Calls the function to obtain the cluster with the maximum log-likelihood ratio or minimum DIC of all the clusters with the same center and start and end dates.

Description

This function orders the regions according to the distance to a given center and selects the regions with distance to the center less than sqrt(rr). Then it calls glmAndZIP.iscluster() to obtain the cluster with the maximum log-likelihood ratio or minimum DIC of all the clusters with the same center and start and end dates, and where the maximum fraction of the total population inside the cluster is less than fractpop.

Usage

CalcStatClusterGivenCenter(
  point,
  stfdf,
  rr,
  minDateCluster,
  maxDateCluster,
  fractpop,
  model0,
  ClusterSizeContribution
)

Arguments

Value

vector containing the coordinates of the center, the size, the start and end dates, the log-likelihood ratio or DIC, the p-value and the risk of the cluster with the maximum log-likelihood ratio or minimum DIC.

Obtains the clusters with the maximum log-likelihood ratio or minimum DIC for each center and start and end dates.

Description

This function explores all possible clusters changing their center and start and end dates. For each center and time periods, it obtains the cluster with the maximum log-likelihood ratio or minimum DIC so that the maximum fraction of the total population inside the cluster is less than fractpop, and the maximum distance to the center is less than radius.

Usage

CalcStatsAllClusters(
  thegrid,
  CalcStatClusterGivenCenter,
  stfdf,
  rr,
  typeCluster,
  sortDates,
  idMinDateCluster,
  idMaxDateCluster,
  fractpop,
  model0,
  ClusterSizeContribution,
  numCPUS
)

Arguments

Value

data frame with information of the clusters with the maximum log-likelihood ratio or minimum DIC for each center and start and end dates. It contains the coordinates of the center, the size, the start and end dates, the log-likelihood ratio or DIC, the p-value and the risk of each of the clusters.

Creates grid over the study area.

Description

If the argument thegrid of DetectClustersModel() is null, this function is used to create a rectangular grid with a given step. If step is NULL the step used is equal to 0.2*radius. The grid contains the coordinates of the centers of the clusters explored.

Usage

CreateGridDClusterm(stfdf, radius, step)

Arguments

Value

two columns matrix where each row represents a point of the grid.

Detects clusters and computes their significance.

Description

Searches all possible clusters with start and end dates within minDateUser and maxDateUser, so that the maximum fraction of the total population inside the cluster is less than fractpop, and the maximum distance to the center is less than radius. The search can be done for spatial or spatio-temporal clusters. The significance of the clusters is obtained with a Monte Carlo procedure or based on the chi-square distribution (glm, glmer or zeroinfl models) or DIC (inla models).

Usage

DetectClustersModel(
  stfdf,
  thegrid = NULL,
  radius = Inf,
  step = NULL,
  fractpop,
  alpha,
  typeCluster = "S",
  minDateUser = NULL,
  maxDateUser = NULL,
  R = NULL,
  model0,
  ClusterSizeContribution = "Population"
)

Arguments

Value

data frame with information of the detected clusters ordered by its log-likelihood ratio value or DIC. Each row represents the information of one of the clusters. It contains the coordinates of the center, the size, the start and end dates, the log-likelihood ratio or DIC, the p-value, the risk of the cluster, and a boolean indicating if it is a cluster (TRUE in all cases). It also returns alpha_bonferroni which is the level of significance adjusted for multiple testing using Bonferroni correction. Thus, rows that should be considered clusters are the ones with p-value less than alpha_bonferroni.

References

Bilancia M, Demarinis G (2014) Bayesian scanning of spatial disease rates with the Integrated Nested Laplace Approximation (INLA). Statistical Methods & Applications 23(1): 71 - 94. doi:10.1007/s10260-013-0241-8

Jung I (2009) A generalized linear models approach to spatial scan statistics for covariate adjustment. Statistics in Medicine 28(7): 1131 - 1143. Gómez-Rubio V, Molitor J, Moraga P (2018) Fast Bayesian Classification for Disease Mapping and the Detection of Disease Clusters. In: Cameletti M., Finazzi F. (eds) Quantitative Methods in Environmental and Climate Research. Springer, Cham

Gómez-Rubio V, Moraga P, Molitor J, Rowlingson B (2019). "DClusterm: Model-Based Detection of Disease Clusters." _Journal of Statistical Software_, *90*(14), 1-26. doi: 10.18637/jss.v090.i14 (URL: https://doi.org/10.18637/jss.v090.i14).

Examples

library("DClusterm")
data("NY8")

NY8$Observed <- round(NY8$Cases)
NY8$Expected  <- NY8$POP8 * sum(NY8$Observed) / sum(NY8$POP8)

NY8$x <- coordinates(NY8)[, 1]
NY8$y <- coordinates(NY8)[, 2]


#Model to account for covariates
ny.m1 <- glm(Observed ~ offset(log(Expected)) + PCTOWNHOME + PCTAGE65P +
PEXPOSURE, family = "poisson", data = NY8)

#Indices of areas that are possible cluster centres
idxcl <- c(120, 12, 89, 139, 146)

#Cluster detection adjusting for covariates
ny.cl1 <- DetectClustersModel(NY8,
thegrid = as.data.frame(NY8)[idxcl, c("x", "y")],
fractpop = 0.15, alpha = 0.05,
typeCluster = "S", R = NULL, model0 = ny.m1,
ClusterSizeContribution = "POP8")

#Display results
ny.cl1

Leukemia in an eight-county region of upstate New York, 1978-1982.

Description

This data set provides the number of incident leukemia cases per census tract in an eight-county region of upstate New York in the period 1978-1982. In addition, the data set also includes information about the location of the census tracts, the population in 1980, the inverse of the distance to the nearest Trichloroethene (TCE) site, the percentage of people aged 65 or more, and the percentage of people who own their home.

The dataset also provides the locations of the TCE sites.

File NY8_clusters contains the results of running DetectClustersModel on a null model ('ny.m0') and another one with covariates ('ny.m1'). The results are in 'ny.cl0' and 'ny.cl1', respectively.

Usage

data(NY8)

Format

A SpatialPolygonsDataFrame with 281 polygons representing the census tracts, and the following information about each census tract:

Source

Waller and Gotway (2004) and Bivand et al. (2008)

References

Bivand, R.S., E. J. Pebesma and V. Gómez-Rubio (2008). Applied Spatial Data Analysis with R. Springer.

Waller, L., B. Turnbull, L. Clark, and P. Nasca (1992). Chronic disease surveillance and testing of clustering of disease and exposure: application to leukemia incidence in tce-contamined dumpsites in upstate New York. Environmetrics 3, 281-300

Waller, L. A. and C. A. Gotway (2004). Applied Spatial Statistics for Public Health Data. John Wiley & Sons, Hoboken, New Jersey.

Brain cancer in males in Navarre, Spain, 1988-1994.

Description

This data set contains the male mortality due to brain cancer in the 40 basic health zones (BHZ) in Navarre over the period 1988-1994, and the neighborhood structure of the BHZ in Navarre. In addition, the data set also includes information about the location of the BHZ, the expected cases, the Standardized Mortality Ratio (SMR), relative risk estimates and 95% confidence intervals.

Usage

data(Navarre)

Format

brainnav: A SpatialPolygonsDataFrame with 40 polygons representing the basic health zones (BHZ) in Navarre, and the following information about each BHZ:

brainnavnb: A neighbor (nb) object which contains the index numbers of the neighbors of each BHZ.

Source

Data set obtained from Ugarte et al. (2004). Boundaries downloaded in shapefile format from https://geoportal.navarra.es/es/idena. These have been thinned to reduce space use.

References

Ugarte, M. D., B. Ibáñez, and A. F. Militino (2004). Testing for poisson zero a inflation in disease mapping. Biometrical Journal 46 (5), 526-539.

Ugarte, M. D., B. Ibáñez, and A. F. Militino (2006). Modelling risks in a disease mapping. Statistical Methods in Medical Research 15, 21-35.

Removes the overlapping clusters.

Description

Function DetectClustersModel() detects duplicated clusters. This function reduces the number of clusters by removing the overlapping clusters.

Usage

SelectStatsAllClustersNoOverlap(stfdf, statsAllClusters)

Arguments

Value

data frame with the same information than statsAllClusters but only for clusters that do not overlap.

Examples

library("DClusterm")
data("brainNM")
data("brainNM_clusters")

SelectStatsAllClustersNoOverlap(brainst, nm.cl1)

Constructs a variable that indicates the locations and times that pertain to a cluster.

Description

This function constructs a variable that indicates the locations and times that pertain to a cluster. Each position of the variable is equal to 1 if it corresponds to a location and time inside the cluster, and 0 otherwise. This is one of the explanatory variables used in the glmAndZIP.iscluster function to model the observed cases.

Usage

SetVbleCluster(stfdf, idTime, idSpace)

Arguments

Value

vector with 1's or 0's that indicates the locations and times that pertain to a cluster.

Brain cancer in New Mexico, USA, 1973-1991.

Description

This data set contains the number of incident brain cancer cases in the 32 counties of New Mexico, USA, and each year of the period 1973-1991, and the location of Los Alamos National Laboratory. In addition, the data set also includes for each county and year information about the expected cases, the Standardized Morbidity Ratio (SMR), the FIPS, ...

File brainNM_clusters contains the results of running DetectClustersModel on a null model ('nm.m0') and another one with covariates ('nm.m1'). The results are in 'nm.cl0' and 'nm.cl1', respectively.

Usage

data(brainNM)

Format

brainst: A STFDF object containing the following information for each county and year:

losalamos: A SpatialPoints object which contains the location (in long/lat) of Los Alamos National Laboratory obtained from the Wikipedia: -106.298333, 35.881667.

Source

Data have been downlodad from the SatScan website. Boundaries have been obtained from the U.S. Census Bureau. Cibola and Valencia counties has been merged together.

References

SatScan (c). https://www.satscan.org

Kulldorff, M., W. F. Athas, E. J. Feurer, B. A. Miller, and C. R. Key (1998). Evaluating cluster alarms: a space-time scan statistic and brain cancer in los alamos, new mexico. American Journal of Public Health 88, 1377-1380.

Computes the probability that a model parameter is <=k from inla marginals

Description

This function will be used to calculate the P(coeficient variable cluster <=0)

Usage

computeprob(func, k)

Arguments

Value

probability model coefficient <=k

Extract indices of the areas in the clusters detected

Description

This function returns a categorical vector that identifies to which cluster a given areas belongs. It is the empty string for areas not in a cluster.

Usage

get.allknclusters(spdf, knresults)

Arguments

Value

A categorical vector with value the cluster to which area belongs. It is the empty string for regions not in a cluster.

Gets areas in a spatio-temporal cluster

Description

This function is similar to get.knclusters but it also allows for spatio-temporal clusters.

Usage

get.stclusters(stfdf, results)

Arguments

Value

A list with as many elements as clusters in 'results'

Examples

library("DClusterm")
library("RColorBrewer")

data("brainNM")
data("brainNM_clusters")

stcl <- get.stclusters(brainst, nm.cl0)
#Get first cluster
brainst$CLUSTER <- ""
brainst$CLUSTER[ stcl[[1]] ] <- "CLUSTER"

#Plot cluster
stplot(brainst[, , "CLUSTER"], at = c(0, 0.5, 1.5), col = "#4D4D4D",
  col.regions = c("white", "gray"))

Obtains the cluster with the maximum log-likelihood ratio or minimum DIC of all the clusters with the same center and start and end dates.

Description

This function constructs all the clusters with start date equal to minDateCluster, end date equal to maxDateCluster, and with center specified by the first element of idxorder, so that the maximum fraction of the total population inside the cluster is less than fractpop, and the maximum distance to the center is less than radius. For each one of these clusters, the log-likelihood ratio test statistic for comparing the alternative model with the cluster versus the null model of no clusters (if model is glm, glmer or zeroinfl), or the DIC (if model is inla) is calculated. The cluster with maximum value of the log-likelihood ratio or minimum DIC is returned.

Usage

glmAndZIP.iscluster(
  stfdf,
  idxorder,
  minDateCluster,
  maxDateCluster,
  fractpop,
  model0,
  ClusterSizeContribution
)

Arguments

Value

vector containing the size, the start and end dates, the log-likelihood ratio or DIC, the p-value and the risk of the cluster with the maximum log-likelihood ratio or minimum DIC.

Constructs data frame with clusters in binary format.

Description

This function constructs a data frame with number of columns equal to the number of clusters. Each column is a binary representation of one of the clusters. The position i of the column is equal to 1 if the polygon i is in the cluster or 0 if it is not in the cluster.

Usage

knbinary(datamap, knresults)

Arguments

Value

data frame where the columns represent the clusters in binary format. The position i of the column is equal to 1 if the polygon i is in the cluster or 0 if it is not in the cluster.

Examples

library("DClusterm")
library("RColorBrewer")

data("NY8")
data("NY8_clusters")

stcl <- knbinary(NY8, ny.cl1)
#Get first cluster
NY8$CLUSTER <- stcl[, 1]

#Plot cluster
spplot(NY8, "CLUSTER", at = c(0, 0.5, 1.5), col = "#4D4D4D",
  col.regions = c("white", "gray"))

Merges clusters so that they are identifed as levels of a factor.

Description

Given a data frame with clusters that do not overlap this function merges the clusters and construct a factor. The levels of the factor are "NCL" if the polygon of the map is not in any cluster, and "CL" if the polygon i is in cluster i.

Usage

mergeknclusters(datamap, knresults, indClustersPlot)

Arguments

Value

factor with levels that represent the clusters.

Examples

library("DClusterm")
library("RColorBrewer")

data("NY8")
data("NY8_clusters")

stcl <- mergeknclusters(NY8, ny.cl1, 1:2)
#Get first cluster
NY8$CLUSTER <- stcl

#Plot cluster
spplot(NY8, "CLUSTER", col.regions = c("white", "lightgray", "gray"))

Remove overlapping clusters

Description

This function slims the number of clusters down. The spatial scan statistic is known to detect duplicated clusters. This function aims to reduce the number of clusters by removing duplicated and overlapping clusters.

Usage

slimknclusters(d, knresults, minsize = 1)

Arguments

Value

A subset of knresults with non-overlaping clusters of at least minsize size.

Examples

data("brainNM_clusters")

nm.cl1.s <- slimknclusters(brainst, nm.cl1)
nm.cl1.s