Expert advisory group
We invited practitioner experts in the area of the TW who approached us following our contact with Lapidus and/or publicity following the awarding of the grant to contribute to the project. On invitation to join the project we were unaware of the techniques of TW that they were employing. Although they were all working in different fields, and with slightly different techniques and approaches, they were all practitioners of facilitated TW. Indeed, we were unable to identify any UK-based practitioners of clinically-based unfacilitated TW to invite to join us as advisors. However, one of our authors, CM, had previously conducted a trial of unfacilitated TW.60 The practitioner experts were invited to collaborate during all phases of this project in the role of advisors, in order to inform our understanding of the range of TW interventions and to help reach consensus within the Steering Group Committee (SGC).
Search strategy
All electronic and hand-searches were conducted up to March 2013 by the lead researcher (OPN) in collaboration with a librarian (JB). A mapping search was performed in order to determine the extent of relevant literature (looking for both qualitative and quantitative studies). From the list of studies, appropriately includable studies for the systematic review were selected according to the selection criteria. A single electronic search was performed for both the mapping search and the systematic reviews of effectiveness and economic studies. A further search was conducted by CM in January 2015 to cover the 2 years since the previous search.
Search engines
Studies were systematically identified by searching a total of 22 electronic medical and psychological electronic databases: MEDLINE, EMBASE, PsycINFO, CAB Abstracts, Physiotherapy Evidence Database (PEDro), Published International Literature on Traumatic Stress (PILOTS), The British Library’s Electronic Table of Contents (Zetoc), Science Citation Index (SCI), Social Sciences Citation Index (SSCI), Linguistics and Language Behaviour Abstracts, Periodicals Index Online, Applied Social Sciences Index and Abstracts (ASSIA), Education Resources Information Center (ERIC), Allied and Complementary Medicine Database (AMED), Cumulative Index to Nursing and Allied Health Literature (CINAHL), Cochrane Central Register of Controlled Trials (CENTRAL) and Database of Abstracts of Reviews of Effects (DARE) for primary studies, and NHS Economic Evaluation Database (NHS EED) for economic studies. Grey literature was searched because of the possibility that effect size estimates might have been overestimated owing to selective reporting bias and unpublished studies are known to be less likely to have statistically significant results compared with published studies.61 Information on studies in progress and unpublished research or research reported in the grey literature was also sought by searching relevant databases including the Inside Conferences, Open System for Information on Grey Literature in Europe, Dissertation Abstracts, Current Controlled Trials database and ClinicalTrials.gov, Cochrane Database of Systematic Reviews (CDSR), Health Technology Assessment (HTA) Database, and The Campbell Library was searched for systematic reviews and economic evaluations. In addition, internet searches were also carried out using a specialist search gateway (OMNI), general search engine (Google) and a meta-search engine (ReadCube).The search was first conducted in Ovid for MEDLINE and MEDLINE In-Process & Other Non-Indexed Citations and then translated into the other databases. Similarly, the search in The Cochrane Library retrieved papers from the CDSR, Cochrane DARE and NHS EED.
The update searches (January 2015) included only the databases that had found all of the relevant citations in the previous search (i.e. MEDLINE, EMBASE, PsycINFO, Web of Science and Social SSCI, CINAHL and The Cochrane Library databases).
Additional searches and cross-referencing
Reference lists of included studies and previous reviews of emotional disclosure were screened. Experts in the area were contacted to identify additional unpublished literature. All studies previously included in former systematic reviews were searched for, screened against the inclusion criteria and considered for inclusion in current systematic review.
Search terms
Medical subject headings together with key words and controlled vocabulary were combined to capture two components of the review question: the populations and the interventions of interest. The search was limited to humans and there were no restrictions regarding the study design. The search terms used for each of the databases are listed in Appendix 3. In the update searches (January 2015) sensitive searches using very wide search terms (writing, writ*, etc.) were used in order to find all relevant studies that had been recently published.
The population of interest: long-term conditions
No definite list of LTCs was pre-established, as the potential range of diseases of interest was both extensive and diverse and made it difficult to create an exhaustive list. For the purposes of the search strategy and subsequent steps of the review, the UK DH definition of a LTC was adopted.33 The definition states: ‘Long term conditions are those conditions that cannot, at present, be cured, but can be controlled by medication and other therapies. They include diabetes, asthma, and chronic obstructive pulmonary disease’. Where it was unclear whether or not a condition met criteria discussion was held with the SGC and consensus reached (, column 1).
Long-term conditions discussed and considered for the review
For inclusion, populations had to have received a clinical diagnosis of the condition. In some studies participants had symptoms of LTCs (e.g. student populations) but no formal evidence of a clinical diagnosis and study participants did not report having been diagnosed with these conditions. For such studies the full text was scrutinised before making a decision regarding inclusion. The chronic conditions in these studies included anxiety, chronic stress, closed head injury, depression (usually stated as symptoms of depression), insomnia or poor sleep, migraine or tension headache, and suicidality.
The authors also discussed whether to consider some diagnosed conditions as LTCs, for example newly diagnosed cancer, or whether congenital conditions might be seen by some to reflect a continuum of normality (e.g. congenital deafness). It was decided to include these conditions but to analyse them separately. These are listed below (see , column 3).
It was decided to include all other cancer studies because patients may receive palliative care for prolonged periods, and terminally ill patients in hospices may still be receiving active treatment. Thus the distinction between active treatment and palliation might be difficult to distinguish and, furthermore, disease trajectories are not always predictable. There is a debate around whether or not obesity in the absence of any comorbidity is a disease;62 therefore, studies in people with uncomplicated overweight and obesity were excluded. Studies of addictive conditions (alcohol, smoking, illegal drugs, legal drugs) and learning disability were also included because the results could be useful to the NHS, although these might not meet the current definition of LTC. The following conditions were excluded:
personality traits, such as alexithymia, body dissatisfaction
people who had undergone stressful life events, such as bereavement, domestic violence, child sex abuse (unless PTSD diagnosed)
people found to be at increased risk of developing a LTC.
The intervention of interest: therapeutic writing
Prior to the drafting of the definitive list of search terms used to develop the search strategy, consensus within the SGC was reached on the relevant and appropriate terms related to TW interventions. The terms under debate had been identified as part of the mapping search. The aim was to capture the published literature related to the different types of TW interventions; therefore the main key terms referring to TW were defined, discussed, agreed and validated with the expert advice (, column 1).
Therapeutic writing key words and variants discussed for the search strategy
Some terms were not considered because:
they had been already identified with a more common synonym thought to be equivalent (e.g. writing for healing/writing to cure vs. wellness writing)
the focus of the writing was thought not to be therapeutic (e.g. written divulgation, written exposé or written material/information).
Selection of papers
Results from the electronic searches (titles and abstracts where available) were transferred into a spreadsheet using Microsoft Excel® 2010 (Microsoft Corporation, Redmond, WA, USA) following automatic de-duplication within the citation manager EndNote X 4.02 (Thomson Reuters, CA, USA) and the manual removal of other duplicates.
Peer-reviewed articles and non-peer-reviewed papers (e.g. conference abstracts and dissertations) were then selected for potential inclusion in a two-stage process by one reviewer (OPN), with a random 10% selection of citations independently checked by a second reviewer (LB and CM). The two reviewers independently selected studies that met the predefined inclusion criteria. Disagreements were resolved by discussion and/or arbitration involving a third reviewer or by the full team, depending on the complexity of the issues. When it was not possible to determine the study eligibility by title and abstract alone the full text was retrieved for assessment. Authors of conference abstracts were contacted for full articles.
During the selection of studies, the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) approach was used.63 A diagram was developed to show the numbers of studies in the different categories and the reasons for exclusion of full-text studies. Additionally, an expert in the field of TW was shown the list of included studies to check whether or not they believed that all relevant papers had been identified.
Inclusion criteria
Studies meeting the following inclusion criteria were included in the systematic review:
Studies assessing participants with at least one LTC as per DH definition.
33Studies assessing any form of TW including emotional disclosure/expressive writing, poetry, diaries, etc., with inactive comparators or comparison groups thought to be inactive:
For example, if participants in the control arm were directed to write in a non-emotional way by using, for instance, neutral topics or by describing facts or how they managed their time. When the control group wrote about topics related to their illness or treatment, or when arousal of emotions might occur, these were not considered to be inactive controls and these studies were excluded. However, if descriptions of the comparators used were not provided, the paper was not be excluded on this basis alone.
Studies reporting any relevant clinical outcomes including both disease-specific outcomes and generic outcomes.
Outcomes related to physical (including physiological, haematological/immunological outcomes, pain, or disability), psychological, social and behavioural health. Performance, health-related quality of life (HRQoL), as well as participant mental status, satisfaction and both intervention safety and compliance to treatment for the LTC, were of interest. Resource-use or cost data were also collected for economic consideration. Outcomes could be self-reported or evaluated by a clinician or a carer.
If the study could report relevant outcomes without reporting any usable numerical data. In this particular case, the authors were contacted for unpublished data.
Full versions of prospective randomised or non-randomised trials or observational studies having any form of comparison group including, for instance, RCTs/non-RCTs, cohort, case–control studies and economic evaluations.
Exclusion criteria
The following were excluded:
Studies including participants with acute conditions, stress, bereavement or any acute event.
Studies assessing any form of psychotherapy, counselling, talking to a listener, talking into a tape recorder, mobile phone or similar, where this was the primary mode of delivering the intervention, expressive drama, dance or film-making.
Any study that evaluated other people’s writing.
Any study that evaluated any type of writing as a diagnostic tool instead of as a therapeutic tool in the course of a disease treatment (e.g. patients with agraphia).
A comparative study with any active or probably active control including any form of TW or talking into a tape recorder or mobile telephone.
Studies assessing only intermediate physiological outcomes such as salivary cortisol, immune parameters not routinely measured in the management of LTCs or studies not reporting relevant numerical and usable data and/or where unpublished data could not be obtained.
Inappropriate study design for this review: single case reports, case series (as both have no comparator arm) and studies where results for intervention and control groups were not presented separately. Studies only available in brief abstract form.
Data collection
The forms used for the data extraction are shown in Appendix 4.
Data extraction methods
Study findings were extracted and entered into a spreadsheet by one reviewer (OPN) and checked by a second reviewer (CM, ST, AH, LS and LB) working independently. A purpose-built data extraction form (Excel) was developed and piloted prior to data collection. Any disagreements were resolved by consensus and/or arbitration involving a third reviewer. Missing information was obtained from investigators if it was crucial to subsequent analysis (this was not possible for the five studies identified in the updated search to January 2015). The software GetData Graph Digitizer version 2.26.0.20 (GETDATA Graph Digitizer, Moscow, Russia) was used when numerical data had to be derived from graphs. To avoid introducing bias, unpublished information was coded in the same fashion as published information.
Quality assessment methods
Quality of studies was assessed based on accepted contemporary standards including the Newcastle–Ottawa Scale for case–control studies.64 The Cochrane risk-of-bias tool65 was used for RCTs and quasi-randomised trials. Risk of bias was qualified as high, low or unclear. The first assessment was performed by one reviewer (OPN) on all studies, with a second reviewer (CM, LB, LS, SJCT) independently checking each study.
Data analysis
Synthesis of data
In order to collate, combine and summarise the information from the included studies, narrative and quantitative (meta-analysis) approaches were undertaken. After all included studies were identified, the SGC discussed organisation of the data for analysis. It became clear that the studies fell into two distinct categories: those that were facilitated (such as interactive biblio/poetry therapy) and those that were not (such as unfacilitated TW models and its various elaborations). Discussion with our expert practitioners and consideration of the literature revealed that facilitated and unfacilitated writing interventions are fundamentally different. As explained by the practitioner experts, facilitated TW interventions consist of one or more interactive activities (including TW) between the group facilitator and participant, which allows a live, in-person communication and an element of quality control and tailoring. Usually, the facilitator is in the same room as the participant, and may help with any unexpected concern and/or guide the participant in the usual process of the intervention. For unfacilitated writing, studies were categorised by ICD-10 (International Classification of Diseases, Tenth Edition) code according to the LTC assessed (see Chapter 3).
Analysis of studies
The numerical results from each of the included studies were checked to identify possible data entry problems. For each study, for continuous measures, either the mean and standard deviation (SD) at the various follow-ups, or any other statistic that could be used to calculate SD, such as the standard error (SE), were extracted for further analysis. For categorical measures, dichotomous or binary data, or counts and rates calculated from the number of events that each individual experienced, were collected.
Meta-analyses
Pooled-effect estimations were conducted using the standard software package Review Manager 5.2.6 (RevMan 2012, The Cochrane Collaboration, The Nordic Cochrane Centre, Copenhagen, Denmark). Analyses were stratified according to the type of outcome measured. A comparison was performed when at least three studies used the same (or a similar) instrument to assess the same (or similar) aspects of a given outcome and when sufficient numerical data were reported. In the case of different follow-up periods, results were combined using threshold intervals of ‘immediate’, ‘short term’, ‘medium term’ and ‘long term’, as shown in .
Thresholds intervals for follow-up combinations in meta-analysis
In cases where two studies reported short-term follow-up, and one study reported an immediate follow-up assessment, the studies were meta-analysed and combined into a short-term follow-up comparison. For continuous outcomes, standardised mean differences (SMDs) were used when outcomes were measured with different instruments. Random-effects models were used because of clinical heterogeneity. Statistical heterogeneity of results between studies was assessed using the I2-value. Conclusions regarding the estimates of effect sizes were interpreted cautiously if there was significant heterogeneity.
Unit of analysis issues
Only comparative studies were included. Participants were usually randomised to one group of two groups; however, some studies could compare one experimental TW intervention group against both a standard intervention (such as standard care) and one with placebo writing. Alternatively, two or more experimental interventions could be tested against a standard intervention (or with both a standard intervention and with placebo writing), giving a four-arm trial. There could be also trials with the same outcome assessed at different time points or just measured after the writing session or at the end of the treatment period.
For the systematic review, the interest lay only in the direct comparison between a TW intervention arm and an inactive comparator. If a trial had two intervention groups and two control groups, for meta-analysis the trial was treated as two separate trials one comparing, for instance, a more brief TW intervention against inactive comparator and one comparing a longer TW intervention against inactive comparator. Where there were two intervention groups and one control group, the TW intervention most widely used (i.e. the unfacilitated EW with the standard instructions was the one included in the meta-analysis).
Results across LTCs
Four analyses were performed:
physiological, disease-related and biomarker outcomes (results tabulated)
positive writing across LTCs (results tabulated)
depression (results tabulated and meta-analysed where possible)
anxiety (results tabulated and meta-analysed where possible).
