Are electronic diaries useful for symptoms research? A systematic review

Abstract

Objective

The objective of this study was to evaluate electronic diaries as a tool for investigating associations between physical symptoms and psychological variables.

Methods

A qualitative systematic review of electronic diary studies in pain and symptoms research was performed.

Results

Studies of electronic diary use report good acceptability and accuracy of data recording. A descriptive classification of the findings of electronic diary studies yielded five categories: experience, recording data when symptoms were present; interaction, testing the relationship between variables at different levels; sequential, observing associations at different time lags; process, recording possible mediating constructs such as catastrophization and self-efficacy; and intervention, recording during treatment to identify trends. Most data from diary studies suggest relatively weak associations between variables.

Conclusion

Electronic diaries are a potentially valuable tool for psychosomatic research. We suggest further possibilities for their use.

Introduction

Diary studies, in which patients record their symptoms or thoughts regularly, have been a feature of psychosomatic research for over 25 years [1]. Traditionally, they have been completed on paper and used to track the variation of a single symptom or interactions between symptoms and other variables such as mood [2] or stress [3].

Compared to conventional questionnaire studies, diaries are less prone to recall bias [4]. Recall bias is potentially a major confounder in symptoms research, as retrospective recall is affected by both current state and experiences since the recalled event [5]. Diary studies are associated with their own unique problems, and there have been doubts about both the accuracy and the timing of responses, particularly their susceptibility to being completed from memory after the stated time [3], [6].

Since the early 1990s, developments in handheld computer technology have offered an opportunity to exploit electronic diaries, into which users enter data via a touch screen and a stylus. User-friendliness is achieved by selecting options from on-screen lists or by completing Likert-type or visual analog scales (VAS). All entries are date stamped and time stamped, and data are stored in the device for later retrieval either by connection to another computer for download or by telephone. Electronic recording of symptom information has start-up costs higher than those of other methods (handheld devices typically range in price from 100 to 400 euros or US dollars and require additional modification and programming before use). However, a number of free and commercial software programs for developing handheld computer diaries are available [7], and principles of electronic diary design have been published [8]. Electronic diaries are being increasingly used in a wide range of clinical and research circumstances, both for condition monitoring and as a primary data collection method for pharmacological trials.

Traditionally, paper diaries have been completed once daily, typically toward the end of the day, but this procedure is prone to recall bias, with selective memory for both the most severe and the most recent events [4]. In recent years, Schwartz and Stone [9] have pioneered a technique known as ecological momentary assessment in which multiple data entries are requested each day and the subject is asked to record how one feels at that point in time. Such studies have been particularly valuable in addiction research, where moment-to-moment changes may have major implications for relapse [10]. To avoid repetition at set times of day, electronic diaries for this type of study are set to sound an alarm at semirandom intervals.

Electronic diary studies generate data with many repeated measures of multiple variables, typically with 50-150 time points. While simple monitoring of trends may be sufficient for descriptive studies of day-to-day variation, more complex study designs, such as those seeking associations or causal sequences, require particular caution in interpretation. Most time-series data from diary studies show marked autocorrelation [4], [9] whereby values depend on preceding values in the series. This violates the assumption of independence, which underlies parametric statistical methods such as multivariate regression [11].

Three main analytical approaches are used. In the first approach, data are treated to remove autocorrelation and trend, either by differencing or by the use of specific time-series modeling techniques such as autoregressive moving average models [3]. The second approach uses multilevel modeling (MLM) techniques in order to carry out analysis at both within-person and between-persons levels [11], [12]. Finally, there are specific true multivariate time-series analysis methods, which require both long data series and considerable statistical skills [13]. New techniques from nonlinear science have potential but have not yet been tested adequately [14].