coding information

8/12/2019 Coding Information

1/4

Methods of Information in Medicine F. K. Schatlau er Verlagsgesellschaft mbH (1996)

J H. Hohnloser P KadlecF Puerner

Medizinische Klinik,Klinikum Innenstadt,Ludwig-Maximilians-University,Munich, Germany

IntroductionUnstructured free text is still the

most frequently used form of documentation in medicine. From an information processing point of view, nexibilityand freedom of expression are hampered by major difficulties to processfree text with computers. Therefore,numerous ever-more complex codingand classification systems have been developed [1] using even multi-axial approaches and vocabularies of more than100.000 codable terms [2, 3]. Naturallanguage processing systems. albeit rapidly improving, are still difficult to usein routine clinical settings, particularlyin languages other than English where amore complex syntax and delayed availability of standard coding systems maycomplicate matters even more. Manyclinical depart men ts do not use numericcodes for clinical documentation as aprimary means of entering informationinto computer systems. Thus, for thetime being, the extraction of codableinformation from free text throughmedical stafr rcmains the only viable1 4

Coding Clinical Information:Analysis of Clinicians UsingComputerized CodingAbstract: Data are presented of a controlled experiment with a computerized browsing and encoding tool. Eighteen practicing clinicians extractedmedical concepts from two narrative exercise cases using two approaches,traditional and computer-assisted use of ICD-9.Our results indicate that by using a computerized coding tool the completeness of coding can be improved by up to 55 , that by enforcing mandatory as opposed to optional modifier codes results in lower rates ofincomplete coding (0 and 55 , respectively), higher rates of correct coding{41 to 92 ) and no change in incorrect code, and that manual coding takes

twice as long than coding with the help of the computerized coding tool.Clinicians need 59 more time for processing the whole set of codes thanis suggested by the sum of individual codes. We conclude that the use ofa computerized coding tool can save time and result in higher quality codes.However, the real time spent on coding may be underestimated whenlooking at individual coding times, instead of the whole task of processing aclinical scenario.Keywords Free Text, Coding, ICD-9, Computerized Patient Record, Database Browsing, Encoding Tools, Medical Concepts

approach for many medical institutions,whereas data quality remains a concern[4-7]. In our hospital, coding of freetext data by physicians is a critical element of documenting patient care related processes. A computerized patientrecord system (PADS, Patient Archiving and Documentation System) wasdeveloped at the University of Munichin 1989 [8]. Amongst the systems' features is database managemen t of codeddiagnoses using a database browsingand encoding tool for codes. This articledescribes a controlled experiment withthe PADS lCD-encoding tool comparing it to traditional coding techniques.

Subjects and Methodses[ Prol col

Eighteen medical residents with 1 to4 years of clinical experience and exposure to traditional rCD-coding weretested using two exercise clinical cases(half a page of narrative free text percase) with eight ICD-codable medicalconcepts embedded in each casco The

text was not structured or formatted inany way; the terms in question were nothighlighted. Clinicians were testedtwice (with one exception). Thus, a to talof 560 medical concepts was analysed.Before each test they were given a sheetwith both exercise cases. They spent the.whole test time in a room with only thetest supervisor present encourag ing thephysicians t work as fast as possible.They had to read the text, identify codable medical concepts, use the correctsearch term and then code the termidentified. Timing was started after thetext was read once to familiarise physicians with the scenario. During one session they had to code the identifiedterms using a paperback-type codingmanual, flipping through pages (manualcoding, mc). During the other session acom puterized coding tool was used(computer-assisted coding, cc), allowing the entry of terms not included inthe lCD-9 code but linking the uscr vocabulary to the target terms using a thesaurus. The interval between mc and ccwas at least one month. The sequence- manual or computerized session first

I lvIclh Inform iVIed 1996: 35: 104 7~ ~


2/4

n s. p:5 0.001computerized (n = 88)manual (control, n = 72)

p:5 0.01 n. s.

computerized (n =144)manual (control, n = 136)

p:5 0.01

Main code,correct.

Main code,close.

Main code,incorrect.

Main code,missing.

Modifier code, correct. Modifier code,incorrect. Modifier code,missing.

Fig. 1 Overview of the coding quality ofthe main ICD diagnosiscode in both protocols (computerized vs. manual [control]), demonstrating a positive effect of the computerized approach. Analyzedwere correct codes (increase by 69.8 ), close codes (no difference;

for definition, see text), incorrect codes (reduction by 38.2%) and

Fig. 2 Overview of coding quality of the (optional) V-modifiercode in both protocols (computerized vs. manual [control]), demonstrating a positive effect of the computerized approach. Analyzedwere correct codes (increase by 46.6%), incorrect (no change) andmissing codes (reduction by 33.7%).missing codes (reduction by 61.5 ).

- was chosen at random. Due to a betterdocumentation and transparency ofcomputerized coding some parameterswere available for the cc-group, only.For example, a stopwatch integratedinto the electronic patient record wasdesigned in a way that the entry to andexit from the main browsing and codingdialogue was recorded in a database.Repeated frustrating attempts to codeone diagnosis by browsing the databaseof codes with the computerized searchtool were summarized, resulting in a total coding time for that particular diag-

and giving a realistic estimateTable 1 Datashowing theadvantage ofthecomputerizedapproach over thetraditional manualone. They alsoindicate that thetime to processthe entire narrative case document is 59 longer than expectedfrom the sum ofcoding times forindividual medicalconcepts.

TypeoJcode

MainICDCode

Y-modifier code(optional for cc-group)

f,z-modifier code(mandatory for cc-group)

Time to code

about the real time spent in coding. Inthe mc-group, however, the line between the coding-endpoint of somediagnosis and the coding-starting pointof the next diagnosis could not preciselybe drawn without interference with thecoding process. Therefore, data on coding time for individual cases are notavailable for the mc-group. For bothgroups (cc vs mc) coded terms were analysed in terms of correct, incorrect,close and missing. As close we defined those user-defined lCD-codeswhich were identical with the targetdiagnoses in at least three rCD-digits.

computerizedParameter meanS nCorrect code 8.830.26 55.21 159Close code 2.83 0.47 17.71 53

Incorrect code 3.160.26 19.79 57Missing code 1.00.52 6.25 19

Correct code 2.17 0.48 40.6 58Incorrect code 0.33 0.21 6.3 10Missing code 2.830.54 53.1 76

Correct code 2.67 0.38 92.1 264Incorrect code 0.260.28 8.2 24Missing code OO 0 0

Tune to code/diagnosis (sec) 54.58 12.4 - -Total time to code (min) 14.46 2.47 - -Total time to process text (min) 233.61 - -

Meth. Inform. Med., Vo1.35, No.2, 1996

Coding Databasern our hospital a German clinicalmodification and subset of the WHO'sICD-9 is used [9] and available as a paperback-type manual. This coding database was used for both experimentalgroups. One feature is the use of modifier codes in addition to the main codeused (usually a three-two-five digit nu

merical code). Three modifier codeswere used: The Y-modifier code modifies the main code with terms such asstatus after operat ion for. .. . t is aone-digit numerical code placed in front

manual control)meanS n significance

5.20.44 32.47 88 p:5 0.0013 01 0.32 18.78 51 n.s5.11 0.63 31.91 87 p:5 0.0012.70.21 16.85 6 p:5 0.01

1.480.32 21.9 30 p:5 0.010.330.22 5.6 8 n.s.4.27 0.41 72.5 98 p:5 0.01

1 33 0.47 41.2 111 p=O.OOI0.170.35 5.3 14 n.s.1 77 0.36 54.5 147 p=O.OOI

n.a. - - -n.a. - - -464.54 - - p:5 0.01

105


3/4

of the main code and separated from itwith a hyphen. t is optional for bothgroups (cc and mc). In certain diagnoses, f- or z-modifier codes exist, coding for functional or morphologicalfeatures of a selected syndrome thusenabling a fairly precise assessment ofdisease status. These additional codes(mandatory for the cc-group, optionalfor the mc-group) can be regarded as amulti-axial coding approach to certaindisease entities. For example, in thyroiddisease the z-modifier code describesmorphological status such as multinodular goitre; the f-modifier code describes functional status such as hyperthyroidism. Both the f- and the z-codeare separated by hyphens and appended to the last digit of the main code. Theresulting final code is:

s. p. operation for 4. Y-modifier codegoitre 241.1. ICD main codemultinodular 3. z-modifer codewith hyperthyreoidis m 3 f-modifier code

Statistical AnalysisAll analyses were done using a

spreadsheet. For statistical analyses anon-parametric paired test (Wilcoxonsigned-rank) was used because noconclusion could be drawn on a normaldistribution due to the small samplesize. Unless indicated otherwise, significance is assumed for p :0:;0.01; n. s. signifies not significant .

esultsQuality o Coding Main Code

When analysing the quality of codedterms in the cc-group as opposed to themc-group, 55.21 % (8.83 0.26) vs.32.47% (5.2 0.44, P :0:;0.001 of thecodes were encoded correctly (a 69.8%increase), 17.71% (2.83 0.47) vs.18.78% (3.01 0.32, n. s.) were close(for definition of close see Section onSubjects and Methods) and 19.79%(3.16 0.26) vs. 31.91 % (5.11 0.63,P :50.001) were coded incorrectly (a38.2% reduction). In 6.25% (1.0 0.52)vs. 16.85% (2.7 0.21, n. s.) the codewas missing; a 61.5% reduction (Fig. 1).Acceptable coding (correct plus close)was 73% vs. 51 % in the cc-group andthe mc-group, respectively. The rate of106

p 0.001 n. s. p 0.001computerized (0 = 288)manual (control, 0 = 72)

oF+Z-code, correct. F+Z-code, incorrect. F+Z-code, missing.

Fig.3 Overview of coding quality of the combined f - and z-modifier code mandatoryfor the computerized group fcc-group] in both protocols [computerized vs. manual]),demonstrating a positive effect ofthe computerized approach. Analyzed were correctcodes (increase by 50.9 ), incorrect (no change) and missing codes reduction by 54.5 ).

higher quality codes (i. e., correct, notonly close), however, was 69.8% higherin the cc-group (Table 1).Quality o Coding Modifier Codes

As outlined under Subjects andMethods, the German version of theICD-9 code [8] requires the use of up tothree additional modifier codes. In oursystem the Y-code is optional, f- andz-codes are mandatory, thus allowingcomparison of the use of mandatory asopposed to optional modifier codes inthe same test person. For the optionalY-code the number of correctly codedterms is 40.6% (2.17 0.48) for thecc-group as opposed to 21.9% (1.48 0.32) for the mc-group (p :0:;0.01 . The

p ~ O O O l

number of incorrectly coded terms is6.3% (0.33 0.21) vs. 5.6% (0.33 0.22,n. s.), and the number of missing terms53.1 % (2.83 0.54) vs. 72.5% (4.27 0.41, p :0:;0.01 , respectively (Fig. 2). Forthe mandatory f- and z-codes the combined number of correctly coded termsis 92.1 % (2.67 0.38) vs. 41.2% (1.33 0.47, p :0:;0.001 , the number of incorrectly coded terms 8.2% (0.26 0.28)vs. 5.3% (0.17 0.35, n.s.), and thenumber of missing terms 0% vs. 54.5%(1.77 0.35, P :0:;0.001 (see Fig. 3; fordata overview see Table 1).When comparing only computerprocessed modifier codes, a comparisonbetween optional (Y-) and mandatoryflz-) codes revealed an advantage for fmandatory coding with a 51.5% in-

Code, correct. Code, incorrect . Code, missing .Fig.4 Chart demonstrating the advantage of mandatory s opposed to optional modifiercoding. Both groups used the computerized approach. Correct codes are increased by51.5 , incorrect codes are unchanged, and the number of missing codes is reduced by53.1 to zero.

Meth. Inform. Med., Vol. 35, No.2, 1996


4/4

a. Computerized encoding. b. Computerized encoding. c. Manual encoding.Processing of completeexercise casesSum of all coding times. Processing of

complete exercise cas es

Fig 5 Chart demonstrating advantage in speed when processing two narrative exercisecases with a computerized coding tool as opposed to the traditional approach (b vs. c).Furthermore, a 59 increase of mean time is noted when considering the overall caseprocessing time as opposed to the simple sum of coding times for each medical concept(a vs. b).crease of correct codes, no change inthe rate of incorrect codes (6.3% vs.8.2 ) and a 53.1 % reduction in the rateof missing codes (Fig. 4).Time needed to code

With a total number of 560 medicalconcepts to be analysed, the mean number of extracted diagnoses was 15.7 2.1 and not significantly different inboth groups (cc vs. mc). As shown inTable 1 the cc-group needed a meantime of 54.6 12.4 seconds for each individual diagnosis. These data were notavailable for the mc-group. For overallprocessing time of both narrative free. text case reports, however, the cc-groupdid significantly better 23 3.6 min vs.46 4.5 min, p :50.01). When analysingthe data of the cc-group, 59% moretime was required (14.5 2.5 to 23 3.6minutes) for the overall time to completely process the two cases, when compared with the sum of all individual coding times for these two cases, suggestingextra time needed for coding-relatedprocesses not explained by the codingtime per se (see Fig. 5 and Table 1).

DiscussionWe have attempted to analyse atypical clinical coding scenario understandardized conditions, applying acontrolled experimental setting. Clinicians had to perform one of their dailytasks, extracting codable medical con-

Meth. Inform. Med., Vol. 35, No.2, 1996

cepts from narrative free text.The following observations seemnoteworthy:- Using the computerized coding tool(a generic browsing and encodingutility as part of our electronic patient record system PADS) the timerequired to code distinct medicalconcepts or diagnoses could bereduced by about 50%.Coding quality was improved substantia llyas indicated by higher ratesof correctly encoded terms, lowerrates of incorrect terms, and lowerrates of missing terms.When applying mandatory as opposed to optional (modifier) codes,comparable to multi-axial codingschemes such as SNOMED, usersresponded favorably with lower ratesof incomplete coding, higher rates ofcorrect coding, and no higher rates ofincorrect coding.Adding individual coding time intervals for individual terms resulted in asignificantly shorter time than thereal total time interval required toprocess the free-text document.This excluded the first reading of thedocument. Summarizing coding timesfor individual diagnoses significantlydistorted the real coding time burdenfor clinicians. Apparently, significantextra time for either mental conceptchange or interaction with the codinginstrument (close to 60%) was spent byclinicians even under maximum timepressure as present in the experiment.In the case of manual coding this inter-

action was intelligent searching andflipping through the pages of the codingmanual. In the case of the computerizedbrowsing/coding tool, identifying andentering the right search terms into aspecific dialogue were relevant, a process we facilitated through an extensivethesaurus.We conclude that with the help of acomputerized coding tool time can besaved and completeness of coding canbe increased. However, there is evidence to suggest that clinicians needsubstantially more time to extract codable information from free text than issuggested by the speed of coding individual diagnoses.

REFERENCES1. International Classification of Diseases Basic

tabulation list with Alphabetical Index (9thRev. Ed., 2 vols). Geneva: World HealthOrganisation 1978.2. Rothwell Dl, Hause LL. SNOMED and microcomputers in anatomic pathology. Med Inf1983; 8: 23-31.3 Unified Medical Language System Fact SheetBethesda Md: National Library of Medicine1989.4. Hohnloser IH, Konig A, Fischer MR, Emmerich B. Data quality in computerized patientrecords: Analysis of a hematology biopsy report database. Int 1 Clin Monit Comp 1994;11: 233-40.5. Klar R, Kaufmehl K. Die QualiUit der Diag

nosenstatistik nach der neuen Bundespflegesatzverordnung. In: Oberla K, Rienhoff 0Victor N, eds. Medizinische Informatik unStatistik Heidelberg: Springer Verlag 1988;23-6.6. Lloyd SS, Rissing IP. Physician and codingerrors in patient records. lAMA 1985; 10:1330-6.7. Nietzschke E, Wiegand M. Fehleranalyse beider Diagnoseverschliisselung nach ICD-9gemaB der Bundespflegesatzverordnung. ZOrthop 1992; 130: 371-7.8. Hohnloser IH, Puerner F. PADS - A PatientArchiving and Documentation System. Int 1Clin Monit Comp 1992; 9: 71-84.9. Scriba PC, Mansky T, Fassl H, Friedrich Hl.Diagnoseschliissel des Zentrums fiir InnereMedizin und des Medizinischen Zentrums.1. Auflage 1986.Address of the authors:Dr. 1. Hohnloser,Electronic Patient Record Group,Medizinische Klinik, Klinikum Innenstadt,Ludwig-Maximilians-Universitat MUnchen,Ziemssenstr. I,D-80336 MUnchen,GermanyPhone: +49 89 5160 4575Fax: +498951602341E-Mail Compuserve:[email protected]: [email protected]

1 7

coding information

Documents