This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Medical Informatics, is properly cited. The complete bibliographic information, a link to the original publication on http://medinform.jmir.org/, as well as this copyright and license information must be included.
Ectopic pregnancy (EP) is a serious complication of assisted reproductive technology (ART). However, there is no acknowledged mathematical model for predicting EP in the ART population.
The goal of the research was to establish a model to tailor treatment for women with a higher risk of EP.
From December 2015 to July 2016, we retrospectively included 1703 women whose serum human chorionic gonadotropin (hCG) levels were positive on day 21 (hCG21) after fresh embryo transfer. Multivariable multinomial logistic regression was used to predict EP, intrauterine pregnancy (IUP), and biochemical pregnancy (BCP).
The variables included in the final predicting model were (hCG21, ratio of hCG21/hCG14, and main cause of infertility). During evaluation of the model, the areas under the receiver operating curve for IUP, EP, and BCP were 0.978, 0.962, and 0.999, respectively, in the training set, and 0.963, 0.942, and 0.996, respectively, in the validation set. The misclassification rates were 0.038 and 0.045, respectively, in the training and validation sets. Our model classified the whole in vitro fertilization/intracytoplasmic sperm injection–embryo transfer population into four groups: first, the low-risk EP group, with incidence of EP of 0.52% (0.23%-1.03%); second, a predicted BCP group, with incidence of EP of 5.79% (1.21%-15.95%); third, a predicted undetermined group, with incidence of EP of 28.32% (21.10%-35.53%), and fourth, a predicted high-risk EP group, with incidence of EP of 64.11% (47.22%-78.81%).
We have established a model to sort the women undergoing ART into four groups according to their incidence of EP in order to reduce the medical resources spent on women with low-risk EP and provide targeted tailor-made treatment for women with a higher risk of EP.
Ectopic pregnancy (EP) is the leading cause of maternal morbidity and mortality during the first trimester, accounting for 5% to 10% of all maternal deaths [
The aim of this study was to establish such a model to rank the women undergoing IVF/ICSI-ET treatment into a few groups according to the incidence of EP. The goals are to reduce medical resources spent on the low-risk EP group, provide more targeted tailor-made treatment for women at a high risk of EP, and further improve the detection rate for this adverse outcome.
This was a retrospective observational cohort study performed from December 2015 to July 2016. Datasets of all fresh ET cycles were recorded. Data were entered into a database by the clinical support staff. The database was used to collect basic and clinical characteristics of patients including age, body mass index, baseline sex hormone levels, main causes of infertility, endometrial thickness on the day of hCG used for triggering ovulation, details of ovarian stimulation protocols, insemination method, date of insemination, date of ET, numbers of ETs, date of hCG examination, serum concentrations of hCG, fertilization results, and pregnancy types, including EP, biochemical pregnancy (BCP), and intrauterine pregnancy (IUP). The inclusion criteria were (1) serum hCG level >5 IU/L on days 14 (hCG14) and 21 post-ET (hCG21); (2) hCG examinations were tested in our own lab (the same platform); and (3) hCG levels were tested exactly on day 14 or 21 post-ET. Of these, 1703 cycles were selected. The cycles were further divided into three outcome groups: EP, IUP, or BCP. A flowchart of this process is shown in
Flowchart of the data selection strategy. hCG14 and hCG21: serum hCG levels on days 14 and 21 post–embryo transfer; EP: ectopic pregnancy; ET: embryo transfer; IUP: intrauterine pregnancy; BCP: biochemical pregnancy.
Basic and clinical characteristics in related to different pregnancy outcomes.
Characteristic | Intrauterine pregnancy, (n=1576) | Ectopic pregnancy, (n=78) | Biochemical pregnancy, (n=49) | |
Age in years, mean (quartile) | 32 (29-35) | 32 (29-35) | 32 (30-35) | |
Body mass index (kg/m2), mean (quartile) | 22.1 (20.3-24.5) | 22.5 (19.5-24.5) | 22.6 (20.1-25.5) | |
|
|
|
|
|
|
Male infertility | 530 (33.6) | 16 (20.5) | 24 (49.0) |
|
Endometriosis | 46 (2.9) | 1 (1.3) | 2 (4.1) |
|
Anovulatory infertility | 81 (5.2) | 9 (11.5) | 5 (10.2) |
|
Tubal factor | 639 (40.5) | 35 (44.9) | 15 (30.6) |
|
Unexplained and others | 280 (17.8) | 17 (21.8) | 3 (6.1) |
Retrieved oocytes, mean (quartile) | 10 (7-14) | 10 (7-13) | 10 (6-14) | |
|
|
|
|
|
|
Cleavage | 1540 (97.7) | 76 (97.4) | 46 (93.9) |
|
Blastocyst | 36 (2.3) | 2 (2.6) | 3 (6.1) |
|
|
|
|
|
|
1 | 105 (6.7) | 5 (6.4) | 5 (10.2) |
|
2 | 1471 (93.3) | 73 (93.6) | 44 (89.8) |
hCG14b, mean (quartile) | 827 (524-1300) | 186 (103-289) | 139 (71-300) | |
hCG21c, mean (quartile) | 15,570 (9954-22,626) | 1870 (815-3107) | 95 (27-275) | |
Ratio of calculated 48-hour rising, mean (quartile) | 2.3 (2.1-2.4) | 1.9 (1.5-2.4) | 0.9 (0.6-1.1) | |
hCG21/hCG14, mean (quartile) | 17.5 (13.8-22.0) | 10.3 (4.1-20.4) | 0.7 (0.2-1.4) |
aET: embryo transfer.
bhCG14: serum level of human chorionic gonadotropin on 14th day post–embryo transfer.
chCG21: serum level of human chorionic gonadotropin on 21st day post–embryo transfer.
The ovarian stimulation protocols used in our center include a gonadotrophin releasing hormone (GnRH) antagonist protocol, a GnRH agonist long protocol, a GnRH agonist short protocol, and mild stimulation protocols, as described previously [
An IUP was defined as one or more intrauterine gestational sacs detected by transvaginal sonography (TVS) at 30 or 37 days after embryo transfer. As the heartbeat is not necessarily present on the 30th or 37th day post-ET, as long as the gestational sac is seen within the uterus on the 30th or 37th day post-ET it is an IUP, which includes a certain proportion of first-trimester miscarriage. An EP was diagnosed by visualization of one or more gestation sacs outside the uterus detected by TVS. A BCP was indicated by a temporary rise of serum hCG without gestational sacs inside or outside the uterus detected by TVS.
The serum β-hCG level of each patient was assessed from December 2015 to July 2016 using an Access UniCel DxI 800 chemiluminescence system and an Access total β-hCG assay kit (both Beckman Coulter Inc), standardized to the highly purified World Health Organization 5th International Standard for hCG. Quality controls used were the Lyphochek trilevel Immunoassay Plus Controls (catalogue 370; lot number 40320; Bio-Rad Laboratories). The interassay variation was 7.9% in low-level Bio-Rad immunoassays and controls, 7.4% in mid-level controls, and 4.1% in high-level controls.
Normally distributed variables were presented as mean and standard deviation. Nonnormally distributed variables were presented as median and quartile. Before further analysis, a generalized additive model was used to explore the suitable function between explanatory variables and outcome. The outcome variables were classified into three subgroups: EP, IUP, and BCP. Multinomial logistic regression was used because there were more than two outcome variables. Before analysis, the dataset was partitioned into a training set and a validation set at the proportion of 0.75:0.25, and multinomial logistic regression was performed on the training set to establish the prediction model. Specifically, the hCG21 and hCG21/hCG14 ratio were entered as quadratic forms, and the cause of infertility was treated as a dummy variable with reference to male-factor infertility. Akaike’s information criterion (AIC) and Schwarz-Bayesian information criterion (SBIC) were used to compare various models to determine the best-fitting model; the model having the smallest AIC and BIC values was preferred. The model was then applied to the validation set, and the areas under the receiver operating curve (AUC) and misclassification rates were calculated for model evaluation. To build a more targeted predictive model, according to the incidence of EP, we partitioned cases into 12 groups based on the prediction probability of EP and BCP in each group, using the actual outcome proportions of the three categories. An exact (Clopper-Pearson) confidence limits or Wald confidence limits method was used to calculate the 95% confidence intervals of EP incidence. The data were analyzed with JMP Pro version 14.0 software (SAS Institute Inc), and a 2-sided
As the early pregnancy outcome was an EP, IUP, or BCP, we used univariate multinomial logistic regression to test the relationships between each independent variable and the outcome variable. Considering the strong correlation between hCG14 and hCG21 (
The independent variables identified in the univariate analysis were further examined by multivariate multinomial logistic regression. Cleavage or blastocyst embryo transfer was not of significance in predicting pregnancy outcomes because after removing this independent variate, the SBIC and AIC were reduced from 385.44 to 371.83 and 283.06 to 279.64, respectively. To distinguish EP and non-EP, we explored the cutoff value of the predictive model. The default cutoff value of the software is 0.5, which can be adjusted with reference to the prevalence of EP. Based on the incidence of EP in our data and referring to Van Calster’s [
The ability of the model to predict one outcome versus the other two outcomes in the training and validation sets was evaluated by the AUC analysis and misclassification rate, as shown in
Sensitivity and specificity of the model in the training and validation sets.
The performance of the predicting model.
Datasets | Area under the receiver operating curve | MRd | |||
|
IUPa, (n=1576) | EPb, (n=78) | BCPc, (n=49) |
|
|
Training set | 0.978 | 0.962 | 0.999 | 0.038 | |
Validation set | 0.963 | 0.942 | 0.996 | 0.045 |
aIUP: intrauterine pregnancy
bEP: ectopic pregnancy.
cBCP: biochemical pregnancy.
dMR: misclassification rate
The predicted and actual occurrence of different pregnancy outcomes in our data.
Actual pregnancy outcomes | Predicted pregnancy outcomes, n (%) | |||||
|
Training set | Validation set | ||||
|
IUPa, (n=1208) | EPb, (n=31) | BCPc, (n=38) | IUP, (n=398) | EP, (n=13) | BCP, (n=15) |
IUP | 1172 (97.0) | 9 (29.0) | 1 (2.6) | 387 (97.2) | 5 (38.5) | 2 (13.3) |
EP | 35 (2.9) | 21 (67.7) | 2 (5.3) | 11 (2.8) | 8 (61.5) | 1 (6.7) |
BCP | 1 (0.1) | 1 (3.2) | 35 (92.1) | 0 (0) | 0 (0) | 12 (80.0) |
aIUP: intrauterine pregnancy
bEP: ectopic pregnancy.
cBCP: biochemical pregnancy.
For this, we further explored the grouping method according to the predicted probabilities of pregnancy outcomes. Because the sum of the predicted probabilities of IUP+EP+BCP=1, if two predicted probabilities of EP and BCP are known, the other one is known. So, we divided the whole population into more groups based on the predicted probabilities of EP and BCP. As shown in
Classifying the population into subgroups according to the predicted probabilities of IUP, EP, and BCP using a training set and a validation set of data. IUP: intrauterine pregnancy; EP: ectopic pregnancy; BCP: biochemical pregnancy.
Classification according to the incidence of ectopic pregnancy (n=1703).
Group | Predicted probability | n (%) | Incidence of EPa (%) | Number of actual cases | ||
|
|
|
|
IUPb | EP | BCPc |
1: EP low risk | ProbEP<0.1 & ProbBCP<0.5 | 1460 (85.73) | 0.52 (0.23-1.03) | 1453 | 7 | 0 |
2: predicted BCP | ProbBCP≥0.5 | 52 (3.05) | 5.79 (1.21-15.95) | 2 | 3 | 47 |
3: gray zone | 0.1≤ProbEP<0.5 & ProbBCP<0.5 | 152 (8.93) | 28.32 (21.10-35.53) | 108 | 43 | 1 |
4: EP high risk | ProbEP≥0.5 | 39 (2.29) | 64.11 (47.22-78.81) | 13 | 25 | 1 |
aEP: ectopic pregnancy.
bIUP: intrauterine pregnancy
cBCP: biochemical pregnancy.
Here, through the multivariate multinomial logistic regression method, we have established a mathematical model to predict the probability of having an IUP, EP, or BCP in pregnant women subjected to ART using predictors of hCG21, ratio of hCG21/hCG14, and main cause of infertility. We further classified the whole population into four subgroups according to the incidence of EP in each group in order to rearrange our clinical routine to reduce medical resources spent on women with a low risk of EP and provide more targeted tailor-made treatments for women with a higher risk of EP.
Considering that current routine clinical examinations cannot diagnose EP in early pregnancy, the routine in our reproductive center for a woman undergoing IVF/ICSI-ET treatment is to measure serum hCG levels around day 14 and 21 post-ET and then take two TVS tests to confirm the location of the gestational sac on day 30 and 37 post-ET, with sometimes even another test on day 44 post-ET. Based on the good predictive effect of our model, we are currently developing this regression model into computer software to better manage women in early pregnancy according to their risk of EP. To be specific, for the low-risk EP group (accounting for 85.73% of the whole population), we are considering reducing the frequency of TVS tests to one on day 30 post-ET. For the predicted high-risk EP group, with incidence of EP of 64.11% (95% CI 47.22%-78.81%), an immediate TVS examination is recommended after the hCG21 test. For the grey zone group, with incidence of EP of 28.32% (95% CI 21.10%-35.53%), the original frequency of two TVS visits is recommended. For the predicted BCP group, although the incidence of EP is significantly higher than that in the low-risk EP group, the likelihood of having a spontaneous abortion is also high and these women can be treated as belonging to the low-risk EP group.
The acknowledged M4 model for predicting EP in pregnancies of the unknown location (PUL) population [
An hCG ratio strategy was reported to have a better sensitivity in predicting EP compared with a single serum hCG level [
The idea of predicting EP using multinomial logistic regression was actually derived from the work of Condous et al [
Tubal factor infertility was reported to be the most prominent risk factor for EP after IVF/ICSI-ET treatment [
The prevalence of EP per clinical pregnancy in fresh IVF/ICSI-ET cycles was reported to be 4.6% [
A major limitation in our study is the lack of confirmed efficacy of our model compared with the traditional method; we aim to design a randomized controlled study for this. The outcome measurement is the incidence of EP after the 37th day post-ET. We sought to determine if the incidence of EP detected after that time point in the group using our model is comparable or better than in the group using the traditional clinical routine. Second, although our groups 2 to 4 (
A significant amount of time and resources are spent in ART centers on monitoring women with early pregnancies to identify EP in time to prevent its complications. Early tests for assuring the location of gestational sacs have significant cost burdens on patients and centers. In our study, we established a mathematical model for predicting EP according to the incidence of EP. According to our model, we have sought to rearrange our clinical routine to reduce the medical resources spent on women with low EP risk and provide targeted tailor-made treatment for women with a higher risk of EP. We hope that this method can enable the reasonable use of limited medical resources and improve the efficiency in the management of pregnancies in woman undergoing IVF/ICSI-ET treatments.
Univariate logistic analysis to assess the effect of the independent variables on different pregnancy outcomes.
Multiple logistic analysis to establish the prediction model.
Akaike’s information criterion
assisted reproductive technology
areas under the receiver operating curve
biochemical pregnancy
ectopic pregnancy
embryo transfer
gonadotrophin releasing hormone
human chorionic gonadotropin
day 14 post–embryo transfer
day 21 post–embryo transfer
intracytoplasmic sperm injection
intrauterine pregnancy
in vitro fertilization
pregnancies of unknown location
Schwarz-Bayesian information criterion
transvaginal sonography
This study was supported by the National Key Research and Development Program of China (Grant No. 2016YFC1000201, 2018YFC1002104, 2018YFC1002106, 2016YFC1000302); the National Natural Science Foundation of China (Grant No. 81300373, 81771650); the Capital Health Research and Development of Special Project (Grant No. 2018-1-4091); the program for Innovative Research Team of Yunnan, China (Grant No. 2017HC009); and Major National R&D Projects of China (Grant No. 2017ZX09304012-012).
HYX participated in design, data collection, and manuscript writing. GSF was in charge of statistical analysis and contributed to manuscript writing. YW and YH contributed to data collection, clinical consultation, and manuscript writing. BWM edited this manuscript. HXZ, LYW, and RY contributed to clinical consultation. RL conceived and designed this study, edited the manuscript, and approved the submission. All authors have read and approved the final manuscript.
None declared.