Intra- and inter-laboratory variability in the assessment of sperm morphology by strict criteria: impact of semen preparation, staining techniques and manual versus computerized analysis

Gerardo Barroso1, Ramazan Mercan1, Kemal Ozgur1, Mahmood Morshedi1, Paul Kolm2, Kevin Coetzee3, Thinus Kruger3 and Sergio Oehninger1,4

1 Department of Obstetrics and Gynecology, 2 Bio-statistics, The Jones Institute for Reproductive Medicine, Eastern Virginia Medical School, 601 Colley Avenue, Norfolk, Virginia, 23507, USA and 3 Reproductive Biology Unit, Department of Obstetrics and Gynecology, Tygerberg Hospital, University of Stellenbosch, Capetown, South Africa


    Abstract
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
We designed prospective studies to compare manual and computerized analysis of sperm morphology by strict criteria using different semen processing and staining techniques. A total of 54 semen samples were studied; slides were prepared from each subject from liquefied semen and after washing, and stained with Diff-Quik® or Papanicolaou. An intra-laboratory, blind assessment was performed manually (two observers) and using a computerized analyser (two readings). This demonstrated a very good correlation between manual analysis of liquefied and washed samples with both staining techniques [intraclass coefficient (ICC) = 0.93 and 0.83]. Greater agreement was observed between computerized readings (washed samples) of Diff-Quik® (ICC = 0.93) than of Papanicolaou-stained slides (ICC = 0.66). An excellent intra-laboratory correlation was observed for within-computer readings (ICC = 0.93). There was moderate agreement between inter-laboratory computer readings (two centres, ICC = 0.72). Although there was lower inter-laboratory agreement for manual and manual versus computer readings, overall results of all manual and computer analyses showed good agreement (ICC = 0.73). Diff-Quik® staining is reliable for both manual (liquefied) and computer (washed) analysis of strict sperm morphology. Intra- and inter-computer analyses using this method reached satisfactory levels of agreement. There is still high inter-laboratory variability for the manual method.

Key words: inter-laboratory variability/intra-laboratory variability/normal sperm morphology/staining/strict criteria


    Introduction
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
The assessment of sperm morphology, an integral component of the basic semen evaluation, has been demonstrated to aid the clinician in the management of male infertility (Oehninger and Kruger, 1995Go). It is clear that the evaluation of a single sperm feature or function may not provide enough power for the prediction of fertilization or implantation outcome. This reflects the complexity and multiplicity of events leading to sperm–oocyte interaction and conception. Notwithstanding these considerations, sperm morphology assessed by strict criteria (Kruger et al., 1986Go) has been shown by multiple authors to have a high predictive value not only for the outcome of advanced assisted reproductive technologies [in-vitro fertilization (IVF) and gamete intra-Fallopian transfer (GIFT)] but also for those of intrauterine insemination and in-vivo reproduction (Kruger et al., 1988Go; Oehninger et al., 1988Go; Hinting et al., 1990Go; Enginsu et al., 1991Go; Kobayashi et al., 1991Go; Check et al., 1992Go; Eggert-Kruse et al., 1993Go; Grow et al., 1994Go; Ombelet et al., 1994Go; Toner et al., 1994Go; Grow and Oehninger, 1995Go; Coetzee et al., 1998Go; Duran et al., 1998Go). A few studies, however, have not supported those findings (Coates et al., 1992Go; Matorras et al., 1995Go; Karabinus and Gelety, 1997Go).

The examination of sperm morphology by more standardized and stringent criteria has enhanced objectivity and decreased intra-laboratory variability (Menkveld et al., 1990Go). The World Health Organization (WHO, 1992) has also recommended that stricter criteria should be applied when assessing the morphological normality of the spermatozoon. This has led to the establishment of lower threshold levels for normality (WHO, 1992).

Despite these efforts, a large inter-laboratory variation still exists in the examination of sperm morphology (Comhaire et al., 1994Go; Ombelet et al., 1997aGo, 1998Go). Adequate technician training is of paramount significance to achieve consistent results within a given laboratory. Even when strict criteria (Kruger et al., 1986Go; Menkveld et al., 1990Go) are utilized, inter-laboratory variation is probably the result of various factors including: (i) different semen and smear preparation techniques, (ii) differences in interpretation, and (iii) technician experience (Coetzee et al., 1999Go).

Semen assessment by automated means has the potential to avoid the biases, subjectivity and consequent lack of intra- and inter-laboratory reproducibility inherent in the visual methodologies. Although computerized systems have been available for a few years, problems in their development have not yet allowed for their routine use (Wang et al., 1991Go; Davis et al., 1992Go; Davis and Gravance, 1994Go; Kruger et al., 1995Go; Hofmann et al., 1996Go). Use of the `signature' method (Kruger et al., 1993Go, 1995Go, 1996Go) has enabled a satisfactory level of agreement between manual (visual, optical microscopic) and computerized (incorporating strict criteria into the Hamilton Thorne Research IVOS dimensions sperm analyser) method of evaluation.

We have designed prospective, blind studies to compare manual and computerized analysis of sperm morphology by strict criteria using different semen processing and staining techniques. Our specific aims were (i) to perform an intra-laboratory comparison of morphological examination of liquefied versus washed semen, Papanicolaou versus Diff-Quik® staining techniques, and manual versus computerized analyses; and (ii) to carry out an inter-laboratory (two centres) assessment of sperm morphology comparing liquefied versus washed semen and manual versus computerized analysis. The main objective of these studies was to confirm the accuracy and reproducibility of the existing methods used to judge this important sperm feature through the application of strict criteria.


    Materials and methods
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
The studies received the approval of the Institutional Review Boards at Eastern Virginia Medical School and University of Stellenbosch. A total of 54 ejaculates from eight fertile men (donors currently enrolled in the artificial insemination programme) and 46 subfertile patients (members of infertile couples presently undergoing evaluation) were randomly selected and examined. Semen samples were obtained by masturbation after 3–5 days of sexual abstinence.

Examination of liquefied semen
Following liquefaction, a basic determination of sperm concentration and percentage of progressive motility was performed using an automated semen analyser (Cell Soft®; Cryoresources, NY, USA) and manually monitored with a haemocytometer. Smears were immediately prepared for morphological evaluation using slides pre-cleaned with 70% ethyl alcohol. A 5 µl aliquot of semen was placed on each slide, which was air-dried at 37°C in a warm tray and either stored or immediately processed. Duplicate slides were stained with Diff-Quik® (Dade AG, Dudinger, Switzerland) (Kruger et al., 1987Go). Slides were numerically coded and manually read in a blind fashion by two independent observers using strict criteria (Kruger et al., 1986Go). One hundred cells were assessed in several randomly selected fields per slide with bright field illumination and x100 oil immersion objective.

Examination of washed semen
Slides were also prepared following a previously described washing procedure (Lacquet et al., 1996Go). Briefly, after liquefaction each semen sample was washed twice using Ham's F-10 medium (Gibco Lab., Grand Island, NY, USA) supplemented with 10% bovine serum albumin (Irvine Sci., Santa Anna, CA, USA) and centrifugation for 10 min at 350 g. The final pellet was diluted with the appropriate volume of medium to obtain the original semen volume. A concentration-dependent droplet of the washed samples was then thinly smeared across cleaned slides and allowed to dry at room temperature. The sperm concentrations required for computer evaluations of the washed samples are therefore approximately double those required for manual evaluations (Coetzee et al., 1999Go). Coded, duplicate slides were then stained with Diff-Quik® and morphology was assessed blindly both manually as described above and using a computerized analyser.

Computerized analysis
The IVOS (Hamilton Thorne Research Version 2.1 Dimension Program; Hamilton Thorne Research, Beverly, MA, USA) cell analyser was used to examine sperm morphology. This system is unique in that it uses the `signature' method to evaluate the shape of the sperm head in addition to performing the standard axial measurements to determine normality (Kruger, 1995Go; Kruger et al., 1995Go, 1996Go; Coetzee et al., 1999Go). A cell's valid signature (ellipse) extracted by a polar transformation method is compared with a normal signature using a least mean squared-based method of classification (Kruger et al., 1995Go, 1996Go). For a cell to be classified as normal the parameters of size, shape and acrosome must be within specific limits. The computer settings were defined as those used in each individual laboratory for the evaluation of sperm morphology. Evaluations were performed using 662 nm wavelength illumination in conjunction with a x100 oil immersion objective. One hundred cells were evaluated per slide.

Experimental design
Slides were prepared in Norfolk for each of the 54 subjects using liquefied semen and after washing, and were then stained with Diff-Quik®. Blind assessments were performed manually (two observers) and using computerized analysis (two readings). Of the two examiners, one was a trained scientist with >5 years experience whereas the second one was a scientist with <2 years experience.

Duplicate, air-dried slides were also stained with Diff-Quik® in Tygerberg. At this centre, an additional blind reading was performed of the liquefied semen (manually) and the washed semen (by computerized analysis). The observer was a trained (>5 years experience) scientist.

For the initial 21 of the 54 samples (six fertile and 15 subfertile), additional duplicate slides were stained with Papanicolaou (Kruger et al., 1986Go; Menkveld et al., 1990Go) and analysed in Norfolk. Slides were prepared using liquefied and washed semen, and two different readings were made manually and by computer.

Statistical analysis
To assess the level of agreement between the results obtained from paired combinations of methods, staining techniques and centres, difference-average plots were obtained along with a 95% confidence interval (CI) defined as two standard deviations from the mean difference. If the results of the two methods (or techniques or centres) agree, then the points of the plot should vary randomly about a zero mean difference within the 95% CI. A large number of points falling outside the 95% CI or some trend to the plot indicate a lack of agreement (Bland and Altman, 1986Go). In addition, an intraclass correlation coefficient (ICC) was calculated for each pair. The ICC is an index of agreement ranging from 0 (no agreement) to 1 (perfect agreement). Data are presented as percentages of normal morphology (mean ± standard deviation and ranges).


    Results
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
Intra- and inter-laboratory results from the analysis of Diff-Quik®-stained slides (n = 54) assessed manually and by computer are presented in Table IGo. For the intra-laboratory assessment (Norfolk laboratory), a high degree of agreement was observed for results obtained by the manual method for liquefied samples (inter-observer variability, ICC = 0.82). There was also a very high degree of agreement between the results obtained by manual analysis of liquefied and washed semen samples (ICC = 0.93). In addition, a very high level of agreement was observed for the computer analysis of washed samples (within-computer variability, ICC = 0.93).


View this table:
[in this window]
[in a new window]
 
Table I. Indices of agreement for Diff-Quik® stained slides in Norfolk (N) and Tygerberg (T) analysed by computer (C) or manually (M) (n = 54)
 
The inter-laboratory comparisons (Norfolk and Tygerberg laboratories) demonstrated a relatively low level of agreement between the manual readings obtained from the two centres (inter-laboratory manual assessment variability, ICC = 0.57). On the other hand, a superior degree of agreement was observed for the computerized readings obtained from the two centres (inter-laboratory computer assessment variability, ICC = 0.72).

For both centres, there was only moderate agreement between manual (liquefied semen) and computerized measurements (washed semen) (Norfolk, ICC = 0.48; Tygerberg, ICC = 0.66). This was also the case for results for washed semen assessed manually and by computer (Norfolk, ICC = 0.51). Overall, however, there was good agreement between all manual measurements of liquefied semen and all computerized analyses of washed samples performed in both centres (ICC = 0.73).

Table IIGo presents the results of the comparisons of Papanicolaou and Diff-Quik® staining for liquefied and washed samples, manual and computer assessments, performed at Norfolk (n = 21). There was a high degree of agreement between results from manual analysis of liquefied and washed samples for Papanicolaou-stained samples (ICC = 0.83). There was only moderate agreement for results from the manual analysis of Diff-Quik® and Papanicolaou-stained liquefied samples (ICC = 0.69) or washed samples (ICC = 0.66). There was moderate agreement for the computerized readings of Papanicolaou-stained slides (ICC = 0.66) and lower agreement for the computer results of Papanicolaou and Diff-Quik® stained slides (ICC = 0.51).


View this table:
[in this window]
[in a new window]
 
Table II. Indices of agreement for Papanicolaou (P) and Diff-Quik® (DQ) stained slides (n = 21), analysed by computer (C) or manually (M) in Norfolk
 

    Discussion
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
These studies aimed to compare and optimize the methods and techniques presently used to assess sperm morphology using strict criteria. Liquefied semen is the source of spermatozoa generally used to examine the percentage with normal morphology. Semen washing, on the other hand, has been recommended to ensure a high-quality slide with negligible background staining in order to enhance the accuracy of the computer readings (Kruger et al., 1996Go; Menkveld et al., 1997Go). The two more widely utilized stains are the classic Papanicolaou and the more recently introduced Diff-Quik® method (Coetzee et al., 1998Go). Results of the intra-laboratory evaluation (Norfolk) demonstrated a very high degree of agreement for manual analysis between liquefied and washed samples for both staining techniques. The level of agreement was lower, however, when the two staining techniques were compared in the liquefied and in the washed samples. An excellent correlation for morphology assessment has been reported using Papanicolaou and Diff-Quik® stains in both unwashed and washed samples (Menkveld et al., 1997Go).

For the computer readings (washed samples), a very high degree of agreement was observed for the Diff-Quik® stain, substantially higher than was observed for Papanicolaou stain. Slight variations have been reported (Coetzee et al., 1997Go; Menkveld et al., 1997Go) in the percentage with normal morphology assessed using Diff-Quik® and Papanicolaou staining methods. It was concluded that the washed Diff-Quik® smears provided an optimal preparation method for computerized sperm morphology evaluation, comparing favourably with manual examinations.

The inter-laboratory comparisons (two centres) revealed, at best, only a moderate level of agreement between the manual readings obtained. Even when the same semen processing (liquefied sample), staining technique (Diff-Quik®) and classification system (strict criteria) are used, there are still differences in sperm morphology assessments between laboratories. This fact has been elegantly documented (Ombelet et al., 1997bGo, 1998Go) by a cluster analysis of the methodology and results of sperm morphology assessment in 86 centres using only one method of slide preparation and evaluation. We concur with the quality control proposal put forward by those authors. These strategies should include: (i) the advancement of reliable computerized methods; (ii) standardization of the methodology for sperm morphology assessment; (iii) geographical centralization of the assessment of sperm morphology; and (iv) reference population studies for all laboratories (Ombelet et al., 1998Go).

Here, a very low intra-computer variability (for one centre) was observed. Also, a relatively good level of agreement was found for the computerized readings obtained at the two centres. Overall, there was moderate agreement between all results from manual assessments of liquefied semen and computerized analysis of washed samples. Therefore, it is apparent that trained andrologists and computer analysis can determine sperm morphology similarly over a wide range of percentage normal forms (Hofmann et al., 1996Go; Kruger et al., 1996Go). Although improvements related to speed and overall assessment of anomalies (different types and locations within the cells) are warranted, the use of the computer seems to offer clear advantages over the manual method in terms of improved accuracy, repeatability and objective evaluation. An adequate staining resulting in good contrast, improved cell focusing and high repeatability on a cell by cell and slide by slide basis will permit a more universal and daily clinical use of this equipment (Kruger, 1995Go). In a recently performed multi-centre study (Coetzee et al., 1999Go) the magnitude of the variance produced by computer readings reached the same level as the manual evaluation and a <10% coefficient of variation was obtained if the correct quality control measures were implemented. The computerized assessment of strict morphology has therefore the potential to become the standard method to examine this basic sperm parameter.

We conclude that Diff-Quik® staining is a fast, simple and reliable method for both manual (liquefied) and computer (washed) examinations of strict sperm morphology. Intra- and inter-computer readings of washed/Diff-Quik®-stained samples demonstrated satisfactory levels of agreement and consequently offer a very promising means for the examination of this sperm feature. The high inter-laboratory variability observed for the manual method requires immediate implementation of corrective strategies.


    Notes
 
4 To whom correspondence should be addressed Back


    References
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
Bland, J.M. and Altman, D.G. (1986) Statistical methods for assessing agreement between two methods of clinical measurement. Lancet, ii, 307–310.

Check, J.H., Adelson, H.G., Schubert, B.R. et al. (1992) Evaluation of sperm morphology using Kruger's strict criteria. Arch. Androl., 28, 15–17.[ISI][Medline]

Coates, T.E., Check, J.H., Choe, J. et al. (1992) An evaluation of couples with failure of fertilization in vitro. Hum. Reprod., 7, 978–981.[Abstract]

Coetzee, K., Kruger, T.F., Vandendael, A. et al. (1997) Comparison of two staining and evaluation methods used for computerized human sperm morphology evaluations. Andrologia, 29, 133–135.[ISI][Medline]

Coetzee, K., Kruger, T. F. and Lombard, C.J. (1998) Predictive value of normal sperm morphology: a structured literature review. Hum. Reprod. Update, 4, 73–82.[Abstract/Free Full Text]

Coetzee, K., Kruger, T.F., Lombard, C.J. et al. (1999) Assessment of inter- and intra-laboratory sperm morphology readings using a Hamilton Thorne research IVOS semen analyser. Fertil. Steril., 71, 1–5.[ISI][Medline]

Comhaire, F., Schoonjans, F., Vermeulen, L. et al. (1994) Methodological aspects of sperm morphology evaluation: comparison between strict and liberal criteria. Fertil. Steril., 62, 857–861.[ISI][Medline]

Davis, R.O. and Gravance, C.G. (1994) Consistency of sperm morphology classification methods. J. Androl., 15, 83–91.[Abstract/Free Full Text]

Davis, R.O., Bain, D.E., Siemers, R.J. et al. (1992) Accuracy and precision of the CellForm-Human automated sperm morphometry instrument. Fertil. Steril., 58, 763–769.[ISI][Medline]

Duran, E.H., Gürgan, T., Günalp, S. et al. (1998) A logistic regression model using DNA status and morphology of spermatozoa for prediction of fertilization in vitro. Hum. Reprod., 13, 1235–1239.[Abstract]

Eggert-Kruse, W., Schwarz, H., Rorh, G. et al. (1993) Sperm morphology assessment using strict criteria and relationship with male fertility under in-vivo conditions. In Ombelet, W., Bosmans, E., Vandeput, H. et al. (eds), Modern ART in the 2000s: Andrology in the Nineties. Parthenon Publishing, New York, pp. 61–72.

Enginsu, M.E., Dumoulin, J.C.M., Pieters, M.H.E.C. et al. (1991) Evaluation of human sperm morphology using strict criteria after Diff-Quik® staining: correlation of morphology with fertilization in vitro. Hum. Reprod., 6, 854–858.[Abstract]

Grow, D. and Oehninger, S. (1995) Strict criteria for the evaluation of human sperm morphology and its impact on assisted reproduction. Andrologia, 27, 325–333.[ISI][Medline]

Grow, D., Oehninger, S., Seltman, H. et al. (1994) Sperm morphology as diagnosed by strict criteria: probing the impact of teratozoospermia on fertilization rate and pregnancy outcome in a large IVF population. Fertil. Steril., 62, 559–565.[ISI][Medline]

Hinting, A., Comhaire, F., Vermeulen, L. et al. (1990) Value of sperm characteristics and the result of in vitro fertilization for predicting the outcome of assisted reproduction. Int. J. Androl., 13, 59–66.[ISI][Medline]

Hofmann, G.E., Santilli, B.A., Kindig, S.K. et al. (1996) Intraobserver, interobserver variation of sperm critical morphology: comparison of examiner and computer-assisted analysis. Fertil. Steril., 65, 1021–1025.[ISI][Medline]

Karabinus, D.S. and Gelety, T.J. (1997) The impact of sperm morphology evaluated by strict criteria on intrauterine insemination success. Fertil. Steril., 67, 536–541.[ISI][Medline]

Kobayashi, T., Jinno, M., Nozawa, S. et al. (1991) Sperm morphological assessment based on strict criteria and in vitro fertilization outcome. Hum. Reprod., 6, 983–986.[Abstract]

Kruger, T.F., Menkveld, R., Stander, F.S.H. et al. (1986) Sperm morphologic features as a prognostic factor in in-vitro fertilization. Fertil. Steril., 46, 1118–1123.[ISI][Medline]

Kruger, T.F. (1995) Computer-assisted sperm analysis systems: morphometric aspects. Hum. Reprod., 10 (Suppl. 1), 46–52.[ISI][Medline]

Kruger, T.F., Acosta, A.A., Simmons, K.F. et al. (1987) New method of evaluating sperm morphology with predictive value for human in vitro fertilization. Urology, 30, 248–251.[ISI][Medline]

Kruger, T.F., Acosta, A.A., Simmons, K.F. et al. (1988) Predictive value of abnormal sperm morphology in in-vitro fertilization. Fertil. Steril., 49, 112–117.[ISI][Medline]

Kruger, T.F., DuToit, T.C., Franken, D.R. et al. (1993) A new computerized method of reading sperm morphology (strict criteria) is as efficient as technician reading. Fertil. Steril., 59, 202–209.[ISI][Medline]

Kruger, T.F., DuToit, T.C., Franken, D.R. et al. (1995) Sperm morphology: assessing the agreement between the manual method (strict criteria) and the sperm morphology analyser IVOS. Fertil. Steril., 63, 134–141.[ISI][Medline]

Kruger, T.F., Lacquet, F.A., Sanchez Sarmiento, C.A. et al. (1996) A prospective study on the predictive value of normal sperm morphology as evaluated by computer (IVOS). Fertil. Steril., 66, 285–291.[ISI][Medline]

Lacquet, F.A., Kruger, T.F., DuToit, T.C. et al. (1996) Slide preparation and staining procedures for reliable results using computerised morphology (IVOS). Arch. Androl., 36, 133–138.[ISI][Medline]

Matorras, R., Corcóstegui, B., Perez, C. et al. (1995) Sperm morphology analysis (strict criteria) in male infertility is not a prognostic factor in intrauterine insemination with husband's sperm. Fertil. Steril., 63, 608–611.[ISI][Medline]

Menkveld, R., Stander, F.S.H., Kotze, T.J.W. et al. (1990) The evaluation of morphological characteristics of human spermatozoa according to stricter criteria. Hum. Reprod., 5, 586–592.[Abstract]

Menkveld, R., Lacquet, F.A., Kruger, T.F. et al. (1997) Effects of different staining and washing procedures on the results of human sperm morphology evaluation by manual and computerised methods. Andrologia, 29, 1–7.[ISI][Medline]

Oehninger, S. and Kruger, T. (1995) The diagnosis of male infertility by semen quality. Hum. Reprod., 10, 1037–1041.[ISI][Medline]

Oehninger, S., Acosta, A.A., Morshedi, M. et al. (1988) Corrective measures and pregnancy outcome in in-vitro fertilization in patients with severe sperm morphology abnormalities. Fertil. Steril., 50, 283–287.[ISI][Medline]

Ombelet, W., Flourie, F., Vandeput, H. et al. (1994) Teratozoospermia and in-vitro fertilization: a randomized prospective study. Hum. Reprod., 9, 1479–1484.[Abstract]

Ombelet, W., Bosmans, E., Janssen, M. et al. (1997a) Semen parameters in a fertile versus subfertile population: a need for change in interpretation of semen testing. Hum. Reprod., 12, 987–993.[ISI][Medline]

Ombelet, W., Wouters, E., Boels, L. et al. (1997b) Sperm morphology assessment: diagnostic potential and comparative analysis of strict or WHO criteria in a fertile and a subfertile population. Int. J. Androl., 20, 367–372.[ISI][Medline]

Ombelet, W., Cox, A., Janssen, M. et al. (1998) Methodological differences in sperm morphology assessments. In Ombelet, W., Bosmans, E., Vandeput, H. et al. (eds), Modern ART in the 2000s: Andrology in the Nineties. Parthenon Publishing, New York, pp. 61–72.

Toner, J.P., Mossad, H., Grow, D.R. et al. (1994) Value of sperm morphology assessed by strict criteria for prediction of the outcome of artificial (intrauterine) insemination. Andrologia, 27, 143–148.[ISI]

Wang, C., Leung, A., Tsoi, W. et al. (1991) Computer-assisted assessment of human sperm morphology: usefulness in predicting fertilizing capacity of human spermatozoa. Fertil. Steril., 55, 989–993.[ISI][Medline]

World Health Organization (1992) WHO Laboratory Manual for the Examination of Human Semen and Semen–Cervical Mucus Interaction, 3rd edn. Cambridge University Press, Cambridge, pp. 13–17.

Submitted on February 9, 1999; accepted on April 29, 1999.