Table 2.

Selection, performance, detection and attrition bias

Type of bias	Cause	Studies vulnerable	Measures taken to identify or prevent bias
Selection bias	Differences in intervention groups on study entry	Studies without randomisation (39 studies)	Four of 35 interrupted time series studies where isolation or screening changed presented data allowing comparisons of patient characteristics between phases26,34,40,57
			Two described unquantified changes in case mix.43,46 One study presented partial adjustment for confounders30
Performance bias	Differences in care for patients between treatment groups, apart from interventions under investigation	Studies where specified aspect of care was not under investigation:
	Differences in antibiotic prescribing	31 interrupted time series studies	Four presented details of antibiotic use.18,22,36,63 Two alluded to unquantified changes46,54
	Differences in lengths of stay	29 interrupted time series studies	Changes in length of stay could be assessed in four33,36,47,61
	Differences in bed occupancy and staff workload	31 interrupted time series studies	Comparisons of bed occupancy between phases possible in four18,24,47,58 Changes in staffing levels or workloads could be assessed in five18,24,26,32,47
Detection bias	Differential outcome assessment between intervention groups:	All studies:	Three reported some blinding of outcome assessors34,47,57
	Differences in diagnosis of infections	26 studies	16 specified diagnostic criteria. 14 reported MRSA bacteraemias
	Differences in screening practices	10 studies with colonisation data only	In all cases screening effort either reported not to have changed or to have changed in opposite direction from outcomes, suggesting screening effort could not explain the changes
Attrition bias	Differential loss to follow up between treatment groups.	Studies where outcomes are infections (26 studies) and with substantial changes in length of stay	None: no studies followed up patients after discharge to detect hospital acquired infections
	Since hospital acquired infections may first become apparent after discharge, changes to length of stay could lead to attrition bias