Immortal time bias for life-long conditions in retrospective observational studies using electronic health records

Table 1 Characteristics of the study population using different methods to handling immortal time bias

CHARACTERISTICS	EXPOSED		NON-EXPOSED
CHARACTERISTICS	People with intellectual disabilities N (%) / median (range)		People without intellectual disabilities N (%) / median (range)
Method 1
Total	33,867	(100.00)	980,586	(100.00)
Age (years) at baseline:	29.0	(10–102)	34.0	(10–108)
Length of observation time (years)	6.5	(< 0.1–19.7)	5.0	(< 0.1–19.7)
Method 2
Total	33,867	(100.00)	980,586	(100.00)
Age (years) at baseline:	31.0	(10–102)	34.0	(10–108)
Length of observation time (years)	4.6	(< 0.1–19.7)	5.0	(< 0.1–19.7)
Method 3^a
Total	33,867	(100.00)	338,670	(100.00)
Age (years) at baseline:	31.0	(10–102)	34.0	(10–108)
Length of observation time (years)	4.6	(< 0.1–19.7)	3.6	(< 0.1–19.7)
Method 4^b
Total	33,244	(100.00)	980,586	(100.00)
Age (years) at baseline:	33.0	(10–106)	34.0	(10–108)
Length of observation time (years)	2.2	(< 0.1–19.6)	5.0	(< 0.1–19.7)
Method 5^c
Total	33,867	(100.00)	991,879	(100.00)
Age (years) at baseline:	31.0	(10–102)	34.0	(10–108)
Length of observation time (years)	4.6	(< 0.1–19.7)	5.0	(< 0.1–19.7)

^an = 641,916 individuals from the unexposed population were discarded under Method 3 because they were not matched
^bn = 623 individuals from the exposed population were excluded from this analysis under Method 4 because they entered on or after they were censored/died (i.e. system date linked to the diagnosis was after date of leaving/death)
^cIndividuals could contribute to both the exposed and unexposed populations under Method 5, as reflected in the baseline values

ISSN: 1471-2288