Method: Using material gathered from NLM's archives and from personal interviews with people associated with developing MEDLARS and its associated systems, the author discusses key events in the history of MEDLARS.

Discussion: From the development of the early mechanized bibliographic retrieval systems of the 1940s and to the beginnings of online, interactive computerized bibliographic search systems of the early 1970s chronicled here, NLM's contributions to automation and bibliographic retrieval have been extensive.

Conclusion: As NLM's technological experience and expertise grew, innovative bibliographic storage and retrieval systems emerged. NLM's accomplishments regarding MEDLARS were cutting edge, placing the library at the forefront of incorporating mechanization and technologies into medical information systems.

The US Congress established the National Library of Medicine (NLM) to assist the advancement of medical and related sciences [1]. In response to its Congressional mandate, NLM provides electronic and print access to reliable health information in the form of catalogs, bibliographies, indexes, and online databases [1,2] to aid in the dissemination and exchange of scientific information [1]. During the last fifty years, the health sciences field has produced an explosion of biomedical research and related publications to the extent that NLM accelerated investigation of new information technologies [3] to ensure the world biomedical community's access to scientific information [4,5].

This paper sketches the chronology of the mechanization and computerization of medical indexes and bibliographic searching beginning with the early mechanized systems (1940s) and going forward to the introduction of the online computerized bibliographic search systems (1970s). The paper focuses on NLM's development of new methods for processing, organizing, and disseminating health information and illustrates how the library's experience and expertise developed NLM's Medical Literature Analysis and Retrieval System (MEDLARS) and MEDLARS II. The paper presents a selective rather than a comprehensive discussion of the events and systems leading to MEDLARS given the complexity of factors involved in MEDLARS' development.

The chronicle of NLM's commitment to delivering information to health care professionals is deeply rooted in early historical indexing. NLM's indexing tradition began with Dr. John Shaw Billings, the first director of the Library of the Office of the Surgeon General of the US Army 1867–1895 (later named the National Library of Medicine) [6]. Dr. Billings formulated the library's mission to acquire, catalog, and index the literature of the medical sciences [7] and compiled the first medical index [8], the 1879 Index Medicus [9], followed in 1880 by the Library of the Surgeon General's Office Index-catalogue [10]. The library subsequently published a notable series of comprehensive printed biomedical indexes from 1916 to 1959 [11–13]. Around the close of World War II, the American Library Association (ALA) Survey of the Army Medical Library challenged the library to modernize the printed indexes and recommended a renewed effort to analyze the functions of the three major indexes [14]. Specifically, the indexes were the American Medical Association's Quarterly Cumulative Index Medicus (1927–1956) [11], Current List of Medical Literature (1941–1959) [12], and the Army Medical Library's Index-Catalogue of the Library of the Surgeon General's Office [13], all of which were significantly behind their publication deadlines [15,16].

The labor-intensive process to produce the printed indexes in the mid-1940s demonstrates the reason for the indexes' publication delays and the ALA recommendation. Processing began by typing the subject and author entries on individual three-by-five inch slips. Then the individual slips were numbered, arranged, and hand-mounted by a shingling technique pioneered by the US Department of Agriculture Library and the Library of Congress. Following the preparation of printer's plates for the monthly issues, the staff interfiled by hand thousands of subject and author slips to form the six months' accumulation of an index. Seymour Taine, the library's chief of the bibliographic services division (1951–1964), reported that the production process was an ingenious technique, but the half million slips needed in a six months' volume of Current List approached the utmost limits of the process in terms of size and offered scant prospect for expansion or improvement [15]. The slow, labor-intensive, and repetitive indexing process [17] suggested that much of the work could be done by the machines [18] being tested and used in industry.

Scott Adams, NLM deputy director (1960–1969), described the national information retrieval environment in the 1950s as “frenetic” [7]. The serious application of computers to bibliographic information retrieval began with slow serial searches of small files of bibliographic records on magnetic tape [19]. Notably, Johns Hopkins University's Welch Medical Library Indexing Project, supported by the Armed Forces Medical Library [20], was credited as one of the first major studies of the application of machine methods for indexing the medical literature [16].

US Senators and Congressmen expressed their impatience with the library's inability to use the imaginative solutions proposed by academia and industry to solve problems with science information retrieval [7]. In response to their concerns, Dr. Frank B. Rogers, Army Medical Library and NLM director (1949–1963) [8], confirmed that NLM “felt acutely the need to improve upon its bibliographic techniques … and to provide for the publication of Index Medicus and its byproducts [21].”

The labor-intensive process to prepare the printed indexes and the technological advances in the information retrieval environment [15] convinced NLM to seek new methods to produce its comprehensive indexes. The NLM Index Mechanization Project (July 1958–June 1960) [22], funded by the Council on Library Resources, was NLM's first mechanized project to provide rapid bibliographic access and effective management of the scientific literature [17]. Under the NLM management of Seymour Taine, the Index Mechanization Project focused on two objectives: (1) an improved system of index publication and (2) a by-product bibliographic retrieval system [15].

The publication component of the Index Mechanization Project used the time-saving Eastman Kodak Listomatic camera, a high-speed, precision machine capable of photographing text imprinted or typed at the top of an IBM punched card. The system created a negative film to be read and processed for a photo-offset plate, which allowed the camera to photograph a typical monthly issue of the NLM Current List of Medical Literature in a single day. The Listomatic camera could photograph the annual accumulation of approximately a million punched cards in about two weeks [15]. The project's publication component successfully met the objective of composing and publishing NLM publications. [23].

NLM, however, abandoned the Index Mechanization Project's second objective of producing a bibliographic retrieval system [23]. The Index Mechanization Project Report explained that a retrieval system could not be successfully grafted onto a publication system. Instead, an index mechanization project should start with the design of the retrieval system and then proceed with the publication system [24]. The NLM Annual Report for the Fiscal Year 1961 further concluded that far more sophisticated machines than typewriters would be required before an effective retrieval system could be achieved [25]. Director Rogers reported, however, that in spite of the ill-fated bibliographic retrieval system, NLM derived valuable experience from the mechanization project, which he anticipated would be of greater significance in the long run than the project's successful publication capabilities. The project also provided NLM with the impetus to investigate the application of more sophisticated mechanized techniques [23]. Dr. Rogers later observed that the NLM staff involved in the project did not realize that it was sowing the seeds of MEDLARS [26].

Based on the Index Mechanization Project under the leadership of Director Rogers, NLM began planning a computer-oriented bibliographic retrieval and publication system [27,28] called MEDLARS, an acronym for Medical Literature Analysis and Retrieval System [28]. Director Rogers remarked that it was difficult to state precisely the point in time when MEDLARS was born [25], but NLM formally began work on MEDLARS with written specifications and performance capabilities using a computer with the bibliographical search and retrieval system as the top priority [26] and the publication system as a secondary priority [16].

In 1960, NLM made a formal grant proposal to the National Heart Institute (NHI) to underwrite the MEDLARS computer system. Dr. Martin M. Cummings, AHIP, FMLA, then associate director for research grants at the National Institutes of Health (NIH), recently explained [29] the unpublished details of a grant complication that almost doomed MEDLARS. When NLM sought grant funding from NHI, an auditor questioned whether one government agency could “grant” part of its Congressional budget allocation to another government agency. Two years earlier, Dr. Cummings, as director of research for the Veterans Administration (VA), and Dr. Ernest M. Allen, associate director for the Extramural Program at NIH, allowed NIH to give grants from their federal budget to qualified VA hospital researchers. Thus Associate Director Cummings with support from Dr. Allen, then deputy assistant secretary for health, judged that the NLM MEDLARS grant should be awarded based on the 1958 precedent-setting decision [29].

With the funding issue resolved, NLM developed innovative objectives and specifications for MEDLARS [24,30–32] (Table 1) and sought proposals for a two-year developmental program for the study, design, installation, and trial operation of MEDLARS. In August 1961, NLM selected the General Electric Company (GE) from among twenty-five contenders [26]. GE divided the project into three phases:

• Phase I: the preliminary design of the system
• Phase II: the detailed design, programming of the computer operations, and specifications for special equipment
• Phase III: the implementation of the computer system

Table 1 Medical Literature Analysis and Retrieval System (MEDLARS) pri mary development objectives (1960)

By January 1964, the MEDLARS Honeywell 800 computer subsystem had been planned, acquired, installed, and preliminarily tested [29,34,35] (Figure 1). The Louisville, Kentucky, Courier-Journal Magazine [36] credited NLM Director Rogers for the concept and design of the MEDLARS program. Dr. Cummings praised Director Rogers for the foresight to anticipate that computers could foreshorten the time required to produce the library's printed products and provide new types of bibliographic services to serve the needs of physicians and scientists [29].

Figure 1 Honeywell computerCaption from Feb 1975 NLM News: The Honeywell computer, historic workhorse of MEDLARS I, was responsible for processing over 10 years of Index Medicus and numerous other bibliographies used throughout the world. It processed thousands (more ...)

Graphic Arts Composing Equipment (GRACE) When NLM Director Cummings (1964–1984) arrived at NLM in January 1964, only one unique aspect of the MEDLARS processing system, the innovative offline computer-driven phototypesetter GRACE [37]—designed to meet the high-speed, high-quality, and high-volume publication requirements of MEDLARS [6]— was not operational [29]. Director Cummings and the NLM staff worked diligently to put actions in motion to complete GRACE. Their success was marked by the August 1964 publication of Index Medicus using GRACE. GRACE consisted of three distinctive components: (1) a magnetic tape unit from the Honeywell computer that read the tapes to be printed, (2) an electronic control unit with a console, and (3) an optical unit [38,39] (Figure 2). GRACE accepted electrical signals in bursts of coded characters that represented a full line of legible type from the MEDLARS magnetic-tape transport unit. GRACE then set the type on film or paper to produce its publications [38]. In addition, GRACE simultaneously processed as many as thirty reference searches a day [40] with the estimated typesetting power of fifty-five Linotype typesetting operators [41].

Figure 2 IBM 370/155 master console (right) and high-speed printer (left) with NLM staff

Much of the revolutionary computerized photocomposition of today owes its origin to the development of GRACE [29]. The Washington Evening Star reported that GRACE was a breakthrough in printing technology and quoted Director Cummings as saying, “we feel that the August issue of Index Medicus is an historical document … because of its great significance both to medicine and printing” [42]. Testimony to the Congressional Subcommittee on Appropriations reported that “There is no comparable instrument for speed and accuracy of photocomposition anywhere in the world” [34]. Charles Bourne, author of A History of Online Information Services, 1963–1976, cited the 1974 Encyclopedia Britannica that stated that GRACE was as much a landmark in the history of phototypesetting as the Gutenberg Bible was in printing [19].

American business and industry that used repetitive daily mass printing in their operations observed the success of GRACE. Representatives of telephone companies, Wall Street businesses, and newspapers interested in fast, computer-generated photocomposing visited NLM to observe the functioning of GRACE [29]. Archived correspondence from Director Cummings to both Melvin Day, director of the Scientific and Technical Information Division of the National Aeronautics and Space Administration (NASA) [43], and to the US surgeon general [44] confirmed that NASA used GRACE for preparing indexes for their publication, STAR, in 1967. After admirably serving NLM, GRACE retired to the Smithsonian Institution in 1969 [29].

Although MEDLARS used Medical Subject Headings (MeSH), by 1964 the rapid advances in biomedical sciences created some deficiencies in its use [45]. NLM engaged nonlibrarian Dr. Peter Olch, a surgeon and pathologist working at NIH, and later Dr. Norman Shumway, head of the NLM Medical Subjects Heading Section, to enrich the MeSH terminology with the latest medical concepts for both Index Medicus and MEDLARS [45]. NLM indexer Thelma Charen [29] and thesaurus builder and indexer Claire Schultz [46] also worked to enhance MeSH.

MEDLARS was complete with the installation of GRACE and the initial refining of MeSH. MEDLARS was heralded as the first library application of a computer to handle scientific literature [32,47] with a digital computer and high-quality composing equipment [21]. MEDLARS was also the first large-scale information retrieval project based in a research library to provide both bibliographic access as well as copies of the documents [6]. MEDLARS produced (1) printed products, (2) recurring bibliographies, and (3) demand (nonrecurring) bibliographic searches [6,48,49].

• Printed products: MEDLARS used the high-speed phototypesetter GRACE to compile and print several NLM publications including the monthly Index Medicus, the annual Cumulated Index Medicus, the list of MeSH, and the recurring medical bibliographies. The second slower MEDLARS computer printer produced the demand (nonrecurring) search bibliographies [6,21].
• Recurring bibliographies: NLM's trained search specialists periodically prepared medical bibliographies (lists of citations) for specialized consumer groups. MEDLARS then prepared for printing the bibliographies [45,50] to be distributed by government agencies and nonprofit national professional organizations working in the specialty fields [28]. The recurring published bibliographies included, for example, the Index of Rheumatology for the American Rheumatism Association and the Index to Dental Literature for the American Dental Association and the National Institute of Dental Research [34,51].
• Demand (nonrecurring) bibliographic searches: The major motivation for MEDLARS was to provide rapid computer searching of data files [48]. MEDLARS provided a resourceful method to answer individual requestors' highly complex, bibliographic questions that could not be handled efficiently by referring to the printed indexes and catalogs [4,5,28,31].

To obtain a demand bibliography, physicians, health scientists, educators, and students completed a MEDLARS search request form at NLM or mailed the form to NLM [4]. Medical librarians trained as MEDLARS search specialists performed the requested computer searches [51]. Typists converted the search to machine-readable form on punched paper tape that was read into the computer [21]. The computer then performed a time-consuming search of the entire file of magnetic tapes in sequence to retrieve all of the citations on a given subject [4]. NLM prepared pamphlets to assist patrons with the MEDLARS demand search service [52] and provided detailed descriptions of MeSH, subheadings, and the indexing procedures [53].

The MEDLARS data processing system MEDLARS required high-speed data processing equipment to meet the time requirements and volume of work. Hundreds of published and unpublished pages of detailed procedures and instructions in a wide assortment of manuals and pamphlets described every aspect of MEDLARS. The following brief overview summarizes the basic elements of the three main subdivisions of the MEDLARS' data processing system: (1) input and conversion subsystem, (2) computer manipulation subsystem (retrieval subsystem), and (3) output subsystem (publication) [21,31,32,40,50].

• The input and conversion subsystems' operation combined the intellectual talents of librarians with the processing and storage capabilities of the computer [54] and included three basic types of input: (1) citations from the literature with relevant descriptors (known as unit records); (2) master-file data (MeSH, etc); and (3) operating procedures and instructions for computer operators [6]. Charles J. Austin, NLM head of the data processing section (1962–1966), explained that the input subsystem received and prepared journals with indexing, monographs with cataloging, and search requests. The process began with a staff of indexers assigned to a specific number of new journals from which the indexers selected the articles. Indexers created the journal articles' unit record together with their MeSH terms. Catalogers received non-journal documents such as monographs and performed somewhat similar functions [6]. The input typist transformed the unit record to machine-readable form on punched paper tape using a punched-tape typewriter and then integrated the information on a data sheet prepared by an indexer with journal information and codes that allowed the computer to recognize the unit records' elements. The unit record was then verified by proofreaders, corrected, and returned for preparation of a correction tape that was entered into the computer [51]. Similarly, NLM librarians (literature analysts) designed search requests for demand bibliographic searches with logical search combinations to prepare search requests for patrons. A typist transformed each search into machine-readable form for the computer [6].
• The computer manipulation subsystem (also called the retrieval subsystem) centered on the Minneapolis-Honeywell 800 digital computer. GE designed seven intricate programming modules that were large groups of machine instructions for the computer system. Essentially, the computer accepted the unit records on paper tape, spliced into batches (thus the term batch-mode processing). The computer checked the unit record for correctness, compressed the unit record to facilitate subsequent processing, and stored the unit records on magnetic tape in chronological sequence for at least five years [32,50]. In response to both demand and recurring search requests, the computer searched its accumulation of unit records for those that qualified for the search retrieval and then edited and composed the retrieved records for output on magnetic tape [21].
• The output/subsystem (also called the publication subsystem) converted citations and descriptors characterizing the citation into photosensitive film to produce the MEDLARS printed products. Technically stated, the output system transformed the magnetic-tape output of the computer manipulation subsystem into exposed film to two new printers used to print the MEDLARS products from the stored magnetic tapes [21]: first, GRACE, the offline high-speed computer phototypesetter [37] for the high-volume publication requirements such as Index Medicus and the recurring bibliographies [6,33], and second, the standard medium-speed mechanical printer for demand bibliographic searches requested by patrons [6,21,38] (Figures 3 and 4).

  Figure 3 Computer room at the National Library of Medicine

  Figure 4 IBM magnetic tape drives of the IBM 370/155 with NLM staff

The Courier-Journal Magazine announced that NLM “was the largest collection of medical literature ever assembled anywhere in the world … and within these walls stands mighty MEDLARS and GRACE— two names destined to change the entire concept of rapid assembly and retrieving of information” [36]. Miles noted in A History of the National Library of Medicine that “On the whole the system was one of the largest and most successful library automation projects. Its success marked a milestone in the evolution of modern libraries” [8].

Patron feedback on the usability of MEDLARS demonstrated NLM's need to engage Dr. F. W. Lancaster, an outside consultant to NLM, to perform a major evaluation of MEDLARS' bibliographic searching capability [57]. Dr. Lancaster was a respected and objective outsider who had developed an effective method of evaluating systems like MEDLARS, which was, at that time, searching by centralized processing of written requests in a batch mode. Members of the NLM Board of Regents opposed Dr. Lancaster's 1965 evaluation, concerned that the evaluation would damage the reputation of NLM if the evaluation revealed negative results. The Lancaster evaluation, however, proved to be objective and useful [29], revealing that MEDLARS searching operated with 58% recall and 50% precision and pointing out the need for more careful indexing and further vocabulary development [58]. William Hubbard Jr., NLM Board of Regents chairman (1965– 1967), recently stated that librarians completing MEDLARS searches needed additional training [59] to translate the patrons' demand searches into MEDLARS computer language and codes [47]. In response, NLM refined MEDLARS indexing and developed additional MEDLARS training programs [29].

The Lancaster evaluation resulted in one of the earliest publications with empirical bibliographic recall and relevance data [60] and became recognized as the first large-scale evaluation of any computer-based information system [61]. The evaluation was important for the development of MEDLARS II [47] and became a prototype for future information system evaluations [62].

The workload on the pioneering MEDLARS system— combined with the explosive growth of scientific literature, tremendous expansion of medical research, and medical specialization in the mid-1960s—had increased to the extent that MEDLARS was no longer able to efficiently provide the services users expected [4]. MEDLARS could not meet the two-day delivery requests for the demand search and recurring bibliographies. In fact, the average time from a user request to information delivery reached two weeks [45]. The time-consuming MEDLARS batch-mode searches [63] eventually exceeded the capacity of NLM's human resources dedicated to providing searches, forcing NLM to declare a six-week moratorium on new requests for demand searches [64]. Director Cummings described the times as a “cybernetic revolution bringing forth complex, sophisticated computers and other automated equipment to be applied to library development” [56]. Consequently, NLM began to plan for a new, upgraded MEDLARS system using new emerging computer technology.

NLM hired the Auerbach Corporation to draw up specifications for a new system for the online search and retrieval of bibliographic records [45]. Computer Sciences Corporation (CSC) was contracted to develop and support programming for MEDLARS II but failed to meet its schedule and cost estimates and provide the necessary searching capabilities as expected [65,66]. To remedy the problem, Director Cummings asked for support and moved to utilize System Development Corporation's (SDC's) work with Abridged Index Medicus/Teletypewriter Exchange Network (AIM-TWX), discussed below, as the way to the future of MEDLARS II. Director Cummings subsequently contracted with SDC to complete MEDLARS II [65]. Dr. Cummings recently added that the CSC design for MEDLARS II would only have resulted in a more efficient but out-of-date batch mode system. Therefore, the library decided to seek new designs and devote its resources to creating a complete online information retrieval system [29].

The new plans for MEDLARS II included a high-speed IBM 360/50 [66]. However, at that time, smaller computer companies complained that the government bought only IBM equipment and protested NLM's effort to acquire IBM equipment. Dr. Cummings recently pointed out that the NLM Board of Regents significantly contributed to the solution to the IBM problem. With the support of Alfred Zipf, NLM Board of Regents chairman (1969–70) and the Bank of America's vice president for technical affairs, who built the first computer to mechanize the banking industry, NLM successfully argued in favor of procuring IBM equipment for MEDLARS II [29].

The new MEDLARS II included the powerful, fast-operating IBM 360 series computer system, [45], an automated library system with an automated acquisition and cataloging system [71], an improved indexing and searching aid (online MeSH), an ability to provide chemical compound and toxicological searches, and a high-performance graphic image storage and retrieval system [19,47,72]. MEDLARS II, like the first MEDLARS, continued to be a huge database that stored all the references to the medical literature indexed by NLM since 1964. The new system also provided the technology to more rapidly deliver medical information for on-demand searches and continued to produce all of NLM's publications [73]. On January 3, 1975, the upgraded system was complete, and NLM's computer staff tested the MEDLARS II system and accepted it on behalf of the library [8].

The philosophy at NLM was to plan five years in advance [29]. True to this strategy, NLM was already thinking about a new MEDLARS-type structure [73] to keep pace with new technical developments for the improved management of the biomedical literature when MEDLARS II was completed [74]. In the summer of 1979, Director Cummings appointed Joseph Leiter, NLM associate director of library operations (1965–1983), to lead a team to prepare specifications for a new computerized system that would eventually become NLM's Medical Literature Analysis and Retrieval System Online (MEDLINE) [75]. The new NLM focus on the MEDLINE system would develop into the current MEDLINE database, NLM's premier bibliographic database with over fifteen million references. MEDLINE is presently the primary component of PubMed, which was developed by the National Center for Biotechnology Information (NCBI) at NLM and provides free access to MEDLINE [76].

NLM has continued to develop innovative methods to provide rapid access to the world's medical literature. During the era discussed, NLM's pioneering systems included the Index Mechanization Project with the time-saving Eastman Kodak Listomatic camera; MEDLARS, the computer-oriented bibliographic retrieval and publication system that included the innovative phototypesetter GRACE; AIM-TWX, the experimental online bibliographic retrieval system; and MEDLARS II, a high-performance retrieval system (Table 2). NLM's accomplishments regarding MEDLARS were cutting edge and placed the library at the forefront of incorporating mechanization and technologies into medical information systems. MEDLARS and its associated systems confirmed NLM's dedication to improving library and information services on behalf of society as a whole.

Table 2 MEDLARS timeline

The author is indebted to Dr. Martin Cummings, NLM director emeritus, for interviews from 2004–2006 and encouragement for this research; Robert Mehnert, NLM director of the office of communications and public liaison, and Kent A. Smith, former NLM deputy director, for verifying NLM facts; and Douglas J. Turk, Wanita Scroggs, Alicia Livinski, and Aubrey Saunders for bibliographic assistance.