Problems andbenefits of using computer for laboratory data ...
Transcript of Problems andbenefits of using computer for laboratory data ...
J. clin. Path. 22, suppl. (Coll. Path.), 3, 62-73
Problems and benefits of using a computer forlaboratory data processing
F. V. FLYNN
From the Department of Clinical Pathology, University College Hospital, London
More experience in the routine use of computersprobably exists in the field of pathology than in anyother branch of medicine. Even so, the planning andsetting up of a computer system for handling labora-tory data presents many difficulties. Definitive answersto many of the problems which arise cannot beexpected until a lot more experience has been builtup and it will probably be some years before theideal system emerges and is perfected. However,the undoubted benefits of using a computer makeit worth while struggling to overcome the initialdifficulties.The problems and benefits discussed here are
those which have been experienced at UniversityCollege Hospital, London, when using an off-linecomputer to process biochemical data, and whenplanning for an on-line system which is to embraceall the data-handling activities of the chemicalpathology service. Almost without exception, how-ever, the problems and benefits are likely to be thesame in all branches of pathology; many are dis-cussed in greater technical detail elsewhere (Flynn,Alexander, Chalmers, Grant, Jenkins, Lapage,Robertson Smith, Squire, Stirland, Whitby, White-head, and Wootton, 1968).
THE PROBLEMS
SPECIFYING THE FUNCTIONAL REQUIREMENTS OF THE
COMPUTER SYSTEM The expert designing a computersystem needs to know very precisely what themachine will be expected to do, but most patholo-gists, like other specialists, are unlikely to be goodat communicating their requirements to an outsiderwith no knowledge of their needs. During the detailedand logical examination of all the constituentprocesses of the data-handling system, problemsare certain to arise from the jargon barrier, and fromthe staff either omitting to mention the familiarsteps which are normally carried out withoutthinking or failing to recall all the exceptions tothe general rules. Fortunately construction of aflow chart facilitates communication between
62
the pathologist and the systems analyst and willbe found extremely helpful for reviewing procedureand pinpointing unnecessary steps and bottlenecks.
Difficulties are also likely to arise when thesystems analyst requests quantitative informationabout the data to be handled by the computersystem. He will want to know, for instance, theaverage load of requests each day, the maximumnumber expected in any one day, the variationin load at different times of the day, and the numberof urgent tests to be expected. He will also wantto know such things as the number and lengthof individual reports, the average length of patients'records, the greatest length to be anticipated forany particular record, and the duration for whichpatients' records need to be kept in the active file.Some of these questions are readily answered butothers involve a considerable amount of research(Fig. 1).The systems analyst will also require an estimate
of future needs, as there is no point in designinga system which is only adequate for present loads.It would seem sense to make provision for five toseven years ahead but the question is how to do this.A reasonably accurate prediction can usually bemade by plotting on a logarithmic scale the requestand test loads for several preceding years andextrapolating the lines forward (Fig. 2).
DECIDING WHAT COMPUTER FACILITIES SHOULD BEUSED There are three main possibilities: thelaboratory could have exclusive use of a smallcomputer sited within its precincts, it could sharethe facilities of a hospital computer, or it couldmake use of a computer situated remotely fromthe hospital. In each case there would be a choiceof on-line or off-line working but extensive usein the on-line mode would probably be practicalonly in the case of the laboratory computer. At thepresent time there is insufficient experience avail-able to say which approach will prove best, but inpractice the decision will be governed by what com-puter facilities are already available or planned.
on May 21, 2022 by guest. P
rotected by copyright.http://jcp.bm
j.com/
J Clin P
athol: first published as 10.1136/jcp.s2-3.1.62 on 1 January 1969. Dow
nloaded from
Problems and benefits of using a computer for laboratory data processing
30
20
00
0 L
1 7 1 3 6 1
Days Months YearTime in file
FIG. 1. An analysis of 8,652 patient records accumulatedduring a year on a file of chemical pathology results,showing the percentage of records that are recalled aftervarying intervals of time. Half the records examined hadbeen held in store for at least eight months at the timeofanalysis.
200.000
100,000 _
E 50.000z
TESTS-REQUESTS
1959 '60 62 64 66
YearFIG. 2. A logarithmic plot of the request and test load inchemical pathology at University College Hospital overthe years 1958-1968, showing how the growth in workloadfollows apredictable course.
In any event comprehensive laboratory data pro-cessing systems should not be planned in isolationas they must be compatible with medical recordsystems as a whole.
It seems possible that all laboratory records willeventually be stored in the files of central computerinstallations which will provide rapid access to allthose who need the information. Therefore, whenit is envisaged that a small laboratory computerwill be used, it should be regarded as eventuallyforming a satellite to a bigger machine.
SELECTING HARDWARE Considerable difficulty ariseswhen choosing equipment as there are many alter-natives available and technological advances are sorapid that what seems the best today may be obsoletetomorrow. At the present time, too, when there is solittle practical experience available to call upon, it isvery difficult to get sound advice. The choice shouldbe influenced by factors other than just price andtechnological excellence; the viability of the firm,its experience and reputation in the field, thesoftware offered, the adequacy of service facilities,compatibility with existing equipment and withthat of other manufacturers, and delivery datesshould also be taken into account. Requirementsfor special accommodation should be allowed forwhen comparing items of hardware; apparatusneeding special environmental conditions, withfreedom from dust and close control of temperatureand humidity, will necessitate a separate computerroom with a costly air conditioning installation andadditional communication links with the laboratory.
COLLECTING DATA FOR INPUT Several problemsarise during the collection of data for computerinput, the first being that of ensuring accuracy.Transcription should be avoided whenever possibleand maximum use made of mechanical meansof reproducing data. Handwritten entries shouldalso be cut to a minimum, for example, by usingrequest forms which have the test and specimenheadings already printed so that all the clinicianhas to do is to put a mark against those that areappropriate.A second problem is that of ensuring that the
necessary data are complete. One way of achievingthis is to transfer information in a block, for example,by reproducing patient identification data with anAddressograph plate. Another way is to design aspecial data-collection document which makes itobvious to the user when a particular item ofinformation has not been entered (Fig. 3).A third difficulty is that of ensuring that the
information provided is adequate for its purposeand here patient identification is the chief
63
on May 21, 2022 by guest. P
rotected by copyright.http://jcp.bm
j.com/
J Clin P
athol: first published as 10.1136/jcp.s2-3.1.62 on 1 January 1969. Dow
nloaded from
F. V. Flynn
DETERMINATION OP.PIGMENTS IN LIQUOR AMNIICOMPUTER DATA SHEET
PATIENT
IDENTIFICATIONAND LOCATION
DURATIONOF
PREGNANCY
DATE OPCOLLECTION OF
SPECIMN
0. D.
READINGS
Cr lf
'
Surname First Name Case No. Ward/Dept.BLANKS
cr lr
FSCr it
'I
Number of weeks (enter zero it not known)
Cr If
Cr lf
Cr If
Ar lf
Cr lf
Cr lf
r lf
BLANKS
Date
O.D. reading at 700 "
O.D. readlng at 623 wp
O.D. reading at 576 ,u
O.D. reading at 462 uyp
O.D. reading at 700 4u with dithionite
FIG. 3. An example ofa document designedfor collecting datafor computer input viapaper tape.
1
m
a
64
B1
on May 21, 2022 by guest. P
rotected by copyright.http://jcp.bm
j.com/
J Clin P
athol: first published as 10.1136/jcp.s2-3.1.62 on 1 January 1969. Dow
nloaded from
Problems and benefits of using a computer for laboratory data processing
problem. For the purpose of linking recordson a patient, unique identification data are vital;quoting 'John Smith' may be adequate for thepurpose of identifying a particular individualin a ward but it is totally inadequate for the purposeof updating that patient's laboratory record. Theminimum requirement for identifying a patient atlaboratory level is a unique personal numbertogether with the surname. Wherever possible thereshould be additional data which will aid normalcommunication and provide a means of carryingout a check in the event of discrepancies, ie, fullforenames, date of birth, sex and marital status,maiden name, location, and the requesting doctor'sinitials (Flynn et al, 1968). So essential is reliablepatient identification that it is questionable whetherit is possible to use a computer for full-scale labora-tory data processing without at least an Addresso-graph system of reproducing the information.Even then it is difficult to get clinicians to use thesystem consistently, as house officers, for instance,like to write out request forms in their own rooms.Ideally, laboratory request cards bearing patientidentification data in both conventional and machine-readable characters should be prepared at the timeof registration and held in the patient's folderuntil needed (Whitehead, Becker, and Peters,1968).
GETTING DATA INTO A FORM SUITABLE FOR MACHINEHANDLING To reduce the volume of data thathas to be translated into machine-understandablebinary language, it is frequently necessary to code theinformation first. In selecting suitable codes onehas to reconcile the conflicting preferences of manand machine. The laboratory ideal is probably asimple mnemonic system which, if not already inuse, can very readily be learnt. The computerprogrammer, on the other hand, would probablyopt for digital codes but these necessitate the useof code books by the laboratory personnel. Tworequirements that must not be overlooked whensettling the coding problem are expansion capabilityfor the future and avoidance ofany code overlap.A variety of techniques are available which
enable data to be converted into machine-readableform but at the present time laboratory computersystems invariably make some use of keyboardequipment, which punches holes in paper tape orcards and at the same time prints the correspondingcharacters on a sheet of paper or the punched card.Further problems arise from this operation. First,the whole system tends to become dependenton the clerical staff who perform the task and it isvery important therefore to make sure that thereis more than one person capable of doing this
particular job. Secondly, there is the risk of humanerror entering at this point, which necessitatessome verification procedure; visual proof readingshould pick up most of the errors but if they areto be eliminated entirely the information should berepunched by a second operator using averifier. Errors of format can be avoided by theuse of equipment which permits the layout to beprogrammed. Thirdly, keypunching may seriouslyhold up other work; the delay incurred in punchingrequest information, for instance, could mean thatit is not practical to produce technicians' worklists by computer because they would not becomeavailable early enough in the day. Fourthly, key-punching introduces a noise problem and steps willcertainly have to be taken to dampen the sound ofthe punch.Automatic data acquisition eliminates some of
the problems of keypunching operations but onlycertain portions of laboratory data can be enteredinto a computer system in this manner. Equipmentfor capturing information directly in machine-readable form has mainly been developed for usewith automatic analysis equipment and in particularwith the AutoAnalyzer, and methods ofautomatically entering peak heights (Flynn,1965, Flynn, Piper, and Roberts, 1966) and speci-men identity (Rappoport, Gennaro, and Con-standse, 1967) have now had extensive trials.Automatic data acquisition, while avoiding thelimitations of manual keypunching, brings troublesof its own. The interface equipment itself mayintroduce errors; thus Edwards (1968) has shownhow an imperfect slave slide wire used to interfacethe AutoAnalyzer to a computer caused an abnormalstep-wise frequency distribution of patient results.Another problem with automatic data acquisitionis how the data so entered are to be recognized asbeing valid. The Elliott A.L.A. off-line data-loggingequipment that we have used at University CollegeHospital for several years to capture significantdata from AutoAnalyzers copes well with mostoperating problems such as background noise onthe recordings, swamping of low peaks by precedinghigh ones, and temporary absence of peaks. How-ever, from time to time chemical irregularitiescause the equipment to punch extra or incorrectpeak values, which if not corrected when the datahave been transferred into the computer storewould cause serious errors in the results (Figs.4 and 5). Ideally, automatic acquisition of datashould include validation of all the data, but thereal difficulty is how to imitate the assessment ofpeaks that is achieved byeye.Adequatedifferentiationof valid and invalid peaks has been achieved byBlaivas (1966) by providing an on-line computer
65
on May 21, 2022 by guest. P
rotected by copyright.http://jcp.bm
j.com/
J Clin P
athol: first published as 10.1136/jcp.s2-3.1.62 on 1 January 1969. Dow
nloaded from
F. V. Flynn
Peak read-out1 , , , , , , A,
Timed read-outFIG. 4. An AutoAnalyzer recorder tracing showing an
invalid peak caused by some chemical irregularity in theanalytical system. Such a peak would trigger an incorrectreading to be taken by off-line automatic data acquisitionequipment.
Peakread-out 1 A 1 1 1 1 11 1 1 1 1 11
Iimeu reIwUuu
FIG. 5. An AutoAnalyzer recorder tracing showing a
bifidpeak caused by a very briefblockage of the samplingline by particulate matter in a specimen. Such a peak maycause an extra reading to be taken by off-line automaticdata acquisition equipment.
with a large amount of information describing thetotal shape of the curve and then testing the datato see if a number of criteria are satisfied. Thisapproach needs a considerable amount of program-ming effort and permanently locks up a sizeableportion of the computer's core store.
There is a halfway house between manual key-punching and automatic data acquisition for entryof AutoAnalyzer data. This is represented byequipment such as the d-Mac pencil follower and theNormalair Garrett trace reader. Such equipmentenables laboratory staff to record peak heightssequentially on punched paper tape merely byoperating a switch when a cross-wire or stylus iscentered over the zenith of each peak. With practicethe process can become extremely rapid; using ad-Mac pencil follower at University College Hos-pital we have found that it is possible to convert1,000 peaks into binary code within half an hour.Human error may creep in once more but thevisual vetting of peaks is worth a lot of computingtime and programming effort.
WRITING COMPUTER PROGRAMS Several difficultiesconfront one when computer programs are to bewritten, the first being that of recruiting program-ming staff. Such people are in very shortsupply, especially those with hospital experience.If one takes on a newcomer or elects to train some-one within the hospital, it takes a year or so beforehe is able to programme with any degree of efficiency.
Difficulties are also likely to arise when specifyingprecisely what is to be done by the program. De-cisions have to be taken on such things as thecriteria which must be fulfilled before a set ofcalibration standards are to be regarded as satisfac-tory. This particular question may prove fairlyeasy to answer because it is a familiar one butthe difficulty of defining what comprises an accept-able AutoAnalyzer peak can be readily imagined.The choice of programming language creates
another dilemma. The advantage of a problem-orientated high level language, in which one canwrite programs much more rapidly, may have tobe sacrificed if only a small computer is available.The reason for this is that the same program writtenin a machine-orientated language uses up less of thecore store and executes the job in a shorter time.However, as far as possible one should strive toavoid the machine-orientated languages because theprograms written in them will be machine specific.The final problem arising during programming
is the length of time taken to deliver the perfectarticle. The period allowed for the trial and correc-tion of the program is often underestimated. Anothercause of delay is staff turnover; programmers are
66
on May 21, 2022 by guest. P
rotected by copyright.http://jcp.bm
j.com/
J Clin P
athol: first published as 10.1136/jcp.s2-3.1.62 on 1 January 1969. Dow
nloaded from
Problems and benefits of using a computer for laboratory data processing
very liable to move in order to widen their experi-ence and grave difficulties are liable to be encoun-tered at the time of handing over unless very gooddocumentation has been achieved. Alterations to thespecification of a program can also cause seriousdelays and should be avoided as far as is possibleby giving adequate thought to the problems before-hand.
GETTING DATA OUT OF THE COMPUTER The firstproblem here is that of output speed. This is liable tobe far too slow with a teleprinter, operating at10 to 14 characters per second, and yet in manycases the volume of output is not great enough tojustify the use of a fast line-printer unless it isbeing shared with other users. The dilemma seemslikely to be solved by equipment of intermediateprice which has recently appeared on the market;this operates at speeds between 40 and 140 charac-ters per second.A second problem is that of timing. When an
off-line computer is used in a batch-processingmode, laboratory results may only become availablelate in the day; if errors are revealed at this pointin time it may not be possible to deal with themthe same day (Whitby, Proffitt, and McMaster, 1968).This particular problem should not arise with anon-line computer which is able to detect and reporterrors as they occur and produce reports at frequentintervals.Another problem concerns the dimensions of
stationery which are acceptable to,.standard outputdevices. These are at variance with the internationalpaper sizes recommended for use in the NationalHealth Service; thus, unless trimmed, computer-produced reports are incompatible in size with allthe other documents forming the patient's case notes.One further problem is the noise created by
output devices. This is considerable and is lesstolerable than that from input devices as it proceedsin a continuous and monotonous fashion. It isessential, therefore, to soundproof the area wheresuch equipment is placed.
COMMISSIONING THE COMPUTER SYSTEM Persuadingstaff to accept change is always a problem, especiallywhen they have been in the department for someyears. Much patience and tact are needed becauseof the prevalance of emotional attitudes towardscomputers and because the staff are being askedto operate what is inevitably a more rigid system.It is not only the laboratory staff who have to bewon over if the new system is to work smoothly;some time may have to be spent in explaining thenew requirements of the laboratory to the clinicians,the nurses, and medical records staff. An educational
programme for those directly involved in operatingthe system should begin some months beforeimplementation.Another problem is the amount of extra work to
be faced while the new system is being tested andadjusted. Faults in programs may continue to be dis-covered over a period of months as unusual com-binations of circumstances may be needed to bringthem to light. A prolonged period of parallelrunningof the new and old systems must therefore beexpected and an attempt should be made to obtainadditional clerical help for this stage.
OPERATING THE COMPUTER SYSTEM When a labora-tory has to share an off-line computer with others,adherence to a strict time schedule will often posea major problem. Transport delays, fluctuationsin work load, technical problems, and urgentspecimens all make it extremely difficult to keep to ahard and fast timetable, so some flexibility ofcomputer bookings should be negotiated if at allpossible. The alternative is very rigid organizationwithin the laboratory, but with so many uncontrol-lable factors and a constantly changing staff thisnever seems to work in practice.
Malfunction of equipment can produce errorswhich are very difficult to trace, especially whenfaults are intermittent. This problem arises particu-larly with peripheral equipment having movingparts, such as card or paper tape readers and punches,and magnetic tape handlers. The routine running ofa test program at the start of each day, to check theperformance of all the components, is probablyworthwhile but it does not guarantee freedom fromtroubles later. Investment in good environmentalcontrol, especially control of dust, is an importantprophylactic against malfunction of magnetic storagedevices, and when a built-in check on parity is offeredas an optional extra on an item of off-line hardware itshould always be taken up. To lessen the chance offaulty peripherals corrupting the data being enteredinto the computer, checks on such things as thenumber of characters and the presence of specificmarkers should be included in the program inputroutine whenever possible.Complete breakdown of equipment is inevitable
sooner or later and it is vital therefore to have afull backing-up procedure worked out in advance.The effects of breakdown could be catastrophic inreal-time computing and in this situation it is prob-ably justifiable to duplicate certain key componentsof the system. Routine maintenance is usuallyadvocated but a really satisfactory contract to dealwith breakdowns is more important; what is requiredis an unequivocal assurance that breakdowns willbe dealt with swiftly and that only highly trained
67
on May 21, 2022 by guest. P
rotected by copyright.http://jcp.bm
j.com/
J Clin P
athol: first published as 10.1136/jcp.s2-3.1.62 on 1 January 1969. Dow
nloaded from
F. V. Flynn
and experienced engineers will be sent to carry outthe repairs. The items of equipment which seem tobe particularly prone to breakdown are the air-conditioning plant, teleprinters, and magneticbacking stores. Operational breakdowns may alsoarise from power failures, and as far as possible oneshould ensure that the power supply is isolated fromthe effects of surges on heavy duty circuits andthat it is not likely to be interrupted by an electricianpulling out all the fuses when dealing with a faultsomewhere else in the building. It seems impracti-cable to provide backup for a mains power failurebecause of the expense involved.The updating of programs is another operational
problem. Laboratory procedure does not standstill and it is essential that the software can bealtered as occasion demands.This is likely to provevery difficult unless one can retain the services of theprogrammer who wrote the original version, orunless the documentation of the programs is sogood that it permits anyone else to delve into themwith ease at a later date.The final problem arising when running a com-
puter data processing system concerns the trainingof staff. Those working in highly automated,computerized laboratories get little or no practiceat calculation and this puts them at a disadvantagewhen they come to sit their qualifying examinations.One can but hope that those responsible for theexaminations will take cognisance of this.
JUSTIFYING THE COST Computerization of labora-tory procedure requires a considerable capitalinvestment. The small laboratory computer maywell cost between £20,000 and £50,000 by the timeall the peripheral equipment and the software areadded, and the question that inevitably arises iswhether or not this money is being wisely spent.The Department of Health and Social Securityare financing various experimental systems in order toassess the value of the laboratory computer, but thefindings are unlikely to be forthcoming before 1971.
THE BENEFITS
REEXAMINATION OF EXISTING PROCEDURE Havingto look again at the existing data processing systemis undoubtedly a benefit. Very often present pro-cedure is a legacy of the past and it is a salutaryexercise to have to justify it to an outsider like asystems analyst. The economic constraints of acomputer system are such that one must be certainthat each constituent operation really serves auseful function.
REDUCTION IN CLERICAL WORK FOR TECHNICAL AND
SCIENTIFIC STAFF Existing methods of data pro-
cessing are wasteful of the time of skilled personnel;in highly mechanized biochemical laboratories it hasbeen estimated that technicians spend about 30%of their time on clerical work (Rappoport, 1965).With an on-line computer system it should bepossible to eliminate this type of work almostentirely. With an off-line system it has been possibleto reduce the paper work for the non-clerical staffconsiderably; for some tests at University CollegeHospital there is no writing other than the recordingof which specimen is put into a particular Auto-Analyzer cup and no calculation of urine outputsor clearances. Elimination of this paper work is.currently saving about 31 man hours each week.
REDUCTION IN THE NUMBER OF ERRORS There areseveral reasons for expecting the frequency andmagnitude of errors to be reduced. In the case of theAutoAnalyzer with automatic data acquisition it ispossible to carry out checks on the analyticalperformance quite automatically. By programmingthe computer to establish that the calibrationstandards fall within defined limits, reagents,standards, the performance of the apparatus, andthe order of loading are all validated. The standardof work is raised as a result because the checkingprocedure is consistently applied and cannot becircumvented. Automatic validation of the individualpeaks registered by AutoAnalyzers can eliminateother sources of error. With the Elliott ALAoff-line data acquisition equipment, peaks affectedby serious carryover are automatically marked forrepeat, and with an on-line computer system othertypes of invalid peaks, such as those due to a shortsample, could be consistently detected and im-mediately scheduled for repeat analysis.
Errors due to inaccurate reading of the Auto-Analyzer tracing can also be eliminated. Peakheights can be read more accurately by automaticdata acquisition equipment than is possible by eyebut more important is the elimination of misreadings.The latter are liable to occur when the values ofpeaks are read off manually because the figures atthe top of the reading chart are far removed fromthe point of intersection being noted.
Errors of calculation can also be eliminated.At University College Hospital we have investigatedretrospectively the frequency of errors of calculationwhen using an optical density ratio method todetermine the concentration of pigments in speci-mens of liquor amnii. Computer-calculated resultswere compared with those produced manuallyunder ordinary working conditions. Only addition,subtraction, and multiplication are involved butaltogether there are 13 arithmetical operationsand a sizeable error can result from a slip at an early
68
on May 21, 2022 by guest. P
rotected by copyright.http://jcp.bm
j.com/
J Clin P
athol: first published as 10.1136/jcp.s2-3.1.62 on 1 January 1969. Dow
nloaded from
Problems and benefits of using a computer for laboratory data processing
FREQUENCY OF READING AND RECORDING
Number of Results
TABLE IERRORS WHEN READING AUTOANALYZER PEAKS MANUALLY UNDER
ROUTINE CONDmONS1Error
Magnitude Frequency (%) Type
Urea (mg/100 ml)
Bicarbonate (m-equiv/l)
Chloride (m-equiv/1)
Sodium (m-equiv/l)
Potassium (m-equiv/l)
1,001
1,039
1,010
1
2-34-67-121
24-69-11l2-34-5
10-121230-1020.3-0.41*0-1 1
1,010
1,026
39.718891*720
26.1030202
36.7690.502
41.2230-1
31 9220202
} Reading
Transcription ± reading
} Reading
Transcription ± reading
} Reading
Transcription ± reading
> Reading
Reading
Transcription ± reading
1The correct results were produced by computer without human intervention.
stage. Because of this we designed a special worksheet for performing the calculation, with spacesfor the insertion of figures and instructions as towhat to do at the different stages. Even using thisform, and insisting that in every case the calculationwas checked by a second person, we found that sixout of 81 had a sizeable error. In four cases the errorwas not of clinical significance but in two cases itwas serious in that it might have led to the obste-trician terminating the pregnancy earlier or later thanwas necessary.Computer systems also stand to eliminate those
errors of transcription which inevitably occur underordinary working conditions as a result of dis-tractions and interruptions. Errors of transcriptionare particularly liable to occur during the readingof AutoAnalyzer charts when one technician iswriting down what has been read by another;35 gets misheard and written down as 25, and so on.When we compared the results produced from 5,086AutoAnalyzer peaks by a computer system withthose produced by the laboratory staff under routineconditions (Table I), we found 53 significant errorsin the manual results, only some of which were sub-sequently detected by routine checking procedures.Twenty-seven of these errors were considered to bedue to inaccurate reading of the peak height orinaccurate correction for drift, and 26 due to a
transcription error. Interestingly, when this lattertype of error occurred the figure recorded was veryfrequently biased towards normality.
With a computer system it is possible automaticallyto flag those results which should be scrutinizedcarefully by the pathologist. This should improve theefficiency of the final check and therefore contributeanother means of reducing the error rate. In thecomputer program devised at University CollegeHospital for processing the data from the electrolyteAutoAnalyzer, a cross-check between the plasmaresults is performed by calculating the anion-cationdifference; in the printed table of results producedfor the laboratory an asterisk is placed againstresults with an ion difference greater than 20 orless than 12 m-equiv/l (Fig. 6).
FASTER PRODUCTION OF RESULTS Even with arelatively slow computer, such as the Elliott 803machine, the time taken to perform calculations canbe appreciably reduced. The actual operating timerequired on the 803 to produce and interpret theresults of a calculation of the concentration ofpigments in a specimen of liquor amnii is 13 secondscompared with 10 to 15 minutes by hand. To thisoperating time, however, should be added theone or two minutes taken to punch the data tape, the30 seconds taken to load the necessary programinto the computer, and the two minutes requiredby an off-line teleprinter to print the results fromthe output tape. With more complex calculations anddata handling procedures the benefits in this direc-tion become more impressive. At University CollegeHospital, using the Elliott 803 computer and a tele-
Determination
69
on May 21, 2022 by guest. P
rotected by copyright.http://jcp.bm
j.com/
J Clin P
athol: first published as 10.1136/jcp.s2-3.1.62 on 1 January 1969. Dow
nloaded from
F. V. Flynn
088 0 XM
c]tn h- (
cn tn en Uki L^o
CD H HHCD H H
.4.4.4r' * tr . .
.4 ' a' cr cu co \ a'
n1 a' CoF^* ^ e. X Nc
cD H a' r' X H rv uE
Co t- a' ur' C uE Nu U\
-4 _4 CD 0 U\ U\ H U\
W0 Hn C0 N CU
n --- r-not Uis0 F z
ti 0 \0 %0 \0' $
-
0 AC',c'i c'i CV c' ci Ci Ci
*. A * A A A A
40
LA.000 0oom h.lc 2 8 8.8 8 Q 0Q~ ~~A<cz <U mz o 1 oo
a: L. 0 ] o cj L. U
O H H H H H H0 H .̂0
\0a'~
rv;
0:tH
-4:0H4
O' 0 r- 0 O' Ot0 He.:jJjf1 AC8 X(
t'- H O-- In 0 a' '0 C
9 : z S & &1. -4 -40 0s\
* O* .A' Lom%^-H
H IS LnI\HM CU a'IsU
In 0 i0 C0 i In C Im***m *m VC
A
CU
A A
CU
ACU CU
0\0
C',
0
H11
'0 U\i HS ir 0'a'I
NU tr' '0 In * in r' C
in * '0 . 0 '0 .4 0
H ' a' 0 Q 0 0 HC' In In * t .4 C' *
H
*'0 *- *-. *
ujJDa0eQ DCWoH0 ° ° s ot
InCC-a'HC a
'0 0 . ' '0 InO H C
In H'0'0N.4InU FN
inH H (fl In In C's I
F 0 co a) ao 0 X~~j t cj
Ns m CU1 ;N C19J1 CU (U
*o .* \L am n
-c o U\ L^ U\ F C) U\J m CD CDOD CD C
(,U
C) h \0cr CL CLIt k 'X,c
- * H Inm NmN m .4 m, In.t
LA 14 LA X XE Ll X LL
Hat* H
4
NcuH
gcj0* InO -Hc
Nm a'
U ,4 m- mbb A4 N .4- fmr 1- -I:t Nu .-
X X X Xli . XLA X
2s
z
a' C u'*' -I01flj
* *
0Da
r a\ Ur 0 In
0O H sL^ In cm9 L^ t H H H
\O rn _ In H H
3> -3 t -) .OCC) mn mn m n 4c _iJ{
0
CL *
0*- * * *
* * *
* ~~~~U,
_e -'s
C-
*4 HfH H ' '
C°h z a,cr 8 '°
CC) 1g W c4
3 *
1* 3; * *
*f)
LIX X X Le XX. X
o H0 0 cn Hl No
{ O -| C?J C1 0 0*j * * . 4
cUi R°H 'i\> o
en c)0 U--k '10\n -4 tom -4 0m 0cim
*(1- -c WE 9 z 1
Xi. :X
*
*~~~
0 CL U (n CL U in
70
8
tn JZ. E
o_i
0-.C
-Jz L!J
m
_i
oi W40
_LiI9---3
UJX
XJ1-
_H .
0 A<
N
CO
- t xt
s
o *. ; 2
H .
o ~4
t_ o
"c O
O t oO@,*
X
.
U,
O.0
L.JU,
a'Hs-I
m %D .4 110 _* em _itR 1-4 -4 R 14 14'1.4 -4 -4
tt I L V. $- X
on May 21, 2022 by guest. P
rotected by copyright.http://jcp.bm
j.com/
J Clin P
athol: first published as 10.1136/jcp.s2-3.1.62 on 1 January 1969. Dow
nloaded from
Problems and benefits of using a computer for laboratory data processing
E.XAMI.NATION OF LIQUOR AMNl
G---N JEAN AJ6MM OH OPD
DURATION OF PREGNANCY m 32 WEEKS
DATE OF COLLECTION OF SPECIEN m 11. 7.67
01). READING AT 70O0J - 0.019
O.D. READING AT 623MU * 0,030Q.D. READING AT 576# m 0.0450.1). READING AT i62M m 0.312O.D. READING AT 70U m o.c23
* WITH DITHIOWITE
METHAEWLBUMIN - o.s05 M |f1 ML
OXYHAEMOGLOBIN - 1.37 M;i100l
BILIRUBIN 0.32 MG11O
HAEMOLYTIC DISEASE WILL BE SEVEREAND DELIVERY SHOULD BE EFFECTED BY THE 36TH WEEK
FIG. 7. Output produced by a computer program whichis used to calculate pigment concentrations in specimensoJ liquor amnii. The interpretation of the result takes theduration of pregnancy into account and suggests theclinical action which should follow.
printer operating at 10 characters/second, the totaltime taken to produce 104 sets of urea and electrolyteresults from a data tape bearing peak height inform-ation on 750 AutoAnalyzer peaks can be as little as25 minutes, including eight minutes' computertime. The same output produced entirely by handwould take many hours to appear because it includesa calculation of standard deviations; without thelatter it would require just over one-and-a-half manhours if two technicians worked on the job.
Faster results are not necessarily synonymous withspeedier reporting. The benefits of faster manipu-lation of data tend to be nullified by the use of acomputer in a batch-processing mode, especiallywhen it is devoid of a fast output device such as aline printer. The time lost while the data are heldin the queue awaiting computer processing maynot be offset by the subsequent saving of time.
IMPROVED INTERPRETATION OF RESULTS It is possibleto include routines in a computer program which
automatically add an indication of the normalityor otherwise of a given result, taking into accountsuch factors as sex and age (Fig. 6). Doing thiswithout a computer is quite impracticable becauseit is impossible either to carry the multitude ofnormal ranges in one's head or to find time to lookthem up in an appropriate table. In the program wehave developed for the calculation of pigmentconcentrations in specimens of liquor amnii, wehave gone one step further in getting the computerto suggest the clinical action which should followthe finding of a particular result (Fig. 7). Thissaves having to look up or memorize the correctinterpretation of a test which is not very frequentlycarried out.
IMPROVED PRESENTATION OF RESULTS With a com-puter system, typewritten reports of consistentformat can be automatically produced for theclinicians and for the laboratory. Computer systemsmaking use of the backing stores for retention ofprevious results are compatible with the practiceof cumulative reporting, which lessens the bulkof reports filed in the case notes and provides theclinician with a well organized summary of all thepatient's laboratory results in date sequence (Flynnand Vernon, 1965; Whitby and Owen, 1965).
FASTER COMMUNICATION OF RESULTS At presenta large proportion of the time taken to return aresult to the clinician is taken up in transmitting thereport from the laboratory to the periphery. Fastertransmission could be achieved with a computersystem by making use of a teleprinter link, and thesaving in time could be very significant, especiallywhere laboratory services are centralized. Thelaboratory stands to benefit from such a procedurein so far that it should lessen the enquiries whichdisrupt its clerical services.
IMPROVED QUALITY CONTROL With a computeravailable it becomes practical to employ a statisticalapproach to quality control, using the mean valueof near-normal patient results to assess the per-formance of a method (Whitehead, 1965; Whitby,Mitchell, and Moss, 1967). At University CollegeHospital we have found that this technique providesan earlier warning of a method getting out of controlthan any other procedure and we have now built itinto all the computer programs used for dealingwith the processing of AutoAnalyzer results(Flynn, 1966). An important advantage of havingsuch information produced automatically is thatit is available when the results are being scrutinizedand not retrospectively. An example of a quality
71
on May 21, 2022 by guest. P
rotected by copyright.http://jcp.bm
j.com/
J Clin P
athol: first published as 10.1136/jcp.s2-3.1.62 on 1 January 1969. Dow
nloaded from
72 F. V. Flynn
.45
gaily Mean 35 -
25 FIG. 8. A quality control chart1s ______________________________________________ drawn up from data returned by the
computer when processing informationCumulative 10 ~~~~~~~~~~~froman AutoAnalyzer measuring urea.
sum of s Daily means ofallplasma results betweendifferences 0 S and SO mg/100 ml are plotted, alsoof mean the cumulative sum of the difference
(mg.110OmL) 10 between the daily means and the figuretaken as the average value. When the
... value for the daily mean standardizedDaiy Mean
2deviation exceeds 3, action to get the
stadardised 0 method under control is required anddeviation when it is greater than 2 but less than
3 the staff should be alerted to look
28 4 11 18 25 2 9 16 23 for trouble.AUG SEP OCT
control chart drawn up from data returned by thecomputer is shown in Figure 8.
MORE COMPACT STORAGE OF RESULTS The storageof laboratory records in a very compact form couldbe another important advantage of computersystems. At University College Hospital we havecalculated that when all the chemical pathologyrecords are stored on a magnetic tape file it should bepossible to hold a year's results on one reel of tape12 in. in diameter and 1 in. thick. The same numberol results held on cumulative record cards wouldrequire 12 cubic feet of filing space.
EASIER ANALYSIS OF RECORDS Because of the highspeed of search and retrieval operations and theabsence of misfiling it is possible to analyse recordsheld on a computer file much more easily andreliably than is possible with manual procedures.The benefits in the realm of laboratory managementcould be considerable; for example, regular examin-ation of the workload for changes in its patternshould lead to more rational planning of laboratorydevelopment and allocation of resources. The easeof specific and class retrieval of data and the possi-bility of performing linkage studies with otherparts of the medical record also open up excitingpossibilities for research.
SUMMARY
The problems and benefits of using a computer forlaboratory data processing are discussed in the light
of experience obtained at University CollegeHospital, London, in the last five years.The problems covered include the specification of
the functional requirements of the computer system,the choice of computer facilities and of hardware,the difficulties associated with getting data into andout of the computer, the writing of computerprograms, the commissioning and operation of thesystem, and the justification of the costs.The benefits discussed include the streamlining of
existing procedure, the reduction in clerical work andin errors, the faster production and communicationof results and their improved interpretation andpresentation, better quality control, more compactstorage of results, and the better accessibility ofrecords for analysis.
I wish to acknowledge the contributions of Mrs H.Hawkins, Mr M. J. Healy, Mr K. A. Piper, and Miss G.Richards to some of the work reported here. I am alsoindebted to the Department of Health and SocialSecurity for research grants that made this workpossible and to Mr V. K. Asta for the diagrams.
REFERENCES
Blaivas, M. A. (1966). In Automation in Analytical Chemistry, editedby L. T. Skeggs, p. 452. Mediad, New York.
Edwards, R. G. (1968). Personal Communication.Flynn, F. V. (1965). In Progress in Medical Computing (Symposium,
organized by Elliott Medical Automation, London 1965). p.46.Blackwell, Oxford.
, and Vernon, J. (1965). J. clin. Path., 18, 678.(1966). Proc. roy. Soc. Med., 59, 779.Piper, K. A., and Roberts, P. K. (1966). J. clin. Path., 19, 633.Alexander, M. K., Chalmers, D. G., Grant, G. H., Jenkins,W. J., Lapage, S. P., Robertson Smith, D., Squire, J. R.,Stirland, R. M., Whitby, L. G., Whitehead, T.P., and Wootton,I. D. P. (1968). Ibid., 21, 231.
on May 21, 2022 by guest. P
rotected by copyright.http://jcp.bm
j.com/
J Clin P
athol: first published as 10.1136/jcp.s2-3.1.62 on 1 January 1969. Dow
nloaded from
Problems and benefits of using a computer for laboratory data processing 73
Rappoport, A. E. (1965). In Symposium on Computer-assisted Path-ology, College ofAmerican Pathologists, 1964, p. 4. Chicago.
-, Gennaro, W. D., and Constandse, W. J. (1967). Mod. Hosp.,108, 107.
Whitby, L. G., and Owen, J. A. (1965). J. clin. Path., 18, 668.-, Mitchell, F. L., and Moss, D. W. (1967). Advanc. Clin. Chem.
10, 65.
, Proffitt, J., and McMaster, R. S. (1968). Scot. med. J., 13, 181.Whitehead, T. P. (1965). In Progress in Medical Computing (Sym-
posium organized by Elliott Medical Automation, London,1965), p. 52. Blackwell, Oxford.
-, Becker, J. F., and Peters, M. (1968). In Computers in the ServiceofMedicine, edited by G. McLachlan and R. A. Shegog, Vol. 1,p. 113. Oxford University Press.
on May 21, 2022 by guest. P
rotected by copyright.http://jcp.bm
j.com/
J Clin P
athol: first published as 10.1136/jcp.s2-3.1.62 on 1 January 1969. Dow
nloaded from