Type: Package
Title: Simple Dengue Test and Vaccinate Cost Thresholds
Version: 0.1.2
Description: Provides the mathematical model described by "Serostatus Testing & Dengue Vaccine Cost-Benefit Thresholds" in <doi:10.1098/rsif.2019.0234>. Using the functions in the package, that analysis can be repeated using sample life histories, either synthesized from local seroprevalence data using other functions in this package (as in the manuscript) or from some other source. The package provides a vignette which walks through the analysis in the publication, as well as a function to generate a project skeleton for such an analysis.
License: MIT + file LICENSE
Encoding: UTF-8
LazyData: true
Depends: R (≥ 3.5)
Suggests: data.table, testthat, usethis, devtools, roxygen2, jsonlite, ggplot2, cowplot, directlabels, rmarkdown, knitr
RoxygenNote: 6.1.1
VignetteBuilder: knitr
URL: https://gitlab.com/cabp_LSHTM/denvax
BugReports: https://gitlab.com/cabp_LSHTM/denvax/issues
NeedsCompilation: no
Packaged: 2019-11-30 20:31:28 UTC; carlpearson
Author: Carl A. B. Pearson ORCID iD [aut, cre], Kaja M. Abbas ORCID iD [aut], Samuel Clifford ORCID iD [aut], Stefan Flasche ORCID iD [aut], Thomas J. Hladish ORCID iD [aut]
Maintainer: Carl A. B. Pearson <carl.pearson@lshtm.ac.uk>
Repository: CRAN
Date/Publication: 2019-12-01 19:00:02 UTC

denvax: Simple Dengue Test and Vaccinate Cost Thresholds

Description

Provides the mathematical model described by "Serostatus Testing & Dengue Vaccine Cost-Benefit Thresholds" in <doi:10.1098/rsif.2019.0234>. Using the functions in the package, that analysis can be repeated using sample life histories, either synthesized from local seroprevalence data using other functions in this package (as in the manuscript) or from some other source. The package provides a vignette which walks through the analysis in the publication, as well as a function to generate a project skeleton for such an analysis.

denvax functions

serofit

fits serosurvey data against two-risk model

synthetic.pop

using parameter fit data, generate a sample population

nPxA

estimate dengue infection outcome probabilities based on a synthetic population

ROIcoeffs

using population outcome probabilities, compute ROI equation coefficients

ROI

compute ROIs from setting coefficients and cost scenarios

build.project

create a template project for estimating ROIs

Compute the ROI surfaces given test and vaccine cost fractions.

Description

Compute the ROI surfaces given test and vaccine cost fractions.

Usage

ROI(rcoeffs, nus = seq(0.3, 0.9, by = 0.05), taus = 10^seq(-2, -0.5, by
  = 0.05))

Arguments

rcoeffs

a data.frame with the ROI surface coefficients from ROIcoeffs

nus

the series of normalized vaccine costs to use for ROI calcs

taus

the series of normalized test costs to use for ROI calcs

Details

tabulates ROI

Value

a 'data.frame' ('data.table', if available) with columns:

nu

numeric, the normalized vaccine cost used

tau

numeric, the normalized test cost used

mechanism

character, either "ordinal" or "binary" corresponding to the type of test

A

integer; the age when routine test-then-vaccinate strategy starts (from As)

L

integer; the maximum number of tests for routine test-then-vaccinate strategy (from Ls)

cost

numeric; the intervention cost (as a fraction of second infection cost)

benefit

numeric; the difference in health outcome cost (as a fraction of second infection cost) minus 'cost'; positive values indicate positive net benefit

roi

numeric; return on investment: 'benefit' over 'cost'

Examples

require(denvax);
data(morrison2010) # has counts by age
fit <- with(morrison2010, serofit(sero=Seropositive, N=Number, age.min=Age))
m2010pop <- synthetic.pop(fit, runs = 10, popsize = 10) # small sample size for example run time
m2010lh <- nPxA(m2010pop)
L <- 5
rc <- ROIcoeffs(m2010lh, As=5:10, Ls=L)
rois <- ROI(rc, nus = 0.5, taus = 0.01)
srois <- subset(rois, mechanism == "binary")
mrois <- matrix(srois$roi, nrow = L)
contour(x=unique(srois$L), y=unique(srois$A), z=mrois,
  xlab = "Max # of Tests", ylab = "Initial Age", main="ROI Contour"
)

Compute the Return on Investment (ROI) surface coefficients from population probabilities

Description

Compute the Return on Investment (ROI) surface coefficients from population probabilities

Usage

ROIcoeffs(probabilities, As = 5:20, Ls = (diff(range(As)) + 1):1)

Arguments

probabilities

a data.frame (or data.table) with the probabilities resulting from nPxA. Rows must correspond to ages, starting with age 1

As

the starting age(s) to consider

Ls

the maximum number of tests for each age; should either be an integer per age or a single integer for all ages. The default behavior computes the number of tests (for each age) that makes the maximum of 'As' the maximum testing age Note: results will also be provided for shorter testing intervals, as the intermediate coefficients are calculated as part of computing the value at the maximum L

Details

computes the coefficients for the economic calculations

Value

a data.frame (data.table, if available) with columns:

A

integer; the age when routine test-then-vaccinate strategy starts (from As)

L

integer; the maximum number of tests for routine test-then-vaccinate strategy (from Ls)

vacfrac

numeric; the fraction of individuals participating in this strategy that get vaccinated

pri.offset

numeric; the (additive) reduction in vacfrac if using the ordinal test

Sfrac

numeric; the proportion experiencing second infection costs

Fresp

numeric; the F/S cost fraction term, when comparing vaccination with and without testing

Sgain

numeric; the S term, when comparing vaccination with and without testing

Examples

require(denvax);
data(morrison2010) # has counts by age
fit <- with(morrison2010, serofit(sero=Seropositive, N=Number, age.min=Age))
m2010pop <- synthetic.pop(fit, runs = 10, popsize = 10) # small sample size for example run time
m2010lh <- nPxA(m2010pop)
rc <- ROIcoeffs(m2010lh, As=5:10, Ls=5)
pp <- par()
par(mfrow=c(1, 2))
rcs <- subset(rc, A==10 & L < 11)
with(rcs, plot(
  L, aveTests, type="l",
  xlab="Max # of Tests Allowed",
  ylab="Ave # of Tests Administered",
  main="Starting @ Age 10",
  ylim=c(1, 3)
))
rcs <- subset(rc, A==5 & L < 11)
with(rcs, plot(
  L, aveTests, type="l",
  xlab="Max # of Tests Allowed",
  ylab="",
  main="Starting @ Age 5",
  ylim=c(1, 3)
))
par(pp)

Creates an ROI estimation project.

Description

Creates an ROI estimation project.

Usage

build.project(targetdir, overwrite = FALSE, copy_pub = TRUE)

Arguments

targetdir

path to enclosing directory. If this directory does not exist, will attempt to create it (recursively)

overwrite

overwrite existing files corresponding to the project skeleton elements?

copy_pub

copy the '/pub' folder (which contains the analyses from the publication)

Details

This function sets up the skeleton of an analysis to go from seroprevalence data to the ROI estimation surface. That skeleton uses a series of separate scripts for each analytical step (fitting, simulation, analysis, and application), connected via the command line build tool make. This approach allows clean substitution for various stages (e.g., using a different model to generate life histories). The following files are created:

Makefile

the dependencies for various analysis stages

README.md

brief notes about project parts

fit.R

script for fitting seroprevalence data

synthesize.R

script for generating synthetic populations

digest.R

script for converting life histories into probability coefficients for ROI calculation

simple.R

a quick example start-to-finish analysis

Value

logical, indicating error free creation of the project skeleton; there may still be other warnings

Examples

require(denvax)
tardir <- tempdir() # replace with desired target
build.project(tardir)
list.files(tardir, recursive = TRUE)

The serosurvey data in L'Azou 2016

Description

From "Symptomatic Dengue in Children in 10 Asian and Latin American Countries", Table 4.

Usage

lazou2016

Format

a data.frame (data.table, if installed) with 20 rows and 4 columns:

Country

character, common country name (all Peru for this data)

Age

character, the bounding ages for the sample; format: lower age '-' upper age

Number

integer, the number of samples

Seropositive

integer, the number of seropositive samples

Source

https://doi.org/10.1056/NEJMoa1503877

Examples

require(denvax); require(ggplot2)
data(lazou2016)
ggplot(lazou2016) + aes(Age, Seropositive/Number*100, color = Country) +
  geom_point() + labs(y="Seropositive %", x="Age Group") + lims(y=c(0,100)) +
  theme_minimal()

The serosurvey data in Morrison 2010

Description

From "Epidemiology of Dengue Virus in Iquitos, Peru 1999 to 2005: Interepidemic and Epidemic Patterns of Transmission", combining information from Fig. 2 and Fig. 3. The data from Fig. 3 were extracted using https://automeris.io/WebPlotDigitizer/

Usage

morrison2010

Format

a data.frame (data.table, if installed) with 13 rows and 4 columns:

Country

character, common country name (all Peru for this data)

Age

integer, the age category

Number

integer, the number of samples

Seropositive

integer, the number of seropositive samples

Source

https://doi.org/10.1371/journal.pntd.0000670

Examples

require(denvax)
data(morrison2010)
with(morrison2010, plot(Age, Seropositive/Number*100, ylab="% Seropositive", ylim=c(0,100)))

Compute the nPx(A), C(A) proportions from a population of life histories

Description

Compute the nPx(A), C(A) proportions from a population of life histories

Usage

nPxA(lifehistory)

Arguments

lifehistory

a matrix with rows (sample individuals) and columns (outcome in year of life); see synthetic.pop return value

Details

computes the relevant nPx(A) and C(A): the probabilities of the various life trajectories, by age. See <doi:10.1098/rsif.2019.0234>, SI section II.A (Cost Benefit Equations: Definitions)

Value

a data.frame (data.table, if available) with columns

A

integer; the reference year of life, from 1 to dim(lifehistory)[2]

p_0

numeric; probability of 0 lifetime infections

p_1p

numeric; probability of 1 or more lifetime infections

p_2p

numeric; probability of 2 or more lifetime infections

p0_1

numeric; probability of 0 infections at age A, and 1 lifetime infection

p0_1p

numeric; probability of 0 infections at age A, and 1 or more lifetime infections

p0_2p

numeric; probability of 0 infections at age A, and 2 or more lifetime infections

p1_A

numeric; probability of 1 infection at age A, and 1 or more lifetime infections

p1_2p

numeric; probability of 1 infection at age A, and 2 or more lifetime infections

p1p_A

numeric; probability of 1 or more infections at age A, and 1 or more lifetime infections

p2p_A

numeric; probability of 2 or more infections at age A, and 2 or more lifetime infections

CA

numeric; probability of converting from seronegative to seropositive between age A and A+1

Examples

require(denvax);
data(morrison2010) # has counts by age
fit <- with(morrison2010, serofit(sero=Seropositive, N=Number, age.min=Age))
m2010pop <- synthetic.pop(fit, runs = 10, popsize = 10) # small sample size for example run time
m2010lh <- nPxA(m2010pop)
m2010lh
with(m2010lh,
  plot(A, p0_2p*100, type="l",
    xlab="Age", ylab="%", ylim = c(0, 100),
    main="Individuals w/ No Infections,\nbut that will have 2"
  )
)

Model fitting for serological data

Description

Model fitting for serological data

Usage

serofit(seropositive, N, age.min = use.default(1L:length(seropositive),
  "age.min"), age.max = use.default(age.min, "age.max"))

Arguments

seropositive

the number of seropositive samples for each age group; length(seropositive) must be at least 3

N

the total number of samples for each age group; length(N) must equal length(seropositive)

age.min

the low age in age groups; defaults to '1:length(seropositive)', i.e. assumes the seropositive data corresponds to yearly cohorts starting at age 1.

age.max

the upper age in age groups; defaults to 'age.min', i.e. assumes each category corresponds to a single year

Details

Fits a constant force of infection, two-risk category model using seroprevalence survey data. i.e.:

$$ P_+(A) = p_H * (1-(1-f_H)^A) + (1-p_H) * (1-(1-f_L)^A) $$

This probability is fit to the seroprevalence by age category data, using maximum likelihood and optim.

Value

a list of best-fit parameters, all numeric values:

f_H

force of infection, for the high risk group

f_L

force of infection, for the low risk group

p_H

the proportion of the population at high risk

Examples

require(denvax);
data(morrison2010) # has counts by age
fit <- with(morrison2010, serofit(sero=Seropositive, N=Number, age.min=Age))
if (requireNamespace("data.table", quietly = TRUE)) {
data(lazou2016) # has counts by age range, instead of counts for every year
# this example uses `data.table`` functions to simplify processing
# several groups at once
  lazou2016[,{
    agerange <- data.table::tstrsplit(Age, "-")
    serofit(
      sero     = Seropositive,
      N        = Number,
      age.min  = as.integer(agerange[[1]]),
      age.max  = as.integer(agerange[[2]])
    )
  }, by = Country]
}

Compute synthetic population trajectories from model parameters

Description

Compute synthetic population trajectories from model parameters

Usage

synthetic.pop(pars, runs = 100, popsize = 1000, maxAge = 70,
  rngseed = NULL)

Arguments

pars

a list with elements 'f_H', 'f_L', and 'p_H'; see serofit return value

runs

the number of different serotype timelines to simulate

popsize

the number of individuals to sample for each run

maxAge

the length of each lifetime

rngseed

an optional seed for the random number generator

Details

Using fitted parameters for a two-risk, constant force of infection model, simulate a dengue annual exposures model for the requested number of serotype series ('runs') and individuals ('popsize'). The resulting matrix is a collection of integers, 0-4. 0 indicates no infection, 1-4 infection by the corresponding serotype.

Value

a matrix of integers 0-4, rows 'runs*popsize' x columns 'maxAge'

Examples

require(denvax);
data(morrison2010) # has counts by age
fit <- with(morrison2010, serofit(sero=Seropositive, N=Number, age.min=Age))
m2010pop <- synthetic.pop(fit, runs = 10, popsize = 10) # small sample size for example run time
head(m2010pop)