When to use pretest probability

[ChessMaster3000]

---

Members don't see this ad.

I ran across a question in Rx that basically had you utilize the pretest probability to set up a 2x2 table. If the pretest probability is 70%, they say, then set up a table with a hypothetical population where 70% of the people have the disease.

Has anyone seen the use of pretest probability on the exam? I havent seen it on any UW questions or the two NBMEs I have done to date.

 

[ulikedaggers]

Pretest probability is the same thing as prevalence, as far as I know. Can you give more info about the question?

 

[ChessMaster3000]

Pretest probability is the same thing as prevalence, as far as I know. Can you give more info about the question?

You know I can't remember where I saw it now, but if it is the same thing as prevalence then that would make sense/work with the question. They also brought up the point of likelihood ratio, and that you could multiply pretest probability by likelihood ratio (I believe sensitivity/1-specificity) to get the post-test probability. I would say this is beyond the scope of step 1?

 

[Miracoli]

here is a case that anwers your Q.
Based on all the information currently available, you estimate that the patient in your office has a one in four chance of having a serious disease. You order a diagnostic test with sensitivity of 95% and specificity of 90%.
The result comes back positive. Based on all the information now available, the chance your patient really has the disease is closest to
a. 100%
b. 95%
c. 90%
d. 75%
e. 60%
f. roughly 30%

 

[ulikedaggers]

Is it B?

 

[Amemus]

No, the answer is D. I could write out the steps if you want, but honestly I think this is beyond the scope of Step 1.

Here is what you need to know: pretest probability just means how likely you think it is the patient has the disease before you do any tests. Say that based on the patient's presentation, you think the chances that the patient has strep throat is low (10%). Your decide you will prescribe antibiotics only if you think there is a 50% or greater chance that the patient has strep throat.

A positive likelihood ratio tells you how much more likely it is that the patient has strep throat given the fact that they had a positive rapid strep test. A negative likelihood ratio tells you how much more likely it is that the patient doesn't have strep throat given the fact that they had a negative rapid strep test.

So initially you think the patient probably doesn't have strep throat, but then they test positive. Is that positive test result enough to convince you that the patient needs antibiotics?

The pretest probability isn't necessarily the same thing as prevalence, unless you're talking about the prevalence among people with all of the same characteristics as the patient and not the general prevalence statistic. It's subjective. It makes sense to base your pretest probability on the prevalence of the disease, but as a clinician you might weigh other factors when making that judgment. I don't think this is likely to be tested, but if it is I would imagine it would be more of a conceptual question.

 

[ChessMaster3000]

No, the answer is D. I could write out the steps if you want, but honestly I think this is beyond the scope of Step 1.

Here is what you need to know: pretest probability just means how likely you think it is the patient has the disease before you do any tests. Say that based on the patient's presentation, you think the chances that the patient has strep throat is low (10%). Your decide you will prescribe antibiotics only if you think there is a 50% or greater chance that the patient has strep throat.

A positive likelihood ratio tells you how much more likely it is that the patient has strep throat given the fact that they had a positive rapid strep test. A negative likelihood ratio tells you how much more likely it is that the patient doesn't have strep throat given the fact that they had a negative rapid strep test.

So initially you think the patient probably doesn't have strep throat, but then they test positive. Is that positive test result enough to convince you that the patient needs antibiotics?

The pretest probability isn't necessarily the same thing as prevalence, unless you're talking about the prevalence among people with all of the same characteristics as the patient and not the general prevalence statistic. It's subjective. It makes sense to base your pretest probability on the prevalence of the disease, but as a clinician you might weigh other factors when making that judgment. I don't think this is likely to be tested, but if it is I would imagine it would be more of a conceptual question.

I got the correct answer, but only assuming that pretest probability is prevalence. When I instead use the LR, with LR=sensitivity/1-specificity, I got a LR of .95/1-.9=9.5. However, I thought you simply multiplied the LR by the pre-test probability, which is clearly not the case as we would get a post-test probability >100. So, I realize this is probably beyond the scope, but since I'm halfway there I'd rather just learn it . How do you get from a LR of 9.5 to a post-test probability of 75%?

 

[Amemus]

You don't need to use the likelihood ratio to answer the question. You use the 2x2 table and calculate the positive predictive value. You could use the likelihood ratio, but that is a little more complicated because the likelihood ratio is expressed in terms of odds, which is not the same as probability. You can't add probabilities, so you would have to make a conversion from pretest probability to odds and back to probability.

 

[NeuroLAX]

You don't need to use the likelihood ratio to answer the question. You use the 2x2 table and calculate the positive predictive value. You could use the likelihood ratio, but that is a little more complicated because the likelihood ratio is expressed in terms of odds, which is not the same as probability. You can't add probabilities, so you would have to make a conversion from pretest probability to odds and back to probability.

This. I got 76% (closest to D. 75%) by taking the pre-test probability and constructing a 2x2 table with a hypothetical population of 100 and using the sensitivity and specificity they give you. Once you get your table it is just a simple calculation to get the PPV.

 

[ChessMaster3000]

This. I got 76% (closest to D. 75%) by taking the pre-test probability and constructing a 2x2 table with a hypothetical population of 100 and using the sensitivity and specificity they give you. Once you get your table it is just a simple calculation to get the PPV.

Right--but if you construct a 2x2 table with a hypothetical population, aren't you, as you said a few posts ago, equating pre-test probability with prevalence in your hypothetical population?

If that is the case, regardless of the population characteristics they give you in any test question, if they also give you a pre-test probability, you can just go ahead and create your own population with a hypothetical population of 100 patients?

 

[NeuroLAX]

Right--but if you construct a 2x2 table with a hypothetical population, aren't you, as you said a few posts ago, equating pre-test probability with prevalence in your hypothetical population?

If that is the case, regardless of the population characteristics they give you in any test question, if they also give you a pre-test probability, you can just go ahead and create your own population with a hypothetical population of 100 patients?

I see what you're saying, and now I'm not so sure if my method is sound for this specific type of question. They don't even discuss PLR in FA. I had to look in USMLE Secrets for the PLR --> PPV approach:

Post-test probability (PPV) = (Pre-test probability) x (PLR)

So in the example above, PLR is 9.5. So PPV = (0.25) x (9.5). In terms of a ratio, this can be expressed as 9.5:4, and to get the PPV in terms of %, you would put that as 9.5/(9.5 +4), or 9.5/13.5, which comes out to 70.37%. This also gives you the correct answer, D.

 

[Amemus]

I see what you're saying, and now I'm not so sure if my method is sound for this specific type of question. They don't even discuss PLR in FA. I had to look in USMLE Secrets for the PLR --> PPV approach:

Post-test probability (PPV) = (Pre-test probability) x (PLR)

So in the example above, PLR is 9.5. So PPV = (0.25) x (9.5). In terms of a ratio, this can be expressed as 9.5:4, and to get the PPV in terms of %, you would put that as 9.5/(9.5 +4), or 9.5/13.5, which comes out to 70.37%. This also gives you the correct answer, D.

No, that's not correct. You are mixing probability and odds. They're like different units. You need to convert the 0.25 to odds before you can relate that to the LR+. If you do all that, you should get 74%.

Different methods to get to the same answer. There is also the decision tree method and Baye's theorem - again, beyond the scope of Step 1. Most people stick with the 2x2 method because it's the easiest one to understand. But, if you can wrap your mind around the concept of odds, you should learn the likelihood ratio method because it's used all the time in evidence based medicine. I hate to sound annoying, but learning what the concepts mean is more important than memorizing the steps.

Also don't forget that the formula for prevalence is based on the population at risk. Not just the general population. What you think should be factored in when determining the population at risk and hence the pretest probability is subjective. No one expects you to be able to make that call if you're a med student with little to no experience with evidence based medicine research. If they give you the pretest probability, then they made that decision for you.

 

[ulikedaggers]

Can someone just screenshot their 2x2 because I still can't get to the right answe.r

 

[NeuroLAX]

No, that's not correct. You are mixing probability and odds. They're like different units. You need to convert the 0.25 to odds before you can relate that to the LR+. If you do all that, you should get 74%.

Different methods to get to the same answer. There is also the decision tree method and Baye's theorem - again, beyond the scope of Step 1. Most people stick with the 2x2 method because it's the easiest one to understand. But, if you can wrap your mind around the concept of odds, you should learn the likelihood ratio method because it's used all the time in evidence based medicine. I hate to sound annoying, but learning what the concepts mean is more important than memorizing the steps.

Also don't forget that the formula for prevalence is based on the population at risk. Not just the general population. What you think should be factored in when determining the population at risk and hence the pretest probability is subjective. No one expects you to be able to make that call if you're a med student with little to no experience with evidence based medicine research. If they give you the pretest probability, then they made that decision for you.

Well, damn... I thought I understood how to solve this problem. I appreciate your explanation, but do you mind posting the correct steps? Thanks

Better to make mistakes here.

 

[Amemus]

Probability to odds: O = P/(1-P)
Odds to probability: P = O/(1+O)

1. Convert pretest probability to pretest odds
2. Posttest odds = Pretest odds * Likelihood ratio
3. Convert posttest odds to posttest probability