1–7)	40	4.51	1.02	9	3.96	1.20
Self-reported all-or-nothing behavior (range = 0–20)	40	7.32	3.54	9	7.20	2.44
Self-reported avoidance behavior (range = 0–32)	40	8.44	4.06	9	8.58	4.11

Time-series graphs

Self-reported fatigue and vitality. On vitality, in both analyses, participants showed very little change over time. Mean scores remained rounded to 4 in all phases in both analyses. On fatigue, in the between-groups analysis, as depicted in Figure 2, participants scored highest at the post measurement (M = 80.93, SD = 3.63) and lowest at baseline (M = 73.48, SD = 3.38) with the training measurement (M = 75.61, SD = 3.52) and follow-up (M = 75.79, SD = 3.57) in between. Thus, no training effect (and possibly an opposite effect) was found on fatigue. In the within-groups analysis, mean fatigue scores changed very little over time (staying between 70 and 74).

Figure 2.

Change of fatigue and all-or-nothing behavior over time in the active and control group.

Note. The means depicted are estimated marginal means. Error bars represent the 95% confidence interval.

Self-reported behavior. On all-or-nothing behavior, in the between-group analysis, as depicted in Figure 2, the control group means remained close to 7 in all phases over time. The training group showed a small decrease over time (baseline M = 8.57, SD = 0.76, training M = 8.22, SD = 0.77, post M = 7.68, SD = 0.79, follow-up M = 7.88, SD = 0.79). The within-group analysis showed a possibly delayed training effect. As depicted in Figure 3, after baseline (M = 6.40, SD = 0.84), participants scored highest during training (M = 7.08, SD = 0.86) and post (M = 6.97, SD = 0.85) and decreased during follow-up 1 (M = 6.31, SD = 0.85) and follow-up 2 (M = 5.78, SD = 0.86). On avoidance behavior in the between-group analysis, participants scored around 8 at all phases. In the within-group analysis, participants showed a small decrease over time (see Figure 3).

Figure 3.

Change of all-or-nothing and avoidance behavior over time in the delayed treatment group.

Note. The means depicted are estimated marginal means. Error bars represent the 95% confidence interval.

Effects of the CBM training on self-identity bias

In the between-group analysis, a significant time*group interaction effect was found on self-identity bias (F (3,116) = 2.89, p = .038). As can be seen in Figure 4, the training group changed from slight vitality bias at baseline (M = −0.16) to a moderate vitality bias in the training phase (M = −0.38), stabilizing at post (M = −0.35) and follow-up (M = −0.31). The delayed treatment control group showed a relatively stable slight vitality bias at all time points (baseline: M = −0.27, training: M = −0.23, post: M = −0.17, follow-up: M = −0.28). Between-group effect sizes were calculated per timepoint (Cohen’s d: baseline = 0.30, training = −0.31, post = −0.48, follow-up = −0.09) indicating a medium effect size at post and small effect sizes at all other timepoints. Notably, no significant training effect was found in the within-group analysis of the delayed treatment group (main effect: p = .33). As can be seen in Figure 5, participants did show a training effect with lower scores at training and post, but the change over time was not large enough to be significant.

Figure 4.

Interaction (time x group) effect on self-identity bias.

Note. The means depicted are estimated marginal means. Error bars represent the 95% confidence interval.

Figure 5.

Change of self-identity bias over time in the delayed treatment group.

Note. The means depicted are estimated marginal means. Error bars represent the 95% confidence interval. Effects were not significant.

Discussion

This waitlist-controlled study investigated the feasibility and effectiveness of a novel mobile CBM training to reduce fatigue symptoms in women receiving treatment for breast cancer in both the curative and palliative setting. Results on the feasibility were mixed: recruitment and retention did not meet the predetermined feasibility criteria, but feedback from participants, data completion, and demographic variability were judged positively. Training effects were mixed as well, but the predetermined medium effect size for improvement in bias was found.

As the predetermined criteria for recruitment (53% included instead of 80%) and retention (41% drop-out instead of 20%) were not met, recruitment and retention strategies should be re-evaluated before continuing with a full trial. As predicted, people in the metastasis phase were more difficult to recruit, especially for a study with this duration. This is in line with previous research in palliative care (Chaiviboontham, 2011; Ewing et al., 2004). For the full trial, the authors therefore recommend focusing solely on people in the (neo)adjuvant setting or shortening the study. Furthermore, as only two out of five hospitals included the intended number of (neo)adjuvant participants in the current study, hospital commitment should be improved.

Reasons for the limited inclusion in the three other hospitals, from the authors’ perspective, was the renouncement of the involved oncologist, recruiters thinking that fatigue symptoms were an inclusion criterium, the researcher being in contact with nurses instead of oncologists, and the co-occurrence of legal changes regarding eHealth devices in research, slowing down ethical approval in the participating hospitals. In the last months of recruitment, contact with the health care practitioners was increased with weekly updates to stimulate their involvement and sense of urgency. The recruitment period was also extended by 1 month, but these efforts were insufficient to reach the predetermined criteria. Study inclusion struggles are recognized by literature, as recruitment is described as one of the most challenging aspects of medical research with only 31% of clinical trials being able to reach the targeted number of participants within the designated time frame (Fitzer et al., 2022).

The completion rates of measurement and training sessions did not have a predetermined criterium, but the authors judged the results positively as most people only missed a few training- and/or measurement sessions. Considering the high number of measurement and training sessions, the length of the study, and not all participants receiving the notifications and reminders for training sessions, the authors expected low completion rates. Indeed, high attrition rates are common in CBM literature possibly because of boredom or doubts about effectiveness (Vrijsen et al., 2024), which is in line with evaluative CBM studies (e.g., Beard et al., 2012; Geerts et al., 2023).

Results regarding variability in demographics and outcomes were mixed. No floor or ceiling effects were found in any of the averaged measurements which supports the suitability of the measurements chosen for this study. However, average self-identity bias was already vitality oriented at baseline which may restrict potential room for improvement. This could also explain the non-significant result on the within-group analysis as it was conducted with the group that had the highest vitality bias at baseline.

These ceiling effects could be explained by the fact that most participants participated shortly after diagnosis: even though self-reported fatigue scores were on average close to problematic fatigue at baseline, it may take longer before an adverse cognitive bias is established. Furthermore, in the current study, although not based on statistical testing because of the small sample size, mean fatigue scores in the metastasis group were higher and above the cut-off for problematic fatigue than the (neo)adjuvant group that stayed under the cut-off. These differences confirm patients’ reports of a large increase in fatigue in the first 6 months after diagnosis (Schaab et al., 2023).

The increase in fatigue was also found in the time-series graphs of the self-reported outcomes. No training effects were apparent in the self-reported outcomes with fatigue worsening over time and vitality changing very little. These findings are in line with the authors’ previous research with breast cancer patients (Geerts et al., 2025). In the authors’ previous studies, there was also a lack of correlation between cognitive bias and the self-reported outcomes. However, in this study moderate to strong correlations between self-identity bias and fatigue, vitality, and avoidance behavior were observed (see Additional file 3). These correlational findings give more support for the underlying mechanism and theory behind CBM. Moreover, on all-or-nothing and avoidance behavior, there were some indications of small and delayed effects, but a full trial with larger sample sizes is needed to conclusively confirm effects on these outcomes.

One of the limitations of this study regarding the recruitment strategy is that the researcher was unaware of which setting (curative or palliative) the patient was in before inclusion. The authors noticed in the later stages of recruitment (when one hospital started to recruit more people in the metastasis setting) that more patients contacted by the researcher rejected participation. However, as the authors could not confirm the patients’ medical status at this point, it was not possible to provide exact recruitment, enrolment, and retention rates per patient group. These numbers could have been helpful for future research, especially focused on people in the metastasis setting. Furthermore, this lack of information caused some important deviations from protocol: it became impossible to conduct the study separately per patient group as initially planned (Geerts et al., 2024). Additionally, allocation was semi-randomized in the current study instead of the planned randomization (Geerts et al., 2024). The lack of patient information, as well as the recruitment issues experienced in this study, made it difficult to randomize allocation.

A future full trial should take these limitations into account, as well as the already mentioned considerations. Furthermore, in a full trial, it could be considered to screen people on their fatigue bias and only include people above a certain threshold. However, this could also limit the already challenging recruitment process. Future research should also further investigate the correlation between fatigue bias and fatigue, vitality, and behavioral outcomes so that more conclusive results can answer to the mixed results found in the authors’ studies.

This feasibility study is the first to investigate a CBM training targeting fatigue in breast cancer and gives valuable insights for continuing with a full trial on this rarely researched topic with huge need in patients due to the impact and burden of CRF. Before continuation, the future trial should take the suggestions described in this paper into account, especially considering recruitment and retention. Besides feasibility recommendations, this study shows that the novel CBM training is successful changing self-identity bias compared to a control group.

Appendix A: Stimuli used in the IVY training and the IAT measurement: original Dutch and English translations

Table A1:

Stimuli used in the training and measurements. Original Dutch and English translations.

Target concepts
	Dutch	English	Dutch	English
N	Vitaal	Vital	Moe	Tired
1	Energiek	Energetic	Uitgeput	Exhausted
2	Levenslustig	Full of life	Slap	Weak
3	Fit	Fit	Lusteloos	Apathetic
4	Wakker	Awake	Traag	Slow
5	Actief	Active	Duf	Dull
6	Alert	Alert	Slaperig	Sleepy
7	Krachtig	Powerful	Krachteloos	Powerless
8	Sterk	Strong	Vermoeid	Fatigued
9	Snel	Quick	Futloos	Lifeless
10	Vitaal	Vital	Moe	Tired

ORCID iDs

Jody Geerts https://orcid.org/0000-0002-8826-032X

Marcel Pieterse https://orcid.org/0000-0002-6900-3088

Ester Siemerink https://orcid.org/0000-0001-6946-7651

Peter ten Klooster https://orcid.org/0000-0002-2565-5439

Marleen Wensink https://orcid.org/0000-0001-5306-097X

Ethical considerations

According to the committee of human research [commissie mensgebonden onderzoek, cmo] this study was not applicable (File number: 2021-13261) to the Dutch Medical Research Involving Human Subjects Act and redirected the ethical approval to local ethical committees. Both the ethical committees of the faculty of behavioural, management and Social Sciences of University of Twente (File number 220004) and Local Advice Committee of Hospital Group Twente [ziekenhuisgroep Twente, ZGT] (case number: ZGT22-09) approved this study.

Consent to participate

If the patient agrees to participate, an informed consent form with patient’s name, phone number and email address is signed by the participant and the researcher in duplicate, one of which is returned to the patient, and the other is archived. The patient can contact the researcher, the independent person, or the primary practitioner at any time for questions.

Consent for publication

The consent form informed participants on which and how data would be used, stored and published. This was done conform dutch privacy laws (algemene verordering gegevensbescherming (avg)). No published data is directly retraceable.

Data Availability Statement

The data generated during and/or analysed during the current study are not publicly available due to privacy reasons but are available from the corresponding author on reasonable request. Ethics approval, participant permissions, and all other relevant approvals were granted for this data sharing.*