In this case study an analysis was carried out on a data set consisting of Alzheimer’s patients along with healthy controls. This data set was published on Kaggle by Jacob Boysen and consists of a series of Open Access imaging studies on MRI as well as metrics for each patient involved. These metrics include aspects such as Age, gender, Education level, Socioeconomic status, MMSE, CDR etc. For the the purpose of this case study we will be analyzing Socioeconomic status and Education level along with disease presence in subjects above the age of 60. This will provide us insights into a hypothesis of poorer socioeconomic status and education levels being correlated with higher risks of Alzheimer’s disease.
Alzheimer’s disease is a progressive neurodegenerative condition that is the most common cause of dementia, a general term for a decline in cognitive abilities severe enough to interfere with daily life. Named after Dr. Alois Alzheimer who first described it in 1906, the disease is characterized by symptoms like memory loss, difficulties with problem-solving or language, confusion about time and place, and changes in mood and behavior. Alzheimer’s disease primarily affects individuals over the age of 65, and the risk increases as one gets older. However, there’s also a less common form called early-onset Alzheimer’s, which affects people as young as in their 30s to mid-60s. Alzheimer’s has a plethora of risk factors. This includes things like lifestyle choices and diet. Research also dictates education and socioeconomic status as a part of risk factors for the disease. Socioeconomic status includes aspects of an individual such as education level, income, and family income. It provides an idea into how well off an individual might be. Research into Alzheimer’s disease, education, and socioeconomic status has highlighted the intricate interplay between these factors. Studies have indicated that lower levels of education and socioeconomic status are linked to an increased risk of developing Alzheimer’s disease, although the reasons behind this relationship are complex and multifaceted.Socioeconomic status impacts a variety of factors that influence Alzheimer’s risk. Lower socioeconomic status often means limited access to quality healthcare, healthy nutrition, and opportunities for mental stimulation, all of which can increase the risk of Alzheimer’s. Stress is another key factor - lower socioeconomic status is often associated with higher levels of stress, which has been linked to an increased risk of Alzheimer’s.
Provide insights into correlations or lack of correlation between Socioeconomic status and Alzheimer’s disease in patients above the age of 60
Provide insights into correlations or lack of correlation between Education level and Alzheimer’s disease in patients above the age of 60
Provide an in depth analysis and reasoning for the results
Provide the limitations of the case study and alternatives
Preliminary analysis: Using SQL and Google sheets to perform preliminary analysis. Looking at the data overall and what columns we are working with. Validating that the data is free of errors and ready for cleaning
Data cleaning: Selecting relevant columns, filtering for ages above 60, removing nulls, and making sure the data is free of errors
Data Analysis
Results: stating what results were achieved with the analysis.
Discussion: Explaining the results as well as what limitations were present in the study.
SELECT
(ID,M_F,Age,Educ,SES, MMSE, CDR)
FROM arboreal-moment-390009.Alzheimers.Alzheimers
WHERE SES is not NULL library(ggplot2)
library(dplyr)
library(broom)
library(tidyverse)
library(ggpubr)
library(readr)
library(rmarkdown)
library(knitr)
library(cowplot)
library(tinytex)data<-read.csv("oasis_cross-sectional.csv")For the purpose of this study our clean data should include records above the age of 60 as well as remove any nulls and have the data arranged by age in ascending order.
## Selecting the relevant columns and filtering for patients above the age of 60.
data_clean<-data %>%
select(ID, M.F, Age, Educ, SES, MMSE, CDR) %>%
filter(Age>= 60) %>%
##Arranging in ascending order and removing any remaining NULLS
na.omit(SES) %>%
arrange(Age)
## Taking a look at our cleaned data.
glimpse(data_clean)## Rows: 180
## Columns: 7
## $ ID <chr> "OAS1_0072_MR1", "OAS1_0200_MR1", "OAS1_0109_MR1", "OAS1_0455_MR1…
## $ M.F <chr> "F", "F", "F", "F", "M", "M", "F", "F", "F", "M", "M", "F", "F", …
## $ Age <int> 60, 60, 61, 61, 61, 62, 62, 63, 64, 64, 64, 64, 65, 65, 65, 65, 6…
## $ Educ <int> 5, 2, 4, 2, 5, 2, 3, 3, 3, 2, 5, 4, 2, 5, 3, 3, 1, 2, 2, 5, 3, 2,…
## $ SES <int> 1, 4, 3, 4, 2, 4, 3, 2, 2, 4, 2, 2, 3, 2, 4, 3, 4, 3, 4, 2, 4, 4,…
## $ MMSE <int> 30, 30, 30, 28, 30, 30, 26, 30, 30, 29, 22, 30, 29, 30, 29, 29, 2…
## $ CDR <dbl> 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.5, 0.0, 0.0, …str(data_clean)## 'data.frame': 180 obs. of 7 variables:
## $ ID : chr "OAS1_0072_MR1" "OAS1_0200_MR1" "OAS1_0109_MR1" "OAS1_0455_MR1" ...
## $ M.F : chr "F" "F" "F" "F" ...
## $ Age : int 60 60 61 61 61 62 62 63 64 64 ...
## $ Educ: int 5 2 4 2 5 2 3 3 3 2 ...
## $ SES : int 1 4 3 4 2 4 3 2 2 4 ...
## $ MMSE: int 30 30 30 28 30 30 26 30 30 29 ...
## $ CDR : num 0 0 0 0 0 0 0 0 0 0 ...
## - attr(*, "na.action")= 'omit' Named int [1:18] 5 18 20 24 38 41 52 63 69 75 ...
## ..- attr(*, "names")= chr [1:18] "5" "18" "20" "24" ...To provide us with a clearer picture during our analysis, we should ideally have our data categorized according to medical definitions of ranges regarding metrics such as MMSE and CDR. We will also categorize education level.
We need to incorporate the following tables into our data
| MMSE range | Cognitive Impairment category |
|---|---|
| 25-30 | Normal |
| 20-24 | Mild |
| 15-19 | Moderate |
| 0-14 | Severe |
| CDR rating | Dementia Category |
|---|---|
| 0 | Absent |
| 0.5 | Questionable |
| 1 | Mild |
| 2 | Moderate |
| 3 | Severe |
| 4 | Profound |
| 5 | Terminal |
| Education number | Education level |
|---|---|
| 1 | Primary |
| 2 | Lower secondary |
| 3 | Upper secondary |
| 4 | Post secondary |
| 5 | Tertiary |
data_final <- data_clean %>%
##Assigning labels for MMSE
mutate(MMSE_category = case_when(
MMSE >= 0 & MMSE <= 14 ~ "Severe",
MMSE >= 15 & MMSE <= 19 ~ "Moderate",
MMSE >= 20 & MMSE <= 24 ~ "Mild",
MMSE >= 25 & MMSE <= 30 ~ "Normal",
TRUE ~ "Other")) %>%
## Assigning labels for CDR
mutate(CDR_category = case_when(
CDR == 0 ~ "Absent",
CDR == 0.5 ~ "Questionable",
CDR == 1 ~ "Mild",
CDR == 2 ~ "Moderate",
CDR == 3 ~ "Severe",
CDR == 4 ~ "Profound",
CDR == 5 ~ "Terminal",
TRUE ~ "Other")) %>%
## Assigning labels for Education levels
mutate(edu_category = case_when(
Educ == 1 ~ "Primary education",
Educ == 2 ~ "Lower secondary education",
Educ == 3 ~ "Upper secondary education",
Educ == 4 ~ "Post secondary education",
Educ == 5 ~ "Tertiary education",
TRUE ~ "Other"))## Using the summary function for relevant columns
summary_data<-summary(data_final[ c('Age', 'Educ', 'SES', 'MMSE', 'CDR')])
## Using kable to show the summary data in a table
kable(summary_data, "pipe", col.names = c("Age", "Education level", "SES", "MMSE", "CDR"),allign = c("c", "c", "c", "c", "c"))| Age | Education level | SES | MMSE | CDR | |
|---|---|---|---|---|---|
| Min. :60.0 | Min. :1.000 | Min. :1.000 | Min. :15.0 | Min. :0.0000 | |
| 1st Qu.:71.0 | 1st Qu.:2.000 | 1st Qu.:2.000 | 1st Qu.:26.0 | 1st Qu.:0.0000 | |
| Median :76.5 | Median :3.000 | Median :3.000 | Median :28.0 | Median :0.0000 | |
| Mean :76.7 | Mean :3.094 | Mean :2.622 | Mean :26.9 | Mean :0.3139 | |
| 3rd Qu.:83.0 | 3rd Qu.:4.000 | 3rd Qu.:4.000 | 3rd Qu.:30.0 | 3rd Qu.:0.5000 | |
| Max. :96.0 | Max. :5.000 | Max. :5.000 | Max. :30.0 | Max. :2.0000 |
## Calculating statistics
percentage_female<-120/180 *100
data_summary_stats<- data_final %>%
summarise(mean_age = mean(Age),
,Total = 180, female_count = 120, male_count = 60, percentage_female)
## Showing the statistics in a table
kable(data_summary_stats, "pipe", col.names = c("Mean age", "Total", "Number of females", "Number of males", "Percentage of females(%)"), allign = c("c","c","c","c"))| Mean age | Total | Number of females | Number of males | Percentage of females(%) |
|---|---|---|---|---|
| 76.7 | 180 | 120 | 60 | 66.66667 |
## Calculating statistics
count_Alzheimers<- data_final %>%
filter(MMSE <= 24, CDR >= 0.5) %>%
nrow()
Percentage_Alzheimers<- (count_Alzheimers/180 *100)
summary_Alzheimers<- data_final %>%
summarise( mean_SES = mean(SES), mean_Educ = mean(Educ),
count_Alzheimers, Percentage_Alzheimers)
## Showing the statistics in a table
kable(summary_Alzheimers, "pipe", col.names = c("Mean SES", "Mean Education", "Number of patients with Alzheimers", "Percentage of patients with Alzheimer's(%)"), allign = c("c", "c", "c", "c"))| Mean SES | Mean Education | Number of patients with Alzheimers | Percentage of patients with Alzheimer’s(%) |
|---|---|---|---|
| 2.622222 | 3.094444 | 39 | 21.66667 |
## Turning MMSE_category and edu_category into an ordered factor
data_final$MMSE_category<- factor(data_final$MMSE_category, levels = c("Normal", "Mild", "Moderate"))
data_final$edu_category<- factor(data_final$edu_category, levels = c("Primary education", "Lower secondary education", "Upper secondary education", "Post secondary education", "Tertiary education"))
## Making a count and percentage variable for the graph
percent_data<- data_final %>%
group_by(MMSE_category, edu_category) %>%
tally() %>%
mutate(percent = n/sum(n))
## Making the graph
ggplot(percent_data, aes( x = MMSE_category, y = n, fill = edu_category))+
geom_bar(stat = "identity", position = "fill")+
geom_text(aes(label=paste0(sprintf("%1.1f", percent*100),"%")),
position=position_fill(vjust=0.5), colour="white")+
theme_minimal()+
labs(title = " MMSE categories vs Education level")+
labs(subtitle = " cognitive impairment level and maximum level of education")+
labs(caption = "Figure 1")+
labs(x = "MMSE category", y = "Percentage")+
guides(fill=guide_legend(title="Maximum Education level"))
## Turning MMSE_category and SES into an ordered factor
data_final$MMSE_category<- factor(data_final$MMSE_category, levels = c("Normal", "Mild", "Moderate"))
data_final$SES<- factor(data_final$SES)
## Making a count and percentage variable for the graph
percent_data_SES<- data_final %>%
group_by(MMSE_category, SES) %>%
tally() %>%
mutate(percent = n/sum(n))
## Making the graph
ggplot(percent_data_SES, aes( x = MMSE_category, y = n, fill = SES))+
geom_bar(stat = "identity", position = "fill")+
geom_text(aes(label=paste0(sprintf("%1.1f", percent*100),"%")),
position=position_fill(vjust=0.5), colour="white")+
theme_minimal()+
labs(title = " MMSE categories vs SES score")+
labs(subtitle = " cognitive impairment level and Socioeconomic status")+
labs(caption = "Figure 2")+
labs(x = "MMSE category", y = "Percentage")+
guides(fill=guide_legend(title="Socioeconomic status score"))
## Turning edu_category and CDR_category into an ordered factor
data_final$edu_category<- factor(data_final$edu_category, levels = c("Primary education", "Lower secondary education", "Upper secondary education", "Post secondary education", "Tertiary education"))
data_final$CDR_category<- factor(data_final$CDR_category, levels = c("Absent", "Questionable", "Mild", "Moderate"))
## Making a count and percentage variable for the graph
percent_data_CDR<- data_final %>%
group_by(CDR_category, edu_category) %>%
tally() %>%
mutate(percent = n/sum(n))
## Making the graph
ggplot(percent_data_CDR, aes( x = CDR_category, y = n, fill = edu_category))+
geom_bar(stat = "identity", position = "fill")+
geom_text(aes(label=paste0(sprintf("%1.1f", percent*100),"%")),
position=position_fill(vjust=0.5), colour="white")+
theme_minimal()+
labs(title = " CDR category vs Education level")+
labs(subtitle = " Clinical Dementia Rating and maximum level of education")+
labs(caption = "Figure 3")+
labs(x = "CDR category", y = "Percentage")+
guides(fill=guide_legend(title="Maximum Education level"))
## Turning CDR_category and SES into an ordered factor
data_final$CDR_category<- factor(data_final$CDR_category, levels = c("Absent", "Questionable", "Mild", "Moderate"))
data_final$SES<- factor(data_final$SES)
## Making a count and percentage variable for the graph
percent_data_CDR_SES<- data_final %>%
group_by(CDR_category, SES) %>%
tally() %>%
mutate(percent = n/sum(n))
## Making the graph
ggplot(percent_data_CDR_SES, aes( x = CDR_category, y = n, fill = SES))+
geom_bar(stat = "identity", position = "fill")+
geom_text(aes(label=paste0(sprintf("%1.1f", percent*100),"%")),
position=position_fill(vjust=0.5), colour="white")+
theme_minimal()+
labs(title = " CDR categories vs SES score")+
labs(subtitle = " Clinical Dementia Rating and Socioeconomic status")+
labs(caption = "Figure 4")+
labs(x = "CDR category", y = "Percentage")+
guides(fill=guide_legend(title="Socioeconomic status score"))
## Making the graphs
plot1<- ggplot(data_final, mapping = aes( x = MMSE, y = Educ))+
geom_jitter()+
geom_smooth(method = lm, se = FALSE)+
theme_minimal()+
labs(title = " MMSE scores vs Education levels")+
labs(caption = "Figure 5")+
labs(x = "MMSE scores", y = "Education level")
data_final$SES<- as.numeric(data_final$SES)
plot2<-ggplot(data_final, mapping = aes( x = MMSE, y = SES))+
geom_jitter()+
geom_smooth(method = lm, se = FALSE)+
theme_minimal()+
labs(title = " MMSE scores vs SES scores")+
labs(caption = "Figure 6")+
labs(x = "MMSE scores", y = "SES score")
## Making a plot grid to show both graphs
plot_grid(plot1, plot2)## `geom_smooth()` using formula = 'y ~ x'
## `geom_smooth()` using formula = 'y ~ x'
## Making the graphs
plot3<-ggplot(data_final, mapping = aes( x = CDR, y = Educ))+
geom_jitter()+
geom_smooth(method = lm, se = FALSE)+
theme_minimal()+
labs(title = " CDR rating vs Education levels")+
labs(caption = "Figure 7")+
labs(x = "CDR rating", y = "Education level")
data_final$SES<- as.numeric(data_final$SES)
plot4<-ggplot(data_final, mapping = aes( x = CDR, y = SES))+
geom_jitter()+
geom_smooth(method = lm, se = FALSE)+
theme_minimal()+
labs(title = " CDR rating vs SES scores")+
labs(caption = "Figure 8")+
labs(x = "CDR rating", y = "SES score")
## Making a plot grid to show both graphs
plot_grid(plot3, plot4)## `geom_smooth()` using formula = 'y ~ x'
## `geom_smooth()` using formula = 'y ~ x'
## Making sure both variables are numeric and running a correlation on them
data_final$Educ<- as.numeric(data_final$Educ)
data_final$SES<- as.numeric(data_final$SES)
cor.test(data_final$MMSE,data_final$Educ, method = 'spearman')## Warning in cor.test.default(data_final$MMSE, data_final$Educ, method =
## "spearman"): Cannot compute exact p-value with ties##
## Spearman's rank correlation rho
##
## data: data_final$MMSE and data_final$Educ
## S = 716411, p-value = 0.0003626
## alternative hypothesis: true rho is not equal to 0
## sample estimates:
## rho
## 0.262929## Making sure both variables are numeric and running a correlation on them
data_final$Educ<- as.numeric(data_final$Educ)
data_final$SES<- as.numeric(data_final$SES)
cor.test(data_final$MMSE,data_final$SES, method = 'spearman')## Warning in cor.test.default(data_final$MMSE, data_final$SES, method =
## "spearman"): Cannot compute exact p-value with ties##
## Spearman's rank correlation rho
##
## data: data_final$MMSE and data_final$SES
## S = 1153074, p-value = 0.01227
## alternative hypothesis: true rho is not equal to 0
## sample estimates:
## rho
## -0.1863264## Making sure both variables are numeric and running a correlation on them
data_final$Educ<- as.numeric(data_final$Educ)
data_final$SES<- as.numeric(data_final$SES)
cor.test(data_final$CDR,data_final$Educ, method = 'spearman')## Warning in cor.test.default(data_final$CDR, data_final$Educ, method =
## "spearman"): Cannot compute exact p-value with ties##
## Spearman's rank correlation rho
##
## data: data_final$CDR and data_final$Educ
## S = 1135925, p-value = 0.0236
## alternative hypothesis: true rho is not equal to 0
## sample estimates:
## rho
## -0.1686834## Making sure both variables are numeric and running a correlation on them
data_final$Educ<- as.numeric(data_final$Educ)
data_final$SES<- as.numeric(data_final$SES)
cor.test(data_final$CDR,data_final$SES, method = 'spearman')## Warning in cor.test.default(data_final$CDR, data_final$SES, method =
## "spearman"): Cannot compute exact p-value with ties##
## Spearman's rank correlation rho
##
## data: data_final$CDR and data_final$SES
## S = 855702, p-value = 0.1097
## alternative hypothesis: true rho is not equal to 0
## sample estimates:
## rho
## 0.1196205When plotting MMSE categories against education levels, we see results that reflect our initial hypothesis. A significantly higher percentage of patients with a normal MMSE also had tertiary education as their highest level of education (25.5%). Furthermore, as we move along to mild and moderate the percentages follow a decreasing trend of education. For a MMSE relating to mild cognitive impairment the majority of patients(32.3%) had lower secondary education as their highest attained level of education. With moderate cognitive impairment the majority lies even lower at primary education being the majority(37.5%)(figure 1). This correlates with our initial hypothesis. We can further confirm our findings with figure 5. The jitterplot shows a trend of increasing MMSE as education level is increased. A spearman’s rank correlation coefficient gives us a value of 0.26 with a P value less than 0.05. This signifies a statistically significant weak positive correlation between education level and MMSE scores.
When plotted in a bar graph, the results for MMSE scores and SES are harder to infer. According to figure 2 a huge majority for mild and moderate cognitive impairment scores lie in the 4th score of socioeconomic status( 41.9% and 37.5% respectively. This goes against our initial hypothesis as it validates higher socioeconomic status as a correlation to lower cognitive scores. With further analysis using a jitterplot (Figure 6) we see a negative correlation with higher SES scores equating to a lower MMSE score. A spearman’s rank correlation coefficient gives us a value of -0.18 and a P value less than 0.05 making the correlation negatively correlated and statistically significant.
In a bar graph with CDR and Education level we see tertiary education and lower secondary education as the majority(25.8% each) for healthy patients. For questionable and mild dementia we see lower secondary as the majority and finally for moderate dementia we see both primary and upper secondary as the majority (50% each). This does not exactly line up with our hypothesis however somewhat supports it. Upon further inspection with a jitterplot (Figure 7) we see a negative correlation equating lower education levels to higher ratings of dementia. To confirm a correlation that is statistically significant we use a spearman’s rank correlation coefficient and calculate a value of -0.17 with a P value of 0.02. Since our P value is less than 0.05 we can say the correlation is statistically significant. The results support our initial hypothesis.
When plotting CDR and SES scores in a bar graph we see that lower socioeconomic status(2) is a majority(32%) for healthy patients. As we increase in CDR scores the Socioeconomic status also increases. For both mild and moderate dementia ratings the majority lies in a socioeconomic status of 4 or more. Using a jitterplot we can further validate this(Figure 8). We see a positive correlation with a larger socioeconomic status equating to higher CDR ratings. Finally to validate correlation we use a spearman’s rank correlation coefficient test and receive a value of 0.11 with a P value of 0.1. Since our P value is not less than 0.05 we can say this correlation is not statistically significant.
As seen by our results we see a statistically significant correlation in both SES scores and Education level for Alzheimer’s. A higher Educational level eg. tertiary or upper secondary would directly correlate to higher percentages of normal patients. Most of the patients seen in this study with mild or moderate Alzheimer’s would have had their highest level of education be either primary education or some lower variant of secondary education. Both in MMSE scores and CDR ratings we see similar trends relating higher education to lower disease prevalence. From a statistical point of view this seems to have answered our question, however, unless we delve further into why or how these results might be misleading we cannot be sure in our answer. Firstly lets discuss the MMSE test itself. The Mini Mental State Examination (MMSE) is a set of 11 questions that is commonly used for cognitive impairment. The MMSE contains 5 different domains as follows: Orientation, Repetition, Verbal recall, Attention and calculation, and language and visual construction. Research has shown that induviduals with higher education levels score higher on the MMSE. This remains true to this case study as well. Data points with higher education levels were correlated with a higher MMSE score as well. Therefore, this relationship has a valid basis to it. However, the MMSE is used for cognitive impairment rather than as a straightforward diagnosis tool for Alzheimer’s. Scoring higher on the MMSE does not translate to a lack of Alzheimer’s disease. The MMSE is heavily influenced by non-cognitive aspects such as culture, languages and levels of literacy. Therefore induviduals scoring lower on the MMSE do not necessarily have Alzheimer’s. Moreover, When we take a closer look at our data points in the jitterplot, we realize that a majority of our data is in the healthy range. With only 21% of patients having Alzheimer’s of some extent. Therefore when we look at our data we realize that it is skewed towards the healthy side. There is not enough data present in the Alzheimer’s disease category for us to be sure in our analysis. Only 39 patients out of 180 have Alzheimer’s in this dataset. Our correlation is weak and needs more data for it to be supported with assurity. Moreover, when we look at socioeconomic status we see the opposite relationship to what we expected initially. As SES scores increase, the prevalence of Alzheimer’s increases as well. This is reflected in both MMSE scores and CDR ratings. It should be noted however that this correlation was very weak and only statistically significant when discussing SES and MMSE scores. When we have a relationship this weak its imperative we understand where we could have gone wrong. As previously mentioned, this dataset lacks enough data points in the diseased range. Most of our data relates to healthy patients and with such little data regarding Alzheimer’s patients we cannot be a hundred percent sure of our analysis. Since the correlation discovered between SES and Alzheimer’s is extremely weak in our analysis, we can infer that this is due to some sort of sampling error or bias. Without a large plethora of data, these errors are prone and could potentially lead to wrong results.
In conclusion a weak correlation was found between education levels and Alzheimer’s disease. The correlation was calculated to be statistically significant however there is a lack of data present for us to be a hundred percent sure in this correlation. Due to limitations no solid relationship could be gauged between SES and Alzheimer’s disease from this dataset.