Global Burden of Diseases, Injuries, and Risk Factors Study 2010: Comorbidity

UNIVERSITY OF WASHINGTON

Global Burden of Diseases,

Injuries, and Risk Factors Study

2010: Comorbidity

June 18, 2013

Theo Vos

Professor of Global Health

2

Outline

Exploration of comorbidity in Medical Expenditure Panel Surveys (MEPS) in USA

Comorbidity simulation: “COMO”

Comorbidity in MEPS

3

Medical Expenditure Panel Surveys (MEPS)

o New panel starts every year

o 5 data collection points over two years for each panel

o Main focus on expenditure of any health service contact

o 2000 to 2009

o 192,806 observations from 108,522 individuals

o Diagnostic info on 158 GBD disease and injury categories

o Health status information by SF-12, twice over two years

Mapping SF-12 to GBD disability weights

• Convenience sample of 60 IHME staff who had not worked on GBD

• Asked to fill in SF-12 for a random pick of 50 out of 60 health states spanning the spectrum from very mild to most severe in the disability weight surveys

Very mild: “has some difficulty with distance vision, for example reading signs, but no other problems with eyesight”

Most severe: “hears and sees things that are not real and is afraid, confused, and sometimes violent. The person has great difficulty with communication and daily activities, and sometimes wants to harm or kill himself (or herself)

• Respondents asked to fill in SF-12 for an individual as described in the lay descriptions presented

4

Mapping SF-12 to GBD disability weights

• 394 observations (18% of total) excluded from further analysis as they were more than two standard deviations from the median

• Loess regression of remaining SF-12 scores and disability weights

5

Parsing overall DWs into DWs for each health state

• Assume multiplicative function of comorbid health states:

• Mixed-effects model with a logit-transformed dependent variable

• Logit-transforming the outcome variable offers the benefit of

limiting outcome DW between 0 and 1, and it defines a

multiplicative relationship between the independent parameters,

consistent with the multiplicative model of combining disability

weights for YLD estimation

6

Parsing overall DWs into DWs for each health state• Disability weights were modeled for each m measure of each i

individual over n total conditions in the survey as follows:

where Ui is random intercept on individual to control for variation over multiple observations for the same individual

• This allowed us to measure the severity in GBD disability weight terms for any condition in an individual, while controlling for all other conditions present

7

Age and comorbidity

8

Population-level predictions

9

10

Dependent and independent comorbidity for diabetes

11

Major depression COPD

Asthma Migraine

12

Conclusions from MEPS

o Age is no longer a major predictor of comorbidity if a large

number of health states are accounted for

o A multiplicative model of “combining” disability weights

derived for all ages replicates the age pattern of levels of

disability reported by individuals on SF-12 (and translated

by us into GBD disability weights)

o After correcting for independent comorbidities, adding

dependent probabilities of co-occurring conditions makes

little difference

13

Outline

Exploration of comorbidity in Medical Expenditure Panel Surveys (MEPS) in USA

Comorbidity simulation: “COMO”

Disability in a comorbid case: individual perspective• The experience of living with multiple diseases:

o Disability weights are multiplicative, not additive

o Cumulative (multiplicative) weight is lower than additive

14

Comorbidity corrected DW IHD = 0.1/0.4 * 0.37 = 0.0925Comorbidity corrected DW stroke = 0.3/0.4 * 0.37 = 0.2725

Population perspective

oSimulate hypothetical populations of 10,000 for each age, sex, year, country: 0.25 billion people simulated!

oUse prevalence of each of 1120 health states as probabilities

oDetermine for each individual if they have 0, 1, 2 …n comorbid health states

oUse multiplicative function to get “comorbidity corrected” total DW for each individual

oProportionately reduce the value of each comorbid health state’s DW for that individual

oAverage all DWs for all individuals with a health state after the correction

15

Comorbidity correction by age

16

17

Conclusions

oUseful new insights on comorbidity from dataset with rich diagnostic and health status information

oSearch for similar non-USA datasets, preferably in LMIC, to replicate these analyses: potential candidates in China and Turkey

oDecision to seriously address comorbidity in GBD was most compelling reason to abandon the previous approach of incidence YLD

Incidence