What does SUMSearch do?
SUMSearch is a unique method of searching for medical evidence by using the Internet. SUMSearch combines meta-searching and contingency searching in order to automate searching for medical evidence. Meta-searching, which is used by from general Internet search engines such as from Go2Net, Dogpile, and SavvySearch, means simultaneously searching multiple Internet sites and collating the results into one page. In addition, SUMSearch adds the idea of contingency searching. If SUMSearch finds too many ‘hits’ from an Internet site, SUMSearch will execute more restrictive, contingency searches. For example, if a search finds 1000 articles at PubMed, SUMSearch may do up to four additional searches until an optimal number is received. On the other hand, if SUMSearch finds few hits from an Internet site, it may add a search of another site. For example, if the Database of Abstracts of Reviews of Effectiveness (DARE) provides too few systematic reviews, SUMSearch will add a search for systematic reviews from MEDLINE. In summary, SUMSearch allows the clinician to enter a query one time, and then will: select the best Internet sites to search, format the query for each site, execute contingency searches, then return a single document to the clinician. SUMSearch removes the burden to the clinician of remembering details such as which Internet site truncates with the dollar sign and how to execute a limit for the AIM journals if too many articles are found at MEDLINE.
After searching, SUMSearch organizes the list of links to documents that it returns to the clinician. The links are ordered by breadth of discussion. First, there are links to resources that provide broad discussion: relevant textbooks, followed by traditional review articles, and practice guidelines. Next there are links to resources that provide narrow discussions: systematic reviews, and original research. Thus, the clinician that is searching a topic with which they are not familiar, will find links to easy to read, broad discussions at the top of the list. A clinician that has a specific question within a topic with which they otherwise familiar, will find links to systematic reviews and original research in the second half of the results.
The rationale to organizing the results by the breadth is as follows. Ideally, clinicians always answer medical questions by reading original research and systematic reviews of original research. However, this is not always practical, especially for broad clinical questions. For example, consider a clinician that is confronted by a patient in sepsis yet has no recent experience with the disease. The clinician may ask "Tell me all about sepsis, e.g. how to quickly recognize, prognosticate, and treat my patient". This clinician does not have the time to seek original studies addressing each facet of his question and thus may best benefit from broad discussions provided by SUMSearch. On the other hand, a more experienced clinician with the same patient may ask, "among patients with septic shock, how much do steroids, compare to placebo, reduce mortality?" This clinician has a specific question and may afford the time to seek original studies and systematic reviews to get the most current answer to a specific problem.
Quality counts. We want to provide the clinician with the most valid medical information possible. SUMSearch only queries Internet sites that contain evidence written by qualified professionals. The majority of the links provided by SUMSearch come from three Internet sites - the National Library of Medicine (NLM), DARE, and the National Guideline Clearinghouse (NGC). All three of these sites are government sponsored and likely have limited bias due to conflict of interest. For example, probably little influences the selection of journals, systematic reviews, and practice guidelines that are included by the NLM, DARE, and NGC, respectively. In addition, when querying MEDLINE at the NLM, SUMSearch uses validated search filters as much as possible. These filters have been developed by various researchers in order to optimally search for certain types of articles. For example, when the clinician clicks the 'treatment' focus, SUMSearch includes a search of MEDLINE using a filter validated to find randomized controlled clinical trials (2).
2004-04: Diagnosis filter is revised according to Haynes et al. BMJ 2004;328:1040 PMID: 15073027
Strategy sensitivity specificity
First dx filter sensitiv:.mp. OR diagnos:.mp. OR accuracy.tw. (from Table 3, #2) 98 83
Second dx filter is sensitiv:.mp. OR predictive value:.mp. OR accurac:.tw. (from Table 5, 1)
Searching the Internet live is an ever changing task.. What happens if the Internet is momentarily too congested to used, or one of the target Internet sites, such as the National Library of Medicine is momentarily offline or changes the appearance of their pages so that SUMSearch cannot parse the results? These problems may lead to SUMSearch erroneously reporting that evidence does not exit to answer your question. SUMSearch is actually three programs running on two computers.
In addition to the search program, a program on a second computer uses the Internet to query a third program on the SUMSearch server to examine SUMSearch's recent history every 30 minutes. The second computer will page someone if SUMSearch is not reachable or if potential search errors are detected.
Secondly, several analyses are done at the time of each search to detect the presence of possible changes in the design of the Internet sites that SUMSearch querries. If a possible error is detected, the searcher is alerted that the search may be invalid and should be repeated.
These safeguards reduce the chance of errors introduced by changes in the Internet and allow SUMSearch to be re-programmed as quickly as possible in response to changes in the Internet.
We developed SUMSearch because of our disappointing experience in teaching medical students how to quickly locate medical evidence. For several years we have taught students how to locate medical evidence with MEDLINE; however, we also taught new innovations such as MEDLINE filters and new Internet sites including the DARE and NGC. A controlled trial of our efforts showed that although we taught many students skills necessary for searching, these students did not seek medical information more often than did their peers who were not taught. We also observed that we were repetitively giving the same advice to students who needed help in searching. For example, we were frequently reminding students of tips such as, 'if you did not find what you wanted at the DARE, search MEDLINE with the filter for systematic reviews' and 'if you found too many articles at MEDLINE, restrict your search to the Abridged Index Medicus'. These observations led to theorizing that, with the software available at that time, searching for evidence is still very hard yet can be automated. Thus we created SUMSearch, which automates the search for evidence.
In summary, SUMSearch is a new method to search for medical evidence. We hope to provide an easy-to-use search for links to valid medical evidence. We will continue to improve SUMSearch based on quantified evaluations.
Even more detail.
How often and well is SUMSearch used?
We perform ongoing evaluation SUMSearch in several ways. First, the search terms, focus request, date, and number and types of articles for each search always recorded. In addition, at intervals, we monitor which links searchers click after the results of their search are returned to them. Analysis of this data has led to the following observations. Searches are twice as likely to succeed if the searcher includes no more than two words within each search term (3). Also, searches are twice as likely to succeed if the searcher uses the MeSH browser (4). Consequently, SUMSearch now warns the searcher, before their query is submitted, if their query has more than two words within any search term or if the searcher has not checked any of their terms with the MeSH browser.
Predictors of successful searches for medical evidence by clinicians (12/2000).
Badgett RG, Mulrow CD, Levy LS.
Summary: Searches with more than two words within a single search term are more likely to fail
Observations On How Clinicians Use SUMSearch (1/2000).
Badgett RG, Mulrow CD, Levy LS, Arterburn J. Observations on how clinicians use SUMSearch. J Gen Intern Med 2000;15(supplem 1):99.
Summary: Failed searches: A new warning SUMSearch has reduced the above problem. Use of the MeSH browser if now recorded and shows use of the browser is associated with successful searches.
How often do searchers click documents: Searchers click links to documents in about 50% of searches that retrieve citations (this is similar to other studies on how often literature searches effect patient care)
SUMSearch can store your login information for subscriptions to online journals. SUMSearch also store passwords for institutions and place them in hidden cookies on the computers of users in the appropriate IP range of the institution. For more information, click subscriptions manager from the main page.
Can I make my own web pages to access SUMSearch?
Mirrors to SUMSearch exist at:
http://www.nelh.nhs.uk/management/sumsearch.htmNational Electronic Library of Health
Université de Rouen (French)
Please contact me. We will set up a method to identify SUMSearch queries that come from the search form that you create. This allows me to send you a periodic analysis on how well your clinicians are searching. This analysis can lead to changes in your page when clinicians are not using it well, or changes in our page if your page leads to better searches than our page. Badgett@UTHSCSA.edu
Who is SUMSearch
SUMSearch has no formal fund and not for profit. SUMSearch receives no money for any of the sites it links to. We intend for SUMSearch to always be freely available and without advertising. SUMsearch acknowledges generous support from
The University of Texas Health Science Center provides my salary support.
Verdict provides my office and other indirect support. Verdict funded the network upgrade of 8/2001.
Dr. Cindy Mulrow funded the server upgrade 8/2001
The SUMSearch personnel are
Bob Badgett, MD. Associate Professor of Medicine. Director of Clinical Informatics, Department of Medicine. UTHSCSA.
Linda Levy, MLS, AHIP. Briscoe Medical Library
What Improvements are planned?
SUMSearch is continuously revised in responses to changes in medical information on the Internet. The research and development version of SUMSearch is available by clicking here. You are welcome to use this site to see newest changes. Your comments and suggestions are also welcome.