Table 2.
Comparison of prediction errors for the five sampling strategies, for n = 12 (n = 4 per hospital type)
| Strategy | Prediction error at hospital level | Prediction error at national level | ||||
|---|---|---|---|---|---|---|
| RBC | FFP | PLT | RBC | FFP | PLT | |
| LARG | 40 % | 28 % | 35 % | 40 % | 17 % | 20 % |
| MAXVAR | 15 % | 25 % | 28 % | 2 % | 1 % | 17 % |
| RAND (median; mean, 95 % centiles) | 19 %; 20 % (17 %-26 %) | 30 %; 31 % (25 %-42 %) | 32 %; 32 % (25 %-40 %) | 5 %; 6 % (0 %-18 %) | 10 %; 11 % (1 %-30 %) | 9 %; 10 % (0 %-28 %) |
| REGVAR (median; mean, 95 % centiles) | 19 %; 20 % (16 %-26 %) | 29 %; 30 % (25 %-41 %) | 31 %; 31 % (24 %-39 %) | 6 %; 7 % (0 %-18 %) | 9 %; 11 % (1 %-30 %) | 9 %; 10 % (0 %-27 %) |
| 2REG (median; mean, 95 % centiles) | 20 %; 21 % (17 %-29 %) | 34 %; 49 % (27 %-135 %) | 35 %; 44 % (30 %-43 %) | 5 %; 6 % (0 %-17 %) | 11 %; 26 % (1 %-118 %) | 8 %; 16 % (0 %-76 %) |
LARG = largest hospitals, MAXVAR = maximum variation in number of RBCs, RAND = random, REGVAR = regional variation, 2REG = two regions, RBC = red blood cell products, FFP = fresh frozen plasma products, PLT = platelet products. Output for RAND, REGVAR and 2REG is based on the average of 10 times 1000 simulations and accompanied by 95 % centiles. Outcomes are the prediction error at hospital level (summed absolute errors at hospital level) and the national prediction error (absolute deviation of the national estimate from the population values), both expressed as a percentage from the population values