L icterus

Document Sample
L  icterus Powered By Docstoc
					                                                   BIDIRECTIONAL
                                              INTERFACE    CRTs to LIS
                                                                                                                         References
                                                                                                                         1. Critical    care testing   faces cost pressures.       Clin Chem News




          L.#{176}4;;:_
                                                                                                                         1990;16:1,6.
                                                                                                                         2. Weaver DK, Miller D, Leventhal            ED, Tropeano      V. Evaluation
                                                                                                                         of a computer-directed        pneumatic-tube       system     for pneumatic
                                                                                                                         transport    of blood specimens.    Am J Clin Pathol 1978;70:400-5.

                   _
                                                   UT           OUTPUT           B
            EKTACHEM           A                      .4UW                                                               3. Doumas BT, Hause LL, Simuncak                DM, Breitenfeld     D. Differ-
                                   RK7
                                                                                                                         ences between values for plasma and serum in tests performed                 in
                                              .    .........f
                                     *.....        ......Iq
                                                                                                                         the Ektachem       700 XR analyzer,     and evaluation     of “Plasma Sepa-
                                                  AREA                               EKTACHEM            -S-             rater Tubes (PST).” Clin Chem 1989;35:151-3.
   LOW VACUUM                                     jY                                                                     4. Tilzer LL, Jones RW. Use of bar code labels on collection tubes
                                                                                                                         for specimen     management      in the clinical laboratory.     Arch Pathol
   TUBE STATION


                _____
                                                                ______________________________________________________
                                                                                                                         Lab Med 1988;112:1200-2.
                DILUTION
                STATION                      ‘I,
                                    BAR CODING                    L1
                                                                  LJ
                                                                                                                         Use of Serum     Blank Information      to QuantIfy
                                                                 CENTRIFUGES                                             Chromogenic      Interferents    and Correct Sensftlve
                           D
                                                                                                                         Analyses,    K. Burtner,    M. Huber,   and S. Frye (Advanced
                                                                                                                         Development,    Research      and Development,    Baxter
          r-”        VENIPUNCTUREDROPOFF                                                  +    (EXIT)
                                                                                                                         Healthcare   Corp., Dade Div., 9500 Jerommo Rd., Irvine,
Fig. 1. Efficient layout of the laboratory         work area, based                                 on the               CA 92718)
‘kitchen triangle” design
CAT, cathode ray tube; LIS, laboratory information    system                                                                 Lipemia,         icterus,        or hemolysis               in serum          samples        can
cally; controls       are processed      once a day at 0500 hours. This                                                  cause bias in clinical               assays          (1). The bias may be caused                   by
procedure       has been       acceptable,      with   a strictly       enforced                                         uncorrected          spectral        absorbance,            by chemical           involvement
“delta” checking         program      for patients’    results      in the labo-                                         in the assay            reactions,        or by both. For many                      assays       the
ratory information          system.     Downloads     from the laboratory                                                interference          is purely        photometric             and is removed             by spec-
                                                                                                                         tral subtraction             of a serum blank.                 For others, the interfer-
information       system     are broadcast      to both analyzers         so that
either instrument         can handle any specimen              arriving     in the                                       ence requires            a more complex correction                       scheme.
laboratory       at any time. Every 3-4 h, the analyzer’s memory                                                             The first step in adjusting                         for a nonspectral              bias is to
is deleted     of any unprocessed accession numbers, and a new                                                           quantify       the concentrations                     of lipid, bilirubin,          and hemo-
download        is initiated.        Because     the uploaded results will                                               globin in the serum                     sample.           By measuring              the serum
remove      processed       specimens from the work load, only more                                                      absorbance          at several selected wavelengths                            one can simul-
recent or unprocessed              specimens will be downloaded                again.                                    taneously         quantify          all three           potentially        interfering          sub-
This total duplication              and specimen handling               system al-                                       stances. We have investigated                            several different techniques
lows one instrument               to be arbitrarily       assigned      to stats or                                      for quantifying              lipemia,       icterus,         and hemolysis              in serum
high-priority        samples by simple verbal designation                        from                                    samples.
the operators, while the other instrument                     processes routine                                              A set of 141 samples supplemented                             with various           amounts
specimens.        In addition,        both instruments         function      to back                                     of intralipid,         bilirubin,        and hemolysate                was analyzed             with
each other up in case of an analyzer                    problem and during                                               a Paramax5            analyzer        (Baxter          Healthcare        Corp., Dade Div.,
routine      maintenance           or calibration      periods.       The lack of                                        Irvine,      CA) and used as a calibration                           set. A second            set of
additional       instrumentation           prevents problems of internal                                                 646 supplemented                  samples, analyzed                 with     a second         Para-
methodological          biases. The manufacturer’s                stat capability                                        max, was used as a test set. The Paramax                                        measures         the
of the Ektachem             is never utilized because the time to be                                                     absorbance          of each serum                 blank      bichromatically             at seven
gained     by interaction          with the instrument          is negligible.       In                                  different      wavelength            pairs: 340/630,450/575,575/630,340/
fact, other than the entry of dilution                 factors for specimens                                             405,      525/630,         405/630,        and 550/630                 nm. We used the
requiring       them, the instrument              operator rarely interacts                                              Paramax          triglyceride            assay to quantify                   lipids and the
with the Ektachem             instrument       control panel. The system is                                              Parainax        total bilirubin            assay to evaluate                  icterus.      Hemo-
totally     controlled        by the laboratory          information          system                                     globin was quantified                  by using the DMA cyanmethemoglo-
interface      computer         terminal      for each analyzer            and the                                       bin assay and a Beckman                       DU#{174}-7    spectrophotometer                (Beck-
Kodak    analyzer    output      printer. We use the latter                                   to review                  man Instruments,                  Brea, CA).
results  for instrument        analysis   errors (dilutions,                                  linearity,                     Multiple      linear regression                 (MLR) is an extension                 of linear
sampling     problems,      etc.) and use the laboratory                                       informa-                  least-squares            regression          (2).       Because       the Paramax              mea-
tion system terminal         to control downloads,       delta                                checking,                  sures      the serum             blank      at seven             wavelength           pairs,     the
and verification    of results.                                                                                          prediction        equation has the form
    To evaluate   the turnaround       time of the new system, a
quality-assurance       study, “Emergency         department    turn-                                                                     C=      k0   + k1A1+             k2A2      +    ...     +    k7A7

around     time” (College of American          Pathologists,  North-
field, IL 60093-2750)       was carried out. The study demon-                                                            where        C   =   predicted      concentration       of interferent:       lipid,
strated for all shifts a median      turnaround     time of 26.5 mm                                                      bilirubin,      or hemoglobin;         A = serum blank            absorbance       at
for the analyte   potassium     (n = 76 patients evaluated).      The                                                    wavelength          pair n; and k,., = calibration           coefficients.
study also included         a physician     survey and found an                                                              One set of prediction             coefficients    must be developed           for
expected    mean turnaround        time of 34 mm for potassium       (n                                                  each interferent.            After the coefficients       are developed from
 =   5 physicians).      By this criterion        and the lack of any                                                    the calibration            data, estimates         of the concentrations           of
requests    for additional     stat laboratories,       the system was                                                   lipid,   bilirubin,         and hemoglobin          can be made           from the
determined      to be a success    in achieving      timely service.                                                     seven    absorbance           measurements         of the serum blank.          The

1584            CLINICAL   CHEMISTRY,              Vol. 36, No. 9, 1990
                                                                                                    tion      sample      analysis     indicated      that     two    factors    were   nec-
       Table 1. Comparison     of Predicted   vs Reference
                                                                                                    essary        to predict   lipemia,   three factors for icterus,              and four
      Interferent Concentrations     for Several PredIction
                                                                                                    factors        for hemolysis.     The results   of PLS prediction               on the
                           Techniques
                                                                                                    test data set are shown   in Table 1.
                         Upemla                      Icterus                    Hemolysls               The neural network  (NN), an entirely                        different   approach
 Method             r2              Sa         r2               $              r2               s   from the others previously  discussed, simulates   groups of
MLR               0.956           93.0       0.836             3.88       0.864            76.4     neurons  by using computers   (6). Each processing element
                                                                                                    (neuron)       receives multiple         inputs and produces a single
PCR3b             0.953           96.0       0.928             2.56       0.936            52.4
                                                                                                    output.    The neurons in the network are arranged                      in layers,
PCR.4C            0.935           98.0       0.923             2.66       0.937            52.3
                                                                                                    with the neurons at the boundaries                   performing        input and
PLS               0.954           95.3       0.929             2.55       0.937            52.0
                                                                                                    output and the central neurons performing                      pattern     match-
NN                0.951       98.4    0.915                    2.79       0.940            50.7
                                                                                                    ing. After constructing            the network,        we first trained        it to
  a   Standard error of regression.
                                                                                                    predict     the concentration           of each interferent,           using the
  b   Principal components regression performed                with three factors for each
                                                                                                    same data used to calculate               the calibration       coefficients      for
interferent.
     Pnncipal components regression performed with four factors for each                            the other methods, then used it to predict the concentra-
interferent.                                                                                        tions of the test data (Table 1).
   MLR, multiple linear regression; PLS, partial least-squares (regression); NN,                       Comparison          of the correlation        coefficients     and standard
neural network.                                                                                     errors of the predictions            of the various methods (Table 1)
                                                                                                    reveals      that PLS performed               best for two of the three
correlation       results for the set of 646 test samples              are shown                    interferents       (icterus and hemolysis),           with MLR doing best
in Table 1.                                                                                         for lipemia.         MLR     did poorly       for icterus      and hemolysis
   Principal       component      regression    (PCR) attempts          to reduce                   because most of the wavelength                 pairs for these interferents
errors    in prediction        (compared     with MLR) by dividing              the                 do not have high signallnoise                 ratios. Inclusion         of all the
absorbance         data into signal and noise components.                  Predic-                  wavelengths          adds noise to the prediction             set. Conversely,
tions based on signal alone should be more accurate                           than                  the lipemia absorbance            does occur at all seven wavelengths
MLR       predictions,       made     with signal         contaminated        with                  and thus benefits prediction               by MLR.
noise.                                                                                                 With PCR, the difficulty            in knowing how many factors                 to
    PCR calibration           is a two-step      process:      eigen analysis,                      use in the analysis is compounded                 by the fact that different
followed     by multiple   linear regression.        An eigenvector                                 numbers        of factors may be optimal              for different      interfer-
analysis is made of the absorbance          covariance      matrix.    A                            ents. For example, for the test data here, three factors made
subset of the eigenvectors,      those with the largest eigenval-                                   better predictions         for lipemia       and icterus,     but four factors
ues, is used in the analysis.      The number    of vectors    selected                             did slightly better for hemolysis.              Possibly, the fourth factor
should be the minimum          required   to reproduce      the absor-                              contains signal information              for hemolysis        but is primarily
bance    data to within   experimental      error (3). Multiplying                                  noise for lipemia         and icterus.
the data matrix       by the matrix      of selected     eigenvectors                                  PLS outperforms           PCR for icterus        and hemolysis         because
produces an array of scores. The interference               concentra-                              it not only ignores        wavelengths          with    low signal-to-noise
tions are regressed against the scores to produce a predic-                                         ratios,    but also explicitly      attempts       to correlate      the absor-
tion equation of the form                                                                           bances     with concentrations          during      the filtering      phase.   It
                                                                                                    also gains      from the ability         to use different          numbers     of
                                                                                                    prediction     factors for the three         interferents,       including    all
                         C=k0+k1S1+...+kS,.,
                                                                                                    factors that represent        signal and excluding           noise factors    on
                                                                                                    an interferent-by-interferent       basis.
where C = predicted              interferent     concentration,         S, = score
                                                                                                       Interestingly,      even though it started      with no explicit
on eigenvector           n, and k = calibration             coefficients.
                                                                                                    information       about the relationships      between the serum
   We determined              that between          three     and four factors
                                                                                                    absorbances and the lipid, bilirubin,         and hemoglobin      con-
(eigenvectors)          were needed to model             the calibration      data
                                                                                                    centrations,      the neural network       was able to predict the
accurately,        after testing        the residual       error (the error in
                                                                                                    concentrations       about as well as the other methodologies.
reconstructing           the absorbance         data) and the embedded
                                                                                                       Using the PLS coefficients,       we determined     the distribu-
error. We developed a set of regression                   coefficients for each
                                                                                                    tion of lipemic,       icteric, and hemolytic     samples  from an
interferent       and performed          the analysis      by using both three
                                                                                                    actual    patient population:     6737 samples tested at a mid-
and four factors to predict                  the interferents          of the test
                                                                                                    western   hospital     over three months.        Interestingly,     the
samples (Table 1).
                                                                                                    majority  (86%) of the specimens had little interferent.           Most
   Partial     least-squares        (PLS) regression is similar to PCR
                                                                                                    of the remaining     samples  had a significant     concentration     of
in that it attempts            to remove noise from the data before
                                                                                                    only one interferent      (6% were lipemic,     5% icteric,     and 2%
prediction       (4). However,         PCR attempts to separate absor-
                                                                                                    hemolytic).
bance signal          from noise without           regard      to the predicted
                                                                                                       We have shown that PLS is an accurate                    method for esti-
concentrations,           whereas        PLS attempts            to extract         that   part
                                                                                                    mating      interferent      concentrations      from serum      absorbance
of the absorbance        that best models the interference            concen-
                                                                                                    spectra.    Knowing       the interferent     concentrations,     we can do-
trations    by finding factors that simultaneously             make large
                                                                                                    termine     which test results       are biased, and for some assays we
contributions      to the absorbance       variance     and correlate     well
                                                                                                    can calculate       a correction    based on these concentrations.
with the interferent         concentration       (4). Several   factors    can
be found      for each       interferent;     the number        selected      is
                                                                                                    References
determined      by how well the calibration             data are modeled.
                                                                                                    1. Glick MR, Ryder KW. Interferographs.     Indianapolis:   Science
All the factors      for a single interferent        can be combined      into                      Enterprises,    1987.
a single    vector for prediction (5).The number                         of   factors does          2. Beebe     KR, Kowalski BR. An introduction      to multivariate
not have      to be the same for each interferent.                            The    calibra-       calibration        and analysis.    Anal   Chem     1987;59:1007A-17A.

                                                                                                                          CLINICAL     CHEMISTRY,            Vol. 36, No. 9, 1990       1585
3. Malmowski       ER, Howery     DG. Factor analysis     in chemistry.                         start to finish, takes -5 mm. Results should be read within
New York: Wiley-Interscience,       1980.                                                       30 mm after application               of sample. The test is calibrated
4. Geladi P, Kowalski        BR. Partial   least-squares   regression:        a                 such that hCG at 50 mt. unitslL (WHO Third International
tutorial. Anal Chim Acta 1986;185:1-17.                                                         Standard)       is visibly detectable,        whereas concentrations                <5
5. Haaland     MD, Thomas      BE. Partial least-squares     methods        for
                                                                                                int. units/L are undetectable.               A gradual       increase in assay
spectral analyses.    Anal Chem 1988;60:1193-202.
6. Klimasauskas      CC. An introduction    to neural computing.       Pitts-                   sensitivity       is observed       with time. Specimens                containing
burgh: NeuralWare,       1988.                                                                  hCG <50 i.              units/L     may test positive            after      30 miii.
                                                                                                Positive     results are consistently            observed with specimens
                                                                                                containing       hCG concentrations            as high as 106 mt. units/L.
                                                                                                    Addition      of LH or FSH at 1000 mt. units/L, TSH at 1000
                                                                                                milli-int.     urnts/L, or free beta-hCG            subunit at 100 jtg/L to
One-Step Chromatographic          Immunoassay    for                                            hCG-negative           (0 mt. unitfL)         and hCG-positive               (50 mt.
Qualitative DetermInatIon     of Choriogonadotropin         In                                  units/L)      urine     specimens        showed no cross-reactivity                  or
Urine, Gene Osikowicz,    Michael    Beggs, Paul Brookhart,                                     interference        in the assay. A wide range of potentially
Diane     Caplan,        Shanfun       Ching,     Paul     Eck,    Julian      Gordon,          interfering       substances were tested and did not interfere                     (2).
Russell     Richerson,        Susan       Sampedro,     Don Stimpson,  and                          We compared            the performance          of the TestPack              Plus
Frank      Walsworth         (Abbott      Laboratories,    Abbott Park, IL                      assay with that of two sensitive                   qualitative         urine     hCG
60064)                                                                                          assays:      Abbott       TestPack”        hCG Urine            and Hybritech
                                                                                                ICON#{174} hCG (Hybritech,
                                                                                                            II                              La Jolla, CA). Among                   796
                                                                                                samples       (501 positives,        295 negatives)          tested, including
    Human      choriogonadotropin               (hCG), a 36.7-kfla             glycopro-
                                                                                                urine specimens           from 96 post-menopausal                women, 99.7%
tein hormone         secreted by the human                   placenta      throughout
                                                                                                concordance          (795/796)      was obtained          between         TestPack
gestation,     is an established           indicator        of pregnancy.         hCG is
                                                                                                Plus and TestPack.            One urine specimen that tested positive
composed       of two dissimilar           subunits,        alpha and beta, that
                                                                                                by the TestPack             hCG Urine        assay and negative                by the
are closely homologous                 with the subunits                 of the other
                                                                                                TestPack       Plus hCG Urine assay had an hCG concentration
glycoprotein        hormones:         lutropin        (LH), follitropin            (FSH),
and thyrotropin          (TSH).       Screening          blood or urine with a                  of 12 mt. unitsfL           [determined       by use of a whole-molecule
                                                                                                (intact hCG) quantitative              assay]. Among 430 samples (150
sensitive    and specific qualitative                hCG test provides rapid,
                                                                                                positives, 280 negatives)            tested by both TestPack               Plus and
accurate     information         that allows early detection                     of preg-
                                                                                                ICON II, there was 100% agreement                     of results.
nancy (1).Most current commercially                         available      qualitative
                                                                                                    In summary,         this rapid qualitative         pregnancy          test com-
assays of hCG in urine are two-site immunometric                                   assays
with solid-phase         “capture”        antibodies          and enzyme-labeled                bines excellent          performance         with a one-step procedure,
conjugate      antibodies.       These tests are generally                    sensitive,
                                                                                                eliminating        several      of the manual           steps characteristic           of
                                                                                                current     qualitative       methodologies.
specific,  and rapid, but are complex,                    requiring        as many as
four timed manual            steps.
    We have developed             a rapid, one-step immunochromato-                             References
graphic assay for the qualitative                   detection of hCG in urine                   1. Braunstein    GD. HCG testing, a clinical guide for the testing of
                                                                                                human chorionic gonadotropin. Monograph       97-8982. Abbott Park,
with use of a direct colloidal                label. This Abbott TestPack
                                                                                                IL: Abbott Diagnostics Education Services,    1987.
PlusTM hCG Urine                assay      (Abbott         Laboratories,          Abbott        2 Abbott TestPack Plus hCG Urine package insert. Abbott Lab-
Park, IL) utilizes both monoclonal                     and polyclonal           antibod-        oratories, Abbott Park, IL, 1989.
ies to selectively       identify      hCG in urine with high sensitiv-
ity and specificity.
    A single molded-plastic              disposable          container       (5.6 x 5.6
cm) houses        a diagonally        oriented       strip of laminated             nitro-
cellulose     attached       to a glass fiber pad containing                             the    Considerations            In the Design and Implementation         of
colloidal   selenium        conjugate.         In the test procedure,                three      User Interfaces           between     Laboratory  Instrumentation
drops of urine are transferred                with a disposable            pipette into         and Laboratory            Information    Systems,   David   Chou,1’2
the sample well. The sample                 well directs the urine into the                     Frederick       Van Lente,2   and William     Castellani2     (‘Lab.
glass fiber pad containing              colloidal selenium             particles       (160     Computer         Systems and 2Dept. of Biochem.,          The
nm diameter)           conjugated        to monoclonal              anti-alpha-hCG              Cleveland        Clinic Foundation,    Cleveland,      OH 44195)
antibody.     The rehydrated           conjugate then migrates through
the nitrocellulose          membrane            until it reaches an end-of-                         Computerized           instrument            interfaces,      in which the user,
assay window          (in about 5 mm). As the rehydrated                            conju-      the laboratory           information            system      (US),     and the instru-
gate migrates        along the membrane                 strip, it passes through                ment act as an integrated                      unit, can greatly          improve    pro-
a capture      region consisting             of two bars in the form of a                       ductivity,       user fatigue;           and error rates. We experienced
plus/minus       (+1-).    The horizontal           bar (control bar) contains                  sharp     decreases        in personnel            costs, even for using highly
an immobilized           polyclonal         anti-beta-hCG/hCG                  complex,         complex       interface       software,         that more than offset develop-
which captures          any conjugate.            The vertical         bar (patient’s           ment      costs. However,             many instruments                 do not support
result bar) contains immobilized                    polyclonal        anti-beta-hCG             integration        well. Our development                   of a lipid workstation,
antibody     and captures           conjugate         only in the presence of                   with use of a Hitachi             705 analyzer           (Boehringer        Mannheim,
hCG in the sample.               Accumulated              conjugate        produces         a   Indianapolis,         IN), illustrates            these issues.
colored signal at the capture site in the form of a plus sign                                       Complete        integration           requires       instruments         to provide
(+) for positive specimens                or minus sign (-) for negative                        adequate       control      and status information                 to host computers,
specimens.      This signal is visible in the result window and                                 with data-transfer             characteristics           independent         of the tim-
can be read when a red color in the end-of-assay                                 window         ing limitations           of the host computers.                In our final design,
indicates that the test is complete. The entire process, from                                   the user interacts             with both the host computer                      and the

1586      CLINICAL       CHEMISTRY,          Vol. 36, No. 9, 1990

				
DOCUMENT INFO