Positive vs Negative predictive value
how does sensitivity and specificity affect these values?
PPV = probability of having a disease given a positive result, higher specificity incerases PPV of the test
PPV = TP / (TP+FP)
NPV = probability of not having a disease given a negative result, higher sensitivity increases NPV of the test
NPV = TN / (TN+FN)
likelihood ratio (LR)
expression of sensitivity and specificity that can be used to assess the value of a diagnostic test
positive LR - value of a positive test; probability of a patient with disease testing positive / probability of a patient without disease testing positive
negative LR - value of a negative test; probability of a patient with disease testing negative / probability of a patient without disease testing negative
SMALLER LR = LESS LIKELY DISEASE PRESENCE
interpreting confidence intervals when comparing two studies
wider confidence interval = less likely to be significant
if the interval includes 1 = study is NOT statistically significant (likely due to smaller sample size)
what is the best test to use when the mean values of continuous variables in several groups are compared?
ANOVA - analysis of variance - gives an F statistic that can be used to obtain a p-value.
best test to use to study the association between 2 categorical variables when the number of observations is small (i.e. n is small)
best test to use in a study where patients serve as their own control (ie success/failure before and after treatment in the same subjects)
(kind of like a t-test but an added level of comparison in the same subjects)
best test to use when comparing two paired means (i.e. patients serve as their own control)
in assessing whether or a test is great for ruling out a malignancy, what would be the best statistical measure to look at?
to exclude malignancy, look at tests w/ high sensitivity, as negative results rule out a diagnosis (SnNOut)
(sensitivity is the ability of a test to correctly identify patients w/ disease, highly sensitive tests are more likely to detect a disease and the least likely to give false negative reuslts; most appropriate for screening)w
what is a likelihood ratio?
expression of sensitivity and specificity that is use to assess the value of a diagnostic test.
+ LR = sens / (1 - spec) = prob of patient w/ dz testing + / probability of patient w/o disease testing +
- LR = (1 - sens) / spec = prob of patient w/ dz testing - / probability of a patient w/o disease testing -
a test with LR >1 suggest disease presence; the higher the LR the more likely the disease presence, and vice versa
does not change as the disease prevalence changes
When is OR calculated?
in case-control studies to compare exposure of patients (case) w/ disease to the exposure of patients w/o disease (control)
Definition of PPV? NPV?
PPV = probability that a + test result will correctly identify an individual with disease
NPV = probability that a - test result will correctly identify an individual without disease
when is relative risk calculated?
calculated in cohort studies; participants; patients are followed over time to assess a risk factor for developing a given disease.
RR is the probability of an outcome occuring in the exposed group compared to the probability of it occuring in the non-exposed group.
Verification Bias - aka workup bias - occurs when a study uses gold standard testing selectively in order to confirm the preliminary testing; can result in overestimates/underestimates
Contamination Bias - when the control group unintentionally receives the treatment or the intervention, thereby reducing the difference in outcome between the control and treatment group
Observer Bias - when an observer responsible for recording results is influenced by prior knowledge about participants or study details
Selection Bias - results from the manner in which participants are selected or LOTFU
Susceptbility Bias - experimmental and control groups differ from a prognostic standpoint, possibly due to unforseen confounding variables
number needed to treat (NNT)
ARR = control rate - experimental rate
NNT = 1/ARR
Type I Error
Type II Error
Type I - when the null hypothesis is rejected when it is actually true (i.e a study finds statistically significant difference between two groups when it did not truly exist)
Type II - when one rejects the alternate hypothesis (or accepts the null hypothesis) when the alternate hypothesis is actually true; power of the study increase w/ sample size. Therefore, as a sample size increases and therefore power increases, the chance of a type II error occuring decreases
what would explain the reason for reporting median values as opposed to mean values?
in strongly skewed distributions, the median is a better measure of central tendency (than the mean).
what does it mean if a study comparing drug + placebo have a RR of particular outcome that is close to 1?
if the RR of the particular event is close to 1 when comparing drug to placebo, it means that the patients taking a low dose of the drug have almost an equal risk of the particular outcome compared to patients taking the placebo.
what is the basic premise of intention-to-treat (ITT)
study participants should be analyzed in the groups to which they were randomized, regardless of whether or not they received or adhered to the allocated intervention and regardless of whether they withdrew from treatment
goal of ITT is to preserve randomization (i.e. it's used to avoid the effects of cross-over, drop outs, etc)
what is the stanardized mortality ratio (SMR)?
adjusted measure of the overall mortality; typically used in occupational epidemiology. Compared to the standard population
mortality is typically adjusted for age
SMR = observed number of deaths/expected number of deaths
measure eof excess risk; estimates the proportion of disease among exposed subjects that is attributed to exposure status
AR = (risk exposed - risk unexposed)/risk exposed
measure of effect used in survival time or time-to-event analysis; likelihood of an event occuring in a treatment group relative to the control group.
HR <1 = event less likely to occur in a treatment group than the control group
HR >1 = event more likely to occur in a treatment group than the control group
null value for HR usually represents a baseline risk of the event
case fatality rate
how does it differ from mortality rate?
case fatality rate = proportion of people with a particular condition who end up dying from the condition
ex: people who developed DVTs who ended up dying from DVTs
mortality rate = probability of dying from a particular disease in the general population
ex: people who died from DVTs in the general population (much smaller % than CFR)
measure of association between an exposure and an outcome; represents the odds that an outcome will occur in the presence of a particular exposure.
OR > 1 = exposure associated w/ higher odds of the outcome
OR < 1 = exposure associated w/ lower odds of the outcome
OR = (a/c) / (b/d)
standardized incidence ratio
observed # of cases / expected # of cases
measure used to determine if the occurence of X in a small population is high or low relative to an expected value derived from a large comparison population
SIR = 1 implies that the observed incidence of cancer is equal to the expected value
SIR >1 implies a higher observed incidence than expected
number needed to harm
indicates how many patients need to be exposed before a harmful event occurs in one person
1/absolute risk increase (ARI)
ARI = ∆ between exposed and non-exposed groups (aka attributable risk)
number needed to treat (NNT)
number of patients who need to be exposed/treated with an intervention to prevent 1 bad outcome
NNT = 1/ARR
Correlation coefficient (r)
ranges from -1 to +1 (null value is 0 and denotes no association)
the closer the r value is to its margins (-1, +1), the stronger the association
1 Standard Deviation equals what % of all observations?
1 SD = 68%
2 SD = 95%
3 SD = 99.7%
Difference between primary prevention and secondary prevention?
Primary prevention = action taken BEFORE a patient develops the disease and is targeted at preventing occurrence of the disease itself
Secondary prevention = action that halts/delays the progression of a disease at its initial stage and prevents complications
case control, case series, cross-sectional
group of case reports regarding individual patients w/ similar clinical manifestations or received similar treatment
individuals are assessed at a specific time point (i.e. snapshot) to determine whether or not they have a certain risk factor and a certain disease of interest to compare disease prevalence
study that categorizes patients w/ and w/o a specific disease and compares the risk factor frequency between the two groups (i.e. looks back in time)
case series = group of case reports regarding individual patients w/ similar clinical manifestations or received similar treatment
cross-sectional = individuals are assessed at a specific time point to determine whether or not they have a certain risk factor and a certain disease of interest to compare disease prevalence
case-control = categorizes patients w/ and w/o a specific disease and compares the risk factor frequency between the two groups (i.e. looks back in time)
what is the kappa statistic?
quantitative measure of inter-rater reliability (aka inter-rater concordance)
kappa = 0 = agreement due to chance
kappa <0 = less than chance agreement (possibly intentional disagreement)
kappa >0 = greater than chance agreement
what is a factorial study design? cross over study? how does this differ from cross-sectional study?
factorial study = experimental study that uses ≥2 interventions and ALL combinations of these interventions us
cross over study = type of experimental study design in whihc patients are exposed to different treatments or exposures sequentially.
cross-sectional = type of ovservatiaonal study in which specific population or group is studied at one one shouldshippe
What is a factorial study?
type of experimental study design that utilizes ≥2 interventions and all combinations of these interventions
What is a funnel plot useful for?
assessing publication bias
there is a triangle centered on a summary estimate of the treatment effect with its sides corresponding to standard errors. IF there is NO bias, 95% of the studies should lie WITHIN the triangle.
A larger smaple size increases precision, so small studies will be scattered widely at the base of the triangle, whereas larger studies (more powerful) will be at the top and have a narrow spread.
applicability of a study's results byeond the group that was initially assessed
evaluating the inclusion/exclusion criteria is helpful in determining a study's external validity (ie a study in middle-aged women woudl not necessarily be generalizable to elderly men)
major threat to internal validity?
conclusions regarding cause/effect in a study; answers the questions "are we observing/measuring what we think we are observing/measuring"
Quick way to calculate number needed to treat
Type I error
Type II error
type I (a) - false positive - when a study incorrectly rejects a null hypothesis that is true; reflects the significance of a test (i.e. higher a = higher likelihood of type I error and decrease likelihood of type II error.
type II (ß) - false negative - when a study fails to reject a null hypothesis that is false; related to how much power a study has to detect a difference when a difference actually exists. Power = 1 - ß