Basics: the death certificate

by revere, cross-posted at Effect Measure

A couple of days ago we discussed the murky questions surrounding the death of accused anthrax attacker Dr. Bruce Ivins. At the center of stipulating the cause and manner of death were the procedures for filling out the state of Maryland’s death certificate by the medical examiner. Determining and recording the cause of death is important for many other things besides the circumstances surrounding the unexplained deaths of anthrax attackers. In the US you can’t legally dispose of a body without a properly recorded death certificate and its a document survivors use for all manner of other purposes, from claiming a “compassionate fare” discount from an airlines to insurance money. Compiling the information also forms the basis of many public health decisions and policies. As an epidemiologist I have used death certificates for much of my research work. Here is a brief guide to your future paperwork (you won’t have to fill it in, of course; but someone will).

Standard birth and death certificates are relatively recent in the US. When my mother was born in 1904 she was among the first to have her entry into the world recorded on a standard birth certificate. When she died last year the event was recorded on a standard death certificate which was barely older than she was. Collecting this information is a local function, so the system is highly decentralized. But it is also standardized. Keeping vital records (births, deaths, marriages, divorces, etc.) is the responsibility of the states and five territories of the United States, referred to as “registration areas.” Two cities, the District of Columbia (“Washington, DC) and New York City are also their own registration areas. While the registration areas are “independent,” they use a uniform standard for collecting core information in each category. The means for doing this are by Model State Standards and forms. If you want to see what a standard Deatch Certificate looks like you can find a .pdf here. Page 2 has brief instructions for what the entry lines mean.

One of the things you can see is that the Final and Underlying Causes of Death and Other Significant Conditions (item 32) are narrative (free text) entries. If you want to keep statistics on causes of death you have to categorize them so that “stab wound tear of subclavian artery” and “severed subclavian artery from knife wound” are counted under the same heading. Figuring this out is the job of a specialist, called a nosologist, a disease coder. The codes change with time as medical knowledge changes and so far there have been ten different major revisions to the internationally agreed upon coding, called the International Causes of Death codes. Thus since 1999 most of the world has been using the once designated ICD-10. Every time the coding changes there is an issue with the comparability with the previous version. Usually what is done is to spend a year or two coding deaths simultaneously by the old and new ICD versions so that a “cross walk” can be devised to allow comparisons, but sometimes this is pretty difficult so one can see sudden discontinuities in cause specific death rates with a change in ICDs that is due only to a change in coding. It’s a complicated world.

The desire to make comparisons between countries, states or the same state over time is also the reason for presenting mortality statistics with “age adjustment.” For example, CDC, via its Morbidity and Mortality Weekly Reports publication just released 2006 age-adjusted death rates for 2006 for males and females by race:

MMWR.jpg

Just looking at the male female comparison, males had higher age-adjusted mortality than females, 924.6 versus 657.8 per 100,000 population. These are fictitious numbers, however. When I say this I am not saying they are false or made up but rather that they are numbers that have been “age adjusted” so that both are measured along the same yardstick and a meaningful comparison can be made. The reason this is necessary is that males and females (and different race categories) have different age structures. Females live longer than males so there are more older females. But mortality rates also increase with age. If you have trouble visualizing the problem think of comparing the mortality rates in college students versus Medicare recipients. The latter are much older and have a much higher mortality rate, so unless you took account of the age difference the main thing you would be comparing would be the age difference. You don’t need death certificates to know old people die at a higher rate than young people. There are different ways to make the age adjustment but the national data in the CDC bar chart are called Direct Adjustments. Here’s the idea.

What you would like is to compare the two populations as if they had the same age structure. So you count up the number of deaths per 100,000 population for males and females at each age group in the year 2006. Then you calculate how many deaths that would have produced in some common standard population. The standard population used for this was the US census population in 2000. So you are comparing deaths in males that would have occurred in 2006 if the males in 2006 had the same age structure as the US population in 2000 and the same for females. But the age structure for each was not the age structure of the population for the year 2000 neglecting sex. So the two numbers, 924.6 versus 657.8 per 100,000 population, are “fictitious” but useful because they allow comparisons. If they were not adjusted, i.e., what we call crude rates, the difference would be much less or even reversed in direction because there are many more older females than males. The crude rates would be the “real” ones, however, i.e., the actual number of deaths per 100,000 that occurred in 2006 in each sex.

You might wonder why we don’t compare mortality at each age level for males and females separately. That would also take care of the comparison problem. It would, except we would then have many comparisons to make, one for each age range. The age adjustment provides a single composite number for the comparison, although it doesn’t tell us which age ranges are producing the difference.

There is lots more to the problem of counting up causes of death than I have discussed here. The ICD-10 code book is huge, several inches thick of fine print, and it takes training and skill to learn how to use it. Then there are all the problems in how to interpret what we see and how to compare one thing with another.

Counting things up is an essential part of epidemiology but obviously there is much more to it. But if counting were the only thing, it would be complicated enough.

Comments

  1. #1 Mark
    August 15, 2008

    One minor note on this excellent post: The ICD stands for “International Classification of Disease” rather than Death. It is used for many of the living in the US as an essential part of medical reimbursement. If you have a lab test or a medical office visit, you will need to have an ICD code for insurance billing purposes.
    The US is ‘stuck’ on ICD-9 while most of the rest of the world has moved on to ICD-10. I think this is because it is so widely used in insurance billing that insurance companies have been reluctant to move to the new version.
    Meanwhile, the World Health Organization is developing the next version, ICD-11.

  2. #2 revere
    August 15, 2008

    Mark: You are right. Thanks for the correction. I think you are referring to ICD-9CM (for clinical modification) but the death certificate is coded in the US by ICD-10 which is the mortality coding.

  3. #3 john
    August 15, 2008

    Recently I went through some works on ascertaining mortality rates:

    “Violence-related Mortality in Iraq from 2002 to 2006” NEJM 2008: 358 pp484-493 This reports on a cluster survey to ascertain numerous family health indicators as well as mortality rates, all-cause and certain cause-specific rates including a category on violent deaths (from suicide bombings, “coalition” forces, the civil war, insurgents….). The report concludes with a calculation with some uncertainty centered between 104,000 to 223,000 violent deaths since the US invasion to June 2006. This estimate is lower than that put forth in an earlier study (400,000 to 800,000) using similar, though arguably less precise, techniques, with fewer study staff, and fewer and larger clusters (Burnham et al. “Mortality after the 2003 invasion of Iraq…” the Lancet, 2006, vol. 368, pp.1421-1428). What first interested me in this study was hearing Tony Blair (former PM of UK) and our own fuzzy-math mathematical wiz-kid, President George W., say ‘the methods employed in this study are flawed.’ Soon after we learned that Blair’s own advisors had told him the methods were not flawed. All this while the US military claims they don’t count dead bodies [among civilians of subjected nation]. Personally, I feel one can have most confidence that the recent NEJM is closest to the actual value. Iraq Body Count which relies only on deaths confirmed by multiple media outlets, hospitals and other authorities, and excludes combatants had estimated 50,000. Turns ones stomach checking the math. Also sad to see the elevated mortality from chronic conditions that were kept in check before the chaos e.g. diseases, perhaps such as diabetes.

    Cluster surveys are used in disasters where there is no health reporting infrastructure: war zones, refugee camps and some African countries. The next best thing is surveillance.

    Surveillance is how the death rate in China is estimated – using a permanent representative sampling system – spots where death registration is complete. India may do the same (there is 2 billion out of 6, population of this planet, covered by surveillance only). See “Review: Mortality registration and surveillance in China: history, current situation and challenges” in Population Health Metrics (http://www.pophealthmetrics.com/content/3/1/3).

    AJPH March 1981, Vol. 71, no.3 “accuracy of cancer death certificates and its effect on cancer mortality statistics” by C. Percy et al. gives some US perspective from 1969-1971, comparing hospital diagnosis with death certificates from the Third National Cancer Survey. In ascertaining underlying cause you can see some of the difficulties such as multiple primaries, or low incidence (~10%) misclassification e.g. cancer of the stomach or esophagus, or pancreas or NH Lymphoma.

Current ye@r *