Methods	During a 3-month period from January 1977 to April 1977, adult patients undergoing cardiac surgery at the John Hopkins Hospital were randomised by odd or even history numbers to receive in the post-operative period either conventional blood bank transfusion therapy or autotransfusion of shed mediastinal blood
Participants	114 patients undergoing cardiac surgery were randomised to one of two groups: Group 1 (Autotransfusion group): n=63; M//F=41//22; mean (sd) age = 53.6 (10. 3) years Group 2 (Control group): n=51; M//F=32//19; mean (sd) age = 53.4 (10.0) years
Interventions	Group 1: Autotransfusion group (Sorenson autotransfusion system) received shed mediastinal blood processed by the Sorenson autotransfusion system (ATS). Blood collected in the ATS bags was considered suitable for autotransfusion only if 400mls or more was collected within 4 hours. If the rate of mediastinal bleeding was slow and 4 hours passed without 400mls volume being collected, this blood was not reinfused. Shed mediastinal blood was given in preference to stored bank blood when volume replacement was necessary. Group 2: Control group received only transfusions of stored bank blood. Autotransfusion was not performed.
Outcomes	Outcomes reported: amount of allogeneic blood transfused, total blood and blood component replacement, mediastinal blood loss, haematological variables, adverse events
Notes	Transfusion threshold: if Hct values were below 35% and left ventricular filling was judged to be adequate, whole blood and/or packed red blood cells were infused to restore intravascular volume. With higher haematocrit values and with low left ventricular filling pressures, patients received an infusion of colloid solution or crystalloid solution (Ringer’s lactate)
Risk of bias
Bias	Authors’ judgement	Support for judgement
Random sequence generation (selection bias)	High risk	Method used to generate allocation sequences was inadequate.
Allocation concealment (selection bias)	High risk	Method used to conceal treatment allocation was inadequate.
Blinding (performance bias and detection bias) All outcomes	High risk