Jay, [presenting stuff that you know already know, not because I don't think you know, but to help flow of reasoning],
Sensitivity is the true positive rate. The higher the true positive rate the lower the false negative rate. We do not want to have false negatives when we are trying to rule out disease.
A 100% perfect sensitive test (a true SnNout) therefore misses no disease - but also does not exist in real life.
If you only look at sensitivity you ignore the children without the disease.
It is correct to say that we can be 95% sure that of the children WITH (or GIVEN) the disease will be missed somewhere between 1/30 and 1/10 of the time.
Unfortunately when we are making the decision on whether to order the test we do not know which are the ones with the disease.
We could use the worst case scenario and assume that all the children with low score have at most a 1:10 chance of CiTBI and decide to CT all these children.
It also follows that if we CT all these children (pretty much everyone) we could pick up additional cases that would have been missed.
What we would be doing however is exposing a large number of "not worst case scenario" children to CT, which in itself has some harm.
Hence we need to know something about the group who do not have disease calculate Negative Predictive Vale, Negative Likelihood Ratios or Number Needed to Harm.
Then we compare the risk of not doing the test to the estimate of risk caused by doing the test.
I have not looked at how Chris calculated NNH but agree that a 95% confidence interval should be quoted for these measures.
Think it was more of an attempt to try to quantify the risk in a more holistic fashion.
On 09/11/2012, at 5:16 AM, Jay Pershad wrote:
> Hi David,
> You said, "One of the problems of not using NNH or NNT type assessments is that we tend to overestimate prevalence of disease and are poor at calculating risk in our head. The way your statement reads it suggests that your CT scan pick up rate of clinically important ICH in children with minor closed head injuries (that are PECARN negative) is more than than 3% (and up to 10%).This does not sound likely and would not be consistent with the published evidence."
> I have to agree with Jay Fisher's approach to risk stratification. For a similar clinical encounter, if the family of a patient who meets low risk criteria based on the PECARN data, enquires about the "worst case scenario," I would have to state that based on a large study of children with TBI, there is 95 % chance that their son's risk of death or operation (CiTBI) may be as high as 1 in 10. The NNH or NNS (number needed to screen) is also very useful, but to me is is most relevant from a population or societal perspective.
> On that note, I am struggling with how you arrived at an NNH of 1 in 2,000. There were 61 patients with CiTBI in the high risk group (i.e. any one factor in the prediction rule being positive) & 2 patient's with CiTBI in the low risk group from a population of 6400 subjects in the validation study. The NNS would be closer to ~ 1:100. Besides, you are including just the point estimate in quoting the NNH. To be truly objective, you really need to examine the 95% CI for the NNH.
> Unrelated, I thought persistent HA or "severe" mechanism were considered high risk criteria within the decision rule? One could then argue that Lakshmi's patient met both of the above, and cranial imaging would have been warranted???
> Very engaging discussion....
> Jay (Pershad)
> Le Bonheur
> For more information, send mail to [log in to unmask] with the message: info PED-EM-L
> The URL for the PED-EM-L Web Page is:
Dr David Herd
Mater Children's Emergency
[log in to unmask]
For more information, send mail to [log in to unmask] with the message: info PED-EM-L
The URL for the PED-EM-L Web Page is: