1
Basic Biostatistic Application in
Research of Anesthesia
Chan Wei-Hung MD
Department of Anesthesiology National Taiwan University
How to Conduct a Study?
Experimental study: best for cause-effect relationship determination
Observational study: only associations are made; not cause-effect relationship
Retrospective Prospective
3
Experimental Study
(Clinical Trial)Patients are assigned into different groups, receiving different intervention in each group. Random, blind, well-controlled (control over other confounding factors) design is key to success.
Power of measurement and cause-and-effect determination are also vital to success.
Observational Study
Descriptive study (case report/series): no comparison is made
Case-control study: patients with an outcome (case) are analyzed along with patients without the outcome (control). ESPECIALLY PRONE TO SAMPLE SELECTION BIAS!
5
Case Control Study
Parturient
C-section NSD
With
Epidural EpiduralWith Without
Epidural EpiduralWithout
Cohort Study
Parturient
With Epidural Without Epidural
C/S NSD C/S NSD
7
Clinical Trial
Parturient
Epidural
Random Grouping
Analgesics Normal Saline
C/S NSD C/S NSD
Attention for Observational Study
Since the cause-effect relationship can not be established in this kind of study, if you want to do such a study, please notice that: The sample size should be big.
9
Random Assignment
Simple random sampling with a random numbers chart
Number of patients can be balanced within a
11
Restricted Randomization
Group A: 20 patients Group B: 20 patients Frame size: 10 patients
No. of A and B are balanced within every 10 patients.
p, α and β Error
p value: the probability that one will wrongly conclude that there is a difference between groups.
Type I error: also called α error, false-positive error. p value
Type II error: also called β error, false-negative error
13
Type II Error (β Error)
False-negative error ( p>0.05 in the presence of difference)
When p>0.05, it is difficult to determine
between lack of true difference or inability to detect the difference.
Most common problems: insufficient sample size, bias in selection, confounding factors
Statistical Power
The ability to detect an effect when it is present. Equal to 1 – false negative error (1-β)
A statistical power around 80% (β<0.2) for a reasonable effect
15
How to Increase the Power?
1. Increase the size number
2. Reduce variation between measurements 3. The effect of intervention should be
Determination of Sample Size
In a t-test N = 2[
(Zα- Zβ) * SD Mean 1 – Mean 2]
2 SD: 正常值 ( 對照組 ) 的標準 差 Mean 1 – Mean 2: 預估偵測到的差別值17
Example in Size Number Determination
Onset of two muscle relaxants will be
compared. You wish to detect a difference of 10 sec. The standard variation of the
onset time is about 5 sec (according to the literature). You desire a p=0.05 and a
statistical power of 80%. The sample size of each group would be how many ?
Example in Size Number Determination
2 x [(1.96+0.825)x5/10]2 =3.87; about 4 in
each group
If you want to detect a difference of 5 sec: 2 x [(1.96+0.825)x5/5]2 =15.5; about 16 in
19
Noncentrality Parameter (φ)
You can also determine the sample size by computing φ and look up the table.
Φ=δ/σ
(the difference of effects / standard deviation of population)
Critical Reviews of the Results
When you want to say there is an effect of intervention give us the p value
(chance of false-positive error)
When you want to say there is no effect of intervention give us the power
Epidural Analgesia Enhances Functional Exercise Capacity and Hea lth-related Quality of Life After Colonic Surgery
25
Determining the Test (I)
What kind of variables are they?
1. Numerical variable 2. Ordinal variable
3. Categorical variable (Nominal)
How many groups are there?
Determining the Test (II)
Are they “normal distribution”?
Parametric vs. nonparametric methods.
T-test Mann-Whitney U test ANOVA Kruskal-Wallis test
27
Determining the Test (III)
Measurements are taken from the same patient for more than one time (before and after treatment); you should use
Paired t-test
Determining the Test (IV)
Common data are analyzed when they are completed (all the measurements are
finished); but there are some studies that data input are still ongoing (5-year analysis for two treatment for lung cancer); basically for this kind of “unfinished studies”.
29
An Example for Survival Analysis
Patients received meperidine or hydrom orphone in the POR.
31
Trick for Study Design
Thorough examination of past similar studies (sample size, statistical
methods, items of measurements --- you can apply them to save you from brain drainage and avoid fatal errors!)
Central Belief
Biostatistics is not a hindrance but an ai d for data analysis.
As long as you have an idea for study, b iostatistics should not be the excuse tha t you cannot finish the study.
33
THE END
Paired t-test
When the two groups of data are obtained from the same subject
(repeated measurements from a subject under different conditions), paired t-test should be used.
35
Wilcoxon Signed Rank Test
In a repeated measurement, the differen ces are usually not “normally distributed ”.
A Wilcoxon signed rank test should be used in the case.
Analysis of Variance (ANOVA)
Comparison of variation conditions of different groups.
37
Screening Test Evaluation
The effectiveness of diagnostic or prognostic tests is assessed.
Sensitivity and specificity are explored in such studies.
Sensitivity and Specificity
Disease Positive Disease Negative
Test Positive A B Test Negative C D Sensitivity = Specificity = A/(A+C) False-negative = 1- sensitivity D/(B+D)
Sensitivity Specificity
False-negative positive
False-Palm print grade>0 1.00 0.57 0 26
Mallampati >1 0.41 0.80 13 12 Mallampati >2 0.50 0.98 21 1 TMD <6 cm 0.14 0.9 19 6 Head extension<35° 0.50 0.70 11 18 BMI > 27 0.23 0.97 17 2 DM > 10 yrs 0.91 0.67 2 20 DM type 0.45 0.51 12 30
Different Criteria to Predict Difficult Intubation in DM Patients