Geographical Variation of Geographical Variation of Noncommunicable Diseases and Noncommunicable Diseases and

Environmental Risk Factors: Environmental Risk Factors: Application of Bayesian Modeling and Application of Bayesian Modeling and

GISGIS

• Elena Moltchanova1

• Mika Rytkönen1

• Anne Kousa2

• Olli Taskinen1

• Jaakko Tuomilehto1

• Marjatta Karvonen1 for the SPAT Study Group

1 National Public Health Institute, Helsinki, Finland2 Geographical Survey of Finland, Kuopio, Finland

MODEL: The DataYik = number of events in cell i age-group k

Nik = population at risk in cell i age-group k

Zi = other cell-specific covariates in cell i

W = neighborhood matrix of the area such that

wij = 1 if cells i and j are neighbors

wij = 0 otherwise

wii = 0 i

i

MODEL: The RelationshipsLikelihood:

Yik ~ Poisson (ikNik)

log(ik) =+0i+ k + Zi

Priors::

ln 0i ~ N ( ln 0-i,*mi)

~ N (0,0.0001)

~ N (0,0.0001)

~ Gamma (0.001,0.001)

W

j

h

i

Nik

Yik Zi

MODEL: DAG

Nik

Yik Zi

Nik

Yik Zi

MODEL: The Parametersα = overall average risk level

β = age group effect on risk/incidence

ξ = effect of cell-specific covariates on risk/incidence

λi = geographical deviation from the mean at cell i for age group 0

τ = overall geographical precision (inverse variation)

The occurrence of coronary heart disease (CHD) varies widelybetween different populations. In industrialized countries it is thegreatest single cause of death. In Finland CHD mortality is higher than in most populations.

The most important single disorder in cardiovascular disease isishaemic heart disease including acute myocardial infarction (AMI).

Earlier research has shown that the incidence of AMI varies widelywithin Finland. Although there has been a steady decrease in incidence during the last two decades, this difference still persists.

Application: AMI

AMI: DataAMI = Acute Myocardial Infarction (ICD9 410-414)

Analysed population-at-risk: 35-74 year old men

numberof cells

numberof cases

populationat risk

age-standardizedincidence

1983 2731 6732 1149855 524.00

1988 2728 6322 1191491 490.05

1993 2734 5892 1252817 428.12

Results

Element Posterior mean 95 % CI

Tot. water hardness (ºdH) -0.0097160 (-0.0213600, -0.0003195)

Zn (µg/l) -0.0006656 (-0.0061290, 0.0048140)

Al (µg/l) -0.0002723 (-0.0007370, 0.0001862)

Cu (µg/l) 0.0400800 (-0.0652800, 0.1477000)

F (mg/l) -0.0317200 (-0.1453000, 0.0898500)

Fe (mg/l) 0.1015000 (-0.1298000, 0.3176000)

NO3 (mg/l) 0.0006068 (-0.0003548, 0.0015870)

Observed age-standardized incidence of AMI among 35-74 year old men in Finland 1983, 1988, 1993

Posterior mean incidence of AMI among 35-74 year old men in Finland in 1983, 1988, 1993

Posterior probability of being a high-risk area of AMI incidence among 35-74 year old men in

Finland in 1983, 1988, 1993

Application: DM1There is a striking variation in the incidence of childhood type 1 diabetes (DM1) between and within populations.

Childhood type 1 diabetes (DM1) is of a particular importance in Finland, where the incidence is the highest in the world and still increasing.

The aetiology of DM1 and the cause or causes of the increase in frequency are unknown. Geographical variations in DM1 can be interpreted as evidence of environmental and genetic factors in the aetiology of the disease.

DM1: Data• 3649 cases from the period 1987-1996• almost 100% ascertainment• 95% supplied with coordinates

• population data available for the years 1987, 1989, 1991, 1993 and 1995

• Urban rural-rural status: 1. urban areas

2. urban-adjacent rural areas,

3. rural heartland areas

4. remote/isolated areas

ResultsParameter mean sd 2.50 % median 97.50 %

θ12 0.0858 0.0725 -0.0572 0.0864 0.2262

θ13 -0.0637 0.0725 -0.2049 -0.0641 0.0800

θ14 0.0127 0.0753 -0.1313 0.0118 0.1614

θ23* 0.1495 0.0513 0.0472 0.1501 0.2484

θ24 0.0731 0.0573 -0.0422 0.0740 0.1819

θ34 -0.0764 0.0531 -0.1821 -0.0757 0.0262

Estimated effects of area rurality on the incidence of DM1 among 0-14 year olds in Finland. θ ij

is the difference between the area types i and j, where 1= remote area , 2 = rural heartland, 3 =

urban-adjacent rural area and 4 = urban area.

Observed age-standardized incidence of DM1 among 0-14 year old children in Finland in 1987-1991 and 1992-1996

Posterior mean incidence of DM1 among 0-14 year old children in Finland in 1987-1991 and 1992-1996

Posterior probability of being a high-risk area of DM1 incidence among 0-14 year old children in

Finland in 1987-1991 and 1992-1996

Conclusions• Disease mapping is an important explorative and hypothesis- generating tool.

• Continuous speedy progress due to GIS, Bayesian methodology and computer technology development.

• Our study has produced an interesting and useful methodological framework & software needed for it’s implementation.

• Future directions of our research include a more detailed exploration of socio-economic aspect, study of other similar diseases of complex aetiology e.g. Parkonsonism and further software development