The Science of Science

Science is all about exploring the world unseen. Epidemiologist Michael LeVasseur explores the world of disease and explains critical issues in theory, practice, and popular perception of disease.

[scienceofscience-blog]

[scienceofscience-blog]

Disease du jour: HSV II (Genital herpes)

Herpes Simplex Virus Type II (HSV II) is the most common source of what we call "Genital Herpes". Although the prevalence of HSV I-associated genital herpes is rising (mainly due to an increase in oral sex), it does not contribute greatly to the overall prevalence of genital herpes. HSV II infection does not prevent infection with HSV I and vice versa.

Pathophysiology

HSV II does not always have a clinical presentation (symptoms). When it does, it typically leads to skin lesions on the affected area. HSV II is spread by direct skin-skin contact with an infected person and travels from small skin breaks or abrasions. Viral shedding (when the virus can be transmitted) can occur at any point whether or not an infected individual is symptomatic, although shedding is most frequent directly before and after an outbreak.

Prevention

Despite great effort, researchers have yet to develop a viable vaccine. Antiviral medications, such as acyclovir and valacyclovir can prevent outbreaks and shedding, but are not an effective preventive measure. Consistent condom use and limiting sexual partners are still considered the best method of prevention for HSV II, although condoms are less effective at preventing female-male transmission.

Public Health Concern

Some evidence suggests that HSV II infection increases the risk of HIV acquisition as much as 2-fold. (1) Although the biological mechanisms for this increased risk are still unknown. Increasing this concern is that as many as 80% of HSV II-infected individuals are unaware of their infection. (2) This is due, in part, to poor diagnostic measures for HSV II. The common test cannot distinguish between HSV I and HSV II, although a test that can is becoming more widely used. Co-infection of HSV II with HIV may lead to increased outbreaks of HSV II, which can also increase the risk of transmission of HIV.

Epidemiology

Global prevalence of HSV II is estimated at 535 million with a seroprevalence of about 16.2%. In America the age-adjusted seroprevalence is around 17.2%. HSV II seroprevalence is highest among women (23.1% compared to 11.2% for males) and non-Hispanic Blacks (40.3% compared to 13.7% among non-Hispanic whites). (3)

Herpes simplex virus type 2 (HSV‐2) seroprevalence in (A) non‐Hispanic white people and (B) non‐Hispanic black people by age on the National Health and Nutrition Examination Survey (NHANES) in 1976–80, 1988–94 and 1999–2004. Note: The percentage of persons is weighted. Error bars indicate 95% confidence intervals. *Age‐adjusted using the 2000 US census civilian, non‐institutionalised population aged 14–49 years as the standard. Source: Xu et al

[scienceofscience-blog]

Disease du jour: Introduction

In this section I'll provide some intimate information on some popular diseases. Feel free to submit your disease du jour via the "Have a burning question?" link at the top.

[scienceofscience-blog]

Does significance signify significant significance significantly? Also: in Soviet Russia, Data drives you!

[scienceofscience-blog]

Wizard of Cause Part III: Type I and II Error

It's important to understand that one of the underlying assumptions in any statistical science is that we accept that sometimes we're just dead wrong. In Epidemiology, our wrongitude has a name: we call it α (or Type I error) and it's usually set to 0.05. This means that we accept that 5% of the time what we are reporting about the world as truth is, in fact, false. The corollary to α is β (or Type II error), which is often set to 0.20, which means that 20% of the time what we see as false, is actually true. The problem with this, of course, is that we don't publish results that appear false.

The start of any research project begins with a literature review, wherein we make sure that A) no one else has looked at this before (afterall, who wants to fund something that's already been done?) B) so that we don't have to reinvent the wheel (Oh, Dr. Jones published on a similar population, how did she recruit her subjects?)  and C) to show that there is an historical and theoretical framework from which your research is drawn (just because you think there's an association between eating peanut butter and a propensity for the color mauve does not mean it has a logical foundation...although...). Do we see yet where the problem lies? Let's say Dr. Jones' study didn't reveal any significant associations and she did not (could not) publish. We could be doing the same exact study! What a waste of time! (especially if we get the same results).

This is termed "Publication Bias" and is a major problem in the field. Studies that produce null (insignificant) results are just as important and valid as studies that show positive (significant) results. It should also be noted that "significant" does not mean "important", but rather that the results are not likely caused by random chance. So to be able to say that peanut butter aficionados do not have a greater penchant for mauve than Arachibutyrophobics does not mean your results aren't important, just that the association does not exist.

While we have yet to discuss Sir Austin Bradford Hill's Causal Criteria, his second of nine criteria is consistency. If different study designs among different study populations across different times each show consistent results, then we can feel comfortable that the association of interest is "real". In fact, two similar studies showing consistent results decreases the liklihood of Type I error from 5% to 0.25% (0.05*0.05=0.0025). Publication, therefore, is incredibly important for showing consistency.

Before I hit the snooze button on the intertubes, I thought I'd mention the 95% confidence interval (CI) - an incredibly important concept in the field. Often times you will see a 95% CI reported in parenthesis next to a measure of effect e.g. OR 3.2 (95% CI 1.7-4.3). This means that we are 95% certain that the "actual" measure of effect is somewhere between 1.7 and 4.3 and that 3.2 is the most likely guess. You calculate an odds ratio by dividing the probability of the disease among people exposed (A/B) by the probability of getting the disease among the non-exposed (C/D). The equation for the 95% CI of an Odds Ratio is actually one of my favorites (since I once pulled it directly out of my ass for a bonus question on a Midterm exam). It looks something like this: the log of the Odds Ratio plus or minus the z-stat (which is 1.96 for an α of 0.05) times the standard error (S.E.). S.E. is calculated by taking the square root of the sum of the inverse counts for the exposed and diseased (A), exposed and not-diseased (B), not-exposed and diseased (C), and not-exposed and not-diseased (D). It looks something like this:

ln(OR) (+/-) 1.96*√[(1/A)+(1/B)+(1/C)+(1/D)]

For example, our table below gives us an odds ratio of 2.08=(18/16)/(39/72) and a standard error of 0.397=√[(1/18)+(1/16)+(1/39)+(1/72)]. Plugging this into our equation we get ln(2.08) (+/-) 1.96*0.397. This gives us two equations: ln(2.08) - 0.7781=-0.0457 ln(2.08) + 0.7781=1.509 All we need to do is exponentiate our new 95% confidence intervals and voila! OR 2.08 (95% CI 0.95-4.52). Since our null hypothesis of no association is that the OR=1 and since our 95% CI includes 1, then our results may be the result of random chance and are therefore not significant at the 0.05 level.

                                    Loves mauve       Does not love mauve Peanut Butterer                 18                                16 Non-peanut butterer          39                                72

[scienceofscience-blog]

After mapping humans' intricate social networks, Nicholas Christakis and colleague James Fowler began investigating how this information could better our lives. Now, he reveals his hot-off-the-press findings: These networks can be used to detect epidemics earlier than ever, from the spread of innovative ideas to risky behaviors to viruses (like H1N1).

Social Network Epidemiology explores how our interactions with one another influence our behaviors.

[scienceofscience-blog]

This is a link to Richard Doll and Sir A. Bradford Hill's original paper on the association between smoking and carcinoma of the lung. If you're a smoker, bothered by smoking, or have any interest in the history of risk-factor epidemiology, this is THE study to read.

[scienceofscience-blog]

The Wonderful Wizard of Cause Part I: Introduction

My academic field lives in the Land of Cause (not to be confused with the Land of the Cos). and although this great land is in Euclidean Space, its landmarks are ruled by a few Heisenbergian principles. This series will explore the theoretical underpinnings (and shortcomings) of Epidemiologic research. Topics will include: error, temporality, confounding, effect modification (interaction), mediation, and perhaps a rereading of some of the classics. Stay tuned

[scienceofscience-blog]

Hantavirus is a potentially fatal disease spread through contact with rodents, usually by aerosolized droppings. Health officials are warning Oregon residents to take care when cleaning this season and to take caution if you come upon mouse droppings.

Emilio Debess, State Public Health Veterinarian says, "You can spray it down with bleach and water, soap and water, anything that will prevent any dust from going into the air. Do not to use a vacuum cleaner right away. The point is not to create any aerosolized dust that may carry the organism."