Doppler assessment of the ductus venosus and the tricuspid valve at 11–13⁺⁶ weeks: Reference ranges and development of sonographic quality assurance standards

Abstract

Objective

To develop reference ranges for the ductus venosus (DV) and tricuspid valve (TV) waveforms at 11–14 weeks and define auditable standards to assess operator measurement performance.

Materials and methods

A single operator prospectively obtained quantitative measurements of a number of DV and TV Doppler indices to develop medians and 90% reference intervals (RIs). Measurement agreement studies between two experienced operators were also performed. The measurement bias of three additional operators was subsequently assessed using the newly defined auditable standards.

Results

Doppler measurements were obtained in 292 patients (DV) and 321 patients (TV). Reference ranges were constructed for DV pulsatility index for veins (PIV) which did not change over gestation (mean 1.06; 90% RI 0.86–1.23) and for the TV E–A ratio reference range which was positively correlated with gestation. Measurement agreement studies on 30 additional patients showed the within‐operator agreement was almost perfect for DV PIV (ICC, intraclass correlation 0.82–0.86) and strong for TV E–A ratio (ICC 0.68–0.78) while the between‐operator agreement was good for both DV PIV and TV E–A ratio measurements (ICC 0.46 for both).

Discussion

Development of local reference ranges enabled the definition of quantitative auditable standards that can be utilised in assessment of operator training and ongoing Doppler measurement quality assurance. Measurements of DV PIV and TV E–A ratio by experienced operators were found to be reproducible.

Introduction

Screening for fetal chromosome abnormality is now pre‐dominantly performed in the first trimester.1 Combined first trimester screening (cFTS) using maternal age, nuchal translucency thickness (NT) and the maternal serum biochemical analytes provides an effective means of screening for trisomy 21 (detection rate: 90%, false‐positive rate: 5%) and other major chromosomal abnormalities.2, 3 The accuracy of the test can be improved by the use of additional fetal haemodynamic markers, monitoring flow through the ductus venosus (DV) and the tricuspid valve (TV). The addition of these markers have been shown to increase the detection rate for trisomy 21 (93–95%) while maintaining a low false‐positive rate.4, 5, 6 Additionally, these markers are associated structural cardiac defects, stillbirth risk and twin–twin transfusion risk in monochorionic twin pregnancies.5, 7, 8

Robust screening tools need to be easy to perform and reproducible. Effective screening relies first on appropriate training of sonographers, demonstration of competence, structured introduction into clinical practice then ongoing audit of performance. While the value of this training and audit cycle is well demonstrated for measurement of NT, there is less published data describing this process for assessment of DV and TV.9, 10, 11, 12 Qualitative assessment of Doppler waveforms [DV: positive, absent or reversed; TV: presence or absence of valvular regurgitation (TR, tricuspid regurgitation)] is considered harder to perform than measuring NT and the assessment is constrained by subjective variability and errors in image interpretation.13 The use of quantitative measurement of Doppler indices6, 14, 15 may reduce subjectivity of assessment, provide simpler audit of sonographer performance and allow these factors to be incorporated into risk algorithms as continuous rather than categorical variables, thus preventing sudden swings in aneuploidy risk.15, 16, 17, 18

The study aims are: (i) to construct references ranges for blood flow characteristics of DV and TV waveforms at 11–13⁺⁶ weeks’ gestation in an Australian obstetric population, and (ii) to examine measurement agreement and to provide a framework for developing auditable quality standards to assess measurement performance of individual operators.

Materials and methods

Main DV and TV assessment study

This was a prospective study performed between February 2012 and February 2013. Patients were recruited from a private obstetric and gynaecological ultrasound practice that provides specialist prenatal screening, counselling and diagnostic services. Women attending for routine cFTS were approached to participate in the study and consent was obtained in accordance with ethical clearance from Charles Sturt University (study no: 2012/60) and Genea Human Research Ethics Committee (study no: GEC008).

Women with a multiple pregnancy or those outside the gestational age window for cFTS (fetal crown‐rump length 45–84 mm) were excluded from the study. Data were not collected from a proportion of women where visualisation of the DV and/or TV was inadequate due to increased maternal body habitus, uterine retroversion or poor fetal position. All examinations were restricted to a maximum of 30 min. Data were excluded from analysis from cases that were subsequently found to have a congenital abnormality, who had frankly abnormal haemodynamic findings (absent or reversed DV a‐wave or TR) or who had a high aneuploidy risk from cFTS.

All measurements were made by a single, experienced, NT‐certified sonographer using the techniques described by the Fetal Medicine Foundation.4, 5 In brief, DV assessment involved magnification of the fetus in a right ventral sagittal section such that the thorax and abdomen occupied the whole image. Colour flow mapping was used to demonstrate the umbilical vein, DV and fetal heart. The pulsed Doppler sample volume was set at 0.7 mm, to avoid contamination from adjacent vessels, and placed in the aliasing area. The insonation angle was less than 30° and the filter was set at a low frequency (50–70 Hz) so that the a‐wave was not obscured. A high sweep speed was utilised (2–3 cm/s) for optimal assessment of the a‐wave (Figure 1).

Figure 1.

Normal ductus venosus Doppler waveform (S: fetal ventricular systolic contraction; D: fetal early ventricular diastole; A: fetal atrial contraction).

Tricuspid valve assessment involved high magnification of an axial section of the thorax such that an apical four‐chamber view of the fetal heart was obtained. A pulsed‐wave Doppler sample volume of 3.0 mm was positioned across the TV at least three times at different valvular segments (medial valve leaflet insertion, centrally and at the lateral insertion site) in order to interrogate the entire valve and exclude TR. The angle of insonation was less than 30° in relation to the inter‐ventricular septum (ideally 0°). The sweep speed was high (2–3 cm/s) so that the waveforms were widely spread for better assessment (Figure 2).

Figure 2.

Normal tricuspid valve Doppler waveform (E: early ventricular filling velocity; A: atrial filling velocity).

Quantitative DV assessment involved measurements of the pulsatility index for veins (PIV), pre‐load index (PLI) and peak velocity for veins (PVIV). TV assessment involved measurement of e‐wave peak velocity, a‐wave peak velocity and calculation of the E–A ratio. Qualitative evaluations of the DV and TV waveforms [DV: positive, absent or reversed; TV: TR absent or present (defined as retrograde flow for >50% of the cardiac cycle to ≥60 cm/s)] were also recorded. The examinations were performed using a Voluson E8 Expert (GE Medical Systems, Milwaukee, WI, USA) with a RAB 4‐8L probe during fetal quiescence. No additional examination time was allocated, and the ALARA principle was observed.19

Measurement agreement study

Thirty patients were examined by two experienced operators. Each patient was scanned by both operators on the same day, in random order, with blinding of DV and TV measurement results for both.

Pilot study of operator measurement audit

Four experienced, NT‐certified operators were instructed in the technique for DV and TV Doppler assessment. Measurement bias was assessed using auditable standards derived from this study.

Statistical analysis

All measurements were included, regardless of outlier status. Trends in the measurement data over gestation were investigated using scatterplots and locally weighted scatterplot smoothing (lowess) curves. Such plots are usual for visually determining both linear and non‐linear patterns without imposing any assumptions and are useful for determining whether heterogeneity and skewness exist in the data. Quadratic, linear and constant regression models were used to formally test any patterns in the data. Distributional assumptions, residual tests, and influence diagnostics were checked using Shapiro–Wilk's test of normality, histograms and scatterplots.20

To assess measurement agreement, single‐measure intraclass correlation (ICC) was calculated which assesses concordance for continuous variables, correcting correlation for systematic bias and measuring the extent to which those variables will yield the same score when assessed in different conditions, namely, between and within observers.21 The following agreement benchmarks were used for ICC characterisation: poor (0–0.2), fair (0.21–0.4), moderate (0.41–0.6), strong (0.61–0.8) and almost perfect (0.81–1.0).21 All statistical analyses were performed using stata version 12.0 (Stata Corp, College Station, TX, USA) and a level of α = 5% was used to determine statistical significance.

Results

In total, 507 patients who presented for cFTS were approached to participate in the study, of whom 436 with a singleton pregnancy (86%) agreed. All cases were in the correct gestational window. Neither DV nor TV could be adequately assessed in 50/402 (12.4%) cases. One or other parameter was not assessed in a further 29 cases (11 DV and 18 TV) but these cases were not excluded. A total of 47 cases were excluded from analysis due to NT > 3.5 mm [5], qualitative abnormality of DV (absent a‐wave [13]; reversed a‐wave [26]) and tricuspid regurgitation [3]. No fetuses with chromosomal abnormality were identified. The final study population included 292 patients for DV assessment and 321 patients for TV assessment. The demographics and characteristics of the DV and TV study participants are presented in Table 1. The majority of the participants were of European origin, the average age was 33.0 years (range: 24–42 years), less than 13% weighed above 80 kg, and approximately 5% had assisted reproduction.

Table 1.

Demographic characteristics of the study participants

Demographic variable	Ductus venosus assessment group (n = 298), n (%)	Tricuspid valve assessment group (n = 321), n (%)	P‐values
Ethnicity
European origin	257 (88.0)	287 (89.4)	0.7
East Asian	24 (8.2)	22 (6.9)	0.65
South Asian	9 (3.1)	9 (2.8)	0.99
Other	2 (0.7)	3 (0.9)	0.87
Age (years)
<25	1 (0.3)	2 (0.6)	0.97
25–29	58 (19.9)	69 (21.5)	0.70
30–34	147 (50.3)	163 (50.8)	0.97
35–39	72 (24.7)	76 (23.7)	0.85
≥40	14 (4.8)	11 (3.4)	0.50
Weight categories (kg)
<50	7 (2.4)	6 (1.9)	0.89
50–59	77 (26.4)	90 (28.0)	0.73
60–69	123 (42.1)	129 (40.2)	0.69
70–79	50 (17.1)	56 (17.5)	0.98
80+	35 (12.0)	40 (12.5)	0.95
Parity
0	122 (41.8)	132 (41.3)	0.97
1	135 (46.2)	147 (45.9)	0.99
2	26 (8.9)	32 (10.0)	0.75
≥3	9 (3.1)	9 (2.8)	0.99
Assisted reproduction
No	277 (94.9)	306 (95.3)	0.97
Yes	15 (5.1)	15 (4.7)	0.97

The mean crown‐rump length (CRL) measurement was 63.6 (SD 6.9; range: 45.7–82.0 mm). The DV Doppler indices PIV, PVIV and PLI were highly correlated (r values: 0.82–0.91), therefore only a single reference range was established for the most commonly utilised index, PIV. The DV PIV data was not normally distributed, requiring square root transformation, and assessment of the lowess curve demonstrated that the DV PIV remained relatively constant across the gestational age range. Regression assessment suggested a constant‐only model best described the data with the following equation: e ^{[DVPImean 2] CRL} = 1.128. The mean DV PIV measurement was 1.06 (90% RI, reference interval 0.86–1.23). The empirical frequencies of measurements falling outside the 90% RI and associated P‐value derived from the exact binomial test met expectation. Figure 3 shows a scatterplot of DV pulsatility index with superimposed mean line, lowess curve and 90% RI. Regression analysis showed no influence of age, parity, BMI, or conception method on the model. Ethnicity was regressed against the square root transformed data and showed that East Asians had a significantly lower DV PIV than Caucasians (P < 0.007).

Figure 3.

Ductus venosus median, lowess curve and 90% reference interval.

Tricuspid valve assessment

The mean CRL measurement was 63.4 (SD 6.8; range: 45.7–82.0 mm). There was a very high correlation between the values for both e‐wave and a‐wave when measured across the central, medial and lateral parts of the TV (r values: 0.78–0.94), thus a single measurement across the valve is sufficient. The e‐wave and a‐wave change reliably over gestation and, as both are represented in the E–A ratio, a reference range was constructed only for the E–A ratio. The E–A ratio values were also highly correlated for the different valve segments (r values: 0.58–0.88) thus a single reference range was sufficient to describe the measurement. The mean E–A ratio was 0.60 (SD 0.05; range: 0.46–0.78). The E–A ratio values are normally distributed and assessment of the lowess curve demonstrated that the TV E–A ratio increased across the gestational age range. Fractional polynomial analysis identified the first‐level model (power −2) as best describing the mean trend in these data, resulting in the equation [Mean (E–A ratio) = 0.6924808−(357.0799 × CRL⁻²); SD 0.05]. There was no significant or important change in the variability of TV E–A ratio measurement residuals over the CRL measurement range. Depicted in Figure 4 is a scatterplot of TV E–A ratio with superimposed mean line, lowess curve and 90% RI. Regression analysis showed no influence of age, parity, BMI, or conception method on the model but, once again, East Asians were noted to have a significantly lower TV E–A ratio than Caucasians (P = 0.001).

Figure 4.

Tricuspid valve E–A ratio mean, lowess curve and 90% reference interval.

Measurement agreement studies

Qualitative agreement

Both operators agreed that the DV waveform was normal in 24/30 (80%) patients. Both operators agreed that the waveform was abnormal (reversed a‐wave) in 2 (7%) patients. Operator 1 found the waveform was abnormal (absent a‐wave,1 reversed a‐wave3) in a further 4 (13%) patients that were found to be normal by Operator 2. The operators agreed that the TV waveform was normal, with no TR, in all 30 patients.

Quantitative agreement

The intra‐observer and inter‐observer measurement agreement were formally tested for the DV PI and TV E–A ratio measurements. Estimated ICC show almost perfect within‐operator agreement for DV PI [ICC: 0.86 (op 1), 0.82 (op 2)] and strong within‐operator agreement for TV E–A ratio [ICC: 0.68 (op 1), 0.78 (op 2)]. The between‐operator agreement is moderate for both variables (ICC: 0.46 for both).

Pilot study of operator measurement audit

Four operators each examined between 13 and 39 patients. Their DV PI and TV E–A ratio measurements were converted to multiples of the gestation‐related median (MoM) values, utilising the mean equations established by this study. The median MoM and standard deviation of the log MoM values were computed for each operator and each variable and are presented in Table 2. Three of the four operators over‐estimated DV PI (1.07–1.13 MoM) with a relatively low spread of results (SD log MoM 0.03–0.06). Half the operators under‐measured the TV E–A ratio (0.93 and 0.94 MoM) again with relatively low spread (SD log MoM 0.04–0.07). Overlaying the 90% RI chart with a scatter plot of individual measurements and computing the percentage of measurements above the 50th percentile gave a visual representation of the measurement bias (Figure 5).

Table 2.

Ductus venosus and tricuspid valve Doppler measurements of the four operators

Operator	Ductus venosus PI				Tricuspid valve E–A ratio
	n	% >50th percentile	Median MoM	SD log MoM	n	% >50th percentile	Median MoM	SD log MoM
1	38	42.1	0.97	0.06	39	41.0	0.98	0.10
2	31	90.3	1.09	0.03	32	37.5	0.94	0.04
3	13	92.3	1.13	0.06	15	13.3	0.93	0.07
4	30	76.7	1.07	0.06	30	40.0	0.98	0.05

PI, pulsatility index; MoM, multiple of median.

Figure 5.

Visual representation of measurement bias in DVPI measurement (operator B overestimates DVPI compared to the appropriate performance of operator A).

Discussion

Means and 90% RIs were constructed for both DV PIV and TV e‐wave: a‐wave ratio at 11–13⁺⁶ weeks’ gestation. The DV PIV mean shows no change over gestational age while the TV E–A ratio is positively related to CRL between 11 and 13⁺⁶ weeks. Both Doppler parameters are unrelated to maternal age, parity, weight or conception method; however, significantly lower values for both DV PIV and TV E–A ratio were found in an East Asian cohort compared with Caucasian participants. The other key findings from this study are: (i) that DV PIV and TV E–A ratio measurements are reproducible when performed by experienced operators and (ii) that all DV Doppler indices and segmental TV measurements were highly correlated, therefore reference ranges for only the DV PIV and TV E–A ratio were constructed and used in the development of auditable standards.

Over the past decade there have been at least seven reported 11–14 week DV PIV reference ranges.22, 23, 24, 25, 26, 27, 28 Two of these lie above the current study mean and four lie below it, including three of the most recent publications. Excluding the study by Teixeira et al.25 which is markedly higher than all others, the PIV values across these various reference ranges differ by 9% at 11 weeks, increasing to 18% by 14 weeks. Four show a small but significant negative relationship with gestation, whereas this study shows no significant change over gestation after careful model evaluation, which is supported by the recent study by Pruksanusak et al.28 PIV values are generally lower when the angle of insonation increases and if there is contamination from surrounding vessels. This study was specifically designed to construct a reference range by a single, highly experienced operator where the measurement criteria were strictly enforced, with a relatively high number of subjects excluded as a consequence. This is unlikely to be the case where reference ranges are drawn from large cohorts in a routine screening service using multiple machines and operators.22, 27 Technical improvements in Doppler imaging, magnification and tracing have resulted in considerable improvements in resolution over the years thus modification of reference ranges may be appropriate over time in the same institution.22 There are several studies reviewing the mean TV E–A ratio at 11–14 weeks 16, 29 but only one where a reference range has been constructed over this gestational range.30 In agreement with the current study, this showed a positive relationship with gestation. The values are generally lower than those in the current study but the difference (range: 3–9%) is not significant.

Measurement agreement studies show both DV and TV assessments to be reproducible by experienced operators. This study shows almost perfect within‐operator agreement for DV PI and strong within‐operator agreement for TV E–A ratio. The between‐operator agreement is moderate for both variables. These results compare favourably with previous studies.15, 17, 23 It is interesting to observe that between‐operator measurement reliability for DV PIV has altered little compared with studies performed more than 5 years ago despite improvements in machine technology and tighter measurement criteria.15, 23 This highlights the fact that DV measurements appear more prone to natural variability within an examination time‐frame, they are technically challenging to perform and contamination of the waveform by other vessels remains a constant concern.12

In 12.3% of examinations, neither the DV nor the TV could be adequately assessed due to maternal or fetal factors, which is higher than noted in previous studies.4, 5 This may be due to the strict adherence to measurement criteria inherent in a reference range construction study with rigorously excluding images showing any degree of neighbouring vessel contamination or difficulty obtaining images without angle correction. Furthermore, in the measurement agreement study there was discordance in the qualitative assessment of the DV Doppler trace in 4 of 30 images (13%) where absence or reversal of the a‐wave was masked by vessel contamination but concordance was reached after image review. Ongoing training and review, even for experienced first trimester operators, is therefore an important consideration before implementation of these measurements into routine clinical practice. Qualitative assessment of the TV waveform has not been extensively studied but Falcon et al.31 found concordance in 36 (87.8%) of the 41 cases in which regurgitation was present and in 85 (97.7%) of the 87 in which it was absent (in 128 cases examined by both obstetricians and fetal cardiologists).

The primary purpose of constructing means and 90% RIs for the DV PIV and TV E–A ratio in this study was to create a platform to assess operator measurement accuracy and develop a method of providing constructive feedback on performance. The simplest form of assessing measurement bias involves the use of central tendency and dispersion.32 Using multiples of the gestation‐related median allows comparison of measurement performance across all 11–14 week screening examinations. The pilot audit study showed 3/4 over‐estimated the DV PIV and half under‐estimated the TV E–A ratio. However, the precision of those measurements was high as indicated by the low spread of results. Image review was able to identify errors in approximately 15% of cases.

The power of visual cues cannot be over‐estimated in providing feedback to operators. Showing the operators scatterplots of their measurements overlaid on the 90% RI for both Doppler measurements improved the understanding of their measurement bias compared with the median MoM value alone. These standards necessarily are applied retrospectively and the opportunities to influence operator's measurement errors are therefore limited. Another approach is the use of cumulative summation (CUSUM) plots as a prospective, visual, quality assurance tool. Measurement deviation beyond the established control limits signal a potential performance deviation. This approach has the advantage of early intervention to assess and correct measurement technique, and was recently found to add value to the distribution parameters as a quality assurance method.18 Further prospective studies evaluating groups of operators with varying levels of experience would also be useful to assess the utility of this type of performance audit.

The strengths of the study include well‐defined inclusion and exclusion criteria, rigorous adherence to a strict examination protocol using a single experienced operator and careful statistical analyses and diagnostic checks.33 Despite these measures, possible inherent weaknesses in the study relate to: (i) the technical challenges of correct performance of both DV and TV measurements at 11–14 weeks gestation, (ii) smoking status influence on the Doppler measurements could not be ascertained, (iii) the modest sample size (particularly in the CRL range 75–84 mm) which may mask a true DV PIV reduction in later gestation, and (iv) the pre‐dominance of Caucasians in the cohort. The findings of significantly lower DV PIV and TV E–A ratio measurements in a relatively small number of East Asian patients needs review in a larger study and possibly construction of local reference ranges or incorporation of appropriate weighted regression factors in the reference range equation.27

First trimester markers including DV and TV have historically been assessed qualitatively, with potential for subjective errors in image interpretation.13 Screening performance can be improved by a quantitative approach.14 This study has demonstrated that quantitative DV and TV Doppler assessment are accurate, reproducible and can be utilised in the assessment of operator performance. This could be further assessed in a larger cohort of patients and operators to determine the most appropriate audit metrics.