National Pregnancy in Diabetes Audit 2021 and 2022 (01 January 2021 to 31 December 2022)

This page describes how the data is collected, processed and analysed before publication.

To minimise the burden of data collection and submission the audit data is partly collected from antenatal diabetes services and partly obtained by linking the directly collected data to:

• National Diabetes Audit (NDA) data
• Hospital Episode Statistics (HES) data
• Patient Episode Database for Wales (PEDW) data
• Neonatal Research Database

A full list of data items collected by the audit is available from www.digital.nhs.uk/npid

Antenatal diabetes services provide each woman eligible for inclusion in the audit with a patient information leaflet and discuss the audit with them.

In England, the data is collected under direction by NHS England. It is assumed that all women who are eligible for the audit, from all eligible sites, are entered. However, this can not be verified as this information is not collected anywhere else, therefore case ascertainment cannot be calculated accurately.

Welsh data is collected under an agreement with the Welsh government. It is assumed that all women who are eligible for the audit, from all eligible sites, are entered. However, this can not be verified as this information is not collected anywhere else, therefore case ascertainment cannot be calculated accurately.

All data collected by the NPID audit should already be stored in clinical records; no additional data should need to be collected.

Staff in antenatal diabetes services can use the NPID audit data collection form  if they wish to collate the relevant data prior to entering it electronically.

The data is electronically transferred to NHS England by manual data entry into a secure online system called the Clinical Audit Platform (CAP). The data is entered by registered users at antenatal diabetes services. The data entry system checks each record for obvious errors and only records that pass these checks can be saved and submitted successfully.

Data can be collected and entered throughout the year. Services are asked to submit all records of pregnancies with a recorded outcome ending in a calendar year by mid-February in the following year. Any records submitted after the deadline will not be included in that year’s report but will be added to the cumulative dataset.

Following the submission deadline, each service receives details of data quality checks run on their data, flagging records to review and potentially amend before a final submission deadline in early April.

The number of errors in the data has reduced greatly since internal consistency checks of each record were introduced at the point of data entry in January 2015, (for example that the HbA1c measurements within pregnancy are between 0 and 40 weeks prior to the pregnancy end date). However some more complex errors are only detectable when the dataset is checked as a whole (for example, that the same outcome data has been attached to more than 1 pregnancy).

Following the final submission deadline in April, data quality checks are re-run, and either:

• Where the error would affect a lot of the analysis, the record is rejected. Across the 2014 - 2022 dataset 771 records were rejected (leaving 37,490 pregnancies).
• Where the other data in the record can still be used for analysis, the data is cleaned by setting the value(s) assumed to be incorrect to missing.

A full list of the rejection reasons and cleaning actions is included in Appendix 1.

For each pregnancy record in the NPID data, lower super output area (LSOA) of residence was obtained from the latest known LSOA recorded in the NDA data.

As mother’s diabetes type was added to the NPID data collection as a mandatory data item from 1 January 2015, this was also obtained from NDA data for the 2014 NPID audit records, again by linking to the most recent relevant NDA record. Where the diabetes type entered on the NPID system was not known, NDA linkage was also used in order to establish a known diabetes type for as many women as possible.

Ethnicity was taken from the most consistently recorded ethnicity across all NDA audits.

Year of diabetes diagnosis was obtained from the earliest NDA record available for each woman in NPID, excluding diagnosis dates that were after the start of the pregnancy or before the woman’s date of birth as recorded in the NPID data.

Onset of labour and mode of delivery were obtained by linking the NPID audit data to HES and PEDW data.

As HES/PEDW data are published for financial years and the NPID audit reports by calendar year, only NPID pregnancies ending in 2020 or 2021 can currently be linked to available HES/PEDW data.

A hospital episode record was accepted as a match when the pregnancy end date in NPID was up to 7 days earlier than the episode start date or up to 7 days after the episode end date.

Details of hypoglycaemia and DKA (diabetic ketoacidosis) episodes during pregnancy were also obtained by linking to HES and PEDW data. A hospital episode record was accepted as a match when the episode started no more than 40 weeks before the estimated delivery date and on or before the pregnancy end date.

Where hypoglycaemia and DKA diagnosis codes were recorded on the same episode, this episode was counted as both an episode with hypoglycaemia and an episode with DKA. This is consistent with the method for counting complications used by the NDA.

The diagnosis codes used to determine if there was a hypoglycaemia or DKA episode during the women’s pregnancy are as follows:

Hypoglycaemia

• E16.0 Drug-induced hypoglycaemia without coma in any diagnosis field and Y42.3 Insulin and oral hypoglycaemia (antidiabetic drugs) in a secondary diagnosis position

OR (in any diagnosis field)

• E161 Other hypoglycaemia
• E162 Hypoglycaemia, unspecified

Diabetic ketoacidosis (DKA)

The following codes in any diagnosis field

• E10.1 Type 1 diabetes mellitus with ketoacidosis
• E11.1 Type 2 diabetes mellitus with ketoacidosis
• E13.1 Other specified diabetes mellitus with ketoacidosis

E14.1 Unspecified diabetes mellitus with ketoacidosis

We used NNRD to obtain information on admissions to neonatal care for babies born to mothers with diabetes.

The NNRD data came in 2 files - a baby file and an admissions file. The data request was based on mother’s NHS number, which meant that the supplied data tables had births going back to around 2004, so all records relating to births prior to 2014 were excluded.

Data from the 2 tables was combined as part of the linkage process. The majority of babies had a single episode spell, but where there was more than 1 episode, the admission date/time for the first episode and the discharge date/time for the final episode were used to calculate length of stay.

Records were linked to NPID if:

• The baby’s NHS number matched
• The mother’s NHS number matched
• The baby's date of birth matched (to within 7 days)
• A length of stay could be calculated

Around 34% of the infants from NPID pregnancies were found in the NNRD data between 2021 and 2022. This has reduced from around 40% between 2014 and 2020. This rate was fairly consistent across NPID years prior to 2021.

More details of the NNRD are available. 

Local measures of deprivation are produced by the Ministry for Housing, Communities and Local Government (MHCLG) for England and the Welsh government for Wales, with the measures for each country considering different factors and using different calculation methods.

For the NPID audit report, England and Wales data have been analysed together. In order to assign a deprivation quintile to each pregnancy record based on the LSOA of residence of the mother (where this could be obtained from linking to NDA data), a ‘combined’ deprivation score was derived and ranked into quintiles. This combined score was based on an equally weighted combination of the individual scores for the employment and income indices.

Birthweight centiles are used to adjust the actual birthweight of babies in line with maternal factors such as ethnicity, height and weight as well as gestational age at delivery.

Birthweight centiles were calculated for all singleton babies in the NPID dataset where the gestation at delivery and birthweight was known, using the Gestation Related Optimal Weight (GROW) centile tool*.

A baby is described as large for gestational age (LGA) if its birthweight is above the 90th centile based on gestation and maternal characteristics. In the general population, 10% of babies would be expected to be above the 90th centile.

An antenatal diabetes service is counted as participating in the audit if at least 1 completed pregnancy record has been submitted by that service for the audit period. A full list of participating services is included in the service level dashboard, which can be found among the supporting documents on our publication page:

The national report mainly shows data for women with type 1 diabetes and women with type 2 diabetes. The ‘All’ diabetes analysis puts together these 2 groups along with Maturity Onset Diabetes of the Young (MODY) or ‘Other’ diabetes.

Similarly, the service level dashboard breaks down type 1 and type 2 groups, and also groups this data together with MODY and other diabetes type in an ‘All’ diabetes group in order to provide some basic feedback to services that do not see a large volume of patients.

Combining these 2 groups as ‘All’ diabetes can be misleading because the care pathways and physiology are often different for women with type 1 diabetes to those women with type 2 diabetes, and the ‘All’ diabetes figure will be a reflection of the mix of diabetes types seen by the service rather than the overall quality of care.

The number of women with MODY or ‘Other’ diabetes is too small to provide analysis for this group at service level.

The analysis of the 3 years’ of NPID data relating to pregnancies ending in 2014, 2015 and 2016 was done for the first time in the report published in 2017. At that time changes over time were not reported as there were no changes to be seen. The analysis published in 2019 covered 5 years’ of NPID data and showed measures over the time period. Again, for the most part, there were few changes over the 5 year period. Where measures were compared over time in the 2020 report (published in 2021), again no changes were seen. For time series analysis within the 2020 report, data for 2014 to 2018 NPID years were combined, and 2019 and 2020 were shown separately.

Within the 2022 report (published in 2023), time series data have often been combined for 2014 to 2018, 2019 and 2020, and 2021 and 2022; or 2014 to 2020, with 2021 and 2022 shown separately where differences were found in trends for these years.

Each table or chart uses all records for which the relevant data is valid and not missing in order to include as many records as possible.

This means that the denominator for percentages varies between the charts and tables.

For example, pregnancies where gestation is unknown because estimated delivery date has been removed during cleaning will be excluded from analysis where gestation is relevant, such as birthweight centiles, but will be included in other analysis where the gestation at delivery is not needed, such as whether the mother was taking 5mg folic acid prior to pregnancy.

For non-mandatory data items such as HbA1c measurements, the denominator will be much lower reflecting the number of records in which this data has not been entered.

Quartiles and 10th and 90th centiles

The national report shows the variation in some measures, such as length of stay in neonatal care, and summarises the extent of variation using the median, centiles and quartile values.

Arranging all the values in order, the median is the middle value. The lower quartile is the value below which the bottom 25% of data values lie and the upper quartile is the value above which the top 25% of data values lie.

The 10th centile is the value below which the bottom 10% of data values lie, and the 90th centile is the value above which the top 10% of data values lie.

Testing for significant difference between 2 proportions

Where the report explicitly compares a proportion for 2 groups, such as the percentage of women with type 1 diabetes taking 5mg folic acid compared to the percentage of women with type 2 diabetes taking 5mg folic acid, the difference between the 2 groups has been tested to determine whether it is a significant. An independent sample (two-sample) z-test has been used with a with a p-value of 0.05.

Confidence interval for a proportion

The 95% confidence intervals for the stillbirth, neonatal death and congenital anomaly rates were calculated using Byar’s method as described in ‘Analytical Tools for Public Health: Commonly used public health statistics and their confidence intervals’** as

where O is the observed number of events and n is the rate denominator.

**http://webarchive.nationalarchives.gov.uk/20170106081009/http://www.apho.org.uk/resource/view.aspx?RID=48617

The service level dashboard published with this report includes data for pregnancies ending in 2020-2022. The number of records for each service will in part depend on when they started participating in the audit. Some services that see a smaller number of patients per year may have more records included than a larger service because they have participated for all 3 audit years while the larger service has only joined during 2022.

The average number of pregnancy records per service in the report is 80.

Service level data is only published where a service has submitted 10 or more completed pregnancy records across the 3 years. As the total number of records for each service is relatively small (an average of 80 records per service), an apparently large difference in percentages between services may only reflect a small difference in the numerator. The service level report contains a list of participating services.

The England and Wales figures in the service level report are similarly based on 3 years of data so will differ from the 2022 figures in the national report.

Disclosure control has been applied to mitigate the risk of patient identification. We have used the standard method of rounding used across the audits which fall under the umbrella of the NDA.

Zeros are reported, and all numbers are rounded to the nearest 5, unless the number is 1 to 7, in which case it is rounded to 5. This allows for more granular data to be made available. Rounded numbers are used to calculate percentages therefore numbers may not sum as expected. Percentages are not calculated where the rounded denominator is 20 or less, and where numbers are small percentages are volatile and should be treated with caution.

At national level, this makes virtually no difference to the resultant percentages. For some services, where the numbers are small, this rounding can have a relatively large impact on the resultant percentages.

A small number of records with data entry errors were rejected because the error meant that the correct complete record could not be identified or mandatory data needed to allocate a record to the correct service was missing.

The total number of records rejected from the 2014-2022 dataset was 771.

A record with pregnancy outcome data completed was rejected if:

• 2 outcomes with the same pregnancy end date were attached to different pregnancy records for the same woman
• The same baby NHS number was entered in outcome data for more than 1 pregnancy
• The same baby NHS number was entered more than once in outcome data for the same pregnancy
• 2 outcomes with substantially different pregnancy end dates were attached to the same pregnancy
• Booking or delivery hospital was missing – these are mandatory data items needed to allocate each record to a service

Following requests from submitters, the data entry system was altered from January 2015 to allow the ‘Alive at 28 days’ question to be completed separately from the other outcome data. However, ‘Alive at 28 days’ was not completed for a number of otherwise complete and valid records which were also rejected.

Dates

The table below lists the cleaning rules applied to the NPID audit data prior to analysis in order to use as much data as possible from each record where errors remained after the data quality review by services.

Where the dates within a record were inconsistent, the pregnancy end date and pregnancy outcome have been assumed to be correct, and dates that were inconsistent with this have been set to missing.

The order of date checking means that where an estimated delivery date has been set to missing, checks which compare other dates to estimated delivery date will result in the removal of those dates.

Table 1: Data cleaning rules for dates used for 2022 NPID audit report