Science and Technology CommitteeWritten evidence submitted by Dr James Kirkbride (Census 16)

Re: Census and Beyond 2011

I am writing to share my views of how Census data is currently used in the research that I conduct. My specialism is epidemiology, and in particular, psychiatric epidemiology. While I will naturally draw upon my own experiences of how the Census data is used for epidemiological research into psychiatric disorders, much of what I say will apply to epidemiologists and public health practitioners interested in a range of other population health issues. To structure my comments, I have used the question headings you provided for this call. I, as other academic colleagues I work with, are grateful that you are taking on board the views of those engaged in social science and public health research in moving forward from the Census.

1. How do social scientists use Census data?

In my field we use Census data to provide an accurate estimate of the population who may be “at-risk” of developing a disorder in a given catchment area. This catchment area may be geographically focussed ie on just a single Borough, or may be broad ie an estimate of the population at the national level. This data is primarily used as the “denominator” number which goes into calculating estimates of the incidence (number of new cases) and prevalence (number of new and existing cases) of a particular disorder. Estimates of incidence and prevalence are vital for several reasons related to population health. These include:

(i)accurate health service planning;

(ii)clues for aetiological (ie causal) research;

(iii)information to develop potential public health campaigns (prevention and health promotion);

(iv)Health economics—to estimate the burden of disorder(s) and potential associated costs in a given population.

Of these, it is the second (ii) with which I have most direct experience. By calculating incidence rates of disorders in specific populations (and communities) we can look at variation in those rates along several factors which might be of interest to us. These include looking at how incidence rates of disorder vary geographically, or vary by other factors such as age, sex, ethnicity or occupation. Identifying variation in incidence rates can provide clues about the aetiology (or cause) of a disorder, which can advance scientific understanding, reduce the burden of disorder (through prevention strategies which follow) and reduce the direct and indirect economic costs to health services and society associated with disease. The census provides essential underlying denominator data, stratified according to such factors, to enable us to estimate these incidence rates. One of the unique aspects of the census here is that the data can be stratified according to more than one variable at a range of spatial geographies. By having this information, we are able to know how many people in a given geographical location belong to each group we are interested in—for example, we can split the population by age, sex and ethnicity. By knowing this, we can estimate the number of cases of a given disorder which occur in particular groups of people, allowing for the estimation of incidence in different groups. By comparing incidence in different groups, we get valuable information about the burden and impact of disease in different communities. For example, we have used census data in this way to show that men have greater rates of psychotic disorder than women at young ages, but that this pattern reverses in mid-life(Kirkbride et al, 2006), and we have also shown that black and minority ethnic groups have higher rates of psychotic disorder than their white British counterparts (Coid et al, 2008, Fearon et al, 2006, Kirkbride et al, 2008b); vital public health information and important in understanding the causes of such disorder. Furthermore, because census estimates of the population can be stratified across several factors at once (age, sex, ethnicity, etc…) this allows us to build this complexity into our models of health and disease, controlling for the effect of one variable upon another, to eliminate confounding—the chance that an association between disease and a risk factor is really due to another factor. For example, in our 2008 paper, we used census data in East London to show that the incidence rate of schizophrenia remained elevated for ethnic minority groups, after taking into account possible differences between these populations and the white British group in terms of age, sex or social class, which could otherwise have explained this association.

Accurate estimates of the population at-risk—provided for by the census from the national level to the level of output area—are vital for understanding variation in the incidence and prevalence of a variety of diseases, which can then be important for aetiology (causal research), health services planning, health economics and public health. Whatever supersedes the census will need to provide accurate estimates of population data, stratifiable according to several factors, down to fine spatial (ie geographical) scales.

2. What impact will the ending of the Census have on social science research?

Without an adequate replacement for the census we will not be able to conduct the type of research outlined above. This will have major impacts on our ability to detect variation in the rates of disease, geographically and according to different sociodemographic groups, meaning that models of service provision will be less accurate, it will become harder to investigate possible aetiological associations between sociodemographic factors and disease, and harder to forecast (predict) which groups are at greatest risk of developing illness. These consequences will have downstream effects for health economics.

A particular aspect of my social science research—and that of many others—which will be affected is our ability to study geographical variation in the incidence of health (well-being) and disease. Currently population estimates from the census for different geographical aggregations (particularly the super output area and the electoral ward) allow us to estimate rates of disorders in small communities. If these rates vary, we can discover whether any social or physical factor in our environments varies with them, which has possible implications for causal understanding. For example, in psychiatric epidemiology we have used census data to show that the incidence of schizophrenia varies across neighbourhoods (Kirkbride et al, 2007, Kirkbride et al, 2006), and that this variation is associated with certain characteristics of neighbourhoods (Kirkbride et al, 2008a), including deprivation and social capital (more urban, more deprived, less socially cohesive neighbourhoods have more schizophrenia). Interestingly, these findings have motivated neuroscientists to study differences in the brains of people living in rural and urban environments, finding that those people in urban environments have a greater reaction to stressful events. Thus, the epidemiology (facilitated by the census) provided a testable hypothesis for neuroscientists, with the results suggesting that urban environments may alter the way we process information in our brain, and very tentatively, govern our risk of developing certain psychiatric disorders.

This is the kind of translational research which is threatened (and which will be lost to overseas competitors) if the census ceases to exist, or if suitable equivalent data is not made available in its place.

3. What alternatives to the Census would provide population and socio-demographic data of equivalent or higher quality?

The census provides valuable, accurate estimates of the population. As with any statistics there is a degree of error in their estimation, but for the kind of social science literature I have outlined, they represent the gold standard, with currently available UK alternatives lagging some way behind in scope, scale and precision. The alternatives which would provide equivalent or better data for population health research of this kind can be separated into (i) research-led initiatives or (ii) government-led initiatives:

(i)Research-led initiatives would include certain study designs—like birth cohorts—which follow-up a group of people born at the same time, with repeated assessments at various stages over the life course, for the duration of their lives. Examples in the UK exist, such as the 1946 or 1958 Medical Research Council birth cohorts. The advantages of such studies is that they provide prospective clues for causal research of health and disease (ie the risk factor was measured before the person became ill), providing great power to detect causal associations. However, such studies are costly, have issues with loss to follow-up, take a long time (several decades) and even the largest are small relative to the rate of certain disorders. For example, the 1946 birth cohort contained only ~5000 singletons in the sample. For disorders such as schizophrenia, what we know from epidemiology is that this would yield only roughly 50 cases over the entire life time of the participant members; too small for meaningful population research. Thus, I believe such options, while powerful in their ability to detect associations between disease and risk factor, do not provide adequate population data for the type of research currently facilitated by the census.

(ii)Government-led initiatives—one possibility, although it shamefully does not currently exist in the UK—is to develop national registers capable of linking up a range of government databases on the population to provide extremely valuable, prospective, nationally-collected linked datasets for the study of health and disease. Such registers are commonplace in Scandinavian countries where databases are linked through people’s social security number. Such data is extremely powerful to detect possible associations between disease and risk factors and provides population level coverage. However, such databases no doubt have implications for how data is shared between government agencies, and the fiasco over ID cards illustrates the difficulties in setting up such a registry system, though it would be a valuable—and in some ways more powerful—method of analysing health data in the population.

4. What other existing sources of population and socio-demographic data could be improved upon?

Unsure. There is nothing currently available in my eyes which comes close to providing the rigour and precision of the census. Random small sampling with extrapolation to the remainder of the population is a possibility, but samples will rarely be representative, even if participation if selected is made compulsory. Another option would be to link up existing government databases for such research, but this has massive cost, logistical and moral challenges. At best, the decision to scrap the census is very disappointing for social science research, and at worst, this could set back some areas of social research by several decades.

I am very grateful to the Committee for giving researchers the opportunity to express their views on this matter. It is of grave concern that this resource should be superseded with something capable of permitting (and ideally improving) upon the type of population-based research made possible by the Census. The estimation of incidence and prevalence rates (of any health outcome) will only be as good as the denominator data from which those data are ascertained.

I have no conflicts of interest to declare.

I am a Wellcome Trust research fellow based at the University of Cambridge.

References

Coid, J W, Kirkbride, J B, Barker, D, Cowden, F, Stamps, R, Yang, M & Jones, P B (2008). Raised incidence rates of all psychoses among migrant groups: findings from the East London first episode psychosis study. Arch Gen Psych 65, 1250-1258.

Fearon, P, Kirkbride, J B, Morgan, C, Dazzan, P, Morgan, K, Lloyd, T, Hutchinson, G, Tarrant, J, Lun Alan Fung, W, Holloway, J, Mallett, R, Harrison, G, Leff, J, Jones, P B & Murray, R M (2006). Incidence of schizophrenia and other psychoses in ethnic minority groups: results from the MRC AESOP Study. Psychological Medicine 36, 1541-50.

Kirkbride, J, Boydell, J, Ploubidis, G, Morgan, C, Dazzan, P, McKenzie, K, Murray, R & Jones, P (2008a). Testing the association between the incidence of schizophrenia and social capital in an urban area. Psychological Medicine 38, 1083-94.

Kirkbride, J B, Barker, D, Cowden, F, Stamps, R, Yang, M, Jones, P B & Coid, J W (2008b). Psychoses, ethnicity and socio-economic status. British Journal of Psychiatry 193, 18-24.

Kirkbride, J B, Fearon, P, Morgan, C, Dazzan, P, Morgan, K, Murray, R M & Jones, P B (2007). Neighbourhood variation in the incidence of psychotic disorders in Southeast London. Social Psychiatry and Psychiatric Epidemiology 42, 438-445.

Kirkbride, J B, Fearon, P, Morgan, C, Dazzan, P, Morgan, K, Tarrant, J, Lloyd, T, Holloway, J, Hutchinson, G, Leff, J P, Mallett, R M, Harrison, G L, Murray, R M & Jones, P B (2006). Heterogeneity in Incidence Rates of Schizophrenia and Other Psychotic Syndromes: Findings From the 3-Center ÆSOP Study. Archives of General Psychiatry 63, 250-258.

Dr James Kirkbride PhD (Cantab), MSc, D.LSHTM, BA (Hons)

29 November 2011