Schneier on Security
A blog covering security and security technology.
« Using Google Earth to Find Unguarded Houses |
| CCTV Cameras »
June 26, 2008
Fever Screening at Airports
I've seen the IR screening guns at several airports, primarily in Asia. The idea is to keep out people with Bird Flu, or whatever the current fever scare is. This essay explains why it won't work:
The bottom line is that this kind of remote fever sensing had poor positive predictive value, meaning that the proportion of people correctly identified as having fever was low, ranging from 10% to 16%. Thus there were a lot of false positives. Negative predictive value, the proportion of people classified by the IR device as not having fever who in fact did not have fever was high (97% to 99%), so not many people with fevers will be missed with the IR device. Predictive values depend not only on the accuracy of the device but also how prevalent fever is in the screened population. In the early days of a pandemic, fever prevalence will be very low, leading to low positive predictive value. The false positives produced at airport security would make the days of only taking off your shoes look good.
The idea of airport fever screening to keep a pandemic out has a lot of psychological appeal. Unfortunately its benefits are also only psychological: pandemic preparedness theater. There's no magic bullet for warding off a pandemic. The best way to prepare for a pandemic or any other health threat is to have a robust and resilient public health infrastructure.
Lots more science in the essay.
Posted on June 26, 2008 at 6:58 AM
• 35 Comments
To receive these entries once a month by e-mail, sign up for the Crypto-Gram Newsletter.
This should be good, since it will single out the drunks for extra screening, and some drunks get belligerent when harassed.
Buy stock in stun gun manufacturing.
The summery is a bit unclear; am I reading it correctly when I conclude that there is a 84-90% false-negative rate, and a 1-3% false-positive rate?
That doesn't make it totally useless, however. It is absolutely useless when trying to identify sick people to prevent then from traveling, and thus spreading the disease, it is useful for identifying a possible source of the disease; if 10% of the travelers from a certain location are measured to have a fever, and the average is only 4% (false-positive and a few people who really do have a fever), something is going on.
If done properly, accounting for the large measurement errors, trend analysis could be useful as an early detection system.
Poor positive prediction of people proportions for pandemic prevention? Posturing piffle! Previous poster propounds plethoras of practical possibilities.
Sorry, I just HAD to! I couldn't help it. It's a disease.... ;-)
It's the reverse of that. 10-16% correct positive is 84-90% false positive, so five in six to nine in ten of the times that the device indicates a fever it will be wrong.
Part (but not all) of the difficulty here is that different individuals have different baseline temperatures and fever responses, so unless you collect tremendous amounts of data and correlate it with the reading, it's not going to get a whole lot better.
the problem is that people incubating the disease are not sick per se, but will be in a few days, rendering any "sick people firewall" useless
"pandemic preparedness theater"
Wow, Bruce, you really hit the jackpot with your coining of "security theater". Your meme is both spreading and mutating. I hope it really does become "pandemic"!
When and if they get this technique down pat, can we get it implemented at the entrance to my workplace?? America has a huge Typhoid Mary problem.
The wording used for the descriptions of positive and negative predictive values is about as clear as mud.
As far as I am aware surface IR temprature measurment of people engaged in a stressfull and physicaly active activity (draging your bags through airport checks etc) is very unreliable and has a variance of 2 or more degrees C (3.6F). This variation is one of the reasons why polygraphs use surface temprature and breathing "delta" not absolute.
To detect the early stages of feaver you need a core body temprature reading (delta).
Somehow I cannot see the general public putting up with body cavity measurment (tympanic, oral or other) two or three times just to travel.
Nor are the health authorities going to be two pleased either as the re-use and handaling of thermometers and other non surgical equipment for medical examination / monitoring is one of the largest vectors in the spread of things like C-Diff / MRSA / VRSA / strep etc. Esspecialy when those being examined are immuno surpressed (not well or under the stress of work, jet lag etc)...
Further for a large number of the viral infections that you would want to stop people can be most infectious in the 24Hours before they develop noticable symptoms such as fever.
All that being said you need to weigh it against the risks, SARS had something like an 80% (initial) fatality rate. Some scientists and researchers are talking about a possible 90% rate if H5N1 becomes aerosal transmissable in humans.
Therefor install the remote sensors and by all means scan people in normal times and use the informatio to produce local predictive values. If people guenuinlay do have the first stages of a feaver offer them medical assistance/advice and note their details and if possible their travel details.
In times of risk by all means step things up most (but unfortunatly not all) people would be unlikley to object unless there have been to many false alerts previously.
One of the problems is that fever as a sign of contagion is not reliable. Screening assumes that 'passenger with fever is contagious, so we keep them off the plane/out of our country and we're safe' ... but it ignores the fact that they've already been standing in line with other travelers, who may very well contract an airborne disease, without immediately exhibiting an abnormal temperature.
We're back to the basic lesson of 'fail-safe' -- what happens if the screening misses someone who is carrying the disease, but not exhibiting a temperature?
The first time I saw this was in 2003 in Hong Kong. The scare at the time was SARS.
The lousy thing was, you were warned several times before you hit the screening that you were going to be subjected to screening and possible quarantine. I had been traveling for about 30 hours and had had several beverages on the flight. I was nervous about being a long way from home and the possibility of being detained.
I glowed a red and yellow head, and passed through without issue.
Obviously, they should be using gamma ray screening. That is guaranteed to prevent any kind of pathogen from spreading, regardless of its bacterial,viral,proton, or human-like nature.
Some people naturally "run hot". It's never the normal that creates problems in mass screening, it's stuff greater than +/- 1 standard deviation from the mean.
OK, maybe I missed something. Is this a 'close enough' solution issue. If I have a device that gets practically 100% of the target population but generates say 10% false positives, isn't it a useful 'level 1' device.
Everyone that scores 'positive' could be more accurately tested in a second pass (or just excluded period).
By use of the 'level 1' filter to reduce the time consuming testing to say 10% of the population, I've saved about 80% of the time required to test everyone.
Of course, this implies a VERY LOW false negative rate and a reasonable false positive rate.
Is that potentially the goal here? (Asks the guy still spinning over the wording of the failure rates).
From the data, it appears simply that they did not use a thermometer that accurately measures core temperature based on skin measurements. This is a problem with the thermometer, not the screening itself.
Thermometers such as those from Exergen (www.exergen.com) have been clearly demonstrated to be at least as accurate as a rectal thermometer, and would have been the correct choice for this application.
During the incubation period when you are not feverish, but shedding plenty of contagion escapes these people how? By the time you are feverish, you are beginning to be too sick to travel.
@ Clive Robinson
>> If people [genuinely] do have the first stages of a [fever] offer them medical assistance/advice and note their details and if possible their travel details.
Where I live ( US, Upper East Coast) - Hospital emergency rooms are already overburdened with folks who have little other access to medical care.
Because our system gives few/no good alternatives, those who cannot afford health insurance will take their children to the Emergency Room for every fever, earache, and sniffle.
If we install any kind of health screening in airports, we invite another avenue of system abuse.
Maybe a lot of folks won't do this because they can't get a ticket; they will continue to go to ERs.
If one can get access to credit (good or stolen) - then buy a cheap or refundable ticket... then they can get a free health screening at the airport.
Meanwhile, all these sick folks - regardless of why they are there - are still spreading germs while they wait in line.
"The best way to prepare for a pandemic or any other health threat is to have a robust and resilient public health infrastructure."
Too bad such a thing is fantasy, never has been and never will be achievable.
But there are millions (who am I kidding, billions!) to be wasted in PHIT (public health infrastructure theater) along the way!
SPEND! SPEND! SPEND!
Perhaps however they can enlist a hoard of grandmothers to aid in sickness detection. My mother-in-law once claimed: "I smell sore throat", for 15 years I've been trying to devise a use for this skill - now its here.
"...clearly demonstrated to be at least as accurate as a rectal thermometer..."
ohgod please don't let the TSA read this. Next thing you know, we'll be told to "remove shoes and coats, and bend over..."
Great, is this going to turn into the next terror scare? Pandemic flu terror attacks?
The next airport sign:
"Sneezing will result in a fine up to $10,000 and/or 2 years guantanamo."
I can see it now..
The problem has to do with prevalence. A 10% positive predictive value means that 10% of those that get picked up are actually positive (in this case for a "Fever" and not actual sickness).
So if 1 in 20 people have an elevated temperature by this device, in reality only 1 in 200 will end up having an elevated temperature. Thus 10 people will need further "screening" to find the 1 who has an actual temperature. Multiply that by the 100s of thousands that fly daily and you suddenly have a huge number of people that some sort of additional scrutiny. That assumes just 0.5% of the population actually has an elevated temperature. Now, add a hot day, lots of people, heavy bags (they charge for bags now, so more people are carrying more on), running to lines, and general stress and you may find that the number of positives is closer to 40-50+%.
That is just for "elevated temperatures." That still hasn't been correlated to illness. In a hospital, fever and infection are going to have a much higher correlation and there is a much higher incidence of people with infections in the hospital. Translating that to a group of non-sick individuals is a big leap.
All and all, this a crap technology trying to solve a problem that is just speculation, by applying the wrong data.
The typical security sales tactic:
* Create a security theater threat.
* Sell a product that "prevents" the threat.
* Make lots of money!
The threat is almost always overblown and the product rarely protects like you think it would. But it always sounds great! I know I don't want to fly with sick people!
Bruce, you should start cataloging these products so we can vote on our favorites. Then you can give an annual award to the best security theater product in addition to your essay contest award!
In this case, if they want to prevent people with a fever from flying why don't they make everyone take their temperature with a disposable thermometer at the security checkpoint? This is a device with proven accuracy that has been used over a long period of time and I can't imagine the cost is very high (and you could tax the flyers to recoup the cost!). But then that doesn't help the IR camera salespeople make their numbers.
"The typical security sales tactic:
* Create a security theater threat.
* Sell a product that "prevents" the threat.
* Make lots of money!"
You got that right. Now extend that to the rest of the government agencies, and you'll have completed your thought.
The typical govt. agency sales tactic:
* Create a [environmental, corporate, social] theater threat.
* Sell a [service, i.e., "We'll regulate the bastards!"] that "prevents" the threat.
* Make lots of [money for your cronies and political power for yourself]!
That is kind of my point, a 'first pass' device to reduce the issue from screening 100's of thousands to 10's of thousands. (Important note: this is the wrong environment for this process in my opinion, but that doesn't stop it from being someone else's magic bullet).
My whole point was: does someone think this is a good method reduce the body count that needs further screening to a manageable level?
I think they are wrong, law of large numbers says BIG number times 2 digit percentage is still a BIG number. But no one said this was thought through more than any other bureaucratic security measure.
My question/statement was more toward: is this deliberate security theater or is it misguided 'science' as a motivation for this 'security process'.
"Next thing you know, we'll be told to 'remove shoes and coats, and bend over...'"
What, that hasn't happened to you yet?
Two pass screening is an accepted technique for all sorts of disease - a sensitive *cheap* first pass, followed up by a specific second pass. The second pass is usually more expensive - costly or invasive. (TSA run a >two-pass system, btw, which is one part of the things they do that makes sense).
This device might be useful as a first pass.
Applying the first pass at an airport is a mega-dumb idea. In this environment, a low predictive value means that it's a device for causing trouble.
Commenters should learn Bayesian statistics before commenting ;-)
"Somehow I cannot see the general public putting up with body cavity measurment (tympanic, oral or other) two or three times just to travel."
The sheeple will put up witjh anything governments tell them to.
And actually modern devices for measuring temperature in the ear are very fast, on the order of 5 seconds. It could be done.
But I agree, it's stupid. Lots of intercontinental tourists pick up minor infections on holiday (they encounter unfamiliar bugs to which they have no natural resistance). Preventing them from going home
is not only pointless, it's inhumane in the extreme.
The problem with using this as a first-pass is the terrible false-negative rate. For such a device, a 10% false-positive might be acceptable, depending on what percentage of travelers actually do have a fever.
Random selection for a more accurate check is pointless, because the purpose of random selection is to scare people to not bring illegal stuff. It only works for things people do on purpose, or for data collection for statistical purposes.
People who are found to have a fever on their flight home, could be put on a "quarantine" plane, with all the other sick people. If you weren't already sick, you are now, and if you were, and because of that, you have a weakened immune system, you have multiple diseases now. So the question is: what are they going to do with people who are sick, away from home?
It's like software development; if you can't recover from an error, there is little point in detecting it. Like checking the return value from printf() (to console). What are you going to do when it fails? Print an error message?
I do feel a movie plot security device coming up: the "inflatable airplane health protection bubble". You step inside before standing in line, with some food and drinks, and stay inside during the flight. Afterwards, it must be discarded, because the outside might be infected. Only 49,99! Isn't your health worth 49,99? Full size, 4-person family bubble for only 149,99! Save 49,97!
This sort of screening doesn't make sense as a normal every day screening measure. However in the case of a serious epidemic that is actually in progress, it does make some amount of sense.
I live in Taiwan and was here during the SARS outbreak. At the airport when entering and leaving you had to have a temperature check and also fill out a questionnaire about possible symptoms.
Not only that but practically every building or large store had temperature checks before you could enter. It was pretty much impossible to go about your business without getting temp checked a few times a day. If you had a temperature you were legally required to go to a hospital to be examined.
That may seem like a lot of effort, but Taiwan was one of the countries seriously affected by SARS. In this context, it was important to do as much as possible to identify every possible case of SARS.
A lot of work? A lot of false positives? Yes, but in a serious epidemic it is worth it to be able to quickly identify possible cases and be able to trace and contain the spread of the disease.
The problem is that these measures are difficult to get rid of. A year after it was clear that SARS was over, airports were still screening for symptoms. And despite Taiwan not being affected by bird flu, there are a lot of stringent rules (e.g. no birds or bird products on public buses) already in place to control it.
You forgot: "... pandemic preparedness pageant."
I apologize for my alliteracy.
5. a pretentious display or show that conceals a lack of real importance or meaning.
Seems like they'd be farther ahead to use trained animals. Dogs are obvious, but perhaps pit vipers would be more accurate.
The idea of temperature checks depends very sensitively not only on the curve of infectivity vs fever, but also on the rate of checks. Perhaps if people are absolutely not infective until they show a fever, and they are checked on a very short cycle you would get somewhere. But even then, sending the person with the fever to the hospital for checking is only a half-measure. To do it right for stopping pandemic, you'd have to send everyone who had been within 3 meters or so of them since their previous temp check to the hospital for checking as well.
In an airport, of course, by the time you ID someone with a fever who actually has what you're trying to stop, all the people who were in line with them or walking past them or handling their documents have already dispersed.
If anything, fever screening at airports warrants further study before labeling it "pandemic theater". There are many infectious diseases that could be caught by effective screening. Someone at the original article linked this CDC research summary:
"Airport fever screening alone identified 40 (83.3%) of 48 of all imported [dengue fever] cases identified by the active surveillance system."
I have quite a few problems with this essay.
First, I suggest interested readers read the original paper (which is linked), which has not yet been peer reviewed. You may notice several serious issues with the study. Notably, they did not attempt to use a variety of methods to see which -- if any -- would be most effective, rather they evaluated just one set up. The essay on the "progressive public health blog" that Bruce quotes then leaps to the conclusion that all possible remote temperature sensing methods are too inaccurate to be useful, which on any fair reading is completely unsupported by this study.
Even more seriously, the study used just one measuring instrument, and they found that it "underestimated body temperature at low values and overestimated it at high values" (page 3.) Surely this must suggest the strong possibility that the instrument is, shall we say, broken !?!
And there are other problems. For example, these researchers took the measurements of their tympanic infrared thermometer as a gold standard of measuring core temperature, but in fact there are serious doubts about the accuracy of some models of those, too (see http://www.biomedcentral.com/1471-2296/6/3 )
In short I do not think this essay can be used to conclude anything except that some people have a political bias against medical screening programs, which is very, very sad.
A couple of "by-the-ways" about SARS: it is characterised by high fever -- usually much higher than the 2°C they were measuring in this study -- which appears very soon after infection, and usually several days before other symptoms such as coughing. Further, contrary to what some commenters on this page have suggested, there are no known cases of contagion before the appearance of obvious symptoms. This is why fever screening was recommended by the WHO in the first place. The effectiveness of the test is of course greatly increased by specifically screening arrivals from a location known to have suffered an outbreak. Such a patient is regarded as a "suspected case" and subject to two more layers of progressively more accurate screening.
The method WAS effective; all places which implemented it contained their outbreaks to a scale two orders of magnitude smaller than the first outbreaks, and SARS vanished (there have been no cases reported for 4 years.)
Also, while the case fatality rate of SARS may have approached 20% in certain limited groups, the average is less than half that. And like 'flu, the fatality rate is also much higher for the elderly; the fatality rate among young people is possibly below 1%.
It is important what disease is being screened for.
As Roger points out, SARS causes symptoms including easily detectable fever before patients become contagious, so fever screening is (and was) very useful. For mosquito-borne diseases such as dengue that don't spread directly from person to person the fever also comes early enough to be useful (and even imperfect screening is worthwhile). For influenza, on the other hand, it's much less useful since so much of the transmission happens in the presymptomatic part of the illness.
It is also probably worth pointing out that the interventions in the SARS outbreak, like those in later phases of the animal foot-and-mouth outbreak in Britain, were mostly not made up out of thin air. The politicians and public health officials were advised by mathematical epidemiologists who tried (pretty successfully) to model the likely effects of interventions. This required simplified models, since calculations had to run in hours rather than days, but it still isn't anywhere near `security theater'. It's a pity that anti-terrorist security is so far behind infectious disease security.
Schneier.com is a personal website. Opinions expressed are not necessarily those of BT.