Implications of conducting trend analyses of emergency department visits using publicly released masked design...

We examine whether using public National Hospital Ambulatory Medical Care Survey (NHAMCS) data with masked design variables produces different estimates, standard errors, and confidence intervals (CIs) for the temporal trend of overall and injury-related emergency department (ED) visits compared with using unmasked data housed at the Centers for Disease Control and Prevention.

We obtained counts of ED visits with their standard errors for unmasked data from published summaries from 1999 through 2006. Public files with masked design variables were used to estimate visit counts and rates with standard errors for each year, using the Taylor series linearization method. Weighted least squares linear regressions were used for trend analysis to estimate the annual change in visits.

Compared with using unmasked data, using masked data produced similar estimates of overall ED visit counts and rates for each year from 1999 through 2006 but overestimated standard errors by 27% on average (range 12% to 45%). According to unmasked data, overall ED visit counts increased by 1.973 million annually (standard error 0.747; 95% CI 0.505 to 3.440 million [corrected] ). With masked data, the estimated change was 1.977 million visits annually (standard error 0.894; 95% CI -0.221 to 3.733 million [corrected] ). As for injury-related ED visit counts, masked data overestimated the standard error by 16% for trend. Although neither unmasked nor masked data suggested a statistically significant annual increase of overall or injury-related ED visit rate, masked data overestimated the standard error by 16%.

Using masked public data overestimated standard errors for trend of counts and rates for overall and injury-related ED visits and resulted in wider CIs.