I have a reflexive skepticism about some conventional flu wisdom. There's so much about flu we don't know and even more we think we know that we find out we're wrong about. But skepticism is an occupational hazard of epidemiologists. Our training and practice focusses on detecting subtle biases that can produce misleading interpretations of data. When it comes to the commonly recommended personal protective measures for pandemic flu, our skepticism is all the greater since there is so little data to be skeptical about. The lack of data isn't an accident. If you think hard about it, you can see that studying whether hand hygiene or wearing face masks "really works" is not very easy. But evidence or not, at least for hand hygiene, the recommendation seems commonsensical enough and potentially valuable for other infectious diseases besides influenza, most notably those of the fecal-oral route (e.g., salmonella, hepatitis). Still, while I think CDC's hand hygiene campaign for flu is a prudent idea, as a scientist I'd prefer some data. Yesterday a study from Hong Kong was published in the Annals of Internal Medicine (full text here) providing some data. We provide some details.
The idea is straightforward. Find households where one member is diagnosed with flu, randomly assign the index case's house to one of three intervention groups, and then see if there is a difference in flu transmission within the household. The intervention groups were a control group where only general education was given on a healthy diet and lifestyle and general supportive care of the patient (index case in the household); households where instruction and supplies for hand sanitation were given as soon after diagnosis as possible but in any event within 48 hours; and households where, in addition to hand hygiene, use of a surgical face mask for the index case and all household contacts was recommended and masks provided, to be used at all times in the house except when sleeping or eating and outside the house when with the index case. The hand sanitizers were bottles of Ivory soap and alcohol based gels (both provided at no cost). Households of index cases were recruited through 45 outpatient clinics in Hong Kong, between January and September of 2008. Hong Kong is semi-tropical and the period was prior to the current swine flu outbreak, so this is a study of differences in secondary attack rates for seasonal influenza A or B in a subtropical climate in Hong Kong households. There was no way to compel compliance, but data on adherence to the assigned protocol was collected by home visits on days 3 and 6.
There are a lot of details that have to be attended to in a study of this sort and if you are interested you can read the study, which is quite understandable to the average lay person and available free at the link (kudos to Annals for this). Still, it's worthwhile to describe some typical results to see what the problems might be. The design is simple and in principle most people would assume straightforward to carry out. There were 2750 people who showed up at the outpatient clinics with symptoms of influenza-like illness. Of these, 407 (14.5%) tested positive with a rapid flu test for flu A or B. While these tests aren't particularly accurate, during flu season, when many of these cases were recruited, the proportion of false positives is fairly small, and while there are a lot of false negatives (people with flu who test negative), not including them in this study isn't a serious limitation. If you want to be precise, you'd say this is a study whose results are limited to households where one of the people feels sick and has a positive rapid flu test. The effect of the interventions might be different on people with flu who test negatively, but I don't think this is a big problem. We still get some data points.
So we have 407 index cases, which means we should have results for 407 households and the contacts in those households. But results are presented for only 259 households. Why the discrepancy?
Of the 2750 potential index patients, 407 (14.8%) were found to have influenza A or B virus infection according to the rapid test result; these persons and their households were randomly allocated. In an unintentional deviation from that protocol, 49 of the 407 randomly allocated persons had a household contact with influenza symptoms at recruitment (a potential co-index patient). We also randomly assigned 6 of 407 persons who had symptoms for slightly more than 48 hours.
After random assignment, 76 (19%) of the households declined home visits or could not be contacted after numerous repeated attempts. We implemented the interventions in 331 households. After initial home visits, 9 households declined further participation and were excluded from further analyses. Thus, 322 (97%) households completed follow-up. Influenza could not be confirmed by RT-PCR in 16 of 322 index patients in these households at baseline, and 16 further households were therefore excluded. A further 47 households were excluded because 1 or more contacts in the households had RT-PCR-confirmed influenza virus infection at baseline. Three household contacts declined to participate and were excluded from analyses. We evaluated and compared secondary attack ratios in the remaining 259 (64%) households, which included 794 household contacts. One hundred sixty of 259 (62%) index patients had influenza A virus infection, and 99 (38%) had influenza B virus infection. (Cowling et al., Ann.Int.Med.)
So from the original 407, a little more than a third of the households were excluded for one reason or another while of the ones left, about a third were influenza B and the other two thirds influenza A. About 1 in 5 households that would have been studied were lost at the outset because of no response or refusal. It is possible they were somehow different than the ones left, although it is hard to see how. In addition the studied population is now only those cases with RT-PCR confirmed cases of flu A or B (the paper also has results for some other outcome criteria, but I'm trying to keep it simple to make some points).
I don't think these things dramatically affect the validity of the study, but I mention them to emphasize that conceptually simple study designs can still be very messy in the real world. We often get comments here suggesting that CDC should study this or that and implying it would be a simple and quick study to do. Simple, quick and truly informative studies exist, but they are rare. In this case the statistical issues are also tricky. The outcomes are right censored (they stop at 7 days; any cases after 7 days aren't counted) and this is a cluster randomized trial, i.e., it isn't individuals that are randomly allocated but clusters of individuals (households). That means you have to model within cluster correlation. The authors had some data to allow this from a pilot study, and it appears they used good methods to analyze the data. But this was not just an exercise in counting. There was some statistical heavy lifting involved.
So what did they find? The main measure was the secondary attack ratio for individuals, i.e., the risk of infection in a household contact of an index case. There were 60 contact cases in the 759 household members for an attack rate of 8%. This is in accord with data from the same group's pilot study which we posted on a little while ago. 49/259 (19%) of the households were involved, mostly with one additional case, but 7 households had 2 cases and 2 households 3 cases. So while the household attack rate was about 20%, the individual attack rate was only about 10%. If the index case was an adult, there was less chance of secondary cases, while if the index case was a child there was a greater chance. A sick child in a household is more likely to pass on the flu to others.
What about hand hygiene and face masks? Looked at overall, there was no statistically significant difference in secondary attack rates between control and intervention households. But statistical significance just means that random variation can't be taken off the table as an explanation for the differences that were seen (assuming the correct error and probability model and no bias). If there were a bigger sample that could change. And indeed the raw data were different. The attack rates were 24% in the control group, 14% in the hand hygiene group, and 18% in the facemask plus hand-hygiene group. The control group had the highest attack rate, followed by the intervention groups. Was the variation just random? The statistical model said that was possible, so we don't know. It could be real or the result of some systematic error (bias).
But in a further analysis, planned prior to the study, differences in households which began interventions within 36 hours were looked at separately, and there a difference was more clearly apparent. An adjusted odds ratio for households of .33 was found for the hand plus mask group compared to control. This means that the odds of having a secondary case was one third of what it would be if you were in a household in the hand plus mask group compared to the control group (odds are always slightly larger than risks/probabilities, but you can think of them similarly). The number .33 wasn't very precise. The 95% confidence interval went from .13 to .87, but it was sufficiently tight to make random variation an unlikely explanation (assuming the correct error model and no bias). Said another way, there was evidence that being in a household in the hand plus mask group made it less likely you would have a secondary case in the house, given that you had one person sick. The intervention was effective.
Notice I said "being in a household" rather than using hand hygiene plus a mask. That's because there was good evidence that compliance wasn't very good, especially for mask wearing. When there was mask compliance it was more likely to be for the index case than the contacts. But on the other side, consider that people weren't given the means and information for intervention immediately. There was a lag. The evidence from this study suggests that the earlier you implement hand hygiene and/or mask wearing the more effective it is, so this probably underestimates the effect. This makes sense. People probably shed more virus in the first day or two of illness than later, when their defenses start to win out over the bug and they get better. There was no statistically significant difference between the hand only and the hand plus mask group in the first 36 hour comparison (odds ratio of .72, CI .21 - 2.48), but again, this might be just a sample size issue. Putting a mask on an index case right away makes sense and the data are consistent with an effect for the 36 hour group (but not, as we saw, overall). It's also possible that this difference is just a random variation and without meaning.
There are a number of analyses presented in this paper I haven't discussed, but the take home message for me is that it does present some hard evidence that hand hygiene and household mask wearing might have some efficacy for preventing flu transmission within a household. How that translates into workplaces and the general environment is another question. And whether the results would be better with tighter masks (e.g., N95) rather than surgical masks yet another. We'll sum it up with the authors' own conclusions, which we consider reasonable and prudent:
In conclusion, our results suggest that hand hygiene and facemasks can reduce influenza virus transmission if implemented early after symptom onset in an index patient. During a pandemic, resources may not be available to isolate all infected individuals, and home isolation of some patients may be required. Our results directly inform the personal protective measures that should be taken in such a scenario. Our results also support the use of these nonpharmaceutical interventions in public health control measures against interpandemic influenza in annual epidemics.
- Log in to post comments
This was a hard study to do, no question about it. Can you imaging making 3 home visits to 407 families over 8 days!
While I understand why the authors decided to eliminate so many families from analysis, by doing so they fell below the number needed within each group to give the researchers a reasonable chance of showing a significant difference between interventions.
We can quibble about this but either we live by the evidence based medicine sword or we die by it. If we are going to require interventions to meet statistical significance before being widely adopted, then we need to stick with that standard.
When an underpowered study like this one is published that fails to be conclusive using the standard we apply to all other interventions, then we need to say this and conclude that there is nothing to conclude about the effectiveness of hand washing and mask wearing.
Of course the big problem with this study was getting the patients and their families to cooperate with the protocol. The authors point this out in the statement below:
"Limitation: The delay from index patient symptom onset to intervention and variable adherence may have mitigated intervention effectiveness".
It is difficult to get medical personnel to adhere to tight infection control protocols, so I can imagine how difficult it is to get everyday people to do so and do it well.
The latency in the time from when the index case became symptomatic and the interventions were implemented was the major problem with this study and why it probably failed to show significant benefit. Since an index adult case begins shedding virus about 48 hours before becoming symptomatic and for several days afterward, even if the first home visit occurred 36 hours after the patient's first clinic visit, the family would have been exposed for 3.5 days to the contagious sick person without the benefit of the intervention.
The statistically significant reduction in relative risk for secondary infections within the family for those who began the intervention within 36 hours of the index patients initial visit then was surprising; in fact unexpected. So unexpected that this result could have been due to chance alone, a problem seen in underpowered studies with unexpected outcomes.
However, this study is already being ballyhooed in the media and will be a cause celeb for a while despite its limitations. TPTB will point to it with enthusiasm and in the end will use it to reassure the public that PAN-flu can be defeated by HYGEIA.
During pandemic conditions for the vast majority of families with an index patient, there will be no initial clinic visit, no home nurse visits, no masks, no alcohol based waterless hand cleaner. Their might be PSAs, gossip, rumors, and old wives tails but not much else.
The circumstances that the intervention reported upon in this thread then are totally artificial and so unlike what the real world conditions would be like and even then, they failed to be effective.
Grattan Woodson, MD
Gratt: Statistical significance is an artificial convention meant to provide some information about one of the things that influences outcomes (random variation). It should never be used as you suggest, as a bright line. That leads to ignoring information of public health significance and giving prominence to information that may be of no public health significance. This paper presents some of the very few data points in the literature on an important topic and the question is how we interpret them, which is what the post was about. Note that by your criterion, the statistically significant findings for the 36 hour intervention would seal the case as being useful, but of course it doesn't. There is much else to consider. The important thing is that a scientific question was posed and an attempt to answer it with a reasonable study design whose power might have been sufficient given the size of the effect. That the real world intervenes is a given. Study designs are just designs. You still hve to carry them out and that's a difficult thing to do. I am a skeptic about handwashing and certainly about masks, but I do look at data when it's relevant, and these data are relevant and suggest some efficacy. The question about whether the study circumstances are unrealistic for the reasons you gave is totally separate and not relevant to the scientific question.
These results correspond very well with the earlier Australian study found (free and in its entirety*) here:
http://www.pubmedcentral.nih.gov/articlerender.fcgi?tool=pubmed&pubmedi…
It also found that compliance was a major issue with facemask use (both for surgical masks and N-95 equivalents), but that there was evidence that they helped when used consistently.
*One of the many unglamorous but important things your tax dollars pay for through the CDC (the journal Emerging Infectious Diseases) & the NIH (PubMed and PubMed Central)
"I have a reflexive skepticism about some conventional flu wisdom."
Really?
Did you know that all the early research that created the conventional flu wisdom that resistance to Tamiflu would weaken a virus was based on almost ten years of research on Relenza(before Tamiflu had been invented)?
Did you know that Gilead published the first clinical trial of Tamiflu and included the fact that H274Y mutated influenza A was already in the environment and oblivious to Tamiflu?
Gilead, in the discussion of the data, stated that H274Y would be irrelevant to humanity.
Did you know that the recent announcements that Relenza resistance is real, isn't the claim of the scientists who wrote the original article,
http://pubget.com/paper/19641000
I'm still waiting for the facts behind the "Relenza will bronchospasm you to death" conventional flu wisdom that ends most discussions about zanamivir v Tamiflu at Effect Measure.
complicated and confusing presentation.
Apparantly facemasks only make it worse, so let's ignore them
and consider hand-hygiene alone.
I think the main question is:
if a household member is infected, what is the probability to get it too
a) without measure
b) with optimal hand hygiene but otherwise same behaviour.
my interpretation : a)=8% b)=2.5%
(but not according to table 3, which confuses me)
The take-home message: Regardless whether or not the hygiene measures 'work': Even if they do, their effect is disappointingly small, i.e. barely detectable above statistical significance...
h1n1_watcher: On the contrary, the estimate of the effect measure (OR=.33) is very large. Statistical significance gives info on precision, not magnitude of effect. It is important not to mix those things up.
The main thing about hand hygiene is that it needs to be compulsive and carried out properly. We touch our face (nose, mouth or conjunctiva) up to 200 times a day and each time is equal opportunity for the introduction of infection if the hands are not sanitized. We need to better educate the public on correct technique and to carrying out such practices before touching their mouth or nose or eyes--it is a hard thing not to do since our brains are hardwired to do so.
Check out the 3 OpEds that appeared in the sunday edition of NY times. It discusses hand hygiene in schools, and on planes and at airports and in public.
Revere (and others),
This is off-topic and comes from the more hypochondrical side of things, so please forgive me.
If all goes according to my original plans, I will be a freshman in college at the end of the month. Where do you think the best place to sit in a classroom to limit my exposure to H1N1 (from others coughing/sneezing/breathing on me) would be? I was thinking that one of the front rows on an aisle, close to the door would be a good place to sit, if possible.
I do understand that this probably won't have a big impact on overall exposure, but I thought it was worth pondering. My college is small-medium in size, so classes will probably be fairly small (maybe about 30 students per class?). Otherwise, I plan to carry a lot of hand sanitizer, live off-campus, and take the recommended "flu-fighting" vitamins.
If one uses a hand sanitizer, for how long is it effective? 30 minutes? 2 hours? Until a new3 source of viruses are encountered? I haven't seen an6 information on this.
Thanks for your response Revere.
Let me tell you how much I appreciate your bringing these interesting topics to the Internet everyday for discussion. This is an immense burden that some may not appreciate. Of course some may say this is what they pay you the "big bucks" for but you and I know there are no big bucks rather you all do what you do from the goodness of your hearts. You are acting in the interest of informing the public and for no other reason and for this, we all owe you a significant debt of gratitude.
It takes a great deal of effort to search the news and literature to simply find an event or article to comment on but then to find the jewels within the dustbin and make them into a cogent assessment is where you and all the Reveres shine. Thank you very much.
Please understand that in my opinion, what you Reveres are doing is the heavy lifting and what I as someone who posts here is doing, especially when I take a position in opposition to the editorial content of your efforts should be seen as the light lifting. It is very difficult to create a great Internet site with multiple daily content but very simple to offer criticism.
Again, thanks for establishing a frame of debate on pandemic influenza, on a daily basis that provides all your readers with a focus, one that informs and challenges us.
OK, enough with the praise, here comes the thorns.
If there is any benefit in the application of hand washing and use of face masks by the family with a member ill with flu, it is pretty clear to me that these interventions must be implemented very early during the illness of the index case for them to have optimal impact. As you pointed out, the article under discussion found that the spread of flu within the family with an index case can be reduced by 66% when implemented early.
What's more these interventions must be adhered to comprehensively by as many in the family as possible during the course of the illness of the index case to ensure a positive outcome.
I think it makes sense to use gloves, masks, and air purification via UV or ion methods and plan to implement them all if required. It also seems pretty likely to me that most folks including me who implement these measures might find them effective for a while but in the end they will fail.
That though is not the point of why we should use these Non-Pharmacological Interventions (NPI)s. The point of these and all other NPIs is to slow the rate of transmission of the virus through the population. No one thinks that these or any of the other NPIs including my preferred method, social distancing most particularly early school closure has any chance at all of preventing the spread of pandemic influenza to all people given the fullness of time.
The great benefit of all NPIs is to give us more time to manage the most severe effects of the pandemic, which is an excellent reason to support them. What we want to avoid is have a tsunami of severe cases come too closely together in a way that overwhelms our medical ability to manage them whether that be within the hospital or home.
Grattan Woodson, MD
Yes it benefits family members to wash and if active infection is in the house having the infected wear a simple face mask to prevent the introduction of droplets into the environment.
I work in what is affectionally termed "the pit" in the house of medicine or the ER. I am exposed to infected individuals daily--especially during seasonal influenza season at the peak. I get vaccinated, wash my hands like an anal compulsive psycho type, avoid (but keep catching myself) from touching my nose, keep myself well hydrated and finally I take sambucus nigra berry extract when the state health dept flu curve peaks each year for 2 weeks. The result: I have not had the flu since 1998 and I typically get an URI every 3 yrs rather than the average 2.5/yr the NIH says the average Joe gets. To the soon to be college student: chance plays a role in regard to your exposure so where you sit is meaningless. Practice the transmission interventions above and you will do just fine and good luck but don't consider becoming a doctor since health care is a mess right now.
Gratt: First, thank you for the kind words. They mean much. As for the "thorns," I pretty much agree with what you say. The issue for me is whether there is any evidence hand washing is effective (I skeptical and even more so about masks) but the data seem to suggest they are, at least to some (unknown) extent. They also modify behavior in other ways and affect other disease agents, so that's all to the good. I also think the studies probably underestimated the effect because it was started late. If people think they work (and now there is some data to suggest it) then they can start immediately and get more effect.
BostonERDoc: I know a number of people who act as you do (including your state epidemiologist). I'm not convinced there isn't some magical thinking involved, but whatever works, right? At least now there's some data to keep us at it.
It is interesting to note that mask compliance was greatest with the index cases. Might this imply that masks worked more as a source control than as PPE?
Another thing to consider is that this study was done in Hong Kong. The recent animal models from Mt Sinai suggest that at high relative humidity there is no aerosol transmission of influenza -- only direct contact transmission. If this is true of transmission among humans, then hand washing might be very useful in Hong Kong, and in the US in the summer. But, when the weather cools and relative humidity drops indoors (and especially absolute humidity drops as suggested by more recent analyses of the animal and other experimental data), we may be in a different situation.
The use of masks by the untrained may increase transmission. Hands contacting mask, then eyes and nose is rather easy. Improper donning / doffing. Re-use of same mask. . .
Not a stretch for me to see:(worst to best)
1) Control group
2) Hand hygiene and mask
3) Hand Hygiene
Many radiological contamination's in nuclear power were face / respirator related. Any resulting intakes resulted in "events". That, and the fact that work could be done quicker without them (lowering total body dose) was the reason respirators are only seen in highly contaminated (with airborne) areas within Radiologically Controlled Areas (RCA's) in todays nuke world.
What was that first paragraph? Get to the point. The first paragraph would have made a great last paragraph. Actually, it should have been edited out. As should half the rest of this post.
You have good stuff to say, but learn how to self-edit.