Skarbinski 2009 KEN.

Methods	Cluster RCT; randomized by health facilities Stratified random selection of facilities, by transmission settings (high/low) and facility type (hospitals, health centres and dispensaries) Took into account a design effect of two in sampling Reference slide taken. Results not reported Trial lasted four months
Participants	Inclusion: age ≥ 5 years, irrespective of condition Number of participants randomized: Intervention arm: 799 Number analysed for primary outcome (prescribing of antimalarials): 669 (16.3% loss)
Interventions	Intervention: RDTs for fever patients ≥ 5 years Control: Presumptive treatment of fever Boh groups received training and held guidelines RDT performed by health workers Health workers complied with guidelines partially: 41% of participants with negative RDT results received antimalarials
Outcomes	Primary outcomes: Fever patients prescribed ACT Microscopy negative patients prescribed ACT Secondary outcomes: RDT negative patients prescribed ACT; and RDT positive patients prescribed ACT Patients prescribed ACT presumptively Patients with known alternative diagnosis receiving ACT
Notes	Country: Kenya RDT: paracheck Setting: all referral levels of facilities, 60 in total, 30 in each arm Transmission: Hyperendemic or holoendemic at some sites and low and seasonal at others. Dates: June to September 2006 Funding: USAID
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	High risk	Systematic allocation.
Allocation concealment (selection bias)	Unclear risk	Although probabilistic sampling was used in selecting the participating health facilities, the participants had foreknowledge of intervention assignments.
Blinding of participants and personnel (performance bias) All outcomes	High risk	Both the trial participants and personnel were aware of intervention allocation.
Blinding of outcome assessment (detection bias) All outcomes	High risk	Both the trial participants and personnel were aware of the diagnoses and prescriptions.
Incomplete outcome data (attrition bias) All outcomes	High risk	Per protocol analysis; loss to follow‐up was high, more at the intervention facilities (20.2%) than at the control facilities (11.2%).
Selective reporting (reporting bias)	Low risk	Reported on all pre‐specified outcomes (prescribing of ACT); did not explicitly report on overall antimalarial prescribing, but the summary data were available for inclusion in the review.
Other bias	Low risk	Baseline imbalance minimised by stratifying facilities by level and randomly selecting within each level. Summary data were adjusted for baseline imbalance. Results could be biased towards the null, because some facilities in the comparison arms had RDT. Loss of complete clusters: not reported.