Fig. 1
Flow chart of linkage between HES APC, HES A&E and birth cohort data. Notes: a Linking date was discharge date of A&E attendance (from HES A&E dataset) and date of hospital emergency admission (from HES APC dataset). b Quality control was conducted on linked A&E and APC records according to methodology guidelines from NHS digital. 90% were strong links, 10% good links and only 0.01% poor links (n = 67) which were removed as recommended by the guidelines. Emergency admissions via A&E department that did not link to an A&E attendance record showed 17% were coded as 21 (A&E department) and 83% as code 28 (other means). We prioritised information from APC dataset to indicate emergency admission via A&E department. c Direct emergency admissions were admission method codes 22 (GP,92%), 23 (bed bureau, 2%), & 24 (Consultant clinic, 6%) from HES APC dataset. These were removed prior to linkage and appended in afterwards. d All of the A&E attendances that were not linked to an APC emergency admission were assumed to be an attendance without an admission. e Of observations that did not link with UCL birth cohort, the majority were HESID’s from A&E data (63%). Whereas, 26% were HESID’s found in the A&E and APC linked data, and 11% from APC data only (direct admissions). 48% of the non-linking records were from 2012/13 and therefore may have been records from older infants that were not in the birth cohort (i.e. born before March 2012). This suggests that those not linked might be due to HESID data quality issues in the A&E dataset and HES records from older infants not in the birth cohort for the first year. f 701,680 infants had missing data on any risk factor (16%) and were excluded from the analysis sample; missing values were slightly more frequent in later study years (OR: 1.025, 95%CI: 1.024–1.046). g As the first year 2012/13 did not have full follow-up data from infant born in previous year, we dropped attendances and admissions in 2012/13. The rest of the financial years consisted of full follow-up