Article type
Abstract
Background: Systematic reviews (SRs) for prediction models (PM) are important to evaluate their performance across different settings. Several instruction papers regarding methods for them were reported; however, there were no cross-sectional studies of their methods and reporting characteristics.
Objective: To describe the methods and reporting characteristics of protocol registered published SRs of PM.
Methods: We screened SRs of prognostic or diagnostic categories with published status in an international prospective register of systematic reviews (PROSPERO) for SRs of PMs indexed until Feb 2017. In addition, we searched for papers citing the CHecklist for critical Appraisal and data extraction for systematic Reviews of prediction Modelling Studies (CHARMS) using the Web of Science. Citations were screened and SRs for prediction models with the PROSPERO registration were retained. We defined PM as a multi-factor model for prediction of individual risk or probability of certain conditions or events. We excluded SRs of prediction studies exploring individual predictors of particular outcomes. This study protocol was registered in the University hospital Medical Information Network Clinical Trials Registry (UMIN000026103).
Results: We identified 1058 published SR protocols in the PROSPERO. Of them, 101 protocols were in prognostic or diagnostic categories. We identified 45 records through the Web of Science citation search. A total of 7 SRs (0.66%) of over 65 000 patients’ data were included (Figure 1; Table 1). No Cochrane review was identified. The reporting characteristics were highly variable especially for relevant items of the CHARMS in their protocols. Over half of the reviews did not even pre-specify the outcome to be predicted. Moreover, one study multiplied the outcome at the review phase and did not mention the change in the review.
Conclusion: The present study revealed that a limited number of published SRs of PM was registered in the PROSPERO and that their reporting characteristics were various. A guide for the PROSPERO registration of SRs of PM is needed.
Objective: To describe the methods and reporting characteristics of protocol registered published SRs of PM.
Methods: We screened SRs of prognostic or diagnostic categories with published status in an international prospective register of systematic reviews (PROSPERO) for SRs of PMs indexed until Feb 2017. In addition, we searched for papers citing the CHecklist for critical Appraisal and data extraction for systematic Reviews of prediction Modelling Studies (CHARMS) using the Web of Science. Citations were screened and SRs for prediction models with the PROSPERO registration were retained. We defined PM as a multi-factor model for prediction of individual risk or probability of certain conditions or events. We excluded SRs of prediction studies exploring individual predictors of particular outcomes. This study protocol was registered in the University hospital Medical Information Network Clinical Trials Registry (UMIN000026103).
Results: We identified 1058 published SR protocols in the PROSPERO. Of them, 101 protocols were in prognostic or diagnostic categories. We identified 45 records through the Web of Science citation search. A total of 7 SRs (0.66%) of over 65 000 patients’ data were included (Figure 1; Table 1). No Cochrane review was identified. The reporting characteristics were highly variable especially for relevant items of the CHARMS in their protocols. Over half of the reviews did not even pre-specify the outcome to be predicted. Moreover, one study multiplied the outcome at the review phase and did not mention the change in the review.
Conclusion: The present study revealed that a limited number of published SRs of PM was registered in the PROSPERO and that their reporting characteristics were various. A guide for the PROSPERO registration of SRs of PM is needed.