Searching for clinical prediction rules in MEDLINE

Area of Science:

• Medical Informatics
• Clinical Epidemiology

Background:

• Clinical prediction rules aid in diagnostic, therapeutic, and prognostic assessments.
• Inconsistent terminology hinders the retrieval of clinical prediction rules via computerized searches.
• Validated methods for locating clinical prediction rules are lacking in the literature.

Purpose of the Study:

• To derive and validate an optimal search filter for retrieving clinical prediction rules.
• To improve the discoverability of clinical prediction rules within the MEDLINE database.

Main Methods:

• A retrospective analysis was conducted using a manually identified set of articles on clinical prediction rules (1991-1998) as the gold standard.
• Search filters were derived by analyzing textwords and Medical Subject Headings (MeSH) from a derivation set of articles.
• The accuracy of derived filters was assessed using a validation set of articles.

Main Results:

• A filter combining "predict$" and "rule$" (predict$.ti. AND rule$.) demonstrated superior positive predictive value and positive likelihood ratio.
• Several filters achieved over 90% sensitivity and specificity in retrieving clinical prediction rules.
• The filter "predict$ OR clinical$ OR outcome$ OR risk$" successfully retrieved 98% of clinical prediction rules.

Conclusions:

• Multiple search filters with high retrieval value for clinical prediction rules were identified.
• The choice of filter can be tailored based on the searcher's specific objectives and time constraints.
• These validated filters can enhance the efficient and accurate retrieval of clinical prediction rules.