summarize_data.Rmd

---
title: "Summary of the state of the PACS data set"
csl: the-american-naturalist.csl
output:
  html_document:
    theme: cerulean
    toc: yes
  pdf_document:
    toc: yes
<!-- bibliography: references.bib -->
editor_options: 
  chunk_output_type: console
---

<!--
IMAGES:
Insert them with: ![alt text](image.png)
You can also resize them if needed: convert image.png -resize 50% image.png
If you want to center the image, go through HTML code:
<div style="text-align:center"><img src ="image.png"/></div>

REFERENCES:
For references: Put all the bibTeX references in the file "references.bib"
in the current folder and cite the references as @key or [@key] in the text.
Uncomment the bibliography field in the above header and put a "References"
title wherever you want to display the reference list.
-->

<style type="text/css">
.main-container {
  max-width: 1370px;
  margin-left: auto;
  margin-right: auto;
}
</style>

```{r general options, include = FALSE}
knitr::knit_hooks$set(
  margin = function(before, options, envir) {
    if (before) par(mgp = c(1.5, .5, 0), bty = "n", plt = c(.105, .97, .13, .97))
    else NULL
  },
  prompt = function(before, options, envir) {
    options(prompt = if (options$engine %in% c("sh", "bash")) "$ " else "> ")
  })

knitr::opts_chunk$set(margin = TRUE, prompt = TRUE, comment = "", echo = FALSE,
                      collapse = TRUE, cache = FALSE, autodep = TRUE, message = FALSE,
                      dev.args = list(pointsize = 11), fig.height = 3.5,
                      fig.width = 4.24725, fig.retina = 2, fig.align = "center")

options(width = 137)
```

```{r data}
pacs <- readr::read_csv("data/pacs.csv",
                        col_types = paste(c("icfnD", rep("c", 5), rep("D", 4), rep("f", 3)), collapse = ""))
```

```{r packages}
library(dplyr)
library(ecomore)
library(lubridate)
library(magrittr)
library(padr)
library(sf)
```

```{r problems directory}
if (!dir.exists("problems")) dir.create("problems")
```

The CSV file of the **PACS data set** is
[here](https://github.com/ecomore2/pacs/blob/master/data/pacs.csv). Go
[here](https://raw.githubusercontent.com/ecomore2/pacs/master/data/pacs.csv)
if you want to copy and paste this CSV file to your computer. It looks like

```{r the data}
pacs
```

## Dates of reported cases

```{r time range of pacs}
minmaxdates <- function(f) {
  pacs %>%
    select(onset, hospitalization, consultation, sample_collection) %>% 
    lapply(f, na.rm = TRUE) %>% 
    purrr::reduce(c) %>% 
    f()
}  
```

```{r problems in dates}
threshold <- 15
problems <- pacs %>% 
  add_dates_differences() %>% 
  filter_at_any("diff_", function(x) abs(x) > threshold, starts_with) %>% 
  select(id, onset, hospitalization, consultation, sample_collection)

write.csv(problems, "problems/dates.csv", FALSE, row.names = FALSE)
```

```{r number of cases with no dates}
no_dates <- pacs %>%
  filter(is.na(onset), is.na(hospitalization), is.na(consultation), is.na(sample_collection)) %>% 
  select(id, onset, hospitalization, consultation, sample_collection)
write.csv(no_dates, "problems/no_dates.csv", FALSE, row.names = FALSE)
```

As of `r as.Date(Sys.time())`, the PACS data set contains **`r nrow(pacs)` cases**
reported from **`r minmaxdates(min)`** to **`r minmaxdates(max)`**. There are
`r nrow(problems)` cases for which there are more than `r threshold` days between
at least one pair of dates among `onset`, `hospitalization`, `consultation` and
`sample_collection`. A CSV file of these cases is
[here](https://github.com/ecomore2/pacs/blob/master/problems/dates.csv). Go
[here](https://raw.githubusercontent.com/ecomore2/pacs/master/problems/dates.csv)
if you want to copy and paste this CSV file to your computer. Furthermore, there
are `r nrow(no_dates)` cases with no date at all. A CSV file of these cases is
[here](https://github.com/ecomore2/pacs/blob/master/problems/no_dates.csv). Go
[here](https://raw.githubusercontent.com/ecomore2/pacs/master/problems/no_dates.csv)
if you want to copy and paste this CSV file to your computer. After removing these
**`r nrow(problems)` cases with date problems** as well as the **`r nrow(no_dates)`
cases with no dates at all** and inferring the missing onset dates from
`hospitalization`, `consultation` or `sample_collection`, the
time series of the number of suspected cases per week looks like:

```{r makes weekly incidences}
byweek <- pacs %>% 
  mutate(onset2 = correct_onset(.)) %>% 
  select(-dob, -onset, -hospitalization, -consultation, -sample_collection) %>% 
  filter(! is.na(onset2)) %>% 
  thicken("week") %>% 
  select(-onset2) %>% 
  group_by(onset2_week) %>% 
  tally() %>% 
  ungroup() %>% 
  pad("week")
```

```{r plotting weekly incidences, fig.width = 8}
lab <- min(year(byweek$onset2_week)):(max(year(byweek$onset2_week)) + 1)
ats <- as.Date(paste0(lab, "-01-01"))

with(byweek, plot(onset2_week, n, type = "n", axes = FALSE, xlim = range(ats),
                  xlab = NA, ylab = "number of suspected cases"))

axis(1, ats, lab); axis(2)
abline(v = ats, col = "grey")
abline(h = seq(25, 150, 25), col = "grey")

with(byweek, points(onset2_week, n, type = "h", col = "blue"))
```

## Confirmation tests

Here the presence of a confirmation test is based on the information in the
variables `pcr`, `ns1` and `serotype`. Indeed, some cases without any information
in `pcr` or `ns1` may still have an identified serotype. For example:

```{r link between PCR and NS1 on one hand and serotype on the other hand}
pacs %>% 
  filter(is.na(pcr) | pcr != "positive",
         is.na(ns1) | ns1 != "positive",
         grepl("dengue", serotype)) %>% 
  select(id, pcr, ns1, serotype)
```

```{r adding test information}
pacs %<>% add_tests()
```

The split of data according to the availability of time information and
confirmation test is:

```{r table time and test}
pacs %>% 
  mutate(onset2 = correct_onset(.),
         time_info = ! is.na(onset2)) %$%
  table(tested, time_info) %>% 
  addmargins()
```

Among the cases for which a confirmation test is available, the split of data
according to positivity and time information is:

```{r table time and positivity}
pacs %>% 
  mutate(onset2 = correct_onset(.),
         time_info = ! is.na(onset2)) %>%
  filter(tested) %$% 
  table(confirmed, time_info) %>% 
  addmargins()
```

Out of the `r nrow(pacs)` reported cases,
`r sum(pacs$tested, na.rm = TRUE)` (`r round(100 * sum(pacs$tested, na.rm = TRUE) / nrow(pacs))` %)
have a conclusive confirmation test, of which
`r sum(pacs$confirmed, na.rm = TRUE)`
(`r round(100 * sum(pacs$confirmed, na.rm = TRUE) / sum(pacs$tested, na.rm = TRUE))`
%) are positive:

```{r utilitary functions}
recode_na <- function(x) ifelse(is.na(x), "NA", x)
refactor <- function(x)
  factor(sub("_", " ", x),
         c("positive", "equivocal", "negative", "not finished", "not tested", "NA"))
```

```{r table PCR and NS1 all}
pacs %>% 
  mutate(PCR = refactor(recode_na(as.character(pcr))),
         NS1 = refactor(recode_na(as.character(ns1)))) %$%
  table(PCR, NS1) %>% 
  addmargins()
```

Stratifying by the reported cases with or without problem in missing dates (i.e.
the `r nrow(problems)` cases with date problems as well as the `r nrow(no_dates)`
cases with no dates at all), it gives:

```{r preparing data}
tmp <- pacs %>% 
  mutate(PCR = refactor(recode_na(as.character(pcr))),
         NS1 = refactor(recode_na(as.character(ns1))),
         missing_date = id %in% c(problems$id, no_dates$id))
```

```{r table PCR and NS1 missing date}
tmp %>%
  dplyr::filter(missing_date) %$% 
  table(PCR, NS1) %>% 
  addmargins()
```

and 

```{r table PCR and NS1 non missing date}
tmp %>%
  filter(! missing_date) %$% 
  table(PCR, NS1) %>% 
  addmargins()
```

The status of the serotypes tests is as follow:

```{r serotypes}
pacs %>% 
  mutate(serotype = factor(recode_na(as.character(serotype)),
                           c("dengue_1", "dengue_2", "dengue_3", "dengue_4",
                             "not_identified", "not_finished", "not_tested", "NA"))) %>%
  group_by(serotype) %>% 
  tally() %>% 
  ungroup()
```

## Geography

The number of missing values for province, district and village:

```{r table presence absence province district village}
pacs %>% 
  transmute(province = ! is.na(province),
            district = ! is.na(district),
            village  = ! is.na(village)) %>% 
  lapply(table)
```

where `TRUE` means available informaiton and `FALSE` means missing information.
The combinations of missing values for these 3 variables are:

```{r cross table province district village info}
pacs %>% 
  transmute(province = ! is.na(province),
            district = ! is.na(district),
            village  = ! is.na(village)) %>% 
  group_by(province, district, village) %>% 
  tally() %>% 
  ungroup()
```

The reported cases with village information but missing province information:

```{r village info but missing province info}
pacs %>%
  filter(is.na(province), ! is.na(village)) %>% 
  select(id, province, district, village)
```

The reported cases with village information but missing district information:

```{r village info but missing district info}
(tmp <- pacs %>%
  filter(is.na(district), ! is.na(village)) %>% 
  select(id, province, district, village))
write.csv(tmp, "problems/village_but_no_district.csv", FALSE, row.names = FALSE)
```

A CVS file of all the case is
[here](https://github.com/ecomore2/pacs/blob/master/problems/village_but_no_district.csv). Go
[here](https://raw.githubusercontent.com/ecomore2/pacs/master/problems/village_but_no_district.csv)
if you want to copy and paste this CSV file to your computer.

**WE SHOULD BE ABLE TO FIND THIS DISTRICTS BASED ON VILLAGE (AND PROVINCE) INFORMATION**

```{r the province map}
f <- "gadm36_LAO_1_sf.rds"
if(! file.exists(f))
  download.file(paste0("https://biogeo.ucdavis.edu/data/gadm3.6/Rsf/", f), f)
lao1 <- readRDS(f)
```

The provinces names that are not official Lao province name:

```{r provinces names that are not lao provinces, message = FALSE}
setdiff(unique(pacs$province), lao1$NAME_1)
```

```{r filtering for Vientiane prefecture}
vt <- pacs %>% 
  filter(province == "Vientiane [prefecture]")
```

`r round(100 * nrow(vt) / nrow(filter(pacs, ! is.na(province))))` % of reported
cases (`r nrow(vt)`) are from Vientiane prefecture:

```{r province distribution}
pacs %>% 
  group_by(province) %>% 
  tally() %>% 
  ungroup() %>% 
  arrange(desc(n)) %>% 
  print(n = 22)
```

The distribution of cases among the different districts of Vientiane prefecture
looks like:

```{r district distribution}
vt %>% 
  group_by(district) %>% 
  tally() %>% 
  ungroup() %>% 
  arrange(desc(n))
```

There is village information for `r sum(!is.na(pacs$village))` cases
(`r round(100 * sum(!is.na(pacs$village)) / nrow(pacs))` % of the total number
of cases). In Vientiane prefecture, the split of cases depending on available
information on village, test and time is

```{r split village test time}
(tmp <- vt %>% 
  add_tests() %>% 
  rename(test_info = tested) %>% 
  mutate(time_info = ! is.na(correct_onset(.)),
         village_info = ! is.na(village)) %>%
  group_by(village_info, test_info, time_info) %>% 
  tally() %>% 
  ungroup())
```

which means that there are `r select(filter(tmp, village_info, test_info, time_info), n)`
(`r round(100 * select(filter(tmp, village_info, test_info, time_info), n) / sum(tmp$n))` %)
cases in Vientiane prefecture for which we have village, time and confirmation test
information. If we consider all cases (tested or not), the split then becomes:

```{r table village time}
vt %>% 
  mutate(time_info = ! is.na(correct_onset(.)),
         village_info = ! is.na(village)) %$%
  table(village_info, time_info) %>% 
  addmargins()
```

There are `r sum(select(filter(tmp, village_info, time_info), n))` cases
(`r round(100 * sum(select(filter(tmp, village_info, time_info), n)) / sum(tmp$n))` %)
for which we have both time and village information.

## Ages

```{r making age2}
tmp <- pacs %>% 
  mutate(onset2 = correct_onset(.),
         age2 = as.numeric((onset2 - dob) / 365))
```

For the reported cases below, the dates of birth are not compatible with the age.
It seems that the year of the date of birth has incorrectly been taken as the
same as the year of the onset:

```{r wrong ages}
tmp %>%
  filter(age2 < 0) %>% 
  select(id, dob, age, onset2)
```

```{r looking for suspicious age}
mod <- lm(age2 ~ age, tmp)
cft <- coef(mod)
threshold <- 2
age_check <- tmp %>% 
  filter(age2 > cft[1] + cft[2] * age + threshold | age2 < cft[1] + cft[2] * age - threshold, age2 > 0) %>% 
  select(id, dob, onset, hospitalization, consultation, sample_collection, onset2, age, age2)
write.csv(age_check, "problems/age_check.csv", FALSE, row.names = FALSE)
```

There are `r nrow(age_check)` cases for which the reported age, onset, and date
of birth are not quite compatible:

```{r showing suspicious ages}
age_check
```

where `onset2` is calculated from `onset`, `hospitalization`, `consultation` and
`sample_collection`, and `age2 = onset2 - dob`. A CVS file of all the case is
[here](https://github.com/ecomore2/pacs/blob/master/problems/age_check.csv). Go
[here](https://raw.githubusercontent.com/ecomore2/pacs/master/problems/age_check.csv)
if you want to copy and paste this CSV file to your computer.