Videoconference-delivered group Cognitive Behavioral Stress Management for ME/CFS patients who present with severe PEM: A randomized controlled trial

Abstract

Background:

In Myalgic Encephalomyelitis/Chronic Fatigue Syndrome (ME/CFS), post-exertional malaise (PEM) is associated with greater distress and symptoms. Cognitive Behavioral Stress Management (CBSM) has demonstrated beneficial effects for ME/CFS and may mitigate stress-related triggers of PEM. We tested a virtual CBSM intervention to increase access, and we report on its effects on stress and symptoms in ME/CFS patients with severe PEM.

Methods:

Data were from a randomized controlled trial (NCT01650636) comparing 10-week videoconference-delivered group CBSM (V-CBSM, n=75) to a 10-week Health Information active control (V-HI, n=75) in Fukuda criteria ME/CFS patients (71 classified as highPEM, 79 lowPEM). Linear regression explored PEM-by-Treatment interactions on overall symptom frequency and intensity, perceived stress, and fatigue-specific interference and intensity, at 5-month follow-up. Logistic regression tested V-CBSM effects on 5-month PEM status. Analyses controlled for age, gender, race/ethnicity, mode of symptom onset, and time since diagnosis.

Results:

The sample was middle-aged (47.96±10.89 years), mostly women (87%) and non-Hispanic White (65%), with no group differences on these variables or baseline PEM. For highPEM patients, V-CBSM (versus V-HI) demonstrated medium to large effects on follow-up symptom frequency, symptom intensity, fatigue interference, and fatigue intensity (p’s < .05) and trending to significant reductions in perceived stress (p =.07). Differences were not evident for lowPEM patients. Treatment predicted follow-up PEM status at a trend (p = .058), with patients receiving V-CBSM demonstrating half the risk of highPEM classification versus V-HI.

Conclusions:

V-CBSM demonstrates benefits for ME/CFS patients presenting with severe PEM and may reduce the expression of PEM over time.

Introduction

Various psychological intervention modalities have been evaluated for treating Myalgic Encephalomyelitis/Chronic Fatigue Syndrome (ME/CFS), a debilitating disorder defined by profound fatigue that is not alleviated by rest [1]. Although cognitive behavioral therapies (CBT’s) have shown promise [e.g., 2,3] no gold standard treatment has been identified [1,4]. In ME/CFS intervention research, variability in CBT-based approaches and ME/CFS symptomatology challenges interpretation of results.

Heterogeneity in CBT-based Approaches to ME/CFS

Cognitive restructuring (i.e., modifying unhelpful thoughts) and behavioral activation (i.e., engaging in pleasant activities) are commonly utilized CBT techniques; however, the manner of their application can vary widely. For instance in the PACE trial [5], the largest clinical trial on therapies for ME/CFS [6], CBT was explicitly based on a fear avoidance theory and designed to increase activity levels by promoting physical/mental engagement and restructuring related thoughts [5]. In contrast, cognitive behavioral stress management (CBSM) is based on a neuroimmune model of ME/CFS [7] and aims to increase stress management skills to improve quality of life and functioning [e.g., 8,9], with cognitive restructuring broadly applied to stressors rather than with the specific intent of increasing activity levels.

Studies included in the previously cited systematic review of CFS interventions [2] also varied in their CBT approach. Four appeared to use a fear avoidance model [10–13] whereas one took a strengths-based approach that emphasized tolerance of illness limitations [14]. Another occupied a middle ground of addressing negative fatigue-related beliefs and promoting consistent, realistic activity levels [15]. Despite distinctions, all were classified as CBT [2] obfuscating effects specific to any particular approach.

Heterogeneity of Post-Exertional Malaise in ME/CFS presentation

Variations in case definitions for ME/CFS yield a heterogeneous patient group, and thus subgroups may also respond differently to treatment. Post-exertional malaise (PEM), for example, is an exacerbation of symptom burden following mental or physical exertion [16]. However, it is not a necessary symptom for diagnosis according to the Fukuda criteria for CFS [17] which are polythetic and require the presence of 4 of 8 symptoms (e.g., PEM lasting more than 24 hours, unrefreshing sleep) in addition to fatigue. The Fukuda criteria were at one time frequently used in research and practice [4] but are now applied less often as they do not reflect current understanding of ME/CFS. In 2015, the National Academy of Sciences published an updated case definition for ME/CFS which does require PEM [1] which was then adopted by the Centers for Disease Control and Prevention. More recently, PEM has been identified as among the three most frequently reported symptoms of ME/CFS along with sleep dysfunction and cognitive dysfunction [18] and these are seen as cardinal features of the disorder [1]. Further, the phenomenology of PEM is now more flexibly defined in order to recognize PEM that may be more brief or delayed [19].

Patients diagnosed with CFS who experience high levels of PEM (highPEM) report greater symptom burden as well as greater social disruption, depressive symptoms, and mood disturbance than those experiencing low levels (lowPEM) [20,21]. Associations between PEM and more severe ME/CFS presentation (e.g., being home-bound/bed-ridden) [22] and worse ME/CFS prognosis [23] have also been demonstrated.

PEM is a uniquely distressing symptom of ME/CFS, with higher levels of distress potentially reflecting the challenges of managing a more severe disease. Therefore, PEM may be used to characterize a vulnerable CFS patient subgroup that demonstrates greater need for psychological intervention. Some have advocated for more specific clinical phenotyping of the ME/CFS patient group and consideration of confounding variables, including mode of symptom onset and illness duration, in intervention research [24]. Others suggest that treatment studies should be targeted specifically on PEM due to the unique phenomenological presentation of patients who experience it [25].

CBSM and PEM

The neuroimmune model of ME/CFS holds that immunological/hormonal dysregulation is aggravated by stress, with symptoms prompting distress reactions and cyclically intensifying future symptoms [7,26]. CBSM draws from this model and seeks to interrupt the cycle by increasing adaptive coping skills (e.g., relaxation, coping effectiveness, interpersonal skills) and reducing distress in a supportive group environment. The intervention has demonstrated beneficial effects on health outcomes and neuroimmune processes in several chronic illnesses, including HIV and cancer [e.g., 7,27,28–30].

For patients with CFS, previous trials have evaluated live, face-to-face (L-CBSM) and telephone-delivered (T-CBSM; audio only) CBSM modalities [8,9]. L-CBSM demonstrated large pre-post effects on symptom intensity, symptom frequency, and perceived stress, whereas T-CBSM demonstrated medium effects on perceived stress only [8]. In the present study, we consider data from a third trial of videoconference-delivered CBSM (V-CBSM) for which results reported on the NIH funding database indicated no overall pre-post intervention effects on change in symptom intensity, symptom frequency, or perceived stress (see ClinicalTrials.gov, NCT01650636). We re-analyze these data from the perspective of a PEM subgroup analysis.

All CBSM modalities comprise the same CBSM intervention content, which was delivered across 12 face-to-face group sessions in L-CBSM and across 10 virtual group sessions in V-CBSM. The success of L-CBSM suggests that the CBSM intervention may be used to effectively intervene on the symptom experience of patients with CFS. Indeed, perceived stress management skills (PSMS), which are targeted by CBSM, are associated with reduced fatigue-related neuroendocrine dysregulation in CFS patients [31], and CBSM has been shown to increase PSMS in tandem with decreases in neuroendocrine dysfunction and inflammatory signaling in other populations [32,33]. Differences in effects may partly be due to differences in the control conditions of CBSM trials (i.e., health-related knowledge delivered via a single half-day psychoeducational seminar in the L-CBSM trial, while this control condition was delivered over 10 weekly sessions in T-CBSM and V-CBSM). Since the dosage of the health-related knowledge control condition was greater in T-CBSM and V-CBSM versus L-CBSM, there would have been a greater burden of proof for the CBSM intervention in those trials.

If the CBSM intervention is efficacious, then its virtual modalities deserve particular attention for the population of ME/CFS patients. Travel is often physically and mentally demanding, and this is especially relevant to individuals with fatigue-related concerns. Further, some ME/CFS patients are homebound or bedridden [22,34] and thus entirely unable to access the benefits of in-person treatment. Virtual interventions broaden treatment reach to patients across geographic areas, regardless of access to transportation, and without imposing demands associated with travel.

Recruitment for the V-CBSM trial, which took place prior to the National Academy of Sciences publication [1], utilized the Fukuda criteria for CFS to determine eligibility. Therefore, although PEM is now considered a cardinal feature of ME/CFS [1,18] some of the V-CBSM trial participants did not report notable experience of PEM. Given that PEM has been linked to higher levels of circulating pro-inflammatory cytokines [35] the neuroimmune approach of CBSM may be particularly relevant to patients with ME/CFS and those diagnosed with CFS [17] who experience PEM. Further, PEM is likely to inhibit patients from engaging in physically/mentally exerting tasks, thereby compromising quality of life and resulting in feelings of low agency and hopelessness; stress management skills could intervene on these latter domains. Lastly, previous research suggests that emotional experiences may trigger PEM [25], and as such, emotion regulation skills (e.g., relaxation, coping effectiveness) comprise a large component of the CBSM protocol. Given the importance of PEM in more updated conceptualizations of ME/CFS [1] and the potential fit between the experience of PEM and a CBSM intervention, we sought to re-analyze V-CBSM data separately for patients with and without notable experience of PEM.

Objective

Here, we considered whether a 10-week V-CBSM intervention reduced symptom burden and perceived stress for patients diagnosed with CFS as a function of PEM status [25]. We hypothesized that highPEM patients who received V-CBSM would demonstrate reductions in symptom intensity, symptom frequency, perceived stress, fatigue interference, and fatigue intensity compared to those in the active, time-matched control condition. Since increases in PSMS have been related to decreases in the expression of PEM in cross-sectional analyses [36] we additionally explored whether CBSM affects changes in PEM status over time.

Methods

Sampling and Procedures

We conducted a secondary analysis of a randomized controlled trial of group videoconference-delivered cognitive behavioral stress management (V-CBSM) for CFS patients, funded by the National Institute of Neurological Disorders and Stroke (R01NS072599) and registered under ClinicalTrials.gov (NCT01650636). The study was registered with the Institutional Review Board of the University of Miami and all participants provided informed consent for participation. Participants were identified via physician referral, advertisements, and community support organizations, with the following inclusion criteria: fluency in English, age 21 to 75 years, physician-determined Fukuda criteria [17] CFS diagnosis, partnered (living together/separately). Patients with comorbid conditions (e.g., rheumatoid arthritis) and those taking medications that might influence ME/CFS-associated biological processes (e.g., corticosteroids) were excluded. Due to the psychoeducational aspects of V-CBSM, cognitive impairment in terms of >3 errors on the Short Portable Mental Status Questionnaire [37] was an additional exclusion criterion, and individuals with severe psychiatric conditions (e.g., schizophrenia) were excluded [38] in consideration of group processes.

Participants independently completed assessments via an online platform or hard-copy versions. Following baseline assessment, dyads were randomly assigned to 10 weeks of V-CBSM or the time-matched health information active control condition (V-HI). Follow-up assessments took place at 5- and 9-months post-baseline, and this study focuses on analyses of 5-month follow-up data. Participants were compensated $50 per assessment.

Interventions.

Both V-CBSM and V-HI consisted of weekly, videoconference sessions, each of which was approximately 75m in duration. In V-CBSM, cohorts included ≤6 patient-partner dyads, and each session included two components. The CBT component covered the importance of stressor appraisals in the biological stress response, cognitive distortions and cognitive restructuring, adaptive coping, quality of life, social support, anger management, assertive communication, and acceptance. The relaxation component included practice and discussion of deep breathing, progressive muscle relaxation, special place imagery, and mindfulness meditation.

In V-HI, single-dyad sessions targeted health behaviors and health-related knowledge (e.g., diet, sleep hygiene, patient-physician communication). V-HI was designed as an active control condition to account for common factors (e.g., facilitator attention) and did not provide stress management skills training. Participants in both conditions were afforded access to an online library of ME/CFS resources and expert and session summary videos specific to either V-CBSM or V-HI; in V-CBSM, relaxation exercise demonstration videos were also included.

Predictors/Outcomes

Patient characteristics.

At baseline, participants reported their age, gender, race/ethnicity, employment/disability status, income, education, height, weight, and whether they were cohabitating with their partner. Patients specified the nature of their ME/CFS symptom onset (gradual, sudden) and date of diagnosis.

At baseline and 5-month follow-up, participants indicated whether each of 16 stressful life events from the Life Experiences Survey [39] had occurred within the past 90 days. The number of items with a negative impact served as a proxy for recent negative life experiences.

PEM status, symptom frequency, symptom intensity.

The CDC CFS Symptom Inventory [40] assessed CFS-related symptom intensity and frequency over the previous month, with 19 symptoms (e.g., nausea, chills) based on the Fukuda criteria [17]. A modified 20-item version of the scale, which has been published on previously [20,41] was administered in the trial. The item regarding “unusual fatigue following exertion that lasts for at least 24 hours” served as an index of PEM status and was removed from symptom frequency and intensity score calculations. To align with a clinically useful perspective of PEM status (i.e., presence vs. absence of the symptom, as in case definitions) we dichotomized the PEM intensity variable into highPEM (Severe, Very severe) and lowPEM (Very mild, Mild, Moderate, not endorsed) categories [see 20 for additional rationale for this classification]. Remaining item scores (rated 0–5) were summed, with higher scores indicating greater intensity or frequency (range: 0–95; α_Baseline=.835, α_5-month=.887) or intensity (range: 0–95; α_Baseline=.803, α_5-month=.877).

Perceived stress.

The Perceived Stress Scale (PSS) assessed perceived stress, with higher scores indicating that participants appraised situations in their lives as stressful to a greater degree [42]. Items ask about the frequency of stressful experiences in the last 30 days (e.g., “felt difficulties were piling up so high that you could not overcome them”). Items are rated on a 5-point scale (0=Never; 1=Almost never; 2=Sometimes; 3=Fairly often; 4=Very often) and a scale score is derived by summing the ratings of the 14 items (range: 0–56; α_Baseline=.887, α_5-month=.917).

Fatigue interference and fatigue intensity.

The Fatigue Symptom Inventory (FSI) assessed fatigue-related illness interference and intensity [43]. Interference was assessed with 7 items asking participants to indicate how much fatigue interfered with functioning (e.g., “ability to concentrate”) over the previous week (rated 0–10) which were summed (range: 0–70; α_Baseline = .897, α_5-month = .941). Intensity was assessed with 4 items rating the level of fatigue (rated 0–10) at various points during the previous week; the average of these scores served as a measure of fatigue intensity (range: 0–10; α_Baseline=.894, α_5-month=.899).

Perceived stress management skills.

The Measure of Current Status [MOCS; 44] assessed self-perceived stress management skills (PSMS), with higher scores indicating greater skills. Participants rated their ability to perform each of 17 items related to stress management, including use of social support, coping and cognitive restructuring, tension awareness, and relaxation. Responses were summed in a unifactorial approach [e.g., 28,36] with higher scores on items (rated 0–4) indicating greater perceived ability (range: 0–68; α_Baseline=.881, α_5-month=.900)

Perceived benefit.

A project evaluation questionnaire was administered, following other measures, during the follow-up assessment. Items were designed to solicit participant feedback and inform future studies and included 6 questions regarding perceived benefit of treatment. Patients were asked to report on whether the intervention increased their stress management skills, self-efficacy, quality of life, and understanding of ME/CFS, and whether they found participation to be beneficial and valuable. Higher scores on items (rated 0–2) indicated greater benefit, and scores were summed (range: 0–12, α_5-month=.911).

Statistical Analyses

Analyses were performed in RStudio [45] for R statistical software [46] version 4.0.1. For variables assumed missing at random, missing baseline values were imputed using the mice package [47] and predictive mean matching with 10 iterations for 5 models, and these values were included in the same procedure for follow-up data. Treatment group differences on sociodemographic and clinical characteristics were assessed with Fisher’s exact tests for categorical variables and two-sided independent samples t-tests for continuous variables (α = .05). Linear regression analyses were conducted for manipulation check variables (perceived benefit, perceived stress management skills [PSMS]) and primary outcomes (symptom frequency, symptom intensity, perceived stress, fatigue interference, fatigue intensity), and logistic regression analyses were conducted for predicting follow-up PEM classification.

Regression models.

Baseline scores on outcomes (except for perceived benefit), age (mean-centered), gender (female, male), race/ethnicity (non-Hispanic White, Hispanic/Caribbean Islander/African American/Korean), months since diagnosis (at group start date; mean-centered), and mode of symptom onset (gradual, sudden) were entered as covariates in all regression models. In logistic regression models, neither age nor months since diagnosis variables were transformed, and an integer of 1 was added to the months since diagnosis variable. Dichotomous variables were dummy-coded, with index values (i.e., 1) referencing the following: female, non-Hispanic White, gradual. Standardized regression coefficients (β) were obtained by running models with standardized variables ($\frac{x-\stackrel{-}{x}}{SD}$).

Manipulation checks.

For models of manipulation check variables, we additionally included treatment group (V-CBSM, V-HI; index: V-CBSM) as a predictor. We analyzed whether V-CBSM had improved PSMS at follow-up to a greater degree than V-HI and whether perceived benefit scores were higher in V-CBSM versus V-HI at 5 months.

Primary analyses.

For primary outcomes, we additionally included PEM (highPEM, lowPEM; index: highPEM), treatment group, and a PEM-by-Treatment interaction term in each model. Pairwise comparisons and associated effect sizes (Cohen’s d), and estimated marginal means (i.e., simple effects of treatment) and associated p-values [adjusted for Type I error, see 48] were obtained using the emmeans package for R [49]. Estimates correspond to average baseline values on outcomes and averaging across all other variables. To provide a more interpretable measure of effect size, we additionally calculated overall percent change in primary outcomes for highPEM patients in V-CBSM and highPEM patients in V-HI, as well as the percent of highPEM patients who improved on each of these variables.

Exploratory logistic regression.

Chi-squared tests evaluated treatment group differences in PEM classification at baseline and follow-up. The logistic regression model included baseline PEM status, the covariates described above, and treatment group. We assessed variance explained by the model, and its predictive ability, with pseudo-R² [50] and a confusion matrix.

Assumptions.

In linear models, normality of residuals was ascertained with both Kolmogorov-Smirnov and Anderson-Darling tests (p > .05), as well as consideration of Q-Q plots. Linearity and homoscedasticity were assessed with plots of residuals against fitted values. We additionally applied the gvlma package [51] to ensure assumptions were met. In linear and logistic models, we confirmed absence of multicollinearity (variance inflation factor values ≤ 3 for interactions, ≤ 2.5 otherwise). Studentized residuals (>|3| SD) were used to identify outliers. We compared results of models with outliers retained versus omitted. In logistic regression analyses, the Box-Tidwell test [52] confirmed linear relationships between continuous predictors and the logit of follow-up PEM classification. Across analyses, p-values between .05 and .10 were considered potential trends and we interpreted the magnitude of such effects accordingly.

Results

The perceived benefit manipulation check variable demonstrated a greater proportion of missing data (34.0%) than other variables. These items were administered following other measures, within the project evaluation questionnaire, and missing values were not imputed. Missing values on other predictors and outcomes (8%) were assumed missing at random and imputed. Participants in V-CBSM versus V-HI did not differ on other sociodemographic and clinical characteristics (see Table 1). Descriptive statistics on outcomes are presented in Table 2. Assumptions were met as described unless otherwise noted. Analyses controlled for baseline scores on outcomes (excepting perceived benefit), age, gender, race/ethnicity, mode of symptom onset, and months since diagnosis.

Table 1.

Sociodemographic and Clinical Characteristics by Treatment Group

Sociodemographic and Clinical Characteristics		n V-CBSM, V-HI	V-CBSM $\overline{x}\pm s$, range	V-HI $\overline{x}\pm s$, range	p
Age		75, 75	47.8±11.3, 23–72	47.9±10.5, 20–72	.964
Gender (% female)		75, 75	89.3	85.3	.624
Race/Ethnicity (%)	non-Hispanic White	75, 75	68	62.7	.163
	Hispanic		29	32
	Caribbean Islander		0	4
	African American		2.7	0
	Korean		0	1.3
Educationa (%)	Some high school	74, 74	2.7	1.4	.468
	High school graduate		1.4	5.4
	GED		0	5.4
	Trade school		4.1	4.1
	Some college		27	21.6
	College graduate		40.5	33.8
	Graduate degree		24.3	28.4
Employment statusb (%)	Employed	60, 68	35	39.7	.479
	On Disability		35	25
	Unemployed		16.7	25
	Other		13.3	10.3
Annual incomea (%)	≤ $5000	63, 64	7.9	4.7	.208
	$5,001 - $10,000		1.6	0
	$10,001 - $20,000		6.3	3.1
	$20,001 - $30,000		6.3	4.7
	$30,001 - $40,000		6.3	6.3
	$40,001 - $50,000		7.9	7.8
	$50,001 - $60,000		4.8	12.5
	> $60,000		58.7	60.9
Cohabitating with partner (% yes)		73, 73	96.0	93.3	.620
Recent negative life experiences, baseline		68, 73	1.04±1.38, 0–6	1.34±1.60, 0–6	.239
Recent negative life experiences, follow-up		61, 58	1.02±1.12, 0–5	1.24±1.58, 0–7	.370
BMI		72, 74	27.4±6.7, 17.5–50.3	28.1±6.9, 16.8–56.7	.533
Months since diagnosisc		75, 75	77.3±95.9, 0–477	66.5±85.4, 0–344	.469
Mode of symptom onsetc (% gradual)		75, 75	65.3	69.3	.728

Note. p-values (α = .05) correspond to two-sided independent samples t-tests for continuous variables and Fisher's exact test for categorical variables. Variables were assessed at baseline unless otherwise noted. V-CBSM = Videoconferenced Cognitive Behavioral Stress Management. V-HI = Videoconferenced Health Information, active control condition.

^aEducation and annual income were treated as continuous variables

^bFor employment status, 'Employed' includes part-time and full-time and 'Other' includes student, retired, volunteer worker, and other

^cMissing values were imputed for months since diagnosis, symptom onset, and PEM status.

Table 2.

Descriptive Statistics on Outcome Variables by Treatment Group at Baseline and Follow-up

Variable	Possible Range	Time	$\overline{x}\pm s$, range
			V-CBSM (n = 75)	V-HI (n = 75)
Perceived Stress Management Skills	(0–68)	BL	38.80±10.55, 13–61	36.44±10.91, 13–64
		FU	42.84±11.46, 11–66	37.99±10.79, 18–64
Study Benefit	(0–12)	FU	8.32±3.17, 1–12	6.09±3.43, 0–12
Symptom Intensity	(0–95)	BL	33.00±11.42, 11–66	37.35±15.32, 11–77
		FU	31.43±13.68, 1–70	37.97±16.95, 2–83
Symptom Frequency	(0–95)	BL	34.89±12.77, 13–78	39.56±15.42, 10–87
		FU	31.73±14.26, 1–77	37.84±17.13, 2–85
Perceived Stress	(0–56)	BL	28.64±9.33, 7–50	32.20±9.28, 11–52
		FU	25.28±10.11, 2–56	29.28±10.12, 6–53
Fatigue Interference	(0–70)	BL	42.15±17.24, 7–70	47.35±16.21, 1–70
		FU	36.33±18.77, 1–70	44.24±17.52, 0–70
Fatigue Intensity	(0–10)	BL	5.77±2.02, 1–9	6.59±1.69, 2–10
		FU	5.57±2.13, 1–10	6.30±2.09, 0–10
PEM (n, % highPEM)	highPEM / lowPEM	BL	32, 42.7%	39, 52.0 %
		FU	29, 38.7%	42, 56.0%

Note. V-CBSM = Videoconferenced Cognitive Behavioral Stress Management.V-HI = Videoconferenced Health Information, active control condition. BL = baseline. FU = 5-month follow-up.

Manipulation Checks

The model for perceived stress management skills (PSMS) was significant (F(7, 142) = 25.500, p < .001, R² = .557). Patients who received V-CBSM demonstrated increases in PSMS compared to those who received V-HI (β = 0.276; B = 3.131, SE = 1.278, p = .016). Excepting baseline PSMS, no other significant associations were evident (p’s ≥ .207).

Treatment group differences on perceived benefit were significant in a subsample of 99 patients who provided data (F(6, 92) = 5.494, p < .001; R² = .216). V-CBSM was associated with greater perceived benefit at follow-up relative to V-HI (β = 0.726; B = 2.430, SE = 0.613, p < .001). Non-Hispanic White participants reported less perceived benefit than others across groups (β = −0.713; B = −2.385, SE = 0.736, p = .002). No other variables demonstrated significant associations (p’s ≥ .394).

Analyses of Primary Outcomes

Parameter estimates for models of symptom frequency, symptom intensity, perceived stress, fatigue interference, and fatigue intensity are presented in Table 3. For highPEM patients, V-CBSM was associated with reductions on all variables excepting perceived stress (p’s ≤ .041); no treatment group differences were evident for lowPEM patients (see Table 4). We therefore focus on simple treatment effects and pairwise comparisons for PEM status groups.

Table 3.

Parameter Estimates for Symptom Frequency, Symptom Intensity, Perceived Stress, Fatigue Interference, and Fatigue Intensity

Model Fit	Symptom Frequency				Symptom Intensity				Perceived Stress				Fatigue Interference				Fatigue Intensity
	F(9,140) = 33.720, p < .001				F(9, 140) = 26.080, p < .001				F(9,140) = 16.412, p < .001				F(9, 140) = 8.042, p < .001				F(9,140) = 6.268, p < .001
	R2 = .684				R2 = .626				R2 = .513				R2 = .341				R2 = .287
Parameters	β	B	SE	p	β	B	SE	p	β	B	SE	p	β	B	SE	p	β	B	SE	p
Intercept	−0.397	−5.166	4.054	.205	−0.343	−2.342	4.32	.589	−0.759	−2.077	3.440	.547	−0.941	1.649	6.524	.801	−0.851	1.246	0.817	.130
Baseline y	0.807	0.903	0.059	<.001***	0.782	0.900	0.069	<.001***	0.651	0.709	0.068	<.001***	0.432	0.474	0.081	<.001***	0.415	0.465	0.090	<.001****
Agea	−0.041	−0.060	0.077	.438	−0.050	−0.072	0.083	.386	−0.011	−0.011	0.063	.865	−0.027	−0.046	0.129	.722	−0.100	−0.020	0.015	.209
Genderb	0.208	3.323	2.420	.172	0.118	1.854	2.576	.473	0.294	3.021	1.922	.118	0.513	9.504	4.033	.020*	0.333	0.710	0.485	.146
Race/Ethnicityb	0.219	3.505	1.818	.056	0.286	4.490	1.942	.022*	0.376	3.861	1.431	.008**	0.419	7.761	3.004	.011*	0.317	0.676	0.359	.062
Symp. Onsetb	0.034	0.540	1.734	.756	0.155	2.426	1.855	.193	0.177	1.82	1.383	.191	0.106	1.967	2.904	.499	0.380	0.810	0.348	.021*
Months Diag.	−0.022	−0.004	0.009	.681	−0.021	−0.004	0.010	.721	−0.071	−0.008	0.008	.291	−0.012	−0.003	0.016	.874	−0.004	0.000	0.002	.957
PEMb	0.224	3.584	2.304	.122	0.074	1.154	2.529	.649	0.418	4.302	1.740	.015*	0.601	11.142	3.710	.003**	0.359	0.767	0.455	.095
Treatmentb	0.109	1.752	2.110	.408	−0.018	−0.283	2.251	.900	0.030	0.310	1.691	.855	0.188	3.491	3.561	.328	0.080	0.170	0.427	.691
PEM | Treatment	−0.522	−8.350	3.133	.009**	−0.361	−5.661	3.360	.094	−0.354	−3.639	2.491	.146	−1.085	−20.104	5.228	<.001***	−0.525	−1.120	0.626	.076

Note. fit statistics, standard errors (SE), and p values (i.e., of the intercept) correspond to models of unstandardized variables (B). PEM = post-exertional malaise. V-CBSM = Cognitive Behavioral Stress Management. Symp. Onset = mode of symptom onset. Months Diag. = months since diagnosis at baseline.

^aAge and months since diagnosis were mean-centered in models of unstandardized variables and standardized in models of standardized variables.

^bGender, race/ethnicity, mode of symptom onset, treatment, and PEM are dummy-coded dichotomous variables; index values (i.e., 1) are: female, non-Hispanic White, gradual, high-PEM, V-CBSM

^*p < .050

^**p ≤ .010

^***p ≤ .001

Table 4.

PEM by Treatment Interactions on Symptom Frequency, Symptom Intensity, Perceived Stress, Fatigue Interference, and Fatigue Intensity

Outcome	PEM Status	Condition	Simple Effects						Pairwise Comparisons
			β	B	SE	t(140)	p		β	B	SE	t(140)	p
Symptom Frequency	lowPEM	V-HI	−0.036	−0.580	1.375	−0.422	.674		0.109	1.752	2.110	0.830	.408
	lowPEM	V-CBSM	0.073	1.172	1.322	0.886	.503
	highPEM	V-HI	0.188	3.003	1.401	2.144	.068		−0.412	−6.598	2.292	−2.879	.005**
	highPEM	V-CBSM	−0.225	−3.595	1.441	−2.494	.055
Symptom Intensity	lowPEM	V-HI	0.062	0.980	1.489	0.658	.624		−0.018	−0.283	2.251	−0.126	.900
	lowPEM	V-CBSM	0.044	0.697	1.418	0.491	.624
	highPEM	V-HI	0.136	2.134	1.532	1.393	.332		−0.379	−5.944	2.459	−2.417	.017*
	highPEM	V-CBSM	−0.243	−3.810	1.540	−2.474	.058
Perceived Stress	lowPEM	V-HI	−0.136	−1.396	1.075	−1.298	.389		0.030	0.310	1.691	0.183	.855
	lowPEM	V-CBSM	−0.106	−1.086	1.026	−1.058	.389
	highPEM	V-HI	0.283	2.906	1.064	2.731	.029*		−0.324	−3.329	1.828	−1.821	.071
	highPEM	V-CBSM	−0.041	−0.424	1.147	−0.369	.713
Fatigue Interference	lowPEM	V-HI	−0.124	−2.291	2.260	−1.013	.417		0.188	3.491	3.561	0.981	.329
	lowPEM	V-CBSM	0.065	1.201	2.234	0.537	.592
	highPEM	V-HI	0.478	8.851	2.266	3.907	.001***		−0.897	−16.613	3.800	−4.372	.001***
	highPEM	V-CBSM	−0.419	−7.761	2.418	−3.210	.003**
Fatigue Intensity	lowPEM	V-HI	−0.088	−0.188	0.272	−0.693	.653		0.080	0.170	0.427	0.399	.691
	lowPEM	V-CBSM	−0.008	−0.018	0.272	−0.067	.947
	highPEM	V-HI	0.271	0.578	0.282	2.053	.168		−0.445	−0.950	0.460	−2.065	.041*
	highPEM	V-CBSM	−0.174	−0.372	0.289	−1.288	.400

Note. Simple effects are predictions at the average level of continuous variables in the model (baseline outcome values, age, months since diagnosis), and averaged over levels categorical variables (gender, race/ethnicity, and symptom onset). Standard errors (SE) and p-values correspond to models of unstandardized variables; the false discovery rate adjustment for p−values was applied to simple effects. PEM = post-exertional malaise. V-CBSM = Videoconferenced Cognitive Behavioral Stress Management. V-HI = Videoconferenced Health Information, active control condition.

^*p < .050

^**p ≤ .010

^***p ≤ .001

Symptom frequency.

Simple effects of treatment on symptom frequency were nonsignificant, although highPEM patients in V-CBSM reported reductions in symptom frequency at a trend (β = −0.225, p = .055). For highPEM patients, V-CBSM was associated with medium-to-large reductions in overall symptom frequency as compared to V-HI (β = −0.412, p = .005; d = 0.711, 95%CI[0.216, 1.207]); see Table 4). On average, highPEM patients in the V-CBSM group demonstrated a 14% reduction in symptom frequency scores between baseline and follow-up, and 75% of these patients improved on this variable. HighPEM patients in the V-HI control group demonstrated a mean 2% decrease in symptom frequency between baseline and follow-up.

Symptom intensity.

HighPEM patients who received V-CBSM also reported reductions in symptom intensity at a trend (β = −0.243, p = .058). Pairwise comparisons revealed medium effect size reductions in symptom intensity for highPEM patients who received V-CBSM compared to those who received V-HI (β = −0.379, p = .017; d = 0.600, 95%CI[0.104, 1.097]; see Table 4). On average, highPEM patients in the V-CBSM group demonstrated a 9% reduction in symptom intensity scores between baseline and follow-up, and 53% of these patients improved on this variable. HighPEM patients in V-HI demonstrated a mean 3% increase in symptom intensity between baseline and follow-up.

Perceived stress.

One outlier was present in the model for perceived stress (studentized residual = 4.138). Results for models with this case omitted versus retained were generally equivalent in terms of parameter significance and magnitude, as well as simple effects of treatment, pairwise comparisons, and associated effect sizes. We therefore present the full sample model in Tables 3 and 4. In terms of simple effects, highPEM patients who received V-HI endorsed significantly greater perceived stress scores than others at follow-up (β = 0.283, p = .029; see Table 5). Pairwise comparisons of intervention group for highPEM patients trended towards significance (β = −0.324, p = .071; see Table 4) and indicated a medium effect of CBSM (d = 0.450, 95%CI[−0.041, 0.941]). On average, highPEM patients in the V-CBSM group demonstrated a 6% reduction in perceived stress scores between baseline and follow-up, and 53% of these patients improved on this variable. HighPEM patients in V-HI on average reported a 3% decrease in perceived stress between baseline and follow-up.

Table 5.

Logistic regression model of intervention effects on follow-up PEM status

Model Fit			Χ2(7) = 33.736, p < .001
			AIC = 189.780, pseudo-R2 = .269
Parameters	log Odds	SE	95% CI	Wald Χ2	Odds Ratio	p
Intercept	−0.373	0.741	−1.818, 1.115	−0.503	0.688	.153
Baseline PEM	1.540	0.373	0.823, 2.289	4.131	4.664	<.001***
Age	−0.023	0.019	−0.060, 0.013	−1.242	0.977	.214
Gendera	−0.993	0.595	−2.220, 0.144	−1.668	0.371	.095
Race/Ethnicitya	0.757	0.436	−0.090, 1.627	1.738	2.132	.082
Symp. Onseta	0.343	0.420	−0.474, 1.182	0.815	1.409	.415
Months Diag.	0.000	0.002	−0.004, 0.005	0.131	1.000	.896
Treatmenta	−0.705	0.372	−1.448, 0.018	−1.895	0.494	.058

Note. The model converged after four Fisher iterations. SE = Standard error. 95% CI = 95% Wald Confidence interval. PEM = post-exertional malaise. Symp. Onset = mode of symptom onset. Months Diag. = months since diagnosis at baseline.

^aGender, race/ethnicity, mode of symptom onset, treatment, and PEM are dummy-coded dichotomous variables; index values (i.e., 1) are as follows: female, non-Hispanic White, gradual, highPEM, V-CBSM

^*p < .050

^**p ≤ .010

^***p ≤ .001

Fatigue interference.

Simple treatment effects indicated that highPEM patients receiving V-CBSM reported decreased fatigue interference at follow-up (β = −0.419, p = .003), whereas those in V-HI reported increases (β = 0.478, p = .001). For highPEM patients, group differences were large in magnitude (β = −0.897, p = .001; d = 1.071, 95%CI[0.570, 1.571]; see Table 4). On average, highPEM patients in the V-CBSM group demonstrated a 26% reduction in fatigue interference scores between baseline and follow-up, and 72% of these patients improved on this variable. HighPEM patients in V-HI reported a mean 13% increase in fatigue interference between baseline and follow-up.

Fatigue intensity.

One outlier was present in the model for fatigue intensity (studentized residual = 3.215). Results for models with this case omitted versus retained were generally equivalent, with the exception of the significance of PEM and the PEM-by-treatment interaction (p’s = .110 and .104, respectively, with outlier omitted; p’s = .095 and .076 with outlier retained). Simple effects of treatment and pairwise comparisons, as well as associated effect sizes, were also generally equivalent for both versions of the model. We therefore present the full sample model. Simple effects of treatment on fatigue intensity were not evident for highPEM or lowPEM patients at follow-up. Pairwise comparisons indicated a medium reduction in fatigue intensity for highPEM patients in V-CBSM versus those in V-HI (β = −0.439, p = .037; d = 0.511, 95%CI[0.018, 1.004]; see Table 4). On average, highPEM patients in the V-CBSM group demonstrated a 6% reduction in fatigue intensity scores between baseline and follow-up, and 66% of these patients improved on this variable. HighPEM patients in V-HI on average reported a 1% increase in fatigue intensity between baseline and follow-up.

Exploratory Analyses

Sociodemographic and clinical covariate predictors.

Additional parameters were significant in the regression analyses for primary outcomes (see Table 3). Non-Hispanic White race/ethnicity was associated with increased symptom intensity (β = 0.286, p = .022), perceived stress, (β = 0.376, p = .008), fatigue interference (β = 0.419, p = .011), and fatigue intensity at follow-up (β = 0.365, p = 0.027). Female gender was associated with increased fatigue interference (β = 0.513, p = .020) and gradual symptom onset was associated with greater follow-up fatigue intensity (β = 0.363, p = 0.023).

Intervention effects on PEM classification.

Treatment groups were equivalent on PEM status at baseline (Χ²(1) = 0.963, p = .252). At 5-month follow-up, groups differed significantly (Χ²(1) = 4.520, p = .034), with more highPEM patients in V-HI than in V-CBSM (see Table 2). The logistic regression model was significant and correctly classified 68.7% of cases. The effect of treatment trended towards significance (Odds Ratio = .494, 95%CI[0.235, 1.018], p = .058, pseudo-R² = .269), with individuals who received V-CBSM demonstrating less than half the risk of highPEM classification at follow-up versus those in V-HI (see Table 5).

Discussion

This study was a secondary analysis of a trial testing the effects of a videoconference-delivered group cognitive behavioral stress management intervention (V-CBSM) on symptoms in patients diagnosed with ME/CFS, considering the moderating role of post-exertional malaise (PEM). Although previous trials of CBSM for this population demonstrated large effects on symptom frequency, symptom intensity, and perceived stress (live CBSM) and medium effects on perceived stress (teleconferenced CBSM) [8], the V-CBSM trial revealed no such effects on these variables (see ClinicalTrials.gov). We wondered whether variability in patients’ PEM status, which is now considered a cardinal feature of ME/CFS in updated case definitions [1], obfuscated effects of V-CBSM. We additionally tested effects on fatigue interference and fatigue intensity, as these variables seem to characterize important aspects of patients’ symptom experience and symptom burden.

Patients in our sample were diagnosed with CFS per the Fukuda criteria [17]. Despite that this case definition does not require PEM, 71 of 150 participants (47.3%) reported elevated PEM symptoms (i.e., severe or very severe PEM) at study entry, aligning with PEM prevalence rates reported in other studies [21,35,53]. Thus, we were able to test whether baseline PEM status moderated the efficacy of V-CBSM. Since V-CBSM incorporates relaxation and adaptive coping strategies in addition to traditional CBT techniques (e.g., cognitive restructuring), we reasoned that V-CBSM would be particularly effective for addressing stress-related triggers of PEM. We hypothesized that patients with high levels of PEM would show the greatest effects of V-CBSM on overall and fatigue-related symptoms, as well as perceived stress, compared to an active, time matched health information control (V-HI).

V-CBSM and V-HI participants did not differ on baseline PEM status, recent negative life experiences, or other sociodemographic or clinical characteristics. V-CBSM was associated with increased perceived stress management skills (PSMS) and greater perceived benefit relative to V-HI, as expected. Therefore it appears that assignment to the stress management condition was associated with increases in the intended target skills and greater satisfaction compared to the control condition.

For highPEM patients, controlling for baseline scores on outcomes and other covariates, V-CBSM versus V-HI was associated with medium-to-large effects on symptom frequency, medium effects on symptom frequency, large effects on fatigue interference, and medium effects on fatigue intensity. Across these outcomes, significant effects and non-significant trends suggested overall and fatigue-related symptom reductions for highPEM patients in V-CBSM, and increases for those in V-HI. Given that PEM has previously been linked to worse ME/CFS prognosis [23], worsening symptoms for highPEM patients in the control condition are not unexpected. Medium effects on perceived stress (d = 0.450) only trended towards statistical significance (p = .071) and this may reflect the effectiveness of the V-HI control condition for this variable: it is possible that a dyad meeting with a facilitator on a weekly basis over 10 weeks, even only regarding health-related knowledge and behaviors, improved participants’ overall stress perceptions by decreasing uncertainty, it did not provide specific techniques (e.g., relaxation, cognitive restructuring) that could influence their symptom experience. Moreover, it is important to note that the effects of V-CBSM reflect a medium effect size and this may have more bearing than interpreting the p value in isolation [54].

For primary outcomes, we also calculated overall percent change in scores. Between one half and three quarters of highPEM patients who received V-CBSM showed improvement for any of the five given outcome variables. Reductions were generally around one tenth of baseline scores in terms of magnitude, and on fatigue interference highPEM patients demonstrated a one-quarter reduction in scores. On average, highPEM patients who received V-HI reported increases or small decreases (2%–3% reductions in symptom frequency and perceived stress). It is important to note that percent change is sensitive to changes in variance and therefore vulnerable to bias, and that these statistics do not control for baseline scores on outcomes or other covariates. Nevertheless, they provide an easily interpretable expression of effect size that may more readily translate into real-life relevance for patients and treating clinicians.

Differential treatment effects for highPEM versus lowPEM patients potentially inform the small and nonsignificant effects of telephone-delivered CBSM [8], which did not covary PEM. The virtual visual contact afforded by V-CBSM may also have contributed to greater treatment efficacy than telephone delivery for a group venue. The finding that V-CBSM was associated with increased perceived stress management skills underscores differences between CBSM and other CBT-based approaches for ME/CFS. Considering all CBT-based treatment approaches to ME/CFS, including those based on fear-avoidance models, equivalent likely obfuscates heterogeneity that would be of consequence to ME/CFS patients. Given the degree of patient mistrust surrounding CBT [e.g., 55], facilitating stakeholders’ awareness of differences is vital to ensuring that patients are able to utilize the resources at their disposal.

PEM is associated with a greater degree of psychological concerns and biophysiological dysregulation in ME/CFS [20,21,35], and the neuroimmune model of CBSM appears well-suited to the vulnerable highPEM patient subgroup. Our results suggest that CBSM may intervene on this vulnerability factor, with patients who received V-CBSM demonstrating less than half the risk of follow-up highPEM classification versus those who received V-HI. Although the present study was limited in terms of characterizing PEM status with a single item [see 20], the fact that effects of PEM status were found with our rudimentary measure suggests promising results for future work.

Additional study limitations include the lack of diversity in our sample, which was comprised primarily of non-Hispanic, White, highly educated individuals. Interestingly, we found that members of racial/ethnic minority groups (e.g., Hispanic, African American) showed greater improvements in symptoms over time compared to non-Hispanic White participants. These findings underscore the need for more diverse samples and consideration of these and other patient characteristics in future research. Further, in this study, we report only on findings for ME/CFS patients, despite that both V-CBSM and V-HI were dyadic interventions. It is possible that the interventions affected partners in ways that contributed to patients’ symptoms as well as caregiving burden, and we will be examining these more complex processes in future analyses. Meta-analyses of similar dyadic interventions (e.g., remotely delivered, in the context of chronic physical disease) note that, although caregiver involvement is beneficial for patients, caregiver outcomes are rarely reported [56,57]. More work is needed in this domain.

In the present trial, the V-CBSM intervention was delivered to a group of patient-partner dyads, whereas the V-HI control condition was delivered to single dyads. Thus, individuals who participated in V-CBSM may have benefited from finding support from community. However, a case could be made that those who participated in V-HI may have benefited from exclusive facilitator attention. These two relative benefits may or may not have offset the absence of the other. Nevertheless, the inclusion of a strong, time-matched control condition is a notable strength of this study. Further, in our longitudinal framework, effects on outcomes correspond to 5-month follow-up (i.e., effects that were present 5 months after group start dates, or approximately 10 weeks after the end of the 10-week intervention); this gap was included in order to account for the relapsing nature of ME/CFS and the potential attenuation of intervention effects. Analyses also controlled for baseline scores on outcomes rather than predicting change scores, avoiding artifacts related to regression to the mean and measurement ceilings/floors.

Importantly, the present trial is the first to evaluate the effects of V-CBSM for ME/CFS patients. Given the physical and mental demands of travel, and the fact that limited support is available to the many ME/CFS patients who are homebound [22,34], remote intervention modalities deserve attention in research with this population. Further, our nuanced analysis of treatment efficacy accounts for patient heterogeneity in terms of PEM, aligning with principles of precision medicine and patient-tailored care [e.g., 58,59].

Future directions.

Although the present study was conducted prior to the COVID-19 pandemic, recent research has elucidated similarities between ME/CFS and long-COVID. Qualitative work suggests that over 80% of patients with a clinical course of the primary COVID-19 infection >2 weeks report ongoing illness at 50 days, with symptoms characteristic of ME/CFS (e.g., fatigue, cognitive challenges) that are often exacerbated following exertion [60]. Such symptom exacerbation is similar to PEM, and its varied, persistent, and relapsing nature has been attributed to elevated cytokine response [see 61]. A different study found that a median 54% of patients continue to experience fatigue, flu-like symptoms, and exercise intolerance at ≥6 months [62]. Since viral infection has been implicated as a potential factor in the pathogenesis of ME/CFS, historically termed ‘post-viral syndrome’ [1,63], this phenomenon is not unexpected, and the progression from long-COVID to ME/CFS has also been suggested [61,63,64]. A stress-related mechanism for latent virus reactivation may underlie both long-COVID and ME/CFS, and may explain some of our findings. Up to 67% of long-COVID patients have shown reactivation of Epstein-Barr Virus (EBV) versus 10% of control participants [65]. We previously found that CBSM decreases EBV viral reactivation markers, as well as other herpes virus reactivation markers, in patients with HIV [66,67]. Future studies could examine whether CBSM-associated effects on stress and viral re-activation are related to the CBSM-associated symptom improvement in ME/CFS and those with long-COVID, particularly as the needs of long-COVID patients are likely to overwhelm healthcare resources [62].

Summary.

A 10-week group virtual CBSM intervention demonstrated beneficial effects on overall symptom frequency and intensity, as well as fatigue-specific interference and intensity, at 5-month follow-up for patients diagnosed with CFS [17] who reported severe baseline PEM. A majority of patients with severe PEM who received the V-CBSM intervention improved on each of these outcomes. Overall, these patients reported reductions that were approximately one tenth of baseline scores in terms of magnitude. By considering the moderating effects of heterogeneous patient presentations in the assessment of intervention efficacy, treatment can be more efficiently disseminated to those patients most likely to benefit.

Data Availability Statement:

The data presented in this study are available on request from the corresponding author.