This study was performed following preferred reporting items for PRISMA guidelines (20), and the protocol was registered in PROSPERO (ID: CRD42023493086). A systematic literature search was conducted in multiple databases, including PubMed, Cochrane Library, EMBASE, Web of Science, China National Knowledge Infrastructure (CNKI), and China Biology Medicine Disc, on December 22, 2023. We applied no language restrictions. The complete search used for PubMed was: ((((β-human chorionic gonadotropin) OR (Beta-human chorionic gonadotropin) OR (β-hCG) OR (Beta-hCG))) AND (((assisted reproductive technology) OR (ART) OR (in vitro fertilization) OR (IVF) OR (intracytoplasmic sperm injection) OR (ICSI)))) AND (pregnancy outcomes). We systematically evaluated all potentially eligible studies for inclusion in this meta-analysis, regardless of their primary outcome or language. Additionally, we manually searched the reference lists of each identified primary study to ensure the comprehensive inclusion of all eligible studies.

The inclusion criteria for this study were as follows: (i) prospective or retrospective cohort studies and case-control studies; (ii) the study population comprised individuals who underwent IVF/ICSI cycles, excluding those who received hCG in all luteal phase-support protocols; (iii) measurement of serum β-hCG levels after ET was conducted to predict the occurrence of clinical pregnancy or live birth; (iv) there was sufficient availability of information to construct a 2 × 2 contingency table, which included the true-positive (TP), true-negative (TN), false-positive (FP), and false-negative (FN) test results at specific cutoff values. The exclusion criteria for this study were as follows: research on natural pregnancy, intrauterine insemination, gamete intra fallopian transfer (GIFT), and donor egg cycles. Additionally, studies focusing solely on patients diagnosed with tubal obstructive infertility, polycystic ovary syndrome, autoimmune diseases, or other diseases were excluded. The population in the studies serving as controls for the PGT group was also included in the analysis if they met the aforementioned inclusion and exclusion criteria. Reviews, conference abstracts, case reports, commentaries, experimental animal studies, and studies with insufficient or unavailable data were excluded from the analysis.

Two investigators (WQ and ZR) conducted the literature screening and data extraction, and disagreements were resolved via group discussion. The following data were extracted from the included studies: author, publication time, country, study type, inclusion and exclusion criteria, serum β-hCG assay, stage of transferred embryo, days after ET when β-hCG samples were drawn, cycle type, number of cycles, demographic characteristics of the study population (age and BMI), outcomes, cutoff value, TP, FP, FN, and TN. These data could be obtained directly or calculated indirectly based on the data of the original study. Serum β-hCG concentrations are expressed in mIU/mL (converted to mIU/mL using the conversion formula mIU/mL = 1 IU/L = 1 U/L).

Review Manager (RevMan) 5.4.1 was used to assess the quality of the chosen studies following the quality assessment of diagnostic accuracy studies (QUADAS-2) checklist (21). Two impartial reviewers evaluated the potential bias in each study, with any discrepancies resolved by a third reviewer. Each study’s bias risk was classified as low, high, or unclear as a function of the patient selection, index test, reference standard, and flow and timing.

The statistical calculation was performed using STATA software (version 16.0). The significance level for the statistical analysis of this meta-analysis was set at α < 0.05. After integration, the sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), and diagnostic odds ratio (DOR) were calculated. Calculating the area under the curve (AUC) involved plotting summary receiver operating characteristic curves (SROC) to assess the aggregated diagnostic efficacy of serum β-hCG. Heterogeneity within the study was assessed using Cochrane Q and I² statistics. Fixed effects or random-effects models were used according to the heterogeneity (I² statistic > 50%, random effects models; I² statistic < 50%, fixed effects model). In this analysis, the bivariate random-effects model was employed. A Fagan nomogram was employed to calculate the posterior probability. Deek’s funnel plot was used to assess publication bias among the included studies.

A detailed flowchart outlining the study selection process was presented in Figure 1. We initially identified 1407 records through electronic database searches. After deduplication, 954 unique records were identified for further consideration. During the preliminary screening phase, 789 records were excluded: 41 reviews, conference abstracts, case reports, commentaries, and experimental animal studies, and 748 were unrelated to the research topics. This left 165 studies for full-text review. Then, stringent inclusion and exclusion criteria were applied to ensure the highest relevance and quality of the included studies. Finally, 12 studies were included in the quantitative synthesis (meta-analysis) (8, 16, 22–31).

The attributes of the studies incorporated in the analysis were detailed in Table 1, with each study identified as a retrospective cohort study. Except for one study involving patients who underwent a single cleavage embryo or blastocyst transfer (22), all other studies focused on patients who received a single blastocyst transfer. In the included studies, some have delineated distinct cycle types when investigating the impact of β-hCG on clinical pregnancy or live birth outcomes, differentiating between fresh and frozen embryo cycles. For studies that included data from both fresh and frozen embryo cycles and conducted separate analyses for each, we treated them as two separate study entities within our analysis. Accordingly, we labeled them Study 1 and Study 2 to maintain clarity and distinction. Ultimately, 8 studies evaluated serum β-hCG application in predicting clinical pregnancy, with 2 of these studies encompassing different cycle types, totaling 10 study entities. Nine studies explored serum β-hCG efficacy in predicting live birth, with 2 of these considering different cycle types, resulting in 11 study entities. For detailed information, refer to Tables 2, 3.

Figure 2 showed the QUADAS-2 quality assessment for the 12 included articles. None of the studies fulfilled all the quality criteria. The most common risk of bias was related to patient selection, with two studies (16.7%) exhibiting a high risk of bias, 4 studies (33.3%) presenting an unclear risk of bias, and 6 studies (50.0%) demonstrating a low risk of bias. In the domains of the index test and the reference standard, 4 studies (33.3%) and 2 studies (16.7%), respectively, demonstrated unclear risks of bias. The remaining studies in the two domains showed low risks of bias. Regarding flow and timing, all studies were evaluated as having low risks of bias. In assessing the applicability concerns, two studies (16.7%) exhibited high concern risk, two studies (16.7%) displayed unclear concern risk, and the remaining studies (66.7%) demonstrated low concern risk in patient selection. For the index test, all studies showed a low concern risk. In the reference standard domain, two studies (16.7%) indicated unclear concern risk, while most studies (83.3%) showed low concern risk. Therefore, the included studies demonstrated moderate methodological quality overall.

A threshold effect test was first conducted, with Spearman correlation analysis indicating no threshold effect (r = -0.539, P = 0.108). Given the considerable heterogeneity observed among studies, a random-effects model was utilized to determine the diagnostic metrics of serum β-hCG in predicting clinical pregnancy. The pooled sensitivity and specificity were calculated to be 0.91 (95% confidence interval [CI]: 0.88–0.94) and 0.89 (95% CI: 0.80–0.94), respectively, with the forest plot depicted in Figure 3A. Figure 3B illustrated the pooled DOR, which is 65.07 (95% CI: 30.71–137.86). Furthermore, the SROC curve was plotted, revealing a pooled AUC score of 0.95 (95% CI: 0.93–0.97), as depicted in Figure 3C. These pooled analysis outcomes collectively highlighted a reference informative diagnostic performance of serum β-HCG in forecasting clinical pregnancy.

To explore the potential sources of heterogeneity, subgroup analyses were conducted. Groups were stratified according to the timing of serum β-hCG measurement (9–10 days, 11–13 days, and 14–16 days after ET), cycle type (fresh vs. frozen), and geographic region (China vs. other countries). The results showed no significant heterogeneity between groups based on measurement timing (p = 0.718) (Supplementary Figure 1A) or cycle type (p = 0.209) (Supplementary Figure 1B). Although a statistically significant difference was observed between geographic regions (p < 0.001) (Supplementary Figure 1C), substantial heterogeneity remained within both subgroups under this classification. Taken together, these findings suggested that none of the three grouping factors served as a major source of heterogeneity in this study.

Similarly, Spearman correlation analysis indicated no significant threshold effect for live birth prediction (r = 0.400, P = 0.223). Given the significant variability across studies, the random effects model was again employed to gauge the diagnostic efficacy of serum β-hCG concerning live birth. The pooled sensitivity and specificity were 0.87 (95% CI: 0.80–0.91) and 0.70 (95% CI: 0.65–0.75), respectively, with the corresponding forest plot in Figure 4A. Figure 4B detailed the pooled DOR, which was 15.07 (95% CI: 9.34–24.30). The SROC curve was also constructed, with the pooled AUC depicted in Figure 4C at 0.82 (95% CI: 0.78–0.85). These findings from the pooled analysis indicated that serum β-HCG demonstrates a certain degree of diagnostic capability in predicting live birth.

To explore potential sources of heterogeneity in the prediction of live birth, subgroup analyses were also performed based on the same stratifications. The results showed no significant heterogeneity between groups for measurement timing (p = 0.056) (Supplementary Figure 1D) or cycle type (p = 0.747) (Supplementary Figure 1E). Although a statistically significant difference was observed between geographic regions (p = 0.012) (Supplementary Figure 1F), notable heterogeneity persisted within both subgroups. Therefore, these factors were also not identified as major sources of heterogeneity for live birth prediction.

The potential for publication bias in research on the utility of β-hCG in forecasting clinical pregnancy and live birth was rigorously evaluated using Deek’s funnel plot asymmetry test. Supplementary Figure 2 presented the results, which were notably free from bias. The corresponding P-values for the tests were 0.91 and 0.74, respectively, both comfortably surpassing the threshold of 0.05, highlighting robust integrity in the publication process, with no discernible publication bias detected.

In clinics, physicians expect to utilize initial serum β-hCG levels to forecast pregnancy outcomes after ET, delivering personalized medical advice to enhance the probability of a healthy pregnancy and live birth. This approach also aids in clarifying outcomes for patients and alleviating their uncertainty and anxiety. To elucidate the clinical relevance of our findings, we constructed a Fagan plot to delineate the relationship between pretest probability, likelihood ratio, and post-test probability. Figure 5A illustrated that with a pretest probability of 50%, a positive diagnostic likelihood ratio of 8 elevates the post-test probability to 89%, underscoring the substantial clinical diagnostic utility of serum β-hCG for confirming clinical pregnancy. Similarly, Figure 5B demonstrated that for a pretest probability of 40%, a PLR of 3 results in a post-test probability of 66%, highlighting the serum β-hCG’s value in predicting live birth.

In our analysis, the likelihood ratio scattergram provided insightful distribution patterns for the predictive power of serum β-hCG across different clinical outcomes. When predicting clinical pregnancy, the included studies were predominantly found in the first (left upper, LUQ), second (left lower, LLQ), and fourth (right lower, RLQ) quadrants (Figure 6A). When predicting live birth, most studies concentrated on the RLQ (Figure 6B). These findings indicated that the diagnostic utility of serum β-hCG is restricted in confirming or excluding clinical pregnancy and live birth.