R4SME 15 strategies.Rmd

---
title: "R for SME 15: Strategies of analysis"
author: [Andrea Mazzella](https://github.com/andreamazzella)
output: html_document
---

```{r Load packages}
library(haven)
library(pubh)
library(epiDisplay)
library(magrittr)
library(tidyverse)
```

## Data import and exploration

Make sure you have the mortality.dta dataset in the same folder as this .rmd file.
```{r Import}
mortality <- read_dta("./mortality.dta")
mortality %<>% mutate_if(is.labelled,as_factor)
```

Explore the data.
- What study design?
- How many individuals? 
```{r Explore}
glimpse(mortality)
summary(mortality)
View(mortality)
```

## Data management

- How many variables?
- Which variables code for the main exposure, outcome?
- Which are your fixed confounders?
- What are the other potential confounders?
- How are these variables coded?
- Which variables do we need to recode, rename, categorise?

```{r Data management}
# died -> death (factor)
mortality %<>%
  mutate(death = factor(died,
                        levels = c(0, 1),
                        labels = c("No", "Yes")))
mortality %$% table(death, died)

# District -> factor
mortality$district <- as.factor(mortality$district)
mortality %$% table(district)

# Compound size
mortality %<>% mutate(compound_grp = cut(
  compound_size,
  breaks = c(0, 9, 19, +Inf),
  labels = c("1-9", "10-19", "20-55")
))
mortality %$% table(compound_size, compound_grp)
mortality %$% tabpct(compound_grp, death, percent = "row", graph = F)
mortality %$% tabpct(compound_grp, vimp, percent = "row", graph = F)

# Ethnic: simplify by recoding -> ethnic2
mortality %$% table(ethnic)

mortality %<>%
  mutate(
    ethnic2 = recode(
      ethnic,
      "Fulani" = "Other",
      "Gwari" = "Other",
      "Kiwollo" = "Other",
      " Surubu" = "Other"
    )
  )
mortality %$% table(ethnic, ethnic2)
mortality %$% table(ethnic2)


# Occupation: simplify by recoding -> occ_sector
mortality %$% table(occupation)
mortality %$% tabpct(occupation, death, percent = "row", graph = F)
mortality %$% tabpct(occupation, vimp, percent = "row", graph = F)

mortality %<>%
  mutate(
    occ_sector = recode(
      occupation,
      "Farmer" = "Agriculture",
      "Fishing" = "Agriculture",
      "Hunting" = "Agriculture",
      "Artisan" = "Other",
      "Trader" = "Service",
      "Homemaker" = "Service",
      "Priest/Imam" = "Other",
      "Teacher/Civil servant, etc." = "Other",
      "Traditional healer" = "Other",
      "Child below school age" = "Other",
      "Unemployed" = "Other",
      "Student" = "Other"
    )
  )
mortality %$% tabpct(occ_sector, death, percent = "row", graph = F)

# Education
mortality %$% table(education)
mortality %$% tabpct(education, death, percent = "row", graph = F)
mortality %$% tabpct(education, vimp, percent = "row", graph = F)

mortality %<>%
  mutate(
    education_grp = recode(
      education,
      "Koranic education only" = "Adult/Koranic education",
      "Adult education only" = "Adult/Koranic education",
      "Primary" = "Formal education",
      "Secondary" = "Formal education",
      "Post secondary" = "Formal education"
    )
  )
mortality %$% table(education, education_grp)
mortality %$% table(education_grp)
mortality %$% tabpct(education_grp, death, percent = "row", graph = F)
mortality %$% tabpct(education_grp, vimp, percent = "row", graph = F)

# Systolic
mortality %<>% mutate(systolic_grp = cut(
  systolic,
  breaks = c(0, 119, 139, +Inf),
  labels = c("normal", "pre-hypertension", "hypertension")
))
mortality %$% table(systolic, systolic_grp)
class(mortality$systolic_grp)

# HR
mortality %<>% mutate(tachycardia = cut(
  pulse,
  breaks = c(0, 99, +Inf),
  labels = c("normal", "tachycardia")
))
mortality %$% table(pulse, tachycardia)
mortality %$% tabpct(tachycardia, death, percent = "row", graph = F)
mortality %$% tabpct(tachycardia, vimp, percent = "row", graph = F)
```


## Descriptive analysis

NB: an alternative way to create a quick Table 1 is tableone::CreateTableOne() - check it out! (Thanks, Julian!)
### Describe the sample in terms of outcome and exposure

- What's the incidence risk of death?
- What's the prevalence of visual impairment?
```{r}
mortality %$% tab1(death, cum.percent = F, graph = F)

mortality %$% tab1(vimp, cum.percent = F, graph = F)
```

### Describe the sample in terms of the age and sex

```{r}
# Age
mortality %$% tab1(agegrp, cum.percent = F, graph = F)

# Sex
mortality %$% tab1(sex, cum.percent = F, graph = F)

# Age and sex
mortality %>% ggplot(aes(agegrp)) +
  geom_bar(aes(fill = sex)) +
  scale_fill_viridis_d(option = "D") +
  labs(title = "Age and sex distribution", x = "Age", y = "frequency")
```

### Describe the sample in terms of geography

```{r}
# Area
mortality %$% tab1(area, cum.percent = F, graph = F)

# District
mortality %$% tab1(district, cum.percent = F, graph = F)

# Village ID number
mortality %$% tab1(vcode, cum.percent = F, graph = F) # useless, too many villages
```

### Describe sample in terms of socio-economic status

```{r}
# Compound size
print("Compound size")
mortality %$% summary(compound_size)
mortality %>% ggplot(aes(x = compound_size)) +
  geom_histogram()

# Ethnic origin
mortality %$% tab1(ethnic2, cum.percent = F, graph = F)

# Religion
mortality %$% tab1(religion, cum.percent = F, graph = F)

# Occupation
mortality %$% tab1(occ_sector, cum.percent = F, graph = F)

# Education
mortality %$% tab1(education_grp, cum.percent = F, graph = F)
```

### Describe sample in terms of medical status

```{r}
# Systolic BP
mortality %>% ggplot(aes(x = systolic)) +
  geom_histogram()

mortality %$% tab1(systolic_grp, cum.percent = F, graph = F)

# Diastolic BP
mortality %>% ggplot(aes(x = diastolic)) +
  geom_histogram()

# Mean arterial BP
mortality %>% ggplot(aes(x = map)) +
  geom_histogram()

# HR
mortality %>% ggplot(aes(x = pulse)) +
  geom_histogram()

mortality %$% tab1(tachycardia, cum.percent = F, graph = F)

# Weight
mortality %>% ggplot(aes(x = weight)) +
  geom_histogram()
print("Weight")
print("mean")
mortality %$% mean(weight, na.rm = T)
print("sd")
mortality %$% sd(weight, na.rm = T)

# Height
mortality %>% ggplot(aes(x = height)) +
  geom_histogram()
print("Height")
print("mean")
mortality %$% mean(height, na.rm = T)
print("sd")
mortality %$% sd(height, na.rm = T)

# BMI
mortality %>% ggplot(aes(x = bmi)) +
  geom_histogram()
mortality %$% tab1(bmigrp, cum.percent = F, graph = F)

# Microfilaria
mortality %>% ggplot(aes(x = mfpermg)) +
  geom_histogram()
mortality %$% tab1(mfpos, cum.percent = F, graph = F)
mortality %$% tab1(mfgrp, cum.percent = F, graph = F)
```

## Crude analysis

- What is the crude OR for death in the visually impaired?
```{r}
# With chi-squared test
mortality %$% cc(death, vimp, graph = F)

# With logistic regression
glm(death ~ vimp,
    data = mortality,
    family = binomial()) %>% logistic.display()
```

## Stratified analysis - MH method

```{r}
# Age
mortality %$% epiDisplay::mhor(death, vimp, agegrp, graph = F)

# Sex
mortality %$% epiDisplay::mhor(death, vimp, sex, graph = F)
```

```{r}
# Area
mortality %$% epiDisplay::mhor(death, vimp, area, graph = F)

# District
mortality %$% epiDisplay::mhor(death, vimp, district, graph = F)

# Village number
mortality %$% epiDisplay::mhor(death, vimp, vcode, graph = F)

```

```{r}
# Compound size
mortality %$% epiDisplay::mhor(death, vimp, compound_grp, graph = F)

# Ethnic origin
mortality %$% epiDisplay::mhor(death, vimp, ethnic2, graph = F)

# Religion
mortality %$% epiDisplay::mhor(death, vimp, religion, graph = F)

# Occupation
mortality %$% epiDisplay::mhor(death, vimp, occ_sector, graph = F)

# Education
mortality %$% epiDisplay::mhor(death, vimp, education_grp, graph = F)
```

```{r}
# Systolic BP
mortality %$% epiDisplay::mhor(death, vimp, systolic_grp, graph = F)

# Diastolic BP
# Mean arterial BP

# HR
mortality %$% epiDisplay::mhor(death, vimp, tachycardia, graph = F)

# Weight
# Height

# BMI
mortality %$% epiDisplay::mhor(death, vimp, bmigrp, graph = F)

# Microfilaria
mortality %$% epiDisplay::mhor(death, vimp, mfpos, graph = F)
mortality %$% epiDisplay::mhor(death, vimp, mfgrp, graph = F)
```


## Stratified analysis - logistic regression

A possible model:
```{r}
# Age, hypertension, BMI
logit_without <- glm(death ~ vimp + agegrp + systolic_grp + bmigrp,
    data = mortality,
    family = binomial())
logistic.display(logit_without)
```
- How do you interpret these results?

## Analysis for interaction - logistic regression

```{r}
# Interaction between vimp and age
logit_with_age <- glm(death ~ vimp * agegrp + systolic_grp + bmigrp,
    data = mortality,
    family = binomial())
logistic.display(logit_with_age)

lrtest(logit_with_age, logit_without)
```

```{r}
# Interaction between vimp and systolic
logit_with_sys <- glm(death ~ vimp * systolic_grp + agegrp + bmigrp,
                      data = mortality,
                      family = binomial())
logistic.display(logit_with_sys)

lrtest(logit_with_sys, logit_without)
```

```{r}
# Interaction between vimp and BMI
logit_with_BMI <- glm(death ~ vimp * bmigrp + systolic_grp + agegrp,
    data = mortality,
    family = binomial())
logistic.display(logit_with_BMI)
lrtest(logit_with_BMI, logit_without)
```