Q J Med 2001; 94: 341-346
© 2001 Association of Physicians
Review |
The use of capture-recapture techniques in determining the prevalence of type 2 diabetes
1 From the Department of Medicine, University Hospital Aintree, Liverpool and 2 Division of Tropical Medicine, Liverpool School of Tropical Medicine, Liverpool, UK
 |
Introduction |
|---|
 Top Introduction Traditional epidemiology and...Capture-recapture and diabetesThe technique of capture...Two-list capture-recaptureMultiple-list capture-recaptureCR in action for...ConclusionReferences |
|---|
Knowledge of the epidemiology of diabetes and its complications is vital for planning the provision of appropriate health care. Traditional methods of counting diabetes in a given population are sometimes not feasible because of time and financial restraints. A potential answer to these problems may be the use of ‘capture-recapture’ (CR) techniques. These have been applied to diabetes, primarily to assess the incidence of type 1 diabetes in children.1,2 Widespread use of these techniques to determine the prevalence of type 2 diabetes has yet to be realized. This review examines the advantages and limitations of CR techniques, particularly in the assessment of type 2 diabetes prevalence.
|
Traditional epidemiology and diabetes |
|---|
|
TopIntroductionTraditional epidemiology and...Capture-recapture and diabetesThe technique of capture...Two-list capture-recaptureMultiple-list capture-recaptureCR in action for...ConclusionReferences |
|---|
Cross-sectional diagnostic surveys, postal questionnaires, house-to-house surveys, and cohort surveys are the methods most commonly used to determine the prevalence or incidence of diabetes. Applying diagnostic tests such as the oral glucose tolerance test (OGTT) or fasting plasma glucose (FPG) in a given population may estimate the total prevalence of known and unknown (undiagnosed) diabetes. For known (rather than unknown plus known) prevalence, postal questionnaires or house-to-house surveys can be used.3,4 The outcome of such surveys depends very much on appropriate study design and the choice of samples.5 They are all labour-intensive and difficult to do on a regular basis. In certain populations, especially in developing countries, population surveys are difficult to carry out because of population mobility, lack of census data, and sometimes, public distrust.
A useful tool, particularly for prevalence studies, is the concept of population-based diabetes registries.6–9 These can be constructed in various ways and from various sources, but considerable effort is needed to establish and maintain them.6 This is because of the relatively large numbers involved, losses and additions to the population, and different sites of diabetic care. Such problems may lead to relatively low ascertainment, and confidentiality issues may sometimes be a problem.6,10 In the UK, there are a number of established regional registers of childhood diabetes. The Yorkshire Children's Diabetes Register (YCDR), for example, ascertained cases through three independent sources (i.e. hospital clinics, general practitioners, and hospital discharge lists).8 The European Community Concerted Action on Diabetes Epidemiology (EURODIAB) reported the use of drug sales data to estimate the total number of diabetic patients.11
Efficient drug databases, however, are not always available, and are of no use for estimation of the prevalence of type 2 diabetes, as substantial numbers are on dietary treatment alone. Drug databases may, nevertheless, be of use when combined with other sources—for example, hospital diabetic clinic lists, discharge records or general practitioner lists. As mentioned, there are problems of low ascertainment,10 as well as possible errors in baseline population estimates, diagnostic errors, and under-reporting.12 At least some of these problems can be overcome by combining such lists, preferably by computer—a system sometimes known as ‘electronic data linkage’. This approach has been used successfully in Dundee in the ‘DARTS’ (Diabetes Audit and Research in Tayside, Scotland) database.9 Multiple datasets of such quality are unfortunately rarely available.
Diabetes registries are said to be costly, although little definite information is available. Set-up costs for districts with populations in the region of 200 000 to 270 000 have been estimated at £5000–£24 000 (sterling) and annual maintenance costs at £13 000–£17 000.13 Less expensive registers however, can be created,14 and such data have clinical as well as epidemiological usefulness (e.g. complication audits, annual patient review, appointment reminders, etc). Multiple sources also lend themselves to an alternative method to simple electronic linkage, in order to reduce the problem of low ascertainment. This method is known as ‘capture-recapture’ (CR).
|
Capture-recapture and diabetes |
|---|
|
TopIntroductionTraditional epidemiology and...Capture-recapture and diabetesThe technique of capture...Two-list capture-recaptureMultiple-list capture-recaptureCR in action for...ConclusionReferences |
|---|
CR is a method for estimating the number of individuals in a closed population. The method acquired its name because it was developed by zoologists to count wildlife populations. To estimate the size of a particular bird species, for example, an ornithologist catches (captures) a sample of birds of that species, and marks and then releases them. After a given time, a second sample of birds of the same species is caught. The marked birds in the second catch are ‘recaptured’. Using the sample sizes of the two catches and the number recaptured, a simple formula based on the ‘dilution’ of marked birds in the second capture leads to an estimate of the total population size15,16 (see Figure 1
). The same approach may be used to estimate the size of human populations, but without the necessity to capture, mark and recapture. When lists of people with the characteristics of interest (e.g. diabetic patients) are available, these can be used as the ‘captures’. Those people present on two lists, (the duplicates) are in effect ‘recaptured’.17,18
|
CR techniques were used for the first time in human populations by Sekar and Deeming in 1949 to estimate birth and death rates in India.19 Shapiro used the technique to ascertain completeness of birth registration in the US.20 In recent years, the CR technique has been used by epidemiologists as a potentially useful technique for determining the prevalence or incidence of diabetes.21,22 As mentioned previously, however, most such studies have evaluated type 1 diabetes incidence. Some, however, have investigated type 2 diabetes prevalence,23 and these will be discussed in detail later.
|
The technique of capture-recapture |
|---|
|
TopIntroductionTraditional epidemiology and...Capture-recapture and diabetesThe technique of capture...Two-list capture-recaptureMultiple-list capture-recaptureCR in action for...ConclusionReferences |
|---|
Accurate matching of patients in different lists is vital, and cases must therefore have a unique identifier such as name, date of birth, national identification number or postal code. The number of cases appearing on different lists can then be used to calculate the missing cases, total cases and subsequently prevalence (as long as the total population is known). For reliable results, lists must fulfil the following basic assumptions.24
Assumption 1: the study population is closed
There should be no significant changes in the population under study during the given time period, for example by significant migration or death.
Assumption 2: the lists are independent of one another
This is not always the case.25,26 For example, a list of patients in a diabetic clinic may be similar to lists from family practitioners, because the patients seen at diabetes clinics are usually referred by family physicians. Fortunately, there are mathematical ways to identify and allow for such interdependence of sources, such as log-linear modelling.25,27 The use of multiple sources also helps to overcome this problem.
Assumption 3: all members of the population have the same probability of being captured
This may be influenced by choice of location of care, or by the frequency of attendance, especially if the period of the study is short. Since diabetic patients are heterogeneous in terms of age group and pattern of treatment, there is often an unequal probability of patients appearing in each list. For example, type 1 patients have a higher chance of appearing in hospital-based lists. To avoid this, many researchers subdivide their patients into different age groups and methods of treatment: a technique known as stratification.18,28 Total numbers of patients in the various subgroups are estimated, and group-specific prevalence rates determined.
Assumption 4: all population members can be matched on all lists
There should be at least one common identifier (e.g. a national identification number). In practice, first name, family name, and age or date of birth, are usually sufficient minimum data. In Britain, postal codes are also useful. There may be problems in developing countries, where names may be similar or variably used, dates of birth may not be known, and identification numbers may not be available.29
Data sources
In diabetes studies, the common sources used depend on the structure and level of medical care. Hospital diabetic clinics and family doctor lists are potential good sources of capture,23 but there may be problems with both these sources. The completeness of computerization of diabetic patients in both situations is variable. Hospital in-patient data notoriously under-record actual numbers of diabetic cases, sometimes by up to 40–60%.30,31 District diabetic registers are good sources for diabetic capture, usually with relatively high ascertainment levels. Type 1 diabetes provides an easier model, because of lower numbers and ease of diagnosis compared with type 2 diabetes. Wadsworth et al, for example, reported the use of the British Paediatric Surveillance Unit database in which consultant paediatricians reported newly diagnosed cases monthly.2 As a main source of capture, this gave a 78% ascertainment rate. Lists of prescriptions of drugs, syringes and reagent strips are sometimes computerized, but they often vary in their format and accuracy. Bruno and colleagues used these as sources of capture, and reported a 55% ascertainment rate for prescription lists, and 8% for syringe and reagent strip lists.23 Local diabetic association membership lists are a source of capture that gives variable ascertainment, but confidentiality of these lists sometimes makes them impossible to use.32 The optimum number of sources will be discussed in more detail in the next section. In general however, to increase accuracy and reduce the effect of dependence, more than two sources should ideally be used.
Patient identifiers
It is obviously vital that exact and adequate patient identifier characteristics are used. A combination of full name, gender and age is usually sufficient, though the addition of a postcode (if available) is useful.33 National identification numbers are obviously ideal, and have been used, for example in the enumeration of HIV-infected drug-users in Thailand.34 Surprisingly, such numbers are not used in many countries—notably the UK. Although in the UK all people have a unique NHS number, this does not always appear in health-related documents. In some areas, however, there is a locally used unique identifier (e.g. in Tayside, Scotland) and this can be used effectively.9 The accuracy of the identifier cannot be over-emphasized: in the absence of a single national number, different identifiers may give different results.34 Difficulties such as these have limited the effectiveness of CR techniques in developing countries.29
Matching of the patient identifiers is generally not a problem. The numbers involved in most lists make manual inspection laborious and inefficient, but many computer software packages can do the job efficiently and effectively (e.g. the Statistical Package for Social Sciences or SPSS).35
|
Two-list capture-recapture |
|---|
|
TopIntroductionTraditional epidemiology and...Capture-recapture and diabetesThe technique of capture...Two-list capture-recaptureMultiple-list capture-recaptureCR in action for...ConclusionReferences |
|---|
Two-list CR is the simplest technique, and is illustrated by the Venn diagram in Figure 2
. The box represents the total population required to be measured (n), and inside are the 2 overlapping lists (L1 and L2), and surrounding them are the missing cases (m). The total cases, n, can be calculated by Chapman's formula,15 which is as follows: |
|
|
|
|
|
Multiple-list capture-recapture |
|---|
|
TopIntroductionTraditional epidemiology and...Capture-recapture and diabetesThe technique of capture...Two-list capture-recaptureMultiple-list capture-recaptureCR in action for...ConclusionReferences |
|---|
Using three or more lists (see Figure 2
) allows greater accuracy and tends to reduce the difficulties of list dependence or independence. The calculations with multiple-list CR are highly complex, and usually involve Poisson log-linear modelling.36 This gives an estimate of missing cases, and also provides information on the interdependence of sources. Multiple-list CR has been widely used, and is now the preferred method.21 The complex statistics are greatly eased by the use of computer software such as ‘GLIM’.37 One question that arises over multiple-list CR is exactly how many lists should be used. Particularly with type 2 diabetes, there may be many available lists.23 However, the mathematics become increasingly complex beyond three lists, and recent studies have demonstrated that three good non-independent lists are optimal:38 beyond this there is little to be gained. It may therefore be worthwhile ‘collapsing’ four, five, or six lists into three, particularly those that may be dependent. An example of this is the Italian study referred to previously23 in which the hospital diabetic clinic and general practitioner lists were condensed into one. This provided a larger sample list which was more robust from a CR viewpoint, because hospital and GP lists are heavily co-dependent.
|
CR in action for type 2 diabetes prevalence studies |
|---|
|
TopIntroductionTraditional epidemiology and...Capture-recapture and diabetesThe technique of capture...Two-list capture-recaptureMultiple-list capture-recaptureCR in action for...ConclusionReferences |
|---|
As mentioned, CR has mostly been applied to type 1 diabetes, but there is some limited experience of CR in type 2 diabetes prevalence estimation. In Northern Italy, for example, Bruno and colleagues23 conducted a large population-based prevalence study of type 2 diabetes in Northern Italy. They used four sources as follows: (i) diabetic clinic and general practitioner lists; (ii) hospital discharge data; (iii) drug prescription records (insulin and oral agents); and (iv) prescription reimbursement requests. The first capture included both hospital and GP lists because of significant interdependence. An overall diabetes prevalence of 2.8% was found by CR calculations, with 95%CIs of 2.4–3.1%. For type 2 diabetes only, the prevalence was 2.7% (CI 2.6–2.8%). Our studies in Liverpool, UK, used a total of six separate lists:39 (i) hospital adult diabetic clinic register; (ii) children's hospital diabetic clinic list; (iii) hospital discharge data; (iv) hospital retinal clinic list; (v) research database for stroke admissions; and (vi) general practitioner lists. Lists (i) and (ii) were combined (1316 patients), as were lists (iii), (iv) and (v) (843 patients), but list (vi) was left as a single dataset (1468 patients). This provided three large lists with very little dependence. An overall CR-calculated diabetes prevalence of 3.1% (95%CI 3.0–3.2%) was obtained, compared to a figure of 1.5% (95%CI 1.4–1.5%) by electronic data linkage using the same datasets. The specific type 2 diabetes CR-adjusted prevalence rates in this study showed rates varying between 3.2% and 6.7% in different areas of the population, and there was a strong relationship between the prevalence of type 2 diabetes (but not of type 1), and indices of social deprivation.40
The Italian23 and British39,40 studies discussed above show how multiple incomplete lists of type 2 diabetic subjects, if handled correctly, can give rapid assessment of prevalence with ‘tight’ confidence intervals. Stratification by age, sex, type of diabetes, and postcode is simple and allows interesting correlations—such as that with social deprivation40—to be explored. These studies also demonstrate the superiority of CR over simple electronic data linkage.
|
|
Conclusion |
|---|
|
TopIntroductionTraditional epidemiology and...Capture-recapture and diabetesThe technique of capture...Two-list capture-recaptureMultiple-list capture-recaptureCR in action for...ConclusionReferences |
|---|
Capture-recapture has demonstrated its use as an epidemiological tool, and has been widely used for determination of the incidence of type 1 diabetes, but has so far been little used for type 2 diabetes prevalence estimation. We hope that in this review we have demonstrated the potential of the technique for the surveillance of type 2 diabetes. Certainly, with the current global pandemic of type 2 diabetes, simple and accurate techniques such as CR will be required to assess the rapidly increasing healthcare burden.41 However, the technique must be meticulously applied, as otherwise factors such as source interdependence and identifier failures can lead to seriously misleading results.34,42 CR is said to be rapid and inexpensive, but no detailed comparative cost-effectiveness assessments have been made. Similarly, results comparing CR with ‘gold-standard’ population surveys are lacking. CR in developing countries also requires considerably more investigation. Capture-recapture is an epidemiological promise partly fulfilled, but the methodology has much more to offer in many areas of medicine, of which the epidemiology of type 2 diabetes is but one example.
|
Notes |
|---|
Address correspondence to Dr G.V. Gill, Department of Medicine, University Hospital Aintree, Liverpool L9 7AL. e-mail: G.Gill{at}liv.ac.uk
|
References |
|---|
|
TopIntroductionTraditional epidemiology and...Capture-recapture and diabetesThe technique of capture...Two-list capture-recaptureMultiple-list capture-recaptureCR in action for...ConclusionReferences |
|---|
1. Gujural JS, McNally PG, Botha JL, Burden AC. Childhood-onset diabetes in the white and south Asian population in Leicestershire, UK. Diabet Med1994; 11:570–2.[Web of Science][Medline]
2. Wadsworth E, Shield J, Hunt L, Baum D. Insulin-dependent diabetes in children under 5: incidence and ascertainment validation for 1992. Br Med J1995; 11:700–3.
3. Neil HAW, Gatling W, Mather HM, Thompson AV, Thorogood M, Fowler GH et al. The Oxford Community Diabetes Study: evidence for an increase in the prevalence of known diabetes in Great Britain. Diabet Med1987; 539–43.
4. Mather H, Keen H. The Southall Diabetes Survey: prevalence of known diabetes in Asians and Europeans. Br Med J1985; 291:1081–4.
5. Dowse GK, Zimmet P. A model protocol for a diabetes and other non-communicable disease field survey. World Health Stat Quart1992; 45:360–9.[Medline]
6. Burnett SD, Woolf CM, Yudkin JS. Developing a diabetic register. Br Med J1992; 305:627–30.
7. Karvonen M, Tuomilehto J, Libman L, LaPorte RE. A review of the recent epidemiological data on the worldwide incidence of type 1 (insulin-dependent) diabetes mellitus. Acta Diabetologia1993; 36:883–92.
8. McKinney PA, Law GR, Bodansky HJ, Staines A, Williams DDR. Geographical mapping of childhood diabetes in the Northern English County of Yorkshire. Diabet Med1996; 13:734–40.[Web of Science][Medline]
9.
Morris AD, Boyle DIR, MacAlpine R. for the DARTS/MEMO Collaboration. The Diabetes Audit and Research in Tayside, Scotland (DARTS) study: electronic record linkage to create a diabetes register. Br Med J1997; 315:524–82.
10. Dunn NR, Bough P. Standards of care of diabetic patients in a typical English community. Br J Gen Practice1996; 46:401–5.[Web of Science][Medline]
11. Papoz L and the EURODIAB Subarea C Study Group. Utilisation of drug sales data for the epidemiology of chronic disease: the example of diabetes. Epidemiology1993; 421–7.
12. Schraer CD, Lanier AP, Boyko EJ, Gohdes D, Murphy NJ. Prevalence of diabetes mellitus in Alaskan Eskimos, Indian and Aleuts. Diabetes Care1998; 11:693–700.[Abstract]
13. Buxton C, Gibby O, Hall M, Home P, MacKinnon M, Murphy M et al. Diabetes information systems: a key to improving the quality of diabetes care. Diabet Med1996; 13:S122–8.[Web of Science][Medline]
14. Ismail AA, Gill GV, Houghton GM, Beeching NJ. A simple and cost-effective diabetic clinic database. Practical Diabetes Int1999; 16:237–40.
15. Chapman DG. The estimation of biological populations. Ann Math Stat1954; 25:1–15.
16. Seber GF. The Estimation of Animal Abundance and Related Parameters, 2nd edn. Griffin, London, 1982.
17. Wittes J, Colton T, Sidel VW. A generalisation of the simple capture-recapture model with applications to epidemiological research. J Chronic Disease1968; 21:287–301.[Web of Science][Medline]
18. Wittes J, Colton T, Sidel V. Capture-recapture methods for assessing the completeness of case ascertainment when using multiple information sources. J Chronic Dis1974; 27:25–36.[Web of Science][Medline]
19. Sekar C, Deming EW. On a method of ascertaining birth and death rates and extent of registration. J Am Stat Assoc1949; 44:101–5.[Web of Science]
20. Shapiro S. Estimating birth registration completeness. J Am Stat Assoc1949; 45:261–4.
21. LaPorte RE, McCarty D, Bruno G, Tajima N, Baba S. Counting diabetes in the next millennium, application of capture-recapture technology. Diabetes Care1993; 16:528–34.[Abstract]
22.
LaPorte RE. Assessing the human condition: capture-recapture techniques. Br Med J1994; 308:5–6.
23. Bruno G, Bargero G, Vuolo A, Pisu E, Pagano G. A population-based prevalence survey of known diabetes mellitus in northern Italy based upon multiple independent sources of ascertainment. Diabetologia1992; 35:851–6.[Web of Science][Medline]
24.
International Working Group for Disease Monitoring and Forecasting. Capture-recapture and multiple-record systems estimation I: history and theoretical development. Am J Epidemiol1995; 142:1047–58.
25.
McCarty DJ, Tull ES, Moy CS, Kwoh CK, LaPorte RE. Ascertainment corrected rates: application of capture-recapture methods. Int J Epidemiol1993; 22:559–65.
26. Brenner H. Use and limitations of the capture-recapture method in disease monitoring with two dependent sources. Epidemiology1995; 42–8.
27. Bishop YMM, Fienberg SE, Holland PW. Discrete Multivariate Analysis: Theory and Practice. Cambridge, USA, 1975: 229–56.
28. Bruno G, LaPorte RE, Merletti F, Biggeri A, McCarty D, Pagano G. National Diabetes Programs, application of capture-recapture to count diabetes? Diabetes Care1994; 17:538–55.
29.
Black JFP, McLarty DG, Mtasiwa D. Capture-recapture techniques: difficult to use in developing countries (letter). Br Med J1994; 308:531.
30. Williams DDR, Fuller J, Stevens LK. Validity of routinely collected hospital admissions data on diabetes. Diabet Med1989; 320–4.
31. Leslie PJ, Patrick AW, Hepburn DA, Scougal IJ, Frier BM. Hospital in-patient statistics underestimate the morbidity associated with diabetes mellitus. Diabet Med1992; 379–85.
32.
Lynge E. New draft on European directive on confidential data. Br Med J1995; 310:1024.
33. Ismail AA, Gill GV. The epidemiology of type 2 diabetes and its current measurement. In Gill GV, MacFarlane IA, eds. Clinical Endocrinology & Metabolism: Frontiers in Clinical Diabetology. London, Bailliere Tindall, 1999;197–230.
34.
Mastro TD, Kitayaporn D, Weniger BG. Estimating the number of HIV-infected drug users in Bangkok: a capture-recapture method. Am J Public Health1994; 84:1094–9.
35. Norussis MJ. SPSS for Windows Base: System User's Guide, Release 6.0. SPSS, Chicago, 1993.
36. Cormack RM. Log-linear models for capture-recapture. Biometrics1989; 45:395–413.[Web of Science]
37. Francis B, Green M, Payne C. GLIM 4: The Statistical System for Generalised Linear Interactive Modelling. New York, Oxford Science Publications, 1983.
38.
Ismail AA, Beeching NJ, Gill GV, Bellis MA. How many data sources are needed to determine diabetes prevalence by capture-recapture? Int J Epidemiol2000; 29:536–41.
39. Ismail AA, Gill GV, Beeching NJ. The use of capture-recapture techniques to determine ‘missing’ diabetic patients in an urban population. Diabet Med1999; 16 (suppl.1):59.
40.
Ismail AA, Beeching NJ, Gill GV, Bellis MA. Capture-recapture adjusted prevalence rates of type 2 diabetes are related to social deprivation. Q J Med1999; 92:707–10.
41. Zimmet P, Lefebvre P. The global NIDDM epidemic: treating the disease and ignoring the symptoms. Diabetologia1996; 39:1247–8.[Web of Science][Medline]
42. Walters WE. Capture-recapture techniques: more unreliable in humans than birds. Br Med J1994; 308:531.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  T. P. EISELE, K. A. LINDBLADE, D. H. ROSEN, F. ODHIAMBO, J. M. VULULE, and L. SLUTSKER EVALUATING THE COMPLETENESS OF DEMOGRAPHIC SURVEILLANCE OF CHILDREN LESS THAN FIVE YEARS OLD IN WESTERN KENYA: A CAPTURE-RECAPTURE APPROACH Am J Trop Med Hyg, July 1, 2003; 69(1): 92 - 97. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||