Exam 1 Flashcards
Efficacy
How beneficial a specific intervention, procedure, regimen, or service is under ideal conditions (lab)
Effectiveness
How beneficial a specific intervention, procedure, regimen, or service is when deployed in the field in routine circumstances (real world)
Heirarchy of evidence quality
- Systematic reviews and meta-analysis
- Clinical trial in humans (all criteria met)
- Longitudinal cohort studies
- Case-control studies
- Human trial without concurrent controls
- Descriptive and cross-sectional studies
- Case reports & case series
- Personal opinion, subjective impressions, anecdotal accounts
Criteria for clinical trial in humans
- Sufficient & appropriate subjects
- Subjects randomly allocated
- Use placebo in double-blind
- Tested agent is closely assessed
- Reliability of measurements (calibrate machines)
- Sufficient duration
- Minimal loss to follow-up (Has to be random loss)
- Specific endpoints have to be clearly defined in advance
- Statistical analysis is appropriate
Prospective study
Before disease occurs in subjects
Retrospective study
After disease has occurred in subjects
Cohort study
Separate subjects based on exposure, then find the prevalence of disease in each group
Case-Control study
Separate subjects based on whether they have a disease or not, then find exposures of each group
Cross-sectional survey
Disease and exposure assessed at the same time in a very large population (Snapshot)
Not always possible to distinguish whether exposure preceded or followed disease
Case report
Describes experience of a single patient or groups of patients with similar diagnosis
May lead to formulation of a hypothesis
Case series
A collection of individual case reports
Investigating activities of infected individual leads to hypothesis which is tested by comparing those with and w/o disease in a later case-control study
Nominal variables
A scale based on categories
Ex: gender, political party, marital status
Ordinal variables
A scale based on classification of an observation according to its relationship to other observations
Ex: Poor-fair-good rating scale
Crossover study
A longitudinal study in which subjects receive a sequence of different treatments (or exposures).
Interval variables
A scale based on equal units of measurement; distance between any 2 numbers is of known size
Zero point is arbitrary
Ex: Fahrenheit and centigrade temperature scales
Ratio variables
A scale based on equal units of measurement and a true zero point at its origin
Ex: Mass, time
Population vs Sample
Population = Whole group of people (mean=µ, SD-σ) Sample = Part of the population (mean=x̅, SD=S or SD)
Mode
Most frequent measurement
Most useful with nominal scale, but may be used with any
Median
Measurement right in the middle when put in order
Most useful with ordinal, but may be used with higher order scales
Insensitive to extreme values
Mean
Arithmetical average: sum of measurements divided by total # of measurements
Most useful with interval or ratio measurement scales
Range
Difference between largest and smallest measurements
Interquartile range
X25th percentile - X75th percentile
Variance (s^2, MS)
Unbiased estimate of population variance
Sum of squares/total # in sample Subtract 1 (degrees of freedom)
Sum of squares
Subtracting sample mean from each of the measurements, squaring the result to eliminate negative #s, then add all of these
Degrees of freedom (df)
N - 1 (N = # of variables) # of independent observations (# of observations that are "free to vary")
Standard deviation (SD)
Positive square root of variance
Coefficient of variation (CV)
100 x SD/Mean
The comparison of the size of the standard deviation to that of the mean
Standard error of the mean (SE or SEM)
Measurement of how good we think mean is
SD/Square root of sample size (N)
Bigger sample size = lower standard error
Difference between SD and SE
SD is used to measure variability of individual subjects/entities around a sample mean (variability of individual measurements from the mean)
SE is used to assess how accurately a sample mean represents a population mean (variability of the mean if you repeated the experiment many times)
Bar charts
Categorical data (Nominal, ordinal) Labels under each bar
Histograms
Bar charts for continuous data
Axis labels not necessarily centered under bars
Box-whiskers plots
Box represents middle 50% of data
Line represents median
Whiskers extend to remaining data (circles/asterisks represent outliers)
Dot plots
Continuous data in groups
Shows relative location & spread of each group
Funnel effect
Only see half the data in journals because funders will not publish negative findings that go against their products
Impact factor
Total # citations to articles in journal / Total # of articles published
Types of papers published in journals
(1) Research reports
(2) Reviews of the literature to summarize knowledge in a particular area
(3) Commentaries
Difference between statistical and clinical significance
1% change in plaque index is statistically significant but not clinically significant at all
Prevalence of a disease
How many people had it at a specific point in time
Booleans
AND - searches for another word in addition (narrows)
NOT - doesn’t include a certain word (narrows)
OR - search 2 words separately (broadens)
“” - search exact phrase (narrows)
PICO
P- Patient problem
I- Intervention, prognostic factor or exposure
C- Comparison
O- Outcome
Types of hypotheses
Research - What we are trying to prove; prediction
Null - a mathematical statement of no difference
Alternate - covers all that the null doesn’t
Directional (one-tailed) hypothesis
Trying to prove one scenario (guys smarter than girls)
Non-directional (two-tailed) hypothesis
Allowing either outcome to be possible (guys or girls could be smarter) This is what we will use in class
Dependent variable
• The variable we measure and compare
(intelligence)
• Sometimes called the outcome variable or
response variable
Independent variable
• The variable we manipulate or the “grouping
variable” (gender)
• Sometimes called a predictor variable
Type I Error (α or p-value)
Probability of rejecting null hypothesis when it is true
Type II Error (ß)
Probability of accepting null hypothesis when it is false
Methods to increase power
- Increase type I error willing to tolerate (ß + power = 1)
- Increase sample size
- Increase deviation from null hypothesis willing to tolerate
- Decrease variability
- Use a directional alternate hypothesis if appropriate
- Use most efficient/powerful statistical test
Reject H0 if and only if
p ≤ α
Central limit theorem
In random samples of N observations drawn from a population, the sample means will be approximately normally distributed
Z score
z = x̅i - µ
—————-
σ / SqRt(N)
Z score distances
68% = within 1 SD of mean 95% = within 2 SD of mean 99% = within 3 SD of mean
t-distribution
Used when sample size is small (<30)
Calculation is exact same as z score
Has smaller mound & thicker tails
Confidence interval for a mean
Consists of lower confidence bound & upper confidence bound with the population mean contained in the interval (1-α)% of the time
