Assessment of ultrasound features of placenta accreta spectrum in women at high risk: association with outcome and interobserver concordance

To evaluate the prenatal ultrasound features associated with operative complications and to assess the interobserver agreement of prenatal ultrasound assessment with histopathologic confirmation of placenta accreta spectrum (PAS) in a cohort of high‐risk patients with detailed intraoperative and histopathologic data.

operators blinded to clinical details, intraoperative features, outcome and histopathologic findings. The diagnosis of PAS was confirmed by failure of detachment of one or more placental cotyledons from the uterine wall at delivery, and the absence of decidua with distortion of the uteroplacental interface by fibrinoid deposition on histologic examination of the accretic areas obtained by guided sampling of partial myometrial resection or hysterectomy specimens. Patients were categorized as having a low or high likelihood of PAS at birth. Interobserver agreement of prenatal ultrasound assessment with histopathologic confirmation of PAS was assessed using the kappa statistic. Primary outcome was major operative morbidity (blood loss ≥ 2000 mL, unintentional injury to the viscera, admission to intensive care unit or death).

INTRODUCTION
Ultrasound is the primary tool for screening patients at high risk of placenta accreta spectrum (PAS) disorders at birth 1,2 , and numerous ultrasound features have been described 3 . A recent Delphi survey reached robust agreement on seven of the 11 classic standardized transabdominal sonographic signs associated with PAS at birth 4 . The prevalence of PAS is increasing, following the worldwide rise in Cesarean delivery (CD) rates over the last two decades 5,6 . The risk of PAS is highest in patients presenting with an anterior low-lying placenta/placenta previa and one or more prior CD 7 . However, not all patients with these risk factors develop PAS.
Prenatal identification of patients at high risk of PAS decreases the risk of maternal complications at delivery, and in particular the risk of hemorrhagic morbidity 8 . The reported diagnostic accuracy of ultrasound for the prenatal detection of PAS is excellent in some studies 3,9 . By contrast, Bowman and coworkers 10,11 reported that the agreement between several observers for the diagnosis of PAS is only moderate and the diagnostic accuracy may not be as good as reported previously.
The formal diagnosis of PAS is based on clinicohistopathologic findings at birth 12 . However, most authors of cohort series on PAS do not provide information on the intraoperative features and the description of histopathologic findings is limited 13 . The aims of this study were first to evaluate the prenatal ultrasound signs associated with operative complications, and second, to assess interobserver agreement of prenatal ultrasound assessment with histopathologic confirmation of PAS in a cohort of high-risk patients for whom detailed intraoperative and histopathologic data were available.

METHODS
We conducted a retrospective multicenter cohort study of patients at high risk of PAS referred for perinatal care and management by a specialist multidisciplinary team at Cairo University Hospital, Cairo, Egypt, University College Hospital, London, UK or St George's Hospital, London, UK between January 2019 and May 2022. All patients presented with a singleton pregnancy between 32-37 weeks of gestation, had a history of one or more prior CD and were diagnosed prenatally with an anterior low-lying placenta/placenta previa. Multiple pregnancy and pregnancy requiring emergency delivery before 32 weeks were excluded. Patients were managed according to their local unit protocols and were counseled regarding surgical management (hysterectomy or one-stage resection and repair), which was dependent on ultrasound-based preoperative surgical planning and feasibility mandated by intraoperative findings with the primary aim of reducing maternal morbidity and mortality. Preoperative interventional radiology was not used in any case.
All patients had detailed preoperative transabdominal and transvaginal sonographic (TVS) examinations of the placenta, uterus and pelvis within 1 week before the planned delivery date. The placenta was recorded as low-lying when the edge was 0.5-2 cm from the internal os of the uterine cervix on TVS. It was defined as placenta previa (marginal or complete) when the placenta was < 0.5 cm from the internal os or completely covering it 14 .
Transabdominal ultrasound signs suggestive of PAS were recorded using a standardized reporting proforma protocol 15 . In addition, we used the score for placental lacunae, proposed by Finberg and Williams 16 (0, none; 1+, 1-3; 2+, 4-6; 3+, > 6). In all cases ultrasound and intraoperative images were archived electronically. The diagnosis of PAS was confirmed by the abnormal attachment of one or more placental cotyledons to the uterine wall at delivery and the absence of decidua, with distortion of the uteroplacental interface by fibrinoid deposition on histologic examination of the accreta areas obtained by guided sampling as described previously 17 . Pregnancy and delivery data were collected from hospital records.
Deidentified ultrasound images (and videoclips when available) were reviewed retrospectively and independently by two experienced operators (A.B. and E.J.). Both were blinded to clinical background, surgical findings and histopathologic diagnoses. They were unaware of the proportion of PAS vs non-PAS in the study cohort and the images were reviewed in a random sequence. The examiners were required to judge the likelihood of PAS as being low or high, and antenatal assessment was correlated with major operative morbidity. Major morbidity was defined as occurrence of any one of the following: major obstetric hemorrhage, defined as blood loss of ≥ 2000 mL; unintentional injury to the urinary bladder or other pelvic viscera; admission to the intensive care unit; or maternal death. A composite of several types of morbidity constituted the presence of morbidity. An individual may have multiple morbidities, which would be underestimated by a binary definition of presence or absence of morbidity. Therefore, we derived a morbidity count (expressed as median (interquartile range)), in which each type of morbidity was recorded. Interobserver agreement was quantified using the kappa statistic. In a study including the cohort of patients from Cairo University Hospital (n = 90) 18 , we have previously shown that the features associated consistently with the need for a hysterectomy are: the presence of increased vascularity, lacunae, bridging vessels and feeder vessels. As such, we were particularly interested in the degree of agreement of these four features.
Local research ethics approval was obtained in Cairo, Egypt from the Institutional Scientific and Research Ethical Committee (RSEC 021001). In the UK, the NHS Health Research Authority deemed the study as health surveillance and so formal review by the research ethics committee was not necessary, and retrospective patient consent was not required as basic clinical data were collected using a standard clinical audit protocol and all data were fully anonymized before being submitted for central analysis.
Statistical analysis was carried out using SPSS version 28.0.1.1 (IBM Corp, Armonk, NY, USA). Data are expressed as mean ± SD or median (interquartile range) for normally and non-normally distributed data, respectively. Unpaired t-test or Mann-Whitney U-test was used to compare continuous data as appropriate. Proportions are expressed as percentages and the χ 2 test or Fisher's exact test was used for comparisons. P < 0.05 was considered statistically significant. Agreement between the two observers was quantified by calculating the kappa statistic. Kappa statistics and percentage agreement are reported according to Landis and Koch 19 , and kappa values of < 0.20, 0.21-0.40, 0.41-0.60 and 0.61-0.80 were interpreted as poor, fair, moderate and good, respectively, whereas values between 0.81 and 1.00 were interpreted as excellent agreement and negative values denote disagreement or worse-than-expected agreement.

Demographics and outcome
A total of 102 women at high risk of PAS were referred for specialist perinatal care and management. There were 66 cases of PAS and 36 cases with no evidence of PAS at birth. The maternal demographics and outcomes according to diagnosis of PAS at birth are displayed in Table 1. All women included in the cohort presented with an anterior low-lying placenta/placenta previa and had at least one prior CD (98 of 102 women had undergone two or more prior CDs). The placenta covered the internal os in 75 (73.5%) cases.
The frequency of Cesarean hysterectomy was significantly higher (P < 0.001) in the subgroup with confirmed PAS at birth compared to those with no evidence of PAS at delivery (84.8% vs 27.8%, respectively). In the former subgroup, hysterectomy was not performed in 10 cases because the accreta area was small and could be removed by partial myometrial resection, leaving sufficient healthy myometrium to reapproximate the hysterotomy edges and repair the lower uterine segment. In the non-PAS subgroup, hysterectomy was performed because of the inability to repair a large anterior wall dehiscence and/or excessive intraoperative bleeding. Data on operative morbidity are presented in Table 1.

Concordance in ultrasound prediction of PAS at birth
Both examiners agreed on the probability, based on ultrasound features, of PAS at birth (low or high) in 75 cases (73.5%) ( Table 2). The assessment was discordant in the remaining 27. The kappa statistic was 0.47, which     (Table 2).

Main findings
This study indicates that the concordance of two independent assessors regarding the high probability of PAS being present at birth is moderate when they are blinded to other clinical details. We have shown that the correlation with histopathology is strongest when the two observers agree on there being a high probability of PAS at birth. In this subgroup, the morbidity was high and the individual patient was significantly more likely to have more than one type of morbidity. Placental lacunae and increased subplacental vascularity were the most reproducible ultrasound signs with the least interobserver variability.

Comparison with previous studies
Data on interobserver evaluation of ultrasound signs in the prenatal evaluation of PAS are scarce and have yielded variable results. Bowman et al. 11 reported an overall moderate agreement (kappa, 0.47 ± 0.12) between six observers for the diagnosis of PAS, but they did not evaluate individual ultrasound signs. In contrast, Zosmer et al. 20 found good-to-excellent interobserver agreement for loss of clear zone and myometrial thinning on grayscale imaging, the presence of lacunar feeder vessels on two-dimensional color Doppler imaging and crossing vessels and lacunae on three-dimensional color Doppler imaging. Although it is uncertain if the observers were blinded to the ultimate diagnosis, there were no cases without PAS and loss of clear zone was seen in every case. Kliewer et al. 21 recently reported kappa values of 0.77 for intraplacental vascular cluster and 0.59 for traversing vessel in a cohort of 32 patients at risk of PAS. Most cohort studies do not describe the intraoperative and histopathologic findings at birth 13 , and it is therefore difficult to assess the accuracy of the diagnosis of PAS. In the present study, using guided sampling of accreta areas from hysterectomy or partial myometrial resection specimens for microscopic examination to allow accurate diagnosis of PAS at birth, we found moderate interobserver agreement for placental lacunae and increased subplacental vascularity and poor agreement for bridging vessels and feeder vessels on prenatal ultrasound examination.

Clinical implications
There are only limited data on the role of ultrasound imaging in the preoperative evaluation of surgical risk and outcome in patients with a high risk of PAS at birth. Abnormalities of the uterine contour, including the loss of clear zone, myometrial thinning and a bulge-like appearance on ultrasound, have been reported commonly in the literature as essential signs for the prenatal diagnosis of PAS 2,4,15 . Unlike anomalies of the uteroplacental circulation, ultrasound anomalies of the uterine contour are found with a similar incidence in non-PAS and PAS cases 18 . The lowest kappa values in the present study were found for these anomalies, suggesting they are secondary to a large area of dehiscence of the lower uterine segment rather than to accreta placentation 22 . This also suggests that their use in the prenatal evaluation of PAS may lead to a false-positive diagnosis in the absence of anomalies of the uteroplacental circulation.
Overall surgical morbidity is related to operator experience in the management of complex cases. All three centres in this study are regional referral centers of excellence with multidisciplinary teams with over 10 years of experience and the corresponding outcomes are linked directly to the complexity of cases referred to these centers. The incidence of PAS in cases with placenta previa and previous CD is, respectively, 5.8% and 12.7% in prospective and retrospective series 23 . A recent prospective screening study found an incidence of 28% 24 . In the current study, the incidence of PAS was 64.7% and the cohort included only patients at high risk of PAS at birth, and therefore at risk of complex surgical procedures.
Histopathology has always been considered the gold standard for the confirmation of PAS. However, by definition, histopathology reporting is retrospective and does not assist in preoperative planning in high-risk cases, i.e. those with placenta previa and previous CD. As expected, the highest morbidity was seen with a concordant ultrasound diagnosis of PAS. Two out of three cases were associated with morbidity and the likelihood of morbidity was higher with a concordant ultrasound diagnosis of PAS, and significantly more likely to involve more than one type of morbidity. The likelihood of surgical morbidity was still reasonably high even when the prenatal ultrasound assessment or histopathology did not indicate PAS. This is likely to be related to the risk factors of PAS i.e. anterior low-lying placenta/placenta previa and previous CD with associated uterine wall dehiscence.

Strengths and limitations
This study has several strengths. To our knowledge this is the first study to assess interobserver agreement of ultrasound signs in patients at high risk of PAS at birth for both histopathologic confirmation of PAS and risk of intraoperative complications. Data were collected according to a well-defined protocol for reporting on the classic ultrasound signs associated with PAS and included a detailed description of intraoperative findings. Using guided sampling of the accreta areas for microscopic examination allowed us to differentiate accurately between cases with and without PAS.
The primary limitation of this cohort study lies in its retrospective design. In addition, the observers were blinded to the patient's clinical background. While this approach eliminates bias on ultrasound assessment, it does not reflect clinical reality as one would not disregard background clinical information, such as previous CD, and assess solely on the basis of ultrasound imaging. Cases delivered in an emergency were not included. Emergency cases are not managed by the same team and, in many cases, ultrasound assessment did not take place at one of the three participating centers and it was not possible to obtain the ultrasound images.

Conclusion
The probability of histopathologic confirmation of PAS is very high with concordant prenatal assessment suggestive of PAS. The interoperator agreement for preoperative ultrasound assessment with histopathologic confirmation of PAS is only moderate. Histopathologic diagnosis of PAS is also made in one in four cases with a concordant diagnosis of low probability of PAS. Morbidity is associated with both histopathologic diagnosis (not useful in clinical management as it is a retrospective finding) and antenatal evaluation concordant with PAS. However, the concordance of ultrasound features and histopathology is not absolute. Future prospective studies should use those ultrasound signs with the highest kappa values.