dt_code.Rmd

---
title: "merge data"
author: "Lisa"
date: "4 juillet 2022"
output: html_document
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```



# Analysis for manuscript: "Low seroprotection against diphtheria and tetanus in Lao adolescents"




## Background

Code to reproduce the results presented in **Vantava et al. (2022). Low seroprotection against diphtheria and tetanus in Lao adolescents**. 

Accepted at Tropical Medicine & International Health.




## preparation


```{r options}

options(scipen = 999) # suppress scientific notion

```



### packages

```{r packages}

library(MASS) 
library(epitools) 
library(car)
library(tidyverse) 
library(emmeans)
library(sjPlot)
library(ggforce)
library(ggpubr)    
library(effects)  
library(rstatix)
library(gtsummary) 

```



### dataset import


```{r dataset}

# read in data
# data is already pre-processed
# new variables were created
# only variables needed for analysis were selected
data <- read.csv("data/lao-h-078_data_minimal.csv", header=T, dec=",", sep=";",
                 stringsAsFactors = TRUE)


```


### overview

```{r overview}

glimpse(data)

```


```{r variables}

# set the reference levels; check order of categories

# check variable "hcf"
data %>% count(hcf) # factors are not in the right order
data %>% mutate(hcf = factor(hcf, levels=c("ch", "ph", "dh", "hc"))) -> data

# check variable "sex"
data %>% mutate(sex = factor(sex, levels=c("male", "female"))) -> data

# check variable "prov"
# vtn (= vientiane capital) should be the reference
data %>% mutate(prov = factor(prov, levels=c("vtn", "blx"))) -> data

# check variable "district"
data %>% mutate(district = factor(district, levels=c("vientiane", 
                                                     "paksan",
                                                     "pakkading"))) -> data

```



## analysis


```{r table 1 characteristics and serology results overall}

# TABLE participant testing data overview
# Table with participant characteristics is already in previous prublication!

############## NUMBERS FOR TABLE S1 
############## overall (both study sites combined)

data %>% 
  mutate(prov = ifelse(prov == "vtn", "vientiane capital", "bolikhamxay")) %>%
  mutate(prov = factor(prov, levels=c("vientiane capital", "bolikhamxay"))) %>% 
  dplyr::select(prov, 
         age, 
         sex, 
         ethnic_group,
         tetanus,
         diphtheria) %>% # keep only columns of interest
  tbl_summary(     
    #by = prov,     # stratify entire table by outcome
    statistic = list(all_continuous() ~ "{mean} ({sd})",  # stats and format for continuous columns
                     all_categorical() ~ "{n} / {N} ({p}%)"),   # stats and format for categorical columns
    digits = all_categorical() ~ 2,                              # rounding for continuous columns
    #digits = 2,  
    type   = all_categorical() ~ "categorical",  # force all categorical levels to display
    label  = list(                    # display labels for column names
      prov   ~ "province",                           
      age    ~ "age in years",
      sex ~ "sex",
      ethnic_group ~ "ethnicity",
      tetanus ~ "tetanus immunity level",
      diphtheria    ~ "diphtheria immunity level"),
    missing_text = "missing"          # how missing values should display
  ) %>% 
   modify_caption("**Table S1. Participant characteristics and serology results - overall**")


```


```{r table s1 characteristics and serology results by province}

############## NUMBERS FOR TABLE S1 
############## split by recruitment site

data %>%
  mutate(prov = ifelse(prov == "vtn", "vientiane capital", "bolikhamxay")) %>%
  mutate(prov = factor(prov, levels=c("vientiane capital", "bolikhamxay"))) %>% 

  dplyr::select(prov, 
         age, 
         sex, 
         ethnic_group,
         tetanus,
         diphtheria) %>% # keep only columns of interest
  tbl_summary(     
    by = prov,     # stratify entire table by outcome
    statistic = list(all_continuous() ~ "{mean} ({sd})",  # stats and format for continuous columns
                     all_categorical() ~ "{n} / {N} ({p}%)"),   # stats and format for categorical columns
    digits = all_categorical() ~ 2,                              # rounding for continuous columns
    type   = all_categorical() ~ "categorical",  # force all categorical levels to display
    label  = list(                    # display labels for column names
      prov   ~ "province",                           
      age    ~ "age in years",
      sex ~ "sex",
      ethnic_group ~ "ethnicity",
      tetanus ~ "tetanus immunity level",
      diphtheria    ~ "diphtheria immunity level"),
    missing_text = "missing"          # how missing values should display
  ) %>% 
   modify_caption("**Table S1. Participant characteristics and serology results - by province**")


```


```{r median age participants}

# median age not part of table
median(data$age) # 15

```


```{r vaccination records present}
# one of the reviewer asked about the vaccination records; very few participants had any vaccination record

data %>% count(YC_present, other_rec_present) # 16 participants had vaccination records
data %>% count(other_rec_specify_tet)

```


```{r table s1 anti-diphtheria serology}

data %>%
  mutate(prov = ifelse(prov == "vtn", "vientiane capital", "bolikhamxay")) %>%
  mutate(prov = factor(prov, levels=c("vientiane capital", "bolikhamxay"))) %>% 

  dplyr::select(prov, 
         age, 
         sex, 
         ethnic_group,
         #tetanus,
         diphtheria) %>% # keep only columns of interest
  tbl_summary(     
    by = diphtheria,     # stratify entire table by outcome
    statistic = list(all_continuous() ~ "{mean} ({sd})",  # stats and format for continuous columns
                     all_categorical() ~ "{n} / {N} ({p}%)"),   # stats and format for categorical columns
    digits = all_categorical() ~ 1,                              # rounding for continuous columns
    type   = all_categorical() ~ "categorical",  # force all categorical levels to display
    percent = "row",
    label  = list(                    # display labels for column names
      prov   ~ "province",                           
      age    ~ "age in years",
      sex ~ "sex",
      ethnic_group ~ "ethnicity",
      #tetanus ~ "tetanus immunity level",
      diphtheria    ~ "diphtheria immunity level"),
    missing_text = "missing"          # how missing values should display
  ) %>% 
   modify_caption("**Table S1. Participant characteristics and anti-diphtheria serology results**") %>% 
  as_hux_table() %>% 
  huxtable::quick_xlsx(file ="./tables/tab_s1_d_serology_by-all-variables-diph.xlsx",
                       borders = 0.4,
                       open = interactive())

# total for diphtheria
# data %>% group_by(diphtheria) %>% (dplyr::summarise(p = 100 * n() / nrow(data)))

data %>% count(diphtheria)
40/779
538/779
173/779
28/779

```


```{r figure s1 diph serology by age and sex}

data %>%
  group_by(sex, age, diphtheria) %>% 
  summarize(freq = n()) %>% 
  mutate(prop = freq / sum(freq)) %>% 
  ggplot(aes(x=factor(age), 
             y=prop, 
             fill = forcats::fct_rev(diphtheria))) + 
  geom_col(colour="black")  +
  facet_wrap(~ sex) +
  xlab("age (years)") + 
  ylab ("proportion") +
  scale_fill_manual(name = "Interpretation", 
                    values = c("#481567FF",
                               "#39568CFF",
                               "#73D055FF",
                               "#FDE725FF"),
                    labels = c("long-term protection",
                               "protection present",
                               "uncertain protection",
                               "no protection")) +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))+
  ggtitle("anti-diphtheria IgG seroprevalence")  +
  theme_bw()

```


```{r figure 1 - diphtheria}

############## FIGURE 1

#  INCLUDED IN MANUSCRIPT

# some rows will not have an entry after aggregating the percentages of protected
# students per age group 
# I add 0% manually to increase readability of graph
data.frame(hcf = c("ph", "ph", "ph", "ph"),
           age = c(11,11,12,12), 
           sex = c("male","female","male","female"),
           prop = c(0.0,0.0,0.0,0.0)) -> add_rows_d


data %>% 
  dplyr::select(hcf, age, sex, d_interpretation) %>% 
  dplyr::group_by(hcf, age, sex, d_interpretation) %>% 
  summarize(freq = n()) %>% 
  mutate(prop = freq / sum(freq)) %>% 
  filter(d_interpretation == "protected") %>% 
  dplyr::select(hcf, age, sex, prop) %>% 
  ungroup() %>% 
  rbind(add_rows_d) %>% 
  
  ggplot() +
  geom_line(aes(x=age, y=prop, color = sex), size=1.2) + 
  ylim(0,1) +
  xlim(11,18) +
  facet_wrap(~hcf) +
  theme_bw() +
  scale_color_manual(values=c("#440154FF", "#3CBB75FF")) +
  
  ggtitle("proportion of participants with immunisation protection \nagainst diphtheria") +
  ylab("proportion immunisation protection \nagainst diphtheria") +
  xlab("age (years)") -> figure_1_diphtheria

figure_1_diphtheria

```


```{r figure s2 anti-diphtheria titers}

data %>% 
  filter(sex == "male") %>% 
  ggplot(aes(age, d_conc, color=d_outside_range)) + 
  geom_jitter() + 
  geom_smooth(method = "lm", color = "black") +
  labs(colour = "outside range")+
  scale_color_manual(values = c("blue4","orangered")) +
  theme_bw() +
  ylab("anti-diphtheria titer (IU/ml)") +
  xlab("age (years)") +
  geom_hline(yintercept=0.1, color = "red", size=0.8) +
  geom_hline(yintercept = 2, color = "grey16", linetype = "dashed") +
  facet_zoom(ylim = c(0, 0.75)) -> A

data %>% 
  filter(sex == "female") %>% 
  ggplot(aes(age, d_conc, color=d_outside_range)) + 
  geom_jitter() + 
  geom_smooth(method = "lm", color = "black") +
  labs(colour = "outside range")+
  scale_color_manual(values = c("blue4","orangered")) +
  theme_bw() +
  ylab("anti-diphtheria titer (IU/ml)") +
  xlab("age (years)") +
  geom_hline(yintercept=0.1, color = "red", size=0.8) +
  geom_hline(yintercept = 2, color = "grey16", linetype = "dashed") +
  facet_zoom(ylim = c(0, 0.75)) -> B

ggarrange(A, B, ncol = 1, nrow = 2,labels = c("A", "B")) -> figure_S2

figure_S2

```


```{r spearman correlation diphtheria titer}

data %>% group_by(sex) %>% shapiro_test(d_conc) 
data %>% group_by(sex) %>% rstatix::cor_test(age, d_conc, method = "spearman")

```


```{r table s2 bivariate anti-diphtheria}

tbl=table(data$prov, data$d_interpretation)
tbl
chisq.test(tbl) # p-value = 0.0002128
epitab(tbl)

tbl=table(data$district, data$d_interpretation)
tbl
chisq.test(tbl) # p-value = 0.00007027
epitab(tbl)

tbl=table(data$sex, data$d_interpretation)
tbl
chisq.test(tbl) # p-value = 0.00000001516
epitab(tbl)

tbl=table(data$age_group, data$d_interpretation)
tbl
chisq.test(tbl)  # p-value = 0.03484
epitab(tbl)

tbl=table(data$hcf, data$d_interpretation)
tbl
chisq.test(tbl) # p-value = 0.00001344
epitab(tbl)

tbl=table(data$ethnic_group, data$d_interpretation)
tbl
chisq.test(tbl) # p-value = 0.912

tbl=table(data$n_household_members, data$d_interpretation)
tbl
chisq.test(tbl) # 0.6243

tbl=table(data$vaccinations_school, data$d_interpretation)
tbl
chisq.test(tbl) # p-value = 0.5385


################################################ Note regarding variable selection:
# all variables with p<0.2 will be selected for logistic regression modelling
# province, district, age, sex, type of nearest HCF 
# province and district are essentially the same variable; and both of them overlap with type of nearest HCF

```


```{r figure s3 correlation hcf and district}

# The type of nearest HCF will overlap a lot with the variables for district / province
# Here is a visualization of this
# Central hospitals are exclusively in Vientiane Capital
# There is only 1 provincial hospital in this study and that is in Paksan (Bolikhamxay)

data %>%
  mutate(district = ifelse(district == "vientiane", "vientiane capital",
                      ifelse(district == "paksan", "bolikhamxay - paksan", 
                             "bolikhamxay - pakkading"))) %>% 
  mutate(district = factor(district, 
                             levels=c("vientiane capital", 
                                      "bolikhamxay - paksan", 
                                      "bolikhamxay - pakkading"))) %>% 
  group_by(district, hcf) %>% 
  summarize(freq = n()) %>% 
  mutate(prop = freq / sum(freq)) %>% 
  ggplot(aes(x=factor(district), 
             y=prop, 
             fill = hcf)) + 
  geom_col(colour="black")  +
  xlab("location") + 
  ylab ("proportion") +
  scale_fill_manual(name = "HCF", 
                    values = c("#481567FF", "#39568CFF",
                               "#20A387FF", "#73D055FF")) +
  ggtitle("proportion of HCFs according to study location")  +
  theme_bw() -> figure_S3
figure_S3

```


```{r multivariable models - diphtheria}

# What are the variables from bivariate analysis
## age group 
## sex
## district / province
## HCF


############## collinearity for district / province / nearest_HCF


# first check for district and nearest-HCF
mod <- glm(d_interpretation ~ age_group
           + sex
           + district
           + hcf, data = data, family=binomial())
summary(mod) 

vif(mod) # not ok

vcov(mod)
m <- vcov(mod)
cov2cor(m)
mcor <- cov2cor(m)
mcor                 # overlap HCF, district


# first check for province and nearest-HCF
mod <- glm(d_interpretation ~ age_group
           + sex
           + prov
           + hcf, data = data, family=binomial())
summary(mod) 

vif(mod) # ok

vcov(mod)
m <- vcov(mod)
cov2cor(m)
mcor <- cov2cor(m)
mcor                 # overlap HCF, province
# Model should only contain one of the variables



############## Model 1

mod1 <- glm(d_interpretation ~ age_group
           + sex
           + district, data = data, family=binomial())
summary(mod1) 

vif(mod1) # ok

vcov(mod1)
m <- vcov(mod1)
cov2cor(m)
mcor <- cov2cor(m)
mcor

step <- stepAIC(mod1)
summary(step)
step$anova          




############## Model 2

mod2 <- glm(d_interpretation ~ age_group
           + sex
           + prov, data = data, family=binomial())
summary(mod2) 

vif(mod2) # ok

vcov(mod2)
m <- vcov(mod2)
cov2cor(m)
mcor <- cov2cor(m)
mcor


step <- stepAIC(mod2)
summary(step)
step$anova           




############## Model 3

mod3 <- glm(d_interpretation ~ age_group
           + sex
           + hcf, data = data, family=binomial())
summary(mod3) 

vif(mod3) # ok

vcov(mod3)
m <- vcov(mod3)
cov2cor(m)
mcor <- cov2cor(m)
mcor


step <- stepAIC(mod3)
summary(step)
step$anova           


############## Model comparison

AIC(mod1) # [1] 836.55
AIC(mod2) # [1] 839.43
AIC(mod3) # [1] 834.75
# Model 3 has the lowest AIC; but there is not much difference between the models



############## Check for interactions

mod <- glm(d_interpretation ~ (age_group + sex + hcf)^2, data = data, family=binomial())
summary(mod) # yes, there are interactions
# given the sample size, only the interaction between age and sex will be included in model


############## TABLE S3
############## Model 3 with interaction
# Now include interaction between age and sex

mod <- glm(d_interpretation ~ age_group*sex + hcf, data = data, family=binomial())
summary(mod) # 817.94

cbind(round(coef(summary(mod)),4), OR=exp(mod$coefficients), exp(confint(mod))) 

tab_model(mod, show.aic = T) # wrong estimates


```


```{r table S4 final model anti-diphtheria data - before post-hoc tests}

# in the code chunk before, I looked at model options
# I selected the final model
# now we extract the model information

######################## final model - age is categorical
mod <- glm(d_interpretation ~ age_group*sex + hcf, data = data, family=binomial())
summary(mod) # 817.94
# these estimates from the model are not correct
cbind(round(coef(summary(mod)),4), OR=exp(mod$coefficients), exp(confint(mod)))

######################## model comparisons
null <- glm(d_interpretation ~ 1, data = data, family=binomial())
anova(mod, null, test="Chisq") # significant
Anova(mod, type = "II", test.statistic = "Wald") # significant
plot(fitted(mod),rstandard(mod))



########################################################### table s4
# extract numbers from model
# but estimates for interaction are not correct yet 
# estimates for age and sex cannot be interpreted individually
sjPlot::tab_model(mod, show.aic = T, transform = NULL)
sjPlot::tab_model(mod, show.aic = T) # table with OR (wrong OR for variables in interaction)


```


```{r table 1 final model diphtheria and table s3}

######################## final model - age is categorical
mod <- glm(d_interpretation ~ age_group*sex + hcf, data = data, family=binomial())
summary(mod) # 817.94

########################################################### table 1

# true ORs for variables that interact with each other can be calculated by hand from the information we obtained above
# calculation done in excel; NOT part of the script

# alternative: use allEffects()

allEffects(mod) # only probabilities
allEffects(mod) %>% summary() # probabilities and intervals

# real log-odds 
res <- allEffects(mod)    # real log-odds can also be obtained by using allEffects()
res$`age_group:sex`$fit   # same as calculated by hand
# 1 -1.24526764  # age group <=13 y : male
# 2 -2.00991165  # age group 14-16 y : male
# 3 -1.79298298  # age group >16 y : male
# 4 -1.39976917  # age group <=13 y : female
# 5 -0.63056182  # age group 14-16 y : female
# 6  0.04604646  # age group >16 y : female

# exponentiate the log odds -> real odds
# divide by reference -> real odds ratio


######################## plot model
plot(allEffects(mod))
plot_model(mod, type = "int", terms = c("age_group", "sex")) 


######################## post hoc

emmeans_results <- emmeans(mod, ~ age_group * sex)
emmeans_results # emmean corresponds to my own calculations; corresponds to estimates above

emmeans_results_tib <- as_tibble(emmeans_results)
emmeans_results_tib %>% 
  mutate(true_odds = exp(emmean)) %>%   # gives me the true odds
  mutate(OR = true_odds/first(true_odds))

# 95% CI + OR
# gives all the different combinations possible
pairs(emmeans_results, reverse = T)
# we need to sort out those of interest
emmeans_results_tib <- as_tibble(pairs(emmeans_results, reverse = T))
emmeans_results_tib %>% 
  filter(contrast == "age group 2 male - age group 1 male" |
         contrast == "age group 3 male - age group 1 male" |
         contrast == "age group 1 female - age group 1 male" |
         contrast == "age group 2 female - age group 1 male" |
         contrast == "age group 3 female - age group 1 male") %>% 
  mutate(OR = exp(estimate)) %>%
  mutate(LCI = exp(estimate-(1.96*SE))) %>% 
  mutate(UCI = exp(estimate+(1.96*SE)))


# Plot interaction
emmip(mod, sex ~ age_group, CIs=TRUE, plotit=T)+theme_bw()


# nearest health care facility

emmeans_results <- emmeans(mod, ~ hcf)
emmeans_results 
emmeans_results_tib <- as_tibble(emmeans_results)
emmeans_results_tib %>% 
  mutate(true_odds = exp(emmean)) %>%   # gives me the true odds
  mutate(OR = true_odds/first(true_odds))


# 95% CI + OR
# gives all the different combinations possible
pairs(emmeans_results, reverse = T)
# we need to sort out those of interest
emmeans_results_tib <- as_tibble(pairs(emmeans_results, reverse = T))
emmeans_results_tib %>% 
  filter(contrast == "ph - ch" |
         contrast == "dh - ch" |
         contrast == "hc - ch" ) %>% 
  mutate(OR = exp(estimate)) %>%
  mutate(LCI = exp(estimate-(1.96*SE))) %>% 
  mutate(UCI = exp(estimate+(1.96*SE)))


```


```{r table s4 anti-tetatnus serology by characteristics}

data %>%
  mutate(prov = ifelse(prov == "vtn", "vientiane capital", "bolikhamxay")) %>%
  mutate(prov = factor(prov, levels=c("vientiane capital", "bolikhamxay"))) %>% 

  dplyr::select(prov, 
         age, 
         sex, 
         ethnic_group,
         tetanus) %>% # keep only columns of interest
  tbl_summary(     
    by = tetanus,     # stratify entire table by outcome
    statistic = list(all_continuous() ~ "{mean} ({sd})",  # stats and format for continuous columns
                     all_categorical() ~ "{n} / {N} ({p}%)"),   # stats and format for categorical columns
    digits = all_categorical() ~ 1,                              # rounding for continuous columns
    type   = all_categorical() ~ "categorical",  # force all categorical levels to display
    percent = "row",
    label  = list(                    # display labels for column names
      prov   ~ "province",                           
      age    ~ "age in years",
      sex ~ "sex",
      ethnic_group ~ "ethnicity",
      tetanus ~ "tetanus immunity level"),
    missing_text = "missing"          # how missing values should display
  ) %>% 
   modify_caption("**Table S1. Participant characteristics and serology results**") %>% 
  as_hux_table() %>% 
  huxtable::quick_xlsx(file ="./tables/table_serology_by-all-variables-tet.xlsx",
                       borders = 0.4,
                       open = interactive())

data %>% count(tetanus)
round(((431 / 779)*100),1)
round(((107 / 779)*100),1)
round(((80 / 779)*100),1)
round(((112 / 779)*100),1)
round(((49 / 779)*100),1)


```


```{r figure S4}

data %>%
  group_by(sex, age, tetanus) %>% 
  summarize(freq = n()) %>% 
  mutate(prop = freq / sum(freq)) %>% 
  ggplot(aes(x=factor(age), 
             y=prop, 
             fill = forcats::fct_rev(tetanus))) + 
  geom_col(colour="black")  +
  facet_wrap(~ sex) +
  xlab("age (years)") + 
  ylab ("proportion") +
  scale_fill_manual(name = "Interpretation", values = c("#481567FF",
                                                 "#39568CFF",
                                                 "#20A387FF", 
                                                 "#73D055FF",
                                                 "#FDE725FF")) +
    theme(axis.text.x = element_text(angle = 45, hjust = 1))+
  ggtitle("anti-tetanus IgG seroprevalence")  +
  theme_bw()

#ggsave("./graphs/graph_bar_tet.png", plot = last_plot(), dpi = 300, height = 10, width = 20, units = "cm")


```


```{r figure s5 anti-tetanus titer}

data %>% 
  filter(sex == "male") %>% 
  ggplot(aes(age, t_conc_2, color=t_outside_range)) + 
  geom_jitter() + 
  geom_smooth(method = "lm", color = "black") +
  labs(colour = "outside range")+
  scale_color_manual(values = c("blue4","orangered")) +
  theme_bw() +
  ylab("anti-tetanus titer (IU/ml)") +
  xlab("age (years)") +
  geom_hline(yintercept=0.5, color = "red", size=0.8) +
  geom_hline(yintercept = 5, color = "grey16", linetype = "dashed") +
  facet_zoom(ylim = c(0, 7.5)) -> A

data %>% 
  filter(sex == "female") %>% 
  ggplot(aes(age, t_conc_2, color=t_outside_range)) + 
  geom_jitter() + 
  geom_smooth(method = "lm", color = "black") +
  labs(colour = "outside range")+
  scale_color_manual(values = c("blue4","orangered")) +
  theme_bw() +
  ylab("anti-tetanus titer (IU/ml)") +
  xlab("age (years)") +
  geom_hline(yintercept=0.5, color = "red", size=0.8) +
  geom_hline(yintercept = 5, color = "grey16", linetype = "dashed") +
  facet_zoom(ylim = c(0, 7.5)) -> B

ggarrange(A, B, ncol = 1, nrow = 2,labels = c("A", "B")) -> figure_S5

figure_S5


```


```{r spearman correlation tetanus}

data %>% group_by(sex) %>% rstatix::cor_test(age, t_conc_2, method = "spearman")

```


```{r figure 1 - tetanus}

# data to be added
# there are some participants not protected at all; the "0" needs to be added; it is not part automatically in table
data.frame(hcf = c("ph", "ph", "ph", "ph", "ph", "ph"),
           age = c(11,11,12,13, 14, 14), 
           sex = c("male","female","female","female","male","female"),
           prop = c(0.0,0.0,0.0,0.0,0.0,0.0)) -> add_rows_t

data %>% 
  dplyr::select(hcf, age, sex, t_interpretation) %>% 
  dplyr::group_by(hcf, age, sex, t_interpretation) %>% 
  summarize(freq = n()) %>% 
  mutate(prop = freq / sum(freq)) %>% 
  filter(t_interpretation == "protected") %>% 
  dplyr::select(hcf, age, sex, prop) %>% 
  ungroup() %>% 
  rbind(add_rows_t) %>% 
  ggplot() +
  geom_line(aes(x=age, y=prop, color = sex), size=1.2) + 
  ylim(0,1) +
  xlim(11,18) +
  facet_wrap(~hcf) +
  theme_bw() +
  scale_color_manual(values=c("#440154FF", "#3CBB75FF")) +
  ggtitle("proportion of participants with sufficient protection against tetanus") +
  ylab("proportion sufficient protection \nagainst tetanus") +
  xlab("age (years)") -> figure_1_tetanus
  
figure_1_tetanus

```


```{r table s2 bivariate anti-tetanus}

tbl=table(data$prov, data$t_interpretation)
tbl
chisq.test(tbl) # p-value = 0.7025

tbl=table(data$district, data$t_interpretation)
tbl
chisq.test(tbl) # p-value = 0.238

tbl=table(data$sex, data$t_interpretation)
tbl
chisq.test(tbl) # p-value = 0.1187
epitab(tbl)

tbl=table(data$age_group, data$t_interpretation)
tbl
chisq.test(tbl)  # p-value = 0.0001856
epitab(tbl)

tbl=table(data$hcf, data$t_interpretation)
tbl
chisq.test(tbl) # p-value = 0.2475
epitab(tbl)

tbl=table(data$ethnic_group, data$t_interpretation)
tbl
chisq.test(tbl) # p-value = 0.6992
epitab(tbl)

tbl=table(data$n_household_members, data$t_interpretation)
tbl
chisq.test(tbl) # p-value = 0.7406
epitab(tbl)

tbl=table(data$vaccinations_school, data$t_interpretation)
tbl
chisq.test(tbl) # p-value = 0.2042
epitab(tbl)


```


```{r multivariable models - tetanus}

# in bivariate analysis, age and sex of the participants was significantly associated with the outcome

# there are several option for running multivariable analysis for the anti-tetanus data
# age can be used both as numeric variable and with age groups
# throughout the paper, we used age as categorical variable
# for consistency; age will be used as categorical variable


# age as factor
mod1 <- glm(t_interpretation ~ age_group + sex, data = data, family=binomial())
summary(mod1)

vif(mod1) # ok

step <- stepAIC(mod1)
summary(step)
step$anova # t_int_2 ~ age_group + sex

# Check for interaction
# Interaction?
modi <- glm(t_interpretation ~ (age_group + sex)^2, data = data, family=binomial())
summary(modi) # yes, there is an interaction




# age as factor and interaction
mod2 <- glm(t_interpretation ~ age_group * sex, data = data, family=binomial())
summary(mod2)

vif(mod2) # interaction is over 10; due to interaction

step <- stepAIC(mod2)
summary(step)
step$anova # t_int_2 ~ age_group * sex


# model with interaction 
# wrong estimates
tab_model(mod2, show.aic = TRUE)
tab_model(mod2, show.aic = TRUE, transform = NULL)

```


```{r table S5 final model anti-tetanus data - before post-hoc tests}

# in the code chunk before, I looked at model options
# I selected the final model
# now we extract the model information for Table 1



######################## final model - age is categorical
mod <- glm(t_interpretation ~ age_group * sex, data = data, family=binomial())
summary(mod)
# calculate OR and CIs
# estimates are not correct because there is an interaction
cbind(round(coef(summary(mod)),4), OR=exp(mod$coefficients), exp(confint(mod)))

######################## model comparisons
null <- glm(t_interpretation ~ 1, data = data, family=binomial())
anova(mod, null, test="Chisq") # significant
Anova(mod, type = "II", test.statistic = "Wald") # significant
plot(fitted(mod),rstandard(mod))

########################################################### table s6
# extract numbers from model
# but estimates for interaction are not correct yet 
# estimates for age and sex cannot be interpreted individually
tab_model(mod, show.aic = T, transform = NULL)
tab_model(mod, show.aic = T) # Table with OR (wrong OR)

```


```{r table 1 final model tetanus}

mod <- glm(t_interpretation ~ age_group * sex, data = data, family=binomial())
summary(mod)


########################################################### table 1

# true ORs for variables that interact with each other can be calculated by hand from the information we obtained above
# calculation done in excel; NOT part of the script

# alternative: use allEffects()

allEffects(mod) # only probabilities
allEffects(mod) %>% summary() # probabilities and intervals

# real log odds
res <- allEffects(mod)
res$`age_group:sex`$fit  # same as calculated
# 1 -0.5596158 # Age group ≤13 years : Male
# 2 -1.2286654 # Age group 14-16 years : Male
# 3 -0.9480394 # Age group >16 years : Male
# 4 -1.5998685 # Age group ≤13 years : Female
# 5 -0.7731899 # Age group 14-16 years : Female
# 6  0.2933478 # Age group >16 years : Female

# exponentiate the log odds -> real odds
# divide by reference -> real odds ratio



######################## plot model
plot(allEffects(mod))
plot_model(mod, type = "int",terms = c("age_group", "sex")) 


######################## post hoc

emmeans_results <- emmeans(mod, ~ age_group * sex)
emmeans_results # emmean corresponds to my own calculations; corresponds to estimates above

emmeans_results_tib <- as_tibble(emmeans_results)
emmeans_results_tib %>% 
  mutate(true_odds = exp(emmean)) %>%   # gives me the true odds
  mutate(OR = true_odds/first(true_odds))

# 95% CI + OR
# gives all the different combinations possible
pairs(emmeans_results, reverse = T)
# we need to sort out those of interest
emmeans_results_tib <- as_tibble(pairs(emmeans_results, reverse = T))
emmeans_results_tib %>% 
  filter(contrast == "age group 2 male - age group 1 male" |
         contrast == "age group 3 male - age group 1 male" |
         contrast == "age group 1 female - age group 1 male" |
         contrast == "age group 2 female - age group 1 male" |
         contrast == "age group 3 female - age group 1 male") %>% 
  mutate(OR = exp(estimate)) %>%
  mutate(LCI = exp(estimate-(1.96*SE))) %>% 
  mutate(UCI = exp(estimate+(1.96*SE)))


# Plot interaction
emmip(mod, sex ~ age_group, CIs=TRUE, plotit=T)+theme_bw()

```