UNIVERSITY OF WASHINGTON Global Burden of Diseases, Injuries and Risk Factors: methodological aspects and trends 1 st International Conference on the Burden.

UNIVERSITY OF WASHINGTON

Global Burden of Diseases, Injuries

and Risk Factors: methodological

aspects and trends

1st International Conference on the Burden of Diseases Studies in Brazil

November 18th, 2009

Rafael Lozano, MD. MSc.

Outline

• What is “GBD”

• GBD in the last 20 years

• GBD 2005 is not only an update of the reference year

• Where is GBD making contributions to global health

2

7

Burden of Disease

“…the gap between a population’s health status and some reference standard…”

Murray 1996

Outline



• GBD 2005 is not only an update


8

Brief History From Economist to Public Health practitioners

• What are the leading causes of death worldwide?

• Women, men, children, adults, regions

• Which are the health priorities?

• Due to the magnitude of the problem, Is it enough to use mortality as measurement?

• How can we select the best health interventions?

• How we can know if our investments on health system are producing positive effects on the population health?

• How can we improve the allocation of health resources (financial)?

•

9

• Tension into UN agencies: lack of leadership from WHO

• WHO “Health for all” 1987

• WB “Investing in health, 1993”

• Easy Questions, hard to answer: If HIV/AIDS is the first cause of death, what is the next ?

In May of 1993, a group of Mexicans went to Boston to learn the methods of the BOD

Dr Chris Murray was contracted as advisor of the study “The Health and the Economy” conducted by Dr Julio Frenk

1. Public health statistics, which were partial and fragmented

2. Estimates for numbers of people that die or impacted by disease, which were in some cases exaggerated beyond plausible limits or missing estimates entirely

3. Traditional health statistics did not allow policy-makers to compare relative cost-effectiveness of different interventions across diseases

4. Many reports were influenced by politics which diluted truth and prevented effective intervention

10

Global Context before GBD

- First GBD commissioned by World Bank, published :

- 1993 WDR (WB) Investing in Health

- 1994 WHO, Setting Health Priorities

- 1996: GBD and GHS

- Produced estimates for 1990 and projections to 2020

- Led by Christopher Murray and Alan Lopez

- Disentangled epidemiology from advocacy in order to produce objective, plausible estimates

- Measured burden of mortality and non-fatal conditions in a metric that could be compared across diseases (DALY), ages, and regions

11

GBD 1990

Leading Causes of Death and DALYs 1990

%

Ischemic heart disease 12.4

Cerebrovascular disease 8.7

Lower respiratory infec 8.5

Diarrhoeal diseases 5.8

Perinatal conditions 4.8

C.O.P.D. 4.4

Tuberculosis 3.9

Measles 2.1

Road traffic accidents 2.0

Lung Cancer 1.9

Source: Murray and Lopez, 1996

%

Lower Resp infec. 8.2

Diarrhoeal diseases 7.2

Perinatal conditions 6.7

Depression 3.7

Ischemic Heart Dis 3.4

Cerebrovascular Dis 2.8

Tuberculosis 2.8

Measles 2.6

Road traffic accidents 2.5

Congenital anomalies 2.4

Deaths DALYs

• Since 1998, WHO has produced Since 1998, WHO has produced annually up to dates of the GBD, annually up to dates of the GBD, publishing them in the Statistical publishing them in the Statistical Annexes of the WHR, as well in the Annexes of the WHR, as well in the web site web site www.who.int

• From the 8 original regions WHO From the 8 original regions WHO increased them to 14increased them to 14

• Updates of epidemiological estimates Updates of epidemiological estimates of TB, Malaria, HIV/AIDS, of TB, Malaria, HIV/AIDS, Neuropsychiatric diseases, were Neuropsychiatric diseases, were produced for the GBD 2000produced for the GBD 2000

• In order to increase the theory and In order to increase the theory and methods of summary measures, WHO methods of summary measures, WHO published a book in 2002published a book in 2002

UPDATES of BOD 1990

http://www.who.int/

The second round of estimates of the Attributable The second round of estimates of the Attributable Burden due to some risk factors was initiated in 2001. Burden due to some risk factors was initiated in 2001. That study expanded the number of risk factors from 10 That study expanded the number of risk factors from 10 to 29. to 29.

The results were published in WHR 2002, and the detail The results were published in WHR 2002, and the detail literature review and methods in 2004 literature review and methods in 2004

Comparative Risk Assessment

0.0% 1.0% 2.0% 3.0% 4.0% 5.0% 6.0% 7.0% 8.0% 9.0% 10.0%

Underweight

Unsafe sex

High blood pressure

Tobacco

Alcohol

Unsafe water, sanitation, and hygiene

High cholesterol

Indoor smoke from solid fuels

Iron deficiency

High BMI

Zinc deficiency

Low fruit and vegetable intake

Vitamin A deficiency

Physical inactivity

Occupational risk factors for injury

Lead exposure

Illicit drugs

Unsafe health care injections

Lack of contraception

Childhood sexual abuse

Attributable DALY (% of global DALY - Total 1.46 billion)

High-mortality developing

Lower-mortality developing

Developed

New estimates of GBD for New estimates of GBD for 2001, based on the WHO 2001, based on the WHO revisions and more deatil revisions and more deatil sensitivity analysissensitivity analysis

Includes more Includes more documentation of diseases documentation of diseases and risk factor estimatesand risk factor estimates

Disease Control Priorities II (2006)

17

Chile, Costa Rica, Peru, EcuadorTurkeyIranTanzania, MozambiqueMorocco, Tunisia, etc., etc.

Outline



• GBD 2005 is not only an update of the year of reference


18

Reasons for executing a new round

19

Demand for burden data from governments, funders, policy makers

Only piecemeal revisions of epidemiology for conditions since 1996

No comprehensive revision of disability weights since 1996 (most criticized part of study)

No consistent time trend available (methods for ‘00, ‘01, ‘02 not comparable to ‘90)

Methods advances for mortality measurement, cause of death attribution, modelling missing data, DW estimation and data collection techniques

Need for new tools, approaches to share results of GBD study with diverse audiences

Involve collaboration of many more people

Why a new GBD study

New Round

- Funded by the Bill and Melinda Gates Foundation

- Started September 2007, Ending November 2010

Objectives

- Produce specific DALY, YLL, and YLD estimates for over 220+ diseases/injuries and 40+ risk factors by age range, sex, and for 21 regions for the years 1990 and 2005.

- Create simplified analytical tools to facilitate national burden estimates and policy use

20

GBD 2005

21

Who are the key participants

Organization

22

Core TeamExternal

Advisory Board

Tools and Curricula Development Sub-Team

COD Sub-TeamR Lozano, M

Naghavi

CRA Sub-TeamMajid Ezzati

DW Sub-TeamJosh Salomon and

Colin Mathers

Mortality Sub-TeamAlan Lopez and

Chris Murray

YLD Sub-TeamRafael Lozano

and Colin Mathers

Cluster A CVD, COPD, Cancer

Majid EzzatiHarvard University

Cluster B Child/Maternal

Bob BlackJohns Hopkins

University

Cluster C Injuries and Mental

HealthTheo Vos

University of Queensland

Cluster D Communicable

DiseasesNeff Walker

Johns Hopkins University

Cluster ENoncommunicable

DiseasesCatherine MichaudHarvard University

Who are the key participants

How is the work done at IHME contributing to the study

23

YLL

Epidemiological Estimates

DISMOD

YLD DALY

Mortality

COD

Disability Weights

Mortality Envelopes

• Deaths by age, sex and GBD region

• Contains the number of deaths for all causes

• Death is attributed to one cause

• Estimated using all-cause mortality data

• Child mortality (0-4) estimated separately

24

Envelopes

Models

Assessing and adjusting for

incompleteness

Adjusting for biases

Vital Registers*

Surveys

Estimation Process:

Synthesis of Child Mortality: Examples

Estimation Process: Adults

Adjusting for Incompleteness in VR

• Death Distribution Methods (DDM)

• Demographers long-used tools for assessing level of completeness in death registration

– 3 families

– Many variants

Adjusting For Biases

• Survey Data: Sibling Survival Model

Models: Predicting adult mortality

• Leverage relationship between adult and child mortality

• Build model, predict logit (45q15) for Males and Females separately (HIV prevalence, TFR, Country or Regional FE, Adjustment for post-Soviet collapse

26

Synthesis of Adult Mortality: Examples

27

0.2

.4.6

1950 1960 1970 1980 1990 2000 20101950 1960 1970 1980 1990 2000 2010

female male

Complete VR

DDM-adjusted VR

extrapolated completeness

sib histories

Prediction model country FE

Prediction model region FE

outlier

xxxx = DSS site

45q

15

Year

Graphs by sex

Russian Federation RUS

Objectives• To produce estimates of

selected causes of death by country, age and sex.

• To produce estimates of causes of death, based in GBD cause list for 21 GBD regions by age and sex, 1990 and 2005.

• To produce friendly tools to aid cause of death estimates:

• Mapping ICD across time and populations

• Redistribution of garbage codes

• Modeling causes of death for countries without VR

28

24 to 36 causes for 200 countries and territories (IHME)

268 causes for 21 GBD new regions

CODMOD

29

Mapping GBD Cause List with ICD Revisions and Other Tabulated List

BTL1 2 3 4 5 Tab B 6,7 Tab A 89 tab

9 VA 10Tab

10

GBD 2005 Cause List (268 )

GBD 1990 Cause List (100)

CODMOD level 2 (24)

CODMOD level B (36)

ICD and other formats

Availability of COD data

ICD 07A 815 - 100%

ICD 07B 110 - 100%

ICD 08 detail/ICD 08A 70 - 100%

ICD 08A 776 - 100%

ICD 08B 35 - 100%

ICD 09 detail/BTL 664 - 100%

ICD 09 detail (9M) 13 - 100%

ICD 09 BTL 1082

09A 47%

09B 28%

09N 23%

09C (China tabulation) 1%

09I (India tabulation) 1%

ICD 10 702

103 (3-digit) 14%

104 (4-digit) 80%

10M (mixed) 6%

ICD 10 Tabulated 70

101 (WHO tabulation) 73%

10I (India tabulation) 4%

10Ir (Iran tabulation) 10%

10S (Syria tabulation) 4%

10C (China tabulation) 9%

More than 4550 country-years

Distribution of Garbage Codes by Type and Region

0.0

5.0

10.0

15.0

20.0

25.0

30.0

35.0

SSA Asia LA Europe C&E ALL Europe W Caribbean N.America Australasia

SpecialsImmediateSequelaeIntermediateI&D UNSCancerIll Def

% o

f G

C

• ~20% total deaths from VR are GCs

•10 causes accumulate 75%

• Intermediate causes are the most important Garbage Codes

Causes ICD 10 %

Ill-defined R00-R99 26.0

Heart Failure I50 18.0

Renal Failure N18 6.4

Atherosclerosis I70 6.0Malignant neoplasm without specification of site C80 4.8

Septicaemia A41 4.2

Essential (primary) hypertension I10 3.0

Exposure to Unspecified factor X59 2.7

Pulmonary embolism I26 2.2

Respiratory Failure J96 2.0

Percent of deaths with garbage codesSelect Countries of the Americas, circa 2005

Mortalidad por códigos basura en países de la región, 1979-2007

The Problem is how to predict CoD for 100 countries without VR data

• Data quality• Sparseness : Approximately 75% of total country-years missing

• Compositional bias

• Both sampling and non-sampling error

• Sometimes multiple (discrepant) observations per country-year

• Poor covariates• No global time-series available for many important covariates

• The covariates we do have fail to explain much of the variation in the data

• Need predictions• Not only do we need to fit the data we have, but we need to forecast

forward (and backwards in many cases)

11

How to maximize our use of all the data available?

Estimating Cause of Death Strategy

35

METHOD APPLIED

TYPE OF COUNTRY’S DATA

VR

DATABASE

CODMOD

Country Fixed Effect

CODMOD

Region Fixed Effect

A. Complete VR When we have countries with data in 2005

When we have countries with data before 2005, we project for 2005

B. Adults Complete

Children Incomplete (43)

For adults use VR for 2005

Estimates for 0-4 assuming Completeness=1.0, GC = 0.0

C. Adults/Children

Incomplete

Estimates for all ages, assuming

C=1.0, GC = 0.0

D. VA

Estimates assuming

C=1.0, GC = 0.0

Estimates assuming

C=1.0, GC = 0.0

E. No Data Available

Estimates assuming

C=1.0, GC = 0.0

Some results

26

05

01

001

502

00D

eath

Rat

e (p

er

100

,000

)

1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010Year

Observed Data Simple Mixed Effects ModelWeighted Regression Model

United States of AmericaCommunicable Diseases, Males 45-49

50

100

150

200

250

Dea

th R

ate

(pe

r 1

00,0

00)

1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010Year


Russian FederationCommunicable Diseases, Males 45-49

05

001

000

Dea

th R

ate

(pe

r 1

00,0

00)

1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010Year


MexicoCommunicable Diseases, Males 45-49

01

002

003

004

00D

eath

Rat

e (p

er

100

,000

)

1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010Year


ParaguayCommunicable Diseases, Males 45-49

37

Mortality

COD

YLL


DisMod Disability Weights

YLD DALY


38

Expert Groups and Epidemiological Estimates

Neonatal infections

Prenatal risk factors

Intrapartum risk factors

Postnatal risk factors

Parameters for disease model Incidence of the condition Case-fatality rate Case-complication rate (risk of sequelae) & distribution of severity of sequelae Complication-fatality rate

1

1

SepsisPneumonia

Meningitis

Dead

Full recovery

2

234

Exclude? Sequalae*Mild, moderate, severeSingle or multi-domain

4

3

2

2

2

GBD schematic for neonatal infections

40

Did not meet criteria(n= 532)

214 papers

enteredStudies without live

births (n=115)Studies with live

births (n=99)

Clinical sepsis incidence reported(n=24)

Culture proven sepsis incidence Reported (n=48)

Sepsis Case fatality Rate Reported (n=23)

Excluded studies

DATABASESPubMed, Embase, Web of Science, Popline,

WHO regional databases + Reference lists and Key Review articles

SEARCH TERMSInfection OR Sepsis OR Pneumonia OR Meningitis +

Variables of Interest (eg. Incidence etc.)Limits: Publication Date from 1990 to 2008

Neonatal infections incidence – Searches and data selection

Pending Translation (n= 23)

Unavailable {now coming from Boston] (n= 64)

Clinical sepsis incidence Reported

(n=1)

Culture proven sepsis

incidence Reported (n=8)

Sepsis Case fatality Rate Reported (n=19)

Studies remaining after screening title or abstract (n=833)

Total search results (6518)

Data availability for neonatal infections according to countries grouped by level of NMR

Countries according to level of NMR

NMR<5

Level 1

NMR 5 - 14.99

Level 2

NMR > 15

Levels 3 (15-29), 4 (30- 45) and 5>45)

No of countries 49 50 93

Neonatal sepsis data

68 studies with live births

N = 9,726,840

5 studies with live births

N = 310,082

24 studies with live births N = 240,004

Neonatal sepsis incidence (per 1000 births) and outcomes

Clinical:

3.4 - 136/1000 live births

Culture proven

0.6 - 18/1000

CFR (7.1 – 30.3%)

Clinical

8.4 – 24.4/1000 live births

Culture proven

2.3 - 10/1000

CFR (6.7 - 26.5%)

Clinical:

21-170/1000 live births

Culture proven

5.5 - 24.8/1000

CFR (9.6 - 67%)

Higher NMR settings have higher incidence of sepsis and higher case fatality

Lots of work for these groups (systematic review example)

42

Expert Groups and Epidemiological Estimates

Full-text evaluation (Articles and reports)

4626

Citations identified (Titles and/ or abstracts)

64 586 Excluded

59 960

I ncluded

2580 Excluded

2046

Reasons for exclusion 92% – no

relevant data 6% – sample

size<200 2% – other

reasons

Morbidity outcomes

3215 data sets

Mortality outcome

1143 data sets

Reasons for exclusion57% – no relevant data15% – sample size<20011% – no dates reported17% – other reasons

Maternal Conditions Expert Group

Alcohol-attributable disease and injury 2002 (green mainly protective)

Chronic disease:Cancer: Mouth & oropharyngeal cancer, esophageal cancer, liver cancer, female breast cancerNeuropsychiatric diseases: Alcohol use disorders, unipolar major depression, primary epilepsyDiabetesCardiovascular diseases: Hypertensive diseases, ischemic heart disease, ischemic stroke, hemorrhagic strokeGastrointestinal diseases: Liver cirrhosisConditions arising during perinatal period: Low birth weight, FAS

Injury:Unintentional injury: Motor vehicle accidents, drownings, falls, poisonings, other unintentional injuriesIntentional injury: Self-inflicted injuries, homicide, other intentional injuries

New developments with respect to causality: inclusion of alcohol-attributable disease categories

• Colorectal cancer included (IARC monograph meeting; Baan et al., 2007)

• Tuberculosis/pneumonia incidence and worsening the disease course included (see next slides)

• HIV discussed but not included (not enough evidence for causality for incidence); enough evidence for alcohol worsening the disease cause, but not enough data to quantify

• Pancreatitis included (new disease category in GBD)

• Diverse new GBD injury categories (most injury categories have been causally linked to alcohol consumption)

• Revision of determination of risk relationship between alcohol consumption and primary epilepsy (excluding “alcoholic seizures” – in collaboration with epilepsy experts in GBD)

45

Mortality

COD

YLL


DISMOD Disability Weights

YLD DALY


StatesS: healthy (susceptible)C: diseased (condition of interest) D: dead from the diseaseM: dead from all other causes

Transition ratesi: incidencer: remissionƒ: case fatalitym: all other mortality.

Generic Model of Disease Limitations of DisMod I and II

Laborious PreprocessingNo Confidence IntervalsAd-hoc approach to incorporating prior beliefs

47


48

DisMod III: A Bayesian Approach

• Easier Preprocessing

• Model-based confidence intervals

• Systematic incorporation of prior beliefs

Deliverables and connections between the pieces

49

Mortality

COD

YLL


DisMod Disability Weights

YLD DALY

Disability weights

• Disability weights provide the bridge between mortality and non-fatal outcomes

• Disability weights quantify overall health levels associated with different states, on a continuum between perfect health (d.w.=0) and death (d.w.=1)

• GBD 1990: Six disability classes defined in reference to:

• Four domains of disability (recreation, education, procreation, occupation)

• Activities of daily living (e.g. eating, personal hygiene)

• Instrumental activities of daily living (e.g. meal preparation, housework)

Some Changes…

• Ad hoc additions and modifications and additions to 1996 disability weights based on, e.g.

• Dutch disability weights exercise

• Australian National Burden of Disease disability weights

• Further conceptual, methodological and empirical work on health state valuations

• Marrakech conference and volume on summary measures of population health

• Large-scale empirical measurement at WHO including health state valuations in community samples

• Dimensions of disability not appropriate for characterizing child outcomes

• No formal protocol to guide replication of disability weights measurement, e.g. for national burden studies

• Mildest disability class valued at 0.096 which results in insensitivity to very mild decrements

Criticisms of GBD 1990 approach

Approach to revising disability weights

• Step 1: Finalize list of sequelae

• Step 2: Develop lay descriptions of health states with reference to standardized set of domains and key symptoms

• Step 3: New data collection

• Community surveys in five or six field sites (to be chosen from provisional list including sites in India, Indonesia, Vietnam, Tanzania, Mexico, Rwanda)

• Supplemental Internet survey

Who is involved ?Mohsen Naghavi, R Lozano C. Mathers, T Vos, M Ezzati

Ali Mokdad, Kana Fuse, IHME; Joshua Salomon, HarvardCollaborators internationally implementing survey

Outline



• GBD 2005 is not only an update


53

Where is GBD making contributions to global health

Foster Dialogue and Transparency with Experts

Example:

• Host four rounds of Expert Groups meetings: disease representatives present data collected in systematic reviews in Jan/Feb 09

• Opportunity for Expert Groups to present their status, solicit feedback, and express cross-cutting issues

• IHME previews our work on methods development to produce mortality envelopes, cause of death estimates, and the new DISMOD III tool

54

Where is GBD making contributions to global health─ Potential for significant influence on health policy debates. Much

already achieved (eg. tobacco control focus at WHO).

─ Potential to greatly increase reliability/scientific basis for estimates through more and better linkages with scientific community.

─ Data/knowledge base needs to grow at same or faster pace than methodological advances.

─ Four key specific scientific challenges:

i. resolving legitimate VR/EE differences in causes of death;

ii. broaden the data/knowledge base about disease/injury epidemiology;

iii. more rigorous, acceptable and transparent procedures for ensuring epidemiological consistency; and

iv. better methods to quantify disability and population health.

55

56

Jan/Feb 2009 Meetings - Expert Group presentations - Mortality data - COD data

Nov 2010 - Estimates Completed

Aug 2009 - DISMOD III run

Nov 2009 - Peer review completed - DW data collection completed

Iterations… - Track down additional data - Run DISMOD III - Produce updated COD and Mortality Numbers

When - Overall Timing

58

Questions

Resources

• GBD Operations Manual (http://www.globalburden.org/gbdops.html)

• GBD External Website (http://www.globalburden.org/)

• GBD Internal Website (http://globalburden.health-metrics.org/)

http://www.globalburden.org/gbdops.html

http://www.globalburden.org/

http://globalburden.health-metrics.org/

UNIVERSITY OF WASHINGTON Global Burden of Diseases, Injuries and Risk Factors: methodological aspects and trends 1 st International Conference on the Burden.

Documents