Introduction

About 25% of pregnant women experience bleeding in the early stage, and half of them eventually progress to pregnancy loss. Currently, neither effective treatment nor prevention strategy is available to manage this condition [1,2,3]. Despite some medicines are frequently prescribed in clinical practice including progesterone [4], the latter seems only to benefit early pregnant women with a history of one or more previous miscarriages and vaginal bleeding [5].

The vaginal bleeding during the early pregnancy stage is proven to be strongly associated with miscarriage, which is indeed attributed to luteal insufficiency, leading to low progesterone concentration [1,2,3]. Progesterone, a critical hormone during pregnancy, reflects the luteal function until 7 weeks of gestation, and thus serves as a biomarker to predict the pregnancy outcome in threatened miscarriage [6, 7]. However, the concentration of progesterone varies largely among individuals and across gestational age [8]. Previously, two meta-analyses indicated that both less than 10 ng/mL and 6.3 ng/mL could predict a non-viable pregnancy in early pregnant women including threatened miscarriage [9, 10]. Nevertheless, it is still unclear whether a single progesterone measurement can predict miscarriage in early pregnant women with threatened miscarriage, because most of studies recruited mixed patients including biochemical pregnancy, ectopic pregnancy and threatened miscarriage.

Therefore, we conduct a diagnostic meta-analysis to evaluate the performance of single progesterone measurement predicting miscarriage in early pregnant women with threatened miscarriage.

Methods

Protocol and registration

This meta-analysis was registered on PROSPERO (CRD42021255382) and prepared according to systematic reviews and meta-analyses (PRISMA) guidelines [11].

Literature search

We conduct a comprehensive literature search from Jan 1, 2000 to March 31, 2023, without limitation in language. We searched PubMed, Embase, Scopus, Cochrane Library and China national knowledge infrastructure (CNKI), with the terms of ‘Miscarriage’, Abortion’ and ‘serum Progesterone’. The search strategy used on PubMed-Medline was listed in (Supplement Table 1) in detail. Other databases were tailored when it was necessary, according to the strategy used on PubMed-Medline.

Table 1 Characteristics of individual studies included in the meta-analysis
Full size table

Study selection

Inclusion criteria were patients had an ultrasound-confirmed intrauterine pregnancy with gestational age < 12weeks, presented with vaginal bleeding with or without abdominal pain, measured progesterone concentration, reported threshold of progesterone predicting pregnancy outcomes including miscarriage, designed as observational studies or intervention studies (randomized controlled trials), with sample size over 50. Exclusion criteria were women who planned to terminate the pregnancy or have an inevitable miscarriage, got pregnancy via assisted reproductive technologies, took exogenous progesterone prior study, and without threatened miscarriage.

Data extraction and quality assessment

Two authors (YG and TJ) independently extracted the data from included studies, including the study country, sample size, study design, gestation age, progesterone assessment method, diagnostic threshold of progesterone, exogenous progesterone used during study, true and false positive, true negative and false negative, follow up.

The quality of included studies was evaluated according to the QUADAS-2 standard for quality assessment of diagnostic accuracy [12], including the risk of prejudice and applicability concerns. The risk of bias is consisted of four aspects, namely, patient selection, index testing, reference standards and flow and timing. Applicability assessment included patient selection, index testing, and reference standards.

Outcomes assessment

The main outcomes included cumulative area under receiver operating characteristic (ROC) curve, sensitivity, specificity, positive and negative likelihood ratio (LLR), , and diagnostic odds ratio (DOR).

Statistical analysis

STATA17 software (Stata Corp LLC, College Station, Texas, USA) was employed to conduct the statistical analysis when there were 4 or above available dataset, using the ‘midas’ command [13, 14], otherwise, the analysis was conducted by the Metadisc1.4 software. The area under the ROC curve, LLR and DOR and the 95% confidence interval (CI) were calculated. Publication bias was assessed by the deeks funnel plot asymmetry test. Model fits were assessed by quantile plot of residual based goodness-of fit and chi-squared probability plot of squared mahalanobis distances. Subgroup analysis was conducted, stratifying by the threshold of progesterone, recruited location and region, progesterone measurement method, exogenous progesterone supplement and follow up.

Results

Characteristics of included studies

Overall, 12 studies were included in this study (Fig. 1), and 15 datasets were available to pool the results. The included studies were conducted in India, Singapore, China, and Australia, with a sample size ranging from 76 to 1087. The included patients’ gestational age was between 4 and 12 weeks. The threshold of progesterone reported ranged from 8 to 30ng/ml. Other characteristics of included studies were shown in (Table 1) in detail.

Fig. 1
figure 1

PRISMA flowchart

Full size image

Quality assessment

The methodological quality was assessed by the QUADAS-2 tool (Table 2). One study did not describe exclusion criteria [15]. Two studies reported as retrospective studies without any information in detail [16, 17] and three studies did not report the study design [15, 18, 19]. One study was judged with a high risk of bias and four were unclear. The selection of diagnostic thresholds was not stated prior the study in 5 studies [16, 18,19,20,21] and thus were classified as unclear risk for publication bias. , The outcome assessment was conducted by the ultrasound or follow-up of < 20 weeks of gestation in 9 studies, but live birth in 3 studies [15, 16, 22].  Due to migration, not all patients were included to perform the analysis in 3 studies [22,23,24]. No significant publication bias was detected (Supplement Fig. 1).

Table 2 Quality assessment of included studies
Full size table

Diagnostic performance

The sensitivity, specificity, positive LLR and negative LLR of progesterone in each study were summarized (Fig. 2). The pooled sensitivity (0.69, 95% CI 0.60 to 0.77) and specificity (0.89, 95% CI 0.81 to 0.94) of progesterone predicting miscarriage were relative good, providing positive LLR (6.2, 95% CI 4.0 to 9.7) and negative LLR (0.35, 0.27 to 0.43). The synthesized area under the ROC curve (0.85, 95% CI 0.81 to 0.88) had similar performance (Fig. 3).

Fig. 2
figure 2

Diagnostic test accuracy for single study

Full size image
Fig. 3
figure 3

Summary area under the ROC Curve

Full size image

Subgroup analysis

The threshold of < 12 ng/mL was reported in 8 studies [17, 19, 21,22,23,24,25,26], the rest ranging from 12 to 30 ng/mL [15, 16, 18, 20, 22]. When the threshold of < 12 ng/mL was adapted, progesterone yielded a higher ROC (0.90 vs. 0.78), positive LLR (8.3 vs. 3.8) and DOR (22 vs.12) than its counterpart (12 to 30 ng/mL), respectively (Table 3). Other factors including recruited location and region, progesterone measurement method, exogenous progesterone supplement and follow up, did little impacts on the diagnostic performance.

Table 3 Summarized results of the meta-analysis
Full size table

Discussion

Main results

This study indicated that single progesterone measurement can act as a biomarker predicting miscarriage in early pregnant women with threatened miscarriage, and the performance was better when the concentration was < 12 ng/mL.

Strengths and limitations

In this updated meta-analysis, we included 12 studies which focused on investigating the progesterone predicting miscarriage in early pregnant women with threatened miscarriage, providing a more robust evaluation compared with the previous one [27]. In addition, as the incidence of threatened miscarriage varies widely across races, regions and countries, the use of sensitivity and specificity to assess progesterone predicting miscarriage in this condition is inappropriate and inaccuracy [28]. Therefore, together with sensitivity and specificity, the LLR was calculated [29], because the latter had been proposed to be an prefer outcome in diagnostic study [30, 31].

Also, there are several limitations in this study. First, the threshold interval was relatively wide, lowering the accuracy of estimation. Second, high heterogeneity was observed among studies, leading to increased risk of diagnostic bias and unstable evaluation. Last but not least, a meta-regression analysis can not be performed, due to the absent available data and limited studies.

What ideas does our research provide for clinical applications

Previously, it had been reported that about 50% of women facing threatened miscarriage were affected depressive and anxiety symptomatology [32]. When it comes to the case, they might benefit from an effective treatment or prevention strategy, due to the fear of pregnancy loss, especially for those who had an experience of miscarriage. Unfortunately, there is still absent an effective treatment or prevention strategy. Despite many experimental treatments are frequently prescribed in the real world including exogenous progesterone supplement, the latter had been demonstrated did not improve the clinical outcomes in this subpopulation. As the progesterone concentration < 10 ng/mL can predict a non-viable pregnancy [10], the issue is how is the prognostic value of progesterone concentration in early pregnant women with threatened miscarriage, especially for those who initially seek medical care and conduct the progesterone measurement. Our findings suggested that serum progesterone < 12 ng/mL at a single measurement can also effectively predict miscarriage in early pregnant patients with threatened miscarriage. Therefore, progesterone, a useful prognostic biomarker, would provide help to make a decision in practice and reduce the anxiety secondary to the threatened miscarriage.

What progress does our research provide for future research?

During the first trimester, serum progesterone levels fluctuate over a 24-hour period, the concentration at a single measurement would be influenced by a short-term pattern of pulses, distribution of the hormone and the diet [33]. Moreover, 7 to 8 weeks is considered to be the stage of the luteal-placental shift, progesterone levels may be plateaus or even decrease trend [7], attracting the largest attentions during this period with a highest frequency of vaginal bleeding [34]. Whereas few studies explored the performance of progesterone during the period before 7 weeks of gestation. Thus, studies are warranted to be conducted to further validate the prognostic value of progesterone prior 7 weeks of gestation and clarify the optimal time point of single measurement. In addition, the diagnostic performance of progesterone seemed to be little impacted by the exogenous progesterone supplement even thought at a single measurement without specific time point. This deserves to be validated by further studies, given the fact that exogenous progesterone supplement is still widely prescribed in clinical practice [35].

Conclusion

Single progesterone can act as a prognostic biomarker in early pregnant women with threatened miscarriage, it has a better performance when its concentration is < 12 ng/mL.