04-causes_risks.qmd

# Causes and Risks {#sec-04-causes_risks}

```{r}
#| echo: false
library(ggplot2)
book_theme <- theme_minimal() + 
  theme(plot.title=element_text(face="bold"))
ggplot2::theme_set(book_theme)
```

:::::: solutionbox
:::: solutionbox-header
::: solutionbox-icon
:::

Learning Objectives
::::

::: solutionbox-body
-   Identify major causes of disease and associated risk factors using
    health data
-   Formulate clear and focused research questions for health outcomes
    analysis
-   Gain an introductory understanding of causal inference concepts and
    their application in public health research
:::
::::::

\

> **"...fear is the most pervasive emotion of modern
> society..."[@bourke2007]**

What qualifies as a risk\index{Risk} is subject to dynamic social change
[@li2020] , as well as the perception of risk has evolved over time,
influenced by factors such as media coverage and sociopolitical
dynamics.

Historically, major risks included starvation, infections, and violent
conflicts, while modern risks are often associated with lifestyle
choices and chronic diseases such as obesity\index{Obesity},
cardiovascular disease\index{Cardiovascular Disease (CVD)}, and cancer.
Despite advancements in healthcare and increasing life
expectancy\index{Life
expectancy} in post-industrial countries, the focus often shifts to
perceived threats like terrorism, global pandemics such as
COVID-19\index{COVID-19}, and environmental catastrophes. This shift is
reflected in the increasing combination of quantitative analyses and
public health interventions, tracking changes in risk-related discourse
and identifying key risk topics over time.

Furthermore, tools like topic modelling and sentiment analysis help
identify how the public perceives various risks and how these
perceptions evolve over time.

In the field of public health, the latest
GBD\index{Global Burden of Disease (GBD)} results reveal significant
insights into the causes\index{Causes} and risks associated with health
metrics\index{Health metrics} and infectious
diseases\index{Infectious diseases}. The primary risks identified
include behavioural\index{Behavioural},
environmental\index{Environmental}, occupational\index{Occupational},
and metabolic\index{Metabolic} factors.

```{r}
#| echo: false
#| eval: false
#| fig-cap: "FlowChart of Historical vs. Modern Risks"
#| fig-alt: "FlowChart of Historical vs. Modern Risks"

library(DiagrammeR)
library(DiagrammeRsvg)
library(rsvg)

graph <- grViz("
digraph risk_dev{
  
  # Define graph attributes
  graph [layout = dot, rankdir = TB]
  
  # Define nodes
  node [shape = box, style = filled, fillcolor = lightblue] 
  A [label = 'Historical Risks']
  B1 [label = 'Starvation']
  B2 [label = 'Infectious Diseases']
  B3 [label = 'Violent Conflicts']
  
  node [shape = box, style = filled, fillcolor = gold]
  C [label = 'Modern Risks']
  D1 [label = 'Chronic Diseases (Obesity, CVD, Cancer)']
  D2 [label = 'Perceived Threats (Terrorism, Pandemics, Environment)']
  
  node [shape = ellipse, style = filled, fillcolor = lightcoral]
  E [label = 'GBD Risk Factors']
  F1 [label = 'Behavioral']
  F2 [label = 'Environmental']
  F3 [label = 'Occupational']
  F4 [label = 'Metabolic']

  # Define edges
  A -> B1
  A -> B2
  A -> B3
  A -> C
  C -> D1
  C -> D2
  C -> E
  E -> F1
  E -> F2
  E -> F3
  E -> F4
}")

graph %>%
  export_svg %>% 
  charToRaw %>% 
  rsvg_png("images/04_risk-dev.png")
```

![FlowChart of Historical vs. Modern Risks: this flowchart illustrates
how risk perception has evolved over time, shifting from historical to
modern risks. It integrates key risk factors identified in the Global
Burden of Disease (GBD)
study.](images/04_risk-dev.png){#fig-risk-dev
fig-align="center" width="100%"
fig-alt="FlowChart of Historical vs. Modern Risks"}

## Conditions and Injuries

Conditions and injuries associated with the burden of disease and injury
vary according to specific causes and risks. In this book causes and
risk factors include:

-   **Lifestyle choices**\index{Lifestyle}: Poor diet, physical
    inactivity, tobacco use, and excessive alcohol consumption are major
    risk factors for many chronic diseases and injuries, including heart
    disease, stroke, cancer, and liver disease.
-   **Environmental factors**\index{Environmental factors}: Exposure to
    pollutants, such as air pollution and toxic chemicals, can increase
    the risk of certain diseases and injuries.
-   **Infections**\index{Infections}: Many diseases, such as
    tuberculosis, HIV/AIDS, and malaria, are caused by infectious
    agents.
-   **Poverty**\index{Poverty}: People living in poverty are often more
    susceptible to health problems due to limited access to healthcare,
    healthy food, and safe living conditions.
-   **Ageing**\index{Ageing | aging |}: As people get older, they are at
    an increased risk of many health problems, including chronic
    diseases and disabilities.
-   **Genetics**: Some diseases and injuries are caused by genetic
    factors, such as a genetic predisposition to certain cancers.
-   **Injuries**\index{Injuries}: Injuries, such as falls, road traffic
    accidents, and violence, can also contribute to the burden of
    diseases and injuries.

```{r}
#| echo: false
#| label: fig-daly_dag
#| fig-alt: "Causal relationships leading to fueling DALYs value. Weights (W), Years Lived with Disability (YLD), and Years of Life Lost (YLL) are the main components of the DALY metric."
#| fig-cap: "Causal relationships leading to fueling DALYs value. Weights (W), Years Lived with Disability (YLD), and Years of Life Lost (YLL) are the main components of the DALY metric."
#| fig-width: 5
#| fig-height: 3.2
library(ggdag)

dag <- dagify(
  "DALY" ~ "W" + "YLD" + "YLL", 
  "Cause" ~ "age" + "sex" + "place" + "race" + "other", 
  "W" ~ "Cause", 
  "YLD" ~ "W", 
  "YLL" ~ "W",
  coords = time_ordered_coords())

ggdag(dag) + theme_dag_grid()
```

A particular health condition can have multiple causes
(co-morbidities)\index{Comorbidity} and risk factors. For instance, a
poverty status and the lack of access to healthcare facilities, is
proven to be increasing the risk of infectious diseases, while poor diet
and physical inactivity can increase the risk of chronic diseases.
Acting in favour of addressing the *underlying causes and risk
factors*\index{Risk factors} for diseases and injuries is crucial for
prompt public health interventions and can help reduce the overall
burden of disease.

## Risk Measures

In health metrics, *risk* refers to the likelihood that an individual
will experience a specific health outcome, such as illness or injury,
due to certain behaviours, exposures, or conditions. Risk factors are
variables that increase the probability of developing a particular
health condition or experiencing an adverse health outcome; they are
measurable probabilities influenced by factors like lifestyle (e.g.,
smoking or diet), environmental exposures (e.g., air pollution), or
underlying health conditions.

To provide a comprehensive framework for assessing the burden of
different risk factors on population health and guide effective public
health strategies to mitigate these risks, key measures are used to
assess risks and their impact on health outcomes:

-   Risk-specific exposures\index{Risk-specific exposures}
-   Relative risks (RRs)\index{Relative Risks (RRs)}
-   Theoretical Minimum-Risk Exposure Levels
    (TMRELs)\index{Theoretical Minimum Risk Exposure Levels (TMRELs)}
-   Population Attributable Fractions
    (PAFs)\index{Population Attributable Fraction (PAF)}

### Risk-Specific Exposures

The quantification of risks and causes involves the evaluation of a set
of behavioural, environmental and occupational, and metabolic risks.
Pairs of risk-outcome\index{Risk-outcome} are investigated based on
observations and statistical evidence. Convincing evidence consists of
plausible associations between exposure and disease in terms of size,
duration and effects. Common examples of risk exposures in health
metrics include: smoking, physical inactivity, high blood pressure
(hypertension), and others.

Risk combinations can be **additive** (the occurrence of a least one
event, A or B), **multiplicative** (the occurrence of both of two
events, A and B) or just **interactive,** acting to influence other
pairs, this action is generally identified as possible
**confounding**\index{Confounding}, to be distinguished by factors in
the causal pathway between exposure and outcome.

To have an idea of the impact of different risk factors on a cause of
illness, the **Socio-demographic Index
(SDI)**\index{Socio Demographic Index (SDI)} provides insights into the
potential magnitude of social, cultural and demographic factors looking
at the risk exposures and possible paths for policy interventions. The
life expectancy level is closely correlated to the level of the SDI
indicator as it is based on average income per person, educational
attainment, and total fertility rate (TFR)\index{Total Fertility Rate
(TFR)}. Higher SDI values typically indicate better socio-economic
conditions, including improved access to healthcare, education, and
sanitation, which can mitigate various health risks. Conversely, lower
SDI values are associated with higher risk exposure due to limited
access to healthcare, poorer living conditions, and other socio-economic
challenges. An application of the SDI index on time series is on
@sec-08-predictions.

One more element to take into consideration is the **Comparative Risk
Assessment (CRA)**[@stanaway2018]\index{Comparative Risk
Assessment (CRA)} divided into attributable and avoidable burden.
Considering as the objective the potential reduction of future disease
burden, four types of **minimum risk
exposure**\index{Minimum Risk Exposure} distributions are identified:

-   Theoretical\index{Theoretical}
-   Plausible\index{Plausible}
-   Feasible\index{Feasible}
-   Cost-effective\index{Cost-effective}

The following provide a high level overview on quantifying attributable
burden by using the theoretical minimum risk.

### Relative Risks (RRs)

The relative risk (or Risk Ratio)\index{Relative Risk (RR) | Risk Ratio}
is a measure of the strength of the association between an exposure and
an outcome. It compares the likelihood of a particular health outcome,
occurring in individuals exposed to a specific risk factor, to the
likelihood in those who are not exposed. This metric helps quantify how
much a risk factor, like smoking or high blood pressure, increases the
probability of an adverse health effect.

To calculate the risk-outcome pairs, such as mortality and morbidity,
the attributable burden of a risk is decomposed to identify the impact
on disease burden across factors like location, age, sex, and specific
causes of disease[@murray2020]. This decomposition considers the
combined effect of all risk exposures, which are categorised into
metabolic, behavioural, and environmental risk factors. Each category
contributes distinctly to the overall health outcome, enabling a nuanced
understanding of how multiple exposures collectively influence disease
burden and health outcomes.

The relative risk is the ratio between the proportions of exposed and
unexposed groups.

$$
RR= \frac{p_0}{p_1}
$$ {#eq-rr}

where $p_1$ and $p_0$ are the proportions of exposed and unexposed
groups respectively. Or, in terms of population, these group would
approx the values of the real population:

$$
RR = \frac{p_1}{p_0}=\frac{d_1/n_1}{d_0/n_0}
$$ {#eq-rr2}

where $d_1/n_1$ and $d_0/n_0$ are the proportion of the population with
and without the disease.

For example, let's say we are studying the association between smoking
(exposure) and lung cancer\index{Lung cancer} (outcome). We want to
calculate the relative risk of lung cancer among smokers compared to
non-smokers. If the relative risk is 2, it means that smokers are twice
as likely to develop lung cancer compared to non-smokers.

```{r}
exposed <- c(50, 10)
unexposed <- c(20, 5)

# Calculate the relative risk
relative_risk <- function(exposed, unexposed) {
  (exposed[1] / sum(exposed)) / (unexposed[1] / sum(unexposed))
  }

relative_risk(exposed, unexposed)
```

In this case, a relative risk of 1.04 indicates that the exposed group
is 1.04 times more likely to develop the outcome compared to the
unexposed group.

A second example is to calculate the relative risk based on the number
of events and person-time at risk for exposed and unexposed groups.
Let's consider the following scenario:

```{r}
d1 <- 50 # Number of events in the exposed group
n1 <- 10 # Person-time at risk in the exposed group
d0 <- 20 # Number of events in the unexposed group
n0 <- 5 # Person-time at risk in the unexposed group

# Calculate the relative risk
relative_risk_d <- (d1 / n1) / (d0 / n0)

# Print the relative risk
relative_risk_d
```

In this case, the relative risk based on the number of events and
person-time at risk is 1.25, indicating that the exposed group has a
1.25 times higher risk of developing the outcome compared to the
unexposed group.

```{r}
#| label: fig-barplot_relative_risk
#| fig-alt: "Bar chart showing the relative risk of outcome between exposed and unexposed groups."
#| fig-cap: "Bar chart showing the relative risk of outcome between exposed and unexposed groups. The relative risk is calculated based on the number of events and person-time at risk for each group."
#| echo: false

data <- data.frame(
  Group = c("Exposed", "Unexposed"),
  Event_Rate = c(d1 / n1, d0 / n0))

# Plot the bar chart
ggplot(data, aes(x = Group, y = Event_Rate, 
                 fill = Group)) +
  geom_bar(stat = "identity", width = 0.6) +
  labs(title = "Relative Risk Differences",
       subtitle = "Exposed and Unexposed Groups",
       x = "Group",
       y = "Event Rate (Events per Person-Time)") +
  annotate("text", 
           x = 1.5, y = max(data$Event_Rate) * 0.8,
           label = paste("Relative Risk =", 
                         round(relative_risk_d, 2)), 
           size = 5, color = "black") +
  scale_fill_manual(values = c("Exposed" = "salmon", 
                               "Unexposed" = "skyblue")) 
```

In summary, the relative risk can be calculated using two different
formulas:

-   The first formula, $RR=p_1/p_0$, calculates the relative risk
    directly using the proportions of events $p_1$ in the exposed group
    compared to the unexposed group $p_0$ . This formula provides a more
    simplified view of the relative risk based solely on event
    proportions.
-   The second formula, $RR=\frac{d_1/n_1}{d_0/n_0}$, considers both the
    **number of events** $(d1, d0)$ and **person-time at
    risk**\index{Person-time at risk} $(n1, n0)$ for each group. This
    formula takes into account the **incidence
    rate**\index{Incidence rate} in addition to **event
    proportions**\index{Event proportions}, providing a more specific
    understanding of the relative risk by incorporating information
    about the **duration of exposure**\index{Duration of exposure}.

### Relative Risks and Network Analysis

In some cases, relative risks can be modelled using **network
analysis**\index{Network analysis}, a specialised approach within
statistical modelling which extends the concept of mixed effects to
compare multiple treatments while accounting for various factors and
dependencies. The relationship between variables, represented by nodes
and edges, considers the potential interactions or dependencies between
different risk factors and outcomes. This approach is generally
favourable when exploring complex relationships among multiple
variables.

To represent the network we can use a **Directed Acyclic Graph
(DAG)**\index{Directed Acyclic Graph
(DAG)} for drawing causal relationships between variables, such as the
relationship between health risks and diseases.

This is an example of a **Network graph**\index{Network graph}
representing the causal pathways between different variables, such as
smoking, physical inactivity, high blood pressure, lung cancer, heart
disease, and stroke. By visualising the relationships between these
variables, we can identify the direct and indirect effects of risk
factors on health outcomes.

```{r}
#| echo: false
#| fig-width: 7
#| fig-height: 5
#| fig-align: "center"
#| label: fig-networkgraph1
#| fig-cap: "Directed Acyclic Graph (DAG) - Network Analysis of Risk Factor Relationships."
#| fig-alt: "Network graph to identify the relationship between risk factors and health outcomes."

library(igraph)

# Define the nodes (risk factors and health outcomes)
nodes <- c("Smoking", "Physical Inactivity", 
           "High Blood Pressure", 
           "Lung Cancer", "Heart Disease", "Stroke")

# Define the edges (relationships between risk factors and health outcomes)
edges <- c("Smoking", "Lung Cancer", 
           "Smoking", "Heart Disease", 
           "Physical Inactivity", "Heart Disease", 
           "High Blood Pressure", "Heart Disease", 
           "High Blood Pressure", "Stroke")
set.seed(444)
# Create the graph object
g <- graph(edges = edges, 
           directed = FALSE)

layout <- layout_with_fr(g)*2

# Plot the graph
plot(g,
     layout = layout,
     vertex.color = "skyblue",
     vertex.size = 40,
     vertex.label.color = "black",
     vertex.label.font = 2,
     vertex.label.family = "sans",
     vertex.label.cex = 0.5,
     vertex.frame.color = NA, 
     edge.width = 2.5,
     edge.color = "gray",
     xlim = c(-0.5, 1),
     ylim = c(-0.93, 1),
     main = "Network Analysis of Risk Factor Relationships"
)
```

The following code shows one more example of a network graph that would
be helpful to identify the relationship between outcome (O), exposure
(E) and different risk factors made with the `{ggdag}` package and the
`dagify()` function.

```{r}
#| echo: true
#| fig-width: 6
#| fig-height: 3.2
#| fig-align: "center"
#| label: fig-networkgraph2
#| fig-cap: "DAG between outcome (O), exposure (E) and different risk factors."
#| fig-alt: "Network graph to identify the relationship between outcome (O), exposure (E) and different risk factors."

# Load the library
library(ggdag)
set.seed(555)
# Define the DAG structure
dag <- dagify(
  y ~ x,
  x ~ c1 + c2 + c3,
  c1 ~ c2 + c3,
  c1 ~ c3)

# Plot the DAG
ggdag(dag) + theme_dag_grid()
```

#### Simulation of Risk Exposure

The simulation of the risk exposure\index{Risk exposure} can be done
replicating a logistic regression model with a DAG structure. We can use
the `{dagitty}` package, and the `simulateLogistic()` function. The
model estimates the probability of an outcome (O), given exposure (E) to
different risk factors, such as C1, C2, and C3 (confounders). The
relative risk is calculated based on the estimated probabilities of the
outcome given exposure and no exposure to the risk factors.

```{r}
#| echo: true
# Load necessary libraries
library(dagitty)
library(tidyverse)

# Create DAG structure
dag <- dagitty("dag { E -> O
                      C1 -> O
                      C2 -> O
                      C3 -> O }")

dat <- dag %>% tidy_dagitty()
# Generate data
set.seed(123)
n <- 1000
data <- simulateLogistic(dag)

head(data)
```

```{r}
# Fit logistic regression model
model <- glm(O ~ E + C1 + C2 + C3,
             data = data,
             family = "binomial")

# Extract estimated probabilities
pr_outcome_exp <- predict(model, type = "response")
pr_outcome_no_exp <- predict(model,
                             newdata = data.frame(E = "1",
                                                  C1 = data$C1,
                                                  C2 = data$C2,
                                                  C3 = data$C3),
                             type = "response")

# Calculate relative risk
relative_risk <- pr_outcome_exp / pr_outcome_no_exp
```

```{r}
#| layout-ncol: 2
#| label: fig-distribution_rr
#| fig-cap: "Histogram and density distribution of Relative Risk."
#| fig-subcap:
#|   - "Histogram of Relative Risk"
#|   - "Density distribution of Relative Risk"
#| fig-alt: "The histogram and density distribution of Relative Risk. The histogram displays the frequency of relative risk values, while the density plot shows the distribution of relative risk values."   
#| echo: false

data$relative_risk <- relative_risk

# Visualise relative risk distribution
ggplot(data = data, aes(x = relative_risk)) +
  geom_histogram(binwidth = 0.2,
                 fill = "#29306c",
                 color = "black") +
  labs(title = "Distribution of Relative Risk",
       x = "Relative Risk",
       y = "Frequency")

ggplot(data = data, aes(x = relative_risk)) +
  geom_density(fill = "#29306c", color = "black") +
  labs(title = "Density distribution of Relative Risk",
       x = "Relative Risk",
       y = "Density")
```

### Theoretical Minimum-Risk Exposure Levels (TMRELs)

Risk factors associated with a particular health condition are
considered based on the **Theoretical minimum risk exposure levels
(TMRELs)**\index{Theoretical Minimum Risk Exposure Levels
(TMRELs)} and as a function of the risk exposure or **relative risk
(RR)**\index{Relative Risk
(RR)} value. Not all the variables that are thought to be risk factors
increasing causes for a particular health condition are always the
driving cause of the condition, for this reason a minimum level of risk
exposure is established for the risk to be considered involved as
effective in the outcome.

Moreover, disease attributable to a particular risk factor or
combination of risk factors\index{Risk factors} need to be ascertained
by investigating the risk-outcome relationship. Risk factors can also
act indirectly on the outcome via intermediate risks, such as the
association of low fruit consumption and heart disease influenced by
systolic blood pressure which acts as mediator between the two.

```{r}
#| echo: false
#| lable: fig-risk-outcome
#| fig-cap: "DAG: Risk-Outcome leading to heart disease. The relationship between low fruit consumption, systolic blood pressure, and heart disease."
#| fig-alt: "DAG: Risk-Outcome leading to heart disease"
#| fig-width: 6
#| fig-height: 0.5
#| fig-align: "center"
library(dagitty)
fhp <- dagitty('dag{ "low fruit" -> "systolic blood pressure" -> "heart disease"}')
coordinates(fhp) <- list(
  x = c(
    "low fruit" = 0,
    "systolic blood pressure" = 0.5,
    "heart disease" = 1),
  y = c(
    "low fruit" = 0,
    "systolic blood pressure" = 0,
    "heart disease" = 0))

rethinking::drawdag(fhp)
```

Examples of risk factors with established **Theoretical Minimum Risk
Exposure Levels (TMRELs)** include particulate matter air pollution,
high systolic blood pressure, and smoking. For systolic blood pressure,
the TMREL is typically set around 110/70 mmHg. Research has shown that
maintaining blood pressure near this level is associated with the lowest
risk for cardiovascular disease and stroke. Similarly, for particulate
matter air pollution, the TMREL is set at the lowest level of exposure
that is feasible and achievable, typically based on World Health
Organization (WHO) guidelines. For smoking, the TMREL is set at zero, as
any level of smoking is associated with increased health risks.

In terms of **Disability-Adjusted Life Years
(DALYs)**\index{Disability Adjusted Life Years (DALYs)}, the overall
level is significantly influenced by behavioural, environmental, and
occupational risks. Behavioural risks, such as smoking and physical
inactivity, and environmental exposures, like air pollution, contribute
heavily to DALYs by increasing both mortality and disability rates
within affected populations. Occupational risks further add to the
burden, particularly in regions where workplace safety standards are
lower, underscoring the need for targeted interventions across different
population groups.

### Population Attributable Fractions (PAFs)

The **Population Attributable Fraction
(PAF)**\index{Population Attributable Fraction (PAF)} is a measure used
to quantify the proportion of disease incidence in a population that can
be attributed to a specific risk factor. It represents the proportion of
risk that would be reduced in a given year if the exposure to a risk
factor in the past were reduced to an ideal exposure scenario.

PAF is calculated based on the prevalence of the risk factor in the
population and the relative risk associated with that risk factor. The
formula for calculating PAF is as follows:

$$
PAF = \frac{{P_e \times (RR - 1)}}{{1 + P_e \times (RR - 1)}}
$$ {#eq-paf}

Where:

-   $P_e$ is the prevalence of the risk factor in the population.
-   $RR$ is the relative risk associated with the risk factor,
    representing the increased risk of disease among individuals exposed
    to the risk factor compared to those who are not exposed.

The PAF ranges from 0% to 100%. A PAF of 0% indicates that the risk
factor has no impact on the incidence of the disease, while a PAF of
100% indicates that all cases of the disease in the population can be
attributed to the risk factor.

PAF is useful for public health interventions as it provides insight
into the potential impact of reducing or eliminating a specific risk
factor on the incidence of disease in the population. By targeting
interventions to reduce exposure to the risk factor, public health
efforts can effectively reduce the burden of disease in the population
and improve overall health outcomes.

## Causal Inference

Causality\index{Causality} concerns the relationship between two
variables, where one variable (*the cause*) directly influences the
other (*the effect*). For example, regular exercise improves
cardiovascular health, or adequate sleep supports cognitive function.
However, causality is distinct from correlation\index{Correlation},
which indicates a statistical association without implying a direct
influence. For instance, a correlation between television watching and
obesity does not imply a causal link. Establishing causality requires
systematically evaluating alternative explanations and accounting for
confounding factors that could affect both the cause and effect.

Causal inference is essential for understanding the underlying causes of
a condition or phenomenon, even if these causes are not immediately
apparent. It often requires a structured data analysis to uncover hidden
causal relationships within the observed data.

```{r}
#| echo: false
#| eval: false
#| fig-cap: "Causal Inference Flowchart. The flowchart outlines the steps involved in performing causal inference, from formulating a hypothesis to informing public health actions."
library(DiagrammeR)

# Create a causal inference flowchart
grViz("
digraph causal_inference {
  
  # Define node styles
  node [shape = rectangle, style = filled, color = lightblue, fontname = Helvetica, fontsize = 12]
  
  # Define nodes for each step
  hypothesis [label = 'Formulate Hypothesis: Physical Inactivity -> Cardiovascular Disease']
  study_design [label = 'Choose Study Design: Cohort or RCT']
  data_collection [label = 'Collect Data on Physical Activity, Cardiovascular Disease, and Confounders']
  confounders_adjustment [label = 'Adjust for Confounders: Age, Diet, Smoking']
  causal_analysis [label = 'Perform Causal Analysis (e.g., Regression, Propensity Score Matching)']
  interpret_results [label = 'Interpret Results: Assess Association Strength and Significance']
  public_health_action [label = 'Inform Public Health Actions: Intervention Planning']

  # Define edges
  hypothesis -> study_design -> data_collection -> confounders_adjustment -> causal_analysis -> interpret_results -> public_health_action
}
")
```

![Causal Inference Flowchart. The flowchart outlines the steps involved
in performing causal inference, from formulating a hypothesis to
informing public health
actions.](images/04_causal_inference_flowchart.png){#fig-causal_inference
fig-align="center" width="50%" fig-alt="Causal Inference Flowchart"}

Performing causal inference\index{Causal inference} requires setting up
an experiment, where there are **treatment**\index{Treatment} and
**outcome**\index{Outcome} elements. The *treatment* is the intervention
applied to the data. For example, to confirm the statement that *regular
exercise leads to improved cardiovascular health*, an intervention may
be designed to introduce another variable, such as regular fruit
consumption, and observe its combined effect with exercise on
cardiovascular health.

In this investigation, the primary factors are exercise and fruit
intake. The goal is to examine whether their combination improves
cardiovascular health. Once the intervention is analysed by measuring
changes in cardiovascular health (the response variable), the next step
is to apply a control procedure, often using a **counterfactual**
scenario. This approach helps assess what might have happened in the
absence of the treatment, providing a benchmark to confirm the
treatment’s true effect.

## Summarising the Relationship Between Risk and Outcome

The relationship between risk and outcome in epidemiology is central to
understanding the causes of disease and guiding preventive strategies.
This relationship involves assessing how exposure to certain risk
factors affects the likelihood of developing specific health outcomes.
Epidemiological studies quantify the strength of this association
through measures like **Relative Risk (RR)** and **Population
Attributable Fraction (PAF)**.

**Relative Risk (RR)** compares the risk of developing a health outcome
among individuals exposed to a risk factor with those who are not
exposed. An RR greater than 1 indicates an increased risk associated
with the exposure. For example, if smokers have a relative risk of 15
for lung cancer compared to non-smokers, this suggests a strong
association between smoking and lung cancer.

**Population Attributable Fraction (PAF)** estimates the proportion of
disease incidence in a population that can be attributed to a specific
risk factor, helping quantify the potential impact of reducing or
eliminating that exposure on the overall disease burden. For example, if
smoking accounts for 30% of lung cancer cases in a population, the PAF
for smoking-related lung cancer is 0.30.

To establish causality, epidemiologists must demonstrate consistent
associations, dose-response relationships (where increased exposure
heightens risk), temporal precedence (exposure precedes outcome), and
rule out alternative explanations. Ultimately, understanding these
risk-outcome relationships enables evidence-based public health
decisions, informing preventive strategies, interventions, and policies
to improve population health.

```{r}
#| echo: false
#| label: tab-risk_measures
#| tbl-cap: "Table showing risk measures, definitions, and examples."
#| tbl-alt: "Table showing risk measures, definitions, and examples."
library(knitr)

risk_measures <- data.frame(
  Measure = c("Relative Risk (RR)", 
              "Population Attributable Fraction (PAF)", 
              "Theoretical Minimum Risk Exposure Level (TMREL)"),
  
  Definition = c(
    "Measures the likelihood of an outcome occurring in an exposed group relative to a non-exposed group.",
    "Estimates the proportion of disease cases that could be prevented if a specific risk factor were eliminated.",
    "Represents the ideal level of exposure to a risk factor that minimises adverse health effects."
  ),
  
  Example = c(
    "Smokers have a 15x higher relative risk (RR = 15) of lung cancer compared to non-smokers.",
    "If smoking accounts for 30% of lung cancer cases, then the PAF for smoking is 0.30.",
    paste0("The TMREL for PM2.5 (air pollution) is set around 2.4 $\\mu$g/m$^3$, as exposure below this level is associated with minimal health risk.")
  )
)

# Display the table
kable(risk_measures, 
      col.names = c("Risk Measure", "Definition", "Example"), 
      align = "l")
```

In conclusion, the study of risk and outcome has evolved beyond
traditional epidemiological methods to embrace advanced techniques like
transfer learning\index{Transfer learning}. This interdisciplinary
approach enables the application of insights from epidemiology to other
fields and viceversa, deepening our understanding of the complex
relationships between risk factors and health outcomes. Machine learning
and data-driven techniques help identifying patterns, and develop
predictive models that extend beyond conventional frameworks, offering
fresh perspectives on population health and guiding targeted
interventions.