kulldorff: Kulldorff Cluster Detection Method

In SpatialEpi: Methods and Data for Spatial Epidemiology

Kulldorff Cluster Detection Method

Description

Kulldorff spatial cluster detection method for a study region with n areas. The method constructs zones by consecutively aggregating nearest-neighboring areas until a proportion of the total study population is included. Given the observed number of cases, the likelihood of each zone is computed using either binomial or poisson likelihoods. The procedure reports the zone that is the most likely cluster and generates significance measures via Monte Carlo sampling. Further, secondary clusters, whose Monte Carlo p-values are below the α-threshold, are reported as well.

Usage

kulldorff(
  geo,
  cases,
  population,
  expected.cases = NULL,
  pop.upper.bound,
  n.simulations,
  alpha.level,
  plot = TRUE
)

Arguments

geo	an n x 2 table of the (x,y)-coordinates of the area centroids
cases	aggregated case counts for all n areas
population	aggregated population counts for all n areas
expected.cases	expected numbers of disease for all n areas
pop.upper.bound	the upper bound on the proportion of the total population each zone can include
n.simulations	number of Monte Carlo samples used for significance measures
alpha.level	alpha-level threshold used to declare significance
plot	flag for whether to plot histogram of Monte Carlo samples of the log-likelihood of the most likely cluster

Details

If expected.cases is specified to be NULL, then the binomial likelihood is used. Otherwise, a Poisson model is assumed. Typical values of n.simulations are 99, 999, 9999

Value

List containing:

most.likely.cluster	information on the most likely cluster
secondary.clusters	information on secondary clusters, if none NULL is returned
type	type of likelihood
log.lkhd	log-likelihood of each zone considered
simulated.log.lkhd	n.simulations Monte Carlo samples of the log-likelihood of the most likely cluster

Note

The most.likely.cluster and secondary.clusters list elements are themselves lists reporting:

location.IDs.included	ID's of areas in cluster, in order of distance
population	population of cluster
number.of.cases	number of cases in cluster
expected.cases	expected number of cases in cluster
SMR	estimated SMR of cluster
log.likelihood.ratio	log-likelihood of cluster
monte.carlo.rank	rank of lkhd of cluster within Monte Carlo simulated values
p.value	Monte Carlo p-value

Author(s)

Albert Y. Kim

References

SatScan: Software for the spatial, temporal, and space-time scan statistics https://www.satscan.org/ Kulldorff, M. (1997) A spatial scan statistic. Communications in Statistics: Theory and Methods, 26, 1481–1496. Kulldorff M. and Nagarwalla N. (1995) Spatial disease clusters: Detection and Inference. Statistics in Medicine, 14, 799–810.

Examples

## Load Pennsylvania Lung Cancer Data
data(pennLC)
data <- pennLC$data

## Process geographical information and convert to grid
geo <- pennLC$geo[,2:3]
geo <- latlong2grid(geo)

## Get aggregated counts of population and cases for each county
population <- tapply(data$population,data$county,sum)
cases <- tapply(data$cases,data$county,sum)

## Based on the 16 strata levels, computed expected numbers of disease
n.strata <- 16
expected.cases <- expected(data$population, data$cases, n.strata)

## Set Parameters
pop.upper.bound <- 0.5
n.simulations <- 999
alpha.level <- 0.05
plot <- TRUE

## Kulldorff using Binomial likelihoods
binomial <- kulldorff(geo, cases, population, NULL, pop.upper.bound, n.simulations, 
                     alpha.level, plot)
cluster <- binomial$most.likely.cluster$location.IDs.included

## plot
plot(pennLC$spatial.polygon,axes=TRUE)
plot(pennLC$spatial.polygon[cluster],add=TRUE,col="red")
title("Most Likely Cluster")

## Kulldorff using Poisson likelihoods
poisson <- kulldorff(geo, cases, population, expected.cases, pop.upper.bound, 
                    n.simulations, alpha.level, plot)
cluster <- poisson$most.likely.cluster$location.IDs.included

## plot
plot(pennLC$spatial.polygon,axes=TRUE)
plot(pennLC$spatial.polygon[cluster],add=TRUE,col="red")
title("Most Likely Cluster Controlling for Strata")

SpatialEpi documentation built on March 7, 2023, 8 p.m.