Rj. Glynn et al., ESTIMATING THE VARIANCE OF STANDARDIZED RATES OF RECURRENT EVENTS, WITH APPLICATION TO HOSPITALIZATIONS AMONG THE ELDERLY IN NEW-ENGLAND, American journal of epidemiology, 137(7), 1993, pp. 776-786
Usual approaches for estimating the variance of a standardized rate ma
y not be applicable to rates of recurrent events. Where individuals ar
e prone to repeated health events, Greenwood and Yule (J R Stat Soc [A
], 1920;83:255-79) advocated use of the negative binomial distribution
to account for departures from the assumption of randomness of recurr
ent events required by the Poisson distribution. In this paper, the au
thors implemented the negative binomial distribution in the computatio
n of annual hospitalization rates within certain hospital market areas
. Data used were from 1,549,915 New England residents aged 65 years or
more who were enrolled in Medicare between October 1, 1988, and Septe
mber 30, 1989, and who had 458,593 hospital admissions during that yea
r. New England was partitioned into 170 hospital market areas ranging
in population size from 162 to 70,821 elderly Medicare enrollees. The
negative binomial distribution demonstrated substantially better fits
than the Poisson distribution to the numbers of hospitalizations withi
n hospital market areas. Estimated standard errors for indirectly stan
dardized rates based on the negative binomial distribution were 25-51
percent higher than estimated standard errors that assumed an underlyi
ng Poisson distribution. Using regression analysis to smooth overdispe
rsion parameters across hospital market areas produced similar results
. The approach described in this paper may be useful in estimation of
confidence intervals for standardized rates of recurrent events when t
hese events do not recur randomly.