Reproducibility of aortic intima-media thickness in infants using edge-detection software and manual caliper measurements

Atherosclerosis, the pathological basis of cardiovascular disease, has its origins in the neonatal period [1‐3]. Research into the fetal origins of adult-onset cardiovascular disease is increasingly using aortic intima-media thickness as measured by trans-abdominal ultrasound (aIMT) as a putative intermediate phenotype of cardiovascular risk [4]. Aortic IMT has shown to be superior to carotid IMT in both feasibility and association with known cardiovascular risk factors [5, 6]. Small exposure-specific studies in newborn infants have reported associations between increased aIMT and cardiovascular risk factors such as intra-uterine growth restriction (IUGR), macrosomia, maternal diabetes, maternal hypercholesterolaemia, and maternal smoking [7‐10].

Post mortem studies have shown varying degrees of diffuse intimal thickness present in arteries prone to atherosclerosis from birth [1, 11, 12]. Ultrasound measurement of intima-media has strongly correlated with histological thickness vessels prone to atherosclerosis, in particular carotid and coronary arteries [13, 14]. The reproducibility of newborn aIMT as measured by abdominal ultrasound has only been established in small, tertiary hospital-based studies using specialist vascular ultrasonographers [7, 9, 10]. Reported intra-class correlations (ICC) between ultrasound readers are high (β = 0.89-0.94) [7, 8, 15, 16]; the largest reported mean difference between sonographers was 16 μm [17]. These results are encouraging, but their generalisability, especially to population-based studies conducted outside the tertiary setting, is uncertain.

To date the only normative data regarding aIMT during early infancy are from two small studies conducted in term infants on days 1–4 of life. These studies used different inclusion criteria and reported substantial differences in mean values and variation in their aIMT measurements [15, 16]. Furthermore, these data are different from the distributions presented in other studies of infants at the same age (1–4 days) that compared specific exposure groups (e.g. IUGR infants) to ‘normal’ infants as controls [7, 9, 10]. Published data only pertain to the first few days of life and the distribution of aortic IMT in older infants has not been reported. Given these limitations, it is relevant to establish the distribution of aIMT in a large, population-derived cohort over the first weeks to months of life. To our knowledge there have been no reported data on the effect of environmental and individual variables on image quality in early infancy or childhood.

This study aimed to establish the reproducibility and distribution of aIMT during early infancy in a birth cohort study conducted in a community setting. In addition it aimed to compare the aIMT measures obtained using edge-detection software and manual sonographic calipers and investigate whether image quality is affected by individual and environmental variables.

The Barwon Infant Study (BIS) is a population-derived cohort of 1069 infants recruited prior to 33 weeks gestation. Infants were excluded from BIS if they developed a serious illness in the first few days of life, or if they had major congenital malformations or genetic abnormalities. For the current analysis, preterm infants (<37 weeks gestational age) were excluded.

Of the 1069 eligible infants in BIS, 978 (91%) completed their six week visit, during which 844 (86%) had aIMT measured successfully. Excluding preterm infants, aIMT data were available from 814 infants. Five hundred and seventy three of the 814 aIMT images (70%) were assessed as either ‘good’ or ‘adequate’ quality. Gender was evenly represented (53% male), and the average birth weight was 3.5 kg, approximately the 50^th centile in the population [20]. In addition, 290 (35%) infants had aIMT measured using sonographic calipers; these infants were similar to the main study cohort (Table 1).

We have demonstrated that aIMT is reproducible in a large community-based study. There was found high intra and inter-operator ICC between (a) the results obtained using paired ultrasound cineloops obtained by two operators and (b) off-line edge-detection measurements obtained from two readers using the same cineloops, using either edge-detection software or manual caliper measurements. In term infants assessed at 6 weeks, there were differences in aIMT values obtained using different measurement methods; mean aIMT was greater using edge-detection compared to manual sonographic calipers. Furthermore, there was no need for repeat measurements using edge-detection software. With the exception of infant crying, individual and environmental factors did not impact on the quality of images obtained. There was a trend of association between increasing age at scan and higher aIMT measurement.

Previous studies have demonstrated that aIMT is a reproducible measure when conducted by specialist sonographers in a tertiary setting. We have extended these findings by demonstrating that a high reproducibility can be obtained by general research staff following a period of training in the technique. The ICC between research operators performing repeated ultrasounds and readers measuring the same image are consistent with the ICC reported by other groups [7, 9, 10, 16]. Indeed the reproducibility of aIMT measures observed is similar to that reported for adult carotid IMT, an integral part of cardiovascular risk assessment [4]. These findings have important feasibility implications for future population-based research investigating the early life origins of cardiovascular disease.

Our study was able to assess reproducibility but not the dimension of predictive validity –that is we are unable to compare our results to “true” histological measurements of the newborn aorta. We must use intra-observer (test-retest) and inter-observer variability as a proxy for accuracy. A recent study looking at porcine histology as well as aIMT in young children suggested that the use of very-high frequency ultrasound (25–30 Hz) in infants and young children produced more accurate results than high frequency ultrasound may produce more accurate results than high frequency ultrasound, as used in this study [21].Introducing another method of assessment with undoubtedly make comparisons between study results even more difficult.

This study provides the first normal data for healthy infant aIMT data for 6 weeks of age. Values described as ‘normal’ for newborns in the first days of life have been obtained from small studies (n < 100) that have used selected sampling frames and varying definitions of ‘normal’ that are unlikely to be fully representative of the population. For example, a study by Hondappanavar et al. measured aIMT in 100 term infants, but excluded all those below the 50^th centile for birth weight [16]. Another study by Koklu et al. investigated preterm and term infants of greater than 24 weeks gestational age, including 60 term infants [15]. This study excluded any infants with putative risk factors for increased aIMT (congenital abnormality, IUGR, macrosomia, maternal smoking, abnormal lipid profiles, maternal dyslipidaemia, hypertension or maternal history of cardiovascular disease) [15]. In contrast, our study was more inclusive; only preterm infants and those with congenital abnormalities or significant neonatal illness were excluded.

In addition to sampling considerations, reported newborn values of mean aIMT are highly variable. Variability may arise due to differences in ultrasound equipment, transducer frequency, and, as our study demonstrates, measurement methods. Using manual sonographic calipers, the following studies reported substantial variation. Whilst Koklu et al. (n = 60) reported a mean of 385 μm with SD 19 at term [15], Hondappanavar et al. (n = 100) found a mean 510 μm (41) [16]. Other data are available from healthy control groups in exposure-specific studies that again used sonographic calipers for measurement. For example Koklu et al. reported, in two separate exposure-specific studies, control values of 400 μm (30) (n = 40) [9] and 390 μm (30) (n = 30) [7]. In contrast Skilton et al. (n = 25) [10] used edge-detection software and reported control values of 534 μm (58) [10]. Surprisingly, the highest control aIMT measures are from Zanardo’s in utero study (n = 21) of infants at 32 weeks gestation which reported mean 1050 μm (190) [22]. Zanardo et al. used caliper measurement, but a lower frequency transducer to other studies (3.5-5 Hz versus 8-13 MHz/3-12 MHz). Clearly there are also other significant differences with antenatal imaging such as the depth of imaging, angle of insonation and attenuation that will alter the resolution of images. Our data demonstrate important limitations when comparing results between studies that are not obtained using identical methods. Furthermore, similar to adult studies [23], we show that edge-detection software is at least equivalent to manual caliper measurement in reproducibility (higher ICC) and superior in ease of measurement. Ideally, all future studies would adopt the same ultrasound and measurement protocol to allow accurate comparisons between study results.

In contrast to previous studies that examined infants aged 0–4 days, our study participants varied in age around a median of 5.9 weeks [5.1-7.0]. Aortic IMT has previously been reported to increase with both gestation [15] and age [24]. Concordantly, we found a positive association between aIMT measured by edge-detection software and age at time of scan. Thus the potential influence of gestation and age should be considered when investigating the determinants of aIMT during early life and analyses should be adjusted accordingly.

With the exception of infant crying, we found no evidence that individual and environmental variables, including timing of ultrasound scans during study visits, affected image quality. The effect of environmental variables on image quality is particularly relevant beyond the first days of life, as infants become progressively increasingly more physically active, potentially affecting the quality of the ultrasound images. Our measures of aIMT occurred in the context of multiple other research procedures, such as anthropometric and lung function measurements, and this is likely to the case for many studies of a similar nature to BIS. These results reinforce the feasibility of newborn aIMT measurement in a community-based study where other measures are being taken during a single visit.

Aortic IMT is a highly reproducible measure that is suitable for use in population-based studies of cardiovascular health and development. In 6 week old infants, aIMT approximates a normal distribution. Results differ depending on the measurement technique (edge-detection vs manual sonographic calipers), so it is imperative that a uniform technique is employed throughout. Edge-detection software has superior reproducibility to manual sonographic caliper measurements. With the exception of infant crying, there is no evidence that image quality is affected by infant behavioural and environmental variables.