Protocol and registration

The research protocol was registered with PROSPERO, CRD42020146987. We have previously published the overview of systematic reviews, whose literature search underpins this IPDMA [8]. The manuscript was reported according to PRISMA-IPD guideline. The project was granted ethics approval by both The University of Sydney (Project no. 2019/397) from 14th May 2019 and Monash University from 27th April 2021 (Project no. 27571).

Eligibility criteria

Studies were eligible for inclusion if meeting the following criteria:

• Participants: asymptomatic women with singleton pregnancy, with or without additional risk factors for SPTB, where not more than 20% of cohort received treatment to reduce risk of SPTB (progesterone, cerclage, pessary), n ≥ 100.
• Indexed prognostic factor: transvaginal ultrasound performed with a widely accepted technique (i.e., as described by Iams and colleagues [7] or Fetal Medicine Foundation [12]).
• Outcome: when both gestational age at delivery and classification as spontaneous or iatrogenic preterm birth are available in individual participant data (IPD).
• Timing: pregnancy dated by standard methods (early dating ultrasound or last menstrual period), gestational age at transvaginal cervical length measurement between 16 and 26 weeks’ gestation, data collected after the year 2000.
• Design: cohort studies, or placebo/untreated groups from randomised controlled trials.

Identifying studies—Information sources

An information specialist helped design the search strategy; the full details are provided in Table A in S1 Appendix. We searched Medline, Embase, CINAHL, LILACS, Database of Abstracts of Reviews of Effects (DARE), Cochrane database, JBI Database of Systematic Reviews, ClinicalTrials.gov, and Google Scholar. The search was restricted to 1st January 2000 and beyond due to anticipated limited availability of older data. No language restriction was applied. We performed citation tracking and consulted experts in the field. The search was performed on 30th September 2020, with a post-analysis updated search performed on 4th November 2025.

Identifying studies—Search

We searched the following keywords: cervix uteri, uterine cervical incompetence, cervical length measurement, ultrasonography, prenatal, (cervix or cervical) length, (pre-term or preterm or premature), (delivery or birth or labour or labor), abortion, spontaneous (Table A in S1 Appendix).

Study selection processes

Title and abstract screening were performed by two authors (KH, RW) using Covidence systematic review software (Veritas Health Innovation, Melbourne, Australia). Full-text review was completed by two reviewers (KH, HF); any conflicts were resolved by a third reviewer (RW/BWM).

Data collection processes

We sought contact details for the corresponding and/or first author, or, where that was unavailable or obsolete, any listed author. We sent an introductory email to at least one author of each citation, and at least two further emails (with alternate addresses tried where available) over the course of several months if no response was received. Where contact details were outdated, we attempted to locate the author’s current institution via internet and PubMed searching or enquired via other research contacts in a similar geographical area.

If an author responded positively, we arranged an online meeting and/or a data-sharing agreement. De-identified data was transferred by secure file transfer platform or e-mail according to contributor preference and uploaded to a secure university server. Data queries were resolved by e-mail correspondence. IPD was sought for all eligible studies where an author was contactable, including those published only as conference abstracts.

Data items

We requested entire de-identified datasets, at minimum including cervical length, gestational age at measurement, and gestational age at birth. Cervical length measured in millimetres nearest to 20 weeks’ gestation was the prognostic factor of interest. In cases where only a range for gestational age at the time of measurement of the cervical length for the entire individual study was provided rather than individual-level data, the mean of that range was imputed as the gestational age. In cases where multiple cervical length measurements were available per participant, we selected the measurement closest to 20 weeks’ gestation due to it coinciding with the usual morphology ultrasound.

We also requested other potential prognostic factors including but not limited to maternal age, BMI, nulliparity, history of preterm births, cervical surgery, tobacco smoking, and use of artificial reproductive technology. Prognostic factor selection among pre-specified clinically meaningful variables was guided by the availability of data across studies (at least 50% of data points) and descriptive analyses of association between factor and outcome.

The primary outcome was SPTB <37 weeks’ gestation. Preterm prelabour rupture of membranes (PPROM) was counted as a SPTB even if delivery was due to induction of labour or if it occurred after 37 weeks; the rationale being that the pathological process commenced prematurely.

The secondary outcomes included SPTB <34 weeks and SPTB <30 weeks. We also presented the time to SPTB <37 weeks as a time-to-event outcome in a sensitivity analysis where term births were censored at 37 weeks. Where studies also reported medically indicated preterm birth, these were treated as censored observations.

IPD checks

All data was checked with respect to range, internal consistency, missing or extreme values, errors, and consistency with published reports prior to conducting the analyses. Concerns about data inconsistency and integrity were solved by contacting the leading author of the original publication. If concerns could not be addressed adequately, we excluded the dataset in question.

Risk of bias assessment in individual studies

Risk of bias was assessed in six domains using the QUality In Prognosis Studies (QUIPS) tool [13]; this included study participation, study attrition, prognostic factor measurement, outcome measurement, adjustment for other prognostic factors, and statistical analysis and reporting. For domains on adjustment for other prognostic factors and statistical analysis and reporting, the assessments were based on the IPD. Two reviewers (KH, EC) assessed the included studies, conflicts were resolved by consensus or a third reviewer (RW). Summary figures for risk of bias were created with robvis [14].

Synthesis methods

Datasets were compiled with SAS software version 9.4 [15], harmonising and analysis were performed with R [16] and Stata [17]. We performed a two-stage IPDMA because most included studies were of adequate sample size. The first stage comprised building regression models for each study, using all the explanatory variables available in each study. We built logistic regression models, producing odds ratios (ORs) with 95% confidence intervals (CIs). We treated cervical length as a continuous variable and used restricted cubic splines to investigate the non-linear relationship between cervical length and SPTB for each trial. Four knots were selected using the centiles recommended by Harrell [18] (at 5th, 35th, 65th, and 95th centiles) based on the overall dataset including all studies. The same values (instead of centiles) of the four knots were then applied to each study.

We built multiple models for the primary outcome, as follows:

1. Basic model including only cervical length and gestational age at measurement as prognostic factors (Model 1)—primary model
2. Model with only key demographic factors (mother’s age, nulliparity) and cervical length (Model 2)
3. Model with all potential prognostic factors (Model 3a), and model with all potential prognostic factors, only including studies that include all factors (Model 3b)

In the second stage, the results of the study-level regression models were meta-analysed in a random-effects model using the inverse variance method. For linear associations, forest plots were produced for each model, and the percentage of the total variability in study estimates due to between-study variance was determined using I² statistic. Between-study variance was measured with τ² and the 95% prediction interval (PrI) was calculated using a previously described inverse variance method [19]. For non-linear associations, multivariate random-effects meta-analyses using restricted maximum likelihood were performed. The associations were visualised in restricted cubic splines to illustrate the odds of SPTB at different cervical lengths compared to those at a reference cervical length (40 mm). Tables with ORs and 95% CIs were also produced to assist with the interpretation.

Where data were systematically missing at study level, the two-stage meta-analysis described here allowed for outcome estimates using all the available variables in each study. Complete case analysis was performed.

The following pre-specified and post-hoc subgroup analyses (exposure–covariate interaction) were performed when data were available, and only within-study interaction was considered:

• History of excisional cervical surgery (large loop excision of the transformation zone/loop electrosurgical excision, cone biopsy)
• Uterine anomaly
• Previous preterm birth (post-hoc)
• Previous term birth (post-hoc)
• Nulliparous women (post-hoc)
• Gestational age at measurement (post-hoc)

The following sensitivity analyses for the primary model were performed when the data were available:

• Exclude women who receive treatment to reduce the risk of preterm birth (cerclage, progesterone, pessary).
• Exclusion of studies with an overall high risk of bias.
• Time to SPTB (<37 weeks), time-to-event outcome. We built Cox proportional hazard models, producing hazard ratios (HRs) with 95% CIs. Gestational age at measurement was used as the entry time variable, and GA at delivery as the exit time variable.

We also performed the following post-hoc sensitivity analyses suggested by the reviewers:

• Restricting to participants with gestational age of measurement between 18 and 21 + 6 weeks.
• For time to SPTB (<37 weeks), competing risk model was used, and iatrogenic preterm birth was considered as a competing event to SPTB.

Deviations from the statistical analysis plan

Following the advice from peer reviewers, we used logistic regression, instead of Cox regression, as the primary analysis and treated the primary outcome as a binary outcome to make the paper more accessible to all readers. We still present the Cox regression on time to SPTB findings in sensitivity analyses (Figs D1-6 in S1 Appendix).

Search results

The initial search yielded 6,085 citations. After removal of duplicates and full-text screening, IPD was sought for 109 eligible citations, however, for 64 of them, authors were unable to be contacted or did not complete the data sharing process for various reasons. We obtained IPD from 27 datasets, comprising 91,404 pregnancies from an estimated possible 183,903 (50.7% of IPD) as outlined in the PRISMA flow diagram (Fig 1). One dataset was subsequently excluded due to data integrity concerns (n = 200). One dataset (n = 7,954) was excluded due to cervical length measurements only available for 6% of participants. Studies where IPD could not be obtained, and the reasons for unavailability are listed in Table B in S1 Appendix.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. PRISMA flow diagram.

https://doi.org/10.1371/journal.pmed.1004653.g001

At times, two or more citations (denoted “records” in Fig 1) were found to have been produced from a single dataset (denoted “study” in Fig 1). Where authors were unable to be contacted, we could not verify whether one record corresponded to one study, but this was assumed to be the case for recording the number of pregnancies that could not be included in our combined dataset.

Study characteristics

Study characteristics and demographic data are shown in Tables 1 and 2. The studies were conducted in 12 countries between 2001 and 2019. Fifty-five% of the study population were nulliparous (n = 46,195). Mean gestational age at cervical length measurement was 20.3 weeks’ gestation (standard deviation [SD] 1.6 weeks), and mean cervical length was 40 mm (SD 9 mm). Eleven,128 women (up to 12.2% of the cohort) received prophylactic treatment as shown in Table 2. Mean gestational age at birth was 38.9 weeks’ gestation (SD 2.5 weeks); results by study are shown in Table 1. 4442 women (5.2%) had a SPTB <37 weeks’ gestation, and 1,420 (1.6%) before 34 weeks. The rate of SPTB <37 weeks varied between studies (range 3.6%–18.8%). Cervical length and gestational age at delivery were weakly positively correlated (sample correlation estimate = 0.19, 95% CI [0.19.0.20]), see Fig A in S1 Appendix.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Included studies.

https://doi.org/10.1371/journal.pmed.1004653.t001

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Demographic data.

https://doi.org/10.1371/journal.pmed.1004653.t002

Risk of bias assessments

Results for QUIPS risk of bias (RoB) assessment are shown in Fig 2A and 2B. 3/27 were rated as low RoB overall. 11/26 studies were assessed as being at high RoB overall. 19/26 studies were at moderate or high risk of bias in domain two (bias due to attrition).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Risk of bias assessment results.

(A) Quality In Prognosis Studies (QUIPS) traffic light plot. (B) QUIPS summary plot.

https://doi.org/10.1371/journal.pmed.1004653.g002

Individual participant data meta-analysis

The primary model (including only cervical length and gestational age at measurement, Model 1) showed an L-shape non-linear association between cervical length and SPTB (22 studies, 78,047 participants, Fig C1 in S1 Appendix Table C in S1 Appendix). We present the association on the odds scale for easier interpretation (Fig 3). A longer cervical length was associated with steeply lower odds of SPTB until it reached 40 mm, beyond which the odds of SPTB became stable with longer cervical length. Using cervical length of 40 mm as the reference, a woman with a cervical length of 20 mm was associated with 6.22 higher odds of SPTB (95% CI [4.76,8.13]) whereas a woman with a cervical length of 30 mm was associated with 2.10 higher odds of SPTB (95% CI [1.85,2,38]). The detailed summary statistics including the knots, coefficients, 95% confidence intervals, p values, and between-study heterogeneity (tau) for each restricted cubic spline term are presented in Table C in S1 Appendix.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Non-linear association between cervical length and spontaneous preterm birth <37 weeks in the primary model.

This figure illustrates ORs and 95% Cis of spontaneous preterm birth <37 weeks at different cervical length vs. the reference cervical length (40 mm) in the primary analysis (Model 1). Cervical length and gestational age at measurement were the only prognostic factors used in the model.

https://doi.org/10.1371/journal.pmed.1004653.g003

Similar shapes of non-linear association between cervical length and SPTB were observed when adding maternal body mass index (BMI), maternal age and nulliparous status into the model (Model 2, see Figs B2, C in S1 Appendix), adding all potential prognostic factors (Model 3a, Fig 3 and Fig B3 in S1 Appendix), or model with all potential prognostic factors, only including studies that include all factors (Model 3b, Fig 3 and Fig B4 in S1 Appendix). For Models 2 and 3b, the ORs were larger than those in Model 1 when cervical length was short. As only 4 studies including 6,062 participants were included in Model 3b, the CIs for Model 3b are wide on both tails (Fig C4 in S1 Appendix). The detailed statistics for each model are presented in Table C in S1 Appendix.

The prognostic value of cervical length for SPTB <34 weeks was stronger, as shown with higher ORs at each cervical length value under 40 mm, compared to those on SPTB <37 weeks (Figs 4 and C1 in S1 Appendix). Above 40 mm, the ORs for SPTB <34 weeks did not increase with longer cervical length. Similarly, the prognostic value of cervical length for SPTB <30 weeks was even stronger (Figs 4 and C2 in S1 Appendix). By contrast, the prognostic value of cervical length on any preterm birth <37 weeks was weaker than those on SPTB < 37 weeks ( Figs 4 and 5 and C3 in S1 Appendix).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Non-linear association between cervical length and spontaneous preterm birth <37 weeks in different models.

This figure illustrates ORs of spontaneous preterm birth <37 weeks at different cervical lengths vs. the reference cervical length (40 mm) in three models: Model 1: including only cervical length and gestational age at measurement as prognostic factors; adjusted for gestational age at measurement only. Model 2: including only key demographic factors (mother’s age, nulliparity), cervical length, and gestational age at measurement. Adjusted for gestational age at measurement, maternal age, nulliparity. Model 3: including all potential available prognostic factors (Model 3a). Adjusted for gestational age at measurement, maternal age, maternal BMI, nulliparity AND/OR history of preterm birth. If any of these factors were missing within a study, then they were not adjusted for in the first stage for that particular study, but included in the meta-analysis at the second stage. A model with all potential prognostic factors, only including studies that include prognostic factors (Model 3b). Adjusted for gestational age at measurement, maternal age, maternal BMI, nulliparity AND history of preterm birth. If any of these factors were missing within a study, then that study was excluded from the meta-analysis in the second stage.

https://doi.org/10.1371/journal.pmed.1004653.g004

Sensitivity analysis

Sensitivity analyses for SPTB <37 weeks excluding women who received cerclage, progesterone, or pessary, excluding studies with an overall high risk of bias or excluding participants with gestational age of cervical length measurement outside 18 to 21 + 6 weeks showed robust non-linear association curves, compared to the primary analysis (Fig D1–4 in S1 Appendix). The detailed statistics for all sensitivity analyses are presented in Table C in S1 Appendix.

A sensitivity analysis based on Cox regression models was also performed, with a similar shape of non-linear association curve (Fig D5 in S1 Appendix). When limiting to 10 studies reporting both spontaneous and iatrogenic preterm birth, another sensitivity on competing risk model was performed, which was consistent with the findings from the Cox regression (Fig D5 in S1 Appendix).

Subgroup analysis (exposure–covariate interaction)

Planned subgroup analysis showed stronger prognostic value of cervical length on SPTB <37 weeks in women with a history of cervical surgery compared to women without a history of cervical surgery when cervical length was below 28 mm. Such a difference became inconclusive when cervical length was above 28 mm due to wider CIs (Fig E1 in S1 Appendix).

In women with a previous preterm birth, the prognostic value of cervical length on SPTB<37 weeks was weaker than those without such history when cervical length was below 25 mm. Such a difference became inconclusive when cervical length was above 25 mm due to wider CIs (Fig E3 in S1 Appendix).

No evidence of other potential interaction were observed between cervical length and(uterine anomaly, history of term birth, or nulliparity (Fig E2, E4, E5 in S1 Appendix). There was also no evidence of an interaction between gestational age at measurement and cervical length on SPTB <37 weeks (Fig D5 in S1 Appendix). The detailed statistics for all subgroup analyses are presented in Table C in S1 Appendix.

Systematic review of studies identified during updated literature search

Following the literature search update on 4th November 2025, we identified 26 newly eligible studies (46,247 participants) for our IPDMA. In addition, six more studies may be eligible: they did not directly address the association between cervical length and SPTB, but may have information on gestational age, cervical length, and SPTB available in the datasets.

Study characteristics are presented in Table B in S1 Appendix. Of the 26 eligible studies, 11 (42%) were abstract-only reports, and 15 (58%) were full publications. Twelve (46%) had a prospective cohort study design, 14 (54%) had a retrospective cohort design. The median number of participants in the 26 eligible studies was 797 (interquartile range (IQR) 339–1,425), 13 (50%) reported the research question as a prognostic factor research question, and the other half were reported as a diagnostic test accuracy research question. Among the 13 studies addressing the association between cervical length and SPTB as a prognostic research question, 10 analysed cervical length as a categorical prognostic factor (six used 25 mm as the cutoff) and only three analysed cervical length as a continuous prognostic factor. None considered non-linear associations between cervical length and SPTB in their analyses. Dunn and colleagues reported an OR of 0.92 (95% CI [0.87,0.98]) per 1 mm increase in cervical length [47]. Movahedi and colleagues reported an adjusted OR of 0.82 (95% CI [0.75.0.90]) per 1 mm increase in cervical length [48]. Wikstrom and colleagues reported adjusted ORs per 5 mm decrease in cervical length for three risk groups: high-risk women (OR 1.32; 95% CI [1.07,1.63]), nulliparous with no risk factors (OR 1.69; 95% CI [1.45,1.98]), and parous women with no risk factors (OR 1.22; 95% CI [1.00,1.48]) [49]. Therefore, it would not be possible to perform meta-analyses based on aggregate data presented in these studies or to consider non-linear associations as we did in our IPDMA.