The comparison of evaluative effectiveness between antral follicle count/age ratio and ovarian response prediction index for the ovarian reserve and response functions in infertile women.

Conceptualization: Shan-Jie Zhou. Data curation: Shan-Jie Zhou, Ming-Jia Zhao. Formal analysis: Shan-Jie Zhou, Ming-Jia Zhao. Investigation & Measurement: Shan-Jie Zhou, Ming-Jia Zhao, Cui Li, Xing Su. Methodology & Supervision: Shan-Jie Zhou. Writing – original draft: Shan-Jie Zhou, Ming-Jia Zhao. Writing – review & editing: Shan-Jie Zhou.

Clinical data and variable data of subjects presented in Table 1 , and the boxplots of variables’ frequency distribution presented in Fig. 1 A–F. According the diagnosis of patients, the rates of primary infertility type and secondary type were 37.04% and 62.96%, respectively. The rates of male factor, female factor, and double factor infertility were 7.41%, 25.93%, and 66.66%, respectively. Frequencies of 319 subjects’ clinical data and variable data. The boxplots of variables’ frequency distribution. AFC = antral follicle count; AMH = anti-Müllerian hormone; ORPI = ovarian response prediction index. BMI, OV, serum FSH, LH, E2, TT, and FSH/LH ratio were not significantly different among the 4 subgroups; however, the duration of infertility, AFC, serum AMH levels, AFC/Age ratio, AMH/Age ratio, ORPI, and retrieved oocyte numbers were significantly different among the 4 subgroups. Multiple comparisons showed that the durations of infertility were significantly different between Group 1 and Groups 3 and 4 ( P  = .034, P  = .003). The AFC was significantly different between Group 1 and Groups 3 and 4 ( P  = .000, P  = .000), between Group 2 and Groups 3 and 4 ( P  = .001, P  = .000), and between Group 3 and Group 4 ( P  = .004). AFC decreased gradually as age increased, as presented in Fig. 1 A. The AMH levels were significantly different between Group 1 and Group 4 ( P  = .004) and between Group 3 and Group 4 ( P  = .009). The AMH boxplot is presented in Fig. 1 B. ORPI was significantly different between Group 1 and Groups 2, 3, and 4 ( P  = .005, P  = .000, P  = .000), between Group 2 and Group 4 ( P  = .001), and between Group 3 and Group 4 ( P  = .002). ORPI decreased gradually as age increased, as presented in Fig. 1 C. The retrieved oocyte numbers were significantly different between Group 1 and Groups 2 and 4 ( P  = .046, P  = .006), between Group 2 and Group 3 ( P  = .046), and between Group 3 and Group 4 ( P  = .001). The boxplot of retrieved oocyte numbers is presented in Fig. 1 D. The AFC/Age ratio was significantly different between 4 subgroups (Group 1 versus Group 2: P  = .001; P  = .000 for others). AFC/Age ratio decreased gradually as age increased, as presented in Fig. 1 E. The AMH/Age ratio was significantly different between Group 1 and Groups 2, 3, and 4 ( P  = .002, P  = .000, P  = .000), between Group 2 and Group 4 ( P  = .013), and between Group 3 and Group 4 ( P  = .001). The boxplot of AMH/Age ratio is presented in Fig. 1 F. According to multivariate analysis of variance, there was statistical significance in AFC ( P  = .000), OV ( P  = .001), and AFC/Age ratio ( P  = .002) for evaluating the ovarian reserve. And there was statistical significance in BMI ( P  = .010), FSH ( P  = .028), AFC, AMH, AFC/Age ratio, AMH/Age ratio, and ORPI ( P  = .000 for the latter 5 variables) for forecasting ovarian response. AFC demonstrated positive correlation with the OV ( P  = .009), AMH, AFC/Age ratio, AMH/Age ratio, and ORPI ( P  = .000 for latter 4 variables); however, AFC was negatively correlated with age ( P  = .000) and serum FSH level ( P  = .000). The above-mentioned parameters could reflect ovarian reserve. There were positive correlations between the retrieved oocyte numbers and OV ( P  = .020), LH ( P  = .011), AFC, AMH, AFC/Age ratio, AMH/Age ratio, and ORPI ( P  = .000 for the latter 5 variables); however, a negative correlation existed between the retrieved oocyte numbers and age ( P  = .003). The above-mentioned parameters could reflect ovarian response. There were negative correlations between age and AFC ( P  = .000), AMH ( P  = .018), AFC/Age ratio ( P  = .000), AMH/Age ( P  = .042), ORPI ( P  = .000), and retrieved oocyte numbers ( P  = .003). BMI was inversely correlated with FSH ( P  = .015) and FSH/LH ratio ( P  = .010) but did not correlate with AMH or AFC. In general, AFC < 5 was one of the standards indicating decline of ovarian reserve. [ 10 ] Using the abovementioned standards and the ROC curve to evaluate the significant variables of ovarian reserve decrease, the variables included AFC/Age ratio, AMH/Age ratio, age, OV, FSH, and ORPI, the corresponding AUC was 1.000, 0.641, 0.686, 0.641, 0.652, and 0.908, respectively. Although the AUC of AMH was 0.613, it was less valuable to access ovarian reserve ( P  = .0594). AUC, cut-off values, sensitivity, and specificity of variables evaluating ovarian reserve are presented in Table 2 and Fig. 2 . The AUC and cut-off values of variables evaluating ovarian reserve and response. ROC curve of variables evaluating ovarian reserve decrease. AFC = antral follicle count; AMH = anti-Müllerian hormone; FSH = follicle-stimulating hormone; ORPI = ovarian response prediction index; OV = ovarian volume (total); ROC curve = receiver operating characteristic curve. According to our results, ovarian reserve would decrease when subjects’ age and FSH were more than the cut-off values, moreover, the OV, AFC/Age ratio, AMH/Age ratio, and ORPI were less than the cut-off values. According to the retrieved oocyte numbers, ovarian response categories were divided into poor response (collecting oocytes ≤3), normal response (4–14 collecting oocytes) and excessive response (collecting oocytes ≥15). [ 9 , 10 ] We used the abovementioned standards and the ROC curve to forecast the significant variables of excessive ovarian response, which included age, AFC, AMH, AFC/Age ratio, AMH/Age ratio, FSH, FSH/LH ratio, and ORPI, and to forecast the significant parameters of poor ovarian response, which included AMH, AFC/Age ratio, AMH/Age ratio, FSH/LH ratio, LH, and OV. The AUC, cut-off values, and corresponding sensitivity and specificity of variables evaluating ovarian response are presented in Table 2 , Figs. 3 A–B and 4 A–B. ROC curve of variables forecasting excessive ovarian response. AFC = antral follicle count; AMH = anti-Müllerian hormone; FSH = follicle-stimulating hormone; LH = luteinizing hormone; ORPI = ovarian response prediction index; ROC curve = receiver operating characteristic curve. ROC curve of variables forecasting poor ovarian response. AFC = antral follicle count; AMH = anti-Müllerian hormone; FSH = follicle-stimulating hormone; LH = luteinizing hormone; OV = ovarian volume (total); ROC curve = receiver operating characteristic curve.

We retrospectively enrolled 319 consecutive infertile women who had undergone IVF/intracytoplasmic sperm injection (ICSI) treatments from September 2016 to August 2017 in our fertility center. The infertile women aged 21 to 45 years had experienced infertility lasting 1 to 18 years, all subjects experienced at least 1 year without pregnancy success with natural attempts and were the first time to have IVF treatment, and administrated gonadotropin-releasing hormone (GnRH) antagonist protocol to perform ovulation induction. Patient history and clinical information were obtained from their medical records. All of the women were determined to have both ovaries present, no history of ovarian surgery, no severe endometriosis, and no evidence of endocrine disorders. Conventional IVF and/or ICSI were performed according to the cause of infertility. Infertile women were excluded if they were using fertility drugs (e.g., clomiphene, letrozole, and gonadotropin) or had any history of autoimmune and genetic disease, or iatrogenic conditions (e.g., radiation therapy or pelvic surgery), as these factors have been shown to alter the serum reproductive hormone and AMH levels. Patients were stratified into the following age groups: 21 to 29 years (Group 1), 30 to 34 years (Group 2), 35 to 39 years (Group 3), and 40 to 45 years (Group 4). Blood samples were obtained by venipuncture at 7:30 AM to 10:00 AM, and serum samples were measured together in the clinical laboratory of Peking University International Hospital. And the basal serum FSH, LH, estradiol (E2), total testosterone (TT) levels, and serum AMH levels were tested simultaneously on spontaneous cycle days 2–4 and prior to the beginning of IVF/ICSI cycles. The FSH, LH, E2, TT (kit from Abbott Ireland Diagnostics Division Lisnamuck, Longford Co., Longford, Ireland), and AMH (kit from Roche Diagnostics GmbH, Mannheim, Germany) levels were determined with commercial kits and an electrochemiluminescence immunoassay. The lower limits of the FSH, LH, E2, TT, and AMH levels were ≤10 pg/mL, 0.15 nmol/L, <0.05 IU/L, and ≤0.5 IU/L, and 0.010 ng/mL, respectively. The intra-assay coefficients of variation (CVs) for FSH, LH, E2, TT, and AMH were 4.6%, 5.1%, 6.7%, 2.3%, and 4.0%, respectively. The mean inter-assay CVs for FSH, LH, E2, TT, and AMH were 5.6%, 8.1%, 3.7%, 5.5%, and 2.7%, respectively. The experienced and qualified sonographers performed ultrasonographic evaluations for all subjects during spontaneous cycle days 2–4 using a two-dimensional transvaginal probe of 9 MHz frequency (HD11 XE, Philips Ultrasound, Inc., Bothell, WA). The total number of 2 to 9 mm antral follicles in both ovaries was measured and recorded. OV was calculated as the volume of an ellipsoid, that is, 0.52 × Length × Width × Depth. The total basal volume of both ovaries was evaluated in each patient. Height and weight were measured by the investigators of our team and were used to calculate BMI, that is, weight (kg)/height (m 2 ). Serum FSH, LH, and AMH concentration, AFC, and patient age were used to calculate FSH/LH ratio, AFC/Age ratio, AMH/Age ratio, and ORPI. The ORPI was defined by the following equation: ORPI = [AMH (ng/mL) × AFC (number)]/Patient age (years). [ 9 ] The ovarian stimulation protocols performed the GnRH antagonist protocols. The gonadotropin (Gn) was recombinant FSH (Gonal-f, Laboratoires Serono SA, Aubonne, Switzerland) and human menopausal gonadotrophin for injection (Livzon Pharmaceutical Group Inc. Zhuhai, Guangdong Province, China). The antagonist was Ganirelix Injection (ORGALUTRAN, Ravensburg, Germany). After the third day of treatment, the Gn dose was adjusted to the patient response. The total Gn dose was 2076.72 ± 963.33 IU, and the Gn treatment duration was 9.65 ± 2.74 days in each IVF/ICSI cycle. The trigger drug was recombinant human chorionic gonadotropin alpha for injection (OVIDREL 250 μg, Laboratoires Serono S.A. [LSA], Aubonne, Switzerland). The ovarian reserve was defined as women possessed the reproductive potential or the potential of oocyte yield at reproductive stage, and the biomarkers of ovarian reserve were being promoted as potential markers of reproductive potential or “fertility tests.” According to the criterion for the classification, [ 9 , 10 ] the poor ovarian response was defined as collecting ≤3 oocytes, and the excessive ovarian response was defined as collecting ≥15 oocytes after the ovarian stimulation protocol. Microsoft Excel 2013 software (Microsoft Corporation, Redmond, WA), SPSS 21.0 (IBM Corporation, New York, NY), and MedCalc Statistical Software version 18.2.1 (MedCalc Software bvba, Ostend, Belgium; http://www.medcalc.org ; 2018) were used for all statistical analysis. The data were presented as the mean ± standard deviation (SD), as calculated for all subjects and each group. The one-way analysis of variance (ANOVA) test, multiple comparisons and multivariate analysis of variance were used to assess the differences between the mean values of parameters in the different groups. Pearson correlation analysis was used to assess the correlations between different parameters. Receiver operating characteristic curve (ROC curve) analysis was used to analyze the predictive accuracy of variables, and to calculate the area under the curve (AUC), and the cut-off values and corresponding sensitivity and specificity. The Z test was used to assess the differences between the AUC of different parameters. Tests were considered statistically significant if P  <   .05. This study and the accompanying consent forms were approved by the Ethics Committee and Institutional Review Board of Peking University International Hospital, and Tangshan Maternity and Child Healthcare Hospital. Participants were enrolled in the study after written informed consent was obtained.

Ovarian reserve can be altered or reduced due to age, disease, pelvic operation, chemotherapy, radiotherapy, and other risk factors, and it is beneficial to create treatment regimes, survey treatment effects, and forecast prognoses for infertile women by evaluating ovarian reserve or response. Our results indicated that the AFC, serum AMH levels, AFC/Age ratio, AMH/Age ratio, ORPI, and retrieved oocyte numbers were significantly different among the 4 subgroups and that there were negative correlations between age and AFC, AMH, AFC/Age ratio, AMH/Age ratio, ORPI, retrieved oocyte numbers. Moreover, we discovered that AFC, AFC/Age ratio, and ORPI decreased gradually with age. Lee et al [ 11 ] and Raeissi et al [ 12 ] observed increased FSH levels and decreased AMH levels with increasing age in women. Bozkurt et al [ 13 ] reported that AMH was inversely correlated with age; however, AFC revealed a stronger correlation with age in both the fertile and infertile populations compared with basal FSH and AMH; the decrease in ovarian reserve in infertile patients was directly related to age, not infertility. AFC on day 2–4 of the menstrual cycle, evaluated by transvaginal ultrasound, is commonly used to determine ovarian reserve, but AFC measurement is prone to error because of different sonographers. To explore the evaluative effectiveness of various variables, we analyzed the correlation of AFC with other variables and found that AFC showed a positive correlation with OV, AMH, AFC/Age ratio, AMH/Age ratio, and ORPI; AFC showed a negative correlation with age and serum FSH level. However, Somigliana et al [ 14 ] insisted that low serum AMH is not associated with female subfertility. The literature has reported that BMI is not associated with the AMH levels in the general population of infertile women or in patients without polycystic ovary syndrome (PCOS); however, BMI is significantly and inversely correlated with AMH in women with PCOS. [ 15 ] Another study found that age is negatively correlated with AMH and AFC across all races ( P  < .05) and that elevated BMI is negatively correlated with AMH in Caucasian women but not in African-American, Hispanic, or Asian women. [ 16 ] The results from our research showed that BMI was inversely correlated with FSH and FSH/LH ratio but did not correlate with AMH or AFC. In brief, debate still exists regarding the influence of BMI on ovarian function and AMH levels. The ROC analysis results in our study revealed that the significant variables for evaluating ovarian reserve decrease included age, OV, FSH, AFC/Age ratio, AMH/Age ratio, and ORPI. Moreover, the AUC of AFC/Age ratio and ORPI was higher than those of the other 4 variables, and the diagnostic accuracy reached a “high” grade; the cut-off values of AFC/Age ratio and ORPI from ROC analysis were 0.111 (sensitivity 100.00%, specificity 100.00%) and 0.239 (sensitivity 90.20%, specificity 75.00%). Interestingly, the evaluative effectiveness of AFC/Age ratio exceeded that of ORPI (AUC: 1.000 vs 0.908, P  = .0007). Assessments of the AMH and FSH levels in combination with female age could be helpful in predicting ovarian reserve in infertile women. [ 12 ] Many studies [ 17 – 21 ] have discussed how AFC and AMH could be used to assess ovarian reserve. However, our research did not find that these 2 variables acted as a single variable to significantly assess ovarian function. Serum AMH and AFC begin to decline in women between 34 and 35 years old, and AMH predicts biological age earlier than FSH or AFC do, and AFC does so earlier than FSH does. [ 17 ] By age 32, over 50% of women with subfertility had AMH levels categorized as “low fertility” (AMH ≤19.5 pmol/liter), and the rate increased to 75% by age 39, with a decrease in mean AMH of 1.72 pmol/L/y. [ 18 ] The serum AMH cut-off value for the normal ovarian reserve was calculated as 0.37 ng/mL (sensitivity 71.43%, specificity 66.67%, positive prediction 83.33%, negative prediction 50%). [ 19 ] AMH should be considered a more reliable ovarian reserve assessment test compared with FSH because there was a strong positive correlation between the serum AMH level and AFC; further, the use of AMH combined with AFC may improve ovarian reserve evaluation. [ 20 ] The present findings suggest the applicability of AMH determination as a marker for actual fertility in subfertile women with elevated basal FSH levels, as AMH was significantly associated with the timing of reproductive stages (i.e., the occurrence of menopausal transition or menopause during follow-up). [ 21 ] Our results showed that the cut-off value of age was 37.00 for predicting ovarian reserve decline and that the corresponding sensitivity and specificity were 52.38% and 80.00%, respectively. We found that the significant variables forecasting excessive ovarian response included age, AFC, AMH, AFC/Age ratio, AMH/Age ratio, ORPI, FSH, and FSH/LH ratio, and that the significant variables forecasting poor ovarian response included AMH, LH, OV, AFC/Age ratio, AMH/Age ratio, and FSH/LH ratio. Interestingly, ORPI and AFC/Age ratio demonstrated better effectiveness in evaluating ovarian response. When used to predict excessive response, the cut-off value of ORPI from ROC analysis was 0.880 (sensitivity 84.72%, specificity 67.32%). When used to predict poor response, the AUC of AFC/Age ratio presented the largest value despite no statistical difference among the 6 variables, the cut-off value of AFC/Age ratio was 0.311 (sensitivity 63.75%, specificity 100.00%). In addition, we found that it was inconsistent on the significant variables of evaluation between multivariate analysis of variance and ROC, such as AFC and BMI. In recent years, many studies have focused on the value of a single parameter. For example, AMH was strongly associated with oocyte yield after ovarian stimulation and may therefore predicted ovarian response and the quality of oocytes and embryos. [ 2 , 19 , 22 ] AMH and AFC had a higher predictive value for the responders than for FSH, E2, and chronological age, moreover, could predict the risk of ovarian hyperstimulation syndrome (OHSS) among patients. [ 23 , 24 ] Vembu and Reddy [ 25 ] reported ROC curve was plotted to predict the hyper response (OHSS), which showed a serum AMH cut-off value of 6.85 ng/mL with a sensitivity of 66.7% and a specificity of 68.7% for PCOS group and 4.85 ng/mL with a sensitivity of 85.7% and a specificity of 89.7% in non-PCOS group. AFC is superior to AMH in predicting poor ovarian response. The cut-off point for mean AMH and AFC in discriminating between poor and normal ovarian response cycles was 0.94 ng/mL (with a sensitivity of 70% and a specificity of 86%) and 5.5 (with a sensitivity of 91% and a specificity of 91%), respectively. [ 26 ] Iranian women with a basal AMH level >6.95 ng/mL are at a high risk of developing OHSS, and those with AMH level <1.65 ng/mL are poor responders. [ 27 ] Our results showed that the AMH cut-off value for excessive ovarian response and poor response was 3.51 ng/mL (sensitivity 63.41%, specificity 74.07%) and 1.40 ng/mL (sensitivity 77.78%, specificity 70.83%), AFC cut-off value for excessive ovarian response was 14.00 (sensitivity 77.78%, specificity 76.87%), respectively. Currently, 2 popular combined indexes, ORPI and FSH/LH ratio, are used to assess ovarian function. Regarding the probability of collecting ≥4 oocytes, ORPI showed an AUC of 0.91 and an efficacy of 88% at a cut-off of 0.2, but for the probability of collecting ≥15 oocytes, ORPI showed an AUC of 0.89 and an efficacy of 82% at a cut-off of 0.9. [ 9 ] The cut-off value reported by that study approximated our results. Oliveira and Franco [ 28 ] reported the ORPI offered excellent ovarian response prediction (AUC = 0.91), and good predictions for the possibility of collecting >4 metaphase II oocytes (AUC = 0.84) and excessive ovarian response (AUC = 0.89) in infertile women, and ORPI value (≥1.7) was the benchmark that indicated high risk for OHSS. Selcuk et al [ 29 ] found that the level of association between the ovarian response tests and poor ovarian response data was (in descending order): ovarian sensitivity index (OSI), ORPI, AFC, AMH, and age (AUC = 0.976, 0.905, 0.899, 0.864, 0.617, respectively), and OSI and ORPI could be superior to other ovarian responsiveness markers for poor and high ovarian responses on cycles with agonist or antagonist protocols. However, ORPI was more convenient than OSI, because OSI could be calculated after informed of the number of retrieved oocytes. In addition, opposing views on ORPI effectiveness continue to exist. Another study showed that both AMH and AFC were good predictors of ovarian response with an AUC >0.75 but that combining these variables was not necessary as ORPI would not improve the prediction value. [ 30 ] Using the cut-off value derived from ROC analysis, cycles with an FSH/LH ratio ≥3 produced fewer mature oocytes (8.25 vs 11.74) and a higher percentage of poor ovarian response cycles (32.5% vs 14.3%). Additionally, the serum FSH level and FSH/LH ratio at the commencement of gonadotropin stimulation were inversely correlated to the number of mature oocytes. [ 6 ] According to our results and previous reports in the literature, the abovementioned combined indexes had excellent performances in evaluating ovarian reserve and response. Some shortcomings still exist in our research. First, there were no comparison data on ovarian function between fertile and infertile women. Second, we did not focus on predicting the influence of stimulation protocols and cycle cancellations. Previous research has shown that an elevated FSH/LH ratio >3 is more likely to result in the cancellation of the individual's cycle (15% vs 5.24%, P  = .0001) and that the total gonadotropin dosage was greater in the higher-ratio group than in lower-ratio group (2636 vs 2242 IU; significant). [ 31 ] Finally, we did not collect data on embryo quality and pregnancy outcome associated with parameters in this research. Several studies have paid close attention to treatment outcomes. An FSH/LH ratio <1.26 is associated with good oocyte parameters, high-quality embryos, and implantation after ICSI. [ 32 ]

Comparing the effectiveness of evaluating ovarian reserve and predicting ovarian response on age, OV, AFC, serum FSH, AMH, AFC/Age ratio, AMH/Age ratio, FSH/LH ratio, and ORPI, we found that AFC/Age ratio was superior to the other parameters in evaluating ovarian reserve, and it offered excellent effectiveness in predicting poor ovarian response, however, ORPI exceeded the other parameters in predicting excessive ovarian response. Consequently, we agreed that the evaluative effectiveness of a combined index exceeded that of a single parameter for evaluating the ovarian reserve and response of infertile women.

Several studies [ 1 – 4 ] have revealed that age, antral follicle count (AFC), and serum anti-Müllerian hormone (AMH) levels reflect the ovarian reserve admittedly, so these factors are considered valuable predictors of the ovarian response to exogenous gonadotrophins. In addition, AMH was a key for ovarian reserve-related outcomes, and has indeed been widely used in clinical practice. Sixty percent of the respondents from 796 infertility clinics worldwide reported using AMH as a first-line test in in vitro fertilization (IVF) cycles, and 54% reported AMH as the best test for evaluating ovarian reserve; 89% reported that AMH results were relevant to clinical practice. [ 1 ] And AMH was an effective measure of quantitative ovarian reserve and was strongly associated with the ovarian response and oocyte yield after ovarian stimulation, [ 2 , 3 ] moreover, using AMH, AFC, and age together constituted a new model for predicting poor or excessive ovarian response. [ 4 ] However, several scholars have considered the evaluative effectiveness of the above parameters to be unsatisfactory and inaccurate for clinical practice. For example, Fleming et al [ 5 ] stated that an ovarian reserve measure without limitations had not yet been discovered, although both AFC and AMH had good predictive value. Furthermore, debate exists regarding whether a single parameter or a combined index (follicle-stimulating hormone/luteinizing hormone ratio, FSH/LH ratio; ovarian response prediction index, ORPI) is a superior tool for evaluating the ovarian reserve or response. The FSH/LH ratio could be used to differentiate between decreased and normal response cycles, and the elevated day-3 FSH/LH ratio was associated with an inferior outcome in IVF treatment cycles. [ 6 – 8 ] Oliveira et al [ 9 ] innovatively used ORPI to assess ovarian response and found that ORPI exhibited an excellent ability to predict poor or excessive ovarian response, a collection of greater than or equal to 4 metaphase II oocytes and the occurrence of pregnancy in infertile women. The objective of this study was to explore and compare the evaluative effectiveness of the abovementioned parameters and to determine which could be used to assess ovarian reserve and response advantageously.

The authors wish to thank all participants and their families for participating in this study. They gratefully acknowledge the help of Dr Ming-Jia Zhao, Dr Cui Li, and Dr Xing Su for the collection of data from Tangshan Maternity and Child Healthcare Hospital. And they also gratefully acknowledge the help of Dr Dian He for the statistical analysis, Dr He was from Department of Epidemiology and Health Statistics, School of Public Health, Capital Medical University, Beijing, China. They also thank Nature Research Editing Service from Springer Nature ( http://www.springernature.com ) for editing this manuscript.