This paper describes the data acquisition features of the PureMD Computerized Patient Record (CPR) software designed specifically for physicians. The physician uses a stylus to point, draw and handwrite in an Dynamic Dialog Interface that offers the same flexibility as the paper record and numerous other advantages. The clinical data thus entered is highly structured, easily legible and retrievable in many ways. The underlying Medical Knowledge-Base (MKB) was optimized for rapid, intuitive and consistent data entry and automatic coding with minimum handwriting.

16th Annual Symposium on Computer Applications in Medical Care SCAMC 92, Baltimore, November 8-11, 1992; American Medical Informatics Association, McGraw-Hill 1992, pp.261-264.

INTRODUCTION

The need for a Computerized Patient Record is increasingly recognized by various organizations, such as the Institute of Medicine Committee on Improving the Patient Record [1]. Direct data entry by the physician is fundamental for interactive decision support tools and remains a challenge. Fortunately, promising interface technologies are evolving and offer opportunities for new solutions. Some ambulatory care CPR software destined to physicians already use the mouse and a graphical-user interface for improved data entry and coding [2],[3],[4],[5]. While handwriting recognition technology was experimented with nurses [6] and dentists [7] on partial CPR applications, there is no report of using this technology for the complete CPR data entry by the physician.

PureMD is a complete CPR software especially designed for physicians and dentists that uses computer handwriting recognition and graphical user interface. It is being developed by Development Purkinje with a group of physicians, dentists and computer scientists from private clinics, universities and hospitals working in close association. The goal is to increase the doctors' efficiency with novel clinical data management.

METHOD

The software is being developed by Development Purkinje using successive incremental prototype method. The research started two years ago and is being sponsored by five hundred Canadian physicians and dentists.

The software and Medical Knowledge Base and their improvement are first conceptualized by a team of physicians and computer scientists. They are realized in as new prototype and tested by physicians in clinical simulations and finally tested in a real environment.

The medical software is designed and programmed by a team of twenty computer scientists. While the Medical Knowledge Base is developed by forty physicians (specialists and general practitioners) revised by five physician-analysts trained in knowledge engineering. All physicians are working clinicians in a variety of practice environment.

The development cycle ends with testing. Simulations are performed by another group of forty clinicians to validate the Medical Knowledge Base and the Dynamic Dialog Interface while field testing in private clinics started in autumn and is planned for an overall duration of one year and a half. Over three hundred physicians and dentists will participate in testing successive prototypes in experimental settings. The feed-back from the tests are relayed to the development teams for further improvements.

TECHNOLOGICAL CONSIDERATIONS

The PureMD software is an event-driven object-oriented application written in C++.

A UNIX server holds the reference MKB and the patient records will reside on optical disks (WORM). It also contains the programs that relays the MKB and patient records to the physicians' computers. The server's network uses TCP/IP transport protocol over Ethernet channels to communicate with the physicians' portable and desktop computers. Its communication server can also transmit and receive data from laboratories and billing agencies.

Another program runs on MS-DOS pen-based and desktop computers. The secretaries use a conventional keyboard and mouse for data entry, while the physician writes on a portable pen-based computer. It uses no keyboard and requires a stylus to handwrite, draw or select from the interactive screen. Built-in handwriting recognition (HR) algorithms can translate uppercase and lowercase block printing as well as numbers and into standard ASCII codes. Although relatively fast and accurate, the delays involved in HR are sufficient to impede the physician, if he is solely writing in block letters. To achieve the goal of a better computerized record than its paper counterpart, the objective of the research is automatic coding and an major decrease of writing, thus accelerating data entry and retrieval. The PureMD program is designed to display a selection of pertinent clinical items that the physician might decide to record. The program's choices of clinical items are related to the to the current patient. A matrix of data is presented to the physician for quick and precise record filling purpose and can be chosen by simply pointing, nearly eliminating handwriting. This feature is further described in the Dynamic Dialog Interface section.

MEDICAL KNOWLEDGE BASE DESIGN

The Medical Knowledge Base is structured according to the traditional patient record, it contains: a patient identification section, the chief complaints and present illness descriptions, the accompanying symptoms, the personal past medical history and family medical history, the review of systems, the social profile, the physical examination, the laboratory tests and procedures, lists of diagnosis, the investigations and therapies, patient advices and educational notes, follow-up considerations, a complex medication data base with interactions, etc. All of the above are detailed in a comprehensive and coherent logic as items of the MKB. We define an item as the finer granularity of a recorded information. The MKB already contains over 50000 different clinical observations.

Considerable efforts were applied to develop a uniform and logical model of every section of the MKB such as symptoms, signs, etc. The chronology of the symptom is a good example to illustrate the ingenuity needed for a successful natural interface. Because even a simple descriptive field as the chronology can be arduous to display. Simple acute symptoms only need a minimal chronology. While diseases with multiple level of chronology, as the cluster headache, require a complex syntax. A cluster headache might occur every night for two weeks, and every other semester since age 18. Multiple time periods are stated in the last description. One elegant solution could be to offer a single chronology data entry field flexible enough to generate either a convoluted or a simple description. A recursive chronology field based on an episodic symptom concept has been designed and is currently tested by physicians. In the same manner, many natural and intuitive solutions that appear effortless to the physician need careful design.

DATA ENTRY USING THE DYNAMIC DIALOG INTERFACE

To write medical observations, The physician first chooses the context of the encounter. The context can be any or a combination of the following statements: the chief complaint, a problem, a symptom, a syndrome, a follow-up visit, a treatment plan or a disease. The Dynamic Dialog Interface (DDI) consequently acts as a filter on the Medical Knowledge Base and displays the pertinent structured observations according to the previous choice(s). The data is displayed in a scrollable window according to the problem-oriented record-like format (Subjective, Objective, Assessment, Plan). It can be acted upon in numerous ways using the stylus device. 250 clinical context encounter filters have been written by physicians, revised and programmed and another 1000 are under development.

Pointing at a clinical item is the simplest way to record it. By selectively pointing different parts of an item, the physician is recording his statement as well as getting more details. In the latter case, the DDI searches the MKB and displays items from which the physician can choose to nuance his observations.

The doctor can also complete the record with a handwritten note or a drawing, anywhere in the record. Using special gesture recognition capabilities of the pen-based computer, the physician adds detail to an item and records it, all in one stroke. He thus specifies the absence of a symptom, an equivocal or negative sign or an unsuccessful procedure.

Specialized drawings and Graphical Metaphors (GM) are also part of the MKB and can be retrieved by the DDI in the appropriate context. They are used daily by the physicians and are the medical equivalents to shorthand notes. GM are a promising field of cognitive science [8] and merit to be implemented in new CPR. The stylus interface can further enhance the traditional drawings and GM: with automatic coding, they can be translated into words for external reporting. Examples of currently used GM are the Maigne and Lesage star drawing of the spine [9] and the visual recording of the joint examination [10].

In addition, the DDI can be customized for individual physicians without losing its automatic coding capacity. This allows the user the flexibility of writing his own protocols.

INTEGRATED RECORD OPERATION

There are many reasons for adopting a CPR, one of them is avoiding repetitions. Many program features abolish useless repetition. The medication is a good example of this: the physician prescribes a drug once and it is recorded in the time oriented record, added to the current medication profile of the patient, validated against other drugs for interactions and finally printed with education material to give to the patient. On the next encounter, the renewing of the medications is done by pointing at them with a special gesture.

Data exchange is another important gain. Laboratory request can be directly transmitted via modem. Automatic coding permits direct transmission of billing to insurance company or the proper provincial government in Canada. Data transfer is done in the appropriate format. ICD-9 and DSM-3R are currently used and the upcoming SNOMED III is under study.

ACCESSING THE CONTENT OF THE RECORD

The clinician may search the record in a variety of ways. It can be read in a traditional time oriented format (paper metaphor), or in a problem oriented format. Furthermore, it can be browsed according to different search criteria.

The program uses lists to query the record. The main lists are the problem and diagnosis list, the past history list, the allergy list, the past and current medication list, the laboratory list. These lists are automatically updated during data entry. These list compose the patient profile and can be customized by the user.

Time progression of any recorded item or group of items can be queried and displayed in different manners: textual, several column table-like, time oriented graphical, specific precooked, etc. The item might be as general as a symptom or a system or as precise as a positive clinical sign.

DISCUSSION

Extensive semiological refinements have been involved in the design of the MKB. The result is a better understanding of the medical terminology and the multiple levels of abstraction involved in the medical record. The improvement of the patient record can favorably affect the cognitive decision process of the physician. A more comprehensive record might therefore translate into a better understanding of the recording process.

The immediate benefits from using such a system are a legible record, faster data entry and improved retrieval tools, automatic coding (for clinical research, billing purposes, laboratory data transfer, etc.), the simultaneous use of decision support tools and a notable reduction of repetitive tasks.

In conclusion, the CPR model described in this article can offer the same flexibility as its paper counterpart with major improvements.

Acknowledgments

Dr Roger A. Cote and Dr Clement J. McDonald deserve special credit for contributing helpful information on the upcoming SNOMED version and ASTM-1238-91 Standard respectively.