Increased Workload for Systematic Review Literature Searches of Diagnostic Tests Compared With Treatments: Challenges and Opportunities
Bibliographic record
Abstract
BACKGROUND: Comprehensive literature searches are conducted over multiple medical databases in order to meet stringent quality standards for systematic reviews. These searches are often very laborious, with authors often manually screening thousands of articles. Information retrieval (IR) techniques have proven increasingly effective in improving the efficiency of this process. IR challenges for systematic reviews involve building classifiers using training data with very high class-imbalance, and meeting the requirement for near perfect recall on relevant studies. Traditionally, most systematic reviews have focused on questions relating to treatment. The last decade has seen a large increase in the number of systematic reviews of diagnostic test accuracy (DTA). OBJECTIVE: We aim to demonstrate that DTA reviews comprise an especially challenging subclass of systematic reviews with respect to the workload required for literature screening. We identify specific challenges for the application of IR to literature screening for DTA reviews, and identify potential directions for future research. METHODS: We hypothesize that IR for DTA reviews face three additional challenges, compared to systematic reviews of treatments. These include an increased class-imbalance, a broader definition of the target class, and relative inadequacy of available metadata (ie, medical subject headings (MeSH) terms for medical literature analysis and retrieval system online). Assuming these hypotheses to be true, we identify five manifestations when we compare literature searches of DTA versus treatment. These manifestations include: an increase in the average number of articles screened, and increase in the average number of full-text articles obtained, a decrease in the number of included studies as a percentage of full-text articles screened, a decrease in the number of included studies as a percentage of all articles screened, and a decrease in the number of full-text articles obtained as a percentage of all articles screened. As of July 12 2013, 13 published Cochrane DTA reviews were available and all were included. For each DTA review, we randomly selected 15 treatment reviews published by the corresponding Cochrane Review Group (N=195). We then statistically tested differences in these five hypotheses, for the DTA versus treatment reviews. RESULTS: Despite low statistical power caused by the small sample size for DTA reviews, strong (P<.01) or very strong (P<.001) evidence was obtained to support three of the five expected manifestations, with evidence for at least one manifestation of each hypothesis. The observed difference in effect sizes are substantial, demonstrating the practical difference in reviewer workload. CONCLUSIONS: Reviewer workload (volume of citations screened) when screening literature for systematic reviews of DTA is especially high. This corresponds to greater rates of class-imbalance when training classifiers for automating literature screening for DTA reviews. Addressing concerns such as lower quality metadata and effectively modelling the broader target class could help to alleviate such challenges, providing possible directions for future research.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.494 | 0.811 |
| Meta-epidemiology (narrow) | 0.003 | 0.004 |
| Meta-epidemiology (broad) | 0.009 | 0.007 |
| Bibliometrics | 0.031 | 0.028 |
| Science and technology studies | 0.004 | 0.007 |
| Scholarly communication | 0.016 | 0.016 |
| Open science | 0.007 | 0.011 |
| Research integrity | 0.005 | 0.005 |
| Insufficient payload (model declined to judge) | 0.008 | 0.003 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; the direct Gemma label and the distilled Codex classifier agree on what is shown here.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".