Flashcards in Clinical measurement Deck (19)
Describe reference ranges
generally 2.5% to 97.5%
If normally distributed can be calculated by mean +/- 2(1.96)SD
Just because patient has value outwit the reference range doesn't necessarily make them unhealthy- 5% of normal population will be outwith the reference range
What can measurement errors be due to?
lack of accuracy (calibration)
Lack of precision (repeatability)
What is biological variability?
differences between repeated measurements taken at different time points
what are instrumental errors?
inaccurate diagnostic test
What are respondent errors?
trying to appear healthy
What are observer errors?
mistakes, imprecision in measurement or misunderstanding of procedures
What are data processing errors?
coding of variables
copying in of data
What is the coefficient of variation?
often used in lab experiments
it measures the degree of variation associated with repeated measurements
CV = SD/mean X 100
What are two ways of comparing two methods of measurement?
describe using a scatter plot to compare methods of measurement
with the line of equality
gives an impression of how well the two methods agree
Describe the use of a bland altman plot to compare methods of measurement
plots the difference between the measurements against the sum or the average of the observations
the differences are quantified by estimating the bias
The SD of the differences allows a range to be determined in which we would expect 95% of the differences to lie
there should be no pattern inn the plot suggesting that bias or SD is changing as the measurement increases
What is sensitivity?
the probability of a true positive test result if the disease is present a/a+c
What is specificity?
the probability of a true negative test result if the disease is not present
How is PPV calculated?
how is NPV calculated?
How is prevalence calculated?
Describe the Receiving Operating Characteristic (ROC) curve
Used to establish the optimal cut-off of what we want to achieve from a test
Plot sensitivity (true pos rate) against 1-specificity (flash positive rate) on the x-axis for different cut-off points
The number you need to treat with the new treatment to achieve one more "success" than with the old treatment
NNT=1/P(success new)-P(success old)