Quality Assessment of Mortality Information

38
Quality Assessment of Mortality Information March 26, 2012 Rafael Lozano Professor, IHME Causes of death

description

IHME professor Rafael Lozano chaired the 2012 meeting of the Regional Advisory Committee on Health Statistics, Comité Regional Asesor sobre Estadísticas de Salud (CRAES), in Havana, Cuba. Dr. Lozano spoke on quality assessment of mortality information, explaining IHME’s work in the identification and redistribution of cause of death codes. This research supports the Global Burden of Diseases, Injuries, and Risk Factors 2010 Study.

Transcript of Quality Assessment of Mortality Information

Page 1: Quality Assessment of Mortality Information

Quality Assessment of Mortality Information

March 26, 2012

Rafael Lozano

Professor, IHME

Causes of death

Page 2: Quality Assessment of Mortality Information

2

Outline

• Quality and data qualityo definitions and assessment framework

• Measuring quality in Causes of Death under the ICD framework

• Adding value to the CoD quality o Identification of improper codes for UCD

o Defining the cause list

• Results

• Final remarks

Page 3: Quality Assessment of Mortality Information

3

Page 4: Quality Assessment of Mortality Information

4

What is quality?

• Quality (from Latin qualitas) is an attribute or a property o Attributes are given, by a subject, whereas properties are owned

• For Locke, a quality is an idea of a sensation or a perceptiono primary qualities are intrinsic to an object

o secondary qualities are dependent on the interpretation of the subjective mode and the context of appearance

• From the neutral point of view, the quality of something is the sum of its essential attributes or properties

• Something might be good because it iso Useful

o Beautiful

o Exists

Quality means the understanding of

Page 5: Quality Assessment of Mortality Information

5

What is data quality?• It is difficult to determine the exact definition, but in our daily lives we have a

pretty good sense of what is poor data quality

• Sometimes it is easier to identify inaccurate data – data that are not relevant, data that are not timely, data that are misleading, etc.

Page 6: Quality Assessment of Mortality Information

6

What do you mean by “data quality?”

The majority of people consider accuracy the most relevant dimension of data quality. Completeness, currency, and consistency come next on the list. However, we need to understand better the multidimensional concept of data quality.

Page 7: Quality Assessment of Mortality Information

7

Approaches used in the literature to study data quality

• An intuitive is based on the researchers´ experience trying to understand which attributes of data are important.

• A theoretical focuses on how data may become deficient during the data manufacturing process. Highly recommended but with few examples. Through this approach we can assess the intrinsic attributes to a data product.

• An empirical captures the attributes of data quality that are important for consumers. How data fit for use in their task. Capture the voice of customers and reveal characteristics that researchers have not considered.

Page 8: Quality Assessment of Mortality Information

8

Selected attributes to measure data qualityDimension Definition (The extent to which) Objectivity data is unbiased, unprejudiced , and impartial Believability data is regarded as true and credible Accuracy Data is correct, free of error Reputation data is highly regarded in terms of its sources or context Completeness data is not missing and is of sufficient breadth and depth for

the task at hand Value added data is beneficial to provide advantages from its use Relevancy data is applicable and helpful for the task at hard Timeliness data is sufficient up to date for the task at hard Appropriate amount of data

volume of data is appropriate for the task at hand

Concise representation

data is compactly represented

Consistent representation

data is presented in the same format

Ease manipulation data is easy to manipulate and apply to different task Understandability data is easily comprehended Interpretability data is in appropriate language, symbols, units, and the

definitions are clear Accessibility data is available, or easily and quickly retrieved Security access to data is restricted appropriately to maintain its

security

Page 9: Quality Assessment of Mortality Information

9

Assessment framework for CoD statisticsAttribute IndicatorAccuracy  

Coverage % of population covered by medical certification of cause of death Completeness % of deaths with medically-certified cause of death Missing data % of cause-of-death reports for which age/sex data are missing Use of ill-defined categories % of deaths classified under various miscellaneous and ill-defined categories Improbable classifications Number of deaths assigned to improbable age or sex categories per 100,000 coded deaths

Consistency between CoD and general mortality

% of cause-of-death data points deviating more than 2 (or 3) SDs from general mortality based predictions

Relevance  

Routine tabulations By sex, and at least by eight broad age groups—namely, 0, 1–4,5–14, 15–29, 30–44, 45–59, 60–69, and 70+ years

Small area statistics Number of cause-of-death tabulation areas per million population

Comparability  

Over time Consistency of cause specific mortality proportions over consecutive years Across space ICD to certify and code deaths; revision used and code level to which tabulations are

published

Timeliness  

Production time Mean time from end of reference period to publication Regularity SD of production time

Accessibility  

Media Number of formats in which data are released Metadata Availability and quality of documentation User service Availability and responsiveness of user service

Mahapatra P. et al Lancet 2007

Page 10: Quality Assessment of Mortality Information

10

Outline

• Quality and data quality

• Measuring quality in Causes of Death under the ICD framework

• Adding value to the CoD quality o Identification of improper codes for UCD

o Defining the cause list

• Results

• Final remarks

Page 11: Quality Assessment of Mortality Information

11

Page 12: Quality Assessment of Mortality Information

12

Critical concepts

• One cause - one death (UCD)

oGeneral principle and selection rules

oModification of the selected cause

o The modification rules

– Underlying cause of death (UCD)– Intervening cause– Highly improbable, unlikely to cause death– Ill- defined (symptoms, signs and abnormal clinical

and laboratory findings, not elsewhere classified)

Page 13: Quality Assessment of Mortality Information

13

4.1.11 Notes for use in underlying cause of death mortality

• E86 Volume depletion with mention of A00-A09 (intestinal infectious diseases) code A00-A09

• What happen when E86 or I10 appear alone or the sequence turn into I10 as UCD…

Source: ICD 10th Vol II, Second Edition 2010, pages 37 and 39

Page 14: Quality Assessment of Mortality Information

14

Quality Assessment of Causes of Death National Systems

• Mahapatra P. et al India, 2001

• Rao C. and Lopez A China, 2005

• Mathers C. et al. Bull of WHO, 2005

• França E. et al. Brazil, 2008

Page 15: Quality Assessment of Mortality Information

15

Outline

• Quality and data quality

• Measuring quality in Causes of Death under the ICD framework

• Adding value to the CoD quality o Identification of improper codes for UCD

o Defining the cause list

• Results

• Final remarks

Page 16: Quality Assessment of Mortality Information

16

What is an improper code for UCD?

• Ill-Defined causes (Chapter XVIII, ICD 10th )

• Unlikely cause of death (page 175 Vol II, 2010)

• Intermediate or immediate cause of death

• CoD that may be considered as risk factoro Hypertension or Atherosclerosis

• And depending of the granularity of the cause list, other and/or unspecified CoD within ICD chapters

Page 17: Quality Assessment of Mortality Information

17

What is the right name for these codes?• Murray and Lopez,

1996, “Garbage Codes”• Mathers C. et al, 2005,

“Ill-defined codes”• Mahapatra P. et al 2007,

“Ill-defined categories”• Naghavi M. et al, 2010,

“Garbage Codes”

• Unwanted• Inaccurate• Misclassified

• Improper codes for Underlying Cause of Death

Page 18: Quality Assessment of Mortality Information

18

o Causes that cannot or should not be considered as underlying causes of death.

o Intermediate causes of death such as heart failure, septicemia, peritonitis, osteomyelitis, or pulmonary embolism.

o Immediate causes of death that are the final steps in a disease pathway leading to death

o Unspecified causes within a larger cause grouping

• Unlikely to cause deatho ICD

o IHME

• Ill- definedo Specified

o Unspecified

• Intermediate

• Immediate

• Other and unspecified causes within chapters

• Hypertension and Atherosclerosis

Page 19: Quality Assessment of Mortality Information

19

Distribution of improper codes for UCD

40 million of deaths (ICD 10th)26.7% of total deaths

3 digit 4 digitUnlikely CoD ICD 181 1,175Unlikely CoD IHME 85 429Ill-Defined Specified 10 51Ill-Defined Unspecified 76 249Intermediate 30 137Inmediate 3 6Other and Unspecified within chapters 76 155Hypertension and Atherosclerosis 3 9All 464 2,211

ICD 10thType

Page 20: Quality Assessment of Mortality Information

20

Leading improper codes for UCD in the Americas

No. Cause ICD % Type1 Stroke, not specified as haemorrhage or infarction I64 13.1 Other within group2 Other ill-defined and unspecified causes of mortality R99 7.5 Ill-def Unsp3 Unattended death R98 7.3 Ill-def Unsp4 Congestive heart failure I500 6.4 Intermediate5 Septicaemia, unspecified A419 5.2 Intermediate6 Heart failure, unspecified I509 4.9 Intermediate7 Essential (primary) hypertension I10 3.5 H&A8 Malignant neoplasm without specification of site C80 3.3 Other within group9 Person injured in unspecified motor-vehicle accident, traffi c V892 2.5 Other within group

10 Chronic renal failure, unspecified N189 2.3 Intermediate11 Unspecified renal failure N19 2.1 Intermediate12 Sequelae of stroke, not specified as haemorrhage or infarction I694 2.0 Other within group13 Exposure to unspecified factor causing other and unspecified injury X599 1.9 Other within group14 Pneumonitis due to food and vomit J690 1.6 Intermediate15 Generalized and unspecified atherosclerosis I709 1.5 H&A16 Senility R54 1.5 Ill-def Spe17 Gastrointestinal haemorrhage, unspecified K922 1.4 Other within group18 Cardiac arrest, unspecified I469 1.4 Inmediate19 Pulmonary embolism without mention of acute cor pulmonale I269 1.4 Intermediate20 Respiratory arrest R092 1.4 Ill-def Unsp

Rest 27.7All causes 13,646,225

Page 21: Quality Assessment of Mortality Information

21

Cause list for reports

• The list of Cause of Death selected must be confined to a limited number of mutually exclusive categories able to encompass the whole range of Public Health conditions.o The categories have to be chosen to facilitate the statistical study

of CoD phenomena in the Public Health Framework.

o There will be residual categories for other miscellaneous conditions that cannot be allocated to the more specific categories. As few conditions as possible should be classified to residual categories.

o The list should has different levels of detail using a hierarchical structure with subdivisions. The list should retain the ability both to identify specific entities and to allow statistical presentation of data for broader groups, to enable useful and understandable information to be obtained.

Page 22: Quality Assessment of Mortality Information

22

Examples of short cause list for reports

• Taucher E., 1978

• Avoidable Mortality, 1990

• BTL (ICD 9th), 1979

• Tab 1 (ICD 10th), 1994

• PAHO 6/67, 2002

• Becker R. et al 2006

• GBD 1990

• GBD 2010

• New one ??

Level Group I Group II

Group III

Total

First 1 1 1 3

Second 7 8 4 19

Third 50 83 14 147

Fourth 75 128 22 225

A Communicable, maternal, perinatal and nutritional ConditionsA.1 HIV and tuberculosis

A.1.1 TuberculosisA.1.2 HIV/AIDS

A.1.2.1 HIV disease resulting in mycobacterial infection A.1.2.2 HIV disease resulting in other specified or unspecified diseases

A.2 Infectious diseases predominantly in childrenA.2.1 Diarrheal diseases

A.2.1.1 CholeraA.2.1.2 Other salmonella infectionsA.2.1.3 ShigellosisA.2.1.4 Enteropathogenic Escherichia coli infectionA.2.1.5 Enterotoxigenic Escherichia coli infectionA.2.1.6 Campylobacter enteritisA.2.1.7 AmoebiasisA.2.1.8 CryptosporidiosisA.2.1.9 Rotaviral enteritis A.2.1.10 Other diarrheal disease

GBD 2010 Cause list

Page 23: Quality Assessment of Mortality Information

23

I60-I69 Cerebrovascular diseases

• Ischemic stroke o I63 Cerebral infarction

o I65 Occlusion and stenosis of pre-cerebral arteries, not resulting in cerebral infarction

o I66 Occlusion and stenosis of cerebral arteries, not resulting in cerebral infarction

o I67(except I67.4) Other cerebrovascular diseases (Hypertensive encephalopathy)

o I69.3 Sequelae of cerebral infarction

• Hemorrhagic and other non-ischemic strokeo I60 Subarachnoid hemorrhage

o I61 Intra-cerebral hemorrhage

o I62 Other no traumatic intracranial hemorrhage

o I69.0-I69.2 Sequelae of: subarachnoid hemorrhage, intra-cerebral hemorrhage, and other no traumatic intracranial hemorrhage

o I67.4 Hypertensive encephalopathy

• Stroke not specified as hemorrhagic or Ischemic o I64 Stroke, not specified as hemorrhage or infarction

o I69.4 Sequelae of stroke, not specified as hemorrhage or infarction

o I69.8 Sequelae of other and unspecified cerebrovascular diseases

Page 24: Quality Assessment of Mortality Information

24

Assumptions

• The proportion of deaths assigned to unspecified stroke is negatively associated with the proportion of deaths assigned to individual target codes.

• The epidemiological distribution of causes, as well as the nosological paradigms within the stroke universe, tend to be similar within given country.

%Target = α+ β(% Unspecified Stroke) + Ζ(μ) + ε

Where: • %Target = proportion of deaths attributable to a given target code (either

ischemic or hemorrhagic) within the stroke universe • % Unspecified Stroke = proportion of deaths attributable to a unspecified

stroke within the Stroke universe • μ = a vector of normally-distributed random effects with mean Ε(μ)=0

Page 25: Quality Assessment of Mortality Information

25

HypertensiveHeart

Diseases

Nephropathies

DiabetesChronic Kidney

Diseases

New ways to group causes

CKD due to Diabetes: E10.2, E11.2, E12.2, E13.2, E14.2CKD due to Hypertension: I12.0, I12.9, I13.0, I13.1, I13.2, I13.9Other CKD: N02‐N07, N15.0

Page 26: Quality Assessment of Mortality Information

26

Outline

• Quality and data quality

• Measuring quality in Causes of Death under the ICD framework

• Adding value to the CoD quality o Identification of improper codes for UCD

o Defining the cause list

• Results

• Final remarks

Page 27: Quality Assessment of Mortality Information

27

Improper codes for UCD, selected countries, last year available

Page 28: Quality Assessment of Mortality Information

28

Types of improper codes, last year available

Page 29: Quality Assessment of Mortality Information

29

Ill- Defined

Page 30: Quality Assessment of Mortality Information

30

Intermediate

Page 31: Quality Assessment of Mortality Information

31

Others and unspecified within chapters

Page 32: Quality Assessment of Mortality Information

32

Annual change of inappropriate codes fraction, in selected countries

Country All ill def intermediate other & Unsp H&A First LastColombia -3.5% -7.2% -1.1% -4.9% -1.7% 1997 2008Chile -3.0% -5.0% 0.4% -4.6% 0.9% 1997 2007Brazil -2.4% -5.3% -1.0% -1.5% 4.4% 1996 2009Cuba -2.1% 2.2% -1.4% 2.6% -10.3% 2001 2008Ecuador -2.0% -2.9% -2.6% 0.8% -3.6% 1997 2009Peru -1.4% -9.0% 4.3% 1.6% 1.6% 1999 2004Nicaragua -1.2% 0.4% 1.4% -4.4% 2.9% 1997 2008Paraguay -1.2% 3.0% -2.9% -4.3% 1.6% 1996 2008Canada -1.2% -1.1% 0.6% -2.6% 3.4% 2000 2004Panama -0.7% -13.7% 5.3% -0.1% 8.0% 1998 2008Venezuela -0.5% -10.9% -3.6% 2.5% -3.3% 1996 2007United States -0.4% 3.1% 0.7% -1.7% -1.2% 1999 2007Costa Rica -0.1% 1.9% -1.6% 0.7% -1.1% 1997 2009Mexico 0.1% 2.1% -0.3% -1.1% 8.6% 1998 2009Argentina 0.1% 2.6% 0.0% -2.3% -5.1% 1997 2009Guatemala 0.3% -17.4% 3.9% 1.2% 10.3% 2005 2008El Salvador 0.6% -1.3% 2.6% -0.1% 13.5% 1997 2008Uruguay 1.0% 3.0% 1.5% -0.5% -1.4% 1997 2004

Page 33: Quality Assessment of Mortality Information

33

Age ICD Chapter

ChapterIX Diseases of the circulatory system XVII Symptoms, signs and abnormal clinical and laboratory findings, not elsewhere classified XX InjuriesXIV Diseases of the genitourinary system II Neoplasms I Certain infectious and parasitic diseases X Diseases of the respiratory system

Page 34: Quality Assessment of Mortality Information

34

Outline

• Quality and data quality

• Measuring quality in Causes of Death under the ICD framework

• Adding value to the CoD quality o Defining the cause list

o Identification of improper codes for UCD

• Results

• Final remarks

Page 35: Quality Assessment of Mortality Information

35

Conclusions

• The amount of Improper Codes for UCD (based in ICD 10th) is 25% of all deaths in the region and it varies across countries, ages and years

• The amount of improper codes depends on:o the quality of COD registries (70-80%) and

o the cause list for report selected (20-30%)

• Twenty ICD 10th codes accumulate 73% of all deaths associate to improper codes, e.g., heart failure (13%), stroke unspecified (13%), ill-defined (> 20%), etc.

• There are many good experiences in the region from which to learn and also important lags to fix

Page 36: Quality Assessment of Mortality Information

36

What do we need to do on the data quality front?

• Are we “ok” with the current indicators or do we need to expand the scope?

• Do we need a different cause list for reports?

• Shall we set up a common framework as users of data and producers of information? (new studies of validation of the accuracy of death certificates)

• Do we have to explore and learn more from our customers?

Page 37: Quality Assessment of Mortality Information

37

To improve the quality … is not only raising the bar

The current global record is 2.45m and belongs to Javier Sotomayor (Cuba)

In 1968 (Mex), Richard Fosbury (USA) revolutionized the technic and jumped 2.18 m

The most achieved was 1.97 m

Page 38: Quality Assessment of Mortality Information

38

THANKS

Acknowledgments to the Causes of Death Research Team, IHME.Data Analysts: David Philips, Charles Atkinson, Diego Gonzalez-MedinaResearchers: Kyle Foreman MPH, Prof. Mohsen Naghavi, Prof. Christopher Murray