Methods 1: cohort study

Parts of this chapter have been reproduced with permission from the published protocol by Milton et al.,32 practicalities discussion paper by Lugg-Widger et al.33 and Statistical Analysis Plan paper by Cannings-John et al.34 These are Open Access articles distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The text below includes minor additions and formatting changes to the original text.

Cohort study design

The cohort study was a natural experiment using a combination of retrospective and prospective data captured in electronic NHS maternity and neonatal information systems. It was designed as a non-inferiority study, and all primary comparisons were analysed and are presented on this basis.

Study participants: inclusion and exclusion criteria

Local maternity information systems (MISs) capture data on all pregnancies and births in their area, where maternity care is provided by the NHS, including births in OUs, MLU or at home. To identify women who used any form of water immersion during labour or birth, the fields, ‘use of a pool during labour’ and ‘waterbirth’, were used pragmatically. This identified women who used water immersion in a conventional bath or any form of birthing pool during labour.

All registrable births occurring at the 26 participating NHS sites between 1 January 2015 and June 30 2022 were included. Registrable births in the UK include live births at any gestation and stillborn babies, that being babies born without signs of life at or after 24 + 0 weeks of gestation. Data relating to women who opted out from the study and their babies were not received, and births recorded as being unattended by a midwife due to them being classified as ‘born before arrival’ (BBA), in transit or freebirths, were excluded.

Outcomes

Primary outcomes

There were two primary outcomes:

• The maternal primary outcome was severe perineal trauma of OASI.
• The infant primary outcome was a composite of ‘adverse infant outcomes or treatment’ to include:
  1. any NNU admission requiring respiratory support
  2. intravenous antibiotic administration within 48 hours of birth (with or without culture proven infection)
  3. intrapartum stillbirth and all deaths prior to NNU/postnatal ward discharge.

Secondary outcomes

Maternal intrapartum secondary outcomes

• Incidence of shoulder dystocia and required management.
• Management of the third stage of labour (including whether the placenta was intended to be, or delivered, in or out of water).
• Rates and management of postpartum haemorrhage (PPH).
• Need and reason for obstetric involvement in woman’s care (including sepsis, treatment for haemorrhage).
• Incidence and management of perineal trauma.
• Maternal position at birth.
• Mode of birth.

Maternal postnatal secondary outcomes

• Duration of postnatal hospital stay.
• Breastfeeding initiation and continuation.
• Higher-level care.
• Maternal re-admission to hospital within 7 days of birth.

Infant secondary outcomes

• Snapped umbilical cord prior to clamping.
• Timing of cord clamping.
• Apgar scores at 1, 5 and 10 minutes.
• Resuscitation at birth.
• Intrapartum stillbirth or deaths prior to NNU/postnatal ward discharge.
• Cause of intrapartum stillbirth or death prior to NNU/postnatal ward discharge.
• Skin-to-skin contact at birth.
• First breastfeed within first hour.
• NNU admissions.
• Respiratory support.
• Therapeutic hypothermia.
• Brachial plexus injury.
• Treatment for jaundice.
• Successful/attempted lumbar puncture.
• Administration of intravenous antibiotics, including timing and duration.
• Highest C-reactive protein (CRP).
• Blood culture positive with a recognised pathogen (excluding skin commensal organisms).
• Neonatal deaths that occurred within 7 days of birth on a NNU/postnatal ward.
• Re-admission to hospital within 7 days of birth.

Data sources

Identification of data sources

Due to the need to identify a cohort of women who used a pool during labour and to compare outcomes between those who remained in or left the pool for birth, it was essential for the study to identify data sources that captured pool use during labour as well as separately capturing birth in water. Mapping data fields within national maternity data sets against the study objectives in 2017 confirmed that some essential data items were not exported from NHS sites to national data sets. This included a field to identify women who used water immersion during labour but not for birth. As this was central to the research question, this confirmed that national maternity data sets could not be used for this study. In addition, other data items, including some required to inform the neonatal primary outcome, were not captured at either local or national level. Therefore, to design a study to answer the National Institute for Health and Care Research (NIHR) commissioning brief questions, while meeting the requirement to maximise the use of existing routinely collected data, it became apparent that existing routinely collected data at NHS site level could be used but would need to be supplemented with additional data fields.

As local MISs are required to be adaptable to accommodate changing clinical and monitoring requirements, they have the facility for additional data items to be added or adapted. The decision was taken to add study-specific data fields to existing local NHS maternity systems completed by midwives as part of usual maternity care and for the records to be extracted at regular time points during the study. Two data sources were identified – maternity records held in Euroking^® (Magentus Maternity Software Ltd, London, England) MISs and neonatal records held by the National Neonatal Research Database (NNRD) at Imperial College London.

Partnered maternity information system

Electronic MIS forms a comprehensive maternity data set covering the antenatal, intrapartum and postnatal period. MISs provide similar functionality, but due to the complexity of data extraction, a decision was taken by the study team to collaborate with a single large electronic MIS provider. In 2016, at the time discussions began, the Euroking MIS, provided by Wellbeing Software^® (WS, Magentus Maternity Software Ltd, London, England), was the most used system in the UK.35 In 2016, the system captured data relating to 96,951 births, including 6037 (6.2%) waterbirths in NHS sites, ranging in size from approximately 1500 to 10,000+ annual births. As such, these units were able to provide the volume of data required for the planned study, could be adapted to the study need and had the advantage of being familiar to the clinically based investigators.

National Neonatal Research Database

The NNRD36,37 holds individual patient-level data on all infants admitted for NHS neonatal care in England, Wales and Scotland. Approximately, all 200 NNUs in England, Wales and Scotland form the UK Neonatal Collaborative and contribute electronic health record data to the NNRD. The data set held by the NNRD comprises 450 clearly defined variables extracted at patient level from the electronic health record systems used by UK NNUs.

Field identification

To identify which of the existing fields would be required for the study, the statistician, with clinical support, matched the existing Euroking MIS data dictionary to those required for the POOL Study. Required fields included those to define and characterise all women, including whether a pool was used, potential confounders and outcomes.

Fields required to answer the research questions but not present in the existing data dictionary were identified, defined and developed. These fields were: maternal or pregnancy risk factors present at pool entry; the use of continuous electronic fetal monitoring in water; the administration of intravenous oxytocin for labour augmentation in water; births occurring partially in water; umbilical cord snapping prior to clamping; neonatal antibiotic administration on the postnatal ward and clinical markers for neonatal infection; and management of placental delivery following waterbirth.

To be able to compare characteristics of women who did, or did not, use a pool in labour, data were required on characteristics of women who did not use a pool in labour. The two required data sets were identified. The data set for women who did not use a pool in labour consisted of 80 variables compared to 201 variables within the data set for women who used a pool.

Data collection period

The maternal primary outcome analysis included births between 1 January 2015 and 30 June 2022. The neonatal primary outcome analysis included births from the date of individual sites opening to 30 June 2022.

Follow-up period

The duration of follow-up for the study was until discharge from maternity community postnatal care. For NNRD data, the duration of follow-up was up to NNU discharge.

Recruitment of sites

Potential NHS study sites were identified as those NHS Trusts or Health Boards with an ongoing contract with WS and with waterbirth facilities. Each potential site was contacted by the study manager, inviting them to register an interest in study participation. The proposed NHS site principal investigator provided a Curriculum Vitae and a valid Good Clinical Practice certificate. Contracts were set up between Cardiff University (CU) and the NHS sites and once signed, a site initiation visit was undertaken. Site-specific study information materials were prepared and issued to sites.

Once contracts were in place, a statement of works was drafted, describing the required technical interaction between WS and each NHS site, which was subsequently signed off by each NHS site, CU and WS. Following signing of the statement of works, WS would approach the lead for the maternity software system and agree to a timeline for testing and implementation of the 12 additional study-specific data fields. A further approval process was required prior to implementation of the new data items, a ‘Request for Change’, which, in many sites, needed information governance or senior information technology team sign off. As all MISs are bespoke, prior to release into the live clinical system, testing of the new data items in their parallel system was required at each study site. Once the local lead was satisfied, the new data items were compatible with their system and staff were trained in the study, a date for their introduction into the live clinical system was agreed. The final step was for WS to compare each local site’s data dictionary to their central held data dictionary to ensure that all data fields required for the study were available from the site. E-mails seeking formal study participation were sent to known prospective sites from August 2018. The first contract with study sites was signed in January 2019, and the last was signed in August 2020. The period from the date of a contract being signed between the NHS site and CU and new data fields being implemented in the NHS sites’ maternity software systems ranged from 1 to 11 months, with most sites (n = 17) taking between 2 and 6 months (Figure 1).

FIGURE 1 Duration of study set-up at sites. (Figure not reproduced here; see the original article.)

Governance approvals

To ensure results were generalisable, it was important that as few women as possible were excluded, and for this reason, the study was based on an opt-out design without individual consent. The extraction and access of deidentified medical records for research is a common approach for observational studies. However, matching maternity and NNU data required the transfer of identifiable, personal data outside of the NHS. Section 251 of the NHS Act 2006, as granted by the Health Research Authority (HRA) Confidentiality Advisory Group (CAG), allows the transfer of ‘confidential patient information without consent … without being in breach of the common law duty of confidentiality’.38

Data protection regulations were upheld, such as ensuring women were informed and aware of the research activities and had an opportunity to opt out without any impact on their health care. The intention was for all women giving birth at study sites after site opening to be informed about the study through methods selected by individual sites, including leaflets, posters, take-away cards and postings on websites (see Appendix 1). The application to CAG proposed that women could request to opt out of the study by informing their midwife at any point in pregnancy, around the time of birth or after consideration at home. Women who had given birth at study sites in the period from 1 January 2015 up to the date of the site opening could not opt out, as attempting to facilitate this would have been unfeasible.

Ethical approval

The protocol was approved by NHS Wales Research Ethics Committee on 18 September 2018 (18/WA/0291) and the transfer of identifiable data was approved by HRA CAG on 2 November 2018 (18CAG0153).

The approval committees accepted that women giving birth prior to site opening could not be informed that their data would be included, but they were concerned that an individual woman may feel reluctant to indicate her desire to opt out of the study to a midwife providing direct clinical care to herself or her baby. It was requested that we include the proactive provision of information to women using a pool during labour and provide alternative options for women to phone and/or e-mail the maternity unit to request that they are not included the study. For this purpose of opting out, sites were provided with posters for clinical areas and with ‘business card’-sized information to give to all women who used a pool. As part of the modification to the local systems, a study opt-out tick-box was added to the local electronic maternity record, which could be ticked by any member of the clinical team with access to a record until it was closed following discharge from postnatal community maternity care. This flag was visible to the data processor, who did not extract any data related to flagged records. A list of ethical amendments to the ethics are available in Appendix 2.

Data management

To create a single record combining individual maternity and NNU records, without identifiable data (personal identifiers), a method of linkage was required. To ensure that data received by CU from sites and the NNRD could be linked once the identifiable data had been removed, linking numbers were created by the statistician. The linking number comprised a code for the NHS organisation, the quarter and year of data, six randomly generated digits and a flag for the mother or baby (Table 1). Linking fields were unique to each mother/baby and were attached to each maternity record by WS.

The POOL Study data flow is shown in Figure 2. After maternity (maternal and neonatal) data were extracted on a site-by-site basis, the mother and baby linking fields were sent with each baby’s NHS number to the NNRD. The NNRD identified babies who had been admitted to a NNU by matching babies’ NHS numbers sent to those held within the NNRD. Each baby’s data, and the mother’s and baby’s ID (with the NHS numbers removed) were then sent to CU. The NNRD data were then linked to the maternity records via the linking field. It was not possible to identify or longitudinally link the pregnancies of women who had had more than one pregnancy during the study period.

FIGURE 2 POOL Study data flow. (Figure not reproduced here; see the original article.)

Data access, storage and cleaning

Access and transfer of data

Data held at study sites were sent electronically to CU via Secure Shell (SSH – a public-key cryptography to authenticate a remote computer and allow it to authenticate the user). This provided WS with access to a secure folder on CU servicers, which could only be accessed by the data manager (DM) and the account holders from WS. To ensure continued access when the COVID-19 pandemic began, additional access was granted to the study manager and statistician. To access data at participating sites, WS securely logged into the NHS site server and created the extracts. Extracts were then transferred to WS servers and formatted prior to transfer to CU and the NNRD. Once transfer had successfully occurred, extracts were deleted from the WS servers and the DM then moved the files to a folder which could not be accessed by WS staff. All NNRD data were sent to CU via the Imperial College London file exchange, a secure file transfer system.

Data storage

All data were stored in a secure study-specific data storage at CU, with access restricted to the DM and the statistician. Data sets were delivered in password-protected Comma Separated Values format and were moved to Statistical Product and Service Solutions (SPSS) (SPSS Inc., Chicago, IL, USA) via syntax. The SPSS files were then also stored in the secure space.

Data cleaning

A Microsoft Excel^® (Microsoft Corporation, Redmond, WA, USA) spreadsheet detailing data cleaning for all data received from WS was maintained. Each data file received contained 3 months (one-quarter) of data. The data cleaning process was detailed in a data cleaning plan (see Appendix 3). Once each file was imported into IBM SPSS, the initial check was to ensure records with pool use or no pool use could be identified. This quality check doubled up as a check to ensure that the data extracts had been created correctly. Data received from the NNRD were only received on two occasions, firstly during a pilot stage and at the end of data collection. The data cleaning process for NNRD was also detailed in the data cleaning plan; once all files were imported into IBM SPSS, the initial check was to ensure that all variables were received and then to match the identifiers with the WS data.

The second stage of data cleaning was changing the width of all variables so that they were consistent throughout the data set; this meant that when it came to merging the data sets, they could be merged without problems. Following this, the link IDs assigned were checked to ensure that they matched the prescribed format set out by the statistician. A check was made on each data batch received to ensure that they contained the correct variables and that the following variables derived by WS were within feasible or expected ranges: maternal age at delivery, gestational age at birth, parity, deprivation quintile, duration of rupture of the fetal membranes to birth, duration from head delivered to birth (in cases of shoulder dystocia) and duration of postnatal stay. Data outside feasible ranges or not as expected were set to missing. Frequencies were run on all variables and were visually checked for any outlying data or potentially identifiable information within free text fields. The identification of such data by the DM did not need to be reported as a data protection breech, as following a similar breech in the pilot, it was confirmed by the sponsor that the DM would be permitted to check for, and redact, identifiable data. The linking numbers were visually checked to ensure that they were in the correct format and that the mother and baby IDs were identical except the final number. Data sets were checked for duplicate records, and any duplicate record found was deleted. Data set quarters were merged into years and variable labels were added. Once all cleaning was completed for a site, the data files were merged to produce two files per site, one relating to pool use and one to non-pool use.

Full details of pilot work to assess the planned data management processes are described in Appendix 4.