Disease-free life expectancy: Sullivan versus Multi-state revisitedHendriek Boshuizen, Rudolf Hoogenveen, AnnekeBlokstra, Pieter van Baal, Caroline Baan, Monique Verschuren

Context (1)• RIVM Chronic Diseases Model (CDM):

Dynamic population model: - start: current population- Applies disease incidences / mortalities- dependent on risk factorsAnd projects into the future

Context (2)Classical use of RIVM-CDM: compare interventions;• Theoretical (burden due to smoking)• Realistic interventions

In 2005: used for projections data on diseasesare more critical

prev

alen

ce

time Healthy life expectancy(multi-state from current situation)

RIVM Chronic disease model: life expectancy free of diabetes

multi-state, compared with Sullivan

Overweight Activity Cholesterol Bloodpressure Smoking

AMI

Heartfailure

Stroke

Diabetes

Multi-state versus Sullivan

Multi-state: life-expectancy and healthy life expectancybased on constant mortality and incidence rates

Sullivan: based on constant mortality and prevalence rates

General structure

healthy diabetes

death

Mortality

incidence

(remission)

prevalence

Sullivan

healthy diabetes

death

Mortality

incidence

(remission)

prevalence

Multi-state

healthy diabetes

death

Mortality

incidence

(remission)

prevalence

MortalityIn health

Mortality in diabetics

Multi-state model

incidence

mortality

Differences

Steady-state situation: Sullivan = multistate

Non-steady state: changing incidences:Model goes to new steady state

How fast? -Incidence and remission , and incidence >> incidence-remission :

fast

-Only incidence: takes a life time to reach a new steadystate

Life expectancy free of diabetes

• No remission• Steady state not realistic• For projections multi-state better

Need adequate data

Multi-state

healthy diabetes

death

Mortality

incidence

(remission)

prevalence

MortalityIn health

Mortality in diabetics

Data on diabetesIncidence rate

from sentinel General Practitioners Registrations

Prevalence ratefrom sentinel General Practitioners Registrations

Total mortalityStatistics Netherlands

Mortality in those with and without diabetes: ????Excess mortality = Mortality in those with diabetes – mortality

in those without diabetes

Excess mortality: 4 methods

1. From literature: Relative risks on mortality2. From IPM ( “DISMOD”)- model3. From cause-specific mortality (as registered on death

certificate)4. As 3, but related mortality from cardiovascular diseases

added

Method 1: from relative risks (by age)

Comparison mortality: non-diabetic with regular (unselected) diabeticsunadjusted for BMI

Excluding:-Only data before 1980-Non-caucasion study population-Men and women reported together-Age-groups wider than 30 years

age

RR

Method 2: using IPM (“DISMOD”) model

healthy diabetes

death

overall

mortality

incidencePrevalence-trend

Excess Mortality

Data used5 sentinel networks of GP practices:• CMR- Nijmegen (2000-2004)• LINH, contactregistration (2004)• Transition project (2000-2004) • RNH-Limburg (2001-2004) • RNUH-LEO (2001-2004)

Steady state = SullivanHere : including trend of last 10 years

Men, CMRMen, RNH

Women, CMRWomen RNH

Age-standardized prevalence of DM, indexed

Excess mortality

0

0,02

0,04

0,06

0,08

0,1

0,12

0 20 40 60 80

age

exce

ss m

orta

lity

men 2005 data

women 2005 data

men time series data

women time seriesdata

3. Estimated from diabetes as cause of death

• Primary cause of death as given on death certificate (after applying recoding rules of statistics Netherlands)

Published as rate per population number

Excess mortality = rate per populationprevalence rate

4: cause specific mortality from CVD included

Overweight Activity Cholesterol Blood pressure Smoking

CHD

Heartfailure

Stroke

Diabetes

Including CVD:

• Calculate CVD excess mortality from cause-specific mortality

• Calculate excess prevalence of CVD in diabetics, usingrelative risks of DM on CVD

Can be used to calculate extra excess mortality in diabeticsdue to CVD

Predicted prevalence of diabetes, men (multistate model)

20 40 60 80age

0.050.1

0.150.2

0.250.3

0.35prevalence

empirical

from DISMOD

from RR

causesp+CVD

causespecific

Predicted prevalence of diabetes, women

20 40 60 80age

0.050.1

0.150.2

0.250.3

0.35prevalence

empirical

from DISMOD

from RR

causesp+CVD

causespecific

Without diabetes

Withdiabetes

LE

74.770.96.76.281.477.0MS cause-specific

6.5

3.7

5.43.1

Men

7.1

4.3

6.03.6

Women

74.670.881.777.3MS cause-specific

77.073.281.376.9MS dismod

75.571.481.276.8MS basedon RR

78.774.482.277.5Sullivan

WomenMenWomenMen

Summary

• Increasing prevalence in multistate model leads toincreasing mortality lower life expectancy

• Including time trend in DISMOD moderate influenceon prevalence, but still appreciable in lifeexpectancy without diabetes

• Including cardiovascular mortality not very muchinfluence

• Predicted life expectancy with diabetes can be twiceas high depending on method used

Discussion and conclusion

• DISMOD used to doctor excess mortality, but canalso be used to doctor incidence

• direct data necessary on excess mortality