Transcript Statistics for clinicians

Statistics for clinicians

Biostatistics course by Kevin E. Kip, Ph.D., FAHA
Professor and Executive Director, Research Center
University of South Florida, College of Nursing
Professor, College of Public Health
Department of Epidemiology and Biostatistics
Associate Member, Byrd Alzheimer’s Institute
Morsani College of Medicine
Tampa, FL, USA
1
SECTION 1.1
Module Overview
and Introduction
Introduction to biostatistics,
descriptive statistics,
SPSS, and Power Point.
SECTION 1.4
Introduction
to SPSS
Introduction to SPSS
•
•
•
•
•
Database structure
Data view and variable view
Variable names, labels, and formats
Interactive menus
SPSS syntax generated from interactive analyses
SECTION 1.5
Summarizing
Data in Charts
Summarizing Data – Charts
1.
One categorical, >1 proportion/percentage
(i)
Bar chart
(ii) Stacked bar chart
(iii) Stacked bar chart (100%)
2.
One categorical, >1 continuous variable
(i)
Box plot
(ii) High-low
(iii) Line
(iv) Kernel-density plots
3.
Two continuous variables
(i)
X-Y scatter
(ii) Histogram (can be used for 1 variable)
1.
One categorical, >1 proportion/percentage
(i)
Bar chart
 Rectangular bars with lengths proportional to the values that they represent.
 Bars can be plotted vertically or horizontally.
1.
One categorical, >1 proportion/percentage
(ii) Stacked bar chart
 Can be counts or percentages.
 Do not sum to a specified value
% Obese
Age Group
1.
One categorical, >1 proportion/percentage
(iii) Stacked bar chart (100%)
Bar Charts and Stacked Bar Charts
Important to select either row versus column percentages
Example: Race and blood pressure classification
Usually, the row variable is the “predictor”, and the column
variable is the “outcome”.
SPSS:
Analyze
Descriptive statistics
Crosstabs
Bar Charts and Stacked Bar Charts
Column Percentage:
SPSS-CROSSTABS
/TABLES=SCR_RACECAT3 BY SCR_BP_CLASS4
/FORMAT=AVALUE TABLES
/CELLS=COUNT COLUMN
/COUNT ROUND CELL
/BARCHART.
Race * BP classification Crosstabulation
BP classification
Normal
Race
White
Count
% within BP
classification
Black
Count
% within BP
classification
Other
Count
% within BP
classification
Total
Count
% within BP
classification
Prehypertensive
247
397
Hypertensive Hypertensive
Stage 1
Stage 2
294
95
Total
1033
65.2%
58.3%
49.8%
38.0%
54.4%
117
262
275
149
803
30.9%
38.5%
46.6%
59.6%
42.3%
15
22
21
6
64
4.0%
3.2%
3.6%
2.4%
3.4%
379
681
590
250
1900
100.0%
100.0%
100.0%
100.0%
100.0%
Difficult to identify trends
Bar Charts and Stacked Bar Charts
Row Percentage:
SPSS-CROSSTABS
/TABLES=SCR_RACECAT3 BY SCR_BP_CLASS4
/FORMAT=AVALUE TABLES
/CELLS=COUNT ROW
/COUNT ROUND CELL
/BARCHART.
Race * BP classification Crosstabulation
Normal
Race
White
Count
% within Race
Black
Count
% within Race
Other
Count
% within Race
Total
Count
% within Race
BP classification
Hypertensive
Prehypertensive
Stage 1
Hypertensive
Stage 2
247
397
294
95
Total
1033
23.9%
38.4%
28.5%
9.2%
100.0%
117
262
275
149
803
14.6%
32.6%
34.2%
18.6%
100.0%
15
22
21
6
64
23.4%
34.4%
32.8%
9.4%
100.0%
379
681
590
250
1900
19.9%
35.8%
31.1%
13.2%
100.0%
Use row percentages in stacked bar chart (PP)
Power Point Chart
Column
100% Stacked Column
Power Point Chart (Practice)
Column - 100% Stacked Column
Display Quality of Life from Poor to Excellent by Gender
Column Percentages for QOL
Gender * QOL: Health Crosstabulation
QOL: Health
Excellent Very good
Good
Gender
Male
Count
% within QOL: Health
Female
Count
% within QOL: Health
Total
Count
% within QOL: Health
Fair
Poor
128
245
249
58
6
Total
686
40.8%
34.4%
34.7%
26.6%
25.0%
34.5%
186
467
469
160
18
1300
59.2%
65.6%
65.3%
73.4%
75.0%
65.5%
314
712
718
218
24
1986
100.0%
100.0%
100.0%
100.0%
100.0%
100.0%
Row Percentages for QOL
Gender * QOL: Health Crosstabulation
QOL: Health
Excellent Very good
Good
Gender
Male
Count
% within Gender
Female
Count
% within Gender
Total
Count
% within Gender
Fair
Poor
128
245
249
58
6
Total
686
18.7%
35.7%
36.3%
8.5%
.9%
100.0%
186
467
469
160
18
1300
14.3%
35.9%
36.1%
12.3%
1.4%
100.0%
314
712
718
218
24
1986
15.8%
35.9%
36.2%
11.0%
1.2%
100.0%
Power Point Chart
Column
100% Stacked Column
Power Point Chart
Column
100% Stacked Column
2.
One categorical, >1 continuous variable
(i)
Box plot
 Also known as box-and-whisker diagram.
 Displays 5 summary statistics: minimum, lower quartile (Q1), median (Q2),
upper quartile (Q3), and maximum
 No assumptions on underlying statistical distribution – non-parametric
SPSS:
Graphs
Chart Builder
Boxplot
Example: HDL Cholesterol (continuous) distribution by
gender (categorical)
2.
One categorical, >1 continuous variable
(i)
Box plot
Question: Are HDL cholesterol levels positively or negative skewed?
Run SPSS frequencies procedure
2.
One categorical, >1 continuous variable
(i)
Box plot
Question: Are triglycerides positively or negative skewed?
Run SPSS frequencies procedure
2.
One categorical, >1 continuous variable
(i)
Box plot (Practice)
Draw a box plot of the distribution of HDL cholesterol by ethnicity:
Hispanic:
Min=30, Q1=40, Q2=46, Q3=56, Max=86
Non-Hispanic:
Min=21, Q1=46, Q2=56, Q3=66, Max=131
Example:
2.
One categorical, >1 continuous variable
(i)
Box plot (Practice)
Draw a box plot of the distribution of HDL cholesterol by ethnicity:
Hispanic:
Min=30, Q1=40, Q2=46, Q3=56, Max=86
Non-Hispanic:
Min=21, Q1=46, Q2=56, Q3=66, Max=131
2.
One categorical, >1 continuous variable
(ii) High-low
 Can “trick” Power Point to use open-high-low-close chart (i.e. used for financials) to show
distributions of continuous variables
 Upper and lower ends (high-low) can represent any percentiles, such as 5th/95th percentiles
Total Cholesterol
(mg/dl)
P=0.003
White
Black
Ptrend=0.009
EU>85%
EU>40%
EU>25%
EU<40%
EU<25%
White
Black
Black
Black
Self-Report
N
(753)
(464)
Admixture Defined
(753)
(68)
(201)
The filled rectangles depict the interquartile range (25th and 75th percentile). The lower
and upper limits of the vertical lines depict the 5th and 95th percentiles, respectively.
(195)
Total Cholesterol
(mg/dl)
U.S. Black vs. Ghana Urban:
P=0.0001
U.S. Black vs. Ghana Rural:
P<0.0001
Ghana Urban vs. Ghana Rural: P<0.0001
N=594
N=546
N=80
N=111
The filled rectangles depict the interquartile range (25th and 75th percentile). The lower
and upper limits of the vertical lines depict the 5th and 95th percentiles, respectively.
5%
Male
137
Female 153
25%
175
190
75%
224
245
95%
271
295
Total Cholesterol: (Practice in Power Point – first draw by hand)
(mg/dl)
The filled rectangles depict the interquartile range (25th and 75th percentile). The lower
and upper limits of the vertical lines depict the 5th and 95th percentiles, respectively.
Total Cholesterol: (Practice in Power Point)
(mg/dl)
5%
Male
137
Female 153
25%
175
190
75%
224
245
95%
271
295
“Trick” Power Point
Open
High
Low
25%
95%
5%
The filled rectangles depict the interquartile range (25th and 75th percentile). The lower
and upper limits of the vertical lines depict the 5th and 95th percentiles, respectively.
Close
75%
2.
One categorical, >1 continuous variable
(iii) Line chart
 Typically represents trend in data over intervals of time (i.e. time series)
 Often used to show repeated health outcome measurements over time.
Prevalence of Use (%)” Crohn’s Disease Medications
In this example, the “categorical” variable is individual
subject nested within each treatment arm of the trial
2.
One categorical, >1 continuous variable
(iv) Kernel density plots
 Like a histogram, but constructs a “smooth” probability density function
3.
Two continuous variables
(i)
X-Y scatter
 Shows the relationship between two sets of continuous data
 Also called a scatter chart, scattergram, scatter diagram or scatter graph.
Body Density
1.1
1.08
1.06
1.04
1.02
1
0.98
0.96
15
20
25
30
35
40
Body Mass Index
45
50
55
60
3.
Two continuous variables
(ii) Histogram(s)
 Probability distribution of a continuous variable(s) displayed over discrete intervals (bins)
 The bins contain frequency counts, or can be normalized to display relative frequencies
(i.e. proportion of cases that fall into each category (bin) with total area = 1.0)
# subjects
3.
Two continuous variables
(ii) Histogram(s)
 Probability distribution of a continuous variable(s) displayed over discrete intervals (bins)
 The bins contain frequency counts, or can be normalized to display relative frequencies
(i.e. proportion of cases that fall into each category (bin) with total area = 1.0)
SECTION 1.6
SPSS
Data Manipulation
SPSS Data Manipulation and Syntax Editor
1. Recode continuous variable into arbitrarilydefined or pre-defined categories
2. Visual binning of continuous variable
3. Transform a skewed variable
4. Using the SPSS Data Editor
SPSS Data Manipulation and Syntax Editor
1. Recode continuous variable into arbitrarily-defined
or pre-defined categories
Example: Define age into 3 categories (arbitrary)
45-54
55-64
65 and older
SPSS
Transform
Recode into different variables
Input variable is age
Output variable
Name:
age_cat
Label:
Age in 3 categories
Click on old and new values
Range – specify explicitly
45-54 = value 1
54 64 = value 2
65 and older = value 3
SPSS Data Manipulation and Syntax Editor
2.
Visual binning of continuous variable
Example: Body mass index
Put in output name for binned variable
Make cutpoints
Equal percentiles based on scanned cases
Put in labels for frequency display in bar chart
SPSS Code
Visual Binning.
SPSS Data Manipulation and Syntax Editor
3.
Transform a skewed variable
Descriptive statistics for triglycerides in natural scale
Mean, median, SD, min, max, skewness, kurtosis
Chart = histogram with normal curve superimposed
Triglycerides are skewed. Use a transformation to create a new variable and reduce the
skew in triglycerides.
SPSS
Compute variable
Target Variable:
Numeric Expression:
SPSS Syntax:
LOG_TRIG
lg10(LAB_TRIG_VAP)
COMPUTE log_trig=lg10(LAB_TRIG_VAP).
SPSS Data Manipulation and Syntax Editor
4.
Using the SPSS Data Editor
SPSS:
File: New (syntax)
Save the file with a new name
1.
Select males only (scr_sex=1)
Data
Select Cases
If scr_sex=1
USE ALL.
COMPUTE filter_$=(SCR_SEX=1).
VARIABLE LABELS filter_$ 'SCR_SEX=1 (FILTER)'.
VALUE LABELS filter_$ 0 'Not Selected' 1 'Selected'.
FORMATS filter_$ (f1.0).
FILTER BY filter_$.
EXECUTE.
2. Run descriptives for age
3. Copy code and repeat for females (scr_sex=2);
SPSS Data Manipulation and Syntax Editor
4.
Using the SPSS Data Editor
USE ALL.
COMPUTE filter_$=(SCR_SEX=1).
VARIABLE LABELS filter_$ 'SCR_SEX=1 (FILTER)'.
VALUE LABELS filter_$ 0 'Not Selected' 1 'Selected'.
FORMATS filter_$ (f1.0).
FILTER BY filter_$.
EXECUTE.
DESCRIPTIVES VARIABLES=SCR_AGE
/STATISTICS=MEAN STDDEV MIN MAX.
USE ALL.
COMPUTE filter_$=(SCR_SEX=2).
VARIABLE LABELS filter_$ 'SCR_SEX=2 (FILTER)'.
VALUE LABELS filter_$ 0 'Not Selected' 1 'Selected'.
FORMATS filter_$ (f1.0).
FILTER BY filter_$.
EXECUTE.
DESCRIPTIVES VARIABLES=SCR_AGE
/STATISTICS=MEAN STDDEV MIN MAX.