Exercise does not cause post-exertional malaise in Veterans with Gulf War Illness: A randomized, controlled, dose-response, crossover study

Abstract

Chronic multisymptom illnesses (CMI) such as Myalgic Encephalomyelitis/Chronic Fatigue Syndrome, long-COVID, and Gulf War Illness (GWI) are associated with an elevated risk of post-exertional malaise (PEM), an acute exacerbation of symptoms and other related outcomes following exercise. These individuals may benefit from personalized exercise prescriptions which prioritize risk minimization, necessitating a better understanding of dose-response effects of exercise intensity on PEM.

METHODS:

Veterans with GWI (n=40) completed a randomized controlled crossover experiment comparing 20 minutes of seated rest to light-, moderate-, and vigorous-intensity cycling conditions over four separate study visits. Symptoms, pain sensitivity, cognitive performance, inflammatory markers (C-reactive protein and plasma cytokines) were measured before and within 1 hour after exercise and seated rest. Physical activity behavior was measured ≥7 days following each study visit via actigraphy. Linear mixed effects regression models tested the central hypothesis that higher intensity exercise would elicit greater exacerbation of negative outcomes, as indicated by a significant condition-by-time interaction for symptom, pain sensitivity, cognitive performance, and inflammatory marker models and a significant main effect of condition for physical activity models.

RESULTS:

Significant condition-by-time interactions were not observed for primary or secondary measures of symptoms, pain sensitivity, cognitive performance, and a majority of inflammatory markers. Similarly, a significant effect of condition was not observed for primary or secondary measures of physical activity.

CONCLUSIONS:

Undesirable effects such as symptom exacerbation were observed for some participants, but the group-level risk of PEM following light-, moderate-, or vigorous-intensity exercise was no greater than seated rest. These findings challenge several prior views about PEM and lend support to a broader body of literature showing that the benefits of exercise outweigh the risks.

Introduction

Exercise is a safe therapy for numerous physical and mental health conditions.¹ Indeed, a meta-analysis of 773 clinical trials involving a broad spectrum of clinical populations found that exercise therapy did not increase the relative risk of serious adverse events such as hospitalization or death.² However, this same meta-analysis found that the relative risk of “non-serious adverse events” such as acute pain and fatigue responses was increased by 19% and that only 49% of included studies reported data on adverse events.² Thus, despite strong evidence that exercise is a safe and effective therapy for numerous health conditions,¹ there is a gap in the understanding of the potential risks and how to minimize them, especially those assessed via patient reported outcome measures (e.g., symptom exacerbation and burden).

One such risk is post-exertional malaise (PEM), a prevalent occurrence in chronic multisymptom illnesses (CMI) such as Myalgic Encephalomyelitis/Chronic Fatigue Syndrome (ME/CFS), long COVID, and Gulf War Illness (GWI) where symptoms and related outcomes are exacerbated within a few minutes and up to seven days following exercise.^3–7 Personalized, evidence-based exercise prescriptions which balance the goal of maximizing physical and mental health¹ with minimizing the risk of PEM may help address concerns that patient advocates have raised about exercise therapy.⁸ Consequently, studies which investigate and document the effect of different aerobic exercise intensities on PEM are needed to better inform exercise prescription and ensure risk minimization for these individuals.

Inferences about exercise intensity and PEM risk in CMI populations can be drawn from a few prior studies^9,10 but none were designed to systematically test this association. Thus, following a broader initiative to develop evidence-based, individualized exercise prescriptions that maximize benefits and minimize risks,¹¹ we conducted a dose-response study that evaluated acute effects of light-, moderate-, and vigorous-intensity cycling on three types of psychometric outcomes previously shown to be exacerbated by exercise in CMIs – (1) symptom severity,^12–15 (2) experimental pain,^10,16–18 and (3) cognitive performance.^3,19 Further, considering prior assertions of exercise-related interference with daily activities²⁰ and immune function in CMIs,^21–23 we investigated physical activity and inflammatory markers as additional outcomes of interest. We hypothesized that (1) compared to seated rest, exercise would cause greater exacerbation of psychometric outcomes, interference with physical activity and upregulation of inflammatory markers, and (2) these effects would be larger following exposure to higher exercise intensities.

Methods and Materials

This was an unblinded, randomized, controlled, dose-response, crossover study involving 40 Veterans with GWI (see Table 1 for eligibility criteria).²⁴ The Kansas Questionnaire was used to determine GWI status, as defined by whether participants reported developing moderate-to-severe symptoms in at least three of six domains (fatigue, pain, neurological/cognitive/mood, gastrointestinal, skin, respiratory) following deployment to the 1990–91 Persian Gulf War.²⁵ Study procedures took place at the William S. Middleton Memorial Veterans Hospital in Madison, WI, USA between July 2019 and July 2022. Over five total appointments, participants completed an initial orientation and familiarization visit (Visit 1), three different experimental exercise visits and one no-exercise control visit (Visits 2–5) (Fig 1). All procedures were approved by relevant research oversight committees (VA R&D Protocol #1594150 and University of Wisconsin Health Sciences IRB #2018–0476), and all participants provided informed consent before enrollment. Additional details for study methods are provided in Supplemental File 1.

Table 1.

Eligibility criteria for study participation

Inclusion criteria

1. Age between 45–65 years old

2. Deployment to the Persian Gulf region during the 1990–91 Gulf War

3. Meeting Kansas case definition criteria for Gulf War Illness

a. Self-report of moderate-to-severe or multiple symptoms in ≥ three of the following domains: pain, fatigue, neurological/cognitive/mood, skin, gastrointestinal, respiratory

b. Symptoms first becoming a problem during or after the Gulf War, and not explained by a known medical condition

4. Stable use (≥3 mo if taking) of:

a. Psychotropic medications (i.e., antidepressants, antipsychotics, mood stabilizers - except when used to treat exclusionary psychiatric conditions)

b. Opioid analgesics

c. Sedative-hypnotics (>1 considered exclusionary)

d. Anti-convulsants (except when used for exclusionary medical conditions such as psychotic disorders and mood disorders with psychotic features)

Exclusion criteria

1. Being younger than 45 or older than 65 years of age

2. Use of prescription drugs with chronotropic effects (e.g., beta blockers, non-dihydropyridine calcium channel blockers)

3. Use of prescription drugs which may affect symptom reporting (e.g., >1 sedative; non-stable (≥3 mo) anti-convulsant use or use for exclusionary psychiatric conditions)

4. Current use of illegal drugs (except marijuana)

5. Absolute contraindications to exercise testing according to American College of Cardiology/American Heart Association guidelines

6. Psychiatric conditions that might interfere with ability to report symptoms (e.g., bipolar disorder, psychotic disorders, or mood disorders with psychotic features)

7. Conditions listed in the Kansas case-definition: Type 1 Diabetes, Heart disease (other than high blood pressure), Stroke, Lupus, Multiple sclerosis, Rheumatoid Arthritis, Parkinson’s Disease, Lou Gehrig’s Disease, Seizure disorder, Alzheimer’s Disease, Cancer (other than skin cancer), Melanoma, Liver disease, Kidney disease, Schizophrenia, Bipolar disorder, Manic depression, chronic infectious diseases (> 6 months), or hospitalizations in the past 5 years for posttraumatic stress disorder, depression, or alcohol/drug dependence

8. Symptoms explained by other chronic conditions not listed in the Kansas case-definition but with elevated prevalence in aging populations: Untreated sleep apnea or insomnia, Gout, Paget disease, Hypothyroidism, Neuropathy, pulmonary diseases requiring medication (e.g., Bronchitis, COPD, Asthma)

9. Pregnancy or inability to confirm absence of pregnancy (female Veterans only)

Note. Study eligibility screening took place over three phases: (1) an initial pre-screening of medical records, (2) a 15–20 minute scripted phone screening, and (3) a final confirmation of eligibility during the initial study visit.

Fig 1.

Flow chart of study procedures

Note. The study design was a within-subjects, randomized controlled crossover experiment involving three different experimental visits (light, moderate, and vigorous intensity cycling) and one control visit (seated rest). Procedures involved pre-screening of medical records and an initial phone screening to confirm study eligibility, one baseline and familiarization visit (Visit 1), counterbalanced randomization to one of four potential condition orders, and four additional experimental visits (Visits 2–5). During Visits 2–5, symptoms, pain sensitivity, cognitive performance, and inflammatory cytokines were measured before and after the experimental and control conditions. Study visits were separated by ≥7d, during which participants were asked to wear a physical activity monitor and complete daily at-home symptom questionnaires.

Study procedures involved pre-screening medical records and an initial phone screening to confirm study eligibility, one baseline and familiarization visit which included measures of trait symptoms, expectations, demographics, medical history, and military related information (Visit 1), counterbalanced randomization to one of four potential condition orders, and four testing visits (Visits 2–5). Details on the instruments used to assess trait symptoms and military service-related information on Visit 1 are provided in Supplemental File 1. During Visits 2–5, symptoms, pain sensitivity, cognitive performance, and inflammatory cytokines were measured before and after the experimental and control conditions. Study visits were separated by ≥7 days, during which participants were asked to wear a physical activity monitor and complete daily at-home symptom questionnaires.

Exercise testing took place on an electronically braked cycle ergometer (Lode Corival CPET, Groningen, The Netherlands) in a controlled laboratory environment (Mean±SD temperature=20.7±2.2°C; humidity= 40.3±19.3%; barometric pressure= 738.6±5.6 mmHg). Testing lasted ~30 minutes and consisted of a two-minute rest, a five-minute warm-up, 20 minutes of steady-state cycling, and three minutes of unloaded active recovery. Cycling cadence was kept between 50–70 rpm during warm-up and steady-state periods. Per American College of Sports Medicine recommendations, light, moderate, and vigorous intensities correspond to % heart rate reserve (HRR) ranges of 30–39%, 40–59%, and 60–89%, respectively.²⁶

To standardize between-dose intervals and minimize overlap in cardiorespiratory responses, heart rate (HR) was continuously monitored throughout exercise testing (Polar Electro Inc. Bethpage NY) and workload was adjusted 1–10 W/minute during exercise as needed to maintain a target HRR of 35±3% (light), 50±3% (moderate), and 65±3% (vigorous). Dosing standardization was also confirmed via post-test analysis of oxygen consumption (V̇O₂), carbon dioxide production (V̇CO₂), and minute ventilation (V̇E), as measured by metabolic cart (TrueOne^® 2400 Parvomedics, Sandy, UT). To account for the passage of time and attention received during exercise, a no-exercise control condition that involved asking participants to sit quietly for 30 minutes was also included. All measurements taken during the no-exercise control condition were identical to the experimental conditions.

Pre- and post-test laboratory measures

Based on prior related work,^14,15,27 three questionnaires were selected to measure acute symptom responses to exercise. GWI symptom severity was measured via a visual analog scale (VAS) adapted version of the Kansas Questionnaire (Kansas VAS).¹⁴ Additionally, mood and pain symptoms were measured via the Profile of Mood States (POMS)²⁸ and the Short-form McGill Pain Questionnaire-2 (MPQ).²⁹ Questionnaire instructions asked participants to indicate how they feel “right now”. The primary outcome for symptom severity was the Kansas VAS total score that was calculated by taking the average score across all 29 items which comprise the instrument. Secondary outcomes included scores for the (1) Kansas VAS symptom categories, (2) POMS total and subscales (tension, depression, anger, fatigue, confusion, vigor), and (3) MPQ total and subscales (neuropathic, affective, continuous, intermittent).

Despite the potential psychometric advantages of using an instrument which has been validated for PEM measurement such as the DePaul Symptom Questionnaire Short form (DSQ-SF),³⁰ common data elements guidelines for PEM research in GWI promote using questionnaires and item phrasing that capture immediate and/or day-to-day changes, but are also representative of GWI symptoms³¹. Thus, we chose the adapted Kansas VAS as our primary symptom outcome for several reasons. First, the DSQ-SF assesses a small subset of GWI symptoms and is not optimal for measuring acute responses to exercise since it asks about symptoms experienced over the last 6 months. Second, visual analog scales (VAS) are valid measures of symptom severity³² and are sensitive to acute PEM responses to exercise in CMI populations.^33,34 Third, the Kansas Questionnaire is a widely used screening tool in GWI research and its items are based on symptoms reported by Gulf War Veteran cohorts.²⁵

Experimental pain sensitivity was measured using psychophysical procedures employed in our prior research.³⁵ Fourteen thermal stimuli ranging from 43–49°C (two per temperature) were delivered in a randomized order to the thenar eminence (palm) of the non-dominant hand using the Q-Sense Small Fiber Testing System (Medoc Advanced Medical Systems, Durham, NC). Stimulus durations lasted eight seconds followed by a one minute inter-stimulus interval in which sensory (i.e., intensity) and affective (i.e., unpleasantness) pain ratings were measured via two separate 0–20 category-ratio scales.³⁶ To minimize effects of pain anticipation on ratings,³⁷ participants were blinded to stimulus temperature before delivery.

Primary and secondary outcomes for pain sensitivity were intensity and unpleasantness ratings, respectively. Because pain sensitivity testing involved a range of stimulus intensities, an area under the curve (AUC) approach (R package DescTools³⁸) was used to a generate a single score (i.e., mathematical integral) to reflect the overall stimulus-response curve for each participant. This approach has been applied in studies of chronic headache³⁹ and fibromyalgia,⁴⁰ and outperforms other stimulus-response curves methods (e.g., R², slope) in terms of robustness to linearity violations and strength of association with clinically relevant outcomes.⁴¹

The Connors Continuous Performance Task (CPT-3), a computerized test of sustained attention and response inhibition, was selected to measure cognitive performance because it is sensitive to change in periods as little as 30–60 minutes⁴² and has previously revealed small-to-moderate cognitive deficits in Gulf War Illness.⁴³ The CPT-3 was administered via commercial software following standard testing instructions (Multi-Health Systems Inc., North Tonawanda, NY).⁴⁴ Each CPT-3 began with a one-minute practice session followed by a 14 minute testing period. For each test, 360 random letters appeared on the computer screen one at a time. Participants were instructed to press the spacebar when any letter other than an “X” appeared. Average reaction time (ms) for correct responses was treated as the primary outcome. Secondary outcomes were correct responses, commissions, omissions, and detectability.

Plasma levels of inflammatory markers were measured as an indication of systemic inflammation following exercise and included interleukin-6 (IL-6), interleukin-8 (IL-8), interleukin-10 (IL-10), tumor necrosis factor-alpha (TNF-α), and C-reactive protein (CRP). Peripheral whole blood was drawn via venipuncture from the antecubital vein into two 10 mL EDTA tubes pre- and immediately post-exercise. Whole blood was centrifuged at 3500 RPMs (Ultra-8S, LW Scientific, Lawrenceville, GA) for ten minutes at 3°C immediately after collection. Following centrifugation, plasma was isolated from the whole blood and aliquoted into 2.0 mL microcentrifuge tubes (Thermo Scientific, Rochester, NY) and stored in a −80°C freezer. Following study completion, samples were transported to the Wisconsin National Primate Research Center (Madison, WI) for quantitative analysis.

Cytokines and CRP were quantified in duplicate determinations by multi-cytokine array, and CRP was determined in a single-plex assay using an electrochemiluminescence platform and a QuickPlex SQ 120 imager for analyte detection (Meso Scale Discovery, Gaithersburg, MD). Each specimen was referenced to a standard curve generated from 7 calibrators with known concentrations. The lower limit of detection was 0.1 pg/mL, and the assay had a large dynamic range up to 2000 pg/mL. CRP values were converted to mg/L units to be consistent with the extant clinical literature. Each assay also included a pool to monitor for assay drift and generated both intra- and inter-assay coefficients of variation to verify the reliability of the quantitative determinations of analyte levels. To reduce skewedness, inflammatory cytokine and CRP values were first log-transformed and confirmed for normality using both the Shapiro-Wilk Test and density plots. No single inflammatory marker was identified a priori as a primary outcome, so all five markers were considered secondary for the statistical analyses.

At home measures

An ActiGraph GT3X+ triaxial accelerometer (ActiGraph, LLC, Pensacola, FL) was used to measure physical activity following Visits 2–5. The ActiGraph is a reliable device, with a mean intra-instrument coefficient of variation of 4.1% and a mean inter-instrument coefficient of variation of 4.9%.⁴⁵ Participants were instructed to wear the monitor on their hip via an elastic belt or belt clip during all waking hours except for bathing or swimming for ≥7 days. To facilitate data processing and interpretation, participants completed a daily physical activity log to indicate when they (1) woke up, (2) applied and removed the monitor, and (3) went to sleep. Participants were also encouraged to complete daily symptom questionnaires (Kansas VAS, POMS, MPQ) at approximately the same time of day when possible.

Accelerometer data were processed using in-house software to calculate time spent in sedentary, light, moderate, and vigorous activity levels.⁴⁶ Criteria for valid wear time was ≥ 10 hours for a minimum of 3 week days and 1 weekend day.⁴⁷ Moderate-to-vigorous activity (expressed as a percentage of total wear time) was treated as the primary outcome measure of physical activity. Secondary outcomes were percentage of wear time spent engaging in sedentary and light activity, steps per day, and transitions from a sitting to standing position.

Statistical analyses

Participants who completed at least three of four experimental visits were included in the statistical analysis. An a priori power analysis conducted with G*Power (version 3.1.3)⁴⁸ showed that 40 participants provided a power (1-β) of 0.80 to detect a relatively small condition-by-time interaction effect (partial ɳ² =0.03) via repeated measures analysis of variance (RM-ANOVA), assuming a 2-tailed α value of 0.05 and a correlation of r = 0.80 across two repeated measures. However, to better account for (1) missing data, (2) violations of normality, and (3) individual differences, we instead used linear mixed effects analyses to test our hypotheses.

Statistical analyses were conducted using SPSS (Version 29) and R (Version 4.4). Data visualization and graphics were created using both ggplot2⁴⁹ and GraphPad Prism (Version 10.2.2.). One-way analysis of variance was used to confirm standardization of exercise dosing and compliance across conditions. Linear mixed effect regression models (R packages lme4⁵⁰ and lmerTest⁵¹) with fixed (condition, time, condition-by-time interaction) and random (intercept, slope) effects were used to test hypotheses for primary psychometric outcomes and inflammatory markers (α=0.05). Hypothesis testing for the primary physical activity outcome was conducted via a linear mixed effect regression model with fixed (condition) and random (intercept, slope) effects (α=0.05).

Secondary outcomes were analyzed with the same statistical approaches applied to primary outcomes, though Benjamini-Hochberg corrections (p_adj) were applied to account for family-wise error. Exploratory linear mixed effects models with fixed (condition, time, condition-by-time interaction) and random (intercept, slope) effects tested whether symptom responses differed between conditions when symptoms measured 1–7 days following each test visit were included in the analysis of those measured in the controlled laboratory setting (pre, 10 minutes, 50 minutes).

Results

Recruitment and sample characteristics

In total, 129 Veterans were screened, 47 were randomized, and 40 completed the study (see Fig 2 for CONSORT diagram). Descriptive statistics for demographic, health, physical activity, and military-related information are provided in Tables 2–3 and Supplemental Tables S1–S5.

Fig 2.

CONSORT flow diagram

Note. Participants were randomized in a counterbalanced fashion to one of four potential condition orders (10 per order): a) Rest, Light, Moderate, Vigorous, b) Light, Moderate, Vigorous, Rest, c) Moderate, Vigorous, Rest, Light, and d) Vigorous, Rest, Light, Moderate. One participant did not complete the moderate intensity exercise condition for personal reasons. Three participants did not complete the vigorous intensity exercise condition due to blood pressure-related safety concerns (n=2) or pain from a pre-existing knee injury (n=1).

Table 2.

Demographic characteristics of the study sample (n=40)

	Mean	SD
Age (years)	53.34	4.39
Body mass index (kg/m2)	32.46	5.26
	Frequency	Percentage
Biological sex
Male	36	90%
Female	4	10%
Marital status
Married	29	72.5%
Divorced	9	22.5%
Single, never married	2	5%
Employment status
Employed full time	32	80%
Employed part time	3	7.5%
Home maker	1	2.5%
Unemployed for health reasons/medically retired	1	2.5%
Retired	3	7.5%
Race
Black or African American	1	2.5%
White	38	95%
Multiple races	1	2.5%
Education
High School or General Educational Development	3	7.5%
Some college or vocational school	21	52.5%
Four-year college degree	12	30%
Master’s degree or higher	4	10%
Income
Under $20,000	1	2.5%
$20,000–34,999	3	7.5%
$35,000 to 49,999	1	2.5%
$50,000–75,000	7	17.5%
Over $75,000	28	70%

Table 3.

Baseline symptoms reported during study visit 1 (n=40)

Instrument	Frequency	Percentage
Met criteria for moderate-severe symptoms (Kansas Questionnaire) (Yes/No, %)
Fatigue domain	36/4	90%
Pain domain	35/5	87.5%
Neurological/Cognitive/Mood domain	40/0	100%
Gastrointestinal domain	17/13	42.5%
Respiratory domain	13/27	32.5%
Skin domain	9/31	22.5%
Met criteria for Fibromyalgia (Widespread Pain Index - ACR 2010) (Yes/No, %)	9/31	22.5%
Met criteria for post-exertional malaise (DSQ-SF) (Yes/No, %)	6/34	15%
	Mean	SD
VR-36 Physical Component score (0 – 100)	68.45	17.07
VR-36 Mental Component score (0 – 100)	68.78	13.54
Brief Fatigue Inventory (BFI) total score (0 – 10)	3.17	1.99
Multidimensional Fatigue Inventory (MDFI) total score (20 – 100)	58.12	10.5
Perceived Deficits Questionnaires (PDQ) total score (0 – 80)	31.53	15.46
Pittsburgh Sleep Quality Index (PSQI) total score (0 – 21)	11.13	3.3
Profile of Mood States (POMS) total score (−30 – 200)	18.2	18.41
McGill Pain Questionnaire (MPQ) Total score (0 – 10)	1.46	1.23

Note. Table 3 summarizes the profile of symptoms reported by participants during the initial orientation and familiarization visit (Visit 1). Categorical data are expressed as the total number of participants (frequency) and percentage of the total sample who met criteria for (a) moderate-to-severe symptoms across six different Gulf War Illness domains (Kansas Questionnaire), (b) Fibromyalgia (Widespread Pain Index - American College of Rheumatology 2010 Criteria), and (c) post-exertional malaise (DePaul Symptom Questionnaire - PEM Subscale; DSQ-SF). Continuous data are expressed as means and standard deviations (SD). Reference ranges are also provided to facilitate the following score interpretations: VR-36: higher scores indicate lower functionality; BFI: higher scores indicate greater fatigue; MFDI: higher scores indicate greater fatigue; PDQ: higher scores indicate greater cognitive impairment; PSQI: higher scores indicate worse sleep quality; POMS: lower scores indicate greater mood disturbance; MPQ: higher scores indicate greater pain symptoms.

Exercise dosing standardization and compliance

A one-way analysis of variance revealed a significant main effect of condition on HR during exercise testing (F_3,152 = 367.89; p<.001). Planned comparisons and Hedges’ d effect sizes showed large, dose-dependent, between-condition differences (Fig 3). Similar findings were observed for relevant cardiorespiratory and perceptual variables (Supplemental Table S6). The percentage of the total exercise test spent in heart rate zones corresponding to light, moderate, and vigorous intensity (as defined by the American College of Sports Medicine²⁶) differed significantly across condition (F_2,117 = 4.59; p=.012; ɳ²_p=0.07). Post hoc testing revealed significantly lower percentages for the vigorous condition (Mean ± SD: 73.3% ± 33.9%) compared to the light (90.5% ± 21.7%) and moderate (88.4% ± 26.2%) conditions.

Fig 3.

Mean (95% CI) heart rate during the experimental light, moderate, and vigorous intensity exercise conditions, and seated rest control condition.

Note. A one-way analysis of variance revealed that heart rate (treated as an average between min 7–27) significantly differed between conditions (F_3,152 = 367.89; p<.001). Planned comparisons and Hedges’ d effect sizes revealed that (1) light intensity exercise was greater than seated rest (p<.001, d=3.3; 95% CI: 2.6, 4), (2) moderate intensity exercise was greater than light intensity exercise (p<.001, d=2.1; 95% CI: 1.6, 2.7), and (3) vigorous intensity exercise was greater than moderate intensity exercise (p<.001, d=1.8; 95% CI: 1.3, 2.3).

Psychometric outcomes

Descriptive statistics are provided in Supplemental Tables S7–S12. Overall missingness for symptoms, pain sensitivity, and cognitive performance data was 2.7, 8.75, and 2.2%, respectively. Significant condition-by-time interactions were not observed for the Kansas VAS total score (F_{6, 304} = 0.74; p=.62), pain intensity AUC (F_3,142 = 0.50; p=.69), or reaction time for correct responses (F_3,152 = 0.39; p=.76). Similarly, no significant interactions were observed for secondary outcomes (Supplemental File 2).

Inflammatory cytokines and C-reactive protein

Coefficients of variation (CV) indicated that assay results were both accurate (intra-assay CV% < 5%) and precise (inter-assay CV% < 10%) for all inflammatory markers (i.e., IL-6, IL-8, IL-10, TNF-α, and CRP) across three quality control levels (i.e., low, medium, and high). Descriptive statistics and mixed effect model summaries are provided in Supplemental Table S13 and Supplemental File 2, respectively. Overall missingness for cytokine values was 12%. Significant fixed-effect condition-by-time interactions were observed only for IL-6 (F_3,131 = 3.64, p_adj = .04), with mixed effect model coefficients indicating significant pre- to post-exercise effects for moderate (β = 0.05, SE = 0.02, p=.02) and vigorous (β = 0.06, SE = 0.02, p=.01) conditions. Significant condition-by-time interactions were not observed among other cytokines or CRP (Supplemental File 2).

Physical activity

Descriptive statistics are provided in Supplemental Table S14. Overall missingness for accelerometry data was 10.3%. Significant between condition differences were not observed for moderate-to-vigorous activity (F_3,100 = 0.14; p=.94) or secondary outcomes (Supplemental File 2).

Exploratory analyses

Overall missingness for daily symptom data was 35%. Linear mixed effects models did not reveal significant group-by-time interactions for total or subscale scores for the Kansas VAS, POMS, or MPQ. However, significant main effects of time were observed for 18 of 19 models. Visual inspection of plots indicated a general pattern of increased symptom severity 1–7 days after leaving the laboratory relative to when symptoms were measured in the laboratory setting (pre, 10 minutes post, 50 minutes post) (see Supplemental Fig S1–S4).

Discussion

This study evaluated acute dose-response effects of aerobic exercise intensity on PEM and associated outcomes (e.g., symptoms, experimental pain, cognitive performance, inflammatory markers, and physical activity) in Veterans with GWI. Undesirable effects such as symptom exacerbation were observed for some participants, but the overall risk of PEM following light-, moderate-, or vigorous-intensity exercise was no greater than seated rest. Interpretations and implications of these findings are centered around comparisons with prior research and future directions for exercise prescription in GWI and other CMI populations.

Minimal risk of exercise-related worsening

The degree to which our primary findings conflict with prior literature depends on the outcome that was measured. For instance, exercise-related symptom exacerbation is particularly well documented across CMI populations, as indicated by our own laboratory studies involving Veterans with GWI^14,27 and meta-analyses of acute pain¹³ and fatigue¹² responses in people with ME/CFS and Fibromyalgia. Notably, the meta-analytic studies found smaller effects for studies measuring symptoms at earlier (e.g., <2 hours post) versus later (e.g., 4–72 hours post) timepoints,^12,13 so perhaps the null findings in our study are explained by focusing on symptoms measured within one hour post-exercise. However, exploratory analyses of symptoms measured 1–7 days following exercise testing also failed to provide evidence of exercise-induced symptom exacerbation (Supplemental Fig S1–S4).

There are considerably fewer investigations of experimental pain, but these studies broadly provide supporting evidence for elevated risk of exercise-related worsening in CMI populations. One study found that pain pressure threshold improved in healthy controls and worsened in people with Fibromyalgia 48 and 96 hours after exhaustive treadmill exercise, but this effect could be attributed to improved pain sensitivity in healthy controls relative to people with ME/CFS.¹⁷ A study of Gulf War Veterans with chronic widespread musculoskeletal pain found no significant differences in thermal or pressure pain thresholds following vigorous intensity cycling, but there were significant increases in pain ratings for the thermal stimuli following exercise.¹⁶ Further, an ME/CFS study attempted to minimize exacerbation via self-paced, physiologically limited exercise but found that increases in pain sensitivity were comparable to submaximal incremental exercise.¹⁸ Curiously, these findings conflict to some degree with a report of decreased pain pressure thresholds following moderate but not vigorous intensity cycling in people with Fibromyalgia.¹⁰

Relative to other psychometric outcomes, exercise-related worsening of cognitive performance is less consistent across studies. Although, two ME/CFS studies reported decreased cognitive performance immediately¹⁹ and 24 hours³ following exercise, several others have not.^52–55 For instance, Lamanca and colleagues reported “impaired cognitive processing” immediately and 24 hours after exhaustive treadmill exercise, but their data reflected improved performance in healthy controls rather than worsening in people with ME/CFS.⁵² Similarly, Rayhan and colleagues reported “exercise-induced cognitive dysfunction” in Veterans with Gulf War Illness after dividing the sample into those who improved (“increasers”) or got worse (“decreasers”) within one h following repeated exercise exposures.⁵³ However, significant changes in the “increaser” subgroup versus non-significant changes in the “decreaser” subgroup provides stronger evidence for improvement than deficits following exercise.⁵³ Finally, two particularly rigorous studies, one comparing monozygotic twins with and without ME/CFS⁵⁴ and another involving multiple CMI groups and using a randomized, no-exercise control design,⁵⁵ found no evidence of cognitive deficits within 20–25 minutes of exercise.

Peripheral inflammation has been proposed as one mediator of CMI symptom maintenance.^56,57 Prior case-control studies have reported differences in basal levels of IL-10,⁵⁶ TNF-α,⁵⁸ and CRP,⁵⁹ and an overall cytokine profile that would polarize for immune responses toward a Th1 or cellular bias.⁶⁰ Prior exercise-challenge studies of GWI have reported differences in post-exercise immune responses^61–63 and associations between symptom severity and regulatory cytokines like IL-10 ⁶⁴ after exhaustive graded exercise testing. A significant Condition-by-Time interaction was observed for IL-6, with the greatest post-exercise increases occurring after the moderate and vigorous conditions compared to seated-rest. Our findings are consistent with acute exercise studies in healthy adults, demonstrating cytokines (e.g., IL-6) may increase in an intensity-dependent fashion.^65,66 The absence of widespread differences across other tested cytokines (e.g., IL-8, IL-10, TNF-α) and CRP suggests longer durations or higher intensity (e.g., exercising to exhaustion) may be required to further differentiate conditions. However, inflammatory responses were observed across all conditions (i.e., main effects of Time for IL-8, IL-10, and CRP) which again implies the lack of a causal effect of exercise and denotes the importance of the seated rest control condition for interpreting our findings (Supplemental Fig S5). While previous evidence corroborates that certain cytokine levels (e.g., IL-6 & TNF-α) peak immediately post-exercise in healthy^65,67 and GWI⁶² populations, future research should include additional measurement windows (e.g., 8 hours post-exercise) to test whether prolonged elevations in cytokines would become manifest in a dose-dependent fashion.

The hypothesis that CMI populations avoid physical activity in order to prevent exercise-related worsening of symptoms or from fear of further exacerbation²⁰ has been supported by a recent qualitative survey of people with ME/CFS,⁶⁸ but the few studies that objectively measured physical activity following exercise challenge have provided less compelling evidence. An early ME/CFS study examined day-to-day activity and found up to 30% lower overall average activity 5–7 days following exhaustive treadmill exercise.⁶⁹ Because these findings were largely driven by longer durations of wake time (up to 13% min/d) and higher total rest breaks (up to 26.5%), the authors concluded that exercise-related effects were not so severe that people with ME/CFS could not compensate.⁶⁹ Another investigation found that people with ME/CFS spent a greater proportion of wakeful hours lying down 24 hours following moderate-intensity cycling,⁷⁰ but this study did not include a healthy control comparison group and was strictly focused on sedentary behavior. Based on several different actigraphy metrics, we did not find evidence of exercise-related interference with physical activity or sedentary behavior but this is perhaps explained by the small effect of exercise on symptom exacerbation.^20,68

Unsurprisingly, comparisons with prior work are limited by differences in one or more factors such as CMI population, type and timing of outcome measures, and exercise stimulus characteristics. However, a methodological difference which warrants extra attention when translating evidence to exercise prescription in clinical practice is study design. Specifically, most prior studies used a case-control, pre-post exercise design whereas this study used a randomized crossover design and included a no-exercise control condition. As discussed elswhere,⁵⁵ comparing exercise to a no-exercise control condition isolates exercise-related effects from other factors, thereby increasing the ability to make causal inferences. For instance, our exploratory analyses revealed that severity of symptoms measured 1–7 days following exercise testing was higher than when measured in the laboratory setting (see Supplemental Fig S1–S4). However, observing a similar pattern following seated rest implies that these responses were not caused by exercise per se. Thus, while case-control studies have clear utility in understanding CMI pathophysiology,⁷¹ the design used here is perhaps better suited for understanding the effect of exercise on PEM.

Exploratory findings

Basing clinical practice recommendations on exploratory analyses should generally be avoided.⁷² Bearing this caveat in mind, exploratory analyses of symptoms collected up to seven days following in-lab testing add context to primary and secondary analyses of symptom responses in the laboratory setting. Unexpectedly, scores on 18 of 19 total symptom measures were significantly lower in the laboratory setting relative to the natural at-home setting, regardless of condition. Plausible explanations include that (1) interacting with study team members distracted participants from their symptoms when in the laboratory environment, (2) participants felt more comfortable with reporting symptom severity outside of the laboratory environment, or (3) testing procedures themselves caused post-exertional malaise. Further, sensory inputs encountered within the laboratory setting differ from those of daily life⁷³, and the varying impact of setting-specific stimuli (i.e., home-based vs. laboratory-based setting) on perceptual expectations may have influenced participant symptom perception. Importantly, symptoms were also elevated following the no-exercise control condition, indicating that other factors beyond exercise (e.g., testing procedures, travel to study visit, etc.) may contribute to symptom exacerbation.

Limitations and future directions

This study has several potential limitations and future research directions. First, unblinding participants and test administrators to condition assignment may have introduced bias into outcome measures. Nevertheless, non-significant changes in their expectations over the course of the study tempers this concern to some degree (see Supplemental Fig S6).⁷⁴ Second, while our findings may translate to exercise prescription for other CMI populations such as ME/CFS⁷⁵ and Long-COVID,⁷⁶ demographic, etiologic, and exercise-related physiological differences between Veterans with GWI and other CMIs^21,77 may limit the generalizability of our findings. Further, despite meeting case-definition criteria for GWI, the baseline characteristics of our sample reflected less severe symptoms than some prior studies.^14,27 Prior studies of ME/CFS have reported associations between baseline symptom severity and PEM responses,^78,79 suggesting that the most pronounced PEM responses occur in individuals with greater illness severity. Thus, future research which applies the model used here to other CMI groups and a more representative GWI sample would help address these shortcomings. Finally, basing exercise prescriptions solely on group level analyses may have limited application for certain individuals.⁸⁰ For instance, although linear mixed effects analysis indicated the group-level effect of vigorous exercise on Kansas fatigue scores was not different from quiet rest, plotting individual responses shows that fatigue increased for 46% of participants after vigorous exercise versus 15% of participants after quiet rest (see Supplemental Fig S7). Thus, in line with research which has identified risk factors and developed risk stratification models for people with cardiovascular disease,⁸¹ a similar initiative for CMI populations may facilitate the translation of findings from clinical trials to patient centered exercise prescription in clinical practice.⁸⁰

Conclusion

Undesirable effects such as symptom exacerbation were observed for some participants, but the group-level risk of PEM from light-, moderate-, or vigorous-intensity exercise was no greater than seated rest. These findings challenge several prior reports about the risks associated with exercise in individuals with GWI and lend support to a broader body of literature showing that the benefits of exercise outweigh the risks. Future studies should test the replicability of these findings across other CMI populations, ideally using an adequate control condition to strengthen their interpretability and comparability. Further, better documentation of PEM and its association with adherence in clinical trials may inform exercise prescription guidelines and strengthen the nexus between research and clinical practice for these individuals.

Highlights.

• Post-exertional malaise is a worsening of symptoms following exercise

• Gulf War Illness is associated with an elevated risk of post-exertional malaise

• This study examined dose-response effects of exercise intensity in Gulf War Illness

• Undesirable effects of exercise were observed for some participants

• On average, study outcomes were not worsened by exercise compared to rest