Table 4.

Studies providing intermediate levels of evidence

Study	Setting and study population	Design	Main interventions	Outcomes	Assessment of evidence
Arnow et al, 198218	Burns unit, 8 beds	Hybrid retrospective and prospective interrupted time series. Two phases of 8.5 months each	Phase 1: barrier precautions only Phase 2 nurse cohorting, handwashing education, increased screening	MRSA cases: 39 (phase 1); 6 (phase 2). No new cases occurred during periods when nurse cohorting was complete	Evidence supporting control by interventions. Variation in patient-bed days is a plausible alternative explanation. Regression to the mean effects are possible
Blumberg et al, 199522	Intensive care unit (20 beds), paediatric oncology (15 beds), and non-targeted areas of a tertiary care hospital (~3000 beds)	Hybrid interrupted time series. One year cohort study with non-equivalent concurrent controls, one year historical controls, and one year follow up	No control measures before study (historical controls). During intervention year eradication, screening and patient isolation (single rooms and staff cohorting) used in ICU and paediatric oncology. Measures largely abandoned in follow up year.	299 MRSA bacteraemias (43 in areas with interventions) Bacteraemias fell in the intervention year in targeted areas, then rose to intermediate levels in the post-intervention year. They increased each year in non-targeted areas	Evidence supporting control by interventions. Regression to the mean effects likely, and study vulnerable to changes in length of stay
Cox et al, 199527	One general hospital (hospital A) and two long stay or rehab hospitals (B and C). 750 beds in total	Retrospective interrupted time series. Three phases (at hospital A): 5, 4, and 11 months	Phase 1: single rooms and cohorting Phase 2 and 3: isolation wards Eradication and extensive screening throughout, including pre-admission from phase 2	83 MRSA infected patients, 334 colonisations. Hospital A: 1-4 infections/month in all phases. Last month of data collection showed very low colonisation incidence Hospital B: Continual detection of MRSA cases. No clear trend Hospital C: apparent elimination of MRSA 14 months after isolation ward opened	Evidence that combined measures in all phases failed to prevent sustained spread at hospital A. No evidence of control at hospital B. Weak evidence of control at hospital C. Interpretation of hospital B and C data difficult without colonisation on admission data due to interhospital transfers
Esveld et al, 199930	Dutch hospitals with index MRSA cases responding to a questionnaire. 231 returned questionnaires	Two year retrospective cohort study based on systematically collected survey data	Two cohorts defined by isolation policy Isolation cohort: index cases isolated on admission according to Dutch guidelines Non-isolation cohort: other isolation policy or delayed isolation	Isolation cohort: 4 out of 73 cases led to secondary spread Non-isolation cohort: 19 out of 95 cases led to secondary spread. Odds ratio 4.3 (95% Cl 1.3 to 18.2)	Evidence that immediate isolation contributed to control. Other plausible explanations include: differences in strains (prompt isolation was associated with strains originating abroad); differences in characteristics of cohorts and settings; and bias introduced by differential response rates to questionnaires
Jernigan et al, 199638	Neonatal intensive care unit, 33 beds	Hybrid retrospective and prospective interrupted time series. Two phases: 12 days and 9 months	Phase 1: contact isolation (gloves, gowns, masks and use of two bedded side-room if possible) Phase 2: as phase 1 plus eradication from selected patients; weekly screening; handwashing education	Total cases: 16 (5 in phase 1, 11 in phase 2). Large fall in incidence after additional control measures Relative risk of transmission from an unisolated compared to an isolated source 15.6 ((95% Cl 5.3 to 45.6), P<0.0001	Evidence supporting reduction in MRSA transmission by isolation measures Potential bias as no blinding to the isolation status of patients when assessing transmission sources Regression to the mean effects possible
Kac et al, 200040	Wound care centre, 51 beds	Prospective interrupted time series. Two phases: 3 months and 2 years	Phase 1: no measures Phase 2: gowns and gloves, handwashing education, feedback of infection rates, MRSA wounds dressed last	15 wound infections. Reduction in proportion of patients acquiring MRSA wound infections from 6/70 (9%) to 9/583 (1.5%)	Evidence that control measure reduced infection rates, but limited by short baseline and vulnerable to pre-existing trends (due to lack of time series data). Impossible to distinguish cross-infection and autoinfection
Murray Leisure et al, 199046	General hospital, 884 beds	Retrospective interrupted time series. two phases: 32 and 12 months	Phase 1: Single room isolation Phase 2: Isolation ward and changes to screening	177 new MRSA cases MRSA cases increased throughout phase 1 then fell to low levels in phase 2	Evidence consistent with control by isolation ward and screening, but change in numbers colonised on admission provides a plausible alternative explanation
Selkon et al, 198054	Teaching hospital, 1000 beds	Retrospective interrupted time series. Two phases of 5.5 years each	Phase 1: single room isolation Phase 2: isolation ward	965 MRSA infections MRSA infections increased before the opening of isolation ward, and subsequently decreased	Evidence consistent with control by isolation ward Changing antibiotic use provides a plausible alternative explanation