For many diseases we have very little ability to determine who is at high or low risk; the risk factors are unreplicated, complicated, or understudied. However, for other diseases we can do much better. Alzheimer’s disease is a form of senile dementia that is characterised by abnormal clustering of proteins in the brain (right). We know a number of important risk factors for Alzheimer’s, and knowing your own risk factors may seriously change your estimate of the chance of developing the disease. But how can you calculate this risk?

This is going to be somewhat of an information deluge, as I go through everything to think about when you predict a complex disease, including how to calculate genetic and environmental risks, and how important these risks are, both individually and all together. I will demonstrate all of the calculations on the various GNZ contributors, and in particular how I have worked out my own risk.

I’ll measure the risks in terms of odds ratios; you may want to read the introduction to Carl’s post from earlier this year to refresh your mind on what this means. I will also use the disease probability; this is simply the chance of developing Alzheimer’s, or equally, the percentage of people with this set of risk factors who will develop the disease.

Also note that an important factor to consider is the baseline lifetime risk, the total proportion of people who will develop Alzheimer’s before they die. I am going to use a lifetime risk of 9% for men and 17% for women, taken from an Alzheimer’s Association report, but getting a good estimate of this is actually very difficult, and will vary from country to country.

If you want to know more about Alzheimer’s, including prevention, diagnosis and treatment, you can read about the disease on the Mayo Clinic or NHS Choices websites.

#### Inferring APOE status without genotyping

The most important risk factor for Alzheimer’s disease is a gene called Apolipoprotein E, or APOE. This gene has a number of different alleles, called ε2, ε3 and ε4. ε2 and ε3 protect against Alzheimer’s, whereas ε4 increases your risk of developing it. Every individual has two copies of this allele, and the combination you have determines how likely you are to develop the disease; ε2/ε2 is the lowest risk, and ε4/ε4 is the highest.

The APOE alleles are defined by two genetic variants in the gene, rs429358 and rs7412. The different genotypes, along with odds ratios (taken from this paper) are below:

 APOE Allele rs429358 rs7412 OR ε2/ε2 TT TT 0.23 ε2/ε3 TT TC 0.32 ε2/ε4 TC TC 1.23 ε3/ε3 TT CC 0.50 ε3/ε4 TC CC 1.79 ε4/ε4 CC CC 6.90

Now, if you know your genotypes at rs7412 and rs429358, you can get your APOE genotype, and if you are a deCODEme or a 23andMe V3 customer, you can do this easily; look up your SNPs, check the table above, and get your odds ratio. However, most of the GNZ cohort have been genotyped on the 23andMe V2 chip; while this contains rs7412, it does not have rs429358.

However, all is not lost. Variants in the genome are correlated, and it is possible to make educated guesses about variants that you haven’t genotyped, using variants for which you have genotypes. This process is called genotype imputation. The best way of performing imputation is to use one of the many free imputation programs written by statistical geneticists, such as Beagle or IMPUTE2, and to use a state-of-the-art reference set, such as one from the 1000 Genomes Project.

I will write something next week on how you can do this. However, for right now I will just give the results for the GNZ members:

 Name APOE Certainty OR Daniel ε3/ε3 100% 0.5 Luke ε3/ε3 100% 0.5 Dan ε3/ε4 95% 1.70 Caroline ε2/ε4 93% 1.18 Kate ε3/ε3 100% 0.5 Vincent ε3/ε3 100% 0.5 Jeff ε3/ε4 100% 1.79 Jan ε3/ε3 100% 0.5 Joe ε3/ε4 100% 1.79 Don ε3/ε4 98% 1.76 Carl ε3/ε3 100% 0.5 Illana ε3/ε3 100% 0.5

Imputation is probabilistic, and provides us with a certainty for our predictions. The ε3/ε3 guesses are very confident, but the rarer types are less so. Impute2 is pretty uncertain about Caroline’s somewhat strange (and pretty rare) ε2/ε4 genotype, though it did make the correct guess when compared to her deCODEme results. To compensate for the uncertainty, I’ve shrunk the down the odds ratios at uncertain guesses.

My APOE status is definitely ε3/ε3; impute is 100% certain, plus I have checked against other genotyping. My odds ratio is thus 0.5.

#### Other genetic factors

APOE is a very important genetic risk factor for Alzheimer’s, but it is not the only one that we know about. There are at least five other well-supported risk variants (I have taken these from this study):

 Gene Variant Risk Allele 0 1 2 CR1 rs3818361 A 0.89 1.1 1.35 CLU rs11136000 C 0.83 0.97 1.12 PICALM rs541458 T 0.75 0.92 1.13 ACE rs1800764 T 0.83 0.98 1.16 CST3 rs1064039 C 0.75 0.9 1.08

To get your odds ratio, you just count the number of risk alleles you have for each variant, look up the number in the table, and times them together for all the variants. For instance, my genotypes are GG, CT, CT, TT and CC. This is 0,1,1,2 and 2 risk alleles for each variant, and my odds ratio at these sites is therefore 0.89 X 0.97 x 0.92 x 1.13 x 1.08 = 0.97.

I can combine this with my APOE odds ratio, to give 0.5 x 0.97 = 0.485. The odds of a man developing Alzheimer is 0.09 to 0.81, or 0.111 to 1. My odds ratio of 0.485 changes this to 0.485 x 0.111 = 0.054 to 1, or 4.2%. Doing this for all GNZ members gives us these odds ratios and disease probabilities across all the genetic factors:

 Name Combined OR Alzheimer Probability Caroline 1.5 23.6% Joe 2.26 18.2% Jeff 2.08 17.0% Dan 1.97 16.6% Kate 0.6 10.9% Don 1.21 10.7% Ilana 0.48 8.9% Vince 0.875 7.9% Carl 0.81 7.4% Luke 0.49 4.6% Jan 0.45 4.2% Daniel 0.42 4.0%

We can also visualise how much each individauls’ risk changes. In the plot below, red arrows show an increase in risk, and green arrows show a decrease:

Notice that Kate’s overall risk is higher than Don’s, but she is still “luckier” in some sense, because her genetic factors majorly decrease her risk from her elevated level as a female, whereas Don’s risk slightly increases his from his lower male baseline.

#### Environmental factors

While there aren’t any environmental factors that influence Alzheimer’s as strongly as APOE, there are a number of known and suspected risks. Most of these risk factors can only really be applied reliability to older people, and include things like vascular disease, head trauma and the like. However, there are three well-associated traits that can apply to everyone: level of education, physical activity and alcohol consumption:

 Risk Factor Class OR Education 0-8yrs 1.36 9-12ys 0.98 >12yrs 0.72 Regular Exercise Yes 0.89 No 1.28 Alcohol consumption Yes 0.73 No 1.27

There is has been some research on the interaction between genetics and environment in Alzheimer’s, but not enough for me to attempt to model it. Making the (probably wrong) assumption that we can consider genetics and environment separately, we can just multiply the odds ratios together. So as an alcohol-imbibing PhD student who exercises regularly, my environmental odds ratio is 0.72 x 0.89 x 0.73 = 0.47, and my overall environment+genetics odds ratio is thus 0.47 x 0.485 = 0.23. My Alzheimer’s probability is thus 2.2%.

#### How predictive are these factors

So how predictive are these risk factors, when taken together? Well, assuming that they are independent, we can make a good guess.

It is relatively easy to calculate the proportion of total variance explained by these factors, using the method described in this paper:

 Factors Variance explained Sex 3.8% Sex+APOE 18.4% Sex+Genetcs 20.2% Sex+Genetics+Environment 21.9%

So these factors can capture just over a fifth of the total variance in disease risk. Notice that the majority of this risk comes from APOE, with sex, other genetics and environment factors accounting for just a third of the total explained variance.

We can also calculate an AUC value using the GENROC calculator. The value is 0.75. This means that, given two individuals, one of whom has Alzheimer’s, one of whom doesn’t, both of whom have calculated their odds ratio using the method above, the person with the disease will have a larger total odds ratio than the one without the disease 75% of the time.

To look at it a final way, of those who take this test, 5% will find that they have a risk above 31%, and a further 5% will find that they have a risk below 3%.

#### Prediction, privacy and prevention

Via genotype imputation, we can accurately assess APOE risk using 23andMe’s v2 chip, which does not actually genotype the APOE allele. Thus a strong medically predictive factor can be inferred using entirely “non-medical” variants taken from the surrounding region. This really illustrates how the distinction between medically and non-medically relevant genetic variants is pretty artificial. This has been observed before, and raises some real issue about privacy, and the attempt to separate (and separately regulate) medical and non-medical genetic information.

So what can we do with this information? A lot of people (myself included) would just want to know, as health information about themselves, regardless of what can be done with that information. For Alzheimer’s, there are no strong preventative measures, and treatment options are currently limited. If you aren’t doing regular exercise, you can knock 30% off your risk by taking up a sport (which, of course, would have a lots of other benefits), and eating less fat and red meat may knock another few percent off. While these things are worth doing, they are worth doing for everyone regardless of risk, and the effects are pretty small compared to the overall risk anyway. Perhaps the most useful thing that a high-risk prediction can do is to look out for the early signs of the disease; if you have a 2% lifetime risk, you may not be that worried if you start forgetting your keys, but if you risk is 50%, you should be considering seeing a medical professional.

What the personal utility of this information is will vary from individual to individual. What the above risk factors can give you is a large (though by no-means complete) insight into your personal risk. What you will do with this is up to you.

Amyloid plaque image taken from this paper in PLoS Medicine

#### 14 Responses to “Calculating your Alzheimer’s risk”

• Joe Pickrell

Hey, you’ve got me listed as e3/e3, with an OR of 1.79. One of those things must be a misprint; do you mean I’m really e3/e4?

• Nice post. You’ve got the education ors backwards though (I hope)

• Thanks guys. Nick, yes, education was back to front. Joe, you are really e3/e4.

• Joe Pickrell

Thanks. And great post, this is really useful.

• Great work, as usual. I guess this is another reason to have a drink tonight…

• Nice article Luke. One thing that confused me was how you calculated these odds ratios. It wasn’t clear to me what the reference category was or how you got these APOE odds ratios from the paper you cite (Corneveaux et al, HMG, 2010).

• All the odds ratios were adjusted to be relative to the mean risk, rather than a defined reference catagory, so that they directly modify the life-time risk.

As a quick example, suppose you had a high risk factor with a frequency of 25% and an odds ratio of 2 between high and low risk. You mean risk is 0.75*1 + 0.25*2 = 1.25. So the low-risk factor has an odds ratio of 1/1.25 = 0.8, and the high-risk factor has 2/1.25 = 1.6.

• Thanks for the rapid response Luke, that’s clear now.

• Luke,

You might explain “over what period of time” and “what are my odds of dying with Alzheimer’s at 80 compared to dying without having ever gotten it?”

• Luke,

You might explain “over what period of time” and “what are my odds of dying with Alzheimer’s at 80 compared to dying without having ever gotten it?”

Two others points. I know I’ve seen several Navigenics reports and all of them have had an APOE-linked genotype reported. But I don’t subscribe and can’t remember which marker or chip.

Finally, you might explain risk over time. Adrienne Cupples and Lindsay Farrer did a lot of work to assemble some risk curves:
http://journals.lww.com/geneticsinmedicine/Abstract/2004/07000/Estimating_risk_curves_for_first_degree_relatives.4.aspx (Genetics in Medicine 2004)

• You quite rightly point out that genetic and environmental factors are unlikely to be independent in Alzheimer’s.

However, one thing I don’t understand is that when calculating the combined odds ratio for the genetic factors you assume that the different risk variants are not correlated. Surely this can’t be correct?

Doesn’t genotype imputation work because the opposite is true? (i.e. variants in the genome are indeed correlated.)

@Sajid

These variants are actually unlinked. Plus, it doesn’t matter if the genetic factors are linked, providing you take this into accout when you calculate the odds ratios.

• Elizabeth

FYI, the CLU polymorphism is rs11136000, not rs1113600.

• Dr Robert Peers

It might help if we knew the actual cause of the disease! Converging evidence from epidemiology, clinical observation, experiment and lipid biochemistry points strongly at steam-treated polyunsaturated food oils, as the common direct cause of Alzheimer’s. These ubiquitous vitamin E-depleted seed oils cause lipid membrane peroxidation, that is now known to inhibit beta-amyloid degrading enzymes: beta-amyloid peptide (the proximate cause of the disease) therefore accumulates–with glacial slowness–in the human brain (no other species get Alzheimer’s). APOEe4 merely accelerates AD onset, by inhibiting beta-amyloid clearance; in no way does this gene variant cause the disease on its own. The other late-onset AD genes seem to be very weak susceptibility genes, not worth considering in the big picture. Interested readers might care to see my YouTube video–see PEERS ALZHEIMER. I hope to submit my completed hypothesis to The Lancet soon.

Page optimized by WP Minify WordPress Plugin