Table 1.

Description of potential scenarios leading to pregnancy episodes with no recorded outcome and scenario criteria applied

Scenario	How does this appear in the data?	Criteria used to determine if there is evidence in the data that an episode is consistent with the scenario in question
Problem 1: The women was pregnant at the time of the database record, but the outcome was not captured in CPRD primary care data.
1a. The woman was pregnant. She had a delivery, miscarriage or termination of pregnancy (TOP) in hospital or elsewhere and information either was not fed back to the general practice, or was fed back but not coded in the woman’s records.	There will be no evidence of an outcome in CPRD data up to 38 weeks* (for delivery) or up to 20 weeks (for miscarriage or TOP) after the first antenatal record for the pregnancy. However, there may be evidence of delivery/miscarriage/TOP in one of the linked HES APC data.	The woman must be eligible for linkage. There must be at least 1 day of overlap between the data coverage for each HES source and the pregstart+294 days (42 weeks) to give a maximum potential end date. There must be a record in HES of delivery or loss within 294 days (42 weeks).
1b. The pregnancy outcome was recorded in the primary care data but has no event date recorded alongside it and is therefore not picked up by the algorithm.	There will be an outcome code with missing eventdate† within 38 weeks after the first antenatal record of the pregnancy episode (using the systemdate† as a proxy for the event date).	There must be an antenatal code with missing eventdate† recorded with a systemdate† ≥294 days after pregnancy episode starts.
1c. The pregnancy outcome occurred before the patient was registered at their current practice or before the start of the practice up-to-standard follow-up (UTS). When the patient joined the practice, information was recorded about the pregnancy but not the outcome.	The pregnancy episode will occur before the start of the patient’s current registration and/or UTS.	Pregnancy episode end date must be <UTS date† OR ≤current registration date.
Problem 2: The women was pregnant at the time of the database record, but the pregnancy was still ongoing at the end of available follow-up in the database.
2a. The woman moved practices before the end of her pregnancy. If a patient transfers out of a CPRD practice, then follow-up is lost. OR The woman died before the end of her pregnancy.	There will be a transfer out date or death date (in either CPRD or the ONS mortality data) less than 38 weeks after the earliest antenatal record for the pregnancy episode.	The earliest of the woman’s transfer out date† or death date (in either CPRD or the ONS mortality data) minus pregnancy episode start date must be ≤294 days.
2b. The last collection of data from the practice was before the pregnancy outcome.	There will be a last collection date less than 42 weeks after the start of the pregnancy episode.	The woman’s last collection date minus pregnancy episode start date must be ≤294 days.
Problem 3: The patient was not pregnant at the time of the database record.
3a. A historical pregnancy was recorded retrospectively in the first few months after patient joins the practice. In this scenario, information about the pregnancy is recorded with the current date (by GP software default) rather than the date it occurred (different from scenario 1c). This is more likely to occur when a woman joins a practice and the GP may wish to record past pregnancy events which are relevant to her current clinical care.	The pregnancy episode will occur less than 1 year after the women’s current registration date. There will be a record of a pregnancy event which may be clinically useful for future care between the start and end of the pregnancy episode.	Pregnancy episode start date is <365 days after current registration date. There is a record of a pregnancy code from a list identified as likely to be recorded as useful pregnancy history information (online supplemental appendix 7). This must have an eventdate ≥pregstart† & ≤pregend.†
3b. The woman was not pregnant but was planning a pregnancy and discussed this with the GP, for example, due to other medical conditions which may complicate pregnancy.	The pregnancy episode will include a pregnancy advice code, for example, ‘67AF.00 Pregnancy advice for patients with epilepsy’.	The woman has antenatal codes identified as pregnancy advice codes (online supplemental appendix 8) with an eventdate† ≥pregstart† & ≤pregend.†
Problem 4: The pregnancy record belongs to another pregnancy episode in the Register.
4a. There was a delay in recording the outcome of a pregnancy by the practice. Thus, the outcome code has an eventdate† which is later than the true outcome date. The algorithm then calculates the Last Menstrual Period (LMP) date as being later than it was (figure 1, steps 5 and 6). Records which occurred early in pregnancy are then left unassigned to the pregnancy episode and appear as if belonging to a previous pregnancy episode which has no outcome recorded (figure 1, step 8).	As the pregnancy episode without outcome has been created from unassigned records at the beginning of the pregnancy, it will be followed by another pregnancy episode. There is unlikely to be more than a 3-month delay in outcome recording due to the mother attending the practice for postnatal checks and/or infant vaccinations. Therefore, there will be less than 12 weeks between the end of the episode with no recorded outcome and the start of the next pregnancy episode.	The woman must have >1 episode in the Pregnancy Register. Episodes with recorded outcome missing were eligible if they were not the last pregnancy episode for that woman. There must be ≤84 days (12 weeks) between the pregend† of the episode without outcome and the pregstart† of the woman’s next episode.
4b. The LMP is derived from information in the data and is estimated by the algorithm to have occurred later than reality (figure 1, steps 5). This may lead to a short pregnancy episode and unassigned codes before the estimated start of pregnancy. These are then grouped to form a pregnancy episode with no recorded outcome (figure 1, step 8).	The pregnancy episode without outcome will be followed by another pregnancy episode which will be less than 40 weeks long.	The woman must have >1 episode in the Pregnancy Register. The episode after the episode with missing outcome must have a startsource†=2, 4, 5 or 6 (online supplemental appendix 2). The length (gestdays) of the episode must be <280 days.
4c. If there are pregnancy records within 4 weeks before the estimated LMP, the identified pregnancy episode is shifted earlier in time by the algorithm (within plausible limits) to encompass those records (figure 1, step 6). This may leave unassigned pregnancy records which occurred shortly after the new estimated delivery date which will then be grouped to form a pregnancy episode with no recorded outcome (figure 1, step 8).	The pregnancy episode must not be the only pregnancy for this to apply. There will be another pregnancy episode which ends <8 weeks before the first antenatal record of the pregnancy episode without outcome for which the end has been adjusted by the algorithm.	The woman must have >1 episode in the Pregnancy Register. The episode before the one with recorded outcome missing must have an endadj†=2 (online supplemental appendix 2). The pregend† date for the episode with missing outcome must be ≤56 days (8 weeks) after the pregend† for that previous episode.
4d. The GP records a code relating to the patient’s pregnancy outcome history while the patient is pregnant. This is incorrectly identified by the algorithm as the outcome of the current pregnancy (figure 1, step 3). If the actual outcome is ≤25 weeks after for delivery or ≤12 weeks after for pregnancy losses, they will be grouped together as the same outcome. Subsequent antenatal records may then be grouped together to form a new pregnancy episode with no recorded outcome (figure 1, step 8).	The pregnancy episode must not be the patient’s first pregnancy. The pregnancy episode would be within 25 weeks after the previous outcome.	The woman must have >1 episode in the Pregnancy Register. The pregend† date for the episode with missing outcome had to be ≤175 days (25 weeks) after the pregend† for the previous episode.
4e. The outcome of the pregnancy episode has been misclassified as an antenatal event, for example, ‘Failed abortion’, ‘refer to TOP counselling’, ‘premature labour’, etc.	There will be an antenatal code which should have been an outcome code within 38 weeks after the first antenatal record of the pregnancy episode with recorded outcome missing.	There must be an antenatal record from a code list of potentially misclassified outcomes (online supplemental appendix 9) 266 days (38 weeks) of the firstantenatal† record.

*The first antenatal record is assumed to be recorded ≥4 weeks after the LMP as the woman is unlikely to know she is pregnant before then.

†Refers to a CPRD GOLD-specific variable, for example: pregend=the end of episode as defined by the algorithm; pregstart=the start of episode as defined by the algorithm; endadj=an indication that the end of the episode has been adjusted and how; startsource=which data were used to generate the start of the episode. These variables and others are defined in more detail in online supplemental appendix 2.

APC, Admitted Patient Care; CPRD, Clinical Practice Research Datalink; GP, general practitioner; HES, Hospital Episode Statistics; ONS, Office for National Statistics.