MarkdownExample.Rmd

---
title: "Crypto Outbreak Analysis"
author: "Investigation team"
date: "28 September 2018"
output: word_document
---

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

# Background 

Cryptosporidium is a protozoal parasite that causes a diarrhoeal illness in humans known as cryptosporidiosis. It is transmitted by the faeco-oral route, with both animals and humans serving as potential reservoirs. Cryptosporidiosis is a notifiable disease in many countries across the world. 
This report investigated cryptosporidium incidence in 2015 from surveillance data in Country X. 

# Methods 

## Data aquisition and structure 

Routinely reported national surveillance data was used for analysis. This covered the years from 2004-2015 for the whole of Country X. 

The data was case-based and thus one record designated one case of Cryptosporidium. Information on species (e.g. *C. parvum*, *C. hominis*, etc.) was not available in this dataset in country X. Due to funding restrictions, routine speciation of samples was stopped in most laboratories in country X. 

Denominator data was sourced from the office of national statistics in Country X. These population counts were available from the year 2011 and broken down by region and age group. 

## Data analysis 

Using poisson regression, we calculated incidence rate ratios with corresponding 95% confidence intervals and p-values for three analyses. First comparing all years available using 2015 as a reference. Second, between the average of 2012-2014 and 2015. Finally, between urban and rural areas for 2015. 

# Results 



```{r, warning= FALSE, message= FALSE}
#load "knitr": for styling tables
library(knitr)

#load "broom": for simplifying model output
library(broom)
```


Poisson regression showed that several years had significantly different incidence rates for Cryptosporidium when compared with 2015 in Country X (Table 1). 


**Table 1:** Incidence rate ratios for Cryptosporidium in Country X, by year, using 2015 as the reference

```{r}
#### Reading in datasets ####

#load your cleaned dataset
load("cryptoyear.Rda")


#### Incidence rate ratios of all years compared to 2015 data ### 

# change 2015 to reference group using relevel function
cryptoyear$year <- relevel(factor(cryptoyear$year), ref = "2015")

# run poisson regression of counts by year
model1 <- glm(count ~ year , 
              family = poisson(link = "log"), 
              data = cryptoyear, offset = log(den_reg))

# use the tidy function from broom package to simplify the regression output
model1clean <- tidy(model1, exponentiate = TRUE, conf.int = TRUE)


#clean output table of model
kable(model1clean)

```


Further analysis showed that there was no significant difference in incidence rate for Cryptosporidium when comparing 2015 to the mean of years 2012 to 2014 (Table 2). 



**Table 2:** Incidence rate ratios for Cryptosporidium in Country X, comparing 2015 to the mean of years 2012-2014
```{r, warning= FALSE}

#### Incidence rate ratio of 2015 compared to 2012-14 data ####

# collapse your data producing mean counts for 2012-2014 

# create a variable for current year 
cryptoyear$curryear <- NA
cryptoyear$curryear[cryptoyear$year == 2015] <- 1
cryptoyear$curryear[cryptoyear$year %in% c(2012:2014)] <- 0

# aggregate using the mean function
meanyear <- aggregate(count ~ curryear + den_reg, 
                      FUN = mean, 
                      data = cryptoyear)

# run poisson regression of counts by year
model2 <- glm(count ~ curryear, 
              family = poisson(link = "log"), 
              data = meanyear, offset = log(den_reg))

# use the tidy function from broom package to simplify the regression output
model2clean <- tidy(model2, exponentiate = TRUE, conf.int = TRUE)

#clean output table of model
kable(model2clean)
```



The final analysis conducted, compared urban with rural areas and found no significant difference in incidence rates for 2015 (Table 3). 

**Table 3**: Incidence rate ratio of Cryptosporidium for urban compared to rural areas in Country X, 2015
```{r}
#### Incidence rate ratio: urban vs rural areas #### 

# load your cleaned dataset
  # note that it is still called crypto
load("cryptorecodeddenom.Rda")


# only keep 2015 counts
crypto <- subset(crypto, 
                 year == 2015)

# aggregate using the sum function
cryptoag <- aggregate(count ~ urban + den_reg, 
                      FUN = sum, 
                      data = crypto)

# sum rural seperately 
  # aggregate doesnt work because only one urban row
cryptorural <- colSums(cryptoag[1:6, 2:3])

# bind sums together 
cryptosum <- rbind(cryptorural, cryptoag[7, 2:3])

# change urban to binary
cryptosum$urban <- c(0,1)


# run poisson regression of counts by year
model3 <- glm(count ~ urban, 
              family = poisson(link = "log"), 
              data = cryptosum, offset = log(den_reg))

# use the tidy function from broom package to simplify the regression output
model3clean <- tidy(model3, exponentiate = TRUE, conf.int = TRUE)


#clean output table of model
kable(model3clean)

```