Table 2.

Description of potential scenarios leading to conflicting episodes and scenario criteria applied

Scenario	How does this appear in the data?	Criteria applied to pairs of conflicting episodes to determine if there is evidence in the data that the pair is consistent with the scenario in question
Problem 1: Both pregnancies are true, but one is a current pregnancy and one is a historical pregnancy.
1a. The GP records a past delivery during a current pregnancy >25 weeks before the true delivery of that pregnancy. OR a past pregnancy loss >12 weeks before the actual loss of that pregnancy.	Both pregnancies will have the same outcome type. Evidence of current pregnancy codes would be expected to fall within the second pregnancy.	The outcome combination of the two episodes must be delivery/delivery or loss/loss (see online supplemental appendix 10 for outcome classifications). The second episode had an antenatal code from a list deemed likely to only be recorded if the patient was currently pregnant (online supplemental appendix 11) OR a scan record in the HES DID data between firstantenatal* and pregend*.
1b. If a patient has a record relating to a previous loss recorded during a pregnancy ending in delivery or vice-versa, then conflicting episodes will be created by the algorithm. The algorithm first generates episodes for consecutive deliveries; it then does the same thing for pregnancy losses. There is no step in the algorithm to check that the loss episodes do not coincide with the delivery episodes (figure 1, steps 1–6).	The conflicting pregnancies must consist of one loss and one delivery. Evidence of current pregnancy codes would be expected to fall within the second pregnancy.	The outcome combination of the two episodes must be delivery/loss or loss/delivery (see online supplemental appendix 10 for outcome classifications). The second episode had an antenatal code from a list deemed likely to only be recorded if the patient was currently pregnant (online supplemental appendix 11) OR an antenatal scan record in the HES DID data between firstantenatal* and pregend.*
Problem 2: Both pregnancies are historical.
2a. A patient joins a new practice (or has another reason for a full obstetric history to be taken) and has information on historical pregnancies recorded with the current date rather than the actual date of the event. Losses and deliveries recorded on the same date will result in conflicting episodes in the Register as different outcome types are generated separately by the algorithm (figure 1, steps 1–5).	The conflicting pregnancies must consist of one loss and one delivery. The pregnancy end dates will be the same for both pregnancies. Both pregnancies are likely to be <1 year after the patient’s current registration date. We would not expect to find codes indicating current pregnancy.	The outcome combination of the two episodes must be a delivery and a loss. The pregend* dates must be the same. There must be no antenatal codes relating to current pregnancy (online supplemental appendix 11) or HES DID antenatal scan recorded between the firstantenatal* date and the pregend* date of either episode.
Problem 3: Both pregnancies are true and current but the gestation of the second pregnancy estimated by the algorithm is too long.
3a. The woman has two pregnancy losses which are >8 weeks and <12 weeks apart. The second pregnancy has no information about gestation recorded so the algorithm applies a default of 12 weeks and the episodes overlap.	Both conflicting pregnancies must be losses. The maximum overlap between the two pregnancies must be 4 weeks. Evidence of current pregnancy codes could be found in either pregnancy.	The outcome combination of the two episodes must be two losses. The pregend* for the first episode must be ≤28 days after the pregstart* of the second episode.
3b. The woman has two pregnancies close together and the second pregnancy ends in delivery. If the information on the Last Menstrual Period date (LMP) in the data of the second pregnancy is wrong, then the algorithm may generate the start too early resulting in an overlap.	The second pregnancy must be a delivery and have no information about gestation in the data. The overlap must be <15 weeks (otherwise the two outcomes would be <25 weeks apart and would have been grouped as one; see figure 1, step 3). There may be evidence of current pregnancy codes in either pregnancy.	The outcome of the second episode must be a delivery. The startsource* of the second episode must not be equal to 4 or 5 (online supplemental appendix 2). The pregstart* of the second episode must be <105 days (15 weeks) before the pregend* of the first episode.
Problem 4: The pregnancy is true and current but is split into separate episodes by the rules of the algorithm.
4a. The GP records further information about a pregnancy outcome >25 weeks after the delivery date for pregnancies ending in delivery OR >8 weeks but <12 weeks for pregnancies ending in loss. The algorithm assumes this further information is a different pregnancy and generates a new episode, which may overlap with the ‘true’ episode.	Both pregnancies must be of the same outcome type. Evidence of current pregnancy codes would be expected to fall within the first pregnancy.	The outcome combination of the two episodes must be delivery/delivery or loss/loss (online supplemental appendix 12). The first episode had an antenatal code from a list deemed likely to only be recorded if the patient was currently pregnant (online supplemental appendix 11) OR a scan record in the HES DID data between firstantenatal* and pregend*.
4b. The GP records further antenatal information about a pregnancy after delivery or pregnancy loss. This will then be used to generate a new pregnancy without outcome episode by the algorithm. If the code is within 4 weeks of the end of the true pregnancy episode, the two will overlap.	The first pregnancy must be a pregnancy with an outcome recorded in the data. The second pregnancy must be a pregnancy without outcome which consists of one antenatal code not related to a scan.	The first episode must have outcome=1–10 in the Register (online supplemental appendix 2) and must have endadj*=0. The second episode must have no recorded outcome (outcome=13). The second episode must have a gestdays*=28 (likely to consist of one code) and there must NOT be a scan code (online supplemental appendix 13) with an eventdate*=pregend* of the second episode.
4c. The patient has a follow-up scan after a pregnancy loss. This is recorded in the data by the GP as an antenatal scan. The algorithm then creates a second pregnancy episode based on the antenatal scan code which becomes a pregnancy without outcome in the Register.	The first pregnancy must be a pregnancy loss. The second pregnancy must be a pregnancy without outcome which consists of one antenatal code related to a scan.	The outcome combination of the two episodes must be loss/missing. The second episode must have a gestdays*=28 (likely to consist of one code) and there must be a scan code (online supplemental appendix 13) with an eventdate*=pregend* of the second episode.
4d. The GP records information about a pregnancy but no information about the outcome. If records relating to this pregnancy are more than 6 weeks apart, they will be turned into multiple episodes. Once estimated start dates are generated for these episodes based on the data recorded (figure 1, step 8), episodes may overlap. For example, if there is gestational information included in the second episode, the start of this episode will be assigned before the start of the previous episode resulting in a nested pregnancy episode.	Both pregnancies must be pregnancies without outcome in the Register. The end of the first pregnancy must be greater than 6 weeks before the first antenatal of the second.	The outcome combination of the two episodes must be missing/missing. The pregend* of the first episode is >42 days before the firstantenatal* date of the second episode.
4e. The first pregnancy episode ended in delivery and has been shifted backwards by the rules of the algorithm leaving unassigned late pregnancy or third trimester records. These records will then be identified by the algorithm as end of pregnancies (figure 1, step 6) and new conflicting episodes will be created.	The first pregnancy must be a pregnancy with a delivery outcome recorded in the data. The end of the first pregnancy must have been adjusted. The second pregnancy must be a pregnancy where the outcome is based on a late pregnancy or third trimester record.	The first episode must have a delivery outcome code and endadj* variable not=0. The second episode must have outcome=11, 12 or 13.

*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.

CPRD, Clinical Practice Research Datalink; DID, Diagnostic Imaging Dataset; GP, general practitioner; HES, Hospital Episode Statistics.