Biostats Flashcards
What is frequency of disease in population?
prevalence, incidence, and attack rate
How well does test differentiate sick from healthy?
sensitivity and specificity
Of those in population who test as sick or healthy, how true is that?
Predictive value
What is impact of medicine/treatment?
Risk reduction/increase
NNT, NNH
Prevalence
helps understand disease burden or extent of health problem
= # of people with disease at specific point/# of people AT RISK for illness at same point in time
Period prevalence
during a period of time (specific)
Lifetime prevalence
over course of a lifetime
Incidence
helps understand risk of specific health event
= # of NEW people with disease during time period/# of people at risk for illness during time period
if you already have disease -> not at risk anymore
Cumulative incidence
total number reported over time
Attack Rate
type of incidence used during short period of time (specific exposures/outbreaks)
= # new cases/#exposed
Secondary Attack Rate
= # of new cases/(# exposed - primary cases)
- measures person-to-person spread of disease after initial exposure
What affects prevalence and incidence?
Duration of illness (higher prevalence) Number of new cases (higher prevalence) Ill people coming in (higher prevalence) Healthy people leaving (higher prevalence) Recovery/death (lower prevalence) Prevention (lower incidence) Changes in diagnostic criteria
Relationship between prevalence and incidence
Chronic illness –> prevalence = incidence x average duration
Acute illness –> prevalence = incidence
Sensitivity
probability that diseased person will be ID correctly (true-positive)
= true positives/ total # ill people (TP and FN)
True positives = ill people ID as ill
False negative = ill people ID as healthy
Specificity
probability that well person will be ID correctly (true-negative)
= true negative/ total # well people (TN and FP)
True negative = healthy people ID as healthy
False positive = healthy people ID as ill
Highly sensitive test
ID most of all possible disease cases
- will over-diagnose some people without disease
Highly specific test
ID most or all well people
- will under-diagnose some people that do have disease
Predictive value
probability that test will give correct diagnosis
- depends on sensitivity and specificity
- will vary from population to population (depends on prevalence of disease in population)
- looking at rows of 2x2 table
Positive predictive value
probability that person who tests positive for disease truly has it
= TP/TP+FP
Negative predictive value
probability that person who tests negative for disease truly is healthy
= TN/TN+FN
Predictive value with high disease prevalence
higher PPV
lower NPV
Predictive value with low disease prevalence
lower PPV
higher NPV
Risk reduction/NNT
relevant when comparing effects of RCT
- interest in understanding risk of treatment vs no treatment
- what is frequency of bad outcomes in group being treated compared to group not being treated?
Randomized control trials
1 treatment group and 1 control group
- groups can respond positively or negatively
Control Event Rate
proportion of control group participants who have bad outcome after “treatment” (placebo)
Experimental Event Rate
proportion of treatment group participants who have bad outcome after treatment (drug)
Absolute Risk
probability of developing disease or undesired outcome
Absolute Risk Reduction
control event rate is HIGHER than experimental event rate
CER - EER > 0
Absolute Risk Increase
control event rate is LOWER than experimental event rate
CER - EER < 0
NNT
number of patients who need to be treated to get 1 additional patient a favorable outcome
NNT = 1/ARR
NNH
number of patients who, if treated, would result in 1 additional patient being harmed
NNH = 1/ARI