nmode's Git Repositories - Rnaught/blob - R/seqB.R

   1 #' seqB method
   2 #'
   3 #' This function implements a sequential Bayesian estimation method of R0 due to Bettencourt and Riberio (PloS One, 2008).
   4 #' See details for important implementation notes.
   5 #'
   6 #' The method sets a uniform prior distribution on R0 with possible values between zero and \code{kappa}, discretized to a fine grid.
   7 #' The distribution of R0 is then updated sequentially, with one update for each new case count observation.
   8 #' The final estimate of R0 is \code{Rhat}, the mean of the (last) posterior distribution.
   9 #' The prior distribution is the initial belief of the distribution of R0; which in this implementation is the uninformative uniform
  10 #' distribution with values between zero and \code{kappa}. Users can change the value of /code{kappa} only (i.e., the prior distribution
  11 #' cannot be changed from the uniform).  As more case counts are observed, the influence of the prior distribution should lessen on
  12 #' the final estimate \code{Rhat}.
  13 #'
  14 #' This method is based on an approximation of the SIR model, which is most valid at the beginning of an epidemic. The method assumes
  15 #' that the mean of the serial distribution (sometimes called the serial interval) is known. The final estimate can be quite sensitive
  16 #' to this value, so sensitivity testing is strongly recommended. Users should be careful about units of time (e.g., are counts observed
  17 #' daily or weekly?) when implementing.
  18 #'
  19 #' Our code has been modified to provide an estimate even if case counts equal to zero are present in some time intervals. This is done
  20 #' by grouping the counts over such periods of time. Without grouping, and in the presence of zero counts, no estimate can be provided.
  21 #'
  22 #' @param NT Vector of case counts.
  23 #' @param mu Mean of the serial distribution. This needs to match case counts in time units. For example, if case counts
  24 #'           are weekly and the serial distribution has a mean of seven days, then \code{mu} should be set to one. If case
  25 #'           counts are daily and the serial distribution has a mean of seven days, then \code{mu} should be set to seven.
  26 #' @param kappa Largest possible value of uniform prior (defaults to 20). This describes the prior belief on ranges of R0,
  27 #'              and should be set to a higher value if R0 is believed to be larger.
  28 #'
  29 #' @return \code{secB} returns a list containing the following components: \code{Rhat} is the estimate of R0 (the posterior mean),
  30 #'         \code{posterior} is the posterior distribution of R0 from which alternate estimates can be obtained (see examples),
  31 #'         and \code{group} is an indicator variable (if \code{group=TRUE}, zero values of NT were input and grouping was done
  32 #'         to obtain \code{Rhat}). The variable \code{posterior} is returned as a list made up of \code{supp} (the support of
  33 #'         the distribution) and \code{pmf} (the probability mass function).
  34 #'
  35 #' @examples
  36 #' ## ===================================================== ##
  37 #' ## Illustrate on weekly data                             ##
  38 #' ## ===================================================== ##
  39 #'
  40 #' NT <- c(1, 4, 10, 5, 3, 4, 19, 3, 3, 14, 4)
  41 #' ## obtain Rhat when serial distribution has mean of five days
  42 #' res1 <- seqB(NT=NT, mu=5/7)
  43 #' res1$Rhat
  44 #' ## obtain Rhat when serial distribution has mean of three days
  45 #' res2 <- seqB(NT=NT, mu=3/7)
  46 #' res2$Rhat
  47 #'
  48 #' ## ============================================================= ##
  49 #' ## Compute posterior mode instead of posterior mean and plot     ##
  50 #' ## ============================================================= ##
  51 #'
  52 #' Rpost <-     res1$posterior
  53 #' loc <- which(Rpost$pmf == max(Rpost$pmf))
  54 #' Rpost$supp[loc] # posterior mode
  55 #' res1$Rhat # compare with posterior mean
  56 #'
  57 #' par(mfrow=c(2, 1), mar=c(2, 2, 1, 1))
  58 #' plot(Rpost$supp, Rpost$pmf, col="black", type="l", xlab="", ylab="")
  59 #' abline(h=1/(20/0.01+1), col="red")
  60 #' abline(v=res1$Rhat, col="blue")
  61 #' abline(v=Rpost$supp[loc], col="purple")
  62 #' legend("topright", legend=c("prior", "posterior", "posterior mean (Rhat)", "posterior mode"),
  63 #'        col=c("red", "black", "blue", "purple"), lty=1)
  64 #' plot(Rpost$supp, Rpost$pmf, col="black", type="l", xlim=c(0.5, 1.5), xlab="", ylab="")
  65 #' abline(h=1/(20/0.01+1), col="red")
  66 #' abline(v=res1$Rhat, col="blue")
  67 #' abline(v=Rpost$supp[loc], col="purple")
  68 #' legend("topright", legend=c("prior", "posterior", "posterior mean (Rhat)", "posterior mode"),
  69 #'        col=c("red", "black", "blue", "purple"), lty=1)
  70 #'
  71 #' ## ========================================================= ##
  72 #' ## Compute Rhat using only the first five weeks of data      ##
  73 #' ## ========================================================= ##
  74 #'
  75 #' res3 <- seqB(NT=NT[1:5], mu=5/7)     # serial distribution has mean of five days
  76 #' res3$Rhat
  77 #'
  78 #' @export
  79 seqB <- function(NT, mu, kappa=20) {
  80     if (length(NT) < 2)
  81         print("Warning: length of NT should be at least two.")
  82     else {
  83         if (min(NT) > 0) {
  84             times <- 1:length(NT)
  85             tau <- diff(times)
  86         }
  87             group <- FALSE
  88         if (min(NT) == 0) {
  89             times <- which(NT > 0)
  90             NT <- NT[times]
  91             tau <- diff(times)
  92             group <- TRUE
  93         }
  94
  95         R <- seq(0, kappa, 0.01)
  96         prior0 <- rep(1, kappa / 0.01 + 1)
  97         prior0 <- prior0 / sum(prior0)
  98         k <- length(NT) - 1
  99         R0.post <- matrix(0, nrow=k, ncol=length(R))
 100         prior <- prior0
 101         posterior <- seq(0, length(prior0))
 102         gamma <- 1 / mu
 103
 104         for (i in 1:k) {
 105             mm1 <- NT[i]
 106             mm2 <- NT[i+1]
 107             lambda <- tau[i] * gamma * (R - 1)
 108             lambda <- log(mm1) + lambda
 109             loglik <- mm2 * lambda - exp(lambda)
 110             maxll <- max(loglik)
 111             const <- 0
 112
 113             if (maxll > 700)
 114                 const <- maxll - 700
 115
 116             loglik <- loglik-const
 117             posterior <- exp(loglik) * prior
 118             posterior <- posterior / sum(posterior)
 119             prior <- posterior
 120         }
 121
 122         Rhat <- sum(R * posterior)
 123
 124         return(list(Rhat=Rhat, posterior=list(supp=R, pmf=posterior), group=group))
 125     }
 126 }