Choice of primary outcomes in randomised trials and systematic reviews evaluating interventions for preterm birth prevention: a systematic review


  • S Meher,

    Corresponding author
    1. Department of Women and Children's Health, Institute of Translational Medicine, University of Liverpool, Liverpool, UK
    2. Queen Charlotte's and Chelsea Hospital, Imperial College Healthcare NHS Trust, London, UK
    • Correspondence: Dr S Meher, Department of Women and Children's Health, Institute of Translational Medicine, University of Liverpool, Liverpool L8 7SS, UK. Email

    Search for more papers by this author
  • Z Alfirevic

    1. Department of Women and Children's Health, Institute of Translational Medicine, University of Liverpool, Liverpool, UK
    Search for more papers by this author

  • Linked article: This article is commented on by Scott J, p. 1195 in this issue and Williamson P, p. 1196 in this issue.



The inappropriate and inconsistent selection of primary outcomes (POs) in randomised controlled trials (RCTs) and systematic reviews (SRs) can make evidence difficult to interpret, limiting its usefulness to inform clinical practice.


To systematically review the choice and consistency of POs in RCTs and SRs of preventative interventions for preterm birth.

Search strategy

Cochrane Pregnancy and Childbirth Group's Specialised Register of trials and a full list of published reviews and protocols.

Selection criteria

Full reports of RCTs for preterm birth prevention published after CONSORT (January 1997–January 2011), and Cochrane Reviews and protocols relevant to preterm birth prevention, for the same period.

Data collection and analysis

For RCTs, the PO was the outcome used for sample size calculation. For SRs, we included all outcomes listed as ‘primary’. Two review authors selected studies and double-checked the data for accuracy.


Seventy-two different POs were reported by 103 RCTs. The three most common POs were based on length of gestation, with preterm birth before 37 weeks of gestation being the most common (18/103, 18%). Few RCTs chose perinatal morbidity (4/103) or mortality (1/103), or their composites (5/103), as POs. In 33 Cochrane Reviews, 29 different POs were reported. The three most common POs were based on death or morbidity in the baby, with death of the baby being the most common (22/33, 67%). POs were variably defined.


There is a lack of consistency in the choice and definitions of POs in clinical research related to preterm birth prevention. SRs are more likely to report morbidity and mortality as POs, whereas RCTs tend to use length of gestation. Researchers are urged to review the outcomes reported in RCTs and SRs in their respective areas of interest to highlight discrepancies and facilitate the development of core outcome sets.


In the hierarchy of biomedical research, evidence from randomised controlled trials (RCTs) and their systematic reviews is regarded as most reliable,[1] and is used to guide healthcare decisions. Therefore, it is important that RCTs and systematic reviews measure end points that are relevant and meaningful to patients, clinicians, and health service providers. It is also important that there is consistency in the choice and definitions of outcomes across clinical trials and systematic reviews, to allow for the unbiased synthesis of evidence.

High-quality RCTs and systematic reviews are expected to pre-specify primary and secondary outcomes, and to report them in protocols published at the time of trial registration.[2, 3] The primary outcome should be the outcome of highest importance, representing the greatest therapeutic benefit of the intervention, upon which conclusions about the effect of the intervention are largely based.[3] Sometimes, multiple primary outcomes are selected when more than one outcome is thought to be of high, but equal importance. In RCTs, as a rule, the primary outcome is the outcome that is used to calculate the sample size,[2] whereas secondary outcomes are used to evaluate any additional effects of the intervention.

Preterm birth is a leading cause of significant perinatal morbidity and mortality worldwide,[4] and interventions for its prevention have been the subject of substantial research.[5-7] Large, definitive RCTs in this area are rare, and systematic reviews are often needed to provide robust evidence of effectiveness and safety.[8] Therefore, in order to investigate the degree of consensus in the choice and definitions of primary outcomes used in RCTs and systematic reviews, we chose to explore one of the most important health issues related to maternity care: the prevention of preterm birth.


Criteria for the selection of studies

We included full reports of RCTs of interventions for the prevention of preterm birth, published from January 1997, after the first CONSORT statement was published, until January 2011. Populations were restricted to women at risk of preterm labour or with threatened preterm labour at gestations of <37 weeks. We excluded: non-randomised, observational, analytical, or diagnostic studies and feasibility studies; studies primarily aiming to assess pharmacokinetics, the mechanism of drug action, or tolerability; and studies evaluating interventions with the primary aim of improving fetal condition (e.g. steroids for lung maturity, or magnesium sulphate for neuroprotection). We excluded studies published only as abstracts. For systematic reviews, we included all published Cochrane Reviews and protocols that had preterm birth prevention as the main focus of the review, using the criteria described above.

Search strategy

We searched the Cochrane Pregnancy and Childbirth Group's (PCG) Specialised Register for randomised trials relevant to preterm birth. The register is a database of citations to published and unpublished controlled clinical trials in perinatal medicine. It contains over 17 000 records, with an annual accrual of 1000 new records.[9] The register is maintained by monthly searches of the Cochrane Central Register of Controlled Trials (CENTRAL), weekly searches of MEDLINE and EMBASE, hand searches of 30 journals and the proceedings of major conferences, and weekly current awareness alerts for a further 44 journals, plus monthly BioMed Central email alerts. In addition, members of the Cochrane Collaboration contribute to the register from all around the world by sending in reports of RCTs retrieved through sources including local foreign language databases and personal communications. Details of the PCG search strategy can be found at RCTs for this study were retrieved using the register's codes for preterm birth. Full papers for each eligible study were retrieved. No language restrictions were applied, and translations were obtained for non-English language reports where needed. Cochrane Reviews and protocols relevant to preterm birth were identified by obtaining a full list of published Cochrane Reviews and protocols from the editorial base of the Cochrane and Pregnancy Childbirth group in January 2011.

We also conducted a search for core outcome sets in the area of preterm birth in MEDLINE, using the search string ‘core’ AND ‘outcome’ and ‘set OR measures’ and ‘preterm OR pregnancy’.

Data collection and analyses

Studies were assessed for inclusion by two authors (S.M. and Z.A.), independently, using the pre-specified criteria described above. Data were collected on all reported outcomes from full articles of each publication on a pre-defined data extraction form. For RCTs, the primary outcome was defined as the outcome used for a priori sample size calculation. Trials that did not report sample size calculations were excluded. For Cochrane Reviews and protocols, all outcomes listed as ‘primary’ were included. Reviews that did not differentiate between primary and secondary outcomes were excluded. Data were entered into a Microsoft Access database, double checked for accuracy, and analysed on Microsoft Access and Excel.


A total of 1305 citations relevant to preterm birth were identified in the PCG Specialised Register of Trials. Of these, 1080 citations were excluded for various reasons (Figure 1), and a total of 170 studies (225 citations) were included. Forty-nine published Cochrane Reviews and protocols relevant to preterm birth were identified, and 33 fulfilled the pre-specified criteria (Figure 1).

Figure 1.

Flowchart for the selection of randomised trials and Cochrane systematic reviews and protocols.

Primary outcomes in randomised trials

The types of interventions assessed by trials are listed in Table 1. Nearly half the trials (49%) evaluated the efficacy of tocolytics for preventing or delaying preterm birth, 20% assessed the impact of antibiotic treatment, 9% assessed progesterone, 6% assessed cerclage, and the remaining 15% looked at a range of other interventions.

Table 1. Interventions for preterm birth prevention, evaluated in randomised trials published between January 1997 and January 2011
InterventionNumber of trials
Home uterine activity monitoring4
Fetal fibronectin testing4
Management based on ultrasound assessment of cervical length3
Relaxation therapy2
Infection screening2
Activity restriction (bed rest)2
Nurse support1
Magnesium supplementation1
Total trials170

One hundred and three out of 170 (61%) RCTs pre-specified a primary outcome that was underpinned by a sample size calculation. Seventy-two different primary outcomes were reported by these 103 trials. A full list of reported primary outcomes may be found in Appendix S1. The most common primary outcomes, reported by at least three trials, are listed in Table 2. They can be categorised into four groups: (1) those that were related to length of pregnancy, either dichotomised at a particular gestation or length of time gained; (2) neonatal morbidity or mortality; (3) composites of these; and (4) miscellaneous others. The incidence of preterm births <37 weeks of gestation was the most common primary outcome, used as such by 18% of the trials (18/103). Other primary outcomes based on length of gestation included proportion of births at 30, 32, 33, 34, and 35 weeks of gestation. Length of time gained was reported as mean or median hours, days, and weeks, or was dichotomised at more than or <24 hours, 48 hours, and 7 days.

Table 2. Most commonly reported primary outcomes underpinned by sample size calculations in randomised trials, and proportion of Cochrane Reviews reporting the same primary outcomes
Most common primary outcomes in randomised trialsa,bRandomised trials reporting primary outcome (n = 103)Cochrane reviews reporting primary outcome (n = 33)
  1. a

    Only outcomes used for sample size calculations are included.

  2. b

    Outcomes reported by at least three RCTs are listed.

Preterm birth <37 weeks of gestation18 (18%)8 (24%)
Latency mean or median hours, days, or weeks11 (11%)1 (3%)
Latency 7 days11 (11%)0
Latency 48 hours10 (10%)6 (18%)
Composite morbidity or death5 (5%)10 (30%)
Composite preterm delivery or death5 (5%)0
Changes in contractions or cervical dilatation4 (4%)0
Preterm birth <35 weeks of gestation4 (4%)0
Preterm birth <34 weeks of gestation3 (3%)8 (24%)

Very few trials reported single indicators of perinatal morbidity (4/103) or mortality (1/103), or their composites (5/103), as primary outcomes. No single component was common to all composite outcomes (Table 3). Even the same components were variably defined by trialists. For example, criteria for necrotising enterocolitis were either ‘requiring surgery’, ‘grades 2 or 3’, or unspecified.

Table 3. Components of composite adverse neonatal outcomes used as primary outcomesa
Component of composite neonatal outcomesRandomised trials (n = 5)
  1. a

    Only primary outcomes used for sample size calculations are included.

Death (miscarriage, stillbirth, perinatal, neonatal, fetal, or postnatal)4
Intraventricular haemorrhage (grade 3/4, grade 4)4
Necrotising enterocolitis (unspecified, grade 2/3, requiring surgery)4
Bronchopulmonary dysplasia/chronic lung disease4
Periventricular leukomalacia3
Respiratory distress syndrome2
Sepsis (unspecified, within 72 hours)2
Retinopathy of prematurity1
Cerebral abnormality on scan1
Asphyxia of newborn1

The cost of treatment was used as the primary outcome in one trial, and no trials reported long-term morbidity data as a primary outcome.

Of the 170 trial reports included, 130 (76%) were published in 32 specialty journals, whereas 40 (24%) were published in 29 general journals. RCTs published in the specialty journals where more likely to focus on differences in the gestational age between two groups (Table 4). There were no other significant differences in the choice of primary outcomes or frequency of their reporting.

Table 4. Analysis of primary outcome by type of journal: specialist or general
Most common primary outcomes in randomised trialsa,bRandomised trials in specialist journalc (n = 82)Randomised trials in general journal (n = 21) P
  1. a

    Only outcomes used for sample size calculations are included.

  2. b

    Outcomes reported by at least three RCTs are listed.

  3. c

    Journals with a focus on a specific area of health care: i.e. obstetrics and gynaecology.

Preterm birth <37, <35, <34 weeks of gestation19 (23%)6 (29%)0.776
Latency (mean or median hours, days, or weeks; >7 days; >48 hours)30 (37%)2 (10%)0.018
Composite morbidity or death3 (4%)2 (10%)0.58
Composite preterm delivery or death2 (3%)3 (14%)0.062
Changes in contractions or cervical dilatation4 (5%)00.579

Primary outcomes in published Cochrane Reviews and protocols

Among the 33 Cochrane Reviews or protocols, 82% (27/33) pre-specified more than one primary outcome, and 29 different primary outcomes were reported. The most common primary outcome reported by at least three reviews/protocols was death of the baby (Table 5). This was variably defined, and included miscarriage, perinatal death, stillbirth, neonatal death, and various combinations of these. Six of the 33 reviews (18%) did not list morbidity or mortality in the baby as one of their primary outcomes, but focused on gestational age instead (proportion of birth before 37 or 34 weeks of gestation).

Table 5. Most commonly reported primary outcomes in Cochrane Reviews and protocols, and their relative frequency in randomised trials
Most common primary outcomes in Cochrane Reviews and protocolsCochrane Reviews reporting primary outcomea (n = 33)Randomised trials reporting primary outcome (n = 103)
  1. a

    More than one primary outcome in 27/33 reviews.

Death in baby (variously defined)22 (67%)1 (1%)
Composite morbidity or death in baby (variously defined)10 (30%)5 (5%)
Composite morbidity in neonatal period or at follow up (variously defined)8 (24%)1 (1%)
Preterm birth <34 weeks of gestation8 (24%)3 (3%)
Preterm birth<37 weeks of gestation8 (24%)18 (18%)
Latency, 48 hours6 (18%)10 (10%)
Composite maternal morbidity and mortality5 (15%)0
Preterm birth <28 weeks of gestation4 (12%)0
NICU admission for baby3 (9%)0
Maternal death3 (9%)0
Maternal hospital stay3 (9%)1 (1%)

Comparison of primary outcomes between randomised trials and systematic reviews

Four out of the nine most common primary outcomes in RCTs were never used as primary outcomes in reviews. Similarly, 36% (4/11) of the most common primary outcomes reported in systematic reviews were not reported as primary outcomes in any of the randomised trials. Death of the baby was reported as a primary outcome in 67% of the systematic reviews and in 1% of the RCTs, whereas composite morbidity and mortality was reported by 33% of systematic reviews, but by only 5% of RCTs.

Six Cochrane Review authors were also authors on four of the included RCTs. The choice of primary outcome was similar across the RCTs and systematic reviews for two authors (one based on morbidity/mortality, and the other based on surrogate outcomes of time/gestation age). The other two authors chose primary outcomes in RCTs that were based on length of gestation, but primary outcomes in their reviews also included morbidity and mortality for the baby.


Main findings

Our findings demonstrate a lack of consensus on what are the most important clinical outcomes related to preterm birth prevention. Randomised trials tend to focus on duration of pregnancy and systematic reviews tend to focus on deaths and neonatal morbidity. There was a huge variation in the definitions for all of these domains.

Strengths and limitations of the study

To our knowledge, this is the first study that has systematically investigated differences in choice of primary outcomes between randomised trials and systematic reviews in a specific area of healthcare. The findings of this study highlight a significant lack of consensus and standardisation that exists not only amongst clinical trials, but also amongst systematic reviews. The study also shows that systematic reviews are more likely to focus on rare, but clinically more meaningful, end points.

The findings of this study may have been influenced by the exclusion of studies that were published prior to 1997, did not report sample size calculation, or were published as abstracts only. We chose to exclude them to reduce the risk of reporting bias. Trials published prior to the first CONSORT statement in December 1996 were not required to report pre-specified primary outcomes. We used a purist approach to define primary outcome, based on sample size calculation as a method of validating the stated primary outcome. As abstracts are not required to report sample size calculations, it would not have been possible to validate reported primary outcomes if we had included them. The findings of this study may also have been influenced by the restriction of searches for systematic reviews to Cochrane Reviews. This was done because Cochrane Reviews have a clear policy of reporting primary and secondary outcomes.


It is generally accepted that an important and clinically meaningful end point should measure how a patient feels, functions, or survives.[10] Mortality and morbidity for the baby, the most common primary outcomes in systematic reviews of preterm birth, are clearly important end points upon which conclusions about the effectiveness of an intervention can be made. On the other hand, the prolongation of gestation, the most common primary outcome in clinical trials, is generally regarded as a surrogate end point for mortality and morbidity. For such a surrogate end point to be an effective substitute for clinically important outcomes, the effects of the intervention on the surrogate must reliably predict the overall effect. In practice, this requirement frequently fails because interventions may have a range of intended or unintended health effects that are not captured by surrogate end points.[10] For example, a tocolytic agent may be very successful in arresting uterine contractions and delay preterm birth, but an underlying infective process that triggered preterm labour may continue to have an adverse impact and create a hostile uterine environment that increases neonatal morbidity and mortality.

The detection of small differences in relatively rare but critically important end points, such as morbidity and mortality, usually requires large trials that are difficult to fund and take longer to complete. Our results indicate that trialists are more likely to select surrogate primary outcomes that lead to feasible sample sizes. Overcoming this issue requires much wider collaboration on a national and international level, including greater planning towards prospective meta-analyses on the individual patient data level. Using composite end points is another common strategy, but selecting and agreeing components of the composite outcome is complex. The adverse outcomes included are rarely of similar importance to patients and health professionals. Empirical evidence suggests that less important but more frequent outcomes, for example, admission to neonatal intensive care (NICU) for preterm babies, generally have a bigger effect on the results of the composite outcome than more important components.[11]

There is little evidence on the factors that influence choice of primary outcomes in systematic reviews. The availability of data from clinical trials, previous knowledge of results, and the desire to publish positive results could potentially impact on the outcomes reported.[12] Systematic reviews are unlikely to be influenced by financial and time constraints in the same way as clinical trials, and therefore it is not surprising that they tend to select rare end points that measure significant morbidity and mortality.

As a solution to the problem of inappropriate and inconsistent outcome selection, and to facilitate synthesis of the evidence, it has been proposed that outcomes reported by clinical trials should be standardised for different disease conditions through the development of core outcome sets (COS), and that systematic reviewers should build their reviews around these outcomes.[13, 14] The core outcome measures in effectiveness trials (COMET) initiative is developing a framework to facilitate the process of COS development,[13] and COS have been developed successfully in a number of areas.[15-18] However, there is little empirical evidence on the best methods for developing COS and who should develop them. Once developed, it is not completely clear how to ensure that they are endorsed and adhered to, and how to assess their quality and impact on a global level. COMET recommends that consideration should be given to the identification of existing knowledge in the area for developing COS.[13] Whereas most other steps in the development of COS inherently involve an element of subjectivity, a review of outcomes considered to be most important by previous trialists and systematic reviewers gives an objective baseline from which the development of COS may proceed.

There has been an effort to identify a minimum set of outcome measures for evaluating maternity care as a whole.[19] Using the Delphi technique, 48 core clinical outcomes were identified to be important by a range of stakeholders; however, only 47% of participants completed all three Delphi rounds, and there was a predominant midwifery representation (45%), which may have masked the priorities of other groups. The variety of outcomes included in the long list of core outcomes reflects the vast number of topic areas contained within the broad scope of maternity. Choosing the most appropriate outcomes from such a list for a specific topic would inherently be subjective. There is a need to develop specific COS for particular disorders in maternity care to enable the selection of the most appropriate outcomes, which can then be measured consistently across studies. Our search did not identify any published COS for specific topics within maternity care, although work is planned in the area of preterm birth.[20]


Implications for clinical practice and research

Inconsistency and lack of standardisation in the selection of primary outcomes is prevalent not only amongst RCTs but also amongst systematic reviews, with a clear mismatch in choice of primary outcomes between RCTs and systematic reviews. Systematic reviews appear to be more likely to choose rare but clinically more meaningful end points, whereas primary outcomes underpinned by sample size calculations in RCTs are more often surrogate outcomes. The time has come for this gap to be bridged. It is incumbent on every speciality to review the outcomes reported in their trials and systematic reviews to highlight either consensus, or lack of it, over outcomes perceived to be important. The information obtained will provide a basis for the development of COS, and will facilitate a much needed change towards the appropriate selection and standardisation of clinically important outcomes in all areas of health care.

Disclosure of interests

The authors declare no support from any organisation for the submitted work, no financial relationships with any organisations that might have an interest in the submitted work in the previous 3 years, and no other relationships or activities that could appear to have influenced the submitted work. SM is an Associate Editor for the Cochrane and Pregnancy Childbirth Group. ZA is the Coordinating Editor for the Cochrane and Pregnancy Childbirth Group. The results and views expressed in this article are those of the authors and not of any organisations.

Contribution to authorship

SM conducted the searches, performed the data extraction, data entry, and analyses, and co-wrote the article. ZA had the original idea for the study, double-checked the selection of studies, performed data extraction and analyses, and co-wrote the article. SM is the guarantor.

Details of ethics approval

This study did not require ethical approval as the data used have been published previously, and hence are already in the public domain.


This study did not receive any funding or sponsorship from any sources.


We would like to thank Lynn Hampson from the Cochrane Pregnancy and Childbirth Group for providing data from their Specialised Register of Trials.