Introduction 1 7 1 3 8 9 10 To assist clinicians searching for studies on age-specific patient populations, we have developed and tested Medline search strategies for detecting studies for specific age categories as well as tested age-specific search terms pertinent to five age-related clinical specialties. In this paper, we report on the evaluation of the retrieval performance of age-specific search strategies in Medline compared with a manual review (the "gold standard" search) of each article in every issue of 161 journals in the year 2000. 11 15 15 Methods 16 ACP Journal Club, Evidence-Based Medicine, Evidence-Based Nursing, Evidence-Based Mental Health Annals of Internal Medicine BMJ, JAMA, Lancet Archives of General Psychiatry, British Journal of Psychiatry Nursing Research Table 1 4 17 4 Table 1 Comparison of hand searching and Medline MeSH classification of age categories     Hand Search Category Our Definition Medline MeSH Term Category MeSH Definition Fetus Fetus - - Newborn Birth to 1 month Infant, newborn Birth to 1 month Infant > 1 month to < 24 months Infant 1 to 23 months Preschool 2 years to < 6 years Child, preschool 2 to 5 years Child 6 years to < 13 years Child 6 to 12 years Adolescent 13 years to < 19 years Adolescent 13 to 18 years Adult 19 years to < 45 years Adult 19 to 44 years Middle age 45 years to < 65 years Middle aged 45 to 64 years Aged 65 years to < 80 years Aged 65 to 79 years Aged 80 ≥ 80 years Aged, 80 and over 80 years and over ND Nondiscernible - - 17 Table 2 Table 2 Formulae for calculating the sensitivity, specificity, precision, and accuracy of searches for detecting age-specific articles*     Hand Search Detection of Search Terms Meets Criteria Does Not Meet Criteria Detected a b Not detected c d  a+c b+d * Individual search terms with sensitivity > 25% and specificity > 75% for a given age category were incorporated into the development of search strategies that included two or more terms. All combinations of terms used the Boolean "OR." For the development of multiple-term search strategies to either optimize sensitivity or specificity, we tested all two-term search strategies with sensitivity of at least 75% and specificity at least 50%. To construct a comprehensive set of search terms, a list of MeSH terms and textwords was initially generated, and input was sought from clinicians and librarians in the United States and Canada through interviews with known searchers, requests at meetings and conferences, and requests to the National Library of Medicine. These experts were asked which terms or phrases they used when searching for age-specific studies, as well when searching for studies in specific purpose categories. Search terms could be MeSH terms, including publication types and subheadings, or textwords specific to age in titles and abstracts of articles. Various truncations were also applied to the textwords, phrases, and MeSH terms. We compiled a list of 543 age-specific terms (Multimedia Appendix). All terms were tested in Medline using the Ovid Technologies searching system. Table 1 Table 1 Results Tables 3 to 7 show the operating characteristics of top-performing combinations of terms with best sensitivity, best specificity, and best optimization of sensitivity and specificity while minimizing the difference between the two, for detecting studies on geriatric medicine, adult medicine, pediatric medicine, neonatal medicine, and obstetrics in Medline in 2000. Search strategies are reported using Ovid's search engine syntax for Medline (mp = multiple posting—term appears in title, abstract, or subject heading; sh = subject heading [MeSH term]; tw = textword—word or phrase appears in title or abstract; : = truncation; pt = publication type; exp = explode—a search term that automatically includes closely related MeSH terms; tu = therapeutic use as a subheading; xs = exploded subheading). Geriatric Medicine The single term "exp adult" yielded the best sensitivity (96.4%) with a specificity of 55.9% for retrieving articles about geriatric medicine. However, by using the next best sensitivity combination, "aged.sh. OR age:.tw.", a small sacrifice in sensitivity (1% absolute decrease) resulted in a much better specificity compared with the most sensitive term (absolute increase 14.4%) and improved precision (absolute increase 5.2%) and accuracy (absolute increase 13.3%). As expected, precision improved slightly when specificity was maximized (absolute increase 8.6%). The term that yielded the best optimization of sensitivity and specificity, "aged.sh.", resulted in 93.6% sensitivity and 82.7% specificity. Table 3 Combination of terms with the best sensitivity, best specificity, and best optimization of sensitivity and specificity for detecting studies about geriatric medicine (≥ 65 years) in Medline in 2000      * † Sensitivity, % (95% CI) (n = 3309) Specificity, % (95% CI) (n = 45719) ‡ (95% CI) Accuracy, % (95% CI) (n = 49028)                          * † ‡ Adult Medicine The three-term strategy "adult.mp. OR middle aged.sh. OR age:.tw." yielded the best sensitivity (94.9%) and had a specificity of 64.5% for retrieving articles about adult medicine. When specificity was maximized (85.2%) with the single term "middle aged.sh.", sensitivity lowered to 72.3%, but precision improved to 62.1% (absolute increase 14.8%) and accuracy improved as well (absolute increase 9.8%). The best optimization of sensitivity and specificity occurred with the combined terms "middle aged.sh. OR of age.tw.", with values approaching 79%. Table 4 Combination of terms with the best sensitivity, best specificity, and best optimization of sensitivity and specificity for detecting studies about adult medicine (19 to < 65 years) in Medline in 2000      * Operating Characteristics † Sensitivity, % (95% CI) (n = 12307) Specificity, % (95% CI) (n = 39721) ‡ (95% CI) Accuracy, % (95% CI) (n = 49028)                           * † ‡ Pediatric Medicine The three-term strategy "child:.mp. OR adolescent.mp. OR infan:.mp." yielded the best sensitivity of 98.0% with a specificity of 81.2% for retrieving articles about pediatric medicine. When specificity was maximized (97.1%) with the single term "children.tw.", a striking trade-off in sensitivity occurred as it was lowered to 58.2% (absolute decrease 39.8%). Yet, as expected, precision improved (absolute increase 30.9%). The three-term strategy "adolescent.tw. OR children.tw. OR child, preschool.sh." yielded the best optimization of sensitivity and specificity (89.3% and 87.3%, respectively). Table 5 Combination of terms with the best sensitivity, best specificity, and best optimization of sensitivity and specificity for detecting studies about pediatric medicine (> 1 month to < 19 years) in Medline in 2000      * † Sensitivity, % (95% CI) (n = 2845) Specificity, % (95% CI) (n = 46183) ‡ (95% CI) Accuracy, % (95% CI) (n = 49028)                * † ‡ Neonatal Medicine Best sensitivity (95.3%) was achieved by the three-term strategy "infan:.mp. OR child:.mp. OR gestation:.tw.", with a specificity of 83.6% for retrieving articles about neonatal medicine. An expected trade-off occurred in sensitivity (absolute decrease 41.7%) with the most specific term, "infants.tw." (98.7%). However, precision increased to 38.2% (absolute increase 30.8%) and accuracy reached 98.2%. The three-term strategy "infan:.mp. OR gestation:.tw. OR neonatal.tw." yielded the best optimization of sensitivity and specificity, reaching values of 93% (which were the highest among all five specialties). Table 6 Combination of terms with the best sensitivity, best specificity, and best optimization of sensitivity and specificity for detecting studies about neonatal medicine (birth to 1 month) in Medline in 2000      * † Sensitivity, % (95% CI) (n = 663) Specificity, % (95% CI) (n = 48365) ‡ (95% CI) Accuracy, % (95% CI) (n = 49028)                          * † ‡ Obstetrics The combination of terms "gestation:.tw. OR fetal.tw. OR pregnancy.tw." produced the best sensitivity of 82.0%, with a very high specificity of 97.1% for retrieving articles about obstetrics. The maximization of specificity (reaching almost 99%) with the single term "gestation:.tw." yielded a 1.8% increase in specificity but with a marked trade-off in sensitivity, which decreased to 52.0% (absolute decrease 30%). Table 7 Combination of terms with the best sensitivity, best specificity, and best optimization of sensitivity and specificity for detecting studies about obstetrics (fetus) in Medline in 2000      * † Sensitivity, % (95% CI) (n = 516) Specificity, % (95% CI) (n = 48512) ‡ (95% CI) Accuracy, % (95% CI) (n = 49028)                * † ‡ Discussion Our study shows that selected age-specific search strategies can achieve high retrieval of studies for age-specific populations. Our age-specific search strategies performed differently among the five specialties we investigated. The highest sensitivity and specificity were achieved for pediatric medicine (98% and 81.2%, respectively) and neonatal medicine (95.3% and 83.6%, respectively). This finding may be a result of these age groups being more precisely defined and that studies tend to be narrowly focused on them. Search strategies within obstetrics yielded a higher specificity (97.1%) than sensitivity (82%), indicating that this strategy was better at filtering out nonrelevant age-specific articles than retrieving them. The best performing strategy for optimizing sensitivity and specificity was achieved within neonatal medicine (92.5% and 92.6%, respectively). In all cases, precision was low, a consequence of searching in large multi-purpose databases. Future research is focusing on potential ways to improve precision without compromising sensitivity, for example, by searching in journal subsets. 18 The utility of age-specific filters will vary according to the needs of clinicians and researchers who must weigh the consequences of using a sensitive or specific search. Although a sensitive search will not miss many relevant articles, such searches are less precise and entail time-consuming sorting through irrelevant articles. The narrower yield of a specific search will capture many relevant articles and take less weeding, but it has greater potential for missing key articles. Search Examples Table 8 Table 8 Example: best sensitivity (keeping specificity ≥ 50%) search strategies for detecting treatment studies in geriatric medicine (patients ≥ 65 years of age) in Medline (1996 to July week 3, 2005)       * Content Term Boolean Operator Best Sensitivity Combination Strategy for Treatment Studies Boolean Operator Best Sensitivity Combination Strategy for Geriatric Medicine Number of Articles Huntington disease - - - - 5907 Huntington disease AND clinical trial.mp. OR clinical trial.pt. OR random:.mp. OR tu.xs. - - 901 Huntington disease AND clinical trial.mp. OR clinical trial.pt. OR random:.mp. OR tu.xs. AND † 483 * † However, sifting through such a large number of articles would be time-consuming and many of these articles would not be relevant to treatment studies in geriatric medicine. By combining the content term "Huntington disease" with the most sensitive combination of terms for treatment studies (clinical trial.mp. OR clinical trial.pt. OR random:.mp. OR tu.xs.), the search can be narrowed to 901 articles. Further, by adding the most sensitive strategy for geriatric medicine (exp adult) to this search string with the Boolean operator AND, the search is refined to 483 articles, which is much more manageable than the original 5907 articles retrieved from searching the content term only. A sensitive search such as this would be an efficient beginning for researchers interested in conducting systematic reviews. 12 Table 9 Table 9 Example: best specificity (keeping sensitivity ≥ 50%) search strategies for detecting treatment studies in geriatric medicine (patients ≥ 65 years of age) in Medline (1996 to July week 3, 2005)       * Content Term Boolean Operator Best Specificity Combination Strategy for Treatment Studies Boolean Operator Best Specificity Combination Strategy for Geriatric Medicine Number of Articles Huntington disease - - - - 5907 Huntington disease AND Randomized controlled trial.mp. OR randomized controlled trial.pt. - - 46 Huntington disease AND Randomized controlled trial.mp. OR randomized controlled trial.pt. AND aged, 80 and over.sh. OR of age.tw. 5 * Conclusion Selected age-specific search strategies can enhance the retrieval of studies for clinicians and researchers who need information relevant for a well-defined age-category patient population. The optimal trade-off between sensitivity and specificity should be determined according to the needs of the searcher.