Advertisement

The International Diabetes Federation diabetes atlas methodology for estimating global and national prevalence of diabetes in adults

Published:November 21, 2011DOI:https://doi.org/10.1016/j.diabres.2011.10.040

      Abstract

      Introduction

      Diabetes is a major cause of morbidity and mortality and its global prevalence is growing rapidly. A simple and robust approach to estimate the prevalence of diabetes is essential for governments to set priorities on how to meet the challenges of the disease. The International Diabetes Federation has developed a methodology for generating country-level estimates of diabetes prevalence in adults (20–79 years).

      Methods

      Using country-level data sources from peer-reviewed studies, national health statistics reports, commissioned studies on diabetes prevalence, and unpublished data obtained through personal communication, we use logistic regression to generate estimates of the prevalence of diabetes. An approach matching countries on ethnicity, geography, and income group is used to fill in gaps where original data sources are not available. The methodology also uses changes in urbanization and population to generate estimates and projections on the prevalence of diabetes in adults.

      Conclusion

      Diabetes prevalence estimates are very sensitive to the data from which they are derived. The revised IDF methodology for estimating diabetes prevalence is a transparent, reproducible approach that will be updated annually. It takes data-driven approaches to filling in gaps where data are not available and where assumptions have to be made. It uses a qualification system to rank data sources so that only the highest quality data are used.

      Keywords

      1. Introduction

      No country is free of diabetes. The prevalence of diabetes is growing rapidly around the world and is a major cause of morbidity and mortality [
      • Shaw J.E.
      • Sicree R.A.
      • Zimmet P.Z.
      Global estimates of the prevalence of diabetes for 2010 and 2030.
      ,
      • Danaie G.
      • Finucane M.M.
      • Lu Y.
      • Singh G.M.
      • Cowan M.J.
      • Paciorek C.J.
      • et al.
      National, regional, and global trends in fasting plasma glucose and diabetes prevalence since 1980: systematic analysis of health examination surveys and epidemiological studies with 370 country-years and 2·7 million participants.
      ] and it is therefore important to monitor the development of this global epidemic. The pace of change in some countries as a result of rapid urbanization and increases in life expectancy means that regular updates of estimates of the prevalence based surveys or other data sources are necessary to provide evidence-based estimates for policy makers to meet the needs of diabetes.
      The growing number of publications reporting diabetes prevalence has increased the need to develop a systematic way of assessing studies and their viability for estimating the prevalence of diabetes. The International Diabetes Federation (IDF) has produced estimates of diabetes prevalence over the last decade [
      • International Diabetes Federation
      The diabetes atlas.
      ,
      • International Diabetes Federation
      The IDF diabetes atlas.
      ] and this paper describes and examines new methods developed by IDF to systematically selected and process data sources in a transparent manner.

      2. Data search

      We performed a systematic literature search of PubMed, Google Scholar, relevant citations from within papers, and gathered information from the IDF network beginning in November 2010 and ending in April 2011 using the search terms: ‘diabetes’ or ‘impaired glucose tolerance’ and ‘prevalence’ and <country name> or <region/continent>, ‘cardiovascular risk factors’ and <country name> or <region/continent>. Studies conducted prior to 1980 were excluded.
      In addition, we searched Ministry of Health and associated government websites in each country for relevant data. We also obtained reports from the WHO STEPwise approach to Surveillance (STEPS) studies and reports were obtained from the World Health Organization website for each country. Diabetes researchers in each IDF region were contacted and requested to provide information on the prevalence of diabetes for countries within their region. In addition, information was obtained through the IDF volunteer network and member associations.
      The search for data identified 565 studies containing prevalence data for either diabetes, impaired glucose tolerance, or both. These included 483 peer-reviewed publications from scientific journals, 25 national health surveys, four sources provided from personal communication and 53 other official reports from international agencies and government surveys.

      3. Data storage

      Data entry and data cleaning were performed regularly from November 2010 to April 2011. Data from all 565 studies were stored in a MySQL database back-end with an OpenOffice Base front-end. All programming was done using the R Project for Statistical Computing version 2.13.2 [
      • R Development Core Team
      R: a language and environment for statistical computing.
      ] (www.r-project.org). Data were stratified by age, gender, disease (diabetes or IGT) and setting (urban or rural). Where prevalence data were not disaggregated by gender but only by age group, they were stored as ‘unspecified’ gender. Similarly, where prevalence data were not disaggregated into urban and rural populations, it was stored as ‘unspecified’ setting.
      Studies were carefully reviewed to identify possible duplicate data sources. In case of duplication of study data, information for each paper was stored but excluded from further selection and only the paper reporting the most complete data or the most recent paper was used.
      For each data source country, title, abstract, authors, PubMed ID, and publication year were stored in the database. In addition, fields characterizing the study were stored as shown in Table 1. Study information was stored for all sources, but studies were excluded if they met the following criteria: insufficient data for characterization or modelling, duplicate data, an update of a previous study (if a study with the same methodology and study design was repeated in subsequent years for the same population, we used only the most recent data source), inadequate description of the methodology, clinic-based or hospital-based studies.
      Table 1Study characterization fields.
      FieldOptionsDescription
      Type of dataPeer-reviewed publicationPublication in a peer-reviewed scientific journal
      National health surveyA report published exclusively by a Ministry of Health or comparable government body on a national survey for health which includes information on diabetes
      Personal communicationInformation provided by direct contact with an investigator
      Other official reportA publication on health statistics that is carried out by another organization such as the WHO or CDC which is not peer-reviewed or published exclusively by a government agency
      Study designPopulation-basedThe sample was taken at random or systematically from the entire population in a given area
      Registry-basedThe sample was taken from a disease registry, a diabetes-specific registry, or from a disease surveillance system
      Modelled dataResults based on modelling from a number of data sources combined through meta-analysis or similar pooling
      Clinic-basedA study conducted using patients in a clinic or hospital setting
      Sample representationNationally representativeSamples more than one region within a country and represents a significant proportion of the total population or represents the demographics of the national population
      Regional representationSamples more than one city, town or village in a single region of the country
      Local representationSamples a single city, town, or village
      Ethnic or other specific groupSamples a single ethnic group exclusively, or another defined group such as a socio-economic stratum, workers in a particular industry, or people at a polling station
      Diagnostic criteriaOGTTOral glucose tolerance test
      FBGFasting whole or capillary blood glucose
      HbA1cGlycosylated haemoglobin
      GlycosuriaUrine glucose
      Self-reportAny self-identifying as diagnosed with diabetes by a health professional
      Sample sizeThe number of the total sample sizeThe sample size used for the entire study
      Study dateYearThe year in which the study was conducted

      4. Source characterization

      All studies were subject to a study selection process to choose only those of suitable quality to generate national and global estimates of diabetes and IGT. In order to establish a selection process, we convened a panel of experts from each of the IDF regions. We adopted a validated approach commonly used in operations research, the analytic hierarchy process [
      • Saaty T.L.
      Analytic hierarchy process.
      ] which is used to quantify the relative value of different alternatives. We selected the six criteria presented in Table 1 to characterize and score studies. The expert panel was asked to complete a questionnaire rating individual pairs of criteria against each other according to the importance they placed on each item with respect to selecting data for the generation of estimates. Each set of answers was then combined and the median taken to provide a composite response based on the value selected by the group. The results of this process are presented in Fig. 1.
      Figure thumbnail gr1
      Fig. 1Results of the analytic hierarchy process for weighting study criteria*.
      Based on the group preference, a value was assigned for each pairwise comparison. Those pairwise comparisons were then input into a six-factor comparison matrix and assigned a priority weight using matrix algebra. The priority weight reflects the relative importance of one factor with respect to all other factors based on the choices of the group. In addition, the same process was used for ranking the detailed information in each criterion presented in Table 2, that make up the higher level comparisons.
      Table 2Weights determined from the analytic hierarchy process for scoring studies.
      DomainItemWeight
      RepresentationNationally representative0.21049
      DesignPopulation-based0.11313
      AgeLess than 5 years old0.11245
      DiagnosticSelf-report + OGTT0.08851
      RepresentationRegionally representative0.07974
      Size5000 or greater0.07803
      Age5 years to 9 years0.04565
      DiagnosticSelf-report + FBG0.04307
      Size1500–49990.03139
      DiagnosticMedical record or clinical diagnosis0.03072
      RepresentationLocally representative (one city/town or less)0.02806
      DesignDisease-registry based0.02760
      DiagnosticSelf-report + HbA1c0.02527
      Age10–19 years0.01991
      RepresentationA single ethnic group0.01063
      DesignMedical record review0.00993
      Size700–14990.00939
      Age20+ years old0.00813
      DesignStatistical modelling0.00719
      TypePeer-reviewed publication0.00639
      TypeNational health survey report0.00440
      SizeLess than 7000.00385
      DiagnosticSelf-report0.00357
      TypeOther official report0.00192
      TypePersonal communication0.00060
      A composite score was obtained from adding up all the priority weights for each criterion from a study. The resulting weights are presented in Table 2. Nationally representative, population-based sources using oral glucose tolerance test as the primary diagnostic criteria that were conducted in the last 5 years received the highest score. The lowest score corresponded to data sources based on statistical modelling with less than 700 people determined by self-reported diabetes and conducted more than 20 years ago. In general, representativeness was the highest weighted criterion, while type of publication was considered the least important criterion and did not contribute substantially to the overall score. Data sources that were based on medical or linkage records had their sample size capped at 4999 to prevent undue weight being given to those studies which covered a large number of records, but where this did not necessarily reflect the representativeness of the study.

      5. Source selection

      In order to determine which studies to use for the generation of the estimates, a frequency plot of the final scores of all the sources was generated and thresholds were chosen accordingly (Fig. 2). Source scores followed a bimodal distribution, with half of all sources scoring at 0.3 or above. After examining sources scoring above and below 0.3, we set the minimum threshold at that level so that any source scoring below 0.3 was automatically rejected. We chose an upper threshold of 0.52, corresponding to a gap in the distribution, above which sources were all of high quality and were automatically included. If a country had sources with scores in the middle range (0.3–0.52) and no sources in the upper range, the highest scoring source was selected along with any others within 0.1 of that source (equivalent to one standard deviation of that score). A total of 170 sources from 110 countries were selected for generating the estimates of diabetes and IGT.
      Figure thumbnail gr2
      Fig. 2Histogram of scored studies with thresholds for inclusion and exclusion.

      6. Assigning sources for country estimates

      If sources were selected for a particular country, these were the only studies used to generate prevalence estimates for that country. Where there was more than one study selected, the studies were averaged using the weights determined by the score. The weights were rescaled so that the sum of the scores for all sources for that country totaled 1. As a result, a higher scoring source contributed more towards the estimates than a source that was selected but with a lower score.
      If a country did not have any selected data sources, prevalence estimates were generated using an average of available sources from countries matched by IDF region, World Bank Income classification group [], geography, and ethnicity []. Country groupings for missing data are presented in Appendix 1. Because there were fewer sources reporting data on IGT, a broader definition was used for grouping countries for IGT estimates (Appendix 2).

      7. Missing gender information

      Prevalence estimates for diabetes and IGT were calculated separately for males and females. Where a source did not specify gender or provided information for only one gender, we assumed that the prevalence was the same for males and females in each age group.

      8. Information missing by urban and rural setting

      The estimates take into account differences between prevalence of diabetes in urban and rural setting. We used data from studies scoring above 0.3 to generate estimates of the median ratio of urban to rural cases for different countries grouped by geography and income group (Table 3). For nationally representative studies that did not present results stratified by urban and rural setting, the estimated ratio of urban to rural prevalence of disease was used to stratify the results by setting. In some cases, studies did not present the results stratified by urban and rural setting, but did provide a ratio of urban to rural cases. In these cases, the original ratio of the study was used for stratification rather than the median derived ratio from Table 3.
      Table 3Ratio of urban to rural cases from various data sources of diabetes by region and income group.
      Data regionCountries providing dataUrban to rural ratio
      AFR-MIC2.00
      AFR-LICBenin, Gambia, Guinea, Kenya, Mozambique, United Republic of Tanzania2.48
      EUR-HICGreece, Hungary1.16
      EUR-MICTurkey1.30
      EUR-LICUzbekistan1.57
      MENA-HICOman, Saudi Arabia1.73
      MENA-MICAlgeria, Morocco, Occupied Palestinian Territory, Pakistan, Sudan, Tunisia1.58
      MENA-LIC1.80
      NAC-HICUnited States of America, Bermuda1.00
      NAC-MICBelize, Mexico1.13
      NAC-LIC1.30
      SACA-MICCosta Rica, Dominican Republic1.40
      SEA-LICBangladesh, Nepal4.70
      SEA-MICIndia, Sri Lanka1.97
      WP-HICRepublic of Korea1.01
      WP-LICCambodia, Myanmar2.25
      WP-MIC
      Although studies from Fiji, Kiribati, Nauru and Samoa provided information on the urban to rural ratio in prevalence, these small island states reported a 1:1 ratio and were not considered appropriate for extrapolation to the rest of the WP-MIC region.
      Malaysia, Philippines, Thailand1.40
      Note: There were five regions for which there was not enough information on the urban to rural ratio, but where this information was necessary to fill in gaps in the data: AFR-MIC, MENA-LIC, NAC-HIC, NAC-LIC, SACA-HIC. In order to fill in these gaps, we examined the percent differences in the ratios between other income groups for other regions. We found that the smallest differences were in different income groups in Europe (17%). In order to produce more conservative estimates, we used this difference to calculate the urban to rural ratio for the missing income groups of other regions.
      a Although studies from Fiji, Kiribati, Nauru and Samoa provided information on the urban to rural ratio in prevalence, these small island states reported a 1:1 ratio and were not considered appropriate for extrapolation to the rest of the WP-MIC region.
      The proportion of urban population in a country was obtained from the United Nations Population Division 2010 Revision estimates [
      • United Nations
      World population prospects: the 2010 revision.
      ]. These data were used together with the ratio of urban to rural prevalence to disaggregate data with ‘unspecified’ setting into urban and rural prevalence. We assumed a total sample size for each gender-specific age group. The proportion of urban population for that country was used to stratify as follows:
      Nurban=NTotal×TurbanNrural=NTotalNurban


      where NTotal is the number of people examined by gender and age-group, Turban is the proportion of the population for a country living in urban areas, Trural is the proportion of the population for a country living in rural areas, Nurban is the urban population and Nrural is the rural population.
      Corresponding cases were then calculated for this rural population according to the equation:
      Rural cases=Total cases1+(Purban/Prural)×(Nurban/Nrural)


      where total cases is the number of people with the disease by gender and age-group (Purban/Prural) is the ratio of urban to rural prevalence determined from the study or from the appropriate data region (Table 3).
      Prevalence was then calculated for the rural population according to the equation:
      Rural prevalence=Rural casesNrural


      Using this approach, we stratified the prevalence by urban and rural setting for those studies that did not provide stratified results.

      9. Undiagnosed diabetes

      Some selected sources only reported known diabetes. These studies were adjusted to account for undiagnosed diabetes using a calculated median proportion based on sources scoring above 0.3 with available data on undiagnosed diabetes. The same country groupings were used for the calculation of median undiagnosed diabetes proportions as for urban/rural ratios (Table 4).
      Table 4Median proportion of undiagnosed diabetes for countries grouped by geography and income group.
      Data regionCountries providing dataUndiagnosed diabetes (%)
      AFR-LICBenin, Comoros, Guinea, Kenya, Mauritania, Mozambique, Niger, United Republic of Tanzania77.94
      AFR-MICAngola, Reunion, Seychelles, South Africa80.00
      EUR-HICCroatia, Cyprus, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Malta, Portugal, Slovakia, Spain, Sweden, United Kingdom36.59
      EUR-LICUzbekistan29.34
      EUR-MICAlbania, Bulgaria, Poland, Turkey35.88
      MENA-HICOman, Saudi Arabia, United Arab Emirates40.70
      MENA-MICAlgeria, Egypt, Iraq, Jordan, Occupied Palestinian Territory, Pakistan, Sudan, Tunisia61.57
      NAC-HICBarbados, United States of America, US Virgin Islands27.71
      NAC-LICHaiti29.40
      NAC-MICBelize, Guadeloupe, Mexico41.22
      SACA-MICBolivia, Brazil, Chile, Guatemala, Honduras, Nicaragua44.67
      SEA-LICNepal48.06
      SEA-MICBhutan, India, Mauritius, Sri Lanka,51.08
      WP-HICAustralia, Hong Kong China, Republic of Korea, Taiwan46.71
      WP-LICCambodia63.04
      WP-MICChina, Fiji, Indonesia, Malaysia, Mongolia, Philippines, Samoa, Thailand, Tonga56.88

      10. Age adjustments

      It is very common in sources for the upper age limit of the oldest age group to be open-ended. For example, the last age group presented may be 65+. In order to account for this and still provide adequate information to conduct the modelling, we applied the following rules:
      • 1.
        If the upper age limit was unspecified and age groups were in 10 year ranges, then the upper age limit = last age + 9 (e.g., 74 if the last age group was 65+)
      • 2.
        Otherwise, the upper age limit was set at 79
      No age groups with the youngest members being over 79 years were included, but persons over 80 years were included if part of an age group (e.g., 75–84 years).

      11. Additional adjustments

      Based on evidence from Van der Sande et al's study in The Gambia, estimates for sources where the primary diagnostic method was glycosuria were doubled [
      • van der Sande M.A.
      • Walraven G.E.
      • Bailey R.
      • Rowley J.T.
      • Banya W.A.
      • Nyan O.A.
      • et al.
      Is there a role for glycosuria testing in sub-Saharan Africa?.
      ].
      For some sources, point adjustments were made by combining age groups to reduce variability. If the sample size for an individual stratum (combination of age group, gender, and setting) was less than 50 and the distribution of the prevalence was biologically implausible we combined the data from the age group nearest the one with the low sample size to create a larger sized age group with a bigger sample size.

      12. Logistic regression

      Once all adjustments were made to the study data, logistic regression was used to smooth the curves using age (as midpoint of each age group), and the quadratic of age (age2) as independent variables. The quadratic term (age2) was included to allow the possibility of a drop in the prevalence for the oldest age groups. The final binary logistic formula was as follows:
      Y=θ+βXage+βXage2


      where Y is the prevalence as the number of cases and non-cases, Xage is the midpoint of the age group, Xage2 is the quadratic term of age and θ is the intercept.
      A minimum of three age groups was required for input into the model. Studies without at least three age groups were excluded. Using the “predict” function in R we calculated the age-specific prevalence for each study using five-year age-groups (20–24, 25–29, 30–34, …, 75–79).

      13. Imputation

      Evidence from studies that had data for upper age groups (60+) and sufficient detail (typically 5-year age groups) revealed a general trend showing that the prevalence in the oldest age group was close to 90% of the prevalence in the previous age group (Fig. 3). Therefore, where we found large and implausible differences between the modelled distribution and the original data, we imputed a data point for an age group 10 years older than the last age group available, at a value that was 90% that of the last age group. We applied this imputation where the maximum value from the modelled prevalence was 1.2 times greater than that of the maximum value of the raw data and this difference could not be corrected by combining age groups. This single rule was found to significantly improve all cases where the modelled distribution deviated substantially from the raw data. There were 73 single point imputations. Examples of the effects of applying this rule are shown in Appendix 3.
      Figure thumbnail gr3
      Fig. 3Prevalence by age for three data sources on diabetes.

      14. Applying population estimates

      The total numbers of persons with diabetes and IGT for each country were then calculated by applying the calculated age specific prevalence rates to the demographic data from the United Nations Population Prospects 2010 Revision [
      • United Nations
      World population prospects: the 2010 revision.
      ].
      For countries without age and gender distribution descriptions, i.e., those with populations of less than 100,000 for the year 2011, for which only total population data are provided, the average population distribution of the aggregate region containing that country or territory from the UN population estimates for 2011 and 2030 was applied to the total population for the corresponding year. For Taiwan, the UN regional population distribution was applied to the total population as indicated in the US Census Bureau International Database []. The countries/territories without complete UN population data that are included were: Andorra, Anguilla, Antigua and Barbuda, Aruba, Bermuda, British Virgin Islands, Cayman Islands, Dominica, Grenada, Cook Islands, Kiribati, Liechtenstein, Marshall Islands, Monaco, Nauru, Niue, Palau, Saint Kitts and Nevis, San Marino, Seychelles, Taiwan, Tokelau, Tuvalu.

      15. Age standardisation

      In addition to calculating the national rates, age-standardised rates were calculated using the world population based on the distribution provided by the World Health Organization [
      • Ahmad O.B.
      • Boschi-Pinto C.
      • Lopez A.D.
      • Murray C.L.
      • Lozano R.
      • Inoue M.
      Age standardization of rates: a new WHO standard. GPE discussion paper series 31.
      ] and using the distribution determined from the population estimates used in this study [
      • United Nations
      World population prospects: the 2010 revision.
      ].

      16. Discussion

      The reliability and accuracy of diabetes prevalence estimates are highly dependent on the data sources used in the modelling process. The revised methodology for generating the International Diabetes Federation estimates of diabetes prevalence described in this paper draw on expert opinion in a systematic, explicit, and adaptable way to select data sources. The new methods preferentially select data sources that are nationally representative over those that are subnational or community-based, and pooled sources are used only when there is a lack of sufficient data of high quality. In addition, while we did not explicitly exclude studies using fasting blood glucose, self-reported diabetes, or medical record-based data as the identification method, we did preferentially select studies conducted using the 2-h oral glucose tolerance test based on the judgment of the expert panel. Studies that did not score above the highest threshold, but were still considered of sufficient quality for inclusion were then weighted based on their score so as to minimize the impact of potential bias in the design of the studies that would affect their ability to provide representative data. The new methods are a departure from the previous IDF methodology that did not apply a weighting to included data sources and was therefore more susceptible to bias.
      Experts from all IDF regions contributed to the identification of sources and analysis of the results. They provided an important local and regional context with which the assumptions used could be tested. The methods also take a more systematic approach to identifying countries to provide source estimates for those countries missing data. Rather than base the selection of source data on expert opinion alone, we matched countries based on characteristics that have been associated with the prevalence of diabetes including economic development, geographic proximity, and ethnicity. The approach is transparent and reproducible. All the resulting estimates were validated against nationally representative studies available to examine the plausibility of the results.
      The strengths of these methods lie in their simplicity, adaptability, and reproducibility. They take a systematic approach that can be modified as new information becomes available. The estimates will be updated annually to include new data to ensure that precise estimates are available. However, the estimates are limited by the availability of good-quality data sources. A number of studies were found to be lacking in essential age-specific estimates or did not provide enough information in the description of the methods to characterize the study. Where this information was lacking, every effort was made to contact investigators for more detail. In particular, a number of studies were not indexed in PubMed and it was necessary to search for manuscripts from journals, through libraries, or from the investigators themselves once an abstract was identified. Obtaining manuscripts and sufficient data was particularly difficult for Spanish-language studies and those conducted in South and Central America (even though the research team included a native Spanish-speaker).
      This method for estimating diabetes prevalence takes into account changes in age, sex and urbanization but not explicitly for changes in lifestyle and obesity, although urbanization may be a proxy for these changes in some countries. Thus, the prevalence projections in particular are likely to be an underestimate. However, the provision of easily understood and reproducible methods will contribute to the debate on how best to estimate the prevalence of diabetes and challenge the scientific community to examine estimates that are presented. This methodology provides us not only with the opportunity for updating the estimates for diabetes prevalence annually, but also an approach that can be adapted and improved as new information becomes available. The diabetes prevalence estimates that were generated using this methodology are available in the 5th Edition of the IDF Diabetes Atlas and described in detail in the publication by Whiting et al. [
      • Whiting D.
      • Guariguata L.
      • Weil C.
      • Shaw J.
      IDF diabetes atlas: global estimates of the prevalence of diabetes for 2011 and 2030.
      ]

      Funding

      The IDF Diabetes Atlas was supported by the following sponsors: Lilly Diabetes , Merck and Co, Inc. , Novo Nordisk A/S supported through an unrestricted grant by the Novo Nordisk Changing Diabetes initiative, and Sanofi .

      Conflicts of interest

      The authors have no conflicts of interest to report.

      Appendix 1. Country groupings for estimating diabetes prevalence

      Tabled 1
      Data region
      Abbreviations: Africa (AFR), Europe (EUR), Middle East and North Africa (MENA), North America and the Caribbean (NAC), South and Central America (SACA), South-East Asia (SEA), Western Pacific (WP), high-income country (HIC), upper middle-income country (UMIC), lower-middle income country (LMIC), low-income country (LIC).
      Countries with primary dataCountries without primary data
      AFR-LIC-EastKenya, United Republic of TanzaniaBurundi, Rwanda, Uganda
      AFR-LIC-NorthMali, Mauritania, NigerChad, Eritrea, Ethiopia, Somalia, Western Sahara
      AFR-LIC-SouthMalawi, Mozambique, ZimbabweMadagascar, Zambia
      AFR-LIC-WestBenin, Comoros, Gambia, Ghana, GuineaBurkina Faso, Central African Republic, Democratic Republic of Congo, Guinea-Bissau, Liberia, Senegal, Sierra Leone, Togo
      AFR-LMICAngola, CameroonCape Verde, Côte d’Ivoire, Djibouti, Equatorial Guinea, Lesotho, Nigeria, Republic of Congo, Sao Tome and Principe, Swaziland
      AFR-UMICBotswana, South AfricaGabon, Namibia, South Africa
      AFR-UMIC-EastReunion, Seychelles
      Reunion and Seychelles were grouped separately from other countries in Africa because they were found to have particularly high prevalence of diabetes and were unique in that they were small islands.
      EUR-HIC-CentralHungary, Slovenia, Slovakia
      EUR-HIC-NorthDenmark, Estonia, Finland, Iceland, Norway, Sweden
      EUR-HIC-SouthCroatia, Cyprus, Greece, Israel, Italy, Malta, Portugal, Spain
      EUR-HIC-WestAustria, Belgium, France, Germany, Luxembourg, Netherlands, Switzerland, United KingdomAndorra, Channel Islands, Czech Republic, Ireland, Liechtenstein, Monaco, San Marino
      EUR-LICUzbekistanKyrgyzstan, Tajikistan, Uzbekistan
      EUR-LMICAlbaniaAzerbaijan, Georgia, Moldova, Turkmenistan, Ukraine
      EUR-UMICBulgaria, Poland, Russian Federation, TurkeyBelarus, Bosnia and Herzegovina, Kazakhstan, Latvia, Lithuania, Macedonia, Montenegro, Romania, Serbia
      MENA-HICOman, Saudi Arabia, United Arab EmiratesBahrain, Kuwait, Qatar
      MENA-LMIC-EastPakistanAfghanistan
      Afghanistan was grouped with Pakistan even though it is low-income as there are no other low-income countries in the MENA region and Pakistan is in close proximity.
      , Armenia
      MENA-LMIC-CentralIraq, Jordan, Occupied Palestinian TerritoriesSyrian Arab Republic, Yemen
      MENA-LMIC-WestEgypt, Morocco, Sudan, Tunisia
      MENA-UMICAlgeria, Islamic Republic of Iran, LebanonLibyan Arab Jamahiriya
      NAC-HIC-NorthCanada, United States Of America
      NAC-HIC-SouthBarbados, Bermuda, US Virgin IslandsAntigua and Barbuda, Aruba, Bahamas, Cayman Islands, Martinique, Netherland Antilles, Trinidad and Tobago
      NAC-LIC-SouthHaiti
      NAC-LMIC-CentralBelizeGuyana
      NAC-UMIC-NorthMexico
      NAC-UMIC-SouthGuadeloupe, JamaicaAnguilla, British Virgin Islands, Dominica, Grenada, St. Kitts and Nevis, St. Lucia, St. Vincent and the Grenadines, Suriname
      SACA-HIC-NorthPuerto Rico
      SACA-LMIC-NorthGuatemala, Honduras, NicaraguaEl Salvador
      SACA-LMIC-SouthBoliviaEcuador, Paraguay
      SACA-UMIC-CentralBrazilColombia, French Guiana, Venezuela
      SACA-UMIC-NorthCosta Rica, Dominican RepublicCuba, Panama
      SACA-UMIC-SouthArgentina, ChilePeru, Uruguay
      SEA-LICBangladesh, Nepal
      SEA-LMICBhutan, India, Sri LankaMaldives
      SEA-UMICMauritius
      WP-HIC-CentralSingaporeBrunei Darussalam
      WP-HIC-EastAustralia
      WP-HIC-NorthHong Kong, Republic of Korea, TaiwanJapan, Macau
      WP-HIC-SouthNew ZealandCook Islands, French Polynesia, Guam, New Caledonia
      WP-LIC-CentralCambodiaLao People's Democratic Republic, Viet Nam
      WP-LIC-NorthMyanmar
      Democratic People's Republic of Korea was assigned to a group with Myanmar which had available data and separated from Viet Nam and Cambodia because of the similarities in the political systems of those countries despite being geographically disparate.
      People's Democratic Republic of Korea
      WP-LMIC-CentralIndonesia, Philippines, ThailandPapua New Guinea, Timor L’Este
      WP-LMIC-NorthChina, Mongolia
      WP-LMIC-SouthKiribati, Nauru, Solomon Islands, TongaFederated States of Micronesia, Marshall Islands, Tuvalu, Vanuatu
      WP-UMIC-CentralMalaysia
      WP-UMIC-SouthFiji, SamoaNiue, Palau, Tokelau
      a Democratic People's Republic of Korea was assigned to a group with Myanmar which had available data and separated from Viet Nam and Cambodia because of the similarities in the political systems of those countries despite being geographically disparate.
      b Afghanistan was grouped with Pakistan even though it is low-income as there are no other low-income countries in the MENA region and Pakistan is in close proximity.
      c Reunion and Seychelles were grouped separately from other countries in Africa because they were found to have particularly high prevalence of diabetes and were unique in that they were small islands.
      d Abbreviations: Africa (AFR), Europe (EUR), Middle East and North Africa (MENA), North America and the Caribbean (NAC), South and Central America (SACA), South-East Asia (SEA), Western Pacific (WP), high-income country (HIC), upper middle-income country (UMIC), lower-middle income country (LMIC), low-income country (LIC).

      Appendix 2. Country groupings for estimating impaired glucose tolerance prevalence

      Tabled 1
      Data regionCountries with primary dataCountries without primary data
      AFR-LICKenya, United Republic of Tanzania, ZimbabweBenin, Burkina Faso, Burundi, Central African Republic, Chad, Comoros, Democratic Republic of Congo, Eritrea, Ethiopia, Gambia, Ghana, Guinea, Guinea-Bissau, Liberia, Madagascar, Malawi, Mali, Mauritania, Mozambique, Niger, Rwanda, Senegal, Sierra Leone, Somalia, Togo, Uganda, Western Sahara, Zambia
      AFR-MICAngola, Seychelles, South AfricaBotswana, Cameroon, Cape Verde, Cote d’Ivoire, Djibouti, Equatorial Guinea, Gabon, Lesotho, Namibia, Nigeria, Republic of Congo, Reunion, Sao Tome and Principe, Swaziland
      EUR-HICCyprus, Denmark, Estonia, Finland, Germany, Malta, Portugal, Spain, SwedenAndorra, Austria, Belgium, Channel Islands, Croatia, Czech Republic, France, Greece, Hungary, Iceland, Ireland, Israel, Italy, Liechtenstein, Luxembourg, Monaco, Netherlands, Norway, San Marino, Slovakia, Slovenia, Switzerland, United Kingdom
      EUR-LICUzbekistanKyrgyzstan, Tajikistan
      EUR-MICBulgaria, Poland, TurkeyAlbania, Azerbaijan, Belarus, Bosnia and Herzegovina, Georgia, Kazakhstan, Latvia, Lithuania, Macedonia, Moldova, Montenegro, Romania, Russian Federation, Serbia, Turkmenistan, Ukraine
      MENA-HICOman, Saudi Arabia, United Arab EmiratesBahrain, Kuwait, Qatar
      MENA-MICAlgeria, Islamic Republic of Iran, Jordan, Occupied Palestinian Territory, Pakistan, SudanAfghanistan, Armenia, Egypt, Iraq, Lebanon, Libyan Arab Jamahiriya, Morocco, Syrian Arab Republic, Tunisia, Yemen
      NAC-HICUnited States Of AmericaAnguilla, Antigua and Barbuda, Aruba, Bahamas, Barbados, Belize, Bermuda, British Virgin Islands, Canada, Cayman Islands, Dominica, Grenada, Guadeloupe, Guyana, Jamaica, Martinique, Mexico, Netherlands Antilles, St. Kitts and Nevis, St. Lucia, St. Vincent and the Grenadines, Suriname, Trinidad and Tobago, US Virgin Islands
      NAC-LICHaiti
      SACA-MICBolivia, Brazil, NicaraguaArgentina, Chile, Colombia, Costa Rica, Cuba, Dominican Republic, Ecuador, El Salvador, French Guiana, Guatemala, Honduras, Panama, Paraguay, Peru, Puerto Rico, Uruguay, Venezuela
      SEA-LICNepalBangladesh
      SEA-MICBhutan, India, Mauritius, Sri LankaMaldives
      WP-HICAustralia, Hong Kong China, SingaporeBrunei Darussalam, Cook Islands, French Polynesia, Guam, Japan, Macau China, New Caledonia, New Zealand, Republic of Korea, Taiwan
      WP-LICCambodia, MyanmarDemocratic People's Republic of Korea, Lao People's Democratic Republic, Viet Nam
      WP-MICChina, Fiji, Indonesia, Malaysia, Mongolia, Philippines, Samoa, TongaFederated States of Micronesia, Kiribati, Marshall Islands, Nauru, Niue, Palau, Papua New Guinea, Solomon Islands, Thailand, Timor l’Este, Tokelau, Tuvalu, Vanuatu

      Appendix 3. The effects of applying the imputation rule on the distribution of diabetes prevalence
      These figures were produced as part of the data management process. Please note that study numbers refer to internal database numbers, not references in this publication and ‘n’ values are not sample sizes.
      1These figures were produced as part of the data management process. Please note that study numbers refer to internal database numbers, not references in this publication and ‘n’ values are not sample sizes.

      Figure thumbnail fx1

      References

        • Shaw J.E.
        • Sicree R.A.
        • Zimmet P.Z.
        Global estimates of the prevalence of diabetes for 2010 and 2030.
        Diabetes Res Clin Pract. 2010; 87: 4-14
        • Danaie G.
        • Finucane M.M.
        • Lu Y.
        • Singh G.M.
        • Cowan M.J.
        • Paciorek C.J.
        • et al.
        National, regional, and global trends in fasting plasma glucose and diabetes prevalence since 1980: systematic analysis of health examination surveys and epidemiological studies with 370 country-years and 2·7 million participants.
        Lancet. 2011; 378: 31-40
        • International Diabetes Federation
        The diabetes atlas.
        3rd ed. International Diabetes Federation, Brussels2006
        • International Diabetes Federation
        The IDF diabetes atlas.
        4th ed. International Diabetes Federation, Brussels2009
        • R Development Core Team
        R: a language and environment for statistical computing.
        2010
        • Saaty T.L.
        Analytic hierarchy process.
        Encyclopedia of biostatistics. 2nd ed. John Wiley & Sons, Ltd., 2005
      1. The World Bank. Country and lending groups, http://data.worldbank.org/about/country-classifications/countr-and-lending-groups [accessed 23.08.11].

      2. Central Intelligence Agency. The world factbook, https://www.cia.gov/library/publications/the-world-factbook.

        • United Nations
        World population prospects: the 2010 revision.
        United Nations, Geneva2011
        • van der Sande M.A.
        • Walraven G.E.
        • Bailey R.
        • Rowley J.T.
        • Banya W.A.
        • Nyan O.A.
        • et al.
        Is there a role for glycosuria testing in sub-Saharan Africa?.
        Trop Med Int Health. 1999; 4: 506-513
      3. United States Census Bureau. International data base, http://www.census.gov/population/international/data/idb/informationGateway.php.

        • Ahmad O.B.
        • Boschi-Pinto C.
        • Lopez A.D.
        • Murray C.L.
        • Lozano R.
        • Inoue M.
        Age standardization of rates: a new WHO standard. GPE discussion paper series 31.
        World Health Organization, Geneva, Switzerland2001
        • Whiting D.
        • Guariguata L.
        • Weil C.
        • Shaw J.
        IDF diabetes atlas: global estimates of the prevalence of diabetes for 2011 and 2030.
        Diabetes Res Clin Pract. 2011;