- Joined
- Nov 20, 2001
- Messages
- 1,084
- Reaction score
- 6
Can someone please break down what these actually mean? The fancy definitions confuse me and I don't quite get how these are useful when reading over approaches to some conditions.
Specificity vs. Sensitivity
- Thread starter Mike59
- Start date
- Joined
- Jul 25, 2003
- Messages
- 1,614
- Reaction score
- 6
Specificity - the percent of time the test correctly identifies a condition.
Sensitivity - the percent of conditions a test identifies.
So, a test for prostate cancer with a sensitivity of 100% will never miss prostate cancer, but will likely identify a bunch of people as "positive" for prostate cancer who do not actually have the disease. (false positives)
A test for prostate cancer with 100% specificity will never identify someone as having prostate cancer who does not. The downside is that a bunch of people with prostate cancer will be missed.
A good example (continuing with the prostate) is PSA. I'm using hypothetical numbers here Kalel so don't shoot them down - I know they're not right.
Generally, a PSA > 4 has been the action point for PSA. At this point the specificity is relatively low and the sensitivity is relatively high - most people (but not all) with prostate cancer will be identified at this point (sensitivity) but a number of them won't have prostate problems at all (specificity).
If we dropped it to >1, we would catch almost all the cancer (very high sensitivity), but many more of those men would not actually have cancer (very low specificity). If we raised it to >9 almost certainly those men would have cancer (very high specificity) but we would miss many cases of cancer (very low sensitivity).
It is uncommon to find a test that has both a high specificity and sensitivity. Usually it is a trade off. That is why often a combination of tests are used to evaluate for a particular pathology.
C
Sensitivity - the percent of conditions a test identifies.
So, a test for prostate cancer with a sensitivity of 100% will never miss prostate cancer, but will likely identify a bunch of people as "positive" for prostate cancer who do not actually have the disease. (false positives)
A test for prostate cancer with 100% specificity will never identify someone as having prostate cancer who does not. The downside is that a bunch of people with prostate cancer will be missed.
A good example (continuing with the prostate) is PSA. I'm using hypothetical numbers here Kalel so don't shoot them down - I know they're not right.
Generally, a PSA > 4 has been the action point for PSA. At this point the specificity is relatively low and the sensitivity is relatively high - most people (but not all) with prostate cancer will be identified at this point (sensitivity) but a number of them won't have prostate problems at all (specificity).
If we dropped it to >1, we would catch almost all the cancer (very high sensitivity), but many more of those men would not actually have cancer (very low specificity). If we raised it to >9 almost certainly those men would have cancer (very high specificity) but we would miss many cases of cancer (very low sensitivity).
It is uncommon to find a test that has both a high specificity and sensitivity. Usually it is a trade off. That is why often a combination of tests are used to evaluate for a particular pathology.
C
Sensitivity - how many of the true positives are actually detected? This is important for things such as HIV tests where you dont want to miss anyone. The more sensitive it is, the more of those infected are detected.
Specificity - You can imagine it as the reverse. How many of the true negatives are detected? The more specific it is, the greater % of negative people will test negative.
Usually increasing one will decrease the other.
Specificity - You can imagine it as the reverse. How many of the true negatives are detected? The more specific it is, the greater % of negative people will test negative.
Usually increasing one will decrease the other.
- Joined
- Nov 12, 2002
- Messages
- 1,187
- Reaction score
- 7
mathematically....
sensitivity= (#true positives)/(#true positives + #false negatives)
specificity= (#true negatives)/(#true negatives + #false positives)
sensitivity= (#true positives)/(#true positives + #false negatives)
specificity= (#true negatives)/(#true negatives + #false positives)
- Joined
- May 23, 2003
- Messages
- 177
- Reaction score
- 3
Here's how I look at it...
Sensitivity: for people w/ the disease, the liklihood of getting a positive test result.
Specificity: for people w/o the disease, the liklihood of getting a negative test result.
If you were designing a screening test, you'd want it to be as highly sensitive as possible, even if that meant picking up some false negatives.
For a confirmatory test (after you've identified all the positives by your highly sensitive test), you'd want it to be as highly specific as possible.
***Remember, sensitivity rules OUT, specificity rules IN.
For predicitive values...
Positive predictive value: for people w/ a positive test result, the liklihood of having the disease.
Negative predictive value: for people w/ a negative test result, the liklihood of not having the disease.
Predictive values are unique in that they depend on prevalence (the % of population w/ a dz at a given 'snapshot' in time). As prevalence rises, your PPV rises, and NPV decreases. That's because if a disease is highly prevalent, there's a good chance that a positive test result means that person has the dz!
Sensitivity: for people w/ the disease, the liklihood of getting a positive test result.
Specificity: for people w/o the disease, the liklihood of getting a negative test result.
If you were designing a screening test, you'd want it to be as highly sensitive as possible, even if that meant picking up some false negatives.
For a confirmatory test (after you've identified all the positives by your highly sensitive test), you'd want it to be as highly specific as possible.
***Remember, sensitivity rules OUT, specificity rules IN.
For predicitive values...
Positive predictive value: for people w/ a positive test result, the liklihood of having the disease.
Negative predictive value: for people w/ a negative test result, the liklihood of not having the disease.
Predictive values are unique in that they depend on prevalence (the % of population w/ a dz at a given 'snapshot' in time). As prevalence rises, your PPV rises, and NPV decreases. That's because if a disease is highly prevalent, there's a good chance that a positive test result means that person has the dz!
- Joined
- Apr 22, 2004
- Messages
- 391
- Reaction score
- 0
I find the SpIN and SnOUT mneumonic very useful:
SpIN- Specificity rules Positives In - meaning that on a test w. high specificity if you test positive chances are you got the disease (confirmatory)
SnOUT - Sencitivity rules Negatives Out - meaning that on a test with high sencitivity if you test negative chances are you don't have the disease (screening)
I also just memorized the 2x2 and that helps a lot as well
.............disease
.............+.....-
Te-.... + TP....FP...PPV
st...... - FN....TN...NPV
............Sn...Sp
TP=True Positive FP=False Positive FN=False Neg TN=True Neg PPV=Pos Pred Val NPV=Neg Pred Val
To figure each you take the True value and divide by the total for tha row or column
You can convince yourself then that if Sn is high that means TP must be high comapred to FN and you dont get a lot of FN hence if you're negative you prob don't have the disease (SnOUT)...you can do the same for the others
I think you can solve most stats probs with that table...its MONEY
SpIN- Specificity rules Positives In - meaning that on a test w. high specificity if you test positive chances are you got the disease (confirmatory)
SnOUT - Sencitivity rules Negatives Out - meaning that on a test with high sencitivity if you test negative chances are you don't have the disease (screening)
I also just memorized the 2x2 and that helps a lot as well
.............disease
.............+.....-
Te-.... + TP....FP...PPV
st...... - FN....TN...NPV
............Sn...Sp
TP=True Positive FP=False Positive FN=False Neg TN=True Neg PPV=Pos Pred Val NPV=Neg Pred Val
To figure each you take the True value and divide by the total for tha row or column
You can convince yourself then that if Sn is high that means TP must be high comapred to FN and you dont get a lot of FN hence if you're negative you prob don't have the disease (SnOUT)...you can do the same for the others
I think you can solve most stats probs with that table...its MONEY
- Joined
- May 22, 2004
- Messages
- 48
- Reaction score
- 0
Also, recall that spec. and sens. will always depend upon the prevalence in the population. Very important!!!! Depends on prevalence, yo!
- Joined
- May 23, 2003
- Messages
- 177
- Reaction score
- 3
med_stud_07 said:
This is incorrect.... Predictive Values depend on prevalence.... Sensitivity and specificity do NOT!...
See p 116 of First Aid: "Unlike sensitivity and specificity, predictive values are dependent on the prevalence of the disease. The higher the prevalence of a disease, the higher the positive predictive value of a test."
- Joined
- May 22, 2004
- Messages
- 48
- Reaction score
- 0
Actually, what I meant was they calculate the sens. and spec. from the prevalence. The prevalence is used to determine these numbers. Thus, sens. and spec. DO depend on prevalence.
I may be wrong, I frequently am. Let's examine the facts.
I may be wrong, I frequently am. Let's examine the facts.
- Joined
- Jul 22, 2003
- Messages
- 178
- Reaction score
- 0
This is correct!AlexRusso said:
Sensitivity and Specificity will have the same predictive values in whatever population they are undertaken in. Positive Predictive Values (PPV) and Negative Predictive Values (NPV) will vary depending on the incidence of the disease or condition in that population. Foe instance the PPV for Chlamydia in 18-20 year old females will be higher than the PPV for Chlamydia in thirty-32 year olds simply because the incidence is higher in the younger age range.
I remember Sensitivity as "the number of people who test positive who actually have the condition" and Specificity as "the number of people who test negative who do not have the condition".
Similar threads
