Statistics Flashcards
type of study where investigators evaluate characteristics of a certain patient population without performing any interventions?
observational study
List the types of studies in order of increasing strength of evidence that can be obtained
case report, case series, case-control, cohort, randomized controlled double-masked trial, meta-analysis
Describes a finding in regard to a single patient
case report
Investigate a group of patients without standardizing the patient information
case series
What type of study: Retrospectively analyze group of patients with CNV 2/2 AMD and evaluate the benefit of laser v PDT
case series
Compare a group of people with a disease (cataracts) to a group of patients without the disease, and compare the proportions of patients in each group who were exposed to a specific risk factor (smoking)
case-control
Compare a group of people with a specific risk factor (diabetes) compared a group without, and follow the groups to determine relative proportions that develop a disease (heart attack)
cohort study
Researchers select pateitns with and without a disease from a current database and obtain history by survey or review of medical records, thus obtaining data about exposure to risk factors right away
case-control
which type of study design does recall bias have the greatest effect?
case-control
correlates exposure and risk factors with the prevalence of disease without knowing the timing or sequence of exposure and disease development
cross-sectional study
What type of study: collect blood from a sample of patients and compare cholesterol levels to lens status (phakic, pseudophakic, aphakic)
cross-sectional study
randomly assigns patients to different treatment groups
clincal trial
definition of and equation for sensitivity
proportion of patients with a disease who are determined to have the disease by a particular test
sens = true positives / (true pos + false neg)
definition of and equation for specificity
proportion of patients without a disease who are not determined to have the disease by a test
spec = true neg / (true neg + false pos)
definition of and equation for positive predictive value
proportion of patients that are determined to have a disease by a particular test who actually have the disease
PPV = true pos / (true pos + false pos)
definition of and equation for negative predictive value
proportion of patients with a negative test result that actually do not have a disease
NPV = true neg / (true neg + false neg)
low disease prevalence leads to artificially high ___ and artificially low ___
high false positives
low positive predictive value
x and y axis of receiver operating curve
x: 1 - specificity
y: sensitivity
what on a receiver operating curve correlates to the degree of diagnostic precision of a test?
area under the curve
what is the likelihood ratio for a positive test? for a negative test? which is more important if you are trying to diagnose a disease, and which is more important to rule out a disease?
positive: sensitivity / (1 -specificity)
- high positive LR important to diagnose
negative: (1 - sensitivity) / specificity
- low negative LR important to rule out a disease (the smaller the better)
how do you calculate the post-test probability of disease?
pretest probabability X positive likelihood ratio
which of the following is most useful for evaluating the probability of glaucoma in a glaucoma suspect: ROC, likelihood ratio, sensitivity, specificity
likelihood ratio. the other tools are binary (either you have glaucoma or you do not) and do not take into account pretest probability, which is important to consider in a glaucoma suspect
RNFL is measured by OCT, and only for abnormal results a scanning laser perimetry is performed to measure RNFL. What type of study process is this, and what is a pitfall of this study design?
This is an example of using tests in series.
Pitfall: you may think you can multiple the original pretest probability by the product of the two likelihood ratios to get the final post-test probability, however the fact that these two tests are not independent is a confounder (they both measure RNFL, just by different mechanisms). Only tests without correlation can be performed reliably in series
Type of study process: two tests performed at the same time, and a positive result of either one is considered diagnostic. Pitfall?
Tests in parallel. Pitfall: both tests must have high specificity for this to be useful
calculate risk difference and relative risk:
case: 30% died
control: 70% died
difference = 40%, relative risk = 3/7
calculation for odds and odds ratio
odds: ratio of favorable to unfavorable outcomes (odds of rolling a 5 or a 6 is 1/2)
odds ratio: odds of a subject with the disease having the exposure divided by the odds of the subject without the disease having the exposure
what happens to the odds ratio as disease prevalence approaches zero?
odds ratio approaches relative risk
what is the kappa statistic, and what cutoff value indicates an excellent kappa?
the degree of agreement between 2 or more individual methods or entities, accounting for agreement via chance alone. 75% and above is generally considered to be excellent
