setup-exposure-data.Rmd

---
title: "Processing exposure data"
author: "Bryan Mayer"
date: "`r Sys.Date()`"
output: workflowr::wflow_html
---

This Rmarkdown script creates the exposure data for the dose-response analysis.


```{r,  warning = F, message = F, echo = F}
knitr::opts_chunk$set(
  comment = NA,
  fig.align = "center",
  tidy = FALSE
)

```

```{r load-packages-data, warning = F, message = F, echo = F}
library(zoo)
library(tidyverse)
library(kableExtra)

virusMeltedDataDemoAllInfant = read_csv("data/PHICS_transmission_data.csv")
lower_limit = min(subset(virusMeltedDataDemoAllInfant, count > 0)$count)
source("code/processing_functions.R")

save_data = T
if(!save_data) print("Data was not updated or saved on this compile.")
```

# Subset exposure data 

- no HHV-8, EBV, or HSV  (no, late, and limited infections)  
- Exclude family AZ in HHV-6 because no infection and all 0 viral loads   

```{r subset-data, warning = F, message = F}
exposure_times = virusMeltedDataDemoAllInfant %>%
  subset(idpar != "P" & !virus %in% c("HHV-8", "HSV", "EBV") & 
           !(FamilyID == "AZ" & virus == "HHV-6")
         )

#merge later, relic from old code
age_data = select(exposure_times, FamilyID, enrollment_age) %>%
  distinct()

```

# Create exposure variable by household member

There are multiple secondary children (S) to aggregate.

```{r mult-sib}

multS_exposure = exposure_times %>% 
  subset(idpar == "S") %>%
  group_by(FamilyID, virus) %>%
  mutate(
    idpar2 = str_split_fixed(PatientID, "-", n = 2)[,2],
    total_SC = n_distinct(idpar2)
  ) %>%
  subset(total_SC > 1) %>%
  group_by(FamilyID, virus) %>%
  mutate(
    pvalue = kruskal.test(count ~ idpar2)$p.value
  ) 

multS_exposure %>%
  group_by(FamilyID, virus, idpar2, pvalue) %>%
  summarize(
    mean_VL = mean(count),
    total = n(),
    mean_n = stat_paste(mean_VL, total, digits = 2, trailing_zeros = F)
  ) %>%
  select(FamilyID, virus, idpar2, mean_n, pvalue) %>%
  ungroup() %>%
  spread(idpar2, mean_n, fill = "---") %>%
  mutate(flag = pvalue < 0.1) %>%
  arrange(virus, pvalue, FamilyID) %>%
  select(everything(), pvalue, flag) %>%
  kable(digits = 3) %>%
  kable_styling(full_width = F)
  
```


```{r multi-sib-pl}

multS_exposure %>%
  subset(pvalue < 0.1) %>%
  ggplot(aes(x = infant_days, y = count, colour = idpar2)) +
  geom_line() + geom_point() +
  facet_wrap(~virus+FamilyID) +
  theme(legend.position = "top")

```

```{r idpar-variable}

# combines siblings into one exposure

primary_exposures_idpar = exposure_times %>% 
  group_by(FamilyID, idpar, virus, infant_days, momhiv, days_from_final_infant, final_infant_day) %>%
  summarize(
    total_contributed_idpar = n(),
    who_contributed_idpar = paste(str_split_fixed(PatientID, "-", n = 2)[,2], collapse = ", "),
    exposure = log10(sum(10^count, na.rm = T)),
    infected = infantInfection[1]
    )  %>% 
  group_by(FamilyID, virus, idpar) %>%
  mutate(
    unique_id = paste(FamilyID, virus, idpar, sep = "-"),
    min_time_from_end = min(days_from_final_infant),
    exposure = if_else(exposure <= 1, 0, exposure)
  )

with(primary_exposures_idpar, ftable(idpar, total_contributed_idpar))

```

# Combine exposure data and create outcome

Here, we leave counts (exposures) at times relative to infant birth, and create the outcome variable describing infant infection status in the following week.

The outcome variable is defined so that the infectious exposure occured 4-14 days prior to infected detection. 


```{r all-exposure}

# make a new dataset organized by time before swab, use new days, this is for household
# create outcome variable

all_exposures_raw = primary_exposures_idpar %>% 
  rename(count = exposure) %>%
  filter(days_from_final_infant > 0) %>%  # these are either censored cases or infections (negative = post-infection)
  group_by(FamilyID, idpar, virus) %>% 
  mutate(
    final_exposure = days_from_final_infant == min(days_from_final_infant),
    infectious_1wk = if_else(days_from_final_infant <= 14 & final_exposure & infected == 1, 1, 0)
    )

testthat::expect_equal(min(subset(all_exposures_raw, infected == 0)$days_from_final_infant),
                       expected = 7, 
                       info = "check if all uninfected measurements are at least a week from final measurement (ie, no infection one week later)")

testthat::expect_equal(min(subset(all_exposures_raw, infected == 1)$days_from_final_infant),
                       expected = 4, 
                       info = "check if all infected measurements > 4")

```

# Combine exposures into weekly variable

## Create weekly categories

Plot checks that recoding was done right (no overlap on y-axis across steps)

```{r setup-weeks}

wk_cuts = 0:ceiling(max(primary_exposures_idpar$infant_days)/7) * 7
wk_labels = head(wk_cuts, -1)/7
all_exposures_raw$infant_wks = cut(all_exposures_raw$infant_days, include.lowest = T, ordered_result = T,
                                 breaks = wk_cuts, labels = wk_labels)

all_exposures_raw$final_infant_wk = as.numeric(as.character(cut(all_exposures_raw$final_infant_day, 
                                                   include.lowest = T, ordered_result = T,
                                 breaks = 0:100 * 7, labels = F))) - 1

testthat::expect_equal(min(all_exposures_raw$final_infant_wk) ,
                       expected = 0, 
                       info = "check infant_wk rescale")

all_exposures_raw$infant_wks = as.numeric(as.character(all_exposures_raw$infant_wks))

ggplot(arrange(all_exposures_raw, infant_days), aes(y = infant_wks, x = infant_days)) +
  geom_tile()

```

## Fill in missing weeks

- All interpolation is done using the interpolation of the log viral load (`zoo:na.approx`). 
- `map_df` was used so that the data is summarized by a refactored infant_wk so `complete` can be used to find missing weeks for a giving exposure set (which varies by infant and exposure source). This could be done with group_by and nest.

First plot displays extent of left censoring
Second plot shows where intepolation occured.

```{r interpolate-missing-weeks, fig.height=9}

all_exposures = map_df(unique(all_exposures_raw$unique_id), function(uid){
  temp_data = subset(ungroup(all_exposures_raw), unique_id == uid) %>%
    mutate(first_infant_week = min(infant_wks))
  
  # refactor levels for complete()
  temp_data$infant_wks = factor(temp_data$infant_wks, 
                                levels = 0:max(temp_data$infant_wks)) 

  out = temp_data %>%
  group_by(FamilyID, unique_id, momhiv, virus, idpar, 
           infant_wks, first_infant_week, final_infant_wk) %>%
  summarize(
    count = max(count),
    infected = unique(infected),
    infectious_1wk = max(infectious_1wk),
    final_exposure = max(final_exposure)
    ) %>% 
    ungroup() %>%
    complete(infant_wks, nesting(FamilyID, momhiv, virus, idpar, 
                                 first_infant_week, final_infant_wk, 
                               infected, unique_id)) %>%
    arrange(infant_wks) %>%
    mutate(
      interpolate_idpar = if_else(is.na(count), unique(temp_data$idpar), ""),
      infant_wks = as.numeric(as.character(infant_wks)),
      infectious_1wk = na.fill(infectious_1wk,  0),
      final_exposure = na.fill(final_exposure,  0),
      count = na.approx(count, rule = 2)
      )
  if(nrow(temp_data) == 1) return(out)

  testthat::expect_equal(n_distinct(diff(out$infant_wks)), expected = 1, 
                         info=paste("Check infant_wks interpolation worked (common interval)",
                                    unique(out$unique_id)))
  testthat::expect_equal(unique(diff(out$infant_wks)), expected = 1, 
                         info=paste("Check infant_wks interpolation worked (interval = one)",
                                    unique(out$unique_id)))
  if(any(out$interpolate_idpar != "")){
    testthat::expect_equal(unique(out$infectious_1wk[out$interpolate_idpar != ""]), 
                             expected = 0, 
                           info=paste("Check infectious_1wk interpolation",
                                      unique(out$unique_id)))
    testthat::expect_equal(unique(out$final_exposure[out$interpolate_idpar != ""]), 
                             expected = 0, 
                           info=paste("Check infectious_1wk interpolation",
                                      unique(out$unique_id))) 
    }

  out
  
}) %>%
  mutate(
    count = if_else(count >= lower_limit - 1, count, 0) # the 1 is a small tolerance factor
  )


testthat::expect_equal(all_exposures %>% group_by(unique_id) %>% 
  summarize(test = sum(infectious_1wk), test2 = sum(final_exposure)) %>%
  filter(test > 1 | test2 > 1) %>% nrow(), expected = 0, 
  info = "Verifying infectious_1wk and final_exposure after interpolation")
  

all_exposures %>%
  select(FamilyID, virus, idpar, first_infant_week) %>%
  distinct() %>%
  group_by(virus, first_infant_week, idpar) %>%
  summarize(total = n()) %>%
  ggplot(aes(x = factor(first_infant_week), y = total)) +
  geom_histogram(stat = "identity") +
  geom_text(aes(label = total), vjust = 1) +
  facet_grid(idpar~virus)

all_exposures %>%
  group_by(FamilyID, infant_wks, virus) %>%
  summarize(
    interpolate = str_c(interpolate_idpar, collapse = "")
  ) %>%
  arrange(FamilyID, virus) %>%
  ggplot(aes(y = infant_wks, x = FamilyID, fill = factor(interpolate))) +
  geom_tile() +
  scale_fill_manual("", values = c("black", "red", "blue", "gray"),
                    breaks = c("", "S", "MS", "M"),
                    labels = c("", "Interpolate - S", 
                               "Interpolate - M,S", "interpolate - M")) +
  coord_flip() +
  labs(y = "infant weeks post-dob pre-infection") +
  theme_bw() +
  theme(legend.position = "top", axis.text.y = element_text(size = 7)) +
  facet_wrap(~virus, nrow = 2, strip.position = "right", scales = "free_y")

```


```{r wide-data}

all_exposures_wide = all_exposures %>%
  group_by(FamilyID, virus, infant_wks) %>%
  mutate(
    interpolate_idpar = str_trim(str_c(sort(unique(interpolate_idpar)), collapse = " "))
  ) %>%
  ungroup() %>%
  reshape2::dcast(FamilyID + virus + infant_wks + infectious_1wk + final_infant_wk +
                                   infected + momhiv + final_exposure + interpolate_idpar ~ idpar,
                                     data = ., value.var = "count") %>%
  mutate(
    HH = log10(10^M + 10^S),
    HH = if_else(HH <= lower_limit, 0, HH)
    ) %>% 
  ungroup()

exposure_data = all_exposures_wide %>%
  filter(!is.na(S)  & !is.na(M)) %>%
  group_by(FamilyID, virus) %>%
  mutate(
    obs_infected = infected * max(infectious_1wk), 
    final_wk = max(infant_wks),
    outcome_time = ifelse(obs_infected, final_infant_wk, final_wk + 1)
  ) %>%
  ungroup() %>%
  left_join(age_data, by = "FamilyID")


# extra step because empty string patterns are not supported

tmp_chr = "-"
exposure_data_long = exposure_data %>%
  gather(idpar, count, S, M, HH) %>%
  mutate(
    interpolate_idpar_tmp = if_else(interpolate_idpar == "", tmp_chr, interpolate_idpar),
    interpolated = if_else(interpolate_idpar != "" & idpar == "HH", T,
                           str_detect(interpolate_idpar_tmp, idpar))
  ) %>%
  select(-interpolate_idpar_tmp)

testthat::expect_equal(exposure_data %>% group_by(FamilyID, virus) %>% 
  summarize(test = sum(infectious_1wk), test2 = sum(final_exposure)) %>%
  filter(test > 1 | test2 > 1) %>% nrow(), expected = 0, 
  info = "Verifying final exposure has at most one infectious dose per infant")

testthat::expect_equal(exposure_data_long %>% group_by(FamilyID, idpar, virus) %>% 
  summarize(test = sum(infectious_1wk), test2 = sum(final_exposure)) %>%
  filter(test > 1 | test2 > 1) %>% nrow(), expected = 0, 
  info = "Verifying final exposure has at most one infectious dose per infant")

# save the data
if(save_data) {
  write_csv(exposure_data, "data/exposure_data.csv")
  write_csv(exposure_data_long, "data/exposure_data_long.csv")
}

```