BIOSTAT & EPIDEMIOLOGY Flashcards
What information do you get from an ecologic (correlational) study?
Ecologic studies give population-level information, not individual-level information. Applying population-level information to an individual level = ecologic fallacy bias.
Kappa statistic
quantitative measure of inter-rater reliability (aka inter-rater concordance). reflect extent to which inter-rater agreement represents an improvement on chance agreement alone. Kappa values range from -1 (perfect disagreement) to +1 ( perfect agreement), with kappa = 0 suggesting agreement due to chance. kappa<0 suggests less than chance agreement, kappa > 0 greater than chance agreement.
Kaplan-Meier Curve
depicts probability of survival at various time points during the study. calculated based on proportion of subjects who are alive (“at risk”) at a given time. the event-free survival rates of 2+ groups can be compared. they are statistically different only if p-value is <0.05
what do overlapping stand error of measurement (SEM) bars suggest in general
overlapping SEM error bars suggest a non-statistically significant difference .
Case fatality rate
proportion of people with a particular condition who end up dying from the condition
Attack rate
proportion of people in whom an illness develops out of the total population at risk from the disease
median survival time
measure of prognosis when studying big groups. defined as the length of time that it takes for half of the study population to die
Quality adjusted life years & Disability adjusted life years
both are burden of disease measures evaluating the impact of specific diagnoses/treatments on individual patients of the economic impact of health interventions on populations . time-trade-off is commonly used to QALY calculations, while years of life lost and years lived with disability are used for DALY calculations . DALYs represent a loss to be minimized and reflect the difference between the current situation and an ideal situation
Meta-analysis
conducted by pooling data from several studies to increase statistical power (ability to detect difference in outcome of interest between groups, if such a difference exists). meta-analysis increases sample size and therefore increases power. major disadvantage: concomitant “pooling” of the biases and limitations of individual studies into one analysis
Standardized Mortality Ratio
SMR=observed # of deaths divided by expected # of deaths
NNT
NNT the number of patients who need to be treated in order to prevent one add’l bad outcome. NNT is the inverse of absolute reduction (ARR). ARR=control group event rate-experimental group event rate. NNT = 1/ARR
Likelihood Ratio
-assesses value of diagnostic test.
Positive LR
LR+ =sensitivity/ (1-specificity)
probability of patient with dx testing +/probability of pt without dx testing +
Negative LR
LR- = (1-sensitivity) /specificity
probability of patient with the dx testing neg / probability of patient without the disease testing neg
- LR range 0 to infinity. LR> 1 suggest disease present. higher the LR more likely dx presence. LR<1 argues against disease. in general smaller LR less likely dx
- LRs are independent of prevalence
Standardized Incidence Ratio
used to determine if the occurrence of cancer in a small population is high or low relative to the an expected value derived from a larger comparison population. SIR=observed cases/expected cases
Standard deviation
1 SD: 68%
2SD: 95%
3 SD: 99.7%
Standard Error of the Mean
- SEM is a measure of how tightly grouped a data set is
- smaller the SEM the more precise the data
z-score
shows how far above or below a score is compared to the mean
confidence intervals
- indication of precision of data
- if CI of outcome crosses 1, the results aren’t significant
- narrow 95% CI is more precise
- wide CI may be due to insufficient power
- if 95% CI reflects lack of statistical significance then corresponding p-value >.05; if 95% CI is significant then p < .05
T-score (t-test) & Analysis of Variance (ANOVA)
both are used to assess different groups of data between different sets of data that are in more than one group.
- T-test used for 2 groups of data. t-test answers questions “are means between two groups different?”
- ANOVA used when there are 2+ groups
chi-square test
answers questions “are these groups related (or not)?”
compares multiple groups and indicates whether or not they are statistically significant. used for categorical data.
Cohort Study
- observational, prospective; see what happens to groups of patients with certain exposures/underlying illnesses over time. no intervention
- results are assessed with relative risk calculations
Relative Risk
-equation
-looks at risk of dx based on previous exposure to potential danger
RR=risk of exposed group / risk of unexposed group
RR = a/(a+b)
———–
c/(c+d)
RR>1 indicates increased risk in the group in the numerator. RR<1 indicates decreased risk in the group p in the numerator. RR=1 (null value) means no difference in risk between the ferrous. CI that excludes null value reflects statistically significant results.
Attributable Risk Percentage
measures excess risk & estimates proportion of dx among exposed subjects that is attributed to exposure status
ARP = risk in exposed -risk in unexposed)
————————————————–
risk in exposed
= RR-1
——–
RR
Population Attributable Risk
excess risk in total population
PARP = risk in population - risk in exposed
—————————————————
risk in total population
OR
PARP = (prevalence)(RR-1)
——————————
(prevalence)(RR-1) + 1
Case Control
- retrospective study looking for the odds of previous exposure on the development of a rare dx manifestation
- subject to recall bias which leads to misclassifcation of exposure
- assessed with odd ratio
Odds Ratio
- assesses case control studies
- starts with those that have disease and looks at the chance of them having had exposure in the past
OR = cases exposed / cases unexposed
———————————————————
controls exposed / controls unexposed
OR = a/c = ad
——- ——
b/d bc
null value for OR = 1 . if CI includes 1 then result isn’t statistically significant
- OR > 1 exposure is associated with higher odds of the outcome. OR < 1 means that the exposure is associated with lower odds of the outcome.
- odds that an outcome will occur in the presence of a particular exposure divided by the odds that the outcome will occur in the absence of that exposure
Berkson Bias
hospitalized patients as trial subjects instead of general population . minimize by randomization
Hawthorne Effect
those being studied know they are being watched for the effect of a drug or intervention . minimize this by using placebo control and double blinding
Lead-Time bias
in this bias, early detection is confused with increased survival based on treatment . ex early detection
Null Hypothesis & p-value
null hypothesis = intervention no better than random chance
reject null hypothesis = intervention works
p
post market surveillance
if in RCT adverse events occurred and p > 0.05 (insignificant) then need to do post market surveillance for new drug/intervention. may not have been enough power.
type 1 error
- false positive result
- rejecting null hypothesis
- ex. saying drug works when it really doesn’t
- called alpha error = p-value
type II error
- false negative result
- called beta error
- power = 1-beta, power and better inversely related
Receiver Operating Curve
plot sensitivity as function of 1-specificity at various cut off points for a test. a negative result on highly sensitive test helps r/o disease
Sensitivity
likelihood that TEST will detect all ppl with dx
a/a+c
TP/TP+FN
specificity
likelihood that ppl w/o dx are correctly identified as dx negative
=b/b+d
=TN/TN+FP
Negative Predictive Value
Positive Predictive Value
both change with prevalence. need to know pre-test probability.
Negative Predictive Value:
probability that dx is absent given negative test result
NPV = d/c+d = TN/TN+FN
Positive Predictive Value
PPV = a/a+b= TP/TP+FP
Absolute Risk Reduction (ARR)
Number needed to treat (NNT)
ARR = % decrease in risk of death or dx from treatment compared with 100% of the people in the population
NTT = 1/ARR
in general, RRR is larger number & often used to exaggerate effectiveness of medication
Attributable Risk (AR)/ARI
NNH = 1/AR
Test for heterogeneity
- useful for meta-analysis or comparing trials
- commonly used is Q test. no heterogeneity is p >0.05 and the I2 index (25%, 50%, 75% are traditional considered cutoffs for low, moderate, high heterogeneity respectively)
Hazard Ratio
measure of effect used in survival analysis (time-to-event)
null value HR= 1.00–>no difference
HR< 1.00–>protective
HR > 1 –>detrimental
Prevention Strategies primordial primary secondary tertiary quaternary
primordial: prevention of risk factors themselves
primary: action taken before patient develops dx
secondary: axn halts/delays dx progression/complications
tertiary: limit impairments/disabilities from advanced dx
quaternary: lim consq’s of unnecessary/extra intervention
Correlation coefficient (r)
indicates positive/negative direction of association between 2 variables. closer r is to margins [-1, +1] the stronger the association
intention-to-treat analysis
- used in RCTs
- participants analyzed in groups to which they were randomized, regardless of whether they received/achieved the allocated intervention and regardless of if they withdrew
Sensitivity Analyses
refers to repeating primary analysis calculations in a study by modifying certain criteria or variable ranges to determine whether such modifications significantly affect the results initially obtained
Non-inferiority Trials
goal: prove new drug isn’t unacceptably worse than comparator by a given margin
-----------------0------------------------ #1 vertical line at zero: right of 0 = non-inferior & superior #2 vertical line for inferiority. right of line non-inferior, left of line not non-inferior
Funnel Plot
triangle is centered on a summary treatment effect (pooled estimates of ORs in log) with y-axis being standard error. w/o bias 95% points should be inside triangle. should be symmetric
publication bias and heterogeneity–>asymmetry
