This content is current only at the time of printing. This document was printed on 28 April 2017. A current copy is located at http://apvma.gov.au/node/19421
You are here
Transcript for Ms Janis Baines, FSANZ
Australian food consumption data and use in dietary exposure assessments
This presentation was delivered at the APVMA’s science feature session on 15 October 2015. The full video is available on our YouTube channel.
First of all I would really like to thank APVMA for this opportunity to talk about our part of the work. I think it will be a subject that is perhaps closer to home in that it's talking about food and we all know about food, but it's really how we take what we eat and how we eat food into account in our risk assessments. What I want to talk about is dietary exposure assessments for agricultural chemicals, what they are, how we get the data, what sort of data do we need to do an exposure assessment, our food consumption data and the pesticide residue and veterinary drug residue data that we use in these assessments.
Unlike Jim's presentation, this is a very simple equation. Basically, dietary exposure is a combination of how much food we eat by the chemical concentration in that food. Of course, it's not quite that simple. We can adjust it for body weight and some of the complexity of dietary modelling is actually manipulating the data we have to derive the data we need for the dietary exposure assessment.
In this area that you're interested in, in agvet chemicals, there's basically two dietary exposure assessments we would use, two types. We might be interested in chronic dietary assessment, which is the expected dietary exposure to a food chemical over a lifetime or over a long period of time from the total diet. Or, if the toxicologists have set an acute reference dose, we might want to estimate the expected exposure to a food chemical from short‑term exposure, perhaps from a single meal or single food. Usually we do the calculation for a single food or commodity, either from one meal or up to a 24‑hour period. As you can imagine, those sort of different sorts of dietary exposure assessments require different data inputs.
We will go back to the data we need but just to give you the picture of where this information fits in with our risk assessment process. This risk assessment process is followed by both FSANZ and APVMA and it is in line with international Codex procedures. Basically, first of all we do a hazard identification process and then characterise that risk which is really the dose response relationships. That's the area that the toxicology including pharmacokinetics and other information derive health‑based guidance value. We then do an exposure assessment and then those two pieces of information are combined to do a risk characterisation where we compare exposure to the health‑based guidance standard.
At FSANZ we tend to do dietary exposure but of course there are other routes of exposure and at APVMA you might take worker exposure into account or soil or other forms of exposure for some chemicals we need to think about. But for any model, as Jim so rightly said, whatever model we use and whatever model we use for exposure assessments, the quality of the results really are determined by the inputs we put in and how we choose to put those inputs together in our model and the assumptions we make. Jason and Biraj are going to cover this in a bit more detail in their talks but for dietary exposure assessment we have two health‑based guidance values that are important and sometimes these are referred to as reference health standards. We have the acceptable daily intake (ADI) for our chronic dietary exposures and the acute reference dose (ARfD) for the acute exposures. These are set by the Office of Chemical Safety but also internationally they are set by either the FAO/WHO‑joint meeting for pesticide residues for pesticides residues or JECFA which sets veterinary values for veterinary drugs.
In our dietary exposure assessment we do sometimes get situations where a chemical may have a chronic health‑based guidance value set in Australia but not an acute, but JMPR has actually set an acute value. In that case we would use the set of JMPR values so we always use the combination of the two together from whichever agencies set them.
Also we need to obviously talk with the toxicologists because we might need to find out, are there any particularly vulnerable populations that their toxicological studies have highlighted. For example, not a pesticide example but for mercury women of child‑bearing age are a particularly vulnerable group so we might need to do a dietary assessment on a sub‑population group. Other groups with different food consumption patterns that might put them more at risk than others and other different regional differences.
Where do we get our data from? For food consumption data we have actually just got a brand new set of data from the Australian Health Survey which was run from 2011 to 2013. There was a component of that called the National Nutrition and Physical Activity Survey that surveyed people aged two years and over. However we haven't quite integrated that into our risk assessments because it's always regulators lag behind the data that is there, I think. At FSANZ we actually have to do about six‑months work to prepare the data to put it into our modelling systems. Further through the talk I will explain why that work needs to be done.
We are currently using the 2007 National Children's Nutrition and Physical Activity Survey Data which looked at children aged two to 16 years, about 4000 children. For adults we still refer to the 1995 National Nutrition Survey. As you can imagine there are some issues with that because you think what we might have all been eating in 1995 compared with 20 years on, that's actually food consumption patterns have changed, new products are on the market, so that is one of the limitations of the data that we use if you like. The new data is actually available on the Bureau of Statistics website in a summary form.
The background to the new survey
The survey actually interviewed about 50,000participants in face‑to‑face interviews. About 34,000 in the general population surveyed and a subset was done on Aboriginal and Torres Strait Islander populations, about 15,000 of them. Maybe our visitors don't know, that population sub‑group is only about 2% to 3% of our population so that is a real oversampling of that population group to get valid statistical results. It's the first time we have actually got food consumption data in detail from that group.
The general survey started in March 2011 and results have been released since October 2012, with the ones we're interested in, the food and nutrient intake results, released from May 2014. The Aboriginal and Torres Strait Islander people's survey was a year later and the results have just started coming out from that survey too. We've done a little bit of analysis at FSANZ on differences between those groups, and for some of our risk assessments we may consider running a risk assessment for that particular subgroup. They, for example, drink more soft drinks than the general population. They eat more processed meats. So if we had a food additive in one of those products we might be interested in doing a special run. The implications are for pesticide residues and agvet chemicals but it looks like definitely those in remote areas are actually eating less fruit and veggies than the general population.
The component we're particularly interested in is the NNPAS as we call it, the Nutrition and Physical Activity Survey. How they got the data for this is they went into people's homes. About a third of the people taking part in the Australian survey did this component and two‑thirds did more general health questions. People were asked to do quite a complex interview that takes one to one‑and‑a‑half hours going through every meal they have had in the past 24 hours and getting the information on what they have eaten.
For example, if you look at breakfast, the first question will be, "What did you have for breakfast?" And you'll say, "Well, I had coffee and porridge." Then they will say, "What sort of coffee was it? What sort of milk did you put in it? Did you put sugar in it? What sort of porridge was it? Did you make it with milk? What sort of milk was it? What did you add to it? Did you add sugar? Did you add fruit?" That picture of what you have actually eaten builds up. What it is called is a multiple pass method and it is actually based on the NHANES survey that's undertaken in the US. For everybody who took part in that survey apart from those living in our very remote areas, were then asked to do a second day of dietary recall by telephone. It was on a different day of the week and at least a week after the first interview because you don't want people being interviewed the next day and perhaps eating leftovers from the day before and getting very similar food consumption patterns.
How do we actually get this data? How do we encourage people to remember what they have eaten in an accurate way? The interviewers, it was all done by computer so all the answers were logged into a computer and the questions were all computerised which cuts out some of the interview error. It was also prompted by booklets that the respondent was given to try and give some details on how much of a certain food they had actually eaten. If you had had a glass of wine you were asked to pick what the shape of the glass was and how full it was. The slices here, that was to determine how big your pizza was, how big your slice of cake was and how deep it was. From that sort of information we can calculate how much of that product you had. Then at FSANZ we do a lot of work out in the supermarkets and measuring things so we go out and buy pizzas and take them to bits, look at how much a slice of pizza weighs, what the components of the pizza are and sometimes we get to eat them afterwards.
We can then look at this food and the data we get on what people have eaten so you get up to 30, 40 foods per person per record, per individual record, and there is over 12,000 people in this survey so that is a lot of data on food and food consumption. Obviously we want to disaggregate that into different groups of food sometimes depending on what we are interested in.
For example, we might like to look where our energy comes from and we will see lo‑and‑behold, cereal and cereal‑based products actually form between them about 38% of our total energy. The biggest proportion is from the cakes, biscuits, pizzas, all the yummy sort of foods, if you like, and 18% from more basic cereal foods. Meat and poultry dishes about 14%, dairy and milk products about 11%, vegetables 7% and fruit 4%. That gives you some idea in terms of when you are doing a risk assessment, if you have got pesticide residues or agvet chemicals in some of these products, how that is going to pan out when you do a total dietary assessment.
Just out of interest, a lot of you have seen media in both countries about how much energy we get from discretionary foods, about 35% of our energy comes from what the nutritionists call discretionary foods which are those extra foods that we should only have little bits of, i.e. the top right‑hand group.
The first results, just to show you a couple of results of food product groups you might be interested in for meat products and dishes, that is 70% of people in Australia were consumers. Chicken was the most commonly consumed meat with about 31% of people consuming chicken. Approximately 2% of Australians avoided meat for either religious or ethical reasons. There are some side questions alongside the dietary survey as to why you eat the way you do eat. For milk product and dishes which includes anything made from milk, all the dairy products, 84% of people in Australia consume those. Increasing numbers of the reduced fat varieties, so around 39% of milk, 46% of yoghurt and 30% of cheese. About 4.5% of people reported avoiding milk and milk products because of intolerance or allergies.
Again, we've got similar results. I'm just going to give you a couple of group results for all the different food groups. For nutrition surveys foods tend to be grouped as to how we consume them. It's dairy, meat, eggs, fish, fruit, veggies. They are fairly simple food groups. About 60% of people report consuming foods from this group. Apples are our most commonly consumed ones followed by bananas. We are not always that good at meeting our own recommended dietary guidelines. The guideline is to eat two serves of fruit a day and only 54% of Australians met that. If you look at it for vegetables, the recommendation is to eat five serves of vegetables a day and only about 7% of Australians actually met that recommendation, so perhaps if we've got pesticide residues in veggies we don't need to be that worried. Maybe watch it when we all get better at eating our fruit and veggies.
That's the nutrition survey data but how do we need to use this data for our dietary exposure assessments for regulatory purposes? Some of the things we need to think about, what's the purpose of the assessment, what food consumption data do we actually have available and for whom? Are the data in the right format or appropriate for our assessment? What sort of food consumption data are they? Are they weighed? Some surveys on a small scale can actually weigh what people have eaten and do duplicate diet analysis. That's not feasible to do for 13,000 people. Some surveys go over several days of a survey. We do two days of a survey.
There is a lot of evidence to show that if you do more than three or four days of surveys, people actually change their food habits because they don't want to be bothered. If they make their own muesli every day they get fed up telling you how they have made muesli every day and they just say, "I had cornflakes." There is a bit of respondent fatigue if you like in dietary surveys so it is a balance between doing a short few days of survey as opposed to a longer time in terms of accuracy of that data. That actually poses us some problems when we are looking at chronic risk assessments as to how we deal with that. We also need to question, well, do we need to apply recipes to the mixed foods we have eaten to convert those mixed foods to raw commodity equivalents? Because I'm sure those of you in this area know that the MRLs, the regulatory limits we set for pesticide and veterinary drug residues, are actually on raw commodities, and we don't eat raw commodities, we eat mixed foods generally. So how do we go from one to the other?
As I said before, we need different food consumption data for our different sorts of assessments. For chronic assessments what we are interested in is mean daily consumption for the whole population. That might include eaters and non‑eaters of the foods that have the chemical of interest. In our case we have two days of survey data so we have a few options there of how we deal with that. We could just use day one of data but a better picture would be to use maybe an average of the two days of data so you are beginning to approximate a longer term. There are now some really more sophisticated statistical methods that have been developed by the National Cancer Institute in the US to be predict usual intake from two or more days of data that takes into account the frequency of consumption, age, sex parameters, day of the week that you actually eat the food on, and we are starting to bring those into our assessments. We have just set up that modelling capacity at FSANZ as a new innovation, if you like.
For acute assessments we don't want to average food over two days and why wouldn't we want to do that? Because what we are actually interested in there is what is the biggest amount of that food that you might eat in one day or in one meal? There, if we do have two different survey days for each individual we actually count them as two different records. One person will have two records of data and we don't combine them. It is a different way you manipulate the data and the assumptions and limitations are different for those two sorts of assessments. Really, if you are thinking about a chronic assessment, what we are trying to approximate is what happens over a lifetime. Obviously even two days of survey data doesn't really match a lifetime of exposure. In a sense we are using a cross‑section across different ages of people who take part in the survey as a proxy for lifetime of exposure at that point in time.
Which for a lot of chemicals is fine except it is obviously just at that point in history and it doesn't take into account how an older person might have eaten 20, 40, 50 years ago. But for some chemicals like dioxin you have to take other things into account because they have a body burden and so there are more complex dietary exposure assessment techniques where you can actually take the body burden of that chemical for people of different ages into account where you know, for example, like, the dioxin load in populations is decreasing over time so you assign a higher body burden to older people in your assessments.
We also historically have tended to do it chemical by chemical but of course we do know that some chemicals may have the same toxic endpoint. Where chemicals have light dioxin, there is different congeners, we can assign a toxicological factor to them which takes into account their different way they—how seriously they are toxic I suppose in lay terms. We can assign that factor to the different congeners to do a combined dioxin exposure assessment. That is really in its infancy as to how we actually can account for the fact that you might get different chemicals on the same plate of food.
Thinking about getting back to our data, how we actually get from one to the other from reports of what people have eaten. As I said, people report what they have eaten. Every food that people have eaten is coded. We might code it in a major food group like milk product and dishes, assign it number 19, and then within that we might have different classes of food like milk, yoghurt, cheese. You get a three‑digit number that designates that and then within, say, milk there will be different sorts of milk like full fat, semi, skim milk. They all get different codes so you can actually either assign a different nutrient value to those foods or you can actually assign a different residue level. If, for example, your residue sits in the fat, you might want to adjust for that for different sorts of foods.
Where we are trying to get to is actually the Codex classification of foods which is raw commodities. How do we do that? How do we translate what we have actually eaten? Say, all the different ways we have eaten pork, bacon and ham into one classification that is pig meat. How do we retrieve the pork in a sweet and sour stir‑fry so we can add that in too if it happens to have a veterinary chemical residue in it? How do we account for all the pork, bacon and ham, the bacon out of carbonara, how do we pull that out?
Our first step is to do direct mapping so where the food that you have actually eaten matches the raw commodity, as in fresh fruit and veggies, we can directly map it across. Sometimes though the form of the food we have eaten is not the same as the form the MRL has set. For example, the raw commodity might be coffee beans but we don't normally eat those so we drink coffee. We've got to apply an equivalence factor to the coffee to convert it to the equivalent amount of raw coffee beans. That will be different for different coffees, your Turkish coffee versus your instant coffee will have a different conversion factor.
What do we do with mixed foods like a latté which is obviously made of coffee and milk? We have a recipe for how much coffee and how much milk and then we apply the equivalence factor to the coffee to get it back to coffee beans. Where do we get our recipes from? My team are very good at cookbooks, researching the web for manufacturers' formulations, talking with people, looking at food labels and you develop an expertise. For an actual nutrition survey you can actually only have one recipe for lasagne so I'm sure if all of us went home and made lasagne it would all be slightly different. What we do for lasagne is sort of look at all the cookbooks, look at all the different sorts of ways you can make lasagne and do a sort of average lasagne recipe that takes into account the ingredients people normally add to lasagne.
As you can see that whole process of the dietary food consumption data has a lot of assumptions and limitations in terms of data uncertainty potentially in how we have derived that data and particularly how we derive the raw commodity data.
Then we look at the pesticide residue data so we can model with maximum residue limits. We can use analytical trial data or survey data. Sometimes we have to apply processing factors where we know the residue, say, from a raw grain to the flour that we actually put into our recipes, the recipe level would have changed. The trial might be submitted to APVMA as part of the application and that may have information like for example on a supervised median trial level. At FSANZ itself there is a total diet study, we might have included these sorts of chemicals in our data and actually have got levels in foods as consumed. We have got a choice there as to what we put in our model.
One of the things I did want to say though is that we actually need to check the residue definitions. The regulatory residue definitions may not be the residue definition that we need to use for modelling depending on the toxicological basis of the health‑based guidance value. We need to watch that too.
What is the difference between a chronic dietary exposure assessment and an acute? For a chronic we take our simple equation and we actually put in mean food consumption, include both eaters and non‑eaters, and we multiply that usually but a median, sometimes a mean food chemical concentration level. That could be the STMR or a survey level. We actually sum that across the diet for all the foods. I could have eaten, if a pesticide residue has permitted levels in fruit, vegetables and cereals, in my diet you have to include all the places where that chemical might have arisen in those foods, count them all up, retrieve them all from the recipes, add them up per person, get a total food consumption of all of those. Multiple them by the difficult chemical concentrations to get an exposure.
What you end up with is the distribution of exposures. Now, if you do Monte Carlo modelling, if you are lucky enough to have distribution of concentration data, you can actually do a full Monte Carlo assessment to combine those two distributions. But often we only have a single concentration data but we are combining it with 13,000 records of what people have eaten so we get a half a probabilistic approach, if you like. From that we can derive means or high percentile values of exposure and we can do that for everybody in the population or only people who have eaten the foods that contain the chemical you are interested in.
For an acute, as I said, this is a little different. Here in the dietary exposure you actually take the 97.5th food consumption amount but for only people who have eaten that food. That's a much higher amount. For example, you know, if you look at fish, I think the average consumption of fish across the whole population is about 7 grams per day. If you look at people who consume fish, it will be 150 to 160 grams a day, and that's a huge difference in consumption amounts. It can make a big difference to your calculations. We multiple what we call a high eater, high consumer food amount by a high residue level. You are assuming in your one plate you are unlucky enough when you have a big plate of carrots, to get the carrots with the high residue value as well. That is quite a conservative estimate if you like of an acute exposure and that is what you compare to the acute reference dose.
Because we are assuming you eat such a large amount of that one food, it would be really unrealistic to sum that across all foods. We do that individually by food by food by food because none of us eat, you know, the 97.5th amount of every food that we eat. That would be about 40 to 50 kg of food a day whereas on average we actually eat about one or two kilograms of food a day. You have to put some realism into these models.
We obviously make a lot of assumptions. We could do a thousand models and we choose basically to follow international modelling conventions, if you like. Some of the assumptions that we might make are that the actual data we have actually represents current eating patterns, and as I said, that the consumption across different age groups at one point in time represents a lifetime of chronic exposure. Depending what we do with the residue data, we could assume only a certain percentage of the crop was sprayed so we might only apply that concentration data to a proportion of the crop that we have eaten or the market share for a particular product or we might assume all the food contains that chemical. It is really important when we present that data to our risk managers and other risk assessors, that they understand what assumptions we have made so they know how to use that data.
Obviously the limitations, I'll discuss a couple of them. The data are never going to be perfectly accurate. We have got people involved recording what they eat. How many of you remember exactly what you had in the last 24 hours and who forgot to mention the chocolate bar? It may or may not represent our actual food consumption patterns. We have a problem with the age of food consumption data. That last survey took $50 million to do, that Australian Health Survey, we can't afford to do them that often so sometimes our data gets quite old between surveys.
Food consumption may be underreporting or some of it may be over‑reported. People might like to tell you they have eaten more fruit and veggies than they really have because that is what they know they are supposed to be doing. We may or not have good enough chemical concentration data. We can improve those limitations by better sampling approaches, more up‑to‑date data, better ways of manipulating that data. Once we have got that data obviously we compare it to the health‑based guidance value and that determines that comparison. Whether you are above or below that health standard determines whether an MRL can actually be established or the risk assessors need to go back and look at, well, do you need to extend the withholding period, can you not allow that chemical in a particular food because of the dietary exposure assessment results. It is an iterative process back and forwards at that stage once you've got your initial dietary exposure estimate.
This is just a couple of slides to say we also interact with the international community so we submit Australian data to JECFA and JMPR so that can be used in the international risk assessments. We submit data from our national nutrition surveys on the 97.5th and mean consumption of food amounts that is put in the international [0:30:45 JMs] food data that is used internationally.
Actually for pesticide residues they use apparent food consumption data that is developed by our agricultural department. That is really just what food is available for use so what we produce including imports minus exports minus wastage. That is all grouped together and a quite complex cluster analysis done on that information by key foods for countries round the world. Countries are grouped into groups of countries with similar food consumption patterns of those key foods.
Australia, there are 17 different groups. We happen to be with the UK and some Northern European countries and Uruguay, basically because we love cereals and dairy and beer. Interestingly enough New Zealand actually isn't in our group because they tend to have more non‑alcoholic beverages. We are higher on the alcoholic beverages. They have lower amounts of roots and tubers but they have a bit more cereals and poultry and eggs. Those diets are actually used by the JMPR committee when they are developing international MRLs.
The key message is really for any dietary exposure assessment that we really need to first determine why we are doing the assessment. We need to collect all our relevant information and we may need to interact with APVMA to actually get some of that information that we need. We choose the approach that is most suited to our purpose and the data available, and that will actually change nationally or internationally. We need to be really aware and document our assumptions, limitations and data uncertainties and continue to work with and talk with the risk assessors and risk managers in that process.
Errors and omissions excepted; check against delivery.