BMJ 2004;328:731 (27 March), doi:10.1136/bmj.38041.724421.55 (published
19 March 2004)
David C Mohr, associate professor, Stacey L Hart, assistant professor, Laura Julian, postdoctoral fellow, Darcy Cox, assistant professor, Daniel Pelletier, assistant professor
1: Department of Psychiatry, University of California, San Francisco, CA 94142, USA,
2: Department of Medicine, University of California, San Francisco,
3: Department of Neurology, University of California, San Francisco
To quantify the association between stressful lifeevents and exacerbations of multiple sclerosis.
PubMed, PsychInfo, and Psychological Abstractssearched for empirical papers from 1965 to February 2003 withterms "stress", "trauma", and "multiple sclerosis".
Three investigators independently reviewed papersfor inclusion/exclusion criteria and extracted the relevantdata, including methods, sample statistics, and outcomes.
Of 20 studies identified, 14 were included. The meta-analysisshowed a significant increase in risk of exacerbation in multiplesclerosis after stressful life events, with a weighted averageeffect size of d = 0.53 (95% confidence interval 0.40 to 0.65),P < 0.0001. The studies were homogenous, Q = 16.62, P = 0.22,I2 = 21.8%. Neither sampling nor study methods had any effecton study outcomes.
There is a consistent association between stressfullife events and subsequent exacerbation in multiple sclerosis.However these data do not allow the linking of specific stressorsto exacerbations nor should they be used to infer that patientsare responsible for their exacerbations. Investigation of thepsychological, neuroendocrine, and immune mediators of stressfullife events on exacerbation may lead to new behavioural andpharmacological strategies targeting potential links betweenstress and exacerbation.
Multiple sclerosis is a chronic, often disabling disease inwhich the immune system attacks the myelin sheath of axons ofthe central nervous system.1 Most people with multiple sclerosishave a relapsing form of the disease, characterised in partby exacerbations in which symptoms appear suddenly within 24hours. These symptoms remit slowly over the course of weeksor months but often can leave some residual impairment. Symptomsvary considerably across patients and can include loss of functionor feeling in limbs, loss of bowel or bladder control, sexualdysfunction, debilitating fatigue, blindness due to optic neuritis,double vision, loss of balance, pain, loss of cognitive functioning,and emotional changes.
Numerous triggers of exacerbation have been proposed, includingbacterial or viral infections that cause T cells to "mistake"myelin proteins for these antigens, bacterial "superantigens,"physical injury, or stressful life events.2 Of these, the roleof stressful life events has been by far the most controversial.3
The notion that psychological stress might trigger exacerbationdates back more than 100 years to Charcot, who speculated thatgrief, vexation, and adverse changes in social circumstancewere related to the onset.4 Most patients with multiple sclerosisbelieve that stressful events can cause or contribute to theirexacerbations.5 Over the past decades numerous empirical studiesof the question have been published. Goodin et al's qualitativeliterature review on the effects of stress on clinical exacerbationarrived at an equivocal conclusion.3 While the review containedcareful discussions of each paper included, it did not includeall relevant studies and did not use quantitative meta-analytictechniques. Furthermore, several important studies have beenpublished since this review.
To clarify the present state of empirical research we conducteda systematic review and quantitative meta-analysis to evaluateand quantify the association between stress and clinical exacerbationin multiple sclerosis.
Identification of studies
We searched PubMed, PsychInfo, and Psychological Abstracts from1965 to February 2003 using the terms "stress", "trauma", and"multiple sclerosis". Potential unpublished data sources wereobtained by using the same keywords in a search on the databaseof computer retrieval of information on scientific projects(National Institutes of Health). This database includes federallyfunded ongoing projects. On the basis of title and abstractwe manually examined English and non-English reference listsof all articles to locate any other referenced journal articlesnot identified in the preliminary search.
We included studies if they used standardised diagnosistic criteriafor multiple sclerosis (for example, Schumacher6 or Poser7)with a relapsing course equivalent to current classificationsof relapsing-remitting or secondary progressive, or both; exacerbationwas confirmed by a neurologist; they used standardised or standardchecklist methods through interview or questionnaire to measurestressful life events; they used case-control or longitudinaldesign; and they provided enough information about results toallow us to compute an estimate of effect size (such as meansand standard deviations; group percentages, etc). We excludedstudies if stress could not be distinguished from psychopathologyor "temperament"; stress included only physical trauma (forexample, head injury) or a medical condition, as these mightbe confounded with outcome measurement; there was no clear effectsize or test statistic that allowed us to compute an effectsize; or the same data were used for two reports.
Coding of studies
Three authors (DCM, LJ, SLH) reviewed all eligible studies.They coded studies on design (case-control v longitudinal prospective);outcome (first diagnosed exacerbation v exacerbation in a diagnosedsample); type of patients, included relapsing-remitting multiplesclerosis v secondary-progressive multiple sclerosis (studiesbefore this distinction were presumed to contain both); useof validated v unvalidated measure of stress; use of self reportv structured interview assessment of stress; age; proportionof female patients; and method used to identify exacerbationof multiple sclerosis (judgment of neurologist, change on astandardised neurological exam called the expanded disabilitystatus score,8 or retrospective validation by neurologist ofstandardised exacerbation and symptoms reported by the patient).We had intended to code for severity of disease, but markerswere too variable across studies to be aggregated.
The dependent variable was occurrence of exacerbation of multiplesclerosis. The primary measure of stressful life events wasthe independent variable for all studies. Many of the studiesreport additional subanalyses of the effects of specific areasof stress (such as family, work, bereavement, etc) on exacerbation.However, there was little consistency in how these sources ofstress were conceptualised or grouped, making it impossiblefor us to code them reliably. We therefore focused only on globalstressful life events as the independent variable.
Meta-analytic calculations were conducted as described in Lipseyand Wilson,9 with the aid of their effect size calculator computerprogram as well as three macros written by those authors forSPSS. We calculated effect sizes from statistics provided ineach article. An effect size is defined as a standardised indexof the ability of stress as measured in these studies to predictclinical exacerbation of multiple sclerosis, or the magnitudeof differences between the exacerbation group and the controlor comparison group on the measure of stress. Weighted effectsizes were used for aggregation of effect sizes. Each effectsize was weighted by the inverse variance weight, which providesan adjustment based on the sample size and standard error ofthe mean effect size. Once this adjustment was made we calculated95% confidence intervals for the mean effect sizes using thesimple random effects approach, which is considered a statisticallyconservative estimate. The effect sizes and their confidenceintervals resulting from the meta-analysis are reported in Cohen'sd, a standardised effect size measure ((mean1 - mean2)/s)). Severalof the papers did not contain enough information for us to calculateCohen's d. We therefore calculated standard errors of measurementfor Fisher's z, using effect size and sample size, and translatedthem into Cohen's d and multiplied them by 1.96. We used theQ test to test homogeneity of variance. This test examines ifthe observed variability in study effect sizes is within therange that can be expected if all studies shared a common populationeffect size. We also calculated the I2 statistic (I2 = 100%*(Q- df)), which ranges from 0-100% and provides the degree ofinconsistency across studies (that is, larger percentages reflecthigher degrees of inconsistency). We examined the impact ofstudy characteristics, including sample and study design, onstudy effect sizes using the analog to one way analysis of variance(ANOVA) and regression. As we assumed excess variability tobe due to both sampling error and random differences acrossstudies, we fitted the data to a random effects model, whichprovides more conservative estimates than a fixed effects model.
The literature search produced a total of 20 articles. Six paperswere
excluded because they did not meet inclusion criteria.510-14The
table shows the general characteristics of the 14 includedstudies, including
sample sizes, available details of participantsand the disease, study design,
length of follow up, exacerbationcriteria, and measurement of stress. Of
the 14 studies, sevenwere case-control studies and seven were longitudinal
prospectivestudies. Two studies examined first exacerbations, which ledto
a diagnosis of multiple sclerosis, and 12 examined exacerbationsafter diagnosis.
|Study||No of participants||Mean age||Female (%)||Design||Length of follow up||Exacerbation criterion||Stress measure|
|Warren, 198215||200 (100 with MS, 100 medical controls)||N/A||56.5||Case-control||—||Neurologist confirmation||Modification of validated checklist|
|Franklin, 198816||55 with MS||34.9||88.9||Prospective||Mean 20 months||Neurologist confirmation||Validated interview|
|Grant, 198917||79 (39 with MS, 40 healthy controls)||35.65||74.4||Case-control||—||Neurologist confirmation||Validated interview|
|Warren, 199118||190 with MS (95 with exacerbation, 95 in remission)||35.0||56.5||Case-control||—||Neurologist confirmation||Validated checklist|
|Gaiatto, 199219||42 with MS (20 with exacerbation, 22 stable)||49.5||70.0||Case-control||—||Neurologist confirmation||Validated checklist|
|Nisipeanu, 199320||32 with MS||38.2||56.3||Prospective||26 months||Neurologist confirmation||Single traumatic event|
|Stip, 199421||68 (35 with MS, 33 medical controls)||42.5||60.0||Case-control||—||Neurologist confirmation||Validated checklist|
|Morrison, 1994212||17 with MS||73.4||73.4||Prospective||Mean 24 months||0.5 increase on EDSS||Modification of validated checklist|
|Gasperini, 199523||178 with MS (89 with exacerbation, 89 stable controls)||35.6||69.7||Case-control||—||1.0 increase on EDSS||Unvalidated questionnaire|
|Sibley, 199724||304 (170 with MS, 134 healthy controls)||41.5||62.5||Prospective||Mean 5.2 years||Neurologist confirmation||Unvalidated interview|
|Palumbo, 199825||92 (65 with MS, 27 medical control)||45.0||61.5||Case-control||—||Neurologist confirmation||Unvalidated questionnaire|
|Mohr, 200026||36 with MS||44.0||61.1||Prospective||Mean 10 months||1.0 increase on EDSS||Modification of validated checklist|
|Ackerman, 200327||50 with MS||39.4||100||Prospective||12 months||Patient report and retrospective neurologist confirmation||Validated/interview|
|Buljevac, 200328||73 with RRMS||39.9||76.7||Prospective||Mean 17 months||Neurologist confirmation||Weekly diary|
|EDDS = expanded disability status score; N/A=not available; RRMS= relapsing-remitting multiple sclerosis.|
Characteristics of patients with multiple sclerosis (MS), study design, and measures of stress for primary studies included in meta-analysis
The figure shows the mean effect sizes and standard deviationsfor each study. The primary analysis for the main hypothesisfound that the weighted average effect sizes for the impactof stress on exacerbation over the 14 included studies was d= 0.53 (95% confidence interval 0.40 to 0.65), P < 0.0001.The studies' effects sizes were homogenous (Q = 16.62, P = 0.22),with a low degree of inconsistency (I2 = 21.8%).
Effect of stress on exacerbations in multiple sclerosis
Effect sizes were not significantly affected by any study designcharacteristics, including the use of longitudinal prospectivev case-control designs (P = 0.31), the use of first exacerbationv exacerbation after diagnosis as an outcome criterion (P =0.59), the use of validated v unvalidated assessment of stress(P = 0.12), the use of self report v structured interview inthe assessment of stress (P = 0.12), and the inclusion of relapsing-remittingmultiple sclerosis only or relapsing-remitting multiple sclerosisand secondary-progressive multiple sclerosis (P = 0.11). Regardingsample characteristics, neither the proportions of female participants(P = 0.11) nor age of participants (P = 0.89) were significantlyrelated to effect size. None of the methods of determining exacerbationwas related to outcomes, including the use of the expanded disabilitystatus score v neurologist judgment (P = 0.45), the use of expandeddisability status score >= 1.0, or the use of retrospectiveneurologist verification with standardised data27 (P = 0.41).
On the basis of this meta-analysis of 14 empirical studies,our results support the hypothesis that stress is related toexacerbation of multiple sclerosis, with a weighted averageeffect size of d = 0.53. The effect sizes were statisticallyhomogenous, indicating that these studies are comparable. Differencesin study design, methods, or sample characteristics did notsignificantly influence the outcomes.
This effect size is clinically meaningful. As a comparison,a recent meta-analysis of the effects of interferon beta, theprincipal class of disease modifying drug used to treat multiplesclerosis, showed an overall effect of d = 0.36 in reducingexacerbations in the first year and d = 0.30 over the firsttwo years of treatment.29 This comparison is not meant to questionthe use of interferon beta as the underlying biological mechanismsof interferon beta and stress are not necessarily related toeach other. Rather, we suggest that the negative effects ofstress on exacerbation of multiple sclerosis are at least asgreat as the positive effects of a class of drugs widely consideredto produce clinically meaningful results.
Recent neuroimaging data have provided substantial weight tothe clinical evidence on which this meta-analysis was based.The review by Goodin et al suggested that a longitudinal prospectivestudy of stress in multiple sclerosis with gadolinium enhancingmagnetic resonance imaging (Gd+MRI) as the primary outcome would"convincingly test the relationship."3 Gd+MRI can detect breakdownin the blood-brain barrier and is therefore a highly sensitivemarker of lesion formation in multiple sclerosis and is highlypredictive of subsequent clinical exacerbation. Subsequent tothe Goodin et al review, a prospective study of 36 patientswith multiple sclerosis receiving monthly gadolinium enhancingmagnetic resonance imaging showed that the occurrence of interpersonalstressors was associated with a significantly increased riskof a new brain lesion eight weeks later.25
Our meta-analysis is an improvement over the previous qualitativereview, which included a study by Rabins et al.5 We did notinclude this because the data and statistics provided were notsufficient to calculate an effect size. Our meta-analysis alsocontained seven studies that the Goodin et al review omitted.18212224-26
Is all stress the same?
While our findings were statistically homogenous, the studyby Nisipeanu and Korczyn, in contrast to all other publishedreports, found that stress reduced the risk of exacerbation.19These findings raise the important hypothesis that differenttypes of stressors may have different effects. While all otherstudies examined normal everyday stress, Nisipeanu and Korczynexamined the effects of a traumatic, life threatening stressor—namely,being under one month of missile attacks in Tel Aviv duringthe first Gulf war. The finding that traumatic stressors reducedthe risk of exacerbations is consistent with animal models andother biological data. Numerous studies of stress in experimentalautoimmune encephalomyelitis, an animal model of multiple sclerosis,have shown significant reductions in symptoms related to stress.30Stress is known to increase the release of cortisol, and cortisolis known to be a potent anti-inflammatory hormone.31 However,more moderate stressors have been shown to activate experimentalautoimmune encephalomyelitis.32
At least one study has found that reducing distress in peoplewith multiple sclerosis can reduce T cell production of g interferon,33 a proinflammatory cytokine believed to be vital in the pathogenesisof exacerbation.2 However, no biological mechanism linking stressor distress and inflammatory processes in multiple sclerosishas been tested. Animal studies have suggested several potentialmechanisms. Small increases of cortisol concentrations, similarto concentrations seen in non-traumatic stress, have been shownto enhance the sensitivity of T cells to a number of cytokinesand peptides that promote a proinflammatory response.34 Alternatively,sustained increases in cortisol concentration in response tochronic stress produce a counter-regulatory reduction in thenumber, binding capacity, and affinity of glucocorticoid receptorson immune cells,35 increasing risk of inflammation. Finally,mast cells, which reside in the endothelium, can be activatedby increases in corticotropin releasing factor related to stress.Activated mast cells increase the permeability of the blood-brainbarrier and increase inflammation through the release of tumournecrosis factor a, histamines, and tryptase.36 While these potentialmechanisms are intriguing, none has been adequately tested amongpatients with multiple sclerosis. The absence of a clear biologicalmodel is a substantial weakness in the current literature.
Limitations and recommendations
Our meta-analysis had several limitations. The quality of thestudies included varied. Even the best longitudinal prospectivedesigns, which permit stronger inferences than case-controlstudies, do not offer absolute evidence of a causal association.Other unmeasured factors may affect both the perception of stressand exacerbation. For example, changes in normal appearing whitematter may occur months before traditional neuroimaging markersof inflammation or clinical exacerbation.37 Thus, we cannotrule out the hypothesis that decreased ability to manage stressorsor increased perceived stress may be an early marker of changesin normal appearing white matter.
On the basis of our review, we can recommend issues for futureresearch. Because distress resulting from stressful life eventscan increase the perceived severity of symptoms, objective measuresof exacerbation or inflammation are critical. To examine thepotential differential effects of various types of stressors,various dimensions of stressful life events must be carefullyassessed, including severity, chronicity, type, and source.Variability in individual reactions to stressful events, includingcognitive appraisal and subjective distress, will help to teaseapart the effects of environment and individual psychologicaldifferences. The development and evaluation of potential biologicalpathways has received little attention. Finally, the associationbetween stress and exacerbation in multiple sclerosis can beconclusively confirmed only with a clinical trial of a behaviouralintervention that teaches patients to reduce the occurrenceand impact of stress.
In summary, it is important to note that while these findingsshow a
significant association between stress and exacerbationin multiple sclerosis,
the effect size is modest. This associationis not consistent across patients
or even within individualpatients across time. The potential differential
effects ofvarious types of stress or the mechanisms by which stress affectsinflammation
are not known. Thus, the occurrence of any specificexacerbation cannot
yet be linked to any specific stressor.Furthermore, these findings should
in no way be misconstruedto suggest that patients with multiple sclerosis
bear any responsibilityfor exacerbations. Rather, we hope that these findings
willopen investigation into new avenues of managing multiple sclerosis,either
through stress management or through pharmacologicalmanagement of potential
neuroendocrine or immune responses tostress.
What is already known on this topic
This is Version 2 of the paper. In this version, there has beena change in the table to the entry for Gasperini under the heading"Stress measure." Where it read "1.0 increase in EDSS" thishas been corrected to read "Unvalidated questionnaire."
We thank Irina Fonareva for assisting in the preparation ofthe manuscript and presentation of the data.
Contributors: DCM initially conceived the paper and is guarantor.All authors participated in data acquisition, analysis, andwriting.
Funding: US National Institutes of Health grants R01 MH59708-01and R01 HD043323-01 and National Multiple Sclerosis Societygrants FG 1481-A1 and FG 1376A1.
Competing interests: None declared.
Ethical approval: Not required.
(Accepted 22 January 2004)
Copyright © 2004, BMJ