Background
-
Exclusion of irrelevant references (i.e. references not fulfilling the eligibility criteria) through perusal of the titles, and, if available, the abstracts. If doubts exist as to the relevance of a study, the corresponding full text is obtained.
-
The full texts of the potentially relevant publications are obtained. The decision on the inclusion of the study is then made on the basis of these full texts.
Methods
Information sources and literature search
Eligibility criteria and selection of evaluations
Edwards 2002 | Doust 2005 | Pham 2016 | Shemilt 2016 | |
---|---|---|---|---|
Aim of evaluation | Estimate the accuracy and reliability of reviewers when screening records for relevant trials for a systematic review | Assess the reliability and accuracy of reviewers’ screening | Assess the implications of applying methodological shortcuts: one of the shortcuts is a single screening approach | Compare the costs and effects of a single screening approach. |
Other aims/comparisons included in the evaluation | – | The other aim was to assess the sensitivity and precision of five published search strategies | The other shortcuts were: - one bibliographic database plus ancillary sources - limiting the search to bibliographic databases - only papers available electronically | Four variant screening approaches were analysed. The other approaches were: - safety first screening - double screening - single screening with text mining |
Number of - reviews examined | - 1 | - 2 | - 3 | - 1 |
- reviewers involved | - 4 | - 2 | - 2 | - 1 |
- sets of screenings examined | - 6* | - 2 | - 3 | - 1 |
- individual screenings analysed | - 12 | - 4 | - 6 | - 1 |
Screening step assessed | Title/abstract screening only | Title/abstract screening only | Title/abstract screening only | Title/abstract screening only |
Piloting screening | Prior meeting to discuss inclusion criteria | No information provided | Pre-test screening of 50 potentially relevant records | No information provided |
Gold standard | Studies identified as relevant by a double screening approach.Disgreements were resolved by consensus | Studies identified as relevant by a double screening approach. Disagreements were resolved by consensus. | Original reviews’ study pool. No further information provided. | Studies identified as relevant by a double screening approach. Disagreements were resolved by consensus. |
Number of hits needed to be screened | 22,571 hits, (each reviewer was to screen approximately 11,286 records) | Tympanometry: 638 hits Natriuretic peptides: 373 hits | Wilhelm 2011: 1890 hits Greig 2012: 3091 hits Bucher 2015: 690 hits | 12,477 hits |
Study type(s) included in the review | RCTs | Diagnostic test accuracy studies | Wilhelm 2011: n/a; no limitations Greig 2012: limited to experimental (control and challenge trials), quasi-experimental (before-and-after trials), cohort designs Bucher 2015: no limitations on study designs | Studies of any design, using quantitative and/or qualitative methods |
Reviewer experience as reported in the evaluation | “Each reviewer had substantial experience of screening records for systematic reviews, apart from reviewer 2 who was relatively inexperienced.” | “The first reviewer had more content knowledge and more experience in completing systematic reviews” [...] | “Reviewer A was a veterinarian, had a master’s degree in epidemiology and had over 5 years of experience in relevance screening for reviews in agri-food public health. Reviewer B had a master’s degree in public health and over 2 years of experience in relevance screening for reviews in agri-food public health”. | […] “conducted by an experienced team of systematic reviewers with substantial experience in primary care and medical education” |
Reviewer experience: classification | Reviewer 1: experienced Reviewer 2: less experienced Reviewer 3: experienced Reviewer 4: experienced | Reviewer 1: experienced Reviewer 2: less experienced | Reviewer 1: experienced Reviewer 2: less experienced | Reviewer 1: experienced |
Re-analysis meta-analysis | No | No | Yes | Yes |
Number of missed records presented | Yes | Yes | Yes | Yes |
Data extraction and synthesis
Results
Median proportion missed | Sets of screenings | Min. in % | Max. in % | |
---|---|---|---|---|
Overall result | 5% | 23 | 0% | 57.8% |
Edwards 2002 | 5.7% | 12 | 0% | 24.1% |
Doust 2005 | 1.5% | 4 | 0% | 3.0% |
Pham 2016 | 16.6% | 6 | 0% | 57.8% |
Shemilt 2016 | 0.6% | 1 |
–
|
–
|
Evaluation | Reviewer | Review or set of screenings | Number of missed studies | Number of included studies (gold standard) | Proportion of missing studies | Results of the re-analysis of the meta-analysiszz |
---|---|---|---|---|---|---|
Edwards 2002 | 1 | Set A | 0 | 22 | 0% | n.a. |
1 | Set B | 1 | 30 | 3% | n.a. | |
1 | Set C | 3 | 31 | 10% | n.a. | |
2** | Set A | 2 | 22 | 9% | n.a. | |
2** | Set D | 2 | 20 | 10% | n.a. | |
2** | Set E | 7 | 29 | 24% | n.a. | |
3 | Set B | 1 | 30 | 3% | n.a. | |
3 | Set D | 1 | 20 | 5% | n.a. | |
3 | Set F | 5 | 24 | 21% | n.a. | |
4 | Set C | 2 | 31 | 6% | n.a. | |
4 | Set E | 0 | 29 | 0% | n.a. | |
4 | Set F | 1 | 24 | 4% | n.a. | |
Doust 2005 | 1 | Tympanometry | 1 | 33 | 3% | n.a. |
2** | Tympanometry | 1 | 33 | 3% | n.a. | |
1 | Natriuretic peptides | 0 | 20 | 0% | n.a. | |
2** | Natriuretic peptides | 0 | 20 | 0% | n.a. | |
Pham 2016 | 1 | Wilhelm 2011 | 2 | 19 | 11% | Negligible impact on findings |
1 | Greig 2012 | 2 | 36 | 6% | Negligible impact on findings | |
1 | Bucher 2015 | 0 | 18 | 0% | No impact on findings | |
2** | Wilhelm 2011 | 11 | 19 | 58% | Substantial change in findings | |
2** | Greig 2012 | 7 | 36 | 19% | Substantial change in findings | |
2** | Bucher 2015 | 3 | 18 | 17% | Substantial change in findings | |
Shemilt 2016 | 1 | Park 2015 | 1 | 169 | 1% | Negligible impact on findings |
Overall Result |
53
|
733
|
Discussion
Reviewer experience
Research question too vague
Single screening approach as a methodological shortcut
Importance of bibliographic searches
Available evidence
Research gaps
Future research
-
the impact of reviewer experience and poorly described PICO on the number of missed studies,
-
the impact of missed studies on results of meta-analyses,
-
the impact of non-bibliographic information sources on the relevance of studies missed in bibliographic searches,
-
the impact of single full-text screening on sensitivity (vs. double screening)
-
the impact of training or piloting before starting screening
-
the impact of screening tools
-
the impact of the prioritization of references in combination with single or double screening (as analysed in Shemilt 2016).