Objective
Ultrasound has been reported to be greater than 90% sensitive for the diagnosis of accreta. Prior studies may be subject to bias because of single expert observers, suspicion for accreta, and knowledge of risk factors. We aimed to assess the accuracy of ultrasound for the prediction of accreta.
Study Design
Patients with accreta at a single academic center were matched to patients with placenta previa, but no accreta, by year of delivery. Ultrasound studies with views of the placenta were collected, deidentified, blinded to clinical history, and placed in random sequence. Six investigators prospectively interpreted each study for the presence of accreta and findings reported to be associated with its diagnosis. Sensitivity, specificity, positive predictive, negative predictive value, and accuracy were calculated. Characteristics of accurate findings were compared using univariate and multivariate analyses.
Results
Six investigators examined 229 ultrasound studies from 55 patients with accreta and 56 controls for 1374 independent observations. 1205/1374 (87.7% overall, 90% controls, 84.9% cases) studies were given a diagnosis. There were 371 (27.0%) true positives; 81 (5.9%) false positives; 533 (38.8%) true negatives, 220 (16.0%) false negatives, and 169 (12.3%) with uncertain diagnosis. Sensitivity, specificity, positive predictive value, negative predictive value, and accuracy were 53.5%, 88.0%, 82.1%, 64.8%, and 64.8%, respectively. In multivariate analysis, true positives were more likely to have placental lacunae (odds ratio [OR], 1.5; 95% confidence interval [CI], 1.4–1.6), loss of retroplacental clear space (OR, 2.4; 95% CI, 1.1–4.9), or abnormalities on color Doppler (OR, 2.1; 95% CI, 1.8–2.4).
Conclusion
Ultrasound for the prediction of placenta accreta may not be as sensitive as previously described.
See related editorial, page 87
Placenta accreta is defined as an abnormal adherence of placental villi to underlying myometrium with an absence of decidua basalis. Failure to anticipate placenta accreta and prepare for its appropriate management can lead to emergency hysterectomy with profuse, life-threatening hemorrhage, disseminated coagulopathy, renal failure, acute respiratory distress, or even death. Additional surgical complications include cystotomy, ureteral injury, infection, venous thromboembolism or prolonged hospitalization. Accurate antenatal diagnosis of placenta accreta can allow arrangements to be made for a planned delivery at a tertiary care center utilizing a multidisciplinary approach, which has been shown to significantly reduce maternal morbidity.
As the incidence of placenta accreta increases concurrently with an increased incidence of cesarean delivery, patients and providers are more frequently confronted with difficult decisions such as whether to plan for a scheduled hysterectomy or to transfer care to a tertiary care center. With its implications for surgical morbidity and future fertility, this is not a decision taken lightly.
Placenta previa and history of prior cesarean delivery remain the most important predictors of placenta accreta. In addition to clinical risk factors, ultrasound is often used antenatally as an adjunct to clinical history to modify risk estimation for placenta accreta. The accuracy of ultrasound for the prediction of placenta accreta is generally reported to be good with sensitivities ranging from 77–97%. However, prior studies on the accuracy of ultrasound for the prediction of accreta may be subject to bias because of single expert observers, suspicion for accreta, and knowledge of risk factors. Our objective was to assess the accuracy of ultrasound for the prediction of placenta accreta using multiple observers blinded to clinical status.
Materials and Methods
Patients who delivered at the University of Utah between 2000 and 2012 with documentation of a clinical or histopathologic diagnosis of placenta accreta were identified and matched to patients with placenta previa but no accreta by year of delivery. Histopathalogic diagnosis of accreta was confirmed by documentation of placental invasion into the myometrium, and clinical diagnosis of accreta was confirmed by documentation of abnormal adherence of the placenta or evidence of gross placental invasion at the time of surgery. Patients were included if they had ultrasound images of the placenta available at a gestational age of greater than or equal to 16 weeks. For each patient, every image of the placenta was collected, de-identified, and blinded to clinical history. Images from each study were then placed in random sequence using the Microsoft Excel random number generator.
Six investigators consisting of 3 experienced obstetric radiologists (A.M.K., T.C.W., and P.J.W.) and 3 maternal-fetal medicine physicians (A.J.E., D.S.R., and R.M.S.) prospectively reviewed and interpreted each ultrasound study. All 3 radiologists are fellowship trained in abdominal imaging and have more than 10 years’ experience in obstetric ultrasound, including evaluation of accreta. Similarly, all 3 maternal-fetal medicine specialists are fellowship trained and have a minimum of 8 years of experience in diagnosing placenta accreta (>20 years for 2 of the 3 physicians). Investigators were asked to score each imaging study for the presence of accreta (“yes,” “no,” or “unable to determine”) and indicate the presence or absence of specific findings that have been reported to be associated with its diagnosis. These findings included the following: number of lacunae, loss of retroplacental clear space, loss of visualization of the myometrium, and bladder wall irregularity. If color Doppler was used, investigators further identified the presence or absence of the following: subplacental vascularity, vessels bridging from the placenta to the uterine margin, gaps in myometrial blood flow, vessels crossing interface disruption sites, or turbulent lacunae. If a diagnosis was made, investigator confidence for each imaging study was scored on a scale of 0 (none) to 10 (certain), and image quality was scored on a scale from 1 (very poor) to 10 (best).
Study data were collected and managed using REDCap electronic data capture tools hosted at the University of Utah. REDCap (Research Electronic Data Capture) is a secure, web-based application designed to support data capture for research studies, providing: (1) an intuitive interface for validated data entry; (2) audit trails for tracking data manipulation and export procedures; (3) automated export procedures for seamless data downloads to common statistical packages; and (4) procedures for importing data from external sources.
Continuous variables were analyzed with the Student t test. Categorical variables were analyzed by the Wilcoxon-Mann-Whitney, χ 2 or Fisher exact test, where appropriate. Sensitivity, specificity, positive predictive value, negative predictive value, and accuracy were calculated according to standard definitions. For multivariate analyses, all variables of interest were included in a logistic regression model. Covariates were then removed in a stepwise fashion until all covariates in the final model for a particular outcome had a P value of < .2. Receiver operating characteristic curves were generated and the area under the curve computed to determine the discriminative ability of ultrasound to correctly identify placenta accreta. A P value of < .05 was considered statistically significant. All statistical analyses were performed using Stata 12.1 (StataCorp, College Station, TX). The institutional review board of the University of Utah approved this study.
Results
After exclusions, we identified 55 women with placenta accreta with available imaging studies at the University of Utah between 2000 and 2012. Fifty-six women with placenta previa but no accreta and appropriate imaging studies and matched to cases by year of delivery were chosen as controls. Clinical data are summarized in Table 1 . Compared with patients with placenta previa only (no accreta), those with placenta accreta had significantly higher parity, more prior cesarean deliveries and an earlier gestational age at delivery. Maternal age and body mass index at the time of delivery were similar.
| Characteristic | Accreta (Cases) | Controls | P value |
|---|---|---|---|
| Maternal age, y (mean ± SD) | 32.4 ± 5.2 | 31.3 ± 6.9 | .36 a |
| Body mass index, kg/m 2 (mean ± SD) | 29.0 ± 9.4 | 31.9 ± 6.9 | .15 a |
| Gravidity (median, range) | 5 (2–18) | 4 (1–14) | .0047 b |
| Parity (median, range) | 3 (1–7) | 2 (0–10) | .0090 b |
| Cesarean delivery (n, %) | < .001 c | ||
| 0 | 1 (1.8) | 16 (28.6) | |
| 1 | 17 (30.9) | 25 (44.6) | |
| 2 | 16 (29.1) | 9 (16.1) | |
| 3+ | 21 (38.2) | 6 (10.7) | |
| Delivery gestational age, wk (mean ± SD) | 33.8 ± 3.0 | 34.9 ± 4.0 | .12 a |
| Gestational age at study, wk (median, interquartile range) | 29.2 (25.1–32.4) | 29.3 (23.9–33.6) | .11 b |
The 55 women with placenta accreta had a total of 116 ultrasound studies and the 56 women with placenta previa but no accreta had 113 studies. Thus, a total of 229 ultrasound studies were available for review. All ultrasound images were collected, deidentified, and placed in a random order. Each of the 6 investigators reviewed all studies for a total of 1374 independent observations.
A specific diagnosis regarding the presence or absence of placenta accreta was reported for 1205/1374 (87.7%) studies, including 614/678 (90.6%) of controls and 591/696 (84.9%) of cases ( P = .001). Of studies receiving a diagnosis, diagnostic performance characteristics were as follows: 371 true positives (30.8%), 81 false positives (6.7%), 533 true negatives (44.2%), and 220 false negatives (18.3%). 165/1374 studies (12.0%) were designated “unable to determine” and 4/165 studies (2.4%) were not given a diagnosis. Results for sensitivity, specificity, positive predictive value, negative predictive value and overall accuracy are shown in Table 2 . Two analyses are presented; one excluding studies that were interpreted as “uncertain” and one that assigned uncertain diagnoses as “no accreta.” Receiver operator characteristic curves, which accounted for ultrasound as a binary test (absence or presence of accreta), are shown in Figure 1 . Overall, the diagnostic performance characteristics were improved when uncertain diagnoses were excluded.
| Characteristic | Excluding missing/uncertain diagnoses | Missing/uncertain diagnoses assigned as no accreta |
|---|---|---|
| Sensitivity | 62.8 (58.7–66.7) | 53.3 (49.5–57.1) |
| Specificity | 86.8 (83.9–89.4) | 88.1 (85.4–90.4) |
| PPV | 82.1 (78.7–85.0) | 82.1 (78.8–85.4) |
| NPV | 70.8 (68.5–73.0) | 64.8 (62.9–66.7) |
| Accuracy | 75.0 (72.5–77.4) | 65.8 (63.2–68.3) |
Given that individual patients may have had more than one ultrasound study performed, we examined accuracy of diagnoses by patient. Diagnoses for the 56 women with previa only (ie, controls) were correct significantly more often than for the 55 cases (75% vs 60.4%, respectively, P < .0068). Similarly, concordant with higher overall specificity, incorrect results were lower for controls compared with cases (13.8% vs 27.0%, respectively, P = .0093). The ratio of unknown or missing diagnoses was not significantly different between cases and controls (13.3% vs 10.2%, respectively, P = .21).
To determine which findings on ultrasound were associated with the presence of placenta accreta, an initial model was created with accreta as the outcome and included the following specific findings as covariates: number of lacunae, loss of the retroplacental clear space, loss of visualization of the myometrium, irregular bladder wall, or any color Doppler abnormalities. After stepwise removal of covariates from the model until remaining covariates had a P value of < .2, only loss of visualization of the myometrium was excluded in the final model. Placenta accreta was associated with more placental lacunae (odds ratio [OR], 1.4; 95% CI, 1.3–1.6), loss of the retroplacental clear space (OR, 2.2; 95% CI, 1.6–3.0), an irregular bladder wall (OR, 1.3; 95% CI, 1.0–1.6) and color Doppler abnormalities (OR, 1.3; 95% CI, 1.1–1.4). The receiver operating characteristic curve for this model is shown in Figure 2 .
Next, we examined whether specific ultrasound findings and/or patient demographic factors were associated with accurate and inaccurate diagnoses (ie, true positives, true negatives, false positives, false negatives, and uncertain diagnosis). The results of univariate analyses are shown in Table 3 . True positives were significantly associated with lower maternal BMI (26.9 vs 29.1, P = .0034), better image quality (6.1 vs 5.7, P = .0026), more lacunae (5 vs 1.2, P < .001), loss of the retroplacental clear space (91.9% vs 19.0%, P < .001), an irregular bladder wall (39.4% vs 4.3%, P < .001) and color Doppler abnormalities (97.3% vs 65.7%, P < .001). False positives were associated with similar findings, but color Doppler abnormalities were not significantly different from other diagnoses ( P = .45).
| Variable | True positive | True negative | False positive | False negative | Diagnosis given | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Yes | No | P value | Yes | No | P value | Yes | No | P value | Yes | No | P value | Yes | No | P value | |
| Gestational age at study, wk | 28.9 ± 4.9 | 28.3 ± 5.5 | .043 | 28.7 ± 5.6 | 28.3 ± 5.2 | .28 | 28.0 ± 5.3 | 28.5 ± 5.4 | .46 | 27.5 ± 5.2 | 28.7 ± 5.4 | .0034 | 28.5 ± 5.3 | 28.3 ± 5.6 | .59 |
| Maternal age at study, y | 33.0 ± 5.0 | 31.8 ± 6.7 | .0015 | 32.4 ± 5.5 | 31.6 ± 7.5 | .021 | 31.4 ± 5.8 | 32.1 ± 6.4 | .28 | 31.3 ± 5.8 | 32.2 ± 6.4 | .042 | 32.0 ± 6.4 | 33.1 ± 5.6 | .029 |
| Maternal BMI at delivery | 26.9 ± 9.9 | 29.1 ± 10.3 | .0034 | 28.5 ± 10.1 | 28.5 ± 10.4 | .94 | 23.6 ± 11.6 | 28.8 ± 10.1 | < .001 | 31.8 ± 9.5 | 27.7 ± 10.3 | < .001 | 28.4 ± 10.2 | 29.4 ± 10.6 | .37 |
| Number of placental lacunae | 5.0 ± 2.7 | 1.2 ± 2.7 | < .001 | 0.74 ± 1.0 | 3.2 ± 2.8 | < .001 | 2.9 ± 2.4 | 2.2 ± 2.6 | .014 | 1.4 ± 1.4 | 2.4 ± 2.7 | < .001 | 2.3 ± 2.6 | 1.7 ± 1.9 | .009 |
| Image quality a | 6.1 ± 1.9 | 5.7 ± 2.2 | .0026 | 6.2 ± 2.0 | 5.6 ± 2.1 | < .001 | 5.6 ± 2.1 | 5.8 ± 2.1 | .32 | 5.7 ± 2.0 | 5.9 ± 2.1 | .30 | 6.0 ± 2.0 | 4.0 ± 2.0 | < .001 |
| Confidence of interpretation b | 63.2 ± 22.3 | 69.5 ± 22.4 | < .001 | 76.4 ± 19.5 | 61.6 ± 22.6 | < .001 | 48.0 ± 25.8 | 69.0 ± 21.7 | < .001 | 68.9 ± 21.5 | 67.4 ± 22.8 | .39 | 69.0 ± 22.7 | 49.0 ± 9.4 | < .001 |
| Loss of retroplacental clear space | 341 (91.9) | 191 (19.0) | < .001 | 28 (5.3) | 504 (59.9) | < .001 | 70 (86.4) | 462 (35.7) | < .001 | 25 (11.4) | 507 (43.9) | < .001 | 464 (38.5) | 68 (40.2) | < .001 |
| Loss of visualization of the myometrium | 338 (91.1) | 188 (18.7) | < .001 | 28 (5.3) | 498 (59.2) | < .001 | 67 (82.7) | 459 (35.5) | < .001 | 25 (11.4) | 501 (43.4) | < .001 | 458 (38.5) | 68 (40.2) | < .001 |
| Irregular bladder wall | 146 (39.4) | 43 (4.3) | < .001 | 10 (1.9) | 179 (21.3) | < .001 | 18 (22.2) | 171 (13.2) | .014 | 12 (5.45) | 177 (15.3) | .001 | 186 (15.4) | 3 (1.8) | < .001 |
| Any abnormal Doppler finding | 361 (97.3) | 659 (65.7) | < .001 | 322 (60.4) | 698 (83.0) | < .001 | 66 (81.5) | 954 (73.8) | .45 | 149 (67.7) | 871 (75.5) | .022 | 898 (74.5) | 122 (72.2) | < .001 |
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
