Nthumba 2010.

Methods	Cluster‐randomised controlled cross‐over trial Generation of random number sequence: appropriate Allocation concealment: no details given but considered unlikely to be an issue in cluster‐randomisation Blinding: outcome assessors A priori sample calculations: yes Antisepsis protocol: yes Withdrawals: details of patients without postdischarge surveillance Intention‐to‐treat analysis: no Clear inclusion or exclusion criteria: yes
Participants	66 surgeons and trainees; 3317 patients
Interventions	Group 1 ‐ plain soap and water: 4‐5 min clean running water and plain soap; sterile cotton hand towel dry. 5 clusters (n = 1682 patients) Group 2 ‐ As group 1 before first procedure of day and subsequently in case of visible soiling, then alcohol‐based handrub (75% isopropyl alcohol, 1.45% glycerol, 0.125% hydrogen peroxide) for 3 min and kept wet (7‐10 ml per preparation)
Outcomes	SSI (defined using modified US Centers for Disease Control and Prevention definitions for nosocomial infection) detected by tours of hospital wards; reviews in outpatient clinic; telephone contact: diagnosis established jointly by study collaborators
Notes	—
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Surgical hand preparation procedures were assigned randomly to the 6 participating operating theatres by tossing a coin, with a cross‐over every 2 months. There was no indication of baseline imbalance for important variables such as type of surgery, contamination level of the surgery or use of antibiotic prophylaxis.
Allocation concealment (selection bias)	Low risk	Allocation was decided by toss of a coin. Not clear who undertook this process and if it was concealed from the sites. Given that this was a cluster trial with cross‐over, the potential for bias stemming from allocation concealment was limited. As the Cochrane Handbook for Systematic Reviews of Interventions notes, "Cluster‐randomised trials often randomise all clusters at once, so lack of concealment of an allocation sequence should not usually be an issue."
Blinding (performance bias and detection bias) All outcomes ‐ Blinding participants	High risk	Personnel were aware of allocation as procedures differed in obvious ways. Compliance by surgical teams was determined by observation of practices. A trained observer who did not belong to the surgical team checked whether each sink had the correct hand preparation and whether all surgeons (including visiting staff) followed the recommended hand preparation procedures.
Blinding (performance bias and detection bias) All outcomes ‐ Blinding care givers	Unclear risk	No information reported
Blinding (performance bias and detection bias) All outcomes ‐ Blinding outcome assessors	Low risk	Diagnosis, documentation and determination of SSI by personnel without knowledge of allocation. "SSI was diagnosed . . . and documented by a trained nurse who visited the surgical wards three to four times each week during the 30 days after surgery. Patients discharged before this were reviewed in the outpatient clinic or contacted by telephone . . . The nurse was blinded to the method of surgical hand preparation. The diagnosis of SSI was established jointly by the study collaborators; differences in SSI ascertainment were resolved by consensus without knowledge of the study allocation of the patient."
Incomplete outcome data (attrition bias) All outcomes	Low risk	3722 patients underwent a surgical procedure in the operating theatre, and 3317 were included in the 10 study clusters. Postdischarge surveillance data could not be obtained for 184 patients (5.5%).
Selective reporting (reporting bias)	Low risk	The primary objective of this cluster‐randomised, cross‐over trial was to compare the efficacy of plain soap and water with alcohol‐based handrub, using SSI rates as the main outcome measure. The feasibility and affordability of the local production of an alcohol‐based handrub was also investigated, together with an assessment of its acceptability among healthcare workers. All specified outcomes were reported.
Other bias	Low risk	There was no evidence of other sources of bias. Clustered nature of the data was taken into account in the trial.