BACKGROUND

One of the challenges of applying evidence-based practice is the expectation that practitioners will use the most current, relevant research to guide clinical decisions (1).  The methodology for conducting a systematic review of the research-base may not be a routine practice in daily clinical service delivery.  The need to search and critically appraise the evidence, especially for questions that relate to the effectiveness of treatment is expected. 
Word Prediction is a strategy when combined with spelling supports text-entry using an augmentative and alternative communication (AAC) system or computer (2).  Word prediction has been promoted to reduce keystrokes, enhance rate, and improve the quality of written output for individuals with motor and learning disabilities (3).  However, research studies have reported various results on the effectiveness of word prediction related to these outcomes.  Practitioners must make decisions about the benefits of word prediction for specific clients or to meet specific individual abilities and needs.

STATEMENT OF PROBLEM

Questions about the effectiveness of word prediction as a text-entry strategy for AAC and computer access have been asked for over two decades.   However, a comprehensive appraisal of the research-base has not been published.  Clinicians are interested in knowing not only whether word prediction would benefit clients, but specifically what performance has been achieved using word prediction as a text-entry strategy.

METHODOLOGY

 The first procedure involved a multifaceted search based on the review strategy used by Schlosser and Wendt (2008) (4).   Various bibliographic and dissertation databases, manual searches, and ancestry searches were conducted to identify studies written between 1980 and October 2008.  The following electronic databases were searched between November and December, 2008:  Association for Computing Machinery digital library (ACM), EBSCO AAC Journal, EBSCO ERIC, EBSCO Medline, ERIC, Google Scholar, Ovid Medline, Proquest, and PsychINFO.  Multiple searches and advanced searches were conducted using the following key words:  word prediction, text prediction, message prediction, text-entry and prediction, predicting and text-entry.  Limits on the searches were  English, human, and1980-2008.  AAC and/or computer access were not used as search terms, since that may have restricted or expanded the results significantly for our purposes.  

The second procedure was performed by the third author and involved compiling the search articles by title and author, eliminating duplicate results, and locating the abstracts for 144 studies related to word prediction.  Two reviewers were assigned to read the titles and abstracts to confirm that the study was an experiment on word prediction with original data and human subjects.  Inter-judge agreement was used to establish reliability for this process that resulted in 100% agreement for the 37 studies meeting initial inclusion criteria.

For the third procedure a 10 item inclusion checklist was designed to identify articles specific to reporting the effectiveness of word prediction as a treatment.  Three reviewers, two reviewers for each article, were randomly assigned to complete the checklist.  Inter-rater reliability of 93% (range 90-96%) was achieved.  The inter-judge agreement method was used to handle any discrepancies.  The studies identified from this process will undergo a final critical appraisal.  A coding manual consisting of 8 parts of over 50 items was developed for appraising each article.  This process will provide an overview of the effectiveness of word prediction as a clinical treatment or text-entry strategy for AAC and computer access.

RESULTS

Table 1 represents the results from the bibliographic database searches.  The largest results were obtained from the Google Scholar search (n=846).  Proquest Digital had the fewest results, but several items were found in other databases as published manuscripts of a thesis or dissertation.  Medline and PsychINFO had similar results (n=51).  Considerable overlap was found with ERIC (n=102), and original results tended to be related to curriculum and learning questions regarding word prediction.

Table 1. Results of database searches on word prediction.

DATABASE/SEARCH ENGINE/JOURNAL	SEARCH TERMS	CRITERIA	TOTAL NUMBER OF PAPERS FOUND
Medline	Word prediction, Message prediction, Text entry & prediction, Predicting & text entry	keyword, all text, title, abstract	51
Proquest Digital Dissertations	Word prediction, Message prediction, Text entry & prediction, Predicting & text entry	Keyword, title, Abstract	38
Eric	Word prediction, Message prediction, Text entry & prediction, Predicting & text entry	keyword, all text, title, abstract	102
PsychINFO	Word prediction, Message prediction, Text entry & prediction, Predicting & text entry	Keyword, title	51
ACM	Word prediction, Message prediction, Text entry & prediction, Predicting & text entry	Keyword, all text title, abstract	454
Google Scholar	Word prediction, Message prediction, Text entry & prediction, Predicting & text entry	Full phase	846

DISCUSSION

Conducting such a rigorous, multifaceted search for external evidence used considerable resources (personnel and time).  However, the results provide a comprehensive overview of the types of research conducted on word prediction between 1980 through 2008.  The selected electronic databases had considerable overlap in found publications based on the keywords and limits.  Although Google Scholar produced the largest number of results, many of the results were not appropriate or useable.  Use of this search engine takes more time to reduce the results and, therefore, time away from reviewing and appraising.  Ovid Medline and PsychINFO had the largest number of results that were selected for the second procedure review using the criteria checklist.
            The authors agree that a criteria list needs to be identified prior to searching to ensure a principled search and inclusion process.  We asked a broad question about word prediction in order to identify the variety of independent (treatments) and dependent (measured data) variables reported for participants with a variety of disabilities in the literature.  This will allow us to compare the effectiveness of word prediction based on a range of variables.  Individual practitioners should be able to conduct advanced searches with additional keywords that will result in studies hopefully more specific to a client.
            Finally, our review at this stage suggests that only a small number of studies (n=< 5) will meet final criteria for appraising the effectiveness of word prediction on a single dependent variable such as rate or accuracy.  Reliability from the inclusion criteria review form showed that few studies reported quantitative data on participants.  When data were available effectiveness was either not reported or could not be calculated using the published data points.  Consequently, study results would not answer questions of effectiveness conclusively.

CONCLUSION

This paper reports the principled approach used to conduct a systematic review of the evidence related to word prediction.  Tools and procedures developed in the lab can be used by practitioners for a more time efficient, productive search and appraisal of topics of clinical importance.  The final results of our effort will provide a critical appraisal on the effectiveness of word prediction as a strategy for AAC and computer access.
 Strategies for searching and appraising the external evidence need to be learned and practiced to make the task of identifying the best evidence more labor efficient.  The benefits to identifying the treatment designs and effectiveness of treatment identified by appraising the evidence do not have a price tag.  Clinical confidence in the decisions made about strategies and methods used to improve text-entry for AAC and computer access influences performance, user satisfaction, and overall outcomes.