Marja:
Along the lines of Jesse’s and Rich’s comments, here is the PDF of a presentation that I give during workshops on pathogen surveillance.
http://fwf.ag.utk.edu/mgray/SEPARC/PathogenSurveillance2012.pdf
For pathogen surveillance in a population, the question is: (1) required sample size to detect a pathogen vs. (2) required sample size for p-hat (estimate) to be within a designated error margin of p (parameter) with a certain level of
confidence vs. (3) required sample size to detect a statistical difference between 2 or more estimates of population prevalence. Procedures for estimating 1 and 2 are in the presentation (pages 6 and 7), and #3 can be estimated using various statistical packages
such as Minitab. However, I think that your question pertains to “how many captive individuals should be tested to ensure they are ranavirus negative?” I do not believe OIE has made this specification for ranaviruses yet (as Jesse alluded). If the question
is ranavirus detection, required sample size depends on: (1) size of captive population, (2) confidence in detection, and (3) if there is infection, the minimum prevalence you are able to detect. If your captive population is 50 individuals, 35 should be
tested if you want to be 95% confident you detect the ranavirus if >5% are infected (see page 6). If you want to be
>99% certain that none are infected, all individuals need to be tested. This is assuming that the technique you use is highly sensitive. Similar to Jesse, we have found that qPCR (following Picco et al. 2007) is very sensitive in laboratory studies,
with standard curves predicting detection of <10 PFUs. Thus, required sample size is really influenced by 2 parts: (1) statistical probability of detection, and (2) sensitivity of a test. Moreover, it is likely that different ranavirus sampling techniques
(swabs vs. tissue) and different tissues will have different detection probabilities (see
Gray et al. 2012; DAO 99:1-6). Thus, I am afraid at this point, we cannot provide a standard recommendation for required sample size for ranavirus detection given a certain sampling and
molecular technique. Until a standard is established, you might consider following the guidance above or recommendations provided by Jesse or Rich.
GRC:
I think Marja’s inquiry is a very good one. Perhaps, we can devote some time to discussion at the 2013 International Symposium on Ranaviruses.
All the Best---
Matt
______________________________________________________________
Matthew J. Gray, Ph.D.
Center for Wildlife Health
University of Tennessee
274 Ellington Plant Sciences Building
Knoxville, TN 37996-4563
865.974.2740 [ofc] [log in to unmask]
865.974.4714 [fax] http://fwf.ag.utk.edu/personnel/mgray.htm
For more information about Forestry, Wildlife and Fisheries at UTK,
Please visit http://fwf.ag.utk.edu/
or call 865.974.7126
_____________________________________________________________
From: Global Ranavirus Consortium [mailto:[log in to unmask]]
On Behalf Of Jesse Brunner
Sent: Monday, June 18, 2012 2:27 PM
To: [log in to unmask]
Subject: Re: question from the netherlands
Thanks, Rich.
That website is really useful! In addition to the calculator you linked to, there are other calculators that account for sensitivity and and specificity (although unfortunately not in the context of calculating sample sizes) and one that
is especially handy for calculating confidence intervals on proportions (the normal approximation we use falls apart when the prevalence is close to 0 or 1). If there are any R users, I've embodied some of these methods in R code which makes the playing with
numbers / simulation that Rich recommends a bit more automated. I'm happy to share.
Cheers,
Jesse
On Jun 18, 2012, at 10:59 AM, Seigel, Richard wrote:
Jesse et al:
There are many different ways of calculating sample size. A simple (and interactive way) of determining the required sample size for a study is to model this as a Binomial Distribution, since
animals can only be positive or negative. By using one of the many web-based Binomial calculators available (I recommend http://vassarstats.net/binomialX.html),
you can easily model the probability of getting zero positive results in a sample size of N animals, assuming a specific level of prevalence.
Here is how this works for the calculator noted above:
You need to specify three parameters: N (number of individuals you want to use for the total sample), K (the number of positive animals, which you set as zero), and P (the “true” prevalence
rate, from 0-1.0).
Using the calculator noted above, if you fill in these values for Jesse’s example (30 amphibians sampled, prevalence = 0.1, zero found with infections), you will get the output shown on the
attached page. The critical part of this is the part listed under “hypothesis testing” towards the bottom the output:
P: 0 or fewer out of 30
One-Tail Two-Tail
Method 1. exact binomial calculation 0.0423
-this tells you that the probability of sampling 30 animals at random where the actual prevalence is 0.1 and finding zero “hits” is 0.0423. The inverse of this is the confidence level (1-0.0423
= 95.7%). If the actual prevalence rate was 5%, you would need a sample of 60 animals.
The nice thing about this approach is that it is (a) free and (b) can be used for any combination of values of estimated sample size, levels of prevalence, and desired confidence levels.
However, keep in mind that, like any inference-based statistic, you need to be sure that you are sampling at random and that there are no false negatives. Based on Jesse’s last email, I wonder how confident we can be of that?
Rich Seigel
Richard A. Seigel
Professor
Department of Biological Sciences
Towson University
Towson, MD 21252
410-704-3123
From: Global
Ranavirus Consortium [mailto:[log in to unmask]] On Behalf Of Jesse Brunner
Sent: Monday, June 18, 2012 1:12 PM
To: [log in to unmask]
Subject: Re: question from the netherlands
Dear All,
I was just looking through the OIE manual again, based on Greg's suggestion, and was struck by the lack of validated PCR, especially qPCR assays in it. From my experience (and I believe that of others, though I won't speak for them), quantitative
real time PCR is much more (analytically) sensitive than standard PCR (e.g., with the MCP4/5 primers in Mao et al. 1997) and more sensitive even than virus isolation in cell culture. We are consistently detecting the equivalent of <1 plaque forming unit in
asymptomatic animals from experiments and from the wild. Jaramillo et al. have a validated assay for ranavirus in fish that seems to support the assertion that qPCR is more sensitive:
"The assay had 100-fold greater analytical sensitivity compared to virus isolation in Bluegill fry (BF-2) cells, detected all viruses in a panel of 20 ranaviruses from Australia, America,
Europe and the United Kingdom, did not detect two disputed members of the genus Ranavirus (doctorfish
virus and guppy virus 6) and did not detect a megalocytivirus or two cyprinid herpesviruses." (Note: I do understand that analytic sensitivity is not the same as diagnostic sensitivity, but if virus isolation is our gold standard then qPCR cannot,
by definition, have greater diagnostic sensitivity!)
So I guess my question is, if someone is interested in ensuring that their animals are ranavirus free, shouldn't we be pointing them to these more sensitive assays? What are your thoughts? Would it be worth "stressing out" some subset of
the animals to see if animals carrying inapparent infections might recrudesce? The question of whether they can develop full blown, transmissible infections seems very relevant in these cases.
I would also like to point out that the protocols in the OIE manual state, in section 6: "Statistically valid sampling practices need to be used
but these cannot presently be defined for amphibians." I am not sure why amphibians would present a special case in epidemiology, but I was not at these meetings, so perhaps someone could clear up my confusion. Barring that, I might suggest using
the standard formulas for calculating the sample sizes required to be, say, 95% confidence that the prevalence of infection is at or below some level (e.g., 10%) given a population of N (e.g., 10^6)… here n = 30. I'm sure there are other references for the
formula, but this is the one I have:
Cannon, R.M., and R.T. Roe. 1982. Livestock Disease Surveys: A Field Manual For Veterinarieans. Canberra: Australian Bureau of Animal Health
I've attached a spreadsheet I created a while back to help me see the relationships between confidence-level, population sizes, the number of disease animals, and the sample size. It is rather busy, but some playing with the numbers in
different colors should clear up which graph is associated with which equation. Or I'm happy to answer questions about it.
In any case, if there is a better way of establishing statistically valid sample sizes, I would love to hear them. It would be great for the members of this list to be on the same page, more or less, about how best to sample.
Best wishes,
Jesse
Jesse Brunner
School of Biological Sciences
Washington State University
PO Box 644236
Pullman, WA 99164 USA
509-335-3702
Jaramillo D, Tweedie A, Becker JA, Hyatt A, Crameri S et al. (2012) A validated quantitative polymerase chain reaction assay for the detection of ranaviruses (Family Iridoviridae) in fish
tissue and cell cultures, using EHNV as a model. Aquaculture
<Binomial Probabilities for Rv sampling.pdf>