Filtering Medline for a clinical discipline: diagnostic test assessment framework

Clinicians often face challenges when searching for discipline-specific articles in Medline. General searches may return irrelevant results, making it difficult to locate articles within a specific field. Prior research has shown that Medline lacks built-in filters for clinical disciplines. This gap motivated the development of a targeted search framework. The need for discipline-specific filters is especially high in areas like nephrology. No prior work had resolved this issue effectively. Researchers proposed a solution using a diagnostic test assessment framework. This approach aims to improve the precision and efficiency of clinical literature searches.

This study aimed to create a Medline filter for the clinical discipline of nephrology. The goal was to allow clinicians to search within a specific field rather than the entire database. The researchers focused on improving search accuracy and relevance. They selected 4657 articles from 40 journals published in 2006. Manual review of these articles identified 19.8% as relevant to nephrology. The study tested 1,155,087 unique renal filters against this manual review. The objective was to determine the best-performing filters for this discipline. The researchers sought to validate the filters for future use in PubMed.

The study used a diagnostic test assessment framework with development and validation phases. A sample of 4657 articles was manually reviewed for nephrology relevance. Researchers tested 1,155,087 unique filters to identify the most effective ones. Each filter combined two to 14 terms or phrases. The filters included terms like 'kidney' with truncation and 'renal dialysis'. The team evaluated sensitivity, specificity, precision, and accuracy for each filter. Performance was measured against the manually reviewed dataset. The validation phase confirmed the reliability of the best filters.

The best renal filters achieved a peak sensitivity of 97.8% and specificity of 98.5%. These filters combined two to 14 terms or phrases for optimal performance. Filters included terms like 'kidney' with truncation and 'renal replacement therapy'. The validation phase showed consistent performance across different datasets. The filters remained highly accurate and precise in identifying relevant articles. Researchers identified key phrases such as 'proteinuria' and 'glomerul' truncated. The filters outperformed general Medline searches for nephrology. These results suggest the filters could be stored in PubMed for clinical use.

The study demonstrated that Medline can be filtered for nephrology with high reliability. The best filters achieved near-perfect sensitivity and specificity. These filters can help clinicians find relevant articles more efficiently. The validation phase confirmed the filters' consistent performance. Researchers suggest storing these filters in PubMed for everyday use. The methods used in this study can be applied to other clinical disciplines. The results support the development of discipline-specific filters. This approach may improve the accuracy of clinical literature searches.