Clinical Measurement

Question 1

what is uncertainty?

Answer 1

the estimated range of values which the true value lies in

Question 2

what are limits of agreement?

Answer 2

1) a statistical method used to assess agreement between two measurement tools or methods.
2) simplest way to visualise agreement is a scatterplot; we expect values to be correlated and line on the line y = x

Question 3

How do you calculate limits of agreement?

Answer 3

1) Calculate the mean difference between the two methods. (mean(a-b))
2) Calculate the standard deviation of the differences between the two methods. (sd(a-b))
3) Calculate the limit of agreement via
mean difference +/- (critical value * sd difference)
4) create a bland altman plot
5) test assumptions

Question 4

How do you interpret LOA confidence interval?

Answer 4

It means that if the variable is measured by both techniques, their measurements will typically differ by anywhere between x and y.

this means that 95% of the time, the difference between the two methods measurements will be within this range.

Question 5

How do you interpret LOA?

Answer 5

1) If most of the differences between the two methods fall within the limits of agreement, then the two methods can be considered interchangeable within that range.
2) otherwise, they do not agree well

Question 6

what are the assumptions of LOA?

Answer 6

1)The differences between the two methods should follow a normal distribution.
2)The differences between paired measurements should be independent of each other.
3)The mean difference should be constant across all values of the measurement.

Question 7

what is sensitivity?

Answer 7

Probability that a test correctly identifies those with the disease (True Positive Rate).

Question 8

what is specificity?

Answer 8

Probability that a test correctly identifies those without the disease (True Negative Rate).

Question 9

what is a positive predictive value?

Answer 9

Probability that a person who tests positive actually has the disease.

Question 10

what is a negative predictive value?

Answer 10

Probability that a person who tests negative is actually disease-free.

Question 11

how do you calculate sensitivity?

Answer 11

number who are disease positive and test positive / total number who are disease positive

Question 12

how do you calculate specificity?

Answer 12

number who are disease negative and test negative / total number who are disease negative

Question 13

what are the motivations for high sensitivity?

Answer 13

1) minimise false negatives, ensures diseased individuals are not missed
2)essential when missing the disease has serious consequences
3) good for screening where early detection improves outcomes.

Question 14

what are the negatives of high sensitivity?

Answer 14

1) more healthy people mistakenly identified as diseased
2) leads to unnecessary tests, resource-use, and anxiety

Question 15

what are the motivations for high specificity?

Answer 15

1) ensures healthy individuals are not wrongly diagnosed
2) essential when false positives are harmful or costly
3) good for confirmatory tests, where certainty is needed before treatment

Question 16

what are the negatives of high specificity?

Answer 16

1) more unhealthy people are wrongly identified as healthy

Question 17

what is a positive predictive value (PPV)?

Answer 17

Probability of having the disease given a positive result

Question 18

what is a negative predictive value (NPV)?

Answer 18

Probability of being disease negative given a negative test result

Question 19

what is p prev?

Answer 19

the prevalence of the condition

Question 20

what is p sens?

Answer 20

sensitivity

Question 21

what is p spec?

Answer 21

specificity

Question 22

what is the correct table layout for sensitivity and specificity?

Answer 22

1) true positive = disease present + test positive = top left (a)
2) disease presence/absence in columns
3) test positive/negative in rows
4) a b (top row)
5) c d (bottom row)

Question 23

what is the sensitivity formula from the contingency table?

Answer 23

a / (a+c)

Question 24

what is the specificity formula from the contingency table?

Answer 24

b / (b + d)

Question 25

what is the PPV formula from the contingency table?

Answer 25

a / (a + b)

Question 26

what is the NPV formula from the contingency table?

Answer 26

d / (c + d)

Question 27

what is an ROC curve?

Answer 27

1) shows the performance of a classification model at specific thresholds. 2) examines the trade-off between sensitivity and specificity.

Question 28

what does the area under (AUC) the ROC curve show?

Answer 28

1) summarises test performance in a single value 2) < 0.7 = fail/poor 3) 0.7-0.8 = fair 4) 0.8-0.9 = considerable 5) >0.9 = excellent

Question 29

how should an ROC cutoff be chosen?

Answer 29

to maximise both sensitivity and specificity

Question 30

what is a reference interval?

Answer 30

the range of values within which a specified percentage (typically 95%) of a healthy population falls for a particular laboratory measurement

Question 31

what assumption must data meet when calculating reference intervals?

Answer 31

must be normally distributed; log transform if not; use non-parametric method if this doesn't work if data is log transformed, must take the exp (10^data) when reporting reference values

Question 32

what are the issues with the non-parametric reference interval method?

Answer 32

1) relies on percentiles and makes no assumption about underlying distribution 2) if data has long tails, extreme values can be highly variable, leading to wide, unstable reference limits

Question 33

What is the rcode for adding line y=x?

Answer 33

abline(0,1)

Question 34

whats the limits of agreement calculation in r?

Answer 34

mean.diff + ( c(-1.96, 0, 1.96)*sd.diff )

Question 35

How do you plot a Bland-Altman plot?

Answer 35

a) calculate mean of values (a+b)/2 b) calculate difference between measurements (a-b) c) plot these against each other (plot( mean, difference) d) add LOA to plot abline( h = LOA)

Question 36

whats the normal reference interval calculation?

Answer 36

mean +/- (1.96*s.d)

Question 37

How do you calculate confidence intervals for reference limits?

Answer 37

○ Lower Reference Interval ±(1.96 × s.e. of the reference interval) ○ Upper Reference Interval ±(1.96 × s.e. of the reference interval)

Question 38

a test with high specificity is good for.....

Answer 38

confirming a disease

Question 39

a test with high sensitivity is good for.....

Answer 39

ruling out a disease

Question 40

Is NPV or PPV more important for screening tests?

Answer 40

> NPV > Those with negative results from the screening need to be true negatives as they will have no further medical investigations.

Question 41

In a hospital, where would you source healthy controls from?

Answer 41

a) a fracture clinic; if they fall within the normal range of the variable of interest