https://orcid.org/0009-0004-1465-3333
Figures
Abstract
Objective
As the first model in predicting the failure of early medical abortion (EMA) was inefficient, this study aims to develop and validate a risk assessment model for predicting the failure of EMAs more accurately in a clinical setting.
Methods
The derivation cohort was obtained from a comprehensive systematic review and meta-analysis. The clinically significant risk factors were identified and combined with their corresponding odds ratios to establish a risk assessment model. The risk factors were assigned scores based on their respective weightings. The model’s performance was evaluated by an external validation cohort obtained from a tertiary hospital. The outcome was defined as the incidence of EMA failure.
Results
A total of 126,420 patients who had undergone medical abortions were included in the systematic review and meta-analysis, and the pooled failure rate was 6.7%. The final risk factors consisted of gestational age, maternal age, parity, previous termination of pregnancy, marital status, type of residence, and differences between gestational age calculated using the last menstrual period and that measured via ultrasound. The risk factors were assigned scores based on their respective weightings, with a maximum score of 19. The clinical prediction model exhibited a good discrimination, as validated by external verification (402 patients) with an area under the curve of 0.857 (95% confidence interval 0.804–0.910). The optimal cutoff value was determined to be 13.5 points, yielding a sensitivity of 83.3% and specificity of 75.4%.
Conclusion
This study effectively establishes a simple risk assessment model including seven routinely available clinical parameters for predicting EMA failure. In preliminary validation, this model demonstrates good performance in terms of predictive efficiency, accuracy, calibration, and clinical benefit. However, more research and validation are warranted for future application.
Citation: Liu A-H, Xu B, Li X-W, Yu Y-W, Guan H-X, Sun X-L, et al. (2024) Development and validation of a risk assessment model for predicting the failure of early medical abortions: A clinical prediction model study based on a systematic review and meta-analysis. PLoS ONE 19(12): e0315025. https://doi.org/10.1371/journal.pone.0315025
Editor: Andrea Cioffi, University of Foggia: Universita degli Studi di Foggia, ITALY
Received: August 13, 2024; Accepted: November 19, 2024; Published: December 20, 2024
Copyright: © 2024 Liu et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All relevant data are within the paper and its Supporting Information files.
Funding: The author(s) received no specific funding for this work.
Competing interests: The authors have declared that no competing interests exist.
Abbreviations: EMA, early medical abortion; OR, odds ratio; IGF-1, insulin-like growth factor 1 protein; VEGF, vascular endothelial growth factor protein; FP-V2 PGF2α, receptor splice variant 2; AUC, an area under the receiver operating characteristic curve; ROC, receiver operating characteristic curve; PROSPERO, the International Prospective Register of Systematic Reviews; CI, confidence interval; GA, gestational age; MA, maternal age; PA, parity; PT, previous termination of pregnancy; MS, marital status; TR, type of residence; LU, differences between gestational age calculated using the last menstrual period and that measured via ultrasound; VD, only vaginal deliveries and spontaneous delivery of placenta; CS ≥1, caesarean section; MR ≥1, manual removal of placenta; PT, previous termination of pregnancy; PM, previous medical abortions; PS1 ≥1, previous surgical abortion: ≥56 days of gestation; PS2 ≥1, previous surgical abortion: <56 days of gestation; PS3 ≥2, previous surgical abortion: both <56 and ≥56 days of gestation; MC, married or cohabiting; RU, rural areas; SD, standard deviation; IQR, interquartile range; LMP, gestational age calculated using the last menstrual period; US, gestational age measured via ultrasound
1. Introduction
Medical abortions emerged as an alternative to surgical abortions for 1st trimester pregnancies with the availability of prostaglandins in the 1970s and anti-progesterones in the 1980s [1–4]. Currently, an estimated 73 million induced abortions are performed annually worldwide [5, 6]. In the United States in 2021, medical abortions accounting for a substantial proportion (56%) of all induced abortions [7]. In England and Wales in 2021, medical abortions constituted approximately 87% of the total number of abortions [8]. Although misoprostol has demonstrated high efficacy in medical abortion regimens containing mifepristone, with success rates exceeding 90% [9–11], the pooled failure rate of medical abortions was determined to be 6.7% (95% CI 5.3%, 8.3%) based on a meta-analysis of the included studies and this study (Table 1) [11–19]. Based on the aforementioned data, a roughly estimated 3 million pregnant women experience medical abortion failure annually worldwide. Therefore, revealing and summarizing the risk factors associated with early medical abortion (EMA) failure is of paramount importance in clinical practices.
At the molecular level, the expression of insulin-like growth factor 1 protein (IGF-1) and vascular endothelial growth factor protein (VEGF) in chorionic villi, the expression of PGF2α receptor splice variant 2 (FP-V2) in human decidua, and the G2014G genotype in the estrogen receptor 1 gene has been reported to be associated with EMA failure [20–22]. Two primary factors (endometrial thickness and serum β-hCG level) have been identified as predictors of EMA failure after patients undergo medical abortions [23–26]. Although peripheral blood α1-acid glycoprotein has been linked to incomplete EMAs, its clinical validation and application still require further investigation [27].
Furthermore, it is noteworthy that a multitude of factors have been identified as being associated with the risk of EMA failure before patients undergoing medical abortions [11–19, 28–36]. These factors can be categorized into six main groups: different drugs [28–31], different drug dosages [30–32], different administration methods (routes and time intervals) for drugs [30, 31, 33], different settings (home or hospital) [34, 35], different periods [13, 36], and other epidemiological and clinical risk factors [11–19]. According to the specified clinical guidelines from different countries or regions, clinicians are typically unable to modify the former five categories for specific patients and cannot employ them to assess the risk of EMA failure in clinical settings [37–41]. However, risk factors of the 6th category generally exhibit variations for different patients and can be easily applied in clinical practice.
In 2011, Heikinheimo et al. identified five epidemiological and clinical risk factors associated with medical abortion failure (surgical evacuation performed) comprehensively [13]. In 2019, Meaidi et al. developed and validated a risk assessment model for predicting early medical abortion failure (surgical intervention performed) based on epidemiological and clinical data; however, the model was deemed inefficient as indicated by an area under the curve (AUC) of 0.63 [12]. Therefore, the objective of this study is to develop and validate a more accurate risk assessment model for clinically predicting EMA failure, building upon the foundation established by Meaidi and other researchers.
2. Methods
2.1 Source of data
The derivation cohort was established based on a comprehensive systematic review and meta-analysis of 9 high-quality cohort studies, including 2 prospective and 7 retrospective cohort studies (Table 2) [11–19]. This systematic review was registered in the International Prospective Register of Systematic Reviews (PROSPERO), with registration number: CRD42024505032. The electronic databases, including Medline, Embase, Scopus, Web of Science, and Cochrane Library, were systematically searched from their inception to January 19th, 2024. During the search process, specific terms such as "surgical or failed or incomplete or unsuccessful or failure", "medical abortion" and "surgical abortion" were employed (S1 Appendix).
The validation cohort consisted of EMA patients from Zhongshan Hospital of Xiamen University, whose hospital admission dates were between January 1st, 2019 and May 13rd, 2024. Access to the relevant data was granted for research purposes starting from May 13rd, 2024. The authors ensured that no personally identifiable information of individual participants was accessible during or after data collection. Patient anonymity was maintained throughout this study.
Additionally, we declared that the method section was written according to TRIPOD development and validation writing standard [42].
2.2 Study participants
A total of 126,420 patients who had undergone medical abortions were included in our derivation cohort, and they were from Europe, Asia and North America. The range of time of the included studies was from 1994 to 2020. The flowchart of screening studies is presented in Fig 1A. The original information of the literature selection process is presented in S2 Appendix. The inclusion criteria, exclusion criteria, data extraction, and quality assessment are detailed in S3 Appendix [43]. All data extracted from the primary research sources are presented in S4 Appendix. The certainty assessment of meta-analysis is presented in Table 3.
(A) PRISMA flow chart of the literature selection process. (B) Process for the selection of patients in the validation cohort. PRISMA, preferred reporting items for systematic reviews and meta-analyses. EMA, early medical abortion.
The study participants of validation cohort were aged between 16 and 43 years, who received oral administration of mifepristone (150 mg) upon hospitalization, and were followed by oral administration of misoprostol (600 mcg) after a time interval of 72 hours according to the medical abortion guidelines in China [37]. Women were discharged home following medical abortions, which was confirmed through visualization of the products of conception or by ultrasound scanning if necessary. All participants were scheduled to attend a follow-up visit, which included sonographic examination, approximately 6–8 and 13–15 days or longer after the administration of mifepristone.
2.3 Study outcomes
In derivation cohort, the success and failure of medical abortions had been defined clearly among included studies (Table 2) [44–47].
The outcomes of EMAs in validation cohort were classified as follows: (1) Success: Complete abortion without the need for surgical intervention; (2) Failure: Requirement for surgical evacuation to complete the abortion for any reason.
2.4 Clinical predictors
According to the systematic review, all clinical risk factors associated with EMA failure were identified and incorporated into the risk assessment model. We conducted a meta-analysis to combine odds ratios (ORs) and 95% confidence intervals (CIs) of risk factors, or alternatively, we directly extracted ORs and 95% CIs from a high-quality study with a large sample size. Subsequently, we calculated the corresponding β-coefficients (Fig 2). Related mathematical principles are presented in Fig 3 [48]. The scores of risk factors were determined by multiplying the β-coefficients by 10 and rounding them to the nearest whole number [49].All risk factors in the risk assessment model were categorized and assigned scores to establish a risk scoring system based on the systematic review and meta-analysis [50]. The scores of each risk factor were summarized to calculate the total score [51]. The higher the cumulative score, the greater the risk of EMA failure.
GA7, gestational age of 42–48 days. GA8, gestational age of 49–55 days. GA9, gestational age of 56–62 days. GA12, gestational age of 9–12 weeks. MA1, maternal age of 20–24 years. MA2, maternal age of 25–29 years. MA3, maternal age of 30–34 years. MA4, maternal age of 35–39 years. MA5, maternal age of 40–49 years. PA, parity. VD, only vaginal deliveries and spontaneous delivery of placenta. CS, ≥1 caesarean section. MR, ≥1 manual removal of placenta. PT, previous termination of pregnancy. PM, previous medical abortions. PS1, ≥1 previous surgical abortion: ≥56 days of gestation. PS2, ≥1 previous surgical abortion: <56 days of gestation. PS3, ≥2 previous surgical abortion: both <56 and ≥56 days of gestation. MC, married or cohabiting. RU, rural areas. LU, differences between gestational age calculated using the last menstrual period and gestational age calculated via ultrasound.
P, probability of occurrence of an event (the failure of early medical abortion). X, independent variables (risk factors). β, regression coefficients of risk factors. OR, odds ratio. GA7, gestational age of 42–48 days. GA8, gestational age of 49–55 days. GA9, gestational age of 56–62 days. GA12, gestational age of 9–12 weeks. MA2, maternal age of 25–29 years. MA3, maternal age of 30–34 years. MA4, maternal age of 35–39 years. MA5, maternal age of 40–49 years. PA, parity. VD, only vaginal deliveries and spontaneous delivery of placenta. CS, ≥1 caesarean section. MR, ≥1 manual removal of placenta. PT, previous termination of pregnancy. PM, previous medical abortions. PS1, ≥1 previous surgical abortion: ≥56 days of gestation. PS2, ≥1 previous surgical abortion: <56 days of gestation. PS3, ≥2 previous surgical abortion: both <56 and ≥56 days of gestation. MC, married or cohabiting. RU, rural areas. LU, differences between gestational age calculated using the last menstrual period and gestational age calculated via ultrasound.
2.5 Sample size
As the derivation cohort was established based on a systematic review, the study participants were included according to the inclusion and exclusion criteria in development cohort.
In validation cohort, based on the anticipated outcome proportion (6.7%) and the desired margin of error (≤0.05), the required sample size was calculated as n = 96.1 [52].
In order to better test the model, we decided to set the sample size to be four times of the required sample size (96.1⨯4 = 384.4).
2.6 Missing data
Patients who were lost to follow-up were contacted via telephone and queried regarding their medical condition. Patients with incomplete clinical information were additionally contacted via telephone and requested to provide further details. Relevant patients who were still lost to follow-up or had incomplete clinical data were excluded from the study following our efforts in establishing telephone contact.
2.7 Statistical analysis methods
The data obtained from the derivation cohort were utilized for the development of our clinical prediction model. The OR along with its corresponding 95% CI for each risk factor was extracted. The heterogeneity was assessed using the Q-test and I2 statistics. Due to the existence of heterogeneity among included studies in term of population, study design and sampling, the random-effects model was utilized. The statistical analysis of the data was conducted using Review Manager 5.4. A level of P < 0.05 was adopted, unless otherwise specified, to determine statistical significance.
The data obtained from the validation cohort were used for the evaluation of our risk assessment model. A risk score system was employed to compute the total score of clinical variables for each patient within the cohort. The total score was utilized to construct a receiver operating characteristic (ROC) curve. Then, the sensitivity, specificity, optimal cutoff point, and AUC were computed. The AUC represents the predictive performance, with a range of values between 0.5 and 1.0. The greater the value, the higher the level of prediction accuracy. Based on the optimal cutoff point, patients were stratified into low-risk and high-risk groups. The model performance was assessed using a calibration plot and OE ratio [53]. Additionally, the decision curve analysis was used to measure risk threshold and net benefit of the risk prediction model [53]. Moreover, a series of subgroup analyses were conducted within the validation cohort, in order to assess the model performance across different clinical risk parameters. The statistical analysis was conducted using SPSS 22.0, GraphPad Prism 9.5 and R version 4.4.1.
3. Results
3.1 Description of the cohorts
3.1.1 Derivation cohort.
A total of 126,420 patients who underwent medical abortions were included in the derivation cohort, with the majority having gestational ages of less than 14 weeks and maternal ages ranging between 15–49 years old. These patients had been administered mifepristone and/or prostaglandin analogs for the purpose of medical termination of pregnancy. The majority of patients in the cohort (at least 97%) had a gestational age of less than 12 weeks, while their ages mainly ranged from 15 to 49 years old. The included cohort studies all possessed complete baseline data (Table 4). They were evaluated using the Newcastle-Ottawa Scale (Table 5) [54]. During the follow-up time following medical abortion, approximately 5.9% (7,399 patients) experienced medical abortion failure. The main baseline characteristics of the derivation cohort are presented in Table 4. According to the Newcastle-Ottawa scale, nine included studies were all deemed to be of high quality (Table 5). Seven categories of risk factors were gathered, including gestational age (GA), maternal age (MA), parity (PA) and its type, previous termination of pregnancy (PT) and its type, marital status (MS), type of residence (TR), and differences between gestational age calculated using the last menstrual period and that measured via ultrasound (LU). The details of the included studies and corresponding cohorts are provided in Tables 1–5.
3.1.2 Validation cohort.
A total of 518 patients who had undergone EMAs and were admitted to Zhongshan Hospital were included in our study; however, 85 patients lost to follow-up and 31 patients with incomplete clinical data were excluded. Ultimately, 402 patients were included in the validation cohort (Fig 1B). Among the 402 patients who had undergone EMAs, 48 cases (11.9%) experienced early medical abortion failure and 354 cases (88.1%) experienced early medical abortion success. The mean maternal age with the standard deviation (SD) was 27.7±19.1 years, while the median gestational age was 6 0/7 weeks (interquartile range [IQR] = 5 2/7 to 6 2/7 weeks). Among the patients, 30% had exclusively undergo vaginal deliveries with spontaneous placental delivery, and 14% had experienced one or more caesarean sections. About 29% had undergone one or more previous surgical abortions without exact information about gestational age, and 11% had only experienced medical abortions. Additionally, 41% of the patients were married and 32% were from rural areas. The detailed information of the validation cohort patients is provided in S5 Appendix.
3.2 Model derivation
The 7 categories of clinical risk factors identified in the systematic review and meta-analysis were included in the risk assessment model. All related 22 risk factors in our model were as follows: gestational age of 42–48 days (OR = 1.67, 95%CI 1.53–1.82), gestational age of 49–55 days (OR = 2.35, 95%CI 2.15–2.57), gestational age of 56–62 days (OR = 3.22, 95%CI 2.92–3.55), gestational age of 9–12 weeks (OR = 6.47, 95%CI 5.06–8.26), maternal age of 20–24 years (OR = 1.31, 95%CI 1.17–1.47), maternal age of 25–29 years (OR = 1.52, 95%CI 1.35–1.71), maternal age of 30–34 years (OR = 1.67, 95%CI 1.48–1.89), maternal age of 35–39 years (OR = 1.47, 95%CI 1.30–1.67), maternal age of 40–49 years (OR = 1.16, 95%CI 1.00–1.35), parity (PA, OR = 2.60, 95%CI 2.01–3.37), only vaginal deliveries and spontaneous delivery of placenta (the placenta was delivered spontaneously without the need of manual removal at the third stage of labor) (VD, OR = 1.08, 95%CI 1.00–1.61), ≥1 caesarean section (CS, OR = 1.48, 95%CI 1.33–1.64), ≥1 manual removal of placenta (MR, OR = 2.16, 95%CI 1.75–2.67), previous termination of pregnancy (PT, OR = 1.65, 95%CI 1.20–2.27), previous medical abortions (PM, OR = 0.84, 95%CI 0.78–0.91), ≥1 previous surgical abortion: ≥56 days of gestation (PS1, OR = 1.17, 95%CI 1.08–1.27), ≥1 previous surgical abortion: <56 days of gestation (PS2, OR = 1.53, 95%CI 1.35–1.74), ≥2 previous surgical abortion: both <56 and ≥56 days of gestation (PS3, OR = 1.64, 95%CI 1.41–1.90), married or cohabiting (MC, OR = 1.22, 95%CI 1.04–1.42), rural areas (RU, OR = 1.39, 95%CI 1.16–1.68), and differences between gestational age calculated using the last menstrual period and that measured via ultrasound (LU, OR = 1.24, 95%CI 1.01–1.51). The forest plot of risk factors is shown in Fig 2, and the meta-analysis results are presented in Figs 4 and 5. The ORs (95% CI), β-coefficients, and risk scores of the included risk factors are presented in Table 6. The mathematical formula of the risk prediction model is as follows:
3.3 Model validation
The AUC value of the validation cohort was 0.857 (95% CI 0.804–0.910), and the ROC curve is showed in Fig 6A. The optimal cutoff risk score was determined to be 13.5 points, based on the maximum Youden index (0.588), yielding a sensitivity of 83.3% and specificity of 75.4%. The sensitivity, specificity, and Youden indexes of different critical risk scores are provided in Table 7. The frequency distribution histogram of the success and failure groups is presented in Fig 6B. According to the optimal cutoff value and the frequency distribution, a total of 402 patients who had undergone EMAs were stratified into three groups: low-risk (n = 275), moderate-risk (n = 88) and high-risk (n = 39). The corresponding risk scores were ≤13, 14–19 and ≥20, respectively. Compared with the low-risk group (3%), the failure rates of EMAs in the moderate-risk group and high-risk group were 23% and 51% respectively (Fig 6C).
(A) The receiver operator characteristic (ROC) curve analysis for predicting the failure of EMA. The area under the curve (AUC) was 0.857 (95% CI 0.804–0.910, P < 0.0001). (B) Frequency distribution histogram of the success and failure groups of EMA. (C) Failure and success rate of EMA in three risk groups stratified by risk score in the validation cohort. Low risk: ≤13 points. Moderate risk: 14–19 points. High risk: ≥20 points. Brown arrow: the best cutoff value (13.5) with the highest Youden index (0.588).
In addition, the calibration plot of the prediction model in the validation cohort revealed the goodness of fit of the predicted risk and observed probability of EMA failure (Fig 7). Based on the risk threshold of high-risk group (≥20 points), the OE ratio was calculated as 1.231 (95%CI 0.904–1.558), which indicated that this model might exhibit a tendency towards underprediction. Moreover, the decision curve analysis showed that this model yielded a net benefit when the risk threshold of EMA failure was higher than 0.6 (Fig 8).
The long-dotted line represents a perfect prediction, and the solid line represent the predictive performance of the model. The closer the solid line fits the long-dotted line, the better the predictive accuracy of the model.
The red line represents the risk prediction model. The green line represents the intervention for all patients. The blue line represents the intervention for none patients. When the risk threshold of EMA failure is higher than 0.6, this model yields a net benefit.
In subgroup analysis, the ROC curve analyses of 7 clinical risk parameters were conducted within the validation cohort (Fig 9). All the AUC values of three MA subgroups were more than 0.750. The AUC values of one GA subgroup (8–12 weeks) was 0.640 (95%CI 0.417–0.862), and the other two subgroups exhibited that the AUC values were > 0.750. Three PA subgroups, two MS subgroups, two TR subgroups and two LU subgroups all exhibited that the AUC values were more than 0.800. All the AUC values of three PT subgroups were more than 0.740 (Table 8).
(A) The receiver operator characteristic (ROC) curve analysis of three MA subgroups. (B) The ROC curve analysis of three GA subgroups. (C) The ROC curve analysis of three PA subgroups. (D) The ROC curve analysis of three PT subgroups. (E) The ROC curve analysis of two MS subgroups. (F) The ROC curve analysis of two TR subgroups. (G) The ROC curve analysis of two LU subgroups. (H) The ROC curve analysis of the risk assessment model.
3.4 Risk prediction scale
According to the risk prediction model, a simple risk scale was also developed (points): GA (days: 28–41 = 0, 42–48 = 5, 49–55 = 9, 56–62 = 12, weeks: 9–12 = 19), MA (years: 15–19 = 0, 20–24 = 3, 25–29 = 4, 30–34 = 5, 35–39 = 4, 40–49 = 1), PA (times: 0 = 0, ≥1 = 10; types: VD = 1, CS = 4, MR = 8), PT (times: 0 = 0, ≥1 = 5; types: PM = -2, PS1 = 2, PS2 = 4, PS3 = 5), MS (types: single = 0, MC = 2), TP (types: urban = 0, RU = 3), LU (weeks: <1 = 0, ≥1 = 2). This model was recommended to be used for patients aged 15–49 years in order to predict the risk of EMA failure before undergoing EMAs (Table 9). The mathematical formula of the risk prediction scale is as follows:
4. Discussion
To date, an effective risk assessment model for clinical prediction of EMA failure has not been established by previous studies [55]. To the best of our knowledge, this is the second risk prediction model with a relatively good performance [12].
The first prediction model was developed and validated by Meaidi et al. in 2019, based on a high-quality cohort of more than 80 thousand EMA patients, which included 6 clinical parameters (GA, MA, PA, PM, PS and calendar time) with an AUC of 0.63 [12]. Based on the solid foundation established by Meaidi, Heikinheimo and other researchers, we developed and validated a more comprehensive model with 3 additional clinical parameters (MS, TP, and LU), which exhibited a good discrimination. Although the risk of surgical intervention following medical abortions decreased over time from 2008 to 2012 [36], our risk prediction model cannot incorporate this factor due to its current inapplicability. In addition, our risk assessment model was refined into a simplified risk score scale based on rigorous mathematical principles, facilitating its integration into clinical practices for the convenience of clinicians [56].
Moreover, as the study conducted by Meaidi et al. did not provide an assessment of the overall risk ratios for PA or PT [12], we performed a meta-analysis to extract and combine the OR values from high-quality cohort studies [11, 14, 15, 17–19]. Given the frequent lack of precise information about PA (number and type) and PT (type and gestational age) due to the forgetfulness or other factors, our pooled findings on PA and PT can offer doctors a more convenient reference for real-world clinical applications.
According to a high-quality cohort study comprising over 20,000 patients undergoing medical abortions, Heikinheimo et al. (2011) identified single status and residing in rural areas as risk factors associated with medical abortion failure [13]. Although the gestational age of patients was less than 20 weeks, a majority of patients (at least 86%) had undergone EMAs (≤12 weeks) [13]. Consequently, our risk prediction model assimilated these risk factors and transformed them into parameters [50].
In 2023, Gluck et al. reported that the differences between gestational age calculated using the last menstrual period (LMP) and that measured via ultrasound (US) were correlated with EMA failure [16]. Despite the inherent limitations of the cohort study design, a multivariate regression analysis revealed a direct association between differences in LMP-US and the risk of EMA failure [16]. Therefore, this risk factor was also incorporated into our predictive model.
In addition, there were no unified clear definitions of success or failure of medical abortions (Table 2) [44–47]. Winikoff et al. proposed that the definition of failures in medical abortion should encompass any surgical intervention performed for any reason [44]. Fiala et al. concluded that success may be defined as the termination of pregnancy without requiring additional intervention, while failure may encompass ongoing viable pregnancy, the necessity for additional treatment, or expectant management [45]. Zwerling et al. defined a successful abortion as the complete expulsion of the pregnancy through the intended medical method without any additional intervention, and they also defined a failed abortion as the counterpart to their defined successful abortion, for the purpose of their guideline [46]. Kruse et al. considered that the occurrence of medical abortion failure could be defined as the need for a surgical evacuation to complete the abortion for any reason [47]. According to the aforementioned definitions, we defined a successful abortion as complete expulsion without requiring surgical intervention, while the EMA failure was defined as the requirement for surgical evacuation to complete the abortion due to any reason.
Some important confounders were also considered to have a significant impact on the occurrence of EMA failure, such as socioeconomic factors, geographic region-specific practices, and access to healthcare [11–19, 28–33, 37–41]. Several studies reported that the occurrence rates of EMA failure among developed European countries (2–9%) were lower than those in developing countries/regions (9–11%) [11–19]. This exhibited that socioeconomic factors may significantly affect the failure of EMAs. According to the specified clinical guidelines, medical abortion drugs, dosages, and administration methods were different in different countries [37–41]. Recently, some studies elucidated that the incidence of EMA failure was apparently affected by these specific medical practices [28–33]. In 2011, Heikinheimo et al. identified rural areas as a risk factor associated with medical abortion failure, which may reveal the impact of access to healthcare [13].
There are three key strengths in our study. First, this is the second risk assessment model for clinically predicting the risk of EMA failure, with a good discrimination (AUC = 0.857). Second, this study conducts a more comprehensive analysis of the available clinical risk factors, surpassing previous studies in terms of summarization. Third, the risk score scale (Table 9) can be readily employed by clinicians to assess the risk of EMA failure in clinical settings.
Conversely, there are several limitations. Firstly, the presence of heterogeneity among the included studies may impact the reliability of the risk prediction model to a certain extent. Secondly, the validated cohort in this study had a relatively smaller sample size compared to the derivation cohort, which may limit the generalizability of the results across different populations and healthcare settings. Thirdly, the participants of our validation cohort were from a single-center tertiary hospital. Despite these limitations, the risk assessment model derived and validated from this study still potentially provides an effective clinical prediction method for evaluating the risk of EMA failure prior to undergoing medical abortions.
It is recommended that EMAs could be considered for pregnant women classified as low-risk (≤13 points) or moderate-risk (14–19 points). Furthermore, pregnant women in the moderate-risk group should receive comprehensive information regarding the potential occurrence of EMA failure, and undergo close monitoring. Although medical abortion is not recommended for patients in the high-risk group (≥20 points) due to the high failure rate (51%), it is essential to respect and value the patient’s choice of abortion method. Undoubtedly, this prediction model still necessitates rigorous testing, meticulous refinement, and further research prior to its practical implementation.
5. Conclusion
Based on a systematic review and meta-analysis, this study effectively establishes a simple risk assessment model including seven routinely available clinical parameters for predicting EMA failure. In preliminary validation, this model demonstrates good performance in terms of predictive efficiency, accuracy, calibration, and clinical benefit. However, more research and validation are warranted for future application.
Supporting information
S2 Checklist. Human participants research checklist.
https://doi.org/10.1371/journal.pone.0315025.s002
(PDF)
S3 Checklist. TRIPOD checklist: Prediction model development and validation.
https://doi.org/10.1371/journal.pone.0315025.s003
(DOCX)
S1 Appendix. Search strategy of systematic review.
https://doi.org/10.1371/journal.pone.0315025.s004
(DOCX)
S2 Appendix. Original information of the literature selection process.
https://doi.org/10.1371/journal.pone.0315025.s005
(DOCX)
S3 Appendix. The inclusion criteria, exclusion criteria, data extraction, and quality assessment of systematic review.
https://doi.org/10.1371/journal.pone.0315025.s006
(DOCX)
S4 Appendix. All data extracted from the primary research sources for the systematic review and meta-analysis.
https://doi.org/10.1371/journal.pone.0315025.s007
(DOCX)
S5 Appendix. Original data of the validation cohort patients.
https://doi.org/10.1371/journal.pone.0315025.s008
(DOCX)
S1 Fig. PRISMA flow chart of the literature selection process.
PRISMA, preferred reporting items for systematic reviews and meta-analyses.
https://doi.org/10.1371/journal.pone.0315025.s009
(PPTX)
Acknowledgments
We would like to thank the staff of the library of Xiamen University for providing us with the full texts of some articles. We would also like to thank the staff of the medical record management of Zhongshan Hospital of Xiamen University for their friendly help.
References
- 1. Karim SMM. Once-a-month vaginal administration of prostaglandins E2 and F2α for fertility control. Contraception. 1971;3:173–83. https://doi.org/10.1016/0010-7824(71)90030-8
- 2. Karim S, Rao B, Ratnam S, Prasad R, Wong Y, Ilancheran A. Termination of early pregnancy (menstrual induction) with 16-phenoxy-ω-tetranor PGE2 methylsulfonylamide. Contraception. 1977;16:377–81. https://doi.org/10.1016/0010-7824(77)90049-x pmid:200400
- 3. Goldberg JR, Plescia MG, Anastasio GD. Mifepristone (RU 486): current knowledge and future prospects. Arch Fam Med. 1998;7:219–22. pmid:9596454
- 4. Mahajan DK, London SN. Mifepristone (RU486): a review. Fertil Steril. 1997;68:967–76. pmid:9418681
- 5. Kantorová V, Wheldon MC, Ueffing P, Dasgupta ANZ. Estimating progress towards meeting women’s contraceptive needs in 185 countries: a Bayesian hierarchical modelling study. PLoS Med. 2020;17:e1003026. pmid:32069289
- 6. Bearak J, Popinchalk A, Ganatra B, Moller AB, Tunçalp Ö, Beavin C, et al. Unintended pregnancy and abortion by income, region, and the legal status of abortion: estimates from a comprehensive model for 1990–2019. Lancet Glob Health. 2020;8:e1152–e1161. pmid:32710833
- 7. Kortsmit K, Nguyen AT, Mandel MG, Hollier LM, Ramer S, Rodenhizer J, et al. Abortion Surveillance—United States, 2021. MMWR Surveill Summ. 2023;72:1–29. pmid:37992038
- 8.
Office for Health Improvement and Disparities. Abortion statistics, England and Wales. 2021. Available from: https://www.gov.uk/government/statistics/abortion-statistics-for-england-and-wales-2021/abortion-statistics-england-and-wales-2021
- 9. Ferguson I, Scott H. Systematic Review of the Effectiveness, Safety, and Acceptability of Mifepristone and Misoprostol for Medical Abortion in Low- and Middle-Income Countries. J Obstet Gynaecol Can. 2020;42:1532–1542.e2. pmid:32912726
- 10. Cheng L. Medical abortion in early pregnancy: experience in China. Contraception. 2006;74:61–5. pmid:16781263
- 11. Chien LW, Liu WM, Tzeng CR, Au HK. Effect of previous live birth and prior route of delivery on the outcome of early medical abortion. Obstet Gynecol. 2009;113:669–74. pmid:19300333
- 12. Meaidi A, Friedrich S, Gerds TA, Lidegaard O. Risk factors for surgical intervention of early medical abortion. Am J Obstet Gynecol. 2019;220:478.e1–478.e15. pmid:30763542
- 13. Niinimäki M, Suhonen S, Mentula M, Hemminki E, Heikinheimo O, Gissler M. Comparison of rates of adverse events in adolescent and adult women undergoing medical abortion: population register based study. BMJ. 2011;342:d2111. pmid:21508042
- 14. Ashok PW, Templeton A, Wagaarachchi PT, Flett GM. Factors affecting the outcome of early medical abortion: a review of 4132 consecutive cases. BJOG. 2002;109:1281–9. pmid:12452467
- 15. Bartley J, Tong S, Everington D, Baird DT. Parity is a major determinant of success rate in medical abortion: a retrospective analysis of 3161 consecutive cases of early medical abortion treated with reduced doses of mifepristone and vaginal gemeprost. Contraception. 2000;62:297–303. pmid:11239616
- 16. Gluck O, Barber E, Friedman M, Feldstein O, Tal O, Grinstein E, et al. Failure Rate of Medical Treatment for Miscarriage Correlated with the Difference between Gestational Age According to Last Menstrual Period and Gestational Size Calculated via Ultrasound. J Clin Med. 2023;12:6112. pmid:37834756
- 17. Lefebvre P, Cotte M, Monniez N, Norel G. The role of parity in medical abortion up to 49 days of amenorrhoea. Eur J Contracept Reprod Health Care. 2008;13:404–11. pmid:19117257
- 18. Niinimäki M, Martikainen H, Talvensaari-Mattila A. Increased risk for medical abortion failure for multiparous women. Int J Gynaecol Obstet. 2004;87:174–5. pmid:15491574
- 19. Reeves MF, Monmaney JA, Creinin MD. Predictors of uterine evacuation following early medical abortion with mifepristone and misoprostol. Contraception. 2016;93:119–25. pmid:26285178
- 20. Jiang WX, He FF, Shen Q, Tao XJ, Zhao CC, Shen ZJ, et al. Factors related to completeness of medical abortion with mifepristone and misoprostol. J Chin Med Assoc. 2017;80:96–102. pmid:27591155
- 21. Ma C, Feng W, Han W, Lu Y, Liu W, Sui Y, et al. Elevated mRNA expression of PGF2α receptor splice variant 2(FP-V2) in human decidua is associated with incomplete mifepristone-misoprostol-induced early medical abortion by regulation of interleukin-8. J Matern Fetal Neonatal Med. 2016;29:3472–7. pmid:26744271
- 22. Wang N, Zhao H, Han W, He B, Zhang S, Wang J. Association of the G2014G genotype in estrogen receptor 1 gene with failure of the mifepristone-induced termination of early pregnancy. Tohoku J Exp Med. 2010;220:77–82. pmid:20046055
- 23. El-Baradie SMY, El-Said MH, Ragab WS, Elssery KM, Mahmoud M. Endometrial thickness and serum beta-hCG as predictors of the effectiveness of oral misoprostol in early pregnancy failure. J Obstet Gynaecol Can. 2008;30:877–81. pmid:19038070
- 24. Haimov-Kochman R, Arbel R, Sciaky-Tamir Y, Brzezinski A, Laufer N, Yagel S. Risk factors for unsuccessful medical abortion with mifepristone and misoprostol. Acta Obstet Gynecol Scand. 2007;86:462–6. pmid:17486469
- 25. Rørbye C, Nørgaard M, Nilas L. Prediction of late failure after medical abortion from serial beta-hCG measurements and ultrasonography. Hum Reprod. 2004;19:85–9. pmid:14688162
- 26. Schreiber CA, Ratcliffe SJ, Quinley KE, Miller C, Sammel MD. Serum biomarkers may help predict successful misoprostol management of early pregnancy failure. Reprod Biol. 2015;15:79–85. pmid:26051455
- 27. Li F, Shou Y, Zhu R, Chen X, Shi B. Predictive value of peripheral blood α1-acid glycoprotein in medical abortion outcomes with mifepristone and relativity of concentration. Int J Gynaecol Obstet. 2022;158:201–4. pmid:34561864
- 28. Abubeker FA, Lavelanet A, Rodriguez MI, Kim C. Medical termination for pregnancy in early first trimester (≤ 63 days) using combination of mifepristone and misoprostol or misoprostol alone: a systematic review. BMC Womens Health. 2020;20:142. pmid:32635921
- 29. Shimels T, Getnet M, Shafie M, Belay L. Comparison of mifepristone plus misoprostol with misoprostol alone for first trimester medical abortion: A systematic review and meta-analysis. Front Glob Womens Health. 2023;4:1112392. pmid:36970118
- 30. Kulier R, Kapp N, Gülmezoglu AM, Hofmeyr GJ, Cheng L, Campana A. Medical methods for first trimester abortion. Cochrane Database Syst Rev. 2011;2011:CD002855. pmid:22071804
- 31. Zhang J, Zhou K, Shan D, Luo X. Medical methods for first trimester abortion. Cochrane Database Syst Rev. 2022;5:CD002855. pmid:35608608
- 32. WHO Task Force on Post-ovulatory Methods of Fertility Regulation; Special Programme of Research, Development and Research Training; WHO. Comparison of two doses of mifepristone in combination with misoprostol for early medical abortion: a randomised trial. BJOG. 2000;107:524–30. pmid:10759273
- 33. Schaff EA, Fielding SL, Westhoff C, Ellertson C, Eisinger SH, Stadalius LS, et al. Vaginal misoprostol administered 1, 2, or 3 days after mifepristone for early medical abortion: A randomized trial. JAMA. 2000;284:1948–53. pmid:11035891
- 34. Rodriguez MI, Edelman A, Hersh A, Gartoulla P, Henderson J. Medical abortion offered in pharmacy versus clinic-based settings. Cochrane Database Syst Rev. 2021;6:CD013566. pmid:34114643
- 35. Schmidt-Hansen M, Pandey A, Lohr PA, Nevill M, Taylor P, Hasler E, et al. Expulsion at home for early medical abortion: A systematic review with meta-analyses. Acta Obstet Gynecol Scand. 2021;100:727–35. pmid:33063314
- 36. Meaidi A, Heimustovu BH, Nguyen H, Nguyen T. Declining risk of surgical intervention following early medical abortion: A time trend analysis. Contraception. 2021;104:492–5. pmid:33979650
- 37.
Chinese Medical Association Family Planning Branch. Clinical Diagnosis and Treatment Guidelines and Technical Operating Standards (Family Planning Edition). Beijing: People’s Health Press; 2017.
- 38. Practice Bulletins—Gynecology, the Society of Family Planning. Medication Abortion Up to 70 Days of Gestation. Contraception. 2020;102:225–36. pmid:32807380
- 39.
The Royal College of Obstetricians and Gynaecologists. Best practice in abortion care. RCOG Best Practice Papers. 2022. Available from: https://www.rcog.org.uk/guidance/browse-all-guidance/best-practice-papers/best-practice-in-abortion-care-best-practice-paper/
- 40. Vayssière C, Gaudineau A, Attali L, Bettahar K, Eyraud S, Faucher P, et al. Elective abortion: Clinical practice guidelines from the French College of Gynecologists and Obstetricians (CNGOF). Eur J Obstet Gynecol Reprod Biol. 2018;222:95–101. pmid:29408754
- 41. Costescu D, Guilbert E, Bernardin J, Black A, Dunn S, Fitzsimmons B, et al. Medical Abortion. J Obstet Gynaecol Can. 2016;38:366–89. pmid:27208607
- 42. Collins GS, Reitsma JB, Altman DG, Moons KG. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. Br J Cancer. 2015;112:251–9. pmid:25562432
- 43. Guyatt GH, Oxman AD, Schünemann HJ, Tugwell P, Knottnerus A. GRADE guidelines: a new series of articles in the Journal of Clinical Epidemiology. J Clin Epidemiol. 2011;64:380–2. pmid:21185693
- 44. Winikoff B, Ellertson C, Clark S. Analysis of failure in medical abortion. Contraception. 1996;54:323–7. pmid:8968659
- 45. Fiala C, Cameron S, Bombas T, Parachini M, Agostini A, Lertxundi R, et al. Outcome of first trimester medical termination of pregnancy: definitions and management. Eur J Contracept Reprod Health Care. 2018;23:451–7. pmid:30600730
- 46. Zwerling B, Edelman A, Jackson A, Burke A, Prabhu M. Society of Family Planning Clinical Recommendation: Medication abortion between 14 0/7 and 27 6/7 weeks of gestation: Jointly developed with the Society for Maternal-Fetal Medicine. Am J Obstet Gynecol. 2023:S0002-9378(23)00726-3. pmid:37821258
- 47. Kruse B, Poppema S, Creinin MD, Paul M. Management of side effects and complications in medical abortion. Am J Obstet Gynecol. 2000;183:S65–75. pmid:10944371
- 48. Harris JK. Primer on binary logistic regression. Fam Med Community Health. 2021;9:e001290. pmid:34952854
- 49. Chen D, Wang M, Shang X, Liu X, Liu X, Ge T, et al. Development and Validation of an Incidence Risk Prediction Model for Early Foot Ulcer in Diabetes Based on a High Evidence Systematic Review and Meta-Analysis. Diabetes Res Clin Pract. 2021;180:109040. pmid:34500005
- 50. Ren Q, Chen D, Liu X, Yang R, Yuan L, Ding M, et al. Derivation and Validation of a Prediction Model of End-Stage Renal Disease in Patients with Type 2 Diabetes Based on a Systematic Review and Meta-analysis. Front Endocrinol (Lausanne). 2022;13:825950. pmid:35360073
- 51. Glümer C, Carstensen B, Sandbaek A, Lauritzen T, Jørgensen T, Borch-Johnsen K, et al. A Danish Diabetes Risk Score for Targeted Screening: The Inter99 Study. Diabetes Care. 2004;27:727–33. pmid:14988293
- 52. Riley RD, Ensor J, Snell KIE, Harrell FE Jr, Martin GP, Reitsma JB, et al. Calculating the sample size required for developing a clinical prediction model. BMJ. 2020;368:m441. pmid:32188600
- 53. Efthimiou O, Seo M, Chalkou K, Debray T, Egger M, Salanti G. Developing clinical prediction models: a step-by-step guide. BMJ. 2024;386:e078276. pmid:39227063
- 54.
Ottawa Hospital Research Institute 2021. The Newcastle-Ottawa Scale (NOS) for assessing the quality of nonrandomized studies in meta-analyses. 2023. Available from: https://www.ohri.ca/programs/clinical_epidemiology/oxford.asp
- 55. American College of Obstetricians and Gynecologists’ Committee on Practice Bulletins—Gynecology, Society of Family Planning. Medication Abortion Up to 70 Days of Gestation: ACOG Practice Bulletin, Number 225. Obstet Gynecol. 2020;136:e31–e47. pmid:32804884
- 56. Cronin M, Dengler N, Krauss ES, Segal A, Wei N, Daly M, et al. Completion of the Updated Caprini Risk Assessment Model (2013 Version). Clin Appl Thromb Hemost. 2019;25:1076029619838052. pmid:30939900