Writing as therapy

Effects on immune mediated illness need substantiation in independent studies

Recently JAMA published a trial of a “get it off your chest” writing exercise.1 Seventy one patients with asthma or rheumatoid arthritis were randomised to write about the most stressful experience they had ever had or about their plans for the day for three separate 20 minute periods over a few days and then to drop their completed essay into a sealed box. The study apparently showed a significant improvement in standard measures of disease severity in both conditions four months later. An accompanying editorial exhorted readers to abandon the Cartesian split between mind and body, and acknowledge the growing evidence in support of behavioural interventions that reduce emotional stress as therapies for diseases that are mediated in part by the immune system.2 Do these results stand up, and is it therefore time to heed this call?

Therapeutic writing is a hot topic on both sides of the Atlantic. In the United Kingdom the focus tends to be on descriptive accounts and somewhat speculative psychodynamic explanations for subjective improvements in health status. A recent book provides moving case studies of patients who came to terms with physical or psychological illness through creative writing and offers several different options for promoting the use of the pen in the therapeutic encounter.3 In the United States, in contrast, the focus is on formal “scientific” research aimed at validating the impact of short, sharp, and highly standardised writing exercises on physical measures of illness. The emphasis of such research is on showing that measurable things change, even though we may not yet be able to explain why.

Pennebaker was one of several US psychologists who developed a standard writing task some years ago. He tested it extensively on college students and other healthy volunteers4 before popularising it in the lay press as a self help strategy for coping with stress.5 Subsequent work by several authors, summarised in a systematic review by Smyth,6 explored the effect of this and similar standard writing tasks on a wide range of variables in healthy volunteers. The variables were physiological (for example, skin conductance, helper and suppressor lymphocyte function, serum cortisol), psychological (wellbeing, social functioning, “adjustment”), and behavioural (grade point average, visits to the doctor). All 13 primary studies identified in the review showed a positive effect of the writing task on the chosen variable, but there was marked heterogeneity of effect size, suggesting that confounding factors were important in some studies. The possibility of publication bias was not fully explored, there was no convincing “dose-response” effect—that is, the impact of the writing task was not related to the number or length of sessions—and none of the subjects was ill to begin with.

In the JAMA study Smyth et al recruited patients with symptomatic asthma or rheumatoid arthritis through advertisements in newspapers and on clinic noticeboards. Potential participants (who were paid for their participation) were screened by telephone to confirm eligibility, establish commitment, and exclude those “deemed unable to comply with the protocol.” Of 465 people who called to express interest, 126 were randomised, of whom 14 (12 in the experimental group) withdrew before starting the study and another 5 (1 in the experimental group) withdrew before completing it. Participants supplied demographic data and underwent baseline investigations including a general quality of life score before being introduced to the writing exercise. Asthma severity was measured by forced expiratory volume in one second (FEV₁), in which a 15% change from baseline was taken as significant. Arthritis severity was assessed by a standard clinician ranked score (from 0 (asymptomatic) to 4 (very severe)) in which a significant change was defined as one scale point. Clinical assessors were not told the allocation of participants.

Overall nine of 43 controls and 33 of 83 in the experimental group were classed as having improved at four months. The study apparently had sufficient power to have a 90% chance of detecting a clinically significant improvement if one existed. The results at four months reached statistical significance at the P<0.001 level, and by my calculations they translate into a number needed to treat of about five for improvement in disease status (95% confidence intervals 3 to 36). Differences between the groups at intermediate periods of two weeks and two months were less impressive and not overall statistically significant.

What are we to make of these findings? It seems frankly implausible that a total of 60 minutes' writing on a subject unrelated to the disease should have a clinically significant impact on two different chronic diseases four months later. But if we are to reject the findings of a randomised controlled trial we should do so on the grounds of validity and generalisability, not on whether we believe the results. The potential biases in this study are not difficult to spot. The highly selected participants may well have encountered the intervention (and the popular expectations associated with it) in the media previously. Those given the neutral writing exercise probably guessed they were controls. The “objective” measures of disease severity are open to assessment bias, which may have had an influence if the subjects told the assessor what they had done in the writing task. Finally, the outcomes in the two different diseases should probably have been analysed separately rather than summed.

But perhaps my interpretation is biased by my own cultural prejudices. I have an instinctive empathy with Bolton's approach to writing therapy as an art rather than a science3 and a personal distaste for quick fix interventions that smack of pop psychology and have been marketed to the public through the same channels as Billy Graham and the F Plan diet. If others in the UK share this cynicism our scientific community is probably the ideal ground to attempt to replicate Smyth et al's study in an uncontaminated population that has no prior expectations of the intervention.