Autism Epidemic? Not in the NSCH data

17 Aug

There is an epidemic of vaccine-induced autism. Must be, we’ve been hearing this for about 10 years. During that time, many datasets have been manipulated to “prove” the epidemic. The two datasets that come to mind most readily are the U.S. special education data (IDEA data) and the California Department of Developmental Services data (CDDS). Neither were intended for true epidemiology.

That intro should be a warning: be prepared for more armchair epidemiology. Interested in the short answer? A new dataset is out that just doesn’t fit the idea of an “epidemic” of vaccine induced autism. No huge increases in autism counts.

The new dataset available is the 2007 National Survey of Children’s Health. The NSCH data includes questions about autism. So it should come as no surprise that it was spun into support for the “epidemic” already.

As a bit of background, let’s start with the US special education data. It’s a favorite dataset for those pushing the epidemic. Here’s a graph hosted on the Thoughtfulhouse website (click to enlarge):

special ed data supposedly showing an epidemic

special ed data supposedly showing an epidemic

Wow. That’s an epidemic. 16 year olds have an “incidence” (their word) of 0.4 per 10,000 while younger kids are at nearly 20? That’s an increase of 50 times. Impressive. Until we look at the NSCH data.

By coincidence the Thoughfulhouse data are from 2007, the same year as the NSCH data. What happens if we plot the NSCH data like the graph above? Again, click to enlarge. Sorry, I couldn’t format it to look just like the ThoughtfulHouse graph.

National Child Health Survey data on autism

National Child Health Survey data on autism

Doesn’t even look close to the data on the Thoughtfulhouse website. First, the “incidence” numbers for around age 16–0.4 for ThoughtfulHouse’s graph, 93 for the NSCH data. I wanted to graph the data on the same graph, but the newer, NSCH numbers just dwarf the IDEA numbers that ThoughfulHouse used.

Besides the difference in overall magnitude, what about the “epidemic”. What about the huge increases in autism “incidence” or “prevalence”. Numbers like 273% increase, or more, in autism are commonly quoted for the increase in autism diagnoses in the 1990’s. Well, those increases just aren’t observed in the NSCH data. Keep in mind, the 17 year olds in the NSCH data graph were born in 1990. There just isn’t the dramatic increase in an autism count in those data.

There is a a much smaller trend of increased autism count when moving from about age 17 down to about 12. The “incidence” increasing by about 40%. That’s a big number. Not hundreds of percent, but a big number.

The “incidence” goes down for younger kids–which must be partially or completely due to the average age of diagnosis being about 5. There is at least one blogger who is “sparking the debate” that this could be a sign that the removal of thimerosal has resulted in lower autism counts. The same blogger has stated at different times that we should have already seen any drop in the CDDS data and, later, that we won’t be able to see any trends until sometime well past 2010. That’s covering your bases! Either we saw it already, we are seeing it now, or we need to wait until the future to see it.

But, back to the NSCH data. Pretty much, those data are flat for birth year 1995 to 2003. Noisy but flat. That’s when the “epidemic” was in full force.

Or, another way to put it–there is no good evidence of an epidemic in the NSCH data.

About these ads

20 Responses to “Autism Epidemic? Not in the NSCH data”

  1. Dan Hollenbeck August 18, 2009 at 18:20 #

    Hi,

    Is the NSCH graph you show autism prevalence or autism incidence?

    The autism crude incidence graph from Thoughtful House is graphing autism crude incidence (NEW cases per 10K births), not autism prevalence (total cases per 10K births). Since the IDEA data set does not contain data on whether a child’s eligibility for special education is new or existing for a specific year, we can only use a crude incidence. Crude incidence is calculated by subtracting two consecutive year prevalence data to get the estimated new cases.

    The correct interpretation of this crude incidence graph is that the change in autism prevalence is largely, but not entirely, from younger children.

    The Thoughtful House IDEA autism prevalence graph can be found here: http://tinyurl.com/ofh362

    You can also download all of the prevalence data from our website: http://www.fightingautism.org

    Thanks, Dan
    Dan Hollenbeck

    • Sullivan August 18, 2009 at 18:57 #

      Dan Hollenbeck,

      it is neither prevalence nor incidence. It is a survey data based on a rather vague question.

      The data you link to is also not true prevelence data. It is an administrative prevalence. It is quite severely limited by lack of constant criteria over time and geography. The data do not indicate whether a child is eligible for special education services. Rather, they indicate which children have successfully qualified. Anyone who has been through and IEP process knows that a child can qualify and still not be granted an IEP. At the same time, an “educational diagnosis” of autism is a very different thing than a medical diagnosis.

      The correct interpretation of these data (both IDEA and NSCH) is that it is impossible to tell if a true, secular, prevalence is increasing or not. The data are just not of high enough quality.

      • Sullivan August 18, 2009 at 19:27 #

        Dan Hollenbeck,

        if you are asking or pointing out that the data in the two graphs are not really comparable–one being a pseudo “incidence” and the other a pseudo “prevalence”, then you are correct.

        I could compare to your “cumulative growth” graphs and get a similar answer: the NCHS data just doesn’t show the “epidemic”.

        But, if you would like to claim the IDEA data somehow accurately represent a true increase in autism, I’d love to hear how you explain the geographic and race/ethnic differences within those data. Are, say, Hispanics really less prone to autism (as the IDEA data would suggest). Are there states with dramatically lower autism prevalence?

  2. passionlessDrone August 18, 2009 at 19:48 #

    Hi Sullivan –

    if you are asking or pointing out that the data in the two graphs are not really comparable—one being a pseudo “incidence” and the other a pseudo “prevalence”, then you are correct.

    I was wondering if those two values were supposed to be mapping the same thing, even in a psuedo sense. If they were, the data was so far off (~10 to ~ 160 for one age group), then our data is in even worse shape than I thought.

    Are, say, Hispanics really less prone to autism (as the IDEA data would suggest). Are there states with dramatically lower autism prevalence?

    If the answers to those two questions were both no, doens’t this speak towards us hurtling towards a prevalance far higher than 1%? If the second largest minority of children are actually experiencing autism at rates higher than reported this doesn’t really argue against an increase; just our poor counting techniques. Likewise, if the rate in Alabama is more like the rate in New Jersey all this seems to do is take the national average of 1 in 100 and push it lower.

    Does anyone have any ideas by which we could achieve better statistics in a meaningful timeframe? This question as to a true increase is an important one, but mainly the discussion seems to be about the relatively poor nature of our data and the conclusion that we must then accept based on ‘common sense’.

    – pD

    • Sullivan August 18, 2009 at 22:40 #

      pD–

      “If the answers to those two questions were both no, doens’t this speak towards us hurtling towards a prevalance far higher than 1%?”

      No. The prevalence studies that find a 1% (or higher) rate are active surveillance. They find a group of people (typically children in a certain geographic area), screen them for possible autistics and then test those screened for an ASD.

      IDEA data, on the other hand, is passive. They take the rates based on who is reported by the schools as having “autism” on their IEP. No one checks to see if the rest of the kids may have an ASD. IDEA data shows a lower prevalence in most cases than 1%. Increasing the representation of racial/ethnic minorities or of those in rural areas would increase the prevalence. But you wouldn’t be starting from a prevalence of 1% or higher.

      I admit I haven’t looked (yet) at the racial divide in the NCHS data.

      Does anyone have any ideas by which we could achieve better statistics in a meaningful timeframe? This question as to a true increase is an important one, but mainly the discussion seems to be about the relatively poor nature of our data and the conclusion that we must then accept based on ‘common sense’.

      That is a key question in my mind too. One needs an active surveillance in order to make a good estimate here. If the question being tested is “does better ascertainment lead to the higher rate of autism diagnoses observed”, one can’t just look at the passive (reported) data.

      This is really hard with adults. With kids you can say, “OK, everyone who goes to school in this area is in our ‘catchment’ area”. You screen all the kids in that area. It’s easy because they all (or almost all) are going to school there, or are reported to the school district.

      What about adults? You can’t just say, “all adults in this area”. First, adults don’t all participate. Second, and more key, what area? Are you going to take a section of a city? How do you find “all the adults” living in that area? You can’t just look in group homes or other facilities.

      One possibility would be to say “everyone born in this hospital from 1980-1985″ (as an example). Some will be local still. Some will be scattered around the world. But, how do you find the autistic adults?

    • Sullivan August 19, 2009 at 00:19 #

      pD-

      the NCHS data (I keep getting confused between NCHS and NHCS…) doesn’t have as big an ethnic divide as some data

      for those who answered “yes” to “has autism” (again, not the same thing as really testing who is an is not autistic)

      I’m getting

      about 1% of the total population said “yes” to “has autism”
      0.8% of Hispanics “yes to has autism”
      1.1% Whites
      0.6% African American
      1.2% Other

      note that Hispanic, African American and Other had fewer than 100 who answered yes to “has autism”, so the error bars are large.

      The CDF (cumulative distribution function) plots are pretty much the same for White, African American and Other. There doesn’t seem to be much in the way of a difference in the age distributions.

      The CDF plots are rather linear. This, again, goes against the idea of an “epidemic”. It suggest a rather flat distribution of ages for each race/ethnicity.

      Also note that “Hispanic” is not exclusive. I.e. one can be “White and Hispanic”, “Black and Hispanic” etc.

  3. Joseph August 19, 2009 at 02:07 #

    Crude incidence is calculated by subtracting two consecutive year prevalence data to get the estimated new cases.

    Which, as noted several times prior, is not even crude – it’s nonsense. By using this calculation, any stable population would have incidence of zero, when that is not the case. Basically, the “crude” incidence assumes that children don’t lose eligibility, become old, die or move.

    • Sullivan August 19, 2009 at 03:59 #

      Thanks Joseph.

      Their methodology struck me odd since Mr. Hollenbeck’s comment.

  4. MJ August 19, 2009 at 03:31 #

    Joseph, you said –

    “Which, as noted several times prior, is not even crude – it’s nonsense.”

    Not really, this sort of crude change is used all the time in finance. It doesn’t really give you any insight into why something is changing, just whether it is changing and how the rate of change is varying as a function of time.

    The assumption that all of the new cases are in the newest age to enter the data set isn’t too sound but the basic method of showing rate of change is.

    “By using this calculation, any stable population would have incidence of zero”

    Only if the rate of autism is not increasing, which is really the whole point. There are more IDEA cases per year and the rate of change is increasing. You can argue about what it means but you can’t really argue that the number should be zero.

    Regardless, if what Mr. Hollenbeck say above is true, then the whole point is moot as it is meaningless to compare these two sets of data. They represent two very different things over two very different periods of time as such comparing them directly isn’t valid.

    It would be more valid to extract the 2003 data from the NSCH data and then plot the change of cases per 10,000 per age group between the two data sets and then compare that to the data from TH.

    But since the 2003 data showed about half the rate that the 2007 data does I don’t think you are going to want to do that.

    • Sullivan August 19, 2009 at 03:53 #

      It would be more valid to extract the 2003 data from the NSCH data and then plot the change of cases per 10,000 per age group between the two data sets and then compare that to the data from TH.

      But since the 2003 data showed about half the rate that the 2007 data does I don’t think you are going to want to do that.

      You can’t do that comparison, sorry. I’d like to, but I can’t do it in good conscience. The question changed from asking if your kid has “autism” to does your kid have “autism or ASD”. Obviously, much of the change in the counts is the inclusion of PDD-NOS and Asperger syndrome.

      Beyond that, the suggestion you make had the possibility to show an indication of increased awareness–what if the kids born in 1990 had a big increase in autism count?

  5. MJ August 19, 2009 at 04:05 #

    “You can’t do that comparison, sorry. I’d like to, but I can’t do it in good conscience”

    Your current comparison against the TH data is even worse – the data not only isn’t the same exact question it doesn’t even represent the same thing or the same time period.

    Besides, I would be surprised if the data from 2003 did not already reflect PDD-NOS and Asperger in addition to classic autism. My children all are PDD-NOS but if someone asked me if they had autism I would say yes. I think most parents would do the same.

    I think I will do the comparison regardless even if you won’t. it will be interesting to see if the increases are across the board or concentrated in certain age groups.

    • Sullivan August 19, 2009 at 04:25 #

      Your current comparison against the TH data is even worse – the data not only isn’t the same exact question it doesn’t even represent the same thing or the same time period.

      Yep. You are right. I am comparing some fairly straightforward data to some mashup that, as Joseph points out, makes no sense.

      Compare the NCHS data to any data you like which “shows” an epidemic. Explain why the autism counts are pretty much flat in the NCSH data.

      Go ahead and do the 2003 vs. 2007 comparison. Explain to me why the 1990 birth year group had a 0.48% “prevalence” in 2003 and 0.88% in 2007. You see, I already have that dataset open.

      edited to add:

      The 2003 survey asked the question Has a doctor or other health professional ever told you that [selected child] has autism”

      The closest comparison to the 2007 survey would be the quesiton, “”Has a doctor or other health professional ever told you that [selected child] has autism, Asperger’s disorder, pervasive developmental disorder, or other autism spectrum disorder?”

      These would give “rates” for kids born in 1990 of: 0.46% (for the 2003 dataset) and 1.45% (for the 2007 dataset)

      In my original example in this response, I mistakenly used the 2007 question: “does [selected child] currently have autism or ASD]. That has a rate of 0.88%

      So, if anyone would like to explain why the number tripled from 0.46% to 1.45%, I’m open to ideas. The broadening of criteria seems an obvious first step. The possibility that more kids were diagnosed–even after age 13–is open as well.

  6. Joseph August 19, 2009 at 21:47 #

    It doesn’t really give you any insight into why something is changing, just whether it is changing and how the rate of change is varying as a function of time.

    It allows you to measure population growth; that’s it. It should not be called incidence, crude or not. That’s something else. Additionally, if you want to measure population growth, the answer needs to be expressed as a percentage, e.g. 20%. Saying it’s 20 in 10,000 is meaningless. You need to know the size of the population to determine the meaning of that number. If the population grew by 20 in 10,000, but it was already 200 in 10,000, that’s obviously not the same as if the population was 10 in 10,000.

    BTW, how is the crude incidence calculated in the graph? If it’s divided by the whole population, 20 in 10,000 is ridiculously high. Whole population incidence of autism in the US should be around 1 in 10,000 per year, if awareness is stable.

    How is crude incidence for 16 year olds calculated? Is it by calculating the number of teens who got diagnosed when they were 16?

  7. Dan Hollenbeck August 20, 2009 at 16:09 #

    Hi,

    I am sorry that I may have contributed to some of the confusion here by not including an autism prevalence graph on the autism rates page and not providing enough documentation on the terms, calculations, and graphs on the FightingAutism website.

    How is the crude incidence calculated in the graph?

    Crude incidence is calculated by subtracting two consecutive school years autism administrative prevalence to get the change in autism administrative prevalence for each age. For example, the crude incidence for 16 year olds is calculated by subtracting the autism administrative prevalence of 16 year olds in 2006-2007 from the autism administrative prevalence of 16 year olds in 2007-2008. So it is literally the change in the autism administrative prevalence between consecutive birth cohorts. The crude incidence calculation assumes nothing – it is just a calculation. Perhaps, “prevalence change” would be better terminology.

    Since the autism administrative prevalence is already normalized for population size (i.e per 10K births) the crude incidence is normalized for population change between the consecutive birth cohorts.

    I’d love to hear how you explain the geographic differences

    I would defer to http://www.springerlink.com/content/p7738276788316g8/

    I’d love to hear how you explain the race/ethnic differences

    I have not looked at race/ethnic differences yet. Can you provide me with some references?

    Thanks, Dan
    Dan Hollenbeck

    • Sullivan August 20, 2009 at 16:21 #

      I have not looked at race/ethnic differences yet. Can you provide me with some references?

      There is one article in, I believe, “Multiculturalism” which addresses racial/ethnic differences in the IDEA data. The MMWR’s from the CDC show dramatic racial/ethnic divisions in autism prevalence. The IDEA raw data show very large differences in prevalence depending on ethnicity.

      It is quite clear that IDEA data are not an accurate count of the actual number of autistic children. Never has been. One can derive some trends, but one must keep in mind the limitations.

  8. MJ August 22, 2009 at 04:35 #

    “Go ahead and do the 2003 vs. 2007 comparison. Explain to me why the 1990 birth year group had a 0.48% “prevalence” in 2003 and 0.88% in 2007. You see, I already have that dataset open.”

    Here is what it looks like

  9. Sullivan August 22, 2009 at 16:01 #

    Thank you very much for this.

    I needed confirmation of my numbers for a post I am writing. It is serious enough that I wanted to be sure before going ahead.

  10. David N. Brown September 2, 2009 at 09:58 #

    I was struck by the “peaks” on the graph at ages 7 and twelve. What this suggests to me is that autism diagnoses are most often made when the child is making a social transition: to middle school at age twelve, elementary at age seven, and maybe day care and preschool at ages 2-5. The three age groups represent distinct “waves”, with the trend going downward in the latter two. This pattern can’t be regarded as evidence of anything but diagnostic procedure, and also testimony to the potential for autistics to hide or compensate for handicaps when given time to adjust to a stable environment.

  11. temeh July 29, 2010 at 03:31 #

    It appears that the NSCH data from 2007 were in blocks – that is, those administering the survey grouped specific categories of questions to different folks. They did not report it, but it appears that way when I am running analysis on autism and breastfeeding. If you are comparing prevalence rates for 2003 and 2007, you might want to see if they did the same thing in 2003…

Trackbacks/Pingbacks

  1. Science-Based Medicine » “There must be a reason,” or how we support our own false beliefs - August 31, 2009

    [...] the “true” prevalence of autism has probably not changed much over decades; i.e., there is no “epidemic” of autism. In response, some who cling to the mercury hypothesis claim that even a trace of mercury would [...]

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

Follow

Get every new post delivered to your Inbox.

Join 1,090 other followers

%d bloggers like this: