Escolar Documentos
Profissional Documentos
Cultura Documentos
Statistics
• Methods for collecting , classifying,
summarizing & analyzing data
• Descriptive
– Frequency, Histogram, Measure central
Tendency, Measure of spread, Scatter plot
• Inferential
– Conclusion or generalization made about a
population from study using a sample
population
Variables
• Discrete (Nonparametric)
– Nominal: classified into groups with no
particular order or severity (Yes/NO)
• E.g., Sex, mortality, Disease State
– Ordinal: Ranked in specific order but no
consistent level of magnitude between
groups
• E.g., NYHA class, Trauma scores, Likert scales
1
Variables
• Continuous (Parametric)
– Interval: Data are ranked in specific order
with constant change in magnitude with
zero point arbitrary
• E.g., Fahrenheit Temperature
– Ratio: Like interval but with an absolute
zero
• E.g., Heart Rate, Age, Blood pressure
Descriptive Statistics
• Mean: Average
– Only for Continuous data
– Sensitive to outliers
• Median: Point where half of observations fall
below and above
– Used with ordinal & continuous data
– Insensitive to outliers
• Mode: Most common value
– For all data types
• Mode=Median=Mean for Normal Distribution
2
Standard Deviation
• Calculated to reflect range of samples
• Appropriate for normal or nearly normal
data
• Therefore can only use continuous data
Range SD SEM
3
Hypothesis Testing
• Null hypothesis (H0): No difference between
group (X=Y)
• Is used to determine if any observed
difference between groups is due to chance
alone
• If H0 is rejected = statistical difference
between groups
• If H0 is accepted = no statistical difference
(any difference due to chance)
Hypothesis Testing
• Choose correct statistical test based on:
– Type of data (nominal, ordinal, continuous)
– Study design (parallel, crossover)
– Presence of confounding variables
• Depending on statistics value H0 is
accepted or rejected
Decision Errors
H0 True H0 False
Accept H0 No Error (A) Type II Error (B)
True Positive (TP) False negative (FN)
Reject H0 Type 1 Error (C) No Error (D)
False Positive (FP) True Negative (TN)
4
Decision Errors
Power
• Is the probability of making correct decision and
ability to detect difference= 1 - β
• Analogous to β
• β = 1/α i.e. if α=0.05, β=20% & Power = 80%
• As Increase α Decrease β
• Increased by
– Increasing α
– Increasing n
– Magnitude of difference being studied (Δ)
– One tail versus two tail
• Decreased by
– Poor study design
– Incorrect statistical test
5
Statistical Significance
• Size of p-value not related to importance of
results
– Smaller p-values mean less likely chance explains
difference
• Statistically significance does not mean
clinically significance
• Lack of statistical difference does not mean
results are not important
– Lancet 2000;356:2139-43 Dopamine in ARF
Statistical Test
Type of variable Statistical test
6
Correlation
• Used to estimate strength of relationship between 2
variables
• R is correlation coefficient range -1 to +1
• -1 is perfect negative correlation or indirect
relationship
• +1 is perfect positive correlation or direct relationship
• Correlation is quantitative way to measure strength
of relationship OR
• Simply recognizes relation but does not imply
causation (chicken or egg)
Regression
• Regression PREDICTIVE, correlation is not!
• Math method to describe relationship with
goal to develop equation for prediction of one
variable from one or more variables
• Often use line regression where
• Y = MX + B
X is independent variable
Y is dependent variable
R2 is used for regression
• Intensive Care Med 2004;30:1537-43
• BIS XP and RASS R2 0.36 , p=0.011
• Means 36% of time BIS predicted RASS score
Types of Studies
• Case Reports
• Case Studies
• Case – Control Studies
• Cohort
– Retrospective
– Prospective
• Randomized Clinical Trials
7
Case Control Studies
• Advantages
– Causal influence on relatively uncommon
conditions
– Allows for smaller n than cohort
– Allows investigation of many causes
– Can be done inexpensively and fast
• Disadvantages
– Selection of control can be difficult
– Confounding and bias are concerns
Cohort Studies
• Can be prospective or retrospective
• Advantages
– Allows study of more than one disease and /or
exposures
– Less bias than case control
• Disadvantages
– Cost more
– Long time to conduct
– Larger n
– Bias can be introduced via outcome information
8
Relative Risks (RR) and Odds Ratios
(OR)
Disease
Yes No
Exposure Yes A B
No C D
9
Which is the Best Agent?
Application
• Absolute Reduction = % placebo - %
study drug e.g. 30.8%-24.7%
• Relative Reduction
= (placebo – study)/Placebo e.g.,
(30.8%-24.7%)/30.8%
• Number Needed to Treat (NNT)
= 1/(placebo-study) e.g., 1/(30.8-24.7)
Survival Analysis
• Studies entry into study and death
– Kaplan-Meier Curve
– Cox Regression Model AKA
– Proportional hazards regression analysis
• Hazard Ratio (HR)
– Use log rank formula to calculate slopes of curves
– Interpreted similar to OR or RR
– From Clin Infec Dis 2004;39:797-802
HR , 0.27; 95% CI , 0.09-0.78; p=0.011
10
Meta Analysis
• Sum greater that parts
• Advantages
– Combines results of many studies
– Greater statistical power
• Disadvantages
– Studies may not have same inclusion, hypothesis,
outcomes
– Publication bias
• May not fine studies that have not been published
Socrates
Resources
• Ann Emerg Med 1990;86-9.
• Ann Emerg Med 1990;309-15.
• Ann Emerg Med 1990;591-7
• Ann Emerg Med 1990;820-5
• Ann Emerg Med 1990;1054-9
• Ann Emerg Med 1990;1462-8.
• Graphpad.com
• Bmj.com/collections/statbk.index.shtml
11