Performance monitoring in hip fracture surgery--how big a database do we really need?

Systems for collecting information about patient care are increasingly common in orthopaedic practice. Databases can allow various comparisons to be made over time. Significant decisions regarding service delivery and clinical practice may be made based on their results. We set out to determine the number of cases needed for comparison of 30-day mortality, inpatient wound infection rates and mean hospital length of stay, with a power of 80% for the demonstration of an effect at a significance level of p<0.05. We analysed 2 years of prospectively collected data on 1050 hip fracture patients admitted to a city teaching hospital. Detection of a 10% difference in 30-day mortality would require 14,065 patients in each arm of any comparison, demonstration of a 50% difference would require 643 patients in each arm; for wound infections, demonstration of a 10% difference in incidence would require 23,921 patients in each arm and 1127 patients for demonstration of a 50% difference; for length of stay, a difference of 10% would require 1479 patients and 6660 patients for a 50% difference. This study demonstrates the importance of considering the population sizes before comparisons are made on the basis of basic hip fracture outcome data. Our data also help illustrate the impact of sample size considerations when interpreting the results of performance monitoring. Many researchers will be used to the fact that rare outcomes such as inpatient mortality or wound infection require large sample sizes before differences can be reliably demonstrated between populations. This study gives actual figures that researchers could use when planning studies. Statistically meaningful analyses will only be possible with major multi-centre collaborations, as will be possible if hospital Trusts participate in the National Hip Fracture Database.