Table 6.

Issues with different approaches to dealing with uncertain episodes and recommendations

Example uses	Issues with a highly specific approach: excluding all uncertain episodes	Issues with a highly sensitive approach: including all uncertain episodes	Recommended tailored approach: including or excluding uncertain episodes based on scenario criteria
Vaccine uptake study	Underestimate of uptake during pregnancy	Overestimate of uptake during pregnancy where historical episodes are included	Consider using episodes without recorded outcome which continue after data follow-up to maximise the capture of exposure events. Consider using linked data to obtain additional outcomes. Exclude episodes which are likely to be derived from historical data based on our described scenarios.
Drug/vaccine safety study	Underestimation of pregnancies ending in loss Underestimation of pregnancy complications	Misclassification of exposure status Overestimation of outcomes	Consider using linked data to obtain additional outcomes restricting the study population to those patients eligible for linkage. Exclude episodes which are likely to be derived from historical data based on our described scenarios. Consider merging conflicting episodes which are consistent with problem 4 and adjusting the timing accordingly (deciding which of the outcomes is likely to be the true outcome based on the scenarios we have described and then estimating a start date. This should be based on a combination of the patient’s antenatal records and default duration dependent on outcome type3). Consider ensuring pregnancy start is at least 9 months before the last data collection date to allow for attainment of outcomes.
Ascertaining pregnancy history	Underestimation of parity Underestimation of certain pregnancy events Underestimation of pregnancies ending in loss	Overestimation of parity	Consider using linked data to obtain additional outcomes restricting the study population to those patients eligible for linkage. Exclude episodes which are likely to be derived from historical data based on our described scenarios. Consider ensuring pregnancy start is at least 9 months before the last data collection date to allow for attainment of outcomes.
Excluding pregnant women from a study cohort	Reduction in potential study population	Potential misclassification of pregnancy status Potential errors in pregnancy timing	Consider merging conflicting episodes which are consistent with problem 4 and adjusting the timing accordingly (deciding which of the outcomes is likely to be the true outcome based on the scenarios we have described and then estimating a start date. This should be based on a combination of the patient’s antenatal records and a default duration dependent on outcome type3). Consider using linked data to obtain additional outcomes, restricting the study population to those patients eligible for linkage. Exclude episodes which are likely to be derived from historical data based on our described scenarios.