Econstudentlog

A few SSC comments

I recently left a few comments in an open thread on SSC, and I figured it might make sense to crosspost some of the comments made there here on the blog. I haven’t posted all my contributions to the debate here, rather I’ve just quoted some specific comments and observations which might be of interest. I’ve also added some additional remarks and comments which relate to the topics discussed. Here’s the main link (scroll down to get to my comments).

“One thing worth keeping in mind when evaluating pre-modern medicine characterizations of diabetes and the natural history of diabetes is incidentally that especially to the extent that one is interested in type 1 survivorship bias is a major problem lurking in the background. Prognostic estimates of untreated type 1 based on historical accounts of how long people could live with the disease before insulin are not in my opinion likely to be all that reliable, because the type of patients that would be recognized as (type 1) diabetics back then would tend to mainly be people who had the milder forms, because they were the only ones who lived long enough to reach a ‘doctor’; and the longer they lived, and the milder the sub-type, the more likely they were to be studied/’diagnosed’. I was a 2-year old boy who got unwell on a Tueday and was hospitalized three days later. Avicenna would have been unlikely to have encountered me, I’d have died before he saw me. (Similar lines of reasoning might lead to an argument that the incidence of diseases like type 1 diabetes may also today be underdiagnosed in developing countries with poorly developed health care systems.)”

Douglas Knight mentioned during our exchange that medical men of the far past might have been more likely to attend to patients with acute illnesses than patients with chronic conditions, making them more likely to attend to such cases than would otherwise be the case, a point I didn’t discuss in any detail during the exchange. I did however think it important to note here that information exchange was significantly slower, and transportation costs were much higher, in the past than they are today. This should make such a bias less relevant, all else equal. Avicenna and his colleagues couldn’t take a taxi, or learn by phone that X is sick. He might have preferentially attended to the acute cases he learned about, but given high transportation costs and inefficient communication channels he might often never arrive in time, or at all. A particular problem here is that there are no good data on the unobserved cases, because the only cases we know about today are the ones people like him have told us about.

Some more comments:

“One thing I was considering adding to my remarks about survivorship bias is that it is not in my opinion unlikely that what you might term the nature of the disease has changed over the centuries; indeed it might still be changing today. Globally the incidence of type 1 has been increasing for decades and nobody seems to know why, though there’s consensus about an environmental trigger playing a major role. Maybe incidence is not the only thing that’s changed, maybe e.g. the time course of the ‘average case’ has also changed? Maybe due to secondary factors; better nutritional status now equals slower progression of beta cell failure than was the case in the past? Or perhaps the other way around: Less exposure to bacterial agents the immune system throughout evolutionary time has been used to having to deal with today means that the autoimmune process is accelerated today, compared to in the far past where standards of hygiene were different. Who knows? […] Maybe survivorship bias wasn’t that big of a deal, but I think one should be very cautious about which assumptions one might implicitly be making along the way when addressing questions of this sort of nature. Some relevant questions will definitely be unknowable due to lack of good data which we will never be able to obtain.”

I should perhaps interpose here that even if survivorship bias ‘wasn’t that big of a deal’, it’s still sort of a big problem in the analytical setting because it seems perfectly plausible to me to be making the assumption that it might even so have been a big deal. These kinds of problems magnify our error bars and reduce confidence in our conclusions, regardless of the extent to which they actually played a role. When you know the exact sign and magnitude of a given moderating effect you can try to correct for it, but this is very difficult to do when a large range of moderator effect sizes might be considered plausible. It might also here be worth mentioning explicitly that biases such as the survivorship bias mentioned can of course impact a lot of things besides just the prognostic estimates; for example if a lot of cases never come to the attention of the medical people because these people were unavailable (due to distance, cost, lack of information, etc.) to the people who were sick, incidence and prevalence will also implicitly be underestimated. And so on. Back to the comments:

“Once you had me thinking that it might have been harder [for people in the past] to distinguish [between type 1 and type 2 diabetes] than […] it is today, I started wondering about this, and the comments below relate to this topic. An idea that came to mind in relation to the type 1/type 2 distinction and the ability of people in the past to make this distinction: I’ve worked on various identification problems present in the diabetes context before, and I know that people even today make misdiagnoses and e.g. categorize type 1 diabetics as type 2. I asked a diabetes nurse working in the local endocrinology unit about this at one point, and she told me they had actually had a patient not long before then who had been admitted a short while after having been diagnosed with type 2. Turned out he was type 1, so the treatment failed. Misdiagnoses happen for multiple reasons, one is that obese people also sometimes develop type 1, and if it’s an acute onset setting the weight loss is not likely to be very significant. Patient history should in such a case provide the doctor with the necessary clues, but if the guy making the diagnosis is a stressed out GP who’s currently treating a lot of obese patients for type 2, mistakes happen. ‘Pre-scientific method’ this sort of individual would have been inconvenient to encounter, because a ‘counter-example’ like that supposedly demonstrating that the obese/thin(/young/old, acute/protracted…) distinction was ‘invalid’ might have held a lot more weight than it hopefully would today in the age of statistical analysis. A similar problem would be some of the end-stage individuals: A type 1 pre-insulin would be unlikely to live long enough to develop long term complications of the disease, but would instead die of DKA. The problem is that some untreated type 2 patients also die of DKA, though the degree of ketosis varies in type 2 patients. DKA in type 2 could e.g. be triggered by a superimposed cardiovascular event or an infection, increasing metabolic demands to an extent that can no longer be met by the organism, and so might well present just as acutely as it would in a classic acute-onset type 1 case. Assume the opposite bias you mention is playing a role; the ‘doctor’ in the past is more likely to see the patients in such a life-threatening setting than in the earlier stages. He observes a 55 year old fat guy dying in a very similar manner to the way a 12 year old girl died a few months back – very characteristic symptoms, breath smells fruity, Kussmaul respiration, polyuria and polydipsia…). What does he conclude? Are these different diseases?”

Making the doctor’s decision problem even harder is of course the fact that type 2 diabetes even today often goes undiagnosed until complications arise. Some type 2 patients get their diagnosis only after they had their first heart attack as a result of their illness. So the hypothetical obese middle-aged guy presenting with DKA might not have been known by anyone to be ‘a potentially different kind of diabetic’.

‘The Nybbler’ asked this question in the thread: “Wouldn’t reduced selection pressure be a major reason for increase of Type I diabetes? Used to be if you had it, chance of surviving to reproduce was close to nil.”

I’ll mention here that I’ve encountered this kind of theorizing before, but that I’ve never really addressed it – especially the second part – explicitly, though I’ve sometimes felt like doing that. I figured this post might be a decent place to at least scratch the surface. The idea that there are more type 1 diabetics now than there used to be because type 1 diabetics used to die of their disease and now they don’t (…and so now they are able to transmit their faulty genes to subsequent generations, leading to more diabetic individuals over time) sounds sort of reasonable if you don’t know very much about diabetes, but it sounds less reasonable to people who do. Genes matter, and changed selection pressures have probably played a role, but I find it hard to believe this particular mechanism is a major factor. I have included both my of my replies to ‘Nybbler’ below:

First comment:

“I’m not a geneticist and this is sort-of-kind-of near the boundary area of where I feel comfortable providing answers (given that others may be more qualified to evaluate questions like this than I am). However a few observations which might be relevant are the following:

i) Although I’ll later go on to say that vertical transmission is low, I first have to point out that some people who developed type 1 diabetes in the past did in fact have offspring, though there’s no doubt about the condition being fitness-reducing to a very large degree. The median age of diagnosis of type 1 is somewhere in the teenage years (…today. Was it the same way 1000 years ago, or has the age profile changed over time? This again relates to questions asked elsewhere in this discussion…), but people above the age of 30 get type 1 too.

ii) Although type 1 display some level of familia[l] clustering, most cases of type 1 are not the result of diabetics having had children who then proceed to inherit their parents’ disease. To the extent that reduced selection is a driver of increased incidence, the main cause would be broad selection effects pertaining to immune system functioning in general in the total population at risk (i.e. children in general, including many children with what might be termed suboptimal immune system functioning, being more likely to survive and later develop type 1 diabetes), not effects derived from vertical transmission of the disease (from parent to child). Roughly 90% of newly diagnosed type 1 diabetics in population studies have a negative family history of the disease, and on average only 2% of the children of type 1 diabetic mothers, and 5% of the children of type 1 diabetic fathers, go on to develop type 1 diabetes themselves.

iii) Historically vertical transmission has even in modern times been low. On top of the quite low transmission rates mentioned above, until well into the 80es or 90es many type 1 diabetic females were explicitly advised by their medical care providers not to have children, not because of the genetic risk of disease transmission but because pregnancy outcomes were likely to be poor; and many of those who disregarded the advice gave birth to offspring who were at a severe fitness disadvantage from the start. Poorly controlled diabetes during pregnancy leads to a very high risk of birth defects and/or miscarriage, and may pose health risks to the mother as well through e.g. an increased risk of preeclampsia (relevant link). It is only very recently that we’ve developed the knowledge and medical technology required to make pregnancy a reasonably safe option for female diabetics. You still had some diabetic females who gave birth before developing diabetes, like in the far past, and the situation was different for males, but either way I feel reasonably confident claiming that if you look for genetic causes of increasing incidence, vertical transmission should not be the main factor to consider.

iv) You need to be careful when evaluating questions like these to keep a distinction between questions relating to drivers of incidence and questions relating to drivers of prevalence at the back of your mind. These two sets of questions are not equivalent.

v) If people are interested to know more about the potential causes of increased incidence of type 1 diabetes, here’s a relevant review paper.”

I followed up with a second comment a while later, because I figured a few points of interest might not have been sufficiently well addressed in my first comment:

“@Nybbler:

A few additional remarks.

i) “Temporal trends in chronic disease incidence rates are almost certainly environmentally induced. If one observes a 50% increase in the incidence of a disorder over 20 yr, it is most likely the result of changes in the environment because the gene pool cannot change that rapidly. Type 1 diabetes is a very dynamic disease. […] results clearly demonstrate that the incidence of type 1 diabetes is rising, bringing with it a large public health problem. Moreover, these findings indicate that something in our environment is changing to trigger a disease response. […] With the exception of a possible role for viruses and infant nutrition, the specific environmental determinants that initiate or precipitate the onset of type 1 diabetes remain unclear.” (Type 1 Diabetes, Etiology and Treatment. Just to make it perfectly clear that although genes matter, environmental factors are the most likely causes of the rising levels of incidence we’ve seen in recent times.)

ii. Just as you need to always keep incidence and prevalence in mind when analyzing these things (for example low prevalence does not mean incidence is necessarily low, or was low in the past; low prevalence could also be a result of a combination of high incidence and high case mortality. I know from experience that even diabetes researchers tend to sometimes overlook stuff like this), you also need to keep the distinction between genotype and phenotype in mind. Given the increased importance of one or more environmental triggers in modern times, penetrance is likely to have changed over time. This means for example that ‘a diabetic genotype’ may have been less fitness reducing in the past than it is today, even if the associated ‘diabetic phenotype’ may on the other hand have been much more fitness reducing than it is now; people who developed diabetes died, but many of the people who might in the current environment be considered high-risk cases may not have been high risk in the far past, because the environmental trigger causing disease was absent, or rarely encountered. Assessing genetic risk for diabetes is complicated, and there’s no general formula for calculating this risk either in the type 1 or type 2 case; monogenic forms of diabetes do exist, but they account for a very small proportion of cases (1-5% of diabetes in young individuals) – most cases are polygenic and display variable levels of penetrance. Note incidentally that a story of environmental factors becoming more important over time is actually implicitly also, to the extent that diabetes is/has been fitness-reducing, a story of selection pressures against diabetic genotypes potentially increasing over time, rather than the opposite (which seems to be the default assumption when only taking into account stuff like the increased survival rates of type 1 diabetics over time). This stuff is complicated.”

I wasn’t completely happy with my second comment (I wrote it relatively fast and didn’t have time to go over it in detail after I’d written it), so I figured it might make sense to add a few details here. One key idea here is of course that you need to distinguish between people who are ‘vulnerable’ to developing type 1 diabetes, and people who actually do develop the disease. If fewer people who today would be considered ‘vulnerable’ developed the disease in the past than is the case now, selection against the ‘vulnerable’ genotype would – all else equal – have been lower throughout evolutionary time than it is today.

All else is not equal because of insulin treatment. But a second key point is that when you’re interested in fitness effects, mortality is not the only variable of interest; many diabetic women who were alive because of insulin during the 20th century but who were also being discouraged from having children may well have left no offspring. Males who committed suicide or died from kidney failure in their twenties likely also didn’t leave many offspring. Another point related to the mortality variable is that although diabetes mortality might in the past have been approximated reasonably well by a simple binary outcome variable/process (no diabetes = alive, diabetes = dead), type 1 diabetes has had large effects on mortality rates also throughout the chunk of history during which insulin has been a treatment option; mortality rates 3 or 4 times higher than those of non-diabetics are common in population studies, and such mortality rates add up over time even if base rates are low, especially in a fitness context, as they for most type 1 diabetics are at play throughout the entire fertile period of the life history. Type 2 diabetes is diagnosed mainly in middle-aged individuals, many of whom have already completed their reproductive cycle, but type 1 diabetes is very different in that respect. Of course there are multiple indirect effects at play as well here, e.g. those of mate choice; which is the more attractive potential partner, the individual with diabetes or the one without? What if the diabetic also happens to be blind?

A few other quotes from the comments:

“The majority of patients on insulin in the US are type 2 diabetics, and it is simply wrong that type 2 diabetics are not responsive to insulin treatment. They were likely found to be unresponsive in early trials because of errors of dosage, as they require higher levels of the drug to obtain the same effect as will young patients diagnosed with type 1 (the primary group on insulin in the 30es). However, insulin treatment is not the first-line option in the type 2 context because the condition can usually be treated with insulin-sensitizing agents for a while, until they fail (those drugs will on average fail in something like ~50% of subjects within five years of diagnosis, which is the reason – combined with the much (order(/s, depending on where you are) of magnitude) higher prevalence of type 2 – why a majority of patients on insulin have type 2), and these tend to a) be more acceptable to the patients (a pill vs an injection) and b) have fewer/less severe side effects on average. One reason which also played a major role in delaying the necessary use of insulin to treat type 2 diabetes which could not be adequately controlled via other means was incidentally the fact that insulin ca[u]ses weight gain, and the obesity-type 2 link was well known.”

“Type 1 is autoimmune, and most cases of type 2 are not, but some forms of type 2 seem to have an autoimmune component as well (“the overall autoantibody frequency in type 2 patients varies between 6% and 10%” – source) (these patients, who can be identified through genetic markers, will on average proceed to insulin dependence because of treatment failure in the context of insulin sensitizing-agents much sooner than is usually the case in patients with type 2). In general type 1 is caused by autoimmune beta cell destruction and type 2 mainly by insulin resistance, but combinations of the two are also possible […], and patients with type 1 can develop insulin resistance just as patients with type 2 can lose beta cells via multiple pathways. The major point here being that the sharp diagnostic distinction between type 1 and type 2 is a major simplification of what’s really going on, and it’s hiding a lot of heterogeneity in both samples. Some patients with type 1 will develop diabetes acutely or subacutely, within days or hours, whereas others will have elevated blood glucose levels for months before medical attention is received and a diagnosis is made (you can tell whether or not blood glucose has been elevated pre-diagnosis by looking at one of the key diagnostic variables, Hba1c, which is a measure of the average blood glucose over the entire lifetime of a red blood cell (~3-4 months) – in some newly diagnosed type 1s, this variable is elevated, in others it is not. Some type 1 patients will develop other autoimmune conditions later on, whereas others will not, and some will be more likely to develop complications than others who have the same level of glycemic control.

Type 1 and type 2 diabetes are quite different conditions, but in terms of many aspects of the diseases there are significant degrees of overlap (for example they develop many of the same complications, for similar (pathophysiological) reasons), yet they are both called diabetes. You don’t want to treat a type 2 diabetic with insulin if he can be treated with metformin, and treating a type 1 with metformin will not help – so different treatments are required.”

“In terms of whether it’s ideal to have one autistic diagnostic group or two (…or three, or…) [this question was a starting point for the debate from which I quote, but I decided not to go much into this topic here], I maintain that to a significant extent the answer to that question relates to what the diagnosis is supposed to accomplish. If it makes sense for researchers to be able to distinguish, which it probably does, but it is not necessary for support organizers/providers to know the subtype in order to provide aid, then you might end up with one ‘official’ category and two (or more) ‘research categories’. I would be fine with that (but again I don’t find this discussion interesting). Again a parallel might be made to diabetes research: Endocrinologists are well aware that there’s a huge amount of variation in both the type 1 and type 2 samples, to the extent that it’s sort of silly to even categorize these illnesses using the same name, but they do it anyway for reasons which are sort of obvious. If you’re type 1 diabetic and you have an HLA mutation which made you vulnerable to diabetes and you developed diabetes at the age of 5, well, we’ll start you on insulin, try to help you achieve good metabolic control, and screen you regularly for complications. If on the other hand you’re an adult guy who due to a very different genetic vulnerability developed type 1 diabetes at the age of 30 (and later on Graves’ disease at the age of 40, due to the same mutation), well, we’ll start you on insulin, try to help you achieve good metabolic control, and screen you regularly for complications. The only thing type 1 diabetics have in common is the fact that their beta cells die due to some autoimmune processes. But it could easily be conceived of instead as literally hundreds of different diseases. Currently the distinctions between the different disease-relevant pathophysiological processes don’t matter very much in the treatment context, but they might do that at some point in the future, and if that happens the differences will start to become more important. People might at that point start to talk about type 1a diabetes, which might be the sort you can delay or stop with gene therapy, and type 1b which you can’t delay or stop (…yet). Lumping ‘different’ groups together into one diagnostic category is bad if it makes you overlook variation which is important, and this may be a problem in the autism context today, but regardless of the sizes of the diagnostic groups you’ll usually still end up with lots of residual (‘unexplained’) variation.”

I can’t recall to which extent I’ve discussed this last topic – the extent to which type 1 diabetes is best modeled as one illness or many – but it’s an important topic to keep at the back of your mind when you’re reading the diabetes literature. I’m assuming that in some contexts the subgroup heterogeneities, e.g. in terms of treatment response, will be much more important than in other contexts, so you probably need specific subject matter knowledge to make any sort of informed decision about to which extent potential unobserved heterogeneities may be important in a specific setting, but even if you don’t have that ‘a healthy skepticism’, derived from keeping the potential for these factors to play a role in mind, is likely to be more useful than the alternative. In that context I think the (poor, but understandable) standard practice of lumping together type 1 and type 2 diabetics in studies may lead many people familiar with the differences between the two conditions to think along the lines that as long as you know the type, you’re good to go – ‘at least this study only looked at type 1 individuals, not like those crappy studies which do not distinguish between type 1 and type 2, so I can definitely trust these results to apply to the subgroup of type 1 diabetics in which I’m interested’ – and I think this tendency, to the extent that it exists, is unfortunate.

Advertisements

July 8, 2017 - Posted by | autism, Diabetes, Epidemiology, Genetics, Medicine, Psychology

No comments yet.

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

%d bloggers like this: