Article type
Year
Abstract
Background: Searching for prognosis studies in systematic reviews (SRs) is challenging due to poor reporting and inconsistent indexing in databases. Efficient methods are needed to reduce search results that need to be screened due to high sensitivity, but poor precision. This will have an impact on the amount of time SRs take to complete and funding required.
Objective:
1. Present results of a SR of published search filters that aim to identify any type of prognosis study (overall prognosis, prognostic factors, prediction models).
2. Evaluate performance of most promising search filters in a prognostic SR and assess impact on screening burden for reviewers.
Methods: In the SR of search filters, studies reporting development and/or evaluation of search filters designed to retrieve any type of prognosis study were retrieved using sensitive search strategies. Study quality was assessed using a pre-defined appraisal tool. Filter characteristics and performance metrics reported were extracted and tabulated. To allow comparisons, filters were grouped according to database, platform, type of prognosis study, and type of filter (e.g. sensitive, specific or optimised).
Ten prognosis search filters with published sensitivity >95% were selected and individually combined with subject-specific terms in the search strategy for a SR of predictors of poor outcomes in frail older people. Proportions of included studies retrieved (recall) and Number Needed to Read (NNR) to find one relevant study was calculated for each search filter based on the final number of studies included in the SR. Metrics from the frailty SR will be compared to those published in development and evaluation studies.
Results: We identified 128 published search filters. These included multiple and single term filters and 105 of the filters were for searching Medline. Twenty-four filters, mostly from the Canadian Hedges Team aimed to retrieve any type of prognosis study, the remaining filters were for prediction models. There was huge variation in results for the ten highest performing search filters and across different studies reporting metrics (sensitivity 55-100%; specificity 80-97%). Filters with the highest sensitivity and specificity will be presented.
Adding the ten Medline filters to searches for the SR of predictors for poor outcomes in frail older people retrieved 13,492 unique references when combined with other database searches. Total numbers of references retrieved varied between search filters used from 2174 to 4630. Final decisions on included studies will be made during Spring 2020 and results will be presented at the Colloquium.
Conclusions: Recommendations for the most efficient search filters currently available will be shared with the Cochrane Prognosis Group and included in a course on prognosis SRs at Keele University.
Healthcare consumer involvement: Keele’s PPIE group members have contributed to the SR in frail older people to help identify potential predictors of poor outcomes that will be compared with those found in the research and gaps identified.
Objective:
1. Present results of a SR of published search filters that aim to identify any type of prognosis study (overall prognosis, prognostic factors, prediction models).
2. Evaluate performance of most promising search filters in a prognostic SR and assess impact on screening burden for reviewers.
Methods: In the SR of search filters, studies reporting development and/or evaluation of search filters designed to retrieve any type of prognosis study were retrieved using sensitive search strategies. Study quality was assessed using a pre-defined appraisal tool. Filter characteristics and performance metrics reported were extracted and tabulated. To allow comparisons, filters were grouped according to database, platform, type of prognosis study, and type of filter (e.g. sensitive, specific or optimised).
Ten prognosis search filters with published sensitivity >95% were selected and individually combined with subject-specific terms in the search strategy for a SR of predictors of poor outcomes in frail older people. Proportions of included studies retrieved (recall) and Number Needed to Read (NNR) to find one relevant study was calculated for each search filter based on the final number of studies included in the SR. Metrics from the frailty SR will be compared to those published in development and evaluation studies.
Results: We identified 128 published search filters. These included multiple and single term filters and 105 of the filters were for searching Medline. Twenty-four filters, mostly from the Canadian Hedges Team aimed to retrieve any type of prognosis study, the remaining filters were for prediction models. There was huge variation in results for the ten highest performing search filters and across different studies reporting metrics (sensitivity 55-100%; specificity 80-97%). Filters with the highest sensitivity and specificity will be presented.
Adding the ten Medline filters to searches for the SR of predictors for poor outcomes in frail older people retrieved 13,492 unique references when combined with other database searches. Total numbers of references retrieved varied between search filters used from 2174 to 4630. Final decisions on included studies will be made during Spring 2020 and results will be presented at the Colloquium.
Conclusions: Recommendations for the most efficient search filters currently available will be shared with the Cochrane Prognosis Group and included in a course on prognosis SRs at Keele University.
Healthcare consumer involvement: Keele’s PPIE group members have contributed to the SR in frail older people to help identify potential predictors of poor outcomes that will be compared with those found in the research and gaps identified.