This was a single-center retrospective cohort study using the electronic medical records of Peking University First Hospital, Beijing, China, and was approved by the ethics committee of Peking University First Hospital (ID: 2020-411). The hospital is a tertiary referral center with approximately 6000 deliveries per year. PAS patients included those transferred from Northern China, primarily Beijing, Tianjin, and Hebei provinces. The inclusion criteria included singleton delivery, live birth, and previous cesarean section with antenatal PAS diagnosis from November 2016 to October 2021.

All patients received routine ultrasound examination during pregnancy, at 11–13 weeks, 21–24 weeks, 28–32 weeks, 34–36 weeks, and 40 weeks (if applicable). Both transvaginal and transabdominal ultrasound were performed by experienced ultrasound doctors (senior attending) after systematic training, following the FIGO recommendations on ultrasonic features [17]: placenta lacunae with feeder vessels, loss of hypoechoic space, abnormalities of the uterus-bladder interface, and color Doppler abnormalities, including hypervascularity and bridging vessels.

If PAS was suspected, routine extra ultrasound examinations were performed every 1–4 weeks. For patients who were first suspected of PAS at other hospitals and were referred, further diagnosis was performed within 1 week, for a timely and comprehensive assessment. To identify the gestational age of the first detection of PAS by ultrasound examination, we screened all the ultrasound images and reports within (n = 98) or outside (n = 68) our hospital for each patient throughout the pregnancy.

Details were extracted from the surgical records regarding the position of the placenta and the degree and extent of invasion. All patients were diagnosed according to the intraoperative FIGO classification. Prenatal care and subsequent operations were performed by the same group of senior doctors specialized in PAS disorders. When the recommended gestational age of 34–36 weeks could not be reached [18], gestational age at delivery was determined according to the severity and progression of ultrasonic features. A multi-disciplinary surgery team was assembled at the planned delivery, including experts from the Obstetrics and Gynecology Department, Anesthesia Department, and Neonatal Department. Urologists and general surgeons would participate if necessary. Patients showing hypervascularity during color Doppler imaging would be given a pre-operative intra-aortic balloon placement (IABO) to mitigate major hemorrhage [19]. Hysterectomy was decided either preoperatively or during laparotomy.

The primary independent variable was the detection week of PAS, as determined by ultrasound. As a continuous variable, the timing was also grouped in relation to time points of prenatal examinations, first into 2 categories (28 weeks and $\ge$28 weeks [before and after the start of the third trimester]), and then into 6 categories (20, [20–24], [24–28], [28–32], [32–36], $\ge$36 weeks), for clearer description and presentation.

The primary outcomes were intra-operative blood loss and red-blood-cell (RBC)-transfusion volume. Intraoperative hemorrhage was estimated by suction and weighing of swabs. It was recoded as a binary variable (blood loss $\ge$1500 mL or not), and the RBC-transfusion volume was recoded as well ($\ge$800 mL [or 4 units] or not) [20].

The following variables were also extracted: socio-demographic characteristics, gestational complications (gestational diabetes mellitus and hypertension), pregnancy history and characteristics (cesarean section and abortion history, use of assisted reproductive technology, placenta previa, etc.), clinical manifestations (antepartum hemorrhage, abdominal pain, placenta position [anterior or not]), PAS severity (accreta, increta, percreta), FIGO clinical grading of PAS, use of operational procedure (abdominal aorta balloon, tourniquet, uterine artery ligation, uterine cavity tamponade, and hysterectomy), maternal perinatal outcomes (gestational weeks at delivery, length of stay, maternal complications [which included post-partum bleeding, amount of blood transfusion, acute transfusion reaction, disseminated intravascular coagulation (DIC), hemorrhagic shock, infection, anemia, other organ damage]). In our study, we defined placenta previa (partial and complete placenta previa) according to the latest guideline [21]. Fetal outcomes, including neonatal complications (dyspnea, pneumonia, infection), were based on the pediatric records. No maternal or fetal deaths occurred in the study.

For descriptive analysis, categorical data were presented as frequency (percentage). Median (interquartile range, IQR) was reported for continuous variables. Kruskal-Wallis tests and Chi-square tests were used for group comparisons.

To illustrate the association between ultrasound-detection timing and the outcomes, we plotted box charts of blood loss and RBC-transfusion volume over the 6 categories of ultrasound-detection week (20 weeks, 20–24 weeks, 24–28 weeks, 28–32 weeks, 32–36 weeks, $\ge$36 weeks). We also provided 2 additional box plots showing the results of subgroup analysis by gestation week at delivery ($\ge$34 weeks or not, because 34 weeks was recommended as the earliest window for PAS-complicated pregnancy delivery [18]).

For statistical inference, we applied univariate and multivariate Poisson regression models with robust error variance to examine the associations between the gestational week at ultrasound detection (categorized into 2 and 6 groups) and the outcomes of major intraoperative hemorrhage ($\ge$1500 mL) and RBC-transfusion volume $\ge$800 mL (or $\ge$4 units). Due to the relatively small sample size, we designed 1 unadjusted model and 3 predefined adjusted models according to clinical considerations and according to a previous study [22]: Model A: adjusted for PAS severity (invasive PAS or not) and hysterectomy; Model B: additionally adjusted for IABO and gestation age (34 weeks, 34 weeks or $>$34 weeks); and Model C: further adding placenta previa, gestational diabetes mellitus, gestational hypertension, anterior placenta position and history of cesarean delivery. The variance inflation factor (VIF) was used to check for collinearity in the models.

To understand the influence of gestational age at delivery on maternal outcomes, we stratified the sample according to PAS-detection-week categories (28 weeks or $\ge$28 weeks), and used univariate and multivariate robust-error-variance Poisson regression models to examine the associations between the gestational age at delivery (in 6 categories: $\le$32 weeks, 33 weeks, 34 weeks, 35 weeks, 36 weeks and $>$36 weeks) (as independent variables) and intra-operative blood loss and RBC-transfusion volume (as continuous dependent variables). We plotted the predicted blood loss or RBC-transfusion volume (95% CI) for each gestational week at the delivery category. We also conducted a series of sensitivity analyses, in which we restricted our sample to those with blood-loss volume less 5000 mL (n = 164) to minimize the influence of outliers, and the sample (n = 98) in which transferred patients were excluded.

A two-tailed p $\le$ 0.05 was considered statistically significant. The statistical analyses were conducted using the Stata software (v15.0; StataCorp, College Station, TX, USA).

For the 166 included PAS patients, the median age was 34 (IQR: 31–37), and 48.8% were of advanced maternal age ($\ge$35 year). The median ultrasound-detection week was 30 (IQR: 24–34); 39.2% of the patients were diagnosed before 28 weeks of gestation, and 77.1% reached 34 weeks at delivery (median: 35 weeks). More than half of the patients were diagnosed with invasive PAS, whereas 17.5% were classified as having placenta accreta. Patients’ gestational age at delivery, pregnancy history (especially gravidity and history and frequency of abortion and cesarean section), and complicated placenta previa were significantly distinct in the groups, based on PAS-detection week, either for the 2 or the 6 groups (Table 1 and Supplementary Table 1). Patients with a PAS diagnosis before 28 weeks of gestation (84.6%) had both anterior placentation and a history of cesarean section, compared to less than 60% in the late-diagnosis group (Table 1).

The median intra-operative blood-loss volume was 1000 mL (IQR: 600–1800 mL), and 62.7% of the patients had blood-product transfusion. The median volume of RBC transfusion was 238 mL (IQR: 0–1000 mL) and 14.5% had more than 8 units (1600 mL). Patients with PAS ultrasound detection before 28 gestational weeks had higher intra-operative blood-loss volume (median: 1500 mL vs. 800 mL) and RBC-transfusion volume (median: 586 mL vs. 70 mL) (Table 2 and Fig. 1A–C). Severe maternal morbidity occurred very rarely in the sample; the incidence of reoperation, organ injury, DIC, and infection were not different in the various PAS ultrasound-detection week groups (Table 2 and Supplementary Table 2).

Fig. 1.

The trend of intraoperative blood loss and RBC-transfer volume over the 6 ultrasound-detection-week categories. (A) The overall trend showed much lower intraoperative blood loss and RBC transfer volume for patients with ultrasound-detection week over 32. (B,C) After stratification by gestational-age categories, we found that in the delivery $\ge$34 weeks group, as first detection age increased, there was less intraoperative bleeding and blood transfusion.

Patients with ultrasound detection week 28 were significantly more likely to have major blood loss (unadjusted risk ratio [RR]: 2.05; 95% CI: 1.35–3.11); this significant association remained after adjusting for only PAS severity and hysterectomy (adjusted RR: 1.63; 95% CI: 1.09–2.44). After adjusting for more covariates (adjusted models B and C), patients with ultrasound detection week 28 were no longer significantly different. Compared with patients with ultrasound detection week of 32–36, those with PAS detected in 28–32 weeks (aRR: 2.88; 95% CI: 1.19–6.94), 24–28 weeks (aRR: 3.29; 95% CI: 1.35–8.05) and 20 weeks (aRR: 3.04; 95% CI: 1.22–7.55) were associated with a significantly higher risk of intraoperative blood loss of $>$1500 mL (Adjusted model A, Table 3). The association between ultrasound-detection week and RBC-transfusion outcome showed similar results (Table 3). We found no evidence of problematic multicollinearity (all VIFs 10).

For patients in the early detection group (ultrasound-detection week 28), the predicted estimated blood-loss volume increased along with the gestational week of delivery, whereas in the late detection group, there was a decline in predicted estimated blood loss volume as gestational week of delivery increased. The overall estimated blood loss volume was higher in the early detection group (Fig. 2A,B).

Fig. 2.

The predicted blood loss (A) or RBC-transfusion volume (B) (95% confidence interval [CI]) for each gestational week at delivery category ($\mathrm{\le }$32 week, 33 week, 34 week, 35 week, 36 week, and $\mathrm{>}$36 week) (adjusted for invasive PAS and hysterectomy) (n = 166). The red lines represent PAS patients with ultrasound detection week $\ge$28, the blue lines are for PAS patients with PAS-detection week 28. For both blood-loss volume and RBC-transfusion volume, the early detection group generally had worse hemorrhagic outcomes. We observed that for the early detection PAS patients, later delivery gestation week meant higher blood loss and transfusion volume, whereas the trend was reversed in the late detection PAS patients.

When we restricted our sample to those with blood-loss volume less than 5000 mL (n = 164) to minimize the influence of outliers, and the sample (n = 98) in which transferred patients were excluded, the trends were preserved (Supplementary Table 3, Supplementary Figs. 1,2).

The present study had several limitations. As a retrospective study based on routine pregnancy ultrasound screening, the frequency of ultrasound and the gestational age at delivery varied within a 4-week range, which might have confounded the results. For those referred PAS patients from other hospitals at the late third trimester, the ultrasound features before referral could be neglected, even though every patient was re-assessed in our center. We also did not include images and diagnoses by other diagnostic modalities, such as magnetic resonance imaging (MRI). Since we only included patients with prenatal diagnosis of PAS, other missing cases could affect the generalizability of the study. We only reported on patients with a singleton live birth, and mainly focused on their postpartum hemorrhage as the outcome criterion. Multicenter design to boost sample size and adopt a prospective study design should be considered with a further standardized ultrasound information collection process.