TRAINING ON STATISTICAL

ANALYSIS
Hazwan Mat Din
Malaysian Research Institute on Ageing
Universiti Putra Malaysia

Hazwan Mat Din / MyAgeing 1

STATISTICAL ANALYSIS:
Which to choose?

Hazwan Mat Din / MyAgeing 2

Points to consider to select the right
statistical test
• Research question/ hypothesis

• Number of variables

• Type of data

• Number of groups

• Sample distribution

• Sample type
Hazwan Mat Din / MyAgeing 3
Common research questions in medical
research
• Difference between / among means

• Difference between / among proportions

• Association between / among factors

• Difference between / among treatment effects

Hazwan Mat Din / MyAgeing 4

T-TEST

Hazwan Mat Din / MyAgeing 5

 TYPE OF T-TEST
1. One sample t-test
 Was this sample drawn from a population with certain mean?

2. Independent t-test
 Are the groups drawn from populations with different means?

3. Paired t-test
 Two dependent ore related sample
 Each score in sample is paired with a score in another
 What is the mean difference in an outcome variable under the two
conditions?

Hazwan Mat Din / MyAgeing 6

 INDEPENDENT T-TEST
• = Student’s t-test
• Univariable analysis of numerical data
• Compare means of two independent groups
• Dependent variable is continuous
 Weight, age, height, BP, Hb level
• Independent variable is categorical (ordinal/nominal) with two
levels
 Sex (male/female), disease outcome (alive/died)

Hazwan Mat Din / MyAgeing 7

Assumptions for independent t-test

• Random samples by design

• Observations are independent samples

• Data are normally distributed in each group to check

• Sufficient sample size

• Population variance are homogenous between groups  to check

Hazwan Mat Din / MyAgeing 8

Hands on

• Data: pain_medication.sav

• Hypothesis:
 There is no significant mean difference of age between male and female;
H0: µ(male) = µ(female)

 There is significant mean differenc of age between male and female;

HA: µ(male) ≠ µ(female)

Hazwan Mat Din / MyAgeing 9

Assumption: Checking normality

Histograms show normal distribution of

age according to gender

Hazwan Mat Din / MyAgeing 10

Assumption: Equal variance

• Levene’s test equality of variance

 Not significant (p-value more than 0.05)equal variance met

 Significant (p-value less than 0.05)equal variance not met

Hazwan Mat Din / MyAgeing 11

Hands on: Independent t-test

Hazwan Mat Din / MyAgeing 12

Output

• Assumption of homogeneity of variances, • T-test p-value

p-value = 0.854, therefore the variance • There is no significant mean difference of age between male
are homogenous or equal variances and female
assumed

Hazwan Mat Din / MyAgeing 13

Presentation
Table 1. The mean difference of age between gender

Mean age (SD) t-test (df) p value

Male 54.9 (10.6)
-0.378 (198) 0.705
Female 55.4 (10.2)

Hazwan Mat Din / MyAgeing 14

 PAIRED T-TEST

• To compare means of two samples that are paired / related /

dependent eg matched /measured twice eg pre-post test design

• To study paired data, should examine the difference between the

pair

Hazwan Mat Din / MyAgeing 15

Assumptions
• Random samples

• Observation are dependent, e.g. pre- & post- data

• The differences (post minus pre) are normally distributed or

sufficient sample size need to check

Hazwan Mat Din / MyAgeing 16

Hands On
• Data: Diet_pre_post.sav

• Hypothesis
 H0: There is no significant difference of mean weight before diet and after
diet
 HA: There is significant difference of mean weight before diet and after
diet

Hazwan Mat Din / MyAgeing 17

Assumption: Checking normality
Run histogram to check for
Compute mean normality
difference

A new variable is
created
Hazwan Mat Din / MyAgeing 18
Hands on: Paired t-test

Hazwan Mat Din / MyAgeing 19

Output

• Significant p value less than 0.05

• Fail to reject HA
• There is significant difference of mean before diet and after diet

Hazwan Mat Din / MyAgeing 20

Presentation
Table 2. The mean difference of weight before and after diet
Mean of Weight (SD) Mean Difference
t statistics(df)* P value
Pre Post (95% CI)
198.4 (33.4) 190.3 (33.5) -8.0 (-9.6, 6.5) -11.18 (15) <0.001
*Paired t-test

Hazwan Mat Din / MyAgeing 21

Analysis of Variance
(ANOVA)

Hazwan Mat Din / MyAgeing 22

 THE GENERAL LINEAR MODEL
UNIVARIATE
• The general linear model is an extension of multiple linear
regression for a single dependent variable
• Provides regression analysis and analysis of variance for one
dependent variable by one or more factors and/or variable
• Can investigate interactions between factors as well as the effects
of individual factors, the effect of covariates and covariate
interaction with factors

Hazwan Mat Din / MyAgeing 23

 ANOVA
• One way ANOVA
 To test the differences of one continuous variable between one categorical
variable > 2 groups

• Two way ANOVA

 To test the effect of 2 categorical independent variables on one continuous
dependent variable

• Multifactorial ANOVA
 To test the effects of >2 categorical independent variables on one
continuous dependent variable

Hazwan Mat Din / MyAgeing 24

 ONE WAY ANOVA
• Used to determine the effect of single factor on one numerical
outcome (dependent variable)
• To compare the means of more than 2 groups of an independent
variable
• For factors with 2 categories that compare means of 2 groups,
Independent t-test is preferred
• By logic can do independent t-test for each of pair but it will
increase the type 1 error

Hazwan Mat Din / MyAgeing 25

Type 1 error
• Multiple tests to identify differences in pairs of group means
• Set alpha at 0.05, the probability to make correct decision for an
individual test is 1 – 0.05 = 0.95
• If three related t-test are conducted, each alpha set at 0.05, the
probability of making correct decisions on all three of the tests is:
 0.95 x 0.95 x 0.95 = 0.857375
• The probability of making error is 1 – 0.857375 = 0.142625. Thus,
the probability that would erroneously reject the null hypothesis
has increased from desired 0.05 to an actual error rate of
0.142625
Hazwan Mat Din / MyAgeing 26
Assumptions
• Random samples
• Independent samples
• Normality of distribution
 Histogram
 Sufficient sample size (eg. n = 15 per group)
• Homogeneity of variance
 Levene’s test
 If met, use Bonferroni adjustment
 If violated, use Dunnett’s C

Hazwan Mat Din / MyAgeing 27

Hands on
• Data: iqdata.sav

• Hypothesis:
1) HA: There is significant mean difference of IQ level among student
groups

2) HA: There is a significant association between IQ level and student

groups

Hazwan Mat Din / MyAgeing 28

Assumption: Checking normality

Histograms show
normal distribution

Hazwan Mat Din / MyAgeing 29

Hands On: One Way ANOVA

Hazwan Mat Din / MyAgeing 30

Output

Out of 2268.9 variation,

Equal Variance met 431.2 units explained (19.0%)

 The mean of IQ level is significantly different among groups

But which group?

Hazwan Mat Din / MyAgeing 31
Post hoc tests
• A significant F test in ANOVA table does not tell which pairs of
means are significantly different from one another.
• Need additional tests to determine exactly which mean differences
are significant
• Called post hoc test / multiple comparison procedures / posteriori
comparisons
• Which pair(s) that the significant lie?

Hazwan Mat Din / MyAgeing 32

Multiple Comparison Tests
• Bonferroni
• Duncan Multiple range test
Equal variance assumed
• Scheffe’s test
• Turkey’s test

Equal variance not assumed

• Dunnet’s C

Hazwan Mat Din / MyAgeing 33

Correction in Multiple Comparison
• What Bonferroni post hoc does
• To adjust error: divide the error rate by the number of tests, eg.
0.05/3 = 0.0167
 Now alpha is set at 0.0167
 The probability making a correct decision for an individual test is 1 –
0.0167 = 0.983
 The probability making three correct decision would be 0.983 x 0.93 x
0.983 = 0.94986
 The probability of making an error is 1 – 0.94986 = 0.05014 ≈ 0.05

Hazwan Mat Din / MyAgeing 34

Hands On: Post Hoc Test

Hazwan Mat Din / MyAgeing 35

Output

Physics vs Maths, p >0.999

Physics vs Chemistry, p = 0.027
Maths vs Chemistry, p = 0.029

Hazwan Mat Din / MyAgeing 36

Presentation
Table 3: Mean and standard deviation of IQ level among student groups
Group Mean (SD) F-statistics (df)* p-value
Physics student 35.80 (7.79)
Maths student 35.87 (6.94) 4.93 (2,42) 0.012
Chemistry student 42.40 (4.73)
*One Way ANOVA
Physics vs Maths, p >0.999
Physics vs Chemistry, p = 0.027
Maths vs Chemistry, p = 0.028 (Bonferroni)

 There is significant mean difference of IQ level among student groups

 There is a significant association between IQ level and student groups

Hazwan Mat Din / MyAgeing 37

CORRELATION

Hazwan Mat Din / MyAgeing 38

 What is correlation?

• The relationship between two quantitative variables without being

able to infer causal relationships

• Is a statistical technique used to determine the degree to which

two variable are related

Hazwan Mat Din / MyAgeing 39

Correlation-scatter plot
• Two quantitative variables
• Points are not joined

Hazwan Mat Din / MyAgeing 40

CORRELATION(Pearson)

• Data: HbA1c.sav

• Hypothesis
• HA: There is significant correlation between age and level of HbA1c among
study sample

Hazwan Mat Din / MyAgeing 41

Assumptions
1. Random sample
2. Independent observation
3. Linear relationship Checked using scatter
plot, an elliptical shape
4. Bivariate normal distribution

*If assumption 3 &/or 6 are not met, use Spearman correlation

Hazwan Mat Din / MyAgeing 42

Checking assumption 3 & 4

An ‘elliptical shape’ suggests linear

relationship and bivariate norma

Hazwan Mat Din / MyAgeing 43

Hands on - Correlation

r > 0.75  very good-perfect correlation*

r > 0.50 – 0.75  moderate – good correlation*
r > 0.25 – 0.50  fair correlation*
r < 0.25  weak or no correlation*

There is significant correlation between age and

HbA1c level (p<0.001). The observed r is 0.35
which suggests positive and fair correlation

*Colton T.: Statistics in Medicine. Little, Brown, 1974

Hazwan Mat Din / MyAgeing 44

PEARSON CHI-SQUARE &
FISHER’S EXACT TEST

Hazwan Mat Din / MyAgeing 45

 PEARSON CHI-SQUARE/FISHER’S EXACT
TEST
• To test the association between two categorical variables

• Example:
 Gender (Male, Female) vs HIV (Yes, No)

 Race (Malay, Chinese, Indian) vs Smoking Status (Yes, No)

Hazwan Mat Din / MyAgeing 46

Checking Assumptions

• Two variables are independent

• Two variables are categorical

• Expected count of < 5

• Equal or more 20 % - Fisher exact test

• Less than 20% - Pearson Chi-square

Hazwan Mat Din / MyAgeing 47

 PEARSON CHI-SQUARE

• Data: Coronary.sav

• Hypothesis

 HA: There is significant association between gender and coronary artery

disease

Hazwan Mat Din / MyAgeing 48

Hands on – Pearson Chi-Square

Hazwan Mat Din / MyAgeing 49

Output – Pearson Chi-Square

p-value <0.001
There is significant association
between gender and coronary
artery disease

Assumption met for expect count <5 less than 20%

Hazwan Mat Din / MyAgeing 50

MULTIPLE LINEAR
REGRESSION

Hazwan Mat Din / MyAgeing 51

 REGRESSION
• Regression analysis is a statistical tool that utilizes the relationship
between variables so that one variable can be predicted from the
other, or others

• Uses a variable (x) to predict some outcome variable (y)

• Values in y change as a function of changes in values of x

Hazwan Mat Din / MyAgeing 52

 SIMPLE LINEAR REGRESSION
• To explore the nature of relationship between two continuous
variable
• To investigate the change in response
 How much the value Y (dependent variable) varies with one unit of
change in value X (independent variable)

Y = mx + c

m: slope of the line

c: intercept on y axis

Hazwan Mat Din / MyAgeing 53

 REGRESSION COEFFICIENT
• The constant (m) the represent the rate of change of one variable
(y) as a function of changes in other (x)
• The slope of regression line

Hazwan Mat Din / MyAgeing 54

 Coefficient of Determination (r2)
• It provides a measure of how well future outcomes are likely to be
predicted by the model
• How much the independent ‘x’ is explained by dependent ‘y’
• = Square of Correlation coefficient in SLR

Hazwan Mat Din / MyAgeing 55

Hands On

• Data: Cholesterol_MLR.sav

• Hypothesis

 HA: There is significant linear relationship between age, diet, exercise and

socio-demographic index with serum cholesterol level

Hazwan Mat Din / MyAgeing 56

Hands On: SLR

Hazwan Mat Din / MyAgeing 57

Output
Regression coefficient:
y = mx + c

m = 0.06
c = 5.90

Cholesterol = 5.90 + (0.06*age)

Increase of one year of age will

increase 0.06 mmol/L of serum
cholesterol (0.02, 0.09, p = 0.002)

Hazwan Mat Din / MyAgeing 58

SLR
• Simple (SLR) – Only ONE independent variable

Age Cholesterol

Diet Cholesterol

Exercices Cholesterol

Socio-
Cholesterol
economic

Hazwan Mat Din / MyAgeing 59

Presentation
Table 4. Association of age, duration of exercise, diet inventory score and socio-
economic index with serum cholesterol level using simple linear regression
SLRa
Independent variable
b*(95% CI) P value
Age (years) 0.06 (0.02, 0.09) 0.002
Duration of exercise (hrs/week) -0.62 (-0.79, -0.46) <0.001
Diet inventory score 0.45 (0.30, 0.61) <0.001
Socio-economic index 0.21 (0.17, 0.25) <0.001

a Simplelinear regression (dependent variable: cholesterol mmol/L)

* Crude regression coefficient

Hazwan Mat Din / MyAgeing 60

 MULTIPLE LINEAR REGRESSION
• Very often a single independent variable provides an inadequate
description about the variation in a dependent variable

 For example in order to explain variability in cholesterol level, using age

as the only independent variable will be inadequate

 Variable age may explain only small percentage of variability of the total
variability in cholesterol level

 The effect found may not be due to age but may be influenced by other
factors like exercise, diet and socio-economic index
Hazwan Mat Din / MyAgeing 61
MULTIPLE LINEAR REGRESSION

Cholesterol

Age Socio-economic

Diet
Exercise

Hazwan Mat Din / MyAgeing 62

Steps in Analysis
1. Data exploration 
2. SLR
3. Variable selection
 Preliminary main-effect model
4. Checking interaction and multicollinearity
 Preliminary final model
5. Checking model assumptions
 Final model
6. Interpretation & presentation
Hazwan Mat Din / MyAgeing 63
Step 3: Variable selection

• Repeat step for variable selection for:

 Stepwise method

 Forward method

 Backward method

• Choose with the most numbers of significant independent variable

Hazwan Mat Din / MyAgeing 64

Hands on - Variable selection

Hazwan Mat Din / MyAgeing 65

Output – Stepwise method

Hazwan Mat Din / MyAgeing 66

Output – Forward Method

Hazwan Mat Din / MyAgeing 67

Output – Backward Method

Except for forward method, backward and stepwise contained the

same variables. Backward method model chosen as it contained all
significant variable

Preliminary main-effect model: Cholesterol = 7.297 + 0.033*age + 0.394*diet – 0.540*exercise

Hazwan Mat Din / MyAgeing 68

Step 4.1: Checking interaction

• All possible two way interaction should be checked

• Create interaction terms by transform, compute menu

• Add the interaction term and run the model using method “Enter”

• Check ONE interaction at ONE time

• Kept in the model if p-value <0.05

Hazwan Mat Din / MyAgeing 69

Hands on – Checking Interaction

Hazwan Mat Din / MyAgeing 70

Output – Checking Interaction

Hazwan Mat Din / MyAgeing 71

Step 4.2: Checking Multicollinearity

No VIF>10
No problem

Hazwan Mat Din / MyAgeing 72

Step 5: Checking Assumptions
• Random sample
• Linearity
• (Overall fitness) Scatter plot of unstandardized residuals against
unstandardized predicted values (XPYR)
• Independent observations
• Normality
• Histogram of unstandardized residual
• Equal variance
• Same procedure as Linearity (Scatter plot)

Hazwan Mat Din / MyAgeing 73

Hands on – Generate residual & predicted
values

Hazwan Mat Din / MyAgeing 74

Checking Assumptions- Linearity & Equal
Variance

Hazwan Mat Din / MyAgeing 75

Output –Linearity & Equal variance

Example for non-linear & non equal

variance scatter plot
Hazwan Mat Din / MyAgeing 76
Checking Assumption – Normality of
residual

Hazwan Mat Din / MyAgeing 77

Step 6: Presentation
Table 5. Association of age, duration of exercise, diet inventory score and socio-
economic index with serum cholesterol level using multiple linear regression
MLRa
Indipendent variable
Adj. b*(95% CI) P value
Age (years) 0.03 (0.01, 0.06) 0.005
Duration of exercise (hrs/week) -0.54 (-0.66, -0.44) <0.001
Diet inventory score 0.29 (0.29, 0.50) <0.001
aMultiple LinearRegression (R2 = 0.69; The model fits well; Model assumptions are met; There is no interaction effect between
independent variables and no multicollinearity problem)
*Adjusted regression coefficient

For prediction study, it is essential to report final model equation:

Cholesterol = 7.30 + (0.03*Age) + (-0.54*Exercise) + (0.29*Diet)

Hazwan Mat Din / MyAgeing 78

Step 6: Interpretation
• There is significant linear relationship between age and serum
cholesterol level (p = 0.05)
• One year older have cholesterol level higher by 0.03 mmol/L (95% CI: 0.01,0.06
mmol/L)
• There significant negative linear relationship between duration of
exercise and serum cholesterol level (p<0.001)
• Those having 1 hr/wk more exercise have cholesterol level lower by 0.54
mmol/L (95% CI: -0.66,-0.44)
• There is significant linear relationship between diet inventory index
and serum cholesterol level (p<0.001)
• Those with 1 unit more of the index, have cholesterol level higher by 0.39
mmol/L (95% CI: 0.29,0.50 mmol/L)
• With three significant variables, the model explains 69% of the
variation in the serum cholesterol level in study sample
Hazwan Mat Din / MyAgeing 79
MULTIPLE LOGISTIC
REGRESSION

Hazwan Mat Din / MyAgeing 80

 MULTIPLE LOGISTIC REGRESSION

• Multiple logistic regression is the estimation of the relationship

between a dichotomous dependent variable and more than one
independent variables or covariates

• Independent variables are the combination of numerical and

categorical variables

• Outcome is binary categorical variable

Hazwan Mat Din / MyAgeing 81

ODDS
• The odds = the chance
• The odds of an event is the ratio of the number of ways the event
can occur to the number of ways even cannot occur
• Eg.: On average 54 girls are born in every 100 births. What is the
odds of any randomly chosen delivery to be a girl?
• Number of girls/number of boys =54/46 = 1.17
• So, about 1.17 the odds to get a baby girl

Hazwan Mat Din / MyAgeing 82

ODDS RATIO
• Odds ratio is calculated by dividing the 2 odds
• Eg.: What is the odds ratio of men to have CAD compared to
women?
• The odds of men having CAD/The odds of women having CAD

Have CAD = 1 No CAD = 0

Men = 1 a b
Women = 0 c d

Hazwan Mat Din / MyAgeing 83

• The odds of men having CAD =
n of men having CAD (a)
n of men not having CAD (b)

• The odds of women having CAD =

n of women having CAD (c)
n of women not having CAD(d)

• The odds ratio (OR) of men to have CAD compared to women =

(a/b)/(c/d)

Thus, OR = ad/bc

Hazwan Mat Din / MyAgeing 84

CODING OF CATEGORICAL VARIABLES
• Always start with 0
• 0: Reference group (low risk, non-diseased, normal)
• Eg. 2 level categories:
 Smoking status: 0 (non-smoker), 1 (smoker)
 Cancer status: 0 (no cancer), 1 (has cancer)
• Eg. 3 level categories:
 Race: 0 (Malay), 1 (Chinese), 2 (Indian)
 Income level : 0 (low), 1(medium), 3 (high)

Hazwan Mat Din / MyAgeing 85

STEPS IN MULTIPLE LOGISTIC
REGRESSION
1. Data exploration & cleaning
2. Simple logistic regression
3. Variable selection
 Preliminary main effect model
4. Checking interaction & multicollinearity
 Preliminary final model
5. Checking assumption
 Final model
6. Interpretation and presentation
Hazwan Mat Din / MyAgeing 86
Hands on

• Dateset: Coronary.sav

• Research questions:

• What are the factors associated to coronary artery disease among study

sample?

Hazwan Mat Din / MyAgeing 87

Step 2: Simple logistic regression
• To screen for important independent variable

Hazwan Mat Din / MyAgeing 88

Output

• B – regression coefficient
• S.E – Standard error
• Exp (B) – ODDS RATIO
- Exp (0.332) = 1.394
Showing the reference group • 95% CI for Exp (B)/OR
defined in analysis for • Gender is significant to the model
categorical variable • Interpretation:
- Men has 1.4 times the odd to have CAD compared to
women when others confounders were not adjusted
• If the odd ratio less than 1, it is protective (less risk)

Hazwan Mat Din / MyAgeing 89

• For numerical variable

• Interpretation:
- A person with 1mmHg increase in systolic blood pressure
has 1.03 times the odds to have CAD when other
confounders were not adjusted

Repeat step 2 for all independent variables:

oDbp, chol, age, bmi, race

Hazwan Mat Din / MyAgeing 90

Result from Simple Logistic Regression
Table 6. Associated factors of coronary artery disease by simple logistic regression
model
Variable Regression Crude Odds Ratio Wald p-value
coefficient (b) (95 % CI) Statistic
Systolic blood pressure (mmHg) 0.02 1.025 (1.02,1.03) 203.49 <0.001
Diastolic blood pressure 0.05 1.053 (1.05,1.06) 245.58 <0.001
(mmHg)
Serum cholesterol (mmol/L) 0.25 1.28 (1.21,1.37) 63.07 <0.001
Age (years) 0.03 1.04 (1.02,1.05) 45.50 <0.001
Body mass index (unit) -0.01 0.99 (0.97,1.01) 1.15 0.285
Gender
Women 0 1
Men 0.33 1.39 (1.17,1.66) 13.89 <0.001
Race
Malay 0 1
Chinese -0.03 0.97 (0.79,1.19) 0.08 0.772
India -0.09 0.91 (0.74,1.12) 0.74 0.389

Hazwan Mat Din / MyAgeing 91

Step 3: Variable Selection
• Include all significant independent variables from simple logistic
regression.

• May include clinically important independent variable

• Method used for comparison:

• Forward (LR) method

• Backward (LR) method

Hazwan Mat Din / MyAgeing 92

Step 3:Hands on

Hazwan Mat Din / MyAgeing 93

Output

Repeat same process using Backward(LR)

Hazwan Mat Din / MyAgeing 94

Comparing Forward(LR) & Backward(LR)

Forward (LR)

Backward (LR)

Forward (LR) provide model without

non-significant independent
variable. Variable in Forward(LR) is
chosen for preliminary main effect
model.
Hazwan Mat Din / MyAgeing 95
Step 4: Checking Multicollinearity &
Interaction

Checking multicollinearity

Hazwan Mat Din / MyAgeing 96

Output: Multicollonearity

No >0.5
No problem

Hazwan Mat Din / MyAgeing 97

Checking interaction
• Check for possible 2-way interaction in this model
• Dbp & chol
• Dbp & gender
• Chol gender

Hazwan Mat Din / MyAgeing 98

Use Ctrl key

The interaction term (cholesterol & diastolic blood pressure) is

not significant

Hazwan Mat Din / MyAgeing 99

The interaction term (cholesterol & gender) is not significant

The interaction term (diastolic blood pressure & gender) is not

significant

Hazwan Mat Din / MyAgeing 100

Step 5: Checking Assumptions
• The Hosmer-Lemeshow test
• Classification table
• Area under the curve Receiver Operating Characteristic (ROC)
curve

Hazwan Mat Din / MyAgeing 101

 Classification table
Hosmer-Lemeshow test checking

Not significant for

assumption met

>70% for assumption met

Hazwan Mat Din / MyAgeing 102

• Area under the curve checking

Create predicted value

Hazwan Mat Din / MyAgeing 103

Create ROC curve

• Significant p-value means assumption met for area

under the curve
• The model can accurately discriminate 70.9% of the
cases

Hazwan Mat Din / MyAgeing 104

Step 6: Interpretation & presentation

Hazwan Mat Din / MyAgeing 105

Table 7: Associated factors of coronary heart disease by multiple logistic regression model
Variable Regression Adjusted Odds Ratio Wald p-value
coefficient (b) (95 % CI) Statistic
Diastolic blood pressure (mmHg) 0.05 1.05 (1.04,1.06) 212.62 <0.001
Serum cholesterol (mmol/L) 0.14 1.15 (1.07,1.23) 15.66 <0.001
Gender
Women 0 1
Men 0.40 1.49 (1.24,1.78) 18.55 <0.001
aForward LR Multiple Logistic Regression model was applied
Multicollinearity and interaction were checked and not found
Hosmer-Lemeshow test, (p=0.214), classification table (overall correctly classified percentage=86.4%)and area under the
ROC curve (70.9%) were applied to check the model fitness

Hazwan Mat Din / MyAgeing 106

Interpretation
• A person with an increase in 1mmHg of diastolic blood pressure has a
1.05 times the odd to have coronary heart disease (95% CI:1.04,1.06,
p<0.001)when adjusted for gender and serum cholesterol
• A person with an increase in 1mmol/l of serum cholesterol has a 1.15
times the odds to have coronary heart disease (95% CI: 1.07,1.23,
p<0.001) when adjusted for diastolic blood pressure and gender
• Men has 1.49 times the odds compared to women to have coronary
heart disease (95% CI: 1.24,1.78, P<0.001) when adjusted for diastolic
blood pressure and serum cholesterol
• The prediction equation:
Log (p/1-p) = -7.242 + (0.050*dbp) + (0.137*chol) + [0.398*gender(1)]

Hazwan Mat Din / MyAgeing 107

References
• Lecture notes for Medical Statistics course by Prof. Syed Hatim
Noor (2013). Unit of Biostatistics and Research Methodology,
Universiti Sains Malaysia.
• Lecture notes for Medical Statistics course by Dr. Sarimah
Abdullah (2013). Unit of Biostatistics and Research Methodology,
Universiti Sains Malaysia.
• Lecture notes for Medical Statistics course by Assoc. Prof Siti
Norsa’adah Bachok (2013). Unit of Biostatistics and Research
Methodology, Universiti Sains Malaysia.

Hazwan Mat Din / MyAgeing 108

Thank You

Hazwan Mat Din / MyAgeing 109