Introduction

This document presents a complete end-to-end Eye Cancer Diagnosis analysis including:

  • Descriptive Analysis
  • Analysis on dataset
  • K-Means Clustering
  • ANOVA

Explaination: The dataset used in this analysis contains detailed medical records of eye cancer patients, collected from multiple countries. It includes demographic information (age, gender, country), clinical features (cancer type, laterality, stage at diagnosis), and treatment details such as surgery, radiation therapy, and chemotherapy. The dataset also provides patient outcomes and survival time, enabling both diagnostic and prognostic analysis. This rich combination of variables allows comprehensive exploration of patterns, risk factors, and treatment effectiveness in eye cancer cases.

Library used are

library("readr")
library("dplyr")
## 
## Attaching package: 'dplyr'
## The following objects are masked from 'package:stats':
## 
##     filter, lag
## The following objects are masked from 'package:base':
## 
##     intersect, setdiff, setequal, union
library("ggplot2")

Load Data

data_path <- read.csv("~/Downloads/eye_cancer_patients.csv")

Cleaning & Preprocessing

data <- raw %>% janitor::clean_names()

data <- data %>%
mutate(
patient_id = as.character(patient_id),
age = as.numeric(age),
gender = as.factor(gender),
cancer_type = as.factor(cancer_type),
laterality = as.factor(laterality),
date_of_diagnosis = lubridate::ymd(date_of_diagnosis),
stage_at_diagnosis = as.factor(stage_at_diagnosis),
treatment_type = as.factor(treatment_type),
surgery_status = ifelse(tolower(as.character(surgery_status)) %in% c("true","1","yes"), TRUE, FALSE),
radiation_therapy = as.numeric(radiation_therapy),
chemotherapy = as.numeric(chemotherapy),
outcome_status = as.factor(outcome_status),
survival_time_months = as.numeric(survival_time_months),
genetic_markers = as.factor(ifelse(is.na(genetic_markers),"Unknown",genetic_markers)),
family_history = ifelse(tolower(as.character(family_history)) %in% c("true","1","yes"), TRUE, FALSE),
country = as.factor(country)
)

data <- data %>%
mutate(
age_group = cut(age, breaks=c(-Inf,18,65,Inf),
labels=c("Children (0-18)","Adults (19-65)","Seniors (65+)")),
deceased_flag = ifelse(outcome_status=="Deceased",1,0)
)

glimpse(data)
## Rows: 5,000
## Columns: 18
## $ patient_id           <chr> "PID00001", "PID00002", "PID00003", "PID00004", "…
## $ age                  <dbl> 58, 15, 64, 33, 8, 41, 67, 26, 5, 23, 63, 22, 62,…
## $ gender               <fct> F, Other, M, M, Other, Other, Other, Other, Other…
## $ cancer_type          <fct> Retinoblastoma, Retinoblastoma, Retinoblastoma, M…
## $ laterality           <fct> Left, Right, Bilateral, Right, Left, Bilateral, R…
## $ date_of_diagnosis    <date> 2019-01-25, 2021-10-21, 2021-03-12, 2021-05-10, …
## $ stage_at_diagnosis   <fct> Stage IV, Stage III, Stage IV, Stage II, Stage I,…
## $ treatment_type       <fct> Radiation, Chemotherapy, Surgery, Radiation, Chem…
## $ surgery_status       <lgl> FALSE, TRUE, FALSE, TRUE, FALSE, TRUE, FALSE, FAL…
## $ radiation_therapy    <dbl> 15, 69, 47, 36, 14, 11, 50, 46, 48, 42, 45, 65, 4…
## $ chemotherapy         <dbl> 3, 6, 6, 6, 14, 17, 2, 13, 7, 2, 17, 3, 7, 15, 5,…
## $ outcome_status       <fct> Deceased, In Remission, In Remission, Active, In …
## $ survival_time_months <dbl> 85, 10, 3, 40, 26, 15, 93, 9, 12, 116, 99, 25, 30…
## $ genetic_markers      <fct> None, None, BRAF Mutation, None, BRAF Mutation, B…
## $ family_history       <lgl> TRUE, TRUE, FALSE, FALSE, TRUE, TRUE, FALSE, TRUE…
## $ country              <fct> UK, Japan, UK, Canada, USA, UK, Australia, German…
## $ age_group            <fct> Adults (19-65), Children (0-18), Adults (19-65), …
## $ deceased_flag        <dbl> 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 1, 0, 1, 0, 1, 1…

Selection of Relevants Columns for modelling

# Select relevant columns for modelling

model_df <- data %>%
select(age, gender, cancer_type, laterality, stage_at_diagnosis,
treatment_type, surgery_status, radiation_therapy, chemotherapy,
survival_time_months, genetic_markers, family_history,
country, deceased_flag)

# Numeric variables

num_vars <- c("age","radiation_therapy","chemotherapy","survival_time_months")

# Create model_scaled

model_scaled <- model_df
model_scaled[num_vars] <- scale(model_scaled[num_vars])

glimpse(model_scaled)
## Rows: 5,000
## Columns: 14
## $ age                  <dbl> 0.50095650, -1.15707379, 0.73230956, -0.46301460,…
## $ gender               <fct> F, Other, M, M, Other, Other, Other, Other, Other…
## $ cancer_type          <fct> Retinoblastoma, Retinoblastoma, Retinoblastoma, M…
## $ laterality           <fct> Left, Right, Bilateral, Right, Left, Bilateral, R…
## $ stage_at_diagnosis   <fct> Stage IV, Stage III, Stage IV, Stage II, Stage I,…
## $ treatment_type       <fct> Radiation, Chemotherapy, Surgery, Radiation, Chem…
## $ surgery_status       <lgl> FALSE, TRUE, FALSE, TRUE, FALSE, TRUE, FALSE, FAL…
## $ radiation_therapy    <dbl> -0.99425481, 1.63146947, 0.56172995, 0.02686019, …
## $ chemotherapy         <dbl> -1.1701935, -0.6727609, -0.6727609, -0.6727609, 0…
## $ survival_time_months <dbl> 0.70287001, -1.46791120, -1.67051744, -0.59959871…
## $ genetic_markers      <fct> None, None, BRAF Mutation, None, BRAF Mutation, B…
## $ family_history       <lgl> TRUE, TRUE, FALSE, FALSE, TRUE, TRUE, FALSE, TRUE…
## $ country              <fct> UK, Japan, UK, Canada, USA, UK, Australia, German…
## $ deceased_flag        <dbl> 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 1, 0, 1, 0, 1, 1…

Descriptive Analysis

#1. Age Distribution (Stacked by Cancer Type)

age_plot <- data %>%
group_by(age_group, cancer_type) %>%
summarise(count=n(), .groups="drop") %>%
group_by(age_group) %>%
mutate(pct = count/sum(count)*100)

ggplot(age_plot, aes(age_group, count, fill=cancer_type)) +
geom_bar(stat="identity") +
geom_text(aes(label=paste0(sprintf("%.1f",pct),"%")),
position=position_stack(vjust=0.5), color="white") +
labs(title="Age Distribution by Cancer Type", x="Age Group", y="Patient Count") +
theme_minimal()

Survival By Stage

#2. Survival by Stage

ggplot(data, aes(stage_at_diagnosis, survival_time_months, fill=stage_at_diagnosis)) +
geom_boxplot() +
labs(title="Survival Time by Stage", y="Survival (months)") +
theme_minimal() +
scale_fill_brewer(palette="Reds")

patients in earlier stages (Stage I–III) show similar survival patterns, with median survival around 55–60 months. Stage IV shows a slightly higher median only because the dataset contains a few long-surviving Stage IV patients, which pulls the median upward. It does NOT mean Stage IV patients live longer. Overall, the plot confirms that survival decreases as the cancer stage becomes more advanced, but the synthetic data creates small variations.

Treatment and Outcome

#3. Treatment vs Outcome

df <- data %>%
group_by(treatment_type, outcome_status) %>%
summarise(n=n(), .groups="drop") %>%
group_by(treatment_type) %>%
mutate(pct = n/sum(n)*100)

ggplot(df, aes(treatment_type, n, fill=outcome_status)) +
geom_bar(stat="identity") +
geom_text(aes(label=paste0(sprintf("%.1f",pct),"%")),
position=position_stack(vjust=0.5), color="white") +
labs(title="Treatment vs Outcome", x="Treatment Type", y="Count") +
theme_minimal() +
scale_fill_brewer(palette="Set1")

K-Means Clustering

km_data <- model_scaled %>%
select(age, radiation_therapy, chemotherapy, survival_time_months) %>%
na.omit()

# Find optimal k (silhouette)

sil <- sapply(2:6, function(k){
km <- kmeans(km_data, centers=k, nstart=10)
mean(cluster::silhouette(km$cluster, dist(km_data))[,3])
})

best_k <- which.max(sil) + 1

km_fit <- kmeans(km_data, centers=best_k, nstart=25)

factoextra::fviz_cluster(list(data=km_data, cluster=km_fit$cluster)) +
ggtitle(paste("K-means Clustering (k=",best_k,")"))

The K-means clustering algorithm divided the patients into five distinct groups using four numerical variables: age, radiation therapy, chemotherapy, and survival time. Each colored cluster represents patients with similar characteristics. The X and Y axes (Dim1 and Dim2) are PCA-derived dimensions that summarize the main patterns in the data.

Cluster 1 (Red) - Younger patients, Low treatment, Medium survival

Cluster 2 (Yellow/Orange) - Moderate age, Moderate treatment, Wide survival range

Cluster 3 (Green) - Older patients, High radiation & chemo, Lower survival

Cluster 4 (Blue) - Low radiation, Long survival, Possibly early-stage patients

Cluster 5 (Pink/Purple) - Mixed pattern, Medium treatment, Medium survival

ANOVA

aov_fit <- aov(survival_time_months ~ cancer_type, data=data)
summary(aov_fit)
##               Df  Sum Sq Mean Sq F value Pr(>F)
## cancer_type    2     330   165.2   0.138  0.871
## Residuals   4997 5966910  1194.1
ggplot(data, aes(cancer_type, survival_time_months, fill=cancer_type)) +
geom_boxplot() + theme_minimal() +
labs(title="Survival by Cancer Type", y="Survival (months)") +
theme(axis.text.x = element_text(angle=30, hjust=1))

There is no significant difference in survival time between the cancer types (Lymphoma, Melanoma, Retinoblastoma).

Why is the p-value so high?

The average survival time of all three cancer types is very similar. Their boxplots overlap a lot. There is no strong separation between groups.