rubak / sparrpowr

Power Analysis to Detect Spatial Relative Clusters in R

sparrpowR: Power analysis to detect spatial relative clusters

Overview

The sparrpowR package is a suite of R functions to calculate the statistical power to detect clusters using the kernel-based spatial relative risk function that is estimated using the sparr package. Basic visualization is also supported.

Installation

To install the release version from CRAN:

install.packages("sparrpowR")

To install the development version from GitHub:

devtools::install_github("machiela-lab/sparrpowR")

Available functions

Function	Description
spatial_power	Main function. Compute the statistical power of a spatial relative risk function using randomly generated data.
spatial_data	Generate random bivariate data for a spatial relative risk function.
jitter_power	Compute the statistical power of a spatial relative risk function using previously collected data.
spatial_plots	Easily make multiple plots from spatial_power, spatial_data, and jitter_power outputs.
pval_correct	Called within spatial_power and jitter_power, calculates various multiple testing corrections for the alpha level.

Authors

• Ian D. Buller - Occupational and Environmental Epidemiology Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health, Rockville, Maryland - GitHub

• Derek W. Brown - Integrative Tumor Epidemiology Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health, Rockville, Maryland - GitHub

See also the list of contributors who participated in this package, including:

• Tim A. Myers - Laboratory of Genetic Susceptibility, Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health, Rockville, Maryland - GitHub

Usage

set.seed(1234) # for reproducibility

# ------------------ #
# Necessary packages #
# ------------------ #

library(sparrpowR)
library(spatstat.geom)
library(stats)

# ----------------- #
# Run spatial_power #
# ----------------- #

# Circular window with radius 0.5
# Uniform case sampling within a disc of radius of 0.1 at the center of the window
# Complete Spatial Randomness control sampling
# 20% prevalence (n = 300 total locations)
# Statistical power to detect both case and control relative clustering
# 100 simulations (more recommended for power calculation)

unit.circle <- spatstat.geom::disc(radius = 0.5, centre = c(0.5,0.5))

foo <- spatial_power(win = unit.circle,
                     sim_total = 100,
                     x_case = 0.5,
                     y_case = 0.5,
                     samp_case = "uniform",
                     samp_control = "CSR",
                     r_case = 0.1,
                     n_case = 50,
                     n_control = 250)
                     
# ----------------------- #
# Outputs from iterations #
# ----------------------- #

# Mean and standard deviation of simulated sample sizes and bandwidth
mean(foo$n_con); stats::sd(foo$n_con)    # controls
mean(foo$n_cas); stats::sd(foo$n_cas)    # cases
mean(foo$bandw); stats::sd(foo$bandw)    # bandwidth of case density (if fixed, same for control density) 

# Global Test Statistics
## Global maximum relative risk: Null hypothesis is mu = 1
stats::t.test(x = foo$s_obs, mu = 0, alternative = "two.sided")

## Integral of log relative risk: Null hypothesis is mu = 0
stats::t.test(x = foo$t_obs, mu = 1, alternative = "two.sided")

# ----------------- #
# Run spatial_plots #
# ----------------- #

# Statistical power for case-only clustering (one-tailed test)
spatial_plots(foo)

# Statistical power for case clustering and control
clustering (two-tailed test)
## Only showing second and third plot
spatial_plots(foo, cascon = TRUE)

# --------------------------- #
# Multiple Testing Correction #
# --------------------------- #

# Same parameters as above
# Apply a conservative Bonferroni correction

set.seed(1234) # reset RNG

# Run spatial_power()
foo <- spatial_power(win = unit.circle,
                     sim_total = 100,
                     x_case = 0.5,
                     y_case = 0.5,
                     samp_case = "uniform",
                     samp_control = "CSR",
                     r_case = 0.1,
                     n_case = 50,
                     n_control = 250,
                     alpha = 0.05,
                     p_correct = "FDR")
                     
median(foo$alpha) # critical p-value of 3e-6 

# Run spatial_plots() for case-only clustering
## Only showing third plot
spatial_plots(foo, cascon = TRUE)