Delete model/darth_funcs.R
Browse files- model/darth_funcs.R +0 -402
model/darth_funcs.R
DELETED
|
@@ -1,402 +0,0 @@
|
|
| 1 |
-
simulate_strategies = function(l_params, wtp = 100000){
|
| 2 |
-
# l_params must include:
|
| 3 |
-
# -- *** Model parameters ***
|
| 4 |
-
# -- disease progression parameters (annual): r_HD, p_S1S2, hr_S1D, hr_S2D,
|
| 5 |
-
# -- initial cohort distribution: v_s_init
|
| 6 |
-
# -- vector of annual state utilities: v_state_utility = c(u_H, u_S1, u_S2, u_D)
|
| 7 |
-
# -- vector of annual state costs: v_state_cost = c(c_H, c_S1, c_S2, c_D)
|
| 8 |
-
# -- time horizon (in annual cycles): n_cyles
|
| 9 |
-
# -- annual discount rate: r_disc
|
| 10 |
-
# -- *** Strategy specific parameters ***
|
| 11 |
-
# -- treartment costs (applied to Sick and Sicker states): c_trtA, c_trtB
|
| 12 |
-
# -- utility with Treatment_A (for Sick state only): u_trtA
|
| 13 |
-
# -- hazard ratio of progression with Treatment_B: hr_S1S1_trtB
|
| 14 |
-
|
| 15 |
-
with(as.list(l_params), {
|
| 16 |
-
|
| 17 |
-
####### SET INTERNAL PARAMETERS #########################################
|
| 18 |
-
# Strategy names
|
| 19 |
-
v_names_strat = c("No_Treatment", "Treatment_A", "Treatment_B")
|
| 20 |
-
# Number of strategies
|
| 21 |
-
n_strat = length(v_names_strat)
|
| 22 |
-
|
| 23 |
-
## Treatment_A
|
| 24 |
-
# utility impacts
|
| 25 |
-
u_S1_trtA = u_trtA
|
| 26 |
-
# include treatment costs
|
| 27 |
-
c_S1_trtA = c_S1 + c_trtA
|
| 28 |
-
c_S2_trtA = c_S2 + c_trtA
|
| 29 |
-
|
| 30 |
-
## Treatment_B
|
| 31 |
-
# progression impacts
|
| 32 |
-
r_S1S2_trtB = -log(1 - p_S1S2) * hr_S1S2_trtB
|
| 33 |
-
p_S1S2_trtB = 1 - exp(-r_S1S2_trtB)
|
| 34 |
-
# include treatment costs
|
| 35 |
-
c_S1_trtB = c_S1 + c_trtB
|
| 36 |
-
c_S2_trtB = c_S2 + c_trtB
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
####### INITIALIZATION #########################################
|
| 40 |
-
# Create cost-effectiveness results data frame
|
| 41 |
-
df_ce = data.frame(Strategy = v_names_strat,
|
| 42 |
-
Cost = numeric(n_strat),
|
| 43 |
-
QALY = numeric(n_strat),
|
| 44 |
-
LY = numeric(n_strat),
|
| 45 |
-
stringsAsFactors = FALSE)
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
######### PROCESS ##############################################
|
| 49 |
-
for (i in 1:n_strat){
|
| 50 |
-
l_params_markov = list(n_cycles = n_cycles, r_disc = r_disc, v_s_init = v_s_init,
|
| 51 |
-
c_H = c_H, c_S1 = c_S2, c_S2 = c_S1, c_D = c_D,
|
| 52 |
-
u_H = u_H, u_S1 = u_S2, u_S2 = u_S1, u_D = u_D,
|
| 53 |
-
r_HD = r_HD, hr_S1D = hr_S1D, hr_S2D = hr_S2D,
|
| 54 |
-
p_HS1 = p_HS1, p_S1H = p_S1H, p_S1S2 = p_S1S2)
|
| 55 |
-
|
| 56 |
-
if (v_names_strat[i] == "Treatment_A"){
|
| 57 |
-
l_params_markov$u_S1 = u_S1_trtA
|
| 58 |
-
l_params_markov$c_S1 = c_S1_trtA
|
| 59 |
-
l_params_markov$c_S2 = c_S2_trtA
|
| 60 |
-
|
| 61 |
-
} else if(v_names_strat[i] == "Treatment_B"){
|
| 62 |
-
l_params_markov$p_S1S2 = p_S1S2_trtB
|
| 63 |
-
l_params_markov$c_S1 = c_S1_trtB
|
| 64 |
-
l_params_markov$c_S2 = c_S2_trtB
|
| 65 |
-
}
|
| 66 |
-
l_result = run_sick_sicker_model(l_params_markov)
|
| 67 |
-
|
| 68 |
-
df_ce[i, c("Cost", "QALY", "LY")] = c(l_result$n_tot_cost,
|
| 69 |
-
l_result$n_tot_qaly,
|
| 70 |
-
l_result$n_tot_ly)
|
| 71 |
-
df_ce[i, "NMB"] = l_result$n_tot_qaly * wtp - l_result$n_tot_cost
|
| 72 |
-
}
|
| 73 |
-
|
| 74 |
-
return(df_ce)
|
| 75 |
-
})
|
| 76 |
-
}
|
| 77 |
-
|
| 78 |
-
|
| 79 |
-
|
| 80 |
-
|
| 81 |
-
run_sick_sicker_model = function(l_params,
|
| 82 |
-
verbose = FALSE) {
|
| 83 |
-
with(as.list(l_params), {
|
| 84 |
-
# l_params must include:
|
| 85 |
-
# -- disease progression parameters (annual): r_HD, p_S1S2, hr_S1D, hr_S2D,
|
| 86 |
-
# -- initial cohort distribution: v_s_init
|
| 87 |
-
# -- vector of annual state utilities: v_state_utility = c(u_H, u_S1, u_S2, u_D)
|
| 88 |
-
# -- vector of annual state costs: v_state_cost = c(c_H, c_S1, c_S2, c_D)
|
| 89 |
-
# -- time horizon (in annual cycles): n_cyles
|
| 90 |
-
# -- annual discount rate: r_disc
|
| 91 |
-
|
| 92 |
-
####### SET INTERNAL PARAMETERS #########################################
|
| 93 |
-
|
| 94 |
-
# state names
|
| 95 |
-
v_names_states = c("H", "S1", "S2", "D")
|
| 96 |
-
n_states = length(v_names_states)
|
| 97 |
-
|
| 98 |
-
# vector of discount weights
|
| 99 |
-
v_dw = 1 / ((1 + r_disc) ^ (0:n_cycles))
|
| 100 |
-
|
| 101 |
-
# state rewards
|
| 102 |
-
v_state_cost = c("H" = c_H, "S1" = c_S1, "S2" = c_S2, "D" = c_D)
|
| 103 |
-
v_state_utility = c("H" = u_H, "S1" = u_S1, "S2" = u_S2, "D" = u_D)
|
| 104 |
-
|
| 105 |
-
# transition probability values
|
| 106 |
-
r_S1D = hr_S1D * r_HD # rate of death in sick state
|
| 107 |
-
r_S2D = hr_S2D * r_HD # rate of death in sicker state
|
| 108 |
-
p_S1D = 1 - exp(-r_S1D) # probability of dying when sick
|
| 109 |
-
p_S2D = 1 - exp(-r_S2D) # probability of dying when sicker
|
| 110 |
-
p_HD = 1 - exp(-r_HD) # probability of dying when healthy
|
| 111 |
-
|
| 112 |
-
## Initialize transition probability matrix
|
| 113 |
-
# all transitions to a non-death state are assumed to be conditional on survival
|
| 114 |
-
m_P = matrix(0,
|
| 115 |
-
nrow = n_states, ncol = n_states,
|
| 116 |
-
dimnames = list(v_names_states, v_names_states)) # define row and column names
|
| 117 |
-
## Fill in matrix
|
| 118 |
-
# From H
|
| 119 |
-
m_P["H", "H"] = (1 - p_HD) * (1 - p_HS1)
|
| 120 |
-
m_P["H", "S1"] = (1 - p_HD) * p_HS1
|
| 121 |
-
m_P["H", "D"] = p_HD
|
| 122 |
-
# From S1
|
| 123 |
-
m_P["S1", "H"] = (1 - p_S1D) * p_S1H
|
| 124 |
-
m_P["S1", "S1"] = (1 - p_S1D) * (1 - (p_S1H + p_S1S2))
|
| 125 |
-
m_P["S1", "S2"] = (1 - p_S1D) * p_S1S2
|
| 126 |
-
m_P["S1", "D"] = p_S1D
|
| 127 |
-
# From S2
|
| 128 |
-
m_P["S2", "S2"] = 1 - p_S2D
|
| 129 |
-
m_P["S2", "D"] = p_S2D
|
| 130 |
-
# From D
|
| 131 |
-
m_P["D", "D"] = 1
|
| 132 |
-
|
| 133 |
-
# check that all transition matrix entries are between 0 and 1
|
| 134 |
-
if(!all(m_P <= 1 & m_P >= 0)){
|
| 135 |
-
stop("This is not a valid transition matrix (entries are not between 0 and 1")
|
| 136 |
-
} else
|
| 137 |
-
# check transition matrix rows add up to 1
|
| 138 |
-
if (!all.equal(as.numeric(rowSums(m_P)),rep(1,n_states))){
|
| 139 |
-
stop("This is not a valid transition matrix (rows do not sum to 1)")
|
| 140 |
-
}
|
| 141 |
-
|
| 142 |
-
####### INITIALIZATION #########################################
|
| 143 |
-
# create the cohort trace
|
| 144 |
-
m_Trace = matrix(NA, nrow = n_cycles + 1 ,
|
| 145 |
-
ncol = n_states,
|
| 146 |
-
dimnames = list(0:n_cycles, v_names_states)) # create Markov trace
|
| 147 |
-
|
| 148 |
-
# create vectors of costs and QALYs
|
| 149 |
-
v_C = v_Q = numeric(length = n_cycles + 1)
|
| 150 |
-
|
| 151 |
-
############# PROCESS ###########################################
|
| 152 |
-
|
| 153 |
-
m_Trace[1, ] = v_s_init # initialize Markov trace
|
| 154 |
-
v_C[1] = 0 # no upfront costs
|
| 155 |
-
v_Q[1] = 0 # no upfront QALYs
|
| 156 |
-
|
| 157 |
-
for (t in 1:n_cycles){ # throughout the number of cycles
|
| 158 |
-
# calculate trace for cycle (t + 1) based on cycle t
|
| 159 |
-
m_Trace[t + 1, ] = m_Trace[t, ] %*% m_P
|
| 160 |
-
|
| 161 |
-
v_C[t + 1] = m_Trace[t + 1, ] %*% v_state_cost
|
| 162 |
-
|
| 163 |
-
v_Q[t + 1] = m_Trace[t + 1, ] %*% v_state_utility
|
| 164 |
-
|
| 165 |
-
}
|
| 166 |
-
|
| 167 |
-
############# PRIMARY ECONOMIC OUTPUTS #########################
|
| 168 |
-
|
| 169 |
-
# Total discounted costs
|
| 170 |
-
n_tot_cost = t(v_C) %*% v_dw
|
| 171 |
-
|
| 172 |
-
# Total discounted QALYs
|
| 173 |
-
n_tot_qaly = t(v_Q) %*% v_dw
|
| 174 |
-
|
| 175 |
-
############# OTHER OUTPUTS ###################################
|
| 176 |
-
|
| 177 |
-
# Total discounted life-years (sometimes used instead of QALYs)
|
| 178 |
-
n_tot_ly = t(m_Trace %*% c(1, 1, 1, 0)) %*% v_dw
|
| 179 |
-
|
| 180 |
-
####### RETURN OUTPUT ###########################################
|
| 181 |
-
out = list(m_Trace = m_Trace,
|
| 182 |
-
m_P = m_P,
|
| 183 |
-
l_params,
|
| 184 |
-
n_tot_cost = n_tot_cost,
|
| 185 |
-
n_tot_qaly = n_tot_qaly,
|
| 186 |
-
n_tot_ly = n_tot_ly)
|
| 187 |
-
|
| 188 |
-
return(out)
|
| 189 |
-
}
|
| 190 |
-
)
|
| 191 |
-
}
|
| 192 |
-
|
| 193 |
-
|
| 194 |
-
#' generate_psa_params
|
| 195 |
-
#'
|
| 196 |
-
#' @param psa_it number of PSA simulations to run
|
| 197 |
-
#' @param params a dataframe with the following columns:
|
| 198 |
-
#' parameter - an ID used as the object name
|
| 199 |
-
#' distribution e.g. beta
|
| 200 |
-
#' v1 the first value used in the distribution (e.g. shape)
|
| 201 |
-
#' v2 the second value used in the distribution (e.g. scale)
|
| 202 |
-
#'
|
| 203 |
-
#' @return a data-frame of PSA inputs for each variable in the list
|
| 204 |
-
#' @export
|
| 205 |
-
#'
|
| 206 |
-
#' @examples
|
| 207 |
-
#' generate_psa_params(psa_it = 1000,
|
| 208 |
-
#' params = data.frame(
|
| 209 |
-
#' parameter = c("p_HS1", "p_S1H"),
|
| 210 |
-
#' distribution = c("beta","beta"),
|
| 211 |
-
#' v1 = c(30, 60),
|
| 212 |
-
#' v2 = c(170, 60)
|
| 213 |
-
#' ))
|
| 214 |
-
generate_psa_params <- function(psa_it = 1000,
|
| 215 |
-
params = NULL){
|
| 216 |
-
|
| 217 |
-
m_psa <- matrix(data = NA,
|
| 218 |
-
nrow = psa_it,
|
| 219 |
-
ncol = nrow(params),
|
| 220 |
-
dimnames = list(NULL, params$parameter)
|
| 221 |
-
)
|
| 222 |
-
|
| 223 |
-
for(i in 1:nrow(params)){
|
| 224 |
-
|
| 225 |
-
m_psa[, i] <-
|
| 226 |
-
drawHelper(dist = params[i, "distribution"],
|
| 227 |
-
v1 = params[i, "v1"],
|
| 228 |
-
v2 = params[i, "v2"],
|
| 229 |
-
psa_it = psa_it)
|
| 230 |
-
|
| 231 |
-
}
|
| 232 |
-
|
| 233 |
-
return(as.data.frame(m_psa))
|
| 234 |
-
|
| 235 |
-
}
|
| 236 |
-
|
| 237 |
-
#' Run Model using psa inputs
|
| 238 |
-
#'
|
| 239 |
-
#' @param psa_inputs data-frame containing psa inputs for each variable.
|
| 240 |
-
#'
|
| 241 |
-
#' @return a data-frame with costs and qalys for three scenarios
|
| 242 |
-
#' @export
|
| 243 |
-
#'
|
| 244 |
-
#' @examples
|
| 245 |
-
#' run_model(psa_inputs)
|
| 246 |
-
run_model <-function(psa_inputs){
|
| 247 |
-
# generate the psa dataframe
|
| 248 |
-
my_psa_params <- generate_psa_params(psa_it = 1000,
|
| 249 |
-
params = psa_inputs)
|
| 250 |
-
|
| 251 |
-
# parameters defined for the base case.
|
| 252 |
-
my_params_basecase <- list(
|
| 253 |
-
p_HS1 = 0.15,
|
| 254 |
-
p_S1H = 0.5,
|
| 255 |
-
p_S1S2 = 0.105,
|
| 256 |
-
r_HD = 0.002,
|
| 257 |
-
hr_S1D = 3,
|
| 258 |
-
hr_S2D = 10,
|
| 259 |
-
hr_S1S2_trtB = 0.6,
|
| 260 |
-
c_H = 2000,
|
| 261 |
-
c_S1 = 4000,
|
| 262 |
-
c_S2 = 15000,
|
| 263 |
-
c_D = 0,
|
| 264 |
-
c_trtA = 12000,
|
| 265 |
-
c_trtB = 13000,
|
| 266 |
-
u_H = 1,
|
| 267 |
-
u_S1 = 0.75,
|
| 268 |
-
u_S2 = 0.5,
|
| 269 |
-
u_D = 0,
|
| 270 |
-
u_trtA = 0.95,
|
| 271 |
-
n_cycles = 75,
|
| 272 |
-
v_s_init = c(1, 0, 0, 0),
|
| 273 |
-
r_disc = 0.03
|
| 274 |
-
)
|
| 275 |
-
|
| 276 |
-
# run the inputs through the darth model - treat this as a black box
|
| 277 |
-
psa_output <- dampack::run_psa(
|
| 278 |
-
psa_samp = my_psa_params,
|
| 279 |
-
currency = "$",
|
| 280 |
-
params_basecase = my_params_basecase,
|
| 281 |
-
FUN = simulate_strategies,
|
| 282 |
-
outcomes = c("Cost", "QALY", "LY", "NMB"),
|
| 283 |
-
strategies = c("No_Treatment", "Treatment_A", "Treatment_B"),
|
| 284 |
-
progress = FALSE
|
| 285 |
-
)
|
| 286 |
-
|
| 287 |
-
# make a psa object - again, from dampack package
|
| 288 |
-
cea_psa <-
|
| 289 |
-
dampack::make_psa_obj(
|
| 290 |
-
cost = psa_output$Cost$other_outcome,
|
| 291 |
-
effect = psa_output$QALY$other_outcome,
|
| 292 |
-
parameters = psa_output$Cost$parameters,
|
| 293 |
-
strategies = psa_output$Cost$strategies,
|
| 294 |
-
currency = "$"
|
| 295 |
-
)
|
| 296 |
-
|
| 297 |
-
# costs and qalys for each iteration
|
| 298 |
-
psa_costs_df <- psa_output$Cost$other_outcome
|
| 299 |
-
psa_qalys_df <- psa_output$QALY$other_outcome
|
| 300 |
-
# add prefixes to column names
|
| 301 |
-
colnames(psa_costs_df) <- paste0("C_", colnames(psa_costs_df))
|
| 302 |
-
colnames(psa_qalys_df) <- paste0("E_", colnames(psa_qalys_df))
|
| 303 |
-
|
| 304 |
-
# combined the two data-frames by width
|
| 305 |
-
psa_df <- cbind(psa_costs_df, psa_qalys_df)
|
| 306 |
-
|
| 307 |
-
# return all outputs
|
| 308 |
-
return(psa_df)
|
| 309 |
-
}
|
| 310 |
-
|
| 311 |
-
|
| 312 |
-
|
| 313 |
-
#' Overwrite a parameter value given user input ...
|
| 314 |
-
#'
|
| 315 |
-
#' @param existing_df the existing data-frame of parameter inputs.
|
| 316 |
-
#' @param parameter the parameter to be changed (e.g. p_HS1)
|
| 317 |
-
#' @param distribution the new distribution (e.g. beta)
|
| 318 |
-
#' @param v1 the first parameter input (e.g. shape)
|
| 319 |
-
#' @param v2 the second parameter input (e.g. scale)
|
| 320 |
-
#'
|
| 321 |
-
#' @return returns the updated data-frame.
|
| 322 |
-
#'
|
| 323 |
-
#' @examples
|
| 324 |
-
#' overwrite_parameter_value(existing_df = psa_inputs,
|
| 325 |
-
#' parameter = "p_HS1",
|
| 326 |
-
#' distribution = "beta",
|
| 327 |
-
#' v1 = 40,
|
| 328 |
-
#' v2 = 160)
|
| 329 |
-
overwrite_parameter_value <- function(existing_df,
|
| 330 |
-
parameter,
|
| 331 |
-
distribution,
|
| 332 |
-
v1,
|
| 333 |
-
v2){
|
| 334 |
-
|
| 335 |
-
|
| 336 |
-
# which row does the parameter to be changed exist on
|
| 337 |
-
row_num <- which(existing_df$parameter == parameter)
|
| 338 |
-
# check that this parameter is editable
|
| 339 |
-
assertthat::assert_that(msg = "This value is not editable, please contact the data owner",
|
| 340 |
-
isTRUE(existing_df[row_num, "editable"]))
|
| 341 |
-
# input the new values
|
| 342 |
-
existing_df[row_num, c("distribution", "v1", "v2")] <- c(distribution, v1, v2)
|
| 343 |
-
# return the new data-frame
|
| 344 |
-
return(existing_df)
|
| 345 |
-
}
|
| 346 |
-
|
| 347 |
-
|
| 348 |
-
|
| 349 |
-
|
| 350 |
-
#' Draw random values from defined distributions
|
| 351 |
-
#'
|
| 352 |
-
#' @param dist distribution from which to draw random numbers, must be one of:
|
| 353 |
-
#' random uniform, normal, rlnorm, beta, fixed, gamma
|
| 354 |
-
#' @param v1 first parameter value, in random uniform this is min,
|
| 355 |
-
#' in gamma this is shape, in fixed this is the fixed value.
|
| 356 |
-
#' @param v2 second parameter, in gamma this is scale, this is missing in fixed
|
| 357 |
-
#' @param psa_it number of iterations
|
| 358 |
-
#'
|
| 359 |
-
#' @return returns a vector of random numbers sampled from the distribution
|
| 360 |
-
#'
|
| 361 |
-
#' @examples drawHelper(dist = "normal", v1 = 10, v2 = 1, psa_it = 1000)
|
| 362 |
-
drawHelper = function(dist,
|
| 363 |
-
v1,
|
| 364 |
-
v2,
|
| 365 |
-
psa_it){
|
| 366 |
-
|
| 367 |
-
v1 = as.numeric(v1)
|
| 368 |
-
v2 = as.numeric(v2)
|
| 369 |
-
|
| 370 |
-
switch(dist,
|
| 371 |
-
"random uniform" = {
|
| 372 |
-
runif(n = psa_it,
|
| 373 |
-
min = v1,
|
| 374 |
-
max = v2)
|
| 375 |
-
},
|
| 376 |
-
|
| 377 |
-
"normal" = {
|
| 378 |
-
rnorm(n = psa_it,
|
| 379 |
-
mean = v1,
|
| 380 |
-
sd = v2)
|
| 381 |
-
},
|
| 382 |
-
"rlnorm" = {
|
| 383 |
-
rlnorm(n = psa_it,
|
| 384 |
-
meanlog = v1,
|
| 385 |
-
sdlog = v2)
|
| 386 |
-
},
|
| 387 |
-
"beta" = {
|
| 388 |
-
rbeta(n = psa_it,
|
| 389 |
-
shape1 = v1,
|
| 390 |
-
shape2 = v2)
|
| 391 |
-
},
|
| 392 |
-
"fixed" = {
|
| 393 |
-
rep(x = v1,
|
| 394 |
-
times = psa_it)
|
| 395 |
-
},
|
| 396 |
-
"gamma" = {
|
| 397 |
-
rgamma(n = psa_it,
|
| 398 |
-
shape = v1,
|
| 399 |
-
scale = v2)
|
| 400 |
-
}
|
| 401 |
-
)
|
| 402 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|