The significance of cognitive modeling in building healthcare interfaces

International Journal of Medical Informatics (2006) 75, 163—172

The significance of cognitive modeling in buildinghealthcare interfaces

Constance M. Johnsona,b,∗, James P. Turleyb

a The University of Texas M.D. Anderson Cancer Center, 1515 Holcombe Boulevard, #447 Houston,Texas 77030, USAb The University of Texas Health Science Center at Houston, School of Health Information Sciences,Houston, Texas, USA

Received 7 June 2005; accepted 8 June 2005

KEYWORDSNursing informatics;Medical informatics;Cognitive modeling;Healthcare systems;Human—computerinteraction

SummaryBackground: Although there are many reasons that widespread adoption of health-care information systems has not transpired, one reason is a failure to take intoaccount the cognitive needs of the users.Aim: To understand the cognitive needs of nurses and physicians and determine howthese needs should influence the design of healthcare interfaces.Design of study: A qualitative and quantitative study that compares how nurses andphysicians comprehend patient information.Setting: Twenty-four registered nurses and twenty-four physicians working in thespecialties of gastrointestinal or internal medicine.Methods: Each clinician reviewed two mock electronic medical records and sum-marized the cases using a think-aloud protocol. All verbalizations were coded formedical and conceptual information.Results: The nurses included a larger mean proportion (p < 0.001) of recalls than didthe physicians. As compared to the nurses, the physicians included a statisticallysignificant (p < 0.001) larger mean proportion of inferences, conditional statements,and interventions. The nurses concentrated on functional problems, whereas thephysicians focused on diagnosis, treatment, and management.Conclusion: The main cognitive differences between the physicians and the nursesare explained through the differences in their practice models. Therefore, health-care IT must develop separate interfaces for each discipline to address their uniqueneeds.© 2005 Elsevier Ireland Ltd. All rights reserved.

∗ Corresponding author. Tel.: +1 713 794 4177;fax: +1 713 563 4242.

E-mail address: [email protected](C.M. Johnson).

1. Introduction

The current healthcare delivery system in Americasuffers from substandard quality, partly the resultof the relative absence of a healthcare information

1386-5056/$ — see front matter © 2005 Elsevier Ireland Ltd. All rights reserved.doi:10.1016/j.ijmedinf.2005.06.003

164 C.M. Johnson, J.P. Turley

infrastructure [1]. The Committee on Quality ofHealth Care in America concluded that informa-tion technology (IT) must perform a pivotal partin the redesign of the American healthcare sys-tem, including the widespread automation of clini-cal records [1]. Although IT offers the potential tochange the way healthcare is conducted, its abil-ity to adapt to the many facets of the healthcareworkplace will determine the success of healthcareIT. The success of universal healthcare IT imple-mentation will depend not only upon the function-ality of the system, but also upon its usability oruser-friendliness and its ability to add value to thehealthcare workplace.

There are several reasons that electronic med-ical records (EMR) have not been widely adopted.Not only have there been financial, organizational,and technological problems [2], but there hasalso been a deficiency of user-centered interfaces,which has been cited as a major obstacle to accep-tance and standard use of healthcare informationsystems [3—6]. The user-interface deficiencies pri-marily stem from a lack of understanding of thecognitive needs of the users, and a failure tofully take into account human—computer interac-

influences their comprehension. This interactiondirects their problem solving and decision-makingin clinical tasks and settings. Problem solving anddecision-making are dependent upon each person’scollection of knowledge, their comprehension, andtheir familiarity with each situation [10].

User and task analyses are two different waysto learn about the users of a healthcare applica-tion. The user analysis profiles the characteristicsof the users, and the task analysis identifies systemfunctions that must be performed as well as proce-dures and actions that must be carried out in orderto achieve the task goals of the users. One aspectof the user analysis addresses the cultural differ-ences among groups of individuals, such as physi-cians and nurses, who will be using an application.Although examining discipline-based groups, whichare not traditionally defined as having cultural dif-ferences is unusual in a user analysis, doing so isappropriate for the healthcare domain. Actually,this type of analysis is a combination of user andtask analyses. The healthcare domain has a widevariety of practitioners, and should be considereda culturally diverse population for user/task anal-ysis. For example, the manner in which physiciansadtbdlmvps

cmcictdAtAottctms

ph

tion [3]. Patel and Kushniruk (1998) proposed thatadditional basic research is needed in understand-ing the users, their work activities, and their rea-soning processes in order to adequately addresstheir cognitive needs [7].

To understand the cognitive needs of the usersand the human—computer interaction, we havefocused on two interacting components; the usersand their tasks. Specifically, we have examined howthe different backgrounds and tasks of nurses andphysicians affect their comprehension of patientinformation. Understanding the cognitive processesof nurses and physicians, and the differencesbetween the two, provides basic information thatwill assist in the design of ideal medical interfaces.

Successful software reflects the users’ goals,tasks, and processes. Understanding the cognitivemodels of the users is the first step in the soft-ware development from a cognitivist perspective.The cognitivist approach ascertains the user’s cog-nitive model of the work domain and then designsthe interface to be complementary and consis-tent with that model [8]. Determining the cog-nitive models of the users such as clinicians inthe course of their task accomplishments such aspatient care, planning, and evaluation, is essential.This domain understanding is particularly impor-tant when the bulk of the tasks is cognitive innature, such as decision-making [9]. The cognitivistapproach understands that the interaction betweena person’s background knowledge and their tasks

nd nurses gather, interpret, and use patient dataiffers due to the differences in their approacheso patient care. Whereas a physician’s approach isased upon the medical model which centers oniagnosing, treating, and managing medical prob-ems, a nurse’s approach is based upon the wellnessodel, which focuses on identifying high risk indi-

iduals, diagnosing functional problems, observinghysiological status, and reporting changes in thattatus [11].

Understanding the different practice patterns oflinicians is crucial in ensuring that the operationalodel matches the discipline’s model. A pilot study

onducted to determine the impact of a computer-zed medical system in an out-patient setting con-luded that one reason the physicians abandonedhe project was that the documentation templatesid not match their practice requirements [12].mismatch between system functions and prac-

ice models may cause system failure. Brennan andnthony outline how the nursing practice model canffer direction in the design of healthcare informa-ion systems by understanding the roles of nurses,heir communication patterns, their required clini-al content, and their administrative reporting sys-ems [13]. Understanding all healthcare providerodels is necessary in order to adequately define

ystem functionality and usability.Differences in the various types of clinician

roblem-solving skills within the area of medicineave not been well defined. Although many studies

Cognitive modeling in healthcare interfaces 165

have focused on the mental process of physi-cian problem-solving [14—18], few studies havefocused on the mental process of nurse problem-solving. These nursing studies focused on the differ-ences between experts and novices within specificdomains [19—21].

While there is an understanding about the differ-ences between the roles of physicians and nurses,no empirical studies document the differencesbetween physicians and nurses in their understand-ing of patients. A review of the literature showedthat the majority of research on the roles of physi-cians and nurses focuses on their divergent pro-fessional roles as exemplified in the cure versuscare paradigm [22,23]. This paper examines thecognitive tasks of the users from a cognitive per-spective. Healthcare users have many differentneeds, capabilities, and background knowledge. Toensure that the design of healthcare informationsystem matches the tasks of the intended users, webegin with a basic approach of the cognitive taskanalysis.

2. Methods

Tcccsaumpw

2

Tjw2oSwEdbupoeas

Table 1 Demographics of all subjects

Demographics RN MD

Male 1 (4%) 18 (75%)Female 23 (96%) 6 (25%)Average age 45.41 ± 7.45 36.39 ± 10.34Range 28—60 26—61Associates degree 7 (29%)Diploma degree 4 (17%)Bachelors degree 10 (42%)Masters degree 3 (12%)MD degree 24 (100%)

Average practiceyears

20.62 ± 7.35 11.00 ± 10.06a

3.58 ± 1.58b

Range 5—32 1—29a

1—6b

a Attendings: physicians in practice.b PGY: resident physicians.

study was obtained from the relevant universityinstitutional review boards.

The study was conducted using a laptop com-puter in a quiet, private room at the university,or in the private offices of individual clinicians.Both settings allowed the subjects to think out-loud as they read, interpreted, and summarized themedical cases in an undisturbed setting prevent-ing breaches in subject confidentiality. The estab-lished methodology used in determining how peo-ple mentally represent information involves eitheranalyzing text documents or interviewing experts,and thus encoding knowledge and ideas only afterthey have been orally expressed [24]. This has beenaccomplished through the use of the think-aloudtechnique [25] using propositional analysis [26], aformal methodology in cognitive science for repre-senting textual information. Although the subject’sverbalizations were audio taped, their faces werenot videotaped.

Each clinician (24 nurses and 24 physicians)reviewed two cases for a total of 96 case reviews.Nineteen subjects (40%) were male and twenty-ninesubjects (60%) were female. The mean age of thesubjects was 41 ± 10.02 years with an age range of2ocrdttsf

f

his is a qualitative and quantitative study thatompares how physicians and nurses comprehendlinical information. The study uses verbal proto-ols to elicit the cognitive models of patients con-tructed by physicians and nurses. A text-basednalysis was used to identify and compare ideanits. The physicians and nurses reviewed and sum-arized two cases: a gastroenteritis case and aancreatitis case. Their summations of the casesere analyzed and are presented herein.

.1. Population, study setting and sample

his study solicited a purposeful sample of 48 sub-ects consisting of 24 practicing registered nursesorking in the area of gastrointestinal nursing and4 practicing physicians with practice specialtiesf gastrointestinal medicine or internal medicine.ubjects were recruited through advertisement asell as formal and informal presentations. Non-nglish speaking was the only exclusion criteriaue to potential variance in comprehension inducedy linguistic factors. No one was excluded basedpon ethnicity or gender. Due to concerns regardingroblems with recruitment and time, the numberf practice years of each subject was not consid-red during recruitment and thus was considereddelimitation. All subjects were required to give

igned informed consent. Approval to conduct this

6—61 years. The mean practice years and rangef practice years were separated within the physi-ian group to illustrate the differences between theesidents and attending physicians. However, theseata were not further stratified in the remainder ofhese analyses, as the subgroups would have beenoo small to show statistical significance. Table 1hows the details of the demographic informationor both the nurse and physician groups.

In the nurse group, 96% of the subjects wereemale; and in the physician group, 75% of the


Fig. 1 Pancreatitis case: history and habits.

subjects were male. This distribution is very sim-ilar to the national gender distribution of nursesand physicians. Nationally, 95% of nurses are femaleand 76% of physicians are male [27,28]. In thisstudy, on the average, the nurses were older thanthe physicians, 45 years old versus 36 years old,respectively. There was also dissimilarity in therange of practice years between the nurses andthe physicians. The nurses had a greater meannumber of practice years than did the attendingphysicians, 20.62 ± 7.35 years versus 11.00 ± 10.06years, respectively.

2.2. Case construction

A total of three gastrointestinal cases were con-structed using only fictitious data. No real patientdata was used. The diagnoses of the cases wereappendicitis, gastroenteritis, and pancreatitis. Themock cases contained 12 main sections: demo-graphic; history of present illness; histories andhabits; current medications; allergies; review ofsystems; vital signs; input and output; physicalexam; nursing notes; assessment; and initial physi-cian orders. Once these mock cases were con-

2.3. Case display

Once the cases were formalized for subject review,the cases were inserted into Microsoft Access andformatted for ease of subject review. Fig. 1 showsa screen shot of the pancreatitis case, historyand habits section. Navigation through the sectionscould easily be accomplished in any order throughlabeled buttons on the left side of the screen.

2.4. Procedure

Subjects were randomized to review either thegastroenteritis or pancreatitis case first. Counter-balancing was used to control for order effects.Subjects were given one training case (an appen-dicitis case that was not used in any data analy-ses) and two gastrointestinal test cases to reviewand summarize. Subjects were informed that theyshould think-aloud [25] while reviewing the cases,and once they finished their review of each casethey would need to summarize the case out-loudas they normally would to a colleague. Subjectswere not allowed to take notes or to look at themock medical record during summation. The sub-jtttip

structed, a gastroenterologist, medical internist,and two medical/surgical nursing faculty reviewedand corrected the cases for content accuracy. Thecorrected cases were used in the study. The appen-dicitis case was used as a training case and notincluded in the data analysis.

ects were given instructions on the think-aloudechnique which included saying out-loud every-hing that one is thinking or would normally sayo oneself silently. After these instructions weressued, the subjects were given two problems toractice thinking-aloud.


2.5. Data collection and quality

Demographic data, field notes containing subjectcomments, and digitally audio taped verbal proto-cols were collected. The demographic informationincluded gender, age, education level, occupation,and years in practice. The field notes included com-ments made by the subjects after they completedthe experiment.

The digital audio recordings of the subjects’summaries were transcribed into a Microsoft Wordtext file. Once the summaries were transcribed,the text was separated into individual sentencesand numbered sequentially. The sentences werethen divided into idea units or propositions. Propo-sitions are considered hypothetical units that rep-resent the semantic content within the principalframework. They usually consist of a relation suchas a verb, adjective, and adverb and argumentssuch as nouns [29]. For example, the sentence,‘‘her oropharynx showed dry mucous membranes’’is analyzed as one proposition or one idea unit. Thetext of the original cases was dissected in the samemanner as that of the subjects’ summaries for com-parison. The subjects’ propositions were matcheda

2

Dcraevit‘difsebtbN[mItaFdTp

Table 2 Examples of coding scheme

Propositions Code

32.1 He has some tremors Recall53.1 Patient may simply have

gastritisInference

3.1 She did not have anysurgical history

Assumption

12.1 Surgical history not given Negative21.1 If his LFT’s seem to be

increasing then, . . .

Conditional

59.1 We will get an ultrasoundof the right upper quadrant

Intervention

5.1 She denies temperature Error/uncoded

Once coded, the propositions were iterativelyreviewed and recoded until there was consistencywithin the coding scheme for all of the cases. Eachproposition was given a conceptual code. Mean-ingful, repetitive concepts were evident withinthe analysis of the summaries. Once these con-cepts were developed, these data were iterativelyreviewed for consistency of coding until satura-tion occurred. For example, the text, ‘‘he’s gottremors’’ was coded as tremors and the text;‘‘occasional alcohol drinker’’ was coded as alcohol.Only errors were not coded for concept.

2.7. Data analysis

The primary analyses concerned the comparisonsbetween clinician types for differences betweenthe propositional types and concepts. Descriptivestatistics and box plots were initially applied tocharacterize the distribution of the observationsof all of the aggregated data. Descriptive statis-tics were also applied to all of the demographicdata, propositional type and concepts. The analysisof the concepts included only those concepts thatwere included by ≥50% of the physicians and ≥50%ot

ctpcpcb

3

Od

gainst those of the original cases and coded.

.6. Text segment coding

ata reduction was used to classify and direct theoding of these propositions. Through an iterativeeview of these data, a distinct pattern emerged,nd thus the data were coded as recalls, infer-nces, negatives, assumptions, conditionals, inter-entions, or errors/uncoded. This pattern cod-ng provided a more accurate and complete pic-ure of the summaries. Recalls were defined as‘reconstructions of portions of a clinical caserawn directly from the original text’’; whereasnferences were defined as ‘‘transformations per-ormed on original text-based on the subject’specific or general world knowledge’’ [30]. Infer-nces are considered high level processes which areuilt on prior knowledge and expertise. Assump-ions were defined as statements that are stated aseing true without having proof given for them [31].egatives were defined as statements of denial31]. Conditionals were defined as the develop-ent of an argument, such as ‘‘if x, then y’’ [31].

nterventions were defined as the actions taken byhe physician or nurse vis-a-vis the patient, suchs prescribing tests, medications, procedures, etc.inally, errors/uncoded were defined as either aeparture from the facts or incorrect information.able 2 shows an example of different types ofropositions and their associated codes.

f the nurses. The shared and unshared concepts ofhe two groups were compared.

ANOVA was used as the statistical test for signifi-ant differences in the propositional types betweenhe physicians and nurses. Since the number ofropositions was significantly higher for the physi-ians than for the nurses, the proportions of theropositional types for each subject were alsoompared for statistically significant differencesetween the groups.

. Results

n the whole, there was a statistically significantifference by one-way ANOVA, (p < 0.001) between


Fig. 2 Mean proportion of proposition types betweennurses and physicians.

the mean number of propositions generated percase in the physician summaries (43.52 ± 25.55) andin the nurse summaries (22.16 ± 10.62). The mediannumber of propositions in the physician summarieswas 33; that of the nursing summaries was 21.

Differences also existed between the physiciansand nurses in the types of propositions. Since thephysicians generated nearly twice as many proposi-tions as did the nurses, both the proportions andthe number of the propositions were comparedbetween the physicians and nurses. Fig. 2 displaysthe mean proportion of propositions generated bythe physicians and nurses for their 48 reviewedcases, respectively. As compared to the physicians,the nurses generated a statistically significant (byone-way ANOVA p < 0.001), larger mean proportionof recalls (74% versus 41%). The physicians, how-ever, generated a statistically significant (p < 0.001)larger mean proportion of inferences (30% versus17%), conditionals (2% versus 0), and interventions(23% versus 3%) than did the nurses in their sum-maries.

For consistency, we compared the mean num-bers of propositional types between the nurse andphysician groups. Fig. 3 shows the differences in

Fig. 4 RN and MD conceptual graph of pancreatitis case.

cases, respectively. The only inconsistency betweenthe proportion and the mean number of propo-sitions was the mean number of recalls in whichthe physicians showed a larger mean number thanthe nurses (20 versus 16). This is also explained bythe fact that the physicians had overall a largernumber of propositions. This was not statisticallysignificant. However, as compared to the nurses,the physicians included in their summaries a sta-tistically significant (by one-way ANOVA p < 0.001)larger mean number of inferences (12 versus 4),conditionals (0.7 versus 0), and interventions (9 ver-sus 0.9). The physicians had a statistically largermean number of conditionals since the nurses didnot include any conditionals.

In addition to examining the differencesbetween the types of information included inthe physicians’ and nurses’ summaries, we alsoexamined conceptual differences within betweentheir summaries. Since there was some variancein the number of concepts presented by thephysicians and by the nurses in their summariesof the cases, we show in the conceptual graphs,(Figs. 4 and 5) only the concepts that were included

Fc

the mean number of propositional types generatedby the nurses and physicians for their 48 reviewed

Fig. 3 Mean number of propositions between nurses andphysicians.

ig. 5 RN and MD conceptual graph of gastroenteritisase.


Table 3 Differences between clinician number of propositions and concepts per case type

Group Pancreatitis Gastroenteritis

No. of propositions No. of different concepts No. of propositions No. of different concepts

RN 541 98 503 72MD 1050 201 1019 196

by at least half of the nurses and at least half ofthe physicians. Combined for all of the 98 cases,the physicians and nurses had 3113 different codedpropositions (errors/uncoded propositions wereexcluded), with a total of 567 concepts, of which345 concepts were unique concepts. Table 3 showsthe breakdown of the number of propositions andconcepts for each group by case.

Fig. 4 shows the conceptual graph for the pan-creatitis case. Only 4 concepts were included in≥50% of the summaries of the nurses and physi-cians: medications, abdominal pain, vomiting anddemographics. However, an examination of the con-cepts included only by the nurses or only by thephysicians showed concepts that were important tothe practice patterns of the respective clinicians.For example, ≥50% of the nurses were concernedwith functional problems such as pain control, IV’s,and diet. The physicians, however, focused on diag-nosis, treatment, and management.

In the gastroenteritis case (Fig. 5), ≥50% of thephysicians and nurses agreed upon five concepts:diarrhea, vomiting, abdominal pain, demographics,and medication. Again, there were more unsharedconcepts than shared concepts between the two-cbcHencowpa

4

Ifbict

task analysis included the methodological frame-work of verbal protocols and propositional analy-sis, which are considered robust evaluation tools oftext comprehension [24]. Previous studies that haveused propositional analysis, a method of naturallanguage representation, to understand cliniciancomprehension of patient cases have used a cod-ing scheme which only included recalls, inferences,and uncoded as the means of understanding thecognitive processes of physicians [14,30,32,33]. Aniterative review of the clinician summaries in thisstudy revealed that only including recalls, infer-ences, and uncoded would have excluded a fairamount of information. In the nurse summaries, 74%of the propositions were recalls, 17% were infer-ences, and 9% were comprised of the other types ofpropositions, whereas in the physician summaries,only 41% of the propositions were recalls, 30% wereinferences, and the remaining 29% were comprisedof the other types of propositions. Coding theseother types as uncoded would have excluded 9%of the information in the nursing summaries and29% of the information in the physician summaries.The addition of these categorizations enhancedthe understanding of how clinicians comprehendp

bnltptrtirmfonittr

b

linician types. The shared concepts provided aasic picture of the patient, whereas the unsharedoncepts reflected the clinicians’ practice model.owever, there simply were more concepts gen-rated in the physicians’ summaries, than in theurses’ summaries. Again, the differences in con-epts included, represented the practice patternsf the respective clinicians, in which the nursesere concerned with functional issues while thehysicians were focused on diagnosis, treatment,nd management.

. Discussion

n this paper we have established a preliminaryramework for understanding how the differentackgrounds and tasks of nursing and medicinenteract to affect their comprehension of patientases, and thus provide some insight into the effec-ive design of the electronic medical record. This

atient information.Through these coding schemes, we were able to

roadly appreciate the differences between howurses and physicians comprehend a patient prob-em, and thus successfully complete this cognitiveask analysis. The nursing representation of theatient was mainly observational as shown throughheir large percentage of recalls. The physicianepresentation of the patient was causal as illus-rated by their inclusion of a large percentage ofnferences. These different patient representationseflect the differences in their respective practiceodels, whereby nurses are trained to diagnose

unctional problems and monitor changes in physi-logical status and physicians are trained to diag-ose, manage, and treat patients [11]. Although its known that the roles of each are different dueo their respective practice models, the implica-ions of these differences have not been previouslyeported.

The patient representation differences notedetween the nurses and physicians in this study can


be explained through the differences in their prac-tice models. Since the nursing practice model cen-ters partially on observing and reporting changes,it is not surprising that they included more recalledinformation than did the physicians. However, sincethe medical model involves more decision-making,it is not surprising that they included more infer-ences, conditionals, and interventions than did thenurses. In essence, each group was doing what theywere trained to do.

The conceptual differences between the nursesand physicians can also be explained through theirrespective practice models. Whereas the nurs-ing model of the patients centered on demo-graphic information, current functional status ofthe patient, current medications and IV informa-tion, the medical model of the patient focused onthe demographic information, current functionalstatus of the patient, current medications, pastmedical history information, precipitating events,differential diagnoses, and a treatment and man-agement plan. Although the practice models of thenurses and physicians share some information, nei-ther is a subset of the other. They are separate anddistinct practice models and should be considered

The paper medical record serves many purposesin addition to providing a collection of facts abouta patient’s health. It serves as a medium of commu-nication between different groups of practitioners,ensures continuity of care, and provides medico-legal coverage [34]. Yet this medium of commu-nication is fraught with problems. There is oftenconflicting and redundant information, navigationalproblems, organizational problems, illegibility, andpoor availability [35—37]. Much of the design ofthe electronic medical record has been modeledafter the paper record and hence does not take fulladvantage of the power and options available withinformation technology. The medical record shouldprovide an opportunity for improving patient care,not decrease clinical efficiency and increase thecost of clinical care. These records perform a sig-nificant role not only in gathering and storing infor-mation but also in supporting medical work [38]. Inessence these records are not mere repositories ofpatient health information, but are powerful toolsfor problem solving and decision-making. This pointshould be carefully taken into consideration partic-ularly in view of the design of the EMR: the recordneeds to support the work of the clinicians and notit

tAiarcawtimp

tmiutsutTscrnis

as such.

5. Implications for the design ofinformation systems

Taking into account the different cognitive modelsof nurses and physicians, healthcare IT must designinterfaces that reflect the needs of each discipline.Nurses will need displays that make it easier torecall relevant patient information. By contrast,physicians will need displays that assist in mak-ing inferences needed for medical decision-making.The use of a single interface for both disciplines,each of which use data so differently, may not onlycause errors but also potentially threaten patientsafety.

In the current era of advancing information tech-nology, healthcare providers will be challengedwith increasingly complex levels of information,and therefore will have a greater need to utilizetechnologies to efficiently manage such informa-tion. Their ability to easily adopt and implementthese technologies will depend upon the design ofthe tools. Information technology has the potentialto assist nurses and physicians with their respectivetasks by understanding how they each cognitivelyrepresent information and how they use this infor-mation to care for patients. The implications forthe results of this research lie in the future designof electronic medical records.

mpede the process of care or generate extra taskshat deplete cognitive resources.

Electronic medical records need to be designedo achieve their goals and the goals of the users.lthough they all need to contain certain necessary

nformation, the information must be presented inmanner that mimics the thought processes, work

outines, and practices of the users [39]. Health-are is a heterogeneous environment in which therere many different types of clinicians, not all ofhom share the same information space. Although

he concept of a ‘‘common information space’’ [40]s appealing, it is not practical since nursing andedicine each have different perceptions of theatient.

The need for a good design model for the elec-ronic medical record is apparent. Although thereay be a commonality of information between clin-

cian types, the representation of this shared andnshared information needs to support each respec-ive clinician type. It is clear from the research pre-ented here that physicians and nurses have morenshared information than shared information dueo the differences in their backgrounds and tasks.hese data confirm that a common informationpace is not practical for the individual tasks of thelinicians and may even hamper their efforts. Theseesults suggest that what the nurses and physicianseed are patient information displays that specif-cally focus on the unique tasks of their profes-ion. Although this may create some redundancy of


information within the electronic medical record, itshould not create discrepancies within the patientdata, given the fact that these records are createdwith relational databases.

Acknowledgements

This work was supported by the National Libraryof Medicine Applied Informatics Fellowship, GrantNo. 1F38 LM007188-01. The authors also wish toacknowledge the assistance of the following indi-viduals: Todd Johnson, Ph.D., Jiajie Zhang, Ph.D.,and Christopher Amos, Ph.D.

References

[1] Institute of Medicine, in: Committee on Quality of HealthCare in America Institute of Medicine (Ed.), Crossing theQuality Chasm: A New Health System for the 21st Century,National Academy Press, Washington DC, 2001.

[2] J.M. Overhage, B. Middleton, R.A. Miller, R.D. Zielstorff,W.R. Hersh, Does national regulatory mandate of providerorder entry portend greater benefit than risk for health caredelivery? The 2001 ACMI debate, J. Am. Med. Assoc. 9 (3)(2002) 199—208.

[

[

[

[

[

[15] V.L. Patel, G.J. Groen, Knowledge based solution strategiesin medical reasoning, Cogn. Sci. 10 (1986) 91—116.

[16] J.F. Arocha, V.L. Patel, Novice diagnostic reasoning inmedicine: accounting for clinical evidence, J. Learn. Sci.4 (1995) 355—384.

[17] L.D. Coughlin, V.L. Patel, Processing of critical informationby physicians and medical students, J. Med. Educ. 62 (1987)818—828.

[18] H. Schmidt, G. Norman, HP.A. Boshuizen, A cognitive per-spective on medical expertise: theory and implications,Acad. Med. 65 (10) (1990) 611—621.

[19] P. Benner, From Novice to Expert: Excellence and Power inClinical Nursing Practice, Addison-Wesley Publishing Com-pany, Menlo Park, CA, 1984.

[20] M. MacLeod, It is the little things that count: The hiddencomplexity of everyday clinical nursing practice, J. Clin.Nurs. 3 (1994) 361—368.

[21] P. Benner, C. Tanner, C. Chesla, From beginner to expert:gaining a differentiated clinical world in critical care nurs-ing, Adv. Nurs. Sci. 14 (3) (1992) 13—28.

[22] A.O. Baumann, R.B. Deber, B.E. Silverman, C.M. Mallette,Who cares? Who cures? The ongoing debate in the pro-vision of health care, J. Adv. Nurs. 28 (5) (1998) 1040—1045.

[23] E.M. McMahan, K. Hoffman, G.W. McGee, Physician-nurse relationships in clinical settings: a review and cri-tique of the literature, Med. Care Rev. 51 (1994) 83—112.

[24] V.L. Patel, J.F. Arocha, M. Diermeier, R.A. Greenes, E.H.Shortliffe, Methods of cognitive analysis to support the

[

[

[

[

[

[

[

[

[

[

[

[3] P.C. Tang, V. Patel, Major issues in user interface design forhealth professional workstations: summary and recommen-dations, Int. J. Bio-Med. Comput. 34 (1994) 139—148.

[4] K.C. Lun, New user interfaces, Int. J. Bio-Med. Comput. 39(1) (1995) 147—150.

[5] J.M. Coble, J. Karat, M.J. Orland, M.G. Kahn, Iterativeusability testing: ensuring a usable clinical workstation, in:AMIA 1997 Symposium Proceedings, 1997, pp. 744—748.

[6] A. Terazzi, A. Giordano, G. Minuco, How can usability mea-surement affect the re-engineering process of clinical soft-ware procedures? Int. J. Med. Inform. 52 (1—3) (1998)229—234.

[7] V. Patel, A.W. Kushniruk, Interface design for health careenvironments: the role of cognitive science, in: AMIA 1998Symposium Proceedings, 1998, pp. 121—125.

[8] K.J. Vicente, Coherence- and correspondence-driven workdomains: implications for system design, Behav. Inform.Technol. 9 (1990) 493—502.

[9] K.J. Vicente, Cognitive work analysis. Toward safe, produc-tive, and healthy computer-based work, Lawrence ErlbaumAssociates, Publishers, Mahwah, NJ, 1999.

10] J.D. Bransford, M.K. Johnson, Contextual prerequisites forunderstanding: some investigations of comprehension andrecall, J. Verbal Learn. Verbal Behav. 11 (1971) 717—726.

11] R.L. Simpson, Bridging the nursing-physician gap: technol-ogy’s role in interdisciplinary practice, Nurse Admin. 22 (3)(1998) 87—90.

12] N. Folz-Murphy, M. Partin, L. Williams, C.M. Harris, M.S.Lauer, Physician use of an ambulatory medical record sys-tem: matching form and function, in: AMIA 1998 SymposiumProceedings, 1998, pp. 260—264.

13] P.F. Brennan, M. Anthony, Nursing practice models: implica-tions for IS design, in: AMIA 1997 Symposium Proceedings,1997, p. 847.

14] V.L. Patel, J.F. Arocha, D.R. Kaufman, Diagnostic reason-ing and medical expertise, Psychol. Learn. Motiv. 31 (1994)187—252.

design and evaluation of biomedical systems: the case ofclinical practice guidelines, J. Biomed. Inform. 34 (2001)52—66.

25] K.A. Ericsson, H.A. Simon, Protocol Analysis: Verbal Reportsas Data, The MIT Press, Cambridge, MA, 1984.

26] C.H. Frederiksen, Representing logical and semantic struc-ture of knowledge acquired from discourse, Cogn. Psychol.7 (1975) 371—458.

27] American Medical Association, Physician characteris-tics and distribution in the United States, accessed10/23/2003, 2003/2004, available at http://www.ama-assn.org/ama/pub/article/171-195.html.

28] The American Nurses Association, Nursing Facts, accessed2005, available at http://www.ana.org/readroom/fsdemogrpt.htm#summ.

29] J.E. Newman, Pragmatics and discourse, in: G.F. Luger(Ed.), Cognitive science. The science of intelligent systems,Academic Press, San Diego, 1994, pp. 489—516.

30] V.L. Patel, G.J. Groen, C.H. Frederiksen, Differencesbetween students and physicians in memory for clincalcases, Med. Educ. 20 (1986) 3—9.

31] P.A. Angeles, The Harper Collins Dictionary: Philosophy,Harper Perennial, New York, 1992.

32] T.R. Johnson, V.L. Patel, J.P. Turley, Cognitive differencesin chart reading: a comparison of nurses and physicians, in:AMIA 2001 Symposium Proceedings, 2001.

33] C. Johnson, T.R. Johnson, V. Patel, J.P. Turley, The fea-sibility of using propositional analysis as a method forunderstanding differences between physicians and nursesreading EMRs, in: AMIA 2002 Symposium Proceedings, 2002,p. 1054.

34] F. Pinciroli, F. Crippa, C. Combi, G. Reni, D. Fava, Tanclico:tools for the analysis of inter-departmental clinical com-munications, Meth. Inf. Med. 39 (2000) 44—49.

35] D. Gordon, G. Geiger, N. Lowe, J. Jickling, What is anelectronic patient record? in: AMIA 1998 Symposium Pro-ceedings, 1998, pp. 240—244.

http://www.ama-assn.org/ama/pub/article/171-195.html


[36] M.S. Tuttle, W.G. Cole, D.D. Sherertz, S.J. Nelson, Navigat-ing to knowledge Meth. Inf. Med. 34 (1995) 214—231.

[37] H.J. Tange, The paper-based medical record—–is it reallyobsolete? Comput. Meth. Progr. Biomed. 48 (1995) 127—131.

[38] M. Berg, Accumulating and coordinating: occasions forinformation technologies in medical work, Comput. Sup-port. Cooperative Work 8 (1999) 373—401.

[39] E. Goorman, M. Berg, Modeling nursing activities: electronicpatient records and their discontents, Nurs. Inquiry 7 (2000)3—9.

[40] L. Bannon, S. Bodker, Constructing common informationspace, in: Proceedings of the European Conference onComputer-supported Cooperative Work; 1997, Lancaster,UK, 1997, pp. 81—96.

Documents

The significance of cognitive modeling in building healthcare interfaces