Sonographic predictors of surgery in fetal coarctation of the aorta




Isolated fetal coarctation of the aorta (CoA) has high false-positive diagnostic rates by cardiologists in tertiary centers. Isthmal diameter Z-scores (I), ratio of isthmus to duct diameters (I:D), and visualization of CoA shelf (Shelf) and isthmal flow disturbance (Flow) distinguish hypoplastic from normal aortic arches in retrospective studies, but their ability to predict a need for perinatal surgery is unknown. The aim of this study was to determine whether these four sonographic features could differentiate prenatally cases which would require neonatal surgery in a prospective cohort diagnosed with CoA by a cardiologist.


From 83 referrals with cardiac disproportion (January 2006 to August 2010), we identified 37 consecutive fetuses diagnosed with CoA. Measurements of I and I:D were made and the presence of Shelf or Flow recorded. Sensitivity, specificity and areas under receiver–operating characteristics curves, using previously reported limits of I < − 2 and I:D < 0.74, as well as Shelf and Flow were compared at first and final scan. Associations between surgery and predictors were compared using multivariable logistic regression and changes in measurements using ANCOVA.


Among the 37 fetuses, 30 (81.1%) required surgery and two with an initial diagnosis of CoA were revised to normal following isthmal growth, giving an 86% diagnostic accuracy at term. The median age at first scan was 22.4 (range. 16.6–7.0) weeks and the median number of scans per fetus was three (range, one to five). I < − 2 at final scan was the most powerful predictor (odds ratio, 3.6 (95% CI, 0.47–27.3)). Shelf was identified in 66% and Flow in 50% of fetuses with CoA.


Incorporation of these four sonographic parameters in the assessment of fetuses with suspected CoA at a tertiary center resulted in better diagnostic precision regarding which cases would require neonatal surgery than has been reported previously. Copyright © 2012 ISUOG. Published by John Wiley & Sons, Ltd.


Antenatal diagnosis of isolated coarctation of the aorta (CoA), even by experts, lacks sensitivity at screening and specificity1. This leads to increased morbidity and mortality in affected neonates without a prenatal diagnosis, who often collapse and require resuscitation before surgery2. In the tertiary fetal cardiology setting there is a high false-positive rate, resulting in a significant proportion of fetuses which do not eventually require surgery being scanned throughout pregnancy and transferred and assessed at a cardiac unit following delivery.

We have previously published a method that compared the diameters of the aortic and ductal arches in a normal fetal population and derived reference ranges3, reporting that Z-scores of the aortic isthmus diameter and the isthmal to duct diameters ratio of raw measurements could be used at screening to select for further evaluation those lying below the normal range, thus potentially increasing the sensitivity of screening for CoA. We used these measurements retrospectively in a cohort of fetuses referred with suspected CoA and in whom we knew the outcome, to assess the diagnostic ability of this approach compared with subjective assessment of disproportion4, reporting that only 62.5% of those thought to have CoA by the cardiologist actually required surgery in the neonatal period.

In this prospective study we analyzed whether four sonographic parameters (‘predictors’: isthmal Z-scores (I); ratio of isthmus to duct (I:D); visualization of CoA shelf (Shelf); and visualization of continuous diastolic flow at the isthmus (Flow)) could predict at three time points in pregnancy (before 24 weeks; at first scan after 24 weeks; and at final scan) the need for postnatal surgery in fetuses thought by a fetal cardiologist to have CoA because of disproportion detected at screening in the four-chamber and/or three vessels and trachea view.


This study was performed at Queen Charlotte's and Chelsea Hospital, Imperial College London, UK and University Hospital and Masaryk University, Brno, Czech Republic. We recruited 43 fetuses which had been referred sequentially between January 2006 and August 2010, with a primary diagnosis of CoA made by a cardiologist at one of two tertiary centers. Indication for referral was suspected congenital heart disease (n = 39), family history (n = 2), increased nuchal translucency thickness (n = 1) and diabetes mellitus type 1 (n = 1). During the same time interval an additional 40 fetuses were referred with suspected CoA on obstetric ultrasound, but assessment by the fetal cardiologist showed arch measurements within the normal range3 and neither Shelf nor Flow. These were therefore not felt to have CoA and were excluded from the study, but their postnatal outcome was checked on the national Central Cardiac Audit Database ( to ensure that they did not undergo surgery for CoA during the first year after delivery. A postnatal echo was not recommended routinely but postnatal evaluation was offered in a report generated following the fetal consultation if there were any postnatal concerns.

CoA was defined as concordant atrioventricular and ventriculoarterial connection with hypoplasia of the isthmus < Z-score − 2, presence of a CoA shelf (Shelf) or presence of continuous diastolic flow at the isthmus (Flow). The diagnosis of likely CoA was made if any one of the four predictors assessed in this study was present, i.e. I or I:D below the lower limit of normal or if there was evidence of Shelf or Flow. We included fetuses with associated ventricular septal defect, bicuspid aortic valve and persistent left superior caval vein4, but not those with other cardiac abnormalities. We did not exclude fetuses with aneuploidy or extracardiac malformations, but five cases which suffered intrauterine death or termination of pregnancy without confirmed outcome at postmortem were excluded from analysis and one with diaphragmatic hernia was excluded as the disproportion was thought to be secondary to the extracardiac abnormality rather than a marker of CoA. Therefore, 37 fetuses with a fetal diagnosis of CoA were enrolled into the study.

The institutional review board considered ethical approval unnecessary because the sonographic measurements were performed as an integral part of routine clinical visits, for which informed consent had been obtained from the women.


Fetal echocardiography was performed on a GE Voluson E8 (GE Medical Systems, Zipf, Austria) ultrasound machine using a 4-8-D RAB probe. The four predictors comprised serial arch diameter measurements in the three vessels and trachea view, and presence of Shelf and Flow. Z-scores of the isthmus were calculated based on gestational age as described previously3. Sagittal views of the aortic arch were examined for Shelf (Figure 1) and both views were examined with color and power Doppler to detect Flow. No alteration from routine settings was required to detect Flow. If detected, Flow was confirmed on pulsed Doppler (Figure 2). Fetuses were examined at approximately monthly intervals and perinatal management was informed by objective data from the four predictors.

Figure 1.

Sagittal view of the aortic arch in a fetus showing a posterior shelf on two-dimensional ultrasound (a) and color flow mapping (b).

Figure 2.

Doppler flow profile from the isthmus (sagittal view), showing continuous flow throughout systole and diastole.

Outcome measures

The primary outcome measure was the requirement for CoA surgery perinatally.

Statistical analysis

All analyses were carried out using Stata version 11.1 (Stata Statistical Software; StataCorp, College Station, TX, USA). Mean and standard deviation of I, I:D and gestational age at scan were calculated for neonates who required surgery and those who did not, and differences compared using t-tests for first and last scan. Median and range were determined and Mann–Whitney U-tests were used if variables were not normally distributed. For the binary predictors (Shelf, Flow, I < − 2, I:D < 0.74)3 the percentages in each group were calculated and the chi-square test, or Fisher's exact test for cell counts < 5, was used to test for differences.

The relationship between predictors at first and last scan and outcome was assessed by calculating sensitivity, specificity and area under the receiver–operating characteristics (ROC) curves (AUC) using previously defined cut-offs3 of < − 2 for I and < 0.74 for I:D, as well as for the binary variables (Shelf and Flow).

Differences in predictive power at first and last scan were tested using McNemar's test. Differences in AUC were tested using the roccomp procedure in Stata Statistical Software version 11 (StataCorp LP, 2009, College Station, TX, USA)5. Logistic regression was used to test and compare associations between outcome and predictors at first and last scan.

We investigated whether the change in measurements made over time was associated with the need for surgery. The change in I and I:D as continuous variables from first to last scan was compared using ANCOVA, with last scan measurement as the dependent variable, need for surgery as the independent variable and adjusting for measurement at first scan as this is preferable to using change as an outcome6. For fetuses that were scanned before 24 weeks, the results of the first scan before 24 weeks and the first scan after 24 weeks were compared to determine whether diagnostic properties improved.

Reproducibility of measurements was assessed by calculating the limits of agreement7, i.e. the range in which 95% of the differences between pairs of measurements on the same subject carried out by two different observers were expected to lie. Based on 10 pairs of measurements carried out by V.J. and H.M.G., the interobserver variability showed a mean difference of 0.06 (95% limits of agreement, − 0.68 to 0.56).


Of the 37 fetuses enrolled into the study, at a median (range) gestational age of 22.4 (16.6–37.1) weeks, 30 (81%) required surgery and seven were false-positive. A median of three (range, one to five) scans was performed per fetus. Two fetuses did not have I:D measured at the first scan and one did not have Shelf recorded.

Twenty-three of the 30 (77%) requiring surgery were assessed before 24 weeks, compared with two of the seven (29%) that did not (P = 0.03). Overall, eight fetuses had persistent left superior vena cava (LSVC) (three in cases false-positive for CoA) and nine had small ventricular septal defects (VSD) (four in false-positive group; an additional one was undetected prenatally), while six of the 30 cases of true CoA had an associated bicuspid valve reported.

A comparison of measurements for I and I:D and the percentage lying below the lower limit of normal at the first scan and the last scan before delivery is shown in Table 1. All patients requiring surgery had an I:D ratio < 0.74 on the final scan. Shelf was detected in 19/29 (65.5%) fetuses requiring surgery and in 3/7 (43%) that did not (P = 0.4). Flow was seen in half of those requiring surgery (15/30, 50%) and in three of the seven (43%) that did not (P > 0.9).

Table 1. Comparison of measurements at first and last scan in those requiring surgery with those thought normal after delivery
MeasurementNo surgery (n = 7)Surgery (n = 30)P
  • Data presented as median (interquartile range), mean ± SD or % (n).

  • *

    All fetuses requiring surgery had isthmus to duct ratio < 0.74 by the last scan. N/A, not applicable.

First scan   
 Gestational age (weeks)24.3 (19.6–30)22.0 (16.6–37.1)0.14
 Isthmal Z-score− 2.67 ± 0.57− 2.78 ± 1.70.87
 Isthmal Z-score < − 286 (6/7)77 (23/30)0.6
 Isthmus to duct ratio0.57 ± 0.110.57 ± 0.180.95
 Isthmus to duct ratio < 0.74100 (6/6)90 (26/29)0.68
Last scan   
 Gestational age (weeks)34.4 (26.9–36.7)35.0 (22.4–38.4)0.91
 Isthmal Z-score− 2.89 ± 1.5− 3.44 ± 1.80.46
 Isthmal Z-score < − 271 (5/7)90 (27/30)0.20
 Isthmus to duct ratio0.57 ± 0.160.52 ± 0.120.28
 Isthmus to duct ratio < 0.7480 (6/7)100 (29/29)*N/A

Predictive power of echocardiographic parameters

Table 2 shows the predictive properties of Shelf, Flow, I and I:D. There was no significant difference in the predictive ability of I and I:D between the first and last fetal scans. Table 3 presents the odds ratios (OR) for the four predictors. The OR for all predictors except for I at first scan was > 1; however, none was significant and the confidence intervals were very wide.

Table 2. Predictive properties of sonographic features at first and last scan
PredictorSensitivity (% (95% CI))Specificity (% (95% CI))AUC (95% CI)
  1. AUC, area under the receiver–operating characteristics curve.

Presence of shelf (Shelf)   
 Any scan65.5 (45.7–82.1)57.1 (18.4–90.1)0.61 (0.40–0.83)
Presence of continuous flow (Flow)   
 Any scan50.0 (31.3–68.7)57.1 (18.4–90.1)0.54 (0.32–0.75)
Isthmal Z-score   
 First scan76.7 (57.7–90.1)14.3 (0.4–57.9)0.45 (0.3–0.61)
 Last scan90.0 (73.5–97.9)28.6 (3.7–71.0)0.59 (0.40–0.78)
 P0.29> 0.990.32
Isthmus to duct ratio   
 First scan89.7 (72.6–97.8)0 (0–45.9)0.45 (0.39–0.50)
 Last scan100 (88.1–100)14.3 (0.4–57.9)0.57 (0.43–0.71)
 P0.25> 0.990.5
Table 3. Association between need for surgery and the four predictors at first and last scans
PredictorUnivariable regression: OR (95% CI)P
  • *

    All fetuses requiring surgery had isthmus to duct ratio < 0.74 by the last scan. N/A, not applicable; OR, odds ratio.

Shelf2.53 (0.47–13.6)0.28
Flow1.33 (0.25–7.01)0.73
Isthmal Z-score < − 2  
 First scan0.55 (0.56–5.35)0.61
 Last scan3.6 (0.47–27.3)0.22
Isthmus to duct ratio < 0.74  
 First scanN/A*0.12
 Last scanN/A* 

Predictive power of serial measurements

While there was an improvement in sensitivity of I:D on serial scans, with 100% sensitivity on the final scan, ANCOVA analysis showed that there was no difference in the change in I and I:D between fetuses who required surgery and those who did not (difference in I at last scan, adjusting for first scan = − 0.50 (95% CI, − 1.9 to 0.9), P = 0.48; difference in I:D at last scan, adjusting for first scan = − 0.02 (95% CI, − 0.14 to 0.10), P = 0.75).

Predictive accuracy of combinations of factors

The incremental value of combinations of factors was examined (Table 4). Alone, Shelf had the highest AUC (0.61). By combining this with I there was a modest increase in AUC, to 0.64 (95% CI, 0.44–0.84).

Table 4. Predictive accuracy of combinations of echocardiographic signs at first scan
Combination of predictorsSensitivity (% (95% CI))Specificity % (95% CI))AUC (95% CI)
  1. AUC, area under receiver–operating characteristics curve; Flow, presence of continuous flow; I, isthmal Z-score; I:D, isthmus to duct ratio; Shelf, presence of shelf.

Two signs   
 Shelf + Flow36.7 (19.9–56.1)57.1 (18.4–90.1)0.47 (0.25–0.69)
 Shelf + I56.7 (37.4–74.5)71.4 (6.3–29.0)0.64 (0.44–0.84)
 Shelf + I:D60.0 (40.6–77.3)57.1 (18.4–90.1)0.59 (0.37–0.80)
 Flow + I50.0 (31.3–68.7)71.4 (0.3–29.0)0.61 (0.40–0.81)
 Flow + I:D43.3 (25.5–62.6)57.1 (18.4–90.1)0.50 (0.28–0.72)
 I + I:D70.0 (50.6–85.3)28.6 (3.7–71.0)0.49 (0.29–0.69)
Three signs   
 Shelf + Flow + I36.7 (19.9–56.1)71.4 (29.0–96.3)0.54 (0.34–0.74)
 Shelf + Flow + I:D33.3 (17.3–52.8)57.1 (18.4–90.1)0.45 (0.24–0.67)
 Shelf + I + I:D53.3 (34.3–71.7)71.4 (29.0–96.3)0.62 (0.42–0.83)
 Flow + I + I:D43.3 (25.5–62.6)71.4 (29–96.3)0.57 (0.37–0.78)
Four signs   
 Shelf + Flow + I + I:D33.3 (17.3–52.8)71.4 (29–96.3)0.52 (0.32–0.72)

Predictive accuracy before 24 weeks

Of the 25 fetuses scanned before 24 weeks, only two (8%) did not require surgery, so estimates of specificity and ROC were not reliable. Although the sensitivity of I and I:D increased between the first and last scans (I from 73.9% (95% CI, 51.6–89.8) to 81.8% (95% CI, 59.7–94.8) and I:D from 86.4% (95% CI, 65.1–97.1) to 95.2% (95% CI, 76.2–99.9)), these increases were not significant (P = 0.73 for I, P = 0.63 for I:D).

Presence of abnormal karyotype and extracardiac abnormalities

Fourteen of the 37 (37.8%) women accepted prenatal karyotype testing. All results were normal. Among those declining prenatal testing was one case of trisomy 13 (with CoA) and one case of trisomy 21 (without CoA, with VSD) diagnosed postnatally. All children undergoing surgery had normal karyotype.

Seven of the 37 (18.9%) fetuses had additional sonographic findings detected prenatally, including increased nuchal translucency, single umbilical artery, two-vessel cord and cerebellar hypoplasia or cerebral ventriculomegaly. Three chromosomally normal fetuses had extracardiac abnormalities confirmed postnatally: one with duodenal atresia, one with tracheo-esophageal fistula and one with multiple abnormalities including scoliosis and coloboma. Two fetuses had brain abnormalities confirmed on fetal magnetic resonance imaging, associated with trisomy 21 and trisomy 13.

Characteristics of those not requiring surgery

Table 5 details the characteristics of the seven fetuses that did not require surgery postnatally. Six fetuses had a tapering or hypoplastic arch and one did not. Of these six, two showed improved growth on serial scans, reaching an isthmal Z-score within the normal range by term and so were thought unlikely to have CoA. Four had bilateral superior caval veins and five had a VSD identified prenatally (four of which were confirmed postnatally). Two fetuses had trisomies (21 and 13). The only fetus without an additional finding on prenatal ultrasound was found to have a small ventricular septal defect (not seen on fetal echo) after birth. The postnatal echo showed a normal arch in four babies, two had arch hypoplasia and one had an area of isthmal narrowing without CoA. All children were reviewed in a cardiac clinic until they were 1 year old.

Table 5. Characteristics of the seven fetuses with unobstructed aortic arch postnatally
FetusPrenatal echo: arch morphologyShelfDiastolic flowAdditional prenatal featuresIsthmal Z-score (first scan; last scan)Postnatal features
  • *

    Z-scores measured between these time points were more negative. AoV, aortic valve; CoA, coarctation of the aorta; SVC, superior vena cava; VSD, ventricular septal defect.

1No arch hypoplasiaNoNoLeft SVC, bicuspid AoV− 2.5; − 1.4Normal arch, left SVC to coronary sinus
2Tapering archYesYesLeft SVC, multiple VSD, extracardiac abnormalities− 3.5; − 2.8Normal arch, left SVC to coronary sinus, multiple VSD, dysmorphic features
3Tapering arch on early scans, improved growth on later scansYesYesLeft SVC, VSD− 3.26; − 1.72Mild narrowing at isthmus but no CoA, left SVC to coronary sinus, perimembranous outlet VSD
4Tapering arch on later scansYesYesVSD− 1.98; − 5.98Mild arch hypoplasia without CoA, apical muscular VSD, trisomy 21
5No tapering and improved growth of archNoNoVSD (on early scan; not seen on later scan or postnatally)− 2.2; − 2.2*Normal arch, no VSD
6Arch hypoplasiaNoNoLeft SVC, VSD− 2.9; − 3.4Normal arch, malaligned VSD, trisomy 13
7Arch hypoplasiaNoNoNone− 2.3; − 2.7Hypoplastic arch without CoA, VSD, premature delivery at 28 weeks


Making a prenatal diagnosis of CoA requires the combination of high sensitivity at screening (due to the serious consequences of not detecting this lesion before birth8) and high specificity following referral to a tertiary center. Currently, neither of these is satisfactory, with less than one third of cases of CoA being detected at screening2 and a high rate of false-positive diagnosis (low specificity) by fetal cardiologists. False-positive diagnosis can cause unnecessary parental anxiety, alteration of perinatal plans that are disruptive to the family and unnecessary investigations and hospital stay for the baby in a tertiary cardiac center1, incurring costs.

In this prospective study we have demonstrated that the use of four parameters (I, I:D, Shelf and Flow) in combination achieves superior diagnostic precision (86%) in the detection of true fetal CoA requiring perinatal surgery compared with our previously published clinical estimate of 62.5%4. Shelf, Flow and I had ORs > 1 at the final scan and all patients with CoA postnatally had an I:D ratio below the lower limit of normal. Application of these sonographic criteria by the fetal cardiologist during serial review may increase diagnostic specificity and improve clinical management. Furthermore, we confirmed that we could safely exclude from further review approximately half of the original prenatal referrals with suspected CoA using these sonographic parameters as none had CoA diagnosed during infancy. Statistical analysis could not define a single ‘best predictor’ of CoA; however, measurement of the isthmus and duct allow comparison of serial Z-scores (measurements corrected for gestation or size) and in two fetuses we correctly revised our opinion to their being at low risk of CoA because of improvement of isthmal measurements by term.

Previous studies have used sagittal or coronal views of the aortic arch and focused on measurements of the arch9, 10 or on the ratio of transverse arch diameter to diameter of head and neck vessels11. The diagnostic accuracy in our study (86%) was higher than the 62%1 and 30%12 reported previously. We preferred the three vessels and trachea view in our examination protocol because it is the ideal plane in which to compare aortic and ductal arches and measure I and I:D (Figures 3a and 4a). However, Shelf and Flow are best identified in the sagittal view of the aortic arch (Figures 3b and 4b) and in this study in two-thirds of cases of true CoA we observed Shelf and in 50% we observed Flow on at least one examination.

Figure 3.

Appearance in a normal fetus: (a) three vessels and trachea view, with aorta and duct passing to the left of the trachea and a single right superior caval vein (SVC); (b) sagittal view of aortic arch.

Figure 4.

Appearance in a fetus with coarctation of the aorta: (a) three vessels and trachea view, showing marked disproportion of the great arteries; (b) two-dimensional ultrasound and power Doppler sagittal view of aortic arch, showing hypoplasia of the transverse arch with a posterior shelf at the isthmus (*). SVC, superior vena cava. Calipers in (a) show measurement of aortic isthmus and ductal arch diameters.

The high level of false-positive diagnoses reported in most series and the finding of Shelf in only two-thirds of our current series of true CoA cases is not surprising as tubular hypoplasia is the most common morphology in neonatal CoA, although a shelf may co-exist. Tubular hypoplasia may also occur without obstruction4, 12. We did not recognize Shelf in one case with CoA that had an aneurysmal duct covering the isthmus in all sonographic planes. There seems to be a learning curve in the examination of the fetal arch, as in our previous report we recognized a smaller proportion (45%) of fetuses with true CoA to have Shelf4. Conversely, identification of Shelf in three of our cases that did not require surgery is interesting. One was diagnosed with trisomy 21 postnatally (an unusual association with CoA) (Figure 1). This fetus had an apical muscular VSD on postnatal scan and isthmal narrowing without CoA; the second had Shelf described on the initial scan, followed by improvement in measurements to within the normal range as the pregnancy progressed, with Shelf not visible in later scans; the third had multiple abnormalities, including abnormalities of the thoracic spine compatible with CHARGE syndrome.

Pitfalls in the correct diagnosis of fetal CoA in previous studies included fetuses having additional sonographic findings, such as VSD and persistent LSVC that may be a cause of ventricular disproportion particularly in the early fetus because dilatation of the coronary sinus impairs left ventricular filling13, 14. In this study, persistent LSVC was seen as commonly in those with as in those without CoA, as we have reported previously4. Bicuspid aortic valve was diagnosed more frequently in true CoA; however, difficulty in confirmation may make this well-recognized association less helpful prenatally.

The rate of karyotypic abnormalities in this study was lower than that in other published reports15, likely due to the majority of fetuses being referred following first-trimester screening; chromosomally abnormal pregnancies may already have been terminated.

While statistical analysis of serial arch measurements did not show improved diagnostic accuracy, two fetuses had good aortic arch growth and our counseling was altered to reflect this; neither required postnatal surgery, giving an overall diagnostic accuracy of 86%. This is in accord with our previous retrospective study in which those not requiring surgery tended to show improved growth of the aortic arch by term4. We arrange serial follow-up for any fetus found to have any one of the four sonographic parameters outside the normal range and determine perinatal management (place of delivery and postnatal assessment) if abnormality of any of the four predictors persists. If the parameters normalize during follow-up we arrange local delivery and postnatal assessment, provided that there is appropriate local expertise to assess the neonate safely until the duct has closed.


The accuracy of measurements must be interpreted with caution when arch views are suboptimal, particularly in the very small fetus or when the arch is very narrow, although in the latter case CoA is more likely. Use of the I:D may be misleading in late gestation due to ductal dilatation and tortuosity. Augmented diastolic isthmal flow can be identified using standard machine settings, and recognition that it might be present is important. Ultrasound manufacturers commit to a measurement accuracy of 1 mm for small structures (GE Healthcare, pers. comm.), so accuracy may be reduced in early pregnancy. The lack of statistical confirmation of what we believe to be a true clinical scenario is likely due to the limited size of this study; there was limited power to examine the change in predictive power with gestational age at scan due to the small number of babies and the variation in the number and timing of scans.


We propose incorporation into the prenatal assessment of fetal CoA in the tertiary setting the following four sonographic parameters: I:D < 0.74, I < − 2 and presence of Shelf and/or Flow. In our study this combination allowed us to predict CoA correctly in 81% of cases at first assessment and serial measurements enabled us to revise our assessment in two cases, resulting in an accuracy of 86% overall.


This work was supported by the Richard and Jack Wiseman Trust, Tiny Tickers charity (, the NIHR Biomedical Research Centre funding scheme and the Genesis Research Trust, Institute of Reproductive Biology, Queen Charlotte's and Chelsea Hospital, London, UK.