0031-3955/92 $0.00

PEDIATRIC EMERGENCY MEDICINE

+ .20

COMPUTERIZED DIAGNOSTIC REFERENCING IN PEDIATRIC EMERGENCY MEDICINE Joseph E. Simon, AB, MS, MD

We live in the information age. This age is characterized not only by an explosion of knowledge but also, thanks to the microchip, by revolutionary ways of manipulating that knowledge. A secretary can spell check a document in seconds. A ward clerk can produce a graph of a patient's serum sodiums over the last four admissions with a few key strokes. Yet the vast body of medical science is brought to bear on the care of the individual patient in a manner as yet largely uninfluenced by the methods of the information age. Now, as it was 50 years ago, the patient is virtually totally dependent on the memory, experience, and intellect of the treating physician, supplemented by a cumbersome array of written references that are out of date by the time they are printed. Clinical science needs to join the information age. Pediatric emergency medicine just might be one of the most ideal clinical disciplines to lead the way. This article discusses the reasons for this and reviews some of the past and recent experiences with computer applications to clinical medicine, with special emphasis on computerassisted diagnosis and computerized diagnostic referencing. COMPUTER APPLICATIONS IN MEDICINE

Table 1 lists many computer applications in medicine along with examples. A quick review of this list should convince the reader that the computer already has had a dramatic impact on both the science From Pediatric Emergency Services, Scottish Rite Children's Medical Center, Atlanta; and Medical College of Georgia, Augusta, Georgia

PEDIATRIC CLINICS OF NORTH AMERICA VOLUME 39 • NUMBER 5 • OCTOBER 1992

1165

1166

SIMON

Table 1. SELECTED COMPUTER APPLICATIONS IN MEDICINE Application

Example

Administration Patient registration/accounts Medical record storage Medical reference storage Quality assurance Medical education Diagnostic radiology Research Therapy Data transmission Medical diagnosis

Tracking credentialing data Billing Tracking laboratory data Literature searches by key words Tracking outcome data Simulated case histories CT scanning Statistical analysis of data Tracking drug interactions Teleradiography EKG interpretation

and practice of medicine. It is also evident from this list that most of these applications have occurred without the active participation of the practicing physician, however. Two other observations regarding Table 1 are of interest. First, it is, in my judgment, a commentary on our medical system that one of the most prolific applications of computer technology to medicine has been in the area of patient registration and patient accounts. Although a patient's diagnosis and therapy are not yet supported by the power of the computer, the patient's bill is generated in seconds with 50 mHz of power! A second observation concerns the potential power of the computer to perform quality assurance analyses. (Forgive me, I should have said continuous quality improvement [CQI] analyses!) It is already commonplace for CQI monitors to receive computer printouts of return visits, readmissions, and diagnosis related group outlyers and to trend these by physician. As the medical record becomes more computerized, the potential for the computer to identify and trend deviations from standard practice and expected outcomes will increase exponentially. This trend may ultimately be a major force pushing physicians to use the computer in their daily clinical practice. Better to work with a computer in a patient's care rather than to wait for it to work against you after that care has been rendered! Certainly, insurance companies will adopt this perspective even if physicians do not. COMPUTER-ASSISTED DIAGNOSIS: THEORY

There are four theoretic approaches to computer-assisted diagnosis: bayesian logic, patient data banks, algorithms, and computerized diagnostic referencing. The last two approaches are conceptually similar but are discussed separately for the sake of clarity. The first two approaches are unrealistic at this time but are mentioned briefly for the sake of completeness. Bayesian Logic

Bayesian logic is the best mathematical model currently available to mimic the thought processes of the physician. The foundation of bayesian logic is the bayesian theorem:

COMPUTERIZED DIAGNOSTIC REFERENCING

1167

The bayesian theorem describes how the conditional probability of each of a set of possible causes (diagnoses), given an observed outcome (symptom, sign, lab), can be computed from the knowledge of the probability of each cause and the conditional probability of the outcome given each cause. (In plain English, this actually means the following: If the incidences of diseases D and E are known, and if the frequencies with which symptom S occurs in diseases D and disease E are known, then when a patient presents with symptom S, the relative likelihood of disease D versus disease E can be determined. )

To apply the bayesian theorem to medical diagnosis, the probability of the various diagnoses within a given population and the probability of the many different signs and symptoms for each diagnosis must be known. Although these relationships are, in fact, often known or can be reasonably estimated, the problem becomes more complex as more symptoms and signs are considered. For example, although the medical literature may contain information regarding the frequency with which symptom X occurs in disease D, it becomes more difficult to know or estimate the frequency with which symptom X occurs in disease D when symptom Y is present or when symptoms Y and Z are present, and so forth. Yet all of these relationships would need to be known or estimated before bayesian logic could be applied. Despite such a major impediment to its application to medical diagnosis, bayesian logic has been successfully used to a limited extent in selected areas. One notable success has been in the diagnosis of abdominal pain by Dombal et aF and Horrocks et a1. 3 This program was developed over 20 years ago and is actively used in various parts of Great Britain. Recently, Baxt4 developed and tested a diagnostic computer program for adult patients presenting with suspected myocardial infarction. This ingenious program used an artificial neural network to learn from its experience with previous patients. Essentially, the program used its experience to develop the mathematical relationships between symptoms and diagnoses discussed earlier in the bayesian theorem. After "learning" on 351 patients, the program was tested on 331 patients, outperforming experienced clinicians by correctly identifying 97.2% of patients with myocardial infarction versus the 77.7% success rate of the physicians. Patient Data Bank

The patient data bank is a more realistic method of computerassisted diagnosis that awaits computerization of the medical record. When that occurs, it is not difficult to imagine large patient data banks that could be easily queried to provide diagnostic possibilities ordered by frequency of occurrence. For example, one could present the data bank with pertinent patient information (e.g., 6-year-old child with fever and back pain). The computer could search the large patient data bank and print out the diagnoses of all patients in an arbitrary age range (ages 4-10) whose symptom complex included fever and back pain, ordered by frequency of occurrence in that data bank.

1168

SIMON

Algorithms

Algorithms do not require a computer for storage and application. The computer might be the most convenient way of storing and continuously updating complex clinical algorithms, however. The drawback with algorithms is that they involve break points that represent all-or-nothing decision points. There is no room for "maybe" or "rarely." The authors of an abdominal pain algorithm, for example, must pick an upper age limit for the diagnosis of intussusception. If they pick age 5, then the rare 6-year-old child with intussusception will be missed. Any attempt to build all possible exceptions into an algorithm results ultimately in the fourth and final option for computerassisted diagnosis.

Computerized Diagnostic Referencing

This approach is simply the storage of a set of diagnoses along with all of the symptoms and signs and ages ever reported to be associated with each diagnosis. This may sound like an impossible task. In fact, many attempts have been made to develop such data bases. Indeed, one of the earliest and, in some respects, most comprehensive of all attempts was made in the field of pediatrics over 20 years ago by Simmons and Worley. 1, s. 6 Their work and experience are reviewed in some detail as a prototype of a computerized diagnostic reference. The data base compiled by Simmons and Worley, called the MEDITEL-pediatric system, contained over 1500 diagnoses along with their symptoms and signs. Little information is provided in their published work regarding exactly how the data base was compiled. It was possible to access the data base at a cost of $50 per search. The user could designate an age, sex, and set of symptoms and signs. The computer would search for "matches," and a list of possible diagnoses would eventually be printed out and supplied to the caller. Although never explicitly discussed in their published reports, it can be inferred from the information provided that the MEDITEL-pediatric system did not require a perfect match for a diagnosis to be included in the final list. For example, if a diagnosis fulfilled four of the five symptoms and signs being searched, it might appear in the final list. The creators of this data base and software viewed their program as a diagnostic tool akin tp performing a lumbar puncture rather than as a reference. Accordingly, they referred to it as computer-assisted diagnosis and proceeded to systematically study its efficacy when used as part of a patient examination. When applied to patients admitted without a definitive diagnosis, the list generated by their program (often up to 70 possibilities) successfully included the correct diagnosis with 70% to 90% reliability. MEDITEL-pediatric matched up well with experienced physicians who also developed an initial list of possibilities. 6 Furthermore, when the lists were provided to the admitting physicians there

COMPUTERIZED DIAGNOSTIC REFERENCING

1169

was a trend (not statistically significant) toward shorter hospital stays, decreased use of consultants, and a decreased use of costly laboratory studies. 6 This was 20 years ago. Why then has computer-assisted diagnosis failed to permeate daily clinical practice? There are many reasons. Until recently cost was a factor. Tradition and computerphobia, two conditions medical schools have done little to counteract, are in large measure responsible. In my judgment, however, one of the major reasons for the failure of computer-assisted diagnosis to be accepted by physicians is that it has been promoted and viewed as a diagnostic procedure or test, not unlike performing a serum sodium. As such, although its sensitivity was respectable (ability to list the correct diagnosis), its specificity was poor (up to 70 diagnoses in a list) and it was time consuming and costly to perform. In my judgment, diagnostic computer programming should not be viewed as a bedside procedure but as a medical reference. It makes no more sense to ask the accuracy of a diagnostic computer program in determining a patient's diagnosis than to ask the same question of Ludwig and Fleisher's Textbook of Pediatric Emergency Medicine. 2a Diagnostic computer programming's rightful place is as a reference, not a diagnostic test. As such it offers many unique advantages compared to traditional hard copy references with respect to the storage, retrieval, and updating of diagnostic medical information. Accordingly, the remainder of this discussion uses the term computerized diagnostic referencing.

COMPUTERIZED DIAGNOSTIC REFERENCING IN PEDIATRIC EMERGENCY MEDICINE

MEDITEL-pediatric contained 1500 diagnoses and included nonurgent as well as urgent and emergent conditions. The length of a pediatric emergency medicine physician's list of "must know" diagnoses is probably in the range of 300 to 400. Searches performed on a data base of diagnoses of this siie often yield 10 to 30 possibilities, even when quite rare conditions are included. This is one reason why this author suggested at the outset that pediatric emergency medicine might be an ideal field for the application of computerized diagnostic referencing. There are other reasons as well, however. The pediatric emergency medicine physician simply does not have the time to read or even scan one or more textbook chapters or original articles to search out diagnostic information. When pressured, as he or she often is by high patient volume, disturbing clinical situations, fatigue, and sleep deprivation (the list could go on and on), the pediatric emergency physician needs a reference that organizes and displays in seconds diagnostic information immediately relevant to the clinical presentation. Computerized diagnostic referencing has the potential to fulfill this need.

1170

SIMON

Application

For the past year a computerized diagnostic data base in pediatric emergency medicine has been compiled and used in the pediatric emergency department at Scottish Rite Children's Medical Center in Atlanta, Georgia. This system is called PEM-DXP. The data base contains more than 400 diagnoses and more than 800 symptoms and signs. Approximately 140 of the symptoms and signs have been extensively researched to guarantee the inclusion of a relatively complete emergency differential diagnosis for each of those symptoms and signs in the data base of diagnoses. Once constructed, the data base was refined by challenging it with all pediatric emergency clinical pathological conferences published in several major pediatric journals and other references in the past 10 years (approximately 150 cases) and an equal number of cases from the Scottish Rite Children's Medical Center Emergency Department. A search can be performed in 60 to 90 seconds using the patient's age and up to ten symptoms and signs. (In practice, only three or four symptoms and signs are generally used.) Most searches generate lists of 10 to 30 diagnoses. In addition, each diagnostic possibility may be rapidly displayed along with key diagnostic information-associated symptoms, signs, pertinent epidemiologic factors, relevant laboratory tests and radiographs, and other diagnostic keys. Figure 1 is a reproduction of the screen used to query the data base regarding a specific clinical presentation, in this case a 6-year-old child with back pain (B-PN*) and fever (FVER*). The asterisk indicates that the symptom has been researched to guarantee that a complete differential diagnosis is contained in the data base for that symptom. Symptoms whose emergency differential contains fewer than 40 diagnoses are designated specific symptoms and must be entered in the specific symptom fields. A symptom with more than 40 diagnoses in

Diagnosis Query Screen Patient Age,

6.00

Specific Symptom Codes, 1 B-PIP 2 6 7

3

4

5

B

9

10

Nonspecific Symptom Codes: 1 rVER" 2 6 7

3 8

4

9

5 10

Specific Sign Codes, 1 2 6 7

B

4 9

10

Nonspecific Sign Codes, 1 2 6

B

4 9

5 10

3

3

Specific Lab Codes, 1 ____________-==-________ 3 4 =--= -=____________ ATS Inc.

b-------__

Query database,Change querY,Backup,eXit Query(Q/C/B/X)?

Figure 1, Diagnosis query screen.

l'

COMPUTERIZED DIAGNOSTIC REFERENCING

1171

its emergency differential is designated nonspecific and so entered on the diagnostic query screen. Figure 2 (Diagnosis Matchup Screen) is a reproduction of the emergency differential diagnosis generated for the previous clinical presentation by PEM-DXP. The possibilities are listed in three groups: common, infrequent, and rare, based on the overall incidence of the illness, not the likelihood for the given presentation (fever and back pain) being searched. OA and OA2 signal the user that for the so designated diagnoses, although possible, age 6 is not a likely age for those diagnoses to present. From the diagnosis matchup screen the data base for any of the listed diagnoses may be displayed. Figure 3 is a reproduction of the data base for Lyme disease, stage 1 (LYMEl). The advantages of PEM-DXP over traditional diagnostic reference materials are numerous and are listed here. 1. Data Organization. This is the premier advantage of computerized diagnostic referencing. A hard copy reference might contain a list of causes of back pain in a pediatric patient. Extracting a differential diagnosis for a 6-year-old child with fever and back pain from traditional references will be time consuming and cumbersome, however. With PEM-DXP the computer can construct this differential diagnosis while the physician charts the history and physical examination. 2. Updating. As new diagnostic information becomes available (a case report or review article covering the diagnostic approach to a symptom or presentation) it can be immediately added to PEM-DXP. Retrieval of that diagnostic information no longer

rr===Ma=t=c=h=e=s=:==29===== Diagnosis Matchup Screen = = = = = = = = = = = ; 1

i

I I

Diagnosis

Freq

HEMOL MENIN MYCO OBSTU OSTEO PLEUR PNEUM S-ART SCELL UTI ABABS BONET CHOLE ENTAM LEUK LYME LYME 1 N-BLA PANC TB WILMS ALNEP BRUC CCYST DISC EVANS HYRCA LYMPH T-MYE

C C C C C C C C C C I I I I I I I I I I I R R R R R R R

R

Age

Flag

Descriptions

hemolysis

meniniitis

6

m co p a5ma pneumoniae

o structive uropathy

osteom~elitis

pleuri is pneumonla

septic arthritis sickle cell anemia





urinary tract infection intra-abdominal abscess bone tumor

cholecystitis-acalculus entamoeba histolytica leukemia









lyme disease lyme disease-stage 1

neuroblastoma

~ancreatitis

uberculosis

Wilms tumor acute lobar nephronia brucellosis choledochus cyst discitis

~~~~~c;rg~;~~e lymphoma

transverse myelitis

PgUp or PgDn, [F5] Dlsplay,[F6] Quest,[F7] Flag or to return

Figure 2. Diagnosis matchup screen.

ATS Inc."

1172

SIMON

Diagnosis Screen

Diagnosis Code.

Diagnosis Beginning Beginning Beginning

LYME1

Dese: lyme stage 1Frequency. I Age. 2.00 Ending Age. 21.00 ather Age. 0.00 Ending Other Age. Other Age2. 0.00 Ending Other Age2.

Specific Symptom Codes: 1 rash - EeM 3

5 7 9 11 13 15 17

myalgia* arthralgia* rash - maculopapularscrotal swelling* personality change* sore throat'"

Nonspecific Symptoms Codes: 1 fever-N'" 3 headache-N° 5 confusion-N* 7 dizziness-H* 9 chills-N 11 ataxia-H* 13

2.00 0.00

2 tick bite 4

6 photophobia" 8 rash-urticarial'" 10 rash - marginated'" 12 red eye* 14 back pain'" 16 amnesia 18

2 lethargy-N 4 vomiting-H* 6 8 fever-prolonged-N" 10 12

Specific Sign Codes; 2 rash-maculopapular'" 1 rash-ery. chronicum mig. 3 conjunctivitis· 4 rash-marginated'" 6 muscle tenderness· 5 rash-urticarial* 8 pharyngitis" 7 scrotal swelling" 9 RUQ tenderness 19 photophobia" 12 amnesia 11 splenomegally" 13 facial weakness· 14 15 16 17 18 ~__________________________________________________________ ATS Inc

PgUp or PgDn to display pages. to continue.

Figure 3. Diagnosis screen. I/Iustration continued on opposite page

depends on the physician's ability to recall the content of a particular article or its location. In fact, it is only necessary that one physician in a group using PEM-DXP provides frequent updating for PEM-DXP in order to keep his or her colleagues current in the area of recent diagnostic developments. 3. Availability. PEM-DXP may be loaded on any personal computer and transported easily by 3.2S-inch diskette. 4. Instruction. Traditional reference materials do not readily lend themselves to the conference or small group setting. Here, too, time is a critical factor. It becomes unrealistic to interrupt student or resident instruction to page through texts to add to a differential diagnosis or answer a question related to diagnosis. PEMDXP is a reference that is well adapted for instructional use. Pertinent information may be rapidly accessed. Instructors and students can interact with the computer in a manner that adds another dimension to clinical instruction. 5. Charting. PEM-DXP not only develops a differential diagnosis in 60 seconds but it also produces a list of symptoms and signs

COMPUTERIZED DIAGNOSTIC REFERENCING

Diagnosis Screen Nonspecific Signs Codes, 1 fever-H3 lymphadenopathy-gen.-H· 5 nuchal rigidity-H· 7

9 confusion-H* 11

1173

Page 3

2 lethargy-H 4 lymphadenopathy-Iocal-N· 6 hepatomegally-H· 8 10

12

13 ataxia-HLab Codes, 1 SGOT 3 ESR

2 LP 4

5

EPIDEMIOLOGICAL CORRELATES. (see also LYME) *Ticks that have been implicated: -Deer tick (small) -Lone Star tick (found on many animals) ·Only 25\ recall a tick bite ·Summer/fall ·Wooded areas

·Found throughout the US but predominantly on the Northeast coast, northern Midwest, West coast (California and Oregon), and Southwest (Texas and Hevada) *Incubat,on period. 4-20 days after the tick bite ADDITIONAL DIAGNOSTIC KEYS. *R-ECM - annular rash at the tick bite. not painful or pruritic

t

enlarges

with central clearing, 50% have additional annular lesions;

associated with malar erythema, rash at the site of the tick bite may be atypical: necrotic, vesicular. papular ·Stage 1 lasts for 1-4 weeks. may have a constant or relapsing course,

rash and constitutional symptoms predominate but neurological symptoms have been described 1...._ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ATS

Inc.~

[Escape] Exit Memo Display

Figure 3. (Continued)

that need to be pursued relative to that list of diagnoses in this same 60 seconds. The result is a much more detailed history and physical on the patient at hand.

SUMMARY

A well-qualified and dedicated pediatrician in a mid-Atlantic state recently cared for a 4-year-old child with a fever and a truncal rash. On the second visit the child was lethargic but seemed to improve with oral hydration and fever control. Twenty hours later the child's rash became petechial and the child returned moribund and died shortly thereafter. Looking for an emergency differential diagnosis on a 4-yearold child with fever and a truncal rash would have been cumbersome using traditional reference materials. With PEM-DXP, however, a nurse or secretary could have produced for the physician in 60 seconds a list

1174

SIMON

of 12 emergency diagnoses to be considered. Rocky Mountain spotted fever would have appeared on that list. (Ten percent of patients with Rocky Mountain spotted fever have a maculopapular rash, and in 10% of those children the rash begins on the trunk.) This list might have prompted this physician to ask about a tick bite. A positive response would have been obtained. Possibly the tragic outcome, for both the child and the physician, could have been avoided. Children with urgent complaints are frequently cared for by physicians who are busy, tired, sleep-deprived, and stressed in many other ways. These physicians often have not read last month's medical journals in their specialty or even last year's journals. Yet the parents' expectation, and rightly so, is that their pediatrician or emergency physician has at his or her command and is bringing to bear on the care of their child the vast knowledge base of medical science as it pertains to their child's condition. It is a totally unrealistic expectation. By computerizing reference materials in ways that allow for selectively searching for and compiling information pertinent to a given presentation, however, we can move a bit closer to meeting both the parents' expectations of us and our own expectations of ourselves. References 1. Barness LA, Tunnessen WW, Worley WE, et al: Computer-assisted diagnosis. Am J

Dis Child 127:859, 1974 2. Dombal FT, Leaper DJ, Staniland JR, et al: Computer-aided diagnosis of acute abdominal pain. Br Med J 2:9-13, 1972 2a. Fleisher GR, Ludwig S: Textbook of Pediatric Emergency Medicine. Baltimore, Williams and Wilkins, 1988 3. Horrocks JC, McCann AP, Staniland JR, et al: Computer-aided diagnosiS: Description of an adaptation system, and operational experience with 2,034 cases. Br Med J 2:59, 1972 4. Baxt WG: Use of an artificial neural network for the diagnosis of myacardial infarction. Ann Intern Med 115:843, 1991 5. Swender PT, Tunnessen WW, Oski FA: Computer-assisted diagnosis. Am J Dis Child 127:859, 1974 6. Wexler JR, Swender PT, Tunnessen WW, et al: Impact of a system of computerassisted diagnosis. Am J Dis Child 129:203, 1975

Additional Reading 1. O'Shea JS: Computer-assisted pediatric diagnosis. Am J Dis Child 129:199, 1975 2. Lustig JV, Groothuis JR: Computer applications in pediatrics. Adv Pediatr 31:295, 1984 3. Waxman HS, Worley WE: Computer-assisted adult medical diagnosis: Subject review and evaluation of a new microcomputer-based system. Medicine 69:125, 1990

Address reprint requests to Joseph E. Simon, AB, MS, MD Scottish Rite Children's Medical Center 1001 Johnson Ferry Road Atlanta, GA 30342

Computerized diagnostic referencing in pediatric emergency medicine.

A well-qualified and dedicated pediatrician in a mid-Atlantic state recently cared for a 4-year-old child with a fever and a truncal rash. On the seco...
997KB Sizes 0 Downloads 0 Views