Literature search parameters marginally improved the pooled estimate accuracy for ultrasound in detecting deep venous thrombosis

Abstract

Background and Objectives Data collection remains of utmost importance to avoid publication bias in systematic reviews. Our objectives were to compare search strategies with and without methodological terms in Medline, to find out how other databases complement Medline, and how these strategies affect the pooled estimates of the accuracy. Study Design and Setting A study was eligible if it prospectively compared ultrasound to venography for the diagnosis of deep venous thrombosis in the lower limbs. All relevant articles found with all databases, searched from 1966 through 2003, constitute the reference standard. Sensitivity was defined as the proportion of relevant studies found by Medline searches divided by the total number of articles in the reference standard. Results Of a total of 1,473 citations, 237 abstracts were included. The sensitivity of strategies without methodological terms was higher (96%) than those with (95%, 76%, and 95%). Searches of multiple databases found seven articles not found with Medline. Searches of congress proceedings (International Society of Thrombosis and Haemostasis) found three abstracts not published in full. The diagnostic odds ratio OR was 5.66 (95% confidence interval CI = 4.84–6.48) when multiple databases were searched and 5.57 (95% CI = 3.49–7.65) when only English-language articles in Medline were identified. Conclusion Medline searches combining free text and MeSH terms were more sensitive. Single Medline search affects only marginally the pooled estimate accuracy.