Sampling events in space and time: a case study of burn injuries

We describe the sampling plan used to estimate the number of burn injuries seen at emergency departments of hospitals in New England. We present the rationale for each of the options considered and the implications of each. The chosen plan included all 256 hospitals, but used a different systematic one-day-in-ten sample for each of ten subgroupings of the hospitals. The findings suggest that over 47,000 injuries or almost 400 per 100,000 population were seen at New England Emergency Departments in a 1-year period. This incidence is 60 per cent higher than pre-project estimates would have suggested. From the variability of the data in the sample, we calculate that our estimate is probably no more than 3 per cent under or over the 'correct' rate that would have been obtained by a full non-sampling approach, costing almost 10 times as much. Additional support for our strong belief in the accuracy of the sample-based estimate is provided by the fact that if we had used the same sampling approach to estimate the number of hospitalized burn injuries, our estimate would have been in error by only 124 or 4 per cent from the total 3276 obtained by 100 per cent sampling of hospitalization records.