8

*1.5 million people have lupus in America out of a population of 308 million. 90 percent are women. My wife is white, as is 63 percent of the country. Minorities are three times more likely to have lupus than whites.

My wife has been diagnosed with lupus, but the diagnosis may not be correct.

*20 percent of lupus patients have a parent or sibling with lupus. My wife does not.

  • 70 percent of people with lupus have positive double-stranded anti-DNA test. My wife does not.

*70 percent of lupus cases are systemic, which is what she is diagnosed with.

What is the probability she has lupus?

So, to be simple... What are the chances given the first set of stats, the 1.5 million out of 308 million and further reducing stats, that she would have lupus (one out of 9 million, etc)?

second question, can the probablity and odds be calculated from the percentages. There was a one in five chance a sibling or parent has it, plus a 70 percent chance she would have the double dna test, but didn't. As we combine these percentages and figures, can it be deduced?

  • I flagged the original question suggesting it might be better off migrated here. If so, we can merge the present thread with it. – whuber Dec 13 '12 at 00:14
  • Are you serious? – Jonas Meyer Dec 11 '12 at 06:08
  • 15
    It is never lupus. –  Dec 11 '12 at 06:09
  • 6
    I do not think this question, as written, is fit for this site. You should ask an epidemiologist. If my vote were not binding, I'd vote to close. –  Dec 11 '12 at 06:10
  • 6
    In terms of mathematics this is the perfect place for such a question, but I don't know about this...I'd certainly want to base my decision in such a circumstance on more than a stackexchange answer. In terms a constructive mathematical comment, I'd say we need to know how accurate each of your doctors here are at diagnosis, because this is really a conditional probability...but I agree with Mariano, please ask a real expert for medical advice and not math.stackexchange. –  Dec 11 '12 at 06:11
  • 4
    If this is a real question, hire a mathematical consultant. Advice is generally worth what you pay for it. –  Dec 11 '12 at 06:14
  • By the way, I trust 1.5 people is a typo for 1.5 million people. –  Dec 11 '12 at 06:15
  • 2
    I am asking on this site, because it's a site to ask math questions. I'm not seeking medical advice, I'm asking for help in calculating a mathematical probability based on these numbers and stats. Can someone help? –  Dec 11 '12 at 06:16
  • yes, 1.5 million –  Dec 11 '12 at 06:17
  • 4
    If so, then please remove all irrelevant personal information. –  Dec 11 '12 at 06:17
  • 1
    @MarianoSuárez-Alvarez, can this be moved to stats.stackexchange? I did statistics undergraduate, I know some of the actual issues here, but I would prefer this asked of someone who works on it full time. Or, not answered in a more authoritative manner. – Will Jagy Dec 11 '12 at 06:31
  • 1
    @WillJagy I don't think that's the solution here. If the OP would modify the question to phrase it purely as a mathematics question, then I think everyone would be happy to answer it as best as possible. As it stands the OP should be talking to a medical professional, and non-professionals on stackexchange shouldn't be asked to do the work of medical professionals. It would be irresponsible on our part. –  Dec 11 '12 at 06:38
  • 3
    The guy is asking for help with his computation. –  Dec 11 '12 at 06:43
  • 3
    I agree with @copper.hat. I find some of the above comments somewhat patronizing. No-one is being forced to answer, and everyone can (and probably should) precede their answer with a disclaimer pointing out that there are probably further issues to consider that are outside mathematics and should be dealt with by a medical professional. That doesn't mean we can't answer someone's math question if they decide to post it here, however wise or unwise doing that may seem to us; that's for the OP to decide. – joriki Dec 11 '12 at 06:52
  • 4
    In case we're using the close vote cancelling system (I was never sure whether we adopted it), I hereby vote not to close. – joriki Dec 11 '12 at 06:59
  • 3
    Your information only (at best) allows an estimate of the probability that a person of a particular type has lupus at random. Is says nothing about the probability given the observed symptoms. – Henry Dec 11 '12 at 07:05
  • 2
    On the strictly mathematical side of the question, even a cursory reading reveals that some data and (in-)dependence hypotheses are missing, that are needed before one can give a mathematical answer. – Did Dec 11 '12 at 07:05
  • Sorry for the personal info, I'm not familiar with this website. But this is a real life problem I am trying to calculate, and I do appreciate everyone who has tried to answer, but everyone is making this overly complicated. –  Dec 11 '12 at 07:25
  • so, to be simple... What are the chances given the first set of stats, the 1.5 million out of 308 million and further reducing stats, that she would have lupus (one out of 9 million, etc)? –  Dec 11 '12 at 07:27
  • second question, can the probablity and odds be calculated from the percentages. There was a one in five chance a sibling or parent has it, plus a 70 percent chance she would have the double dna test, but didn't. As we combine these percentages and figures, can it be deduced? –  Dec 11 '12 at 07:29
  • 1
    @Craig: It's not a good idea to bury clarifications of the question deep in a long comment thread. Please edit the question itself. – joriki Dec 11 '12 at 07:32
  • Assuming the indicators are independent, use the Bayes theorem. –  Dec 11 '12 at 08:00
  • 1
    I posted a reopen request. – joriki Dec 11 '12 at 10:58
  • @whuber Merging would be good. – Fomite Dec 13 '12 at 00:21
  • @whuber, it seems the migration happened a few seconds ago. Since EpiGrad gave a very nice answer here, I won't be deleting this or doing anything until I get clear instructions. – Will Jagy Dec 13 '12 at 00:21
  • @EpiGrad, merging happened already. I guess it is up to you and whuber to decide whether to keep two questions open, possibly move your answer (exactly what I wanted, by the way, an answer from someone who is not just guessing) to the merged question, etc. – Will Jagy Dec 13 '12 at 00:23
  • 1
    @whuber It'd be my preference, just in terms of the cleanliness of the site, that my question be moved into the main thread, wherein I think this one can be deleted? If that's hard to do in software I'll just copy-and-paste? – Fomite Dec 13 '12 at 00:24
  • Is it possible to alter the image you posted to remove the poster's name? Perhaps the OP took down the question because he had second thoughts about posting health info on a public site. – Steve Lianoglou Dec 13 '12 at 01:15
  • @SteveLianoglou, he never took down the question. Instead, it has been migrated here as a separate question. It does not show up under the "Newest" sort for questions, but it does for "Active." – Will Jagy Dec 13 '12 at 01:18

3 Answers3

15

An answer that explained why the question cannot be answered in its present form was given but then deleted after you commented on it; I'll try to explain in more detail why there is not enough information to state the desired probability. I want to emphasize that I'm not saying this because I disapprove of the question; as I wrote in a comment, I think it's not for us to decide whether you should be asking this question; it's just that there is simply not enough information, and if you really do want to find the desired probability on your own, you'd need to obtain the missing information.

First, we know nothing about the tests that were performed. You can see that the answer must depend on the reliability of the tests by considering the extreme cases: If the test is utterly unreliable and its results bear almost no relation to the actual presence of the disease, then the probability is tiny, namely roughly the same as before the test. If the test is perfectly reliable and never fails, the probability is $1$. That's a huge difference, and the only way to know where between those extremes the actual probability lies is from information about the reliability of the test, which we don't have.

Second, all the correlating properties that you list (ethnicity, sex, relatives, ...) may or may not be correlated among each other. That is, lupus might tend to be congenital in men but not in women or vice versa. Without knowing these correlations, one could only give bounds on the probability by making opposite extreme assumptions on the correlations. These bounds might be slightly more useful than the range "between tiny and $1$" due to the reliability issue, but to get a single probability you'd have to know these correlations or at least make reasonable assumptions about them.

joriki
  • 455
  • 2
  • 6
  • 1
    Well reasoned, joriki! – Kyle. Dec 13 '12 at 01:06
  • 1
    +1 false positive and negative rates are what's missing. But I think the last paragraph is irrelevant. The point with those properties (not the ones in the bullet points) is that you can break it down to a more accurate percentage for your conditional calculation. i.e. you can figure out what proportion of white women have lupus in the US. – naught101 Dec 13 '12 at 01:39
13

My take, as an Epidemiologist: The question really isn't answerable as given, for several reasons:

  • Without very subject specific knowledge, there's no way of knowing if there's effect measure modification between some of those estimates. For example, if you're more likely to have lupus as a woman and more likely to have it as a minority, what happens if you are a female minority member? Are the two independent? Do they interact additively? Multiplicatively?
  • There's a key factor missing: Why is he asking? He's probably not sitting at his desk going "I wonder if she has lupus..." Next week, we're likely not going to get "What is the probability my wife has dengue fever?" There's a reason he thinks this is true, which distorts all those statistics again, as those are population figures, not "Population where a family member suspects you have lupus" figures.
  • The closest thing I could peg as a thing where you could produce a specific value is the "Negative Predictive Value" of the diagnostic test, but a quick googling suggests there are several brands of that particular type of test available, and without more information, you can't really answer it.

What should this poor guy be told? To consult with his wife and her doctor. Trying to apply population level statistics to an individual is exactly not what epidemiological evidence is meant to do.

Fomite
  • 23,134
  • EpiGrad, if you re-post this as an answer to the migrated question, I imagine I can just delete this overall question. Yes, I can see that I do have a delete button below my question. If I don't see that, I will just wait on what you and whuber figure out. – Will Jagy Dec 13 '12 at 00:32
  • Hmmm. Can they migrate this answer including the votes? It's a good answer. – Will Jagy Dec 13 '12 at 00:55
  • Re: the second point. that was included in the original question, but was removed because it was irrelevant on maths.SE. Check it out at http://math.stackexchange.com/revisions/256079/4 – naught101 Dec 13 '12 at 02:27
2

There is no way anyone can give the answer to the question you ask based on the information you provided. Most of the information you do provide might, if it was thoughly completed (you certainly need the correlations between the individual factors you provide) give the a priori prababilty that a white woman with no sibling etc. would have lupus. That probability could also be computed simply by counting the number of lupus cases among people in that exact category compared to the total number of people in that category (but counting lupus cases is not that simple: your question boils down to asking whether your wife should be included in such a count).

However the key ingredient in the question is that your wife has been diagonsed (maybe incorrectly) with lupus. There is no way in which one can derive the effect this has on the probability. Certainly if the diagonosis is worth anything, it greatly increases the probability above the a priori probability. But to know by how much this affects the probability, one would need to know all the factors the diagonsis is based on (and you supply none), together with a very detailed analysis of how those factors correlate with this disease or other ones. So people are right that you should ask an epidimiologist, who has some chance of knowing the relevant information, unlike the poeple on this site. And even for an epidimiologist the question is very hard to give a reliable answer to.