Deville abstract

ISSG structured abstract for:

Deville WLJM, Brezemer PD, Bouter LM. Publications on diagnostic test evaluation in family medicine journals: an optimal search strategy. Journal of Clinical Epidemiology 2000;53(1):65-9.

Structured abstract prepared by: Cynthia Fraser

Objective:

To develop a MEDLINE optimal search strategy to identify diagnostic test evaluation studies which would be applicable to any clinical field.

Methods:

A reference set of diagnostic test evaluation studies was identified by handsearching nine family medicine journals for the years 1992-5. A control set comprised the false-positive records from the same journals and years, retrieved by a MEDLINE search using an adapted diagnostic test filter that had been previously published. Potential search terms were identified from the MEDLINE records of the reference set and forward stepwise logistic regression analysis was carried out to develop search strategies that best discriminated between the reference and control sets. Four search strategies were reported and internal validity was assessed against the reference set. External validity was assessed for one strategy against a small validation set. The performance of the strategies were compared against a previously published filter

Results:

The reference set comprised 75 records and the control set, 137. The most sensitive strategy achieved 89.3% against the reference set and most specific had 98.5% specificity. The most accurate strategy had a sensitivity of 80.0%; specificity of 97.3% and positive predictive value of 48%; an improvement on the performance of the previously published filter measured against the same reference set. External validation was calculated for the most sensitive search strategy but achieved a much lower sensitivity of 61%.

Discussion:

The most accurate search strategy achieved higher sensitivity and specificity than a previously published filter. However, the most sensitive search strategy’s performance against an external set was considerably reduced. The applicability of the strategies to other clinical areas requires further testing.

ISSG commentary:

The strategies were developed on a small data set using relevant studies from a set of family medicine journals that may not be representative of all diagnostic literature. The range of sensitivities achieved was fairly low (70.7% - 89.3%) and the limited evidence presented on external validation suggests that the performance of even the most sensitive strategy is not adequate for the purposes of research that requires sensitive searches such as systematic reviews.