Biostatistics Flashcards
What are the 2 types of variables and their subcategories?
Discrete, Continuous
Discrete: nominal and ordinal
Continuous: interval and ratio
What types of variables are discrete? Continuous?
Nominal and ordinal
Interval and ratio
Give an example of a nominal variable? Ordinal? (Counting variables)
Nominal: blood type, mortality, disease state (yes or no)
Ordinal: NYHA class 1-4, Ramsey scale
What are some examples of continuous variables (measuring)?
Interval: ranked in specific order with change in magnitude, no real zero —> temperature
Ratio: Like interval but with a real zero —> weight, time, SCr, pulse
How do we look at groups of numbers?
Look at measures of central tendencies: mean (average), median (50th percentile, or middle), mode (most frequently occurring)
For what type of variables is best to use the mean as the measure of central tendency?
Use ONLY for continuous data (interval and ratio) but it is sensitive to OUTLIERS
What type of variables is best used to find a median as measure of central tendency?
Used for ordinal or continuous data, good for SKEWED data as it is INSENSITIVE to outliers
What types of variables are best to find the mode as a measure of central tendency?
Used for both discrete or continuous data. *could have more than one mode
What is best measure for each variable if had to pick one?
Nominal: mode
Ordinal: median
Interval/Ratio (nominally distributed): mean
Interval/Ratio (skewed): median
What is Alpha in biostats?
Typically set at:
0.05 (significant p-value, 1 minus confidence interval)
Alpha = Probability of false positive test
What is a type I error vs a type II error?
Type 1 error FALSE POSITIVE finding: when a stat sig diff was found but NO true diff in population
Type 2 error FALSE NEGATIVE finding: when no stat sig diff was found but there IS a true diff in population
What is beta in biostats?
Beta is typically set at 0.1 or 0.2 = probability of a negative test
What is power in biostats?
Power = (1-beta) = The probability to detect a difference when a difference exists, often set at 80 or 90%
What is a confidence interval and when is it significant?
Typically 95% (if p=0.5 is significant)
CI that includes “1” for HR, OR, RR is NOT significant
CI that includes “0” for mean difference is NOT significant
How do you calculate relative risk ratio (RRR)?
RRR = % of outcomes in treatment — % of outcomes in control / % Outcomes in control
How do you calculate absolute risk ratio (ARR)?
% of outcomes in treatment / % of outcomes in control
How do you calculate hazards ratio (HR)?
Requires statistical program to calculate
How do you calculate number needed to treat (NNT)? And what does it mean?
1/ARR = for every # of patients treated, 1 additional outcome will be avoided or prevented
Or 1/ % of outcomes in treatment / % of outcomes in control
* can only be calculated if sig difference exists
How do you calculate number needed to harm (NNH)? And what does it mean?
1 / AAR = For every # of patients treated, 1 additional patient will have outcome
Or 1 / (% of outcomes the treatment / % of outcomes the control)
How do you calculate relative risk (RR)?
RR = [a / (a + b)] / [c / (c + d)]
Group A = a = treatment group and event happened, b = treatment group and event DID NOT happen
Group B = c = control group and even happened, d = control group and even DID NOT happen
*in prospective cohort or RCTs
How do you calculate Odds ratio (OR)?
OR = ad / bc
Group A = a = treatment group and event happened, b = treatment group and event DID NOT happen
Group B = c = control group and even happened, d = control group and even DID NOT happen
- in cross-sectional or case-control studies
What can the Confidence Interval (CI) tell us?
CI = provides info about a range in which the true value lies with a certain degree of probability, as well as about the direction and strength of the demonstrated effect
- a narrower CI indicates a more precise estimate, while a wider CI indicates a less precise estimate
- If the point of parity (1 or 0) is in the confidence interval on certain data types = there is no stat sig
I.e If CI includes 1 for HR, OR, RR = no stat sig or if CI includes 0 for mean difference (MD) = no stat sig
For which data types does it mean no statistical significance if CI contains “1”?
HR, RR, OR
For what data type does it mean no statistical significance if CI contains “0”?
Mean difference (MD)
For what p-values is the null hypothesis rejected and is considered statistically significant?
When p < alpha (usually 0.5)
What is a null hypothesis?
There is NO difference between groups for outcome X
What is the order of causal relationship in types of study design starting with the least relationship?
- Case Report
- Case Series
- Case-control study
- Cohort study
- RCT
What are the types of RCTs?
Designs: Parallel, Cross-over, Factorial
*RCT’s minimize bias via randomization
What are the types of observational trials?
Case-control, prospective cohort, retrospective cohort
What is a factorial design RCT?
Evaluates two interventions compared to a control
As long as no interaction, advantage is to possible do two experiments at one time
What is a cross-over design in RTCs?
Subjects are randomized to sequences of treatments (A then B or B then A)
Uses the patient as own control
Often a “wash-out” period is used to avoid a “carry over” effect
Only appropriate for conditions that are expected to return to baseline levels at the beginning of the second period
What are the 4 types of observational studies?
Cohort: consists of group of individuals from a well-defined population followed over time to observe what happens to which groups and when
Prospective: group of subjects/cohorts identified and evaluated as it occurs, Time-sensitive
Retrospective: identified in the past and then evaluated for event, Medication use evaluations, time and cost efficient, cannot completely remove bias or confounders (need special programs)
Case-control: group os subjects with disease or condition are identified, suitable control group identified without the condition, frequency of exposure or risk factor compared in cases and controls, decreases confounding factors that might explain differences between groups
In order to calculate NNH/NNT, what do you assess first?
First you must look at P values to see if significant. If p >0.05 not significant and cannot calculate NNH/NNT
If data is given in percent, must convert to decimal 35 = .35 etc
Formula is 1/ARR or 1/ (% of outcome in treatment — % of outcome in control) *always round to greater whole number
If answer is positive = NNH, if answer is negative = NNT