getdelaydata: Get delay data for one or more cities for delay analysis

In claudia-codeco/AlertTools: utility functions for the Alerta Dengue System

Description

Gets delay data for one or more cities. Internal function used in the delay fitting using inla.

Usage

getdelaydata(
  cities,
  nyears = 2,
  cid10 = "A90",
  lastday = Sys.Date(),
  datasource = con
)

Arguments

cities	vector with geocodes
nyears	number of years of data used for fitting: Default = 2 years
cid10	disease code, Default is dengue. "A92.0" for chik, "A92.8" for zika
lastday	last digitation day
datasource	valid connection to database

Details

# bayesnowcasting ——————————————————————— #'@description This function estimate the Bayesian nowcast (new version) #'@title Correct incidence data with notification delay (nowcasting). #'@export #'@param obj data.frame with individual cases, containing columns municipio_geocodigo, dt_notific, dt_sin_pri, dt_digita #'@param Dmax for the "bayesian" method. Maximum number of weeks that is modeled #'@param Fim date for the nowcasting (date). Default is today. #'@return data.frame with median and 95percent confidence interval for the #'predicted cases-to-be-notified) #'@examples #' # bayesian #'dd <- getdelaydata(cities=3167202, nyears=1, cid10="A90", datasource=con) #'resfit<-bayesnowcasting(dd) #'resfit

bayesnowcasting <- function(d, Dmax = 10, Fim = Sys.Date())

# check input if(is.null(names(d))) message("bayesnowcasting: no data, returning NULL") return(NULL)

if(nrow(d) < 50) message("bayesnowcasting: few data, returning NULL") return(NULL)

#d contains columns assert_that(all(c("municipio_geocodigo", "dt_notific", "dt_sin_pri", "dt_digita") municipio_geocodigo, dt_notific, dt_sin_pri, dt_digita")

d <- d dt_notific_epiweek = epiweek(dt_notific), dt_notific_aux = as.numeric(format(as.Date(dt_notific), " dt_notific_week = dt_notific + 6 - dt_notific_aux, dt_notific_epiyear = epiyear(dt_notific), dt_digita_epiweek = epiweek(dt_digita), dt_digita_epiyear = epiyear(dt_digita), delay_epiweek = ifelse( dt_digita_epiyear == dt_notific_epiyear, as.numeric(dt_digita_epiweek - dt_notific_epiweek), as.numeric(dt_digita_epiweek - dt_notific_epiweek) + 52 ) )

# casos observados por semama obs <- d group_by(dt_notific_week) summarize(casos = n())

Inicio <- min(d$dt_notific_week) # Ultimo dia com notificacao ou digitacao #Fim <- max(d$dt_digita, d$dt_notific, na.rm = T) #Fim <- max(d$dt_digita, d$dt_notific, na.rm = T) #Fim <- Fim + 6 - as.numeric(format(as.Date(Fim), "

# contruindo a matriz de atraso - running triang tibble(Date = c(Inicio,Fim) ) mutate(Weekday = weekdays(Date) )

tbl.dates <- tibble(dt_notific_week = seq(Inicio, Fim, by = 7)) rowid_to_column(var = "Time") Today = max(tbl.dates$Time)

dados.ag <- d filter( dt_digita <= Fim) mutate( delay_epiweek = ifelse(delay_epiweek > Dmax, NA, delay_epiweek) ) drop_na(delay_epiweek) group_by(dt_notific_week, delay_epiweek) dplyr::summarise( Casos = n() ) # Passando para o formato wide spread(key = delay_epiweek, value = Casos) # Adicianoando todas as data, algumas semanas nao houveram casos full_join( y = tbl.dates, by = "dt_notific_week" ) # Voltando para o formato longo gather(key = "Delay", value = Casos, -dt_notific_week, -Time) mutate( Delay = as.numeric(Delay), # Preparing the run-off triangle Casos = ifelse( test = (Time + Delay) <= Today, yes = replace_na(Casos, 0), no = NA) ) dplyr::rename( Date = dt_notific_week) # Sorting by date dplyr::arrange(Date)

dados.ag.full <- d group_by(dt_notific_week) dplyr::summarise( Total = n() ) # Adicianoando todas as data, algumas semanas nao houveram casos right_join( y = tbl.dates, by = "dt_notific_week" ) mutate( Total = replace_na(Total, 0) ) dplyr::rename( Date = dt_notific_week) # Sorting by date dplyr::arrange(Date)

# Model equation model.dengue <- Casos ~ 1 + f(Time, model = "rw2", constr = T #hyper = list("prec" = list(prior = "loggamma", param = c(0.001, 0.001) )) #hyper = list("prec" = list(prior = half_normal_sd(.1) )) ) + f(Delay, model = "rw1", constr = T # hyper = list("prec" = list(prior = "loggamma", param = c(0.001, 0.001) )) #hyper = list("prec" = list(prior = half_normal_sd(.1) )) ) + # Efeito tempo-atraso f(TimeDelay, model = "iid", constr = T # hyper = list("prec" = list(prior = "loggamma", param = c(0.001, 0.001) )) )

output.dengue <- nowcast.INLA( model.day = model.dengue, dados.ag = dados.ag mutate(TimeDelay = paste(Time, Delay)) )

pred.dengue <- nowcasting(output.dengue, dados.ag, Dm = Dmax, Fim = max(dados.ag$Date))

pred.dengue.summy <- pred.dengue dplyr::summarise( Mean = mean(Casos, na.rm = TRUE), Median = median(Casos, na.rm = TRUE), LI = quantile(Casos, probs = 0.025, na.rm = TRUE), LS = quantile(Casos, probs = 0.975, na.rm = TRUE) ) left_join(obs, by = c("Date" = "dt_notific_week"))

pred.dengue.summy$SE <- daySEday(pred.dengue.summy$Date)$SE

pred.dengue.summy

Value

list with d = data.frame with data.

Examples

dados <- getdelaydata(cities=4100905, nyears=2, cid10="A90", datasource=con)  # Not run without connection

claudia-codeco/AlertTools documentation built on Aug. 12, 2021, 9:58 a.m.