Modelling and Implementation of spatio-temporal disease mapping (STDM) Antonio L´ opez-Qu´ılez
Grup d’Estad´ıstica espacial i temporal en Epidemiologia i Medi Ambient
(GEeitEma – UV)
Facundo Mu˜ noz
Departament d’Estad´ıstica i Investigaci´ o Operativa
Modelling and Implementation of STDM
STDM Workshop 2009
1 / 26
So we need to implement a STDM model...
... but WHICH ONE? Multiple alternatives and proposals → no BEST Selection criteria: automatic fit versatile fast Accurate!
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
2 / 26
So we need to implement a STDM model...
... but WHICH ONE? Multiple alternatives and proposals → no BEST Selection criteria: automatic fit versatile fast Accurate!
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
2 / 26
So we need to implement a STDM model...
... but WHICH ONE? Multiple alternatives and proposals → no BEST Selection criteria: automatic fit versatile fast Accurate!
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
2 / 26
Framework for STDMs
Four stages: 1 2 3 4
Probabilistic model for observations Components of the linear predictor Structures of the effects Inference methodology
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
3 / 26
Framework: Probabilistic model for observations
Poisson, Binomial or Negative Binomial distribution? Relative or absolute risk? External or internal standardization of the population? All depend on the dataset, and do not condition the spatio-temporal modelling, so... As appropriate
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
4 / 26
Framework: Probabilistic model for observations
Poisson, Binomial or Negative Binomial distribution? Relative or absolute risk? External or internal standardization of the population? All depend on the dataset, and do not condition the spatio-temporal modelling, so... As appropriate
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
4 / 26
Framework: Probabilistic model for observations
Poisson, Binomial or Negative Binomial distribution? Relative or absolute risk? External or internal standardization of the population? All depend on the dataset, and do not condition the spatio-temporal modelling, so... As appropriate
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
4 / 26
Framework: Components of the linear predictor
ηijk = g(rijk ) = Int.+Ck + Si + Tj + CSik + CTjk + STij + CSTijk +εijk | {z } | {z } main effects
interaction terms
Intercept: Base risk Main effects: Additional risk of belonging to group k, living in region i and period j respectively. Interaction terms: Contribution to the risk due to a combination of the effects. Extra variability term: Overall effect of other minor factors.
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
5 / 26
Framework: Components of the linear predictor
ηijk = g(rijk ) = Int.+Ck + Si + Tj + CSik + CTjk + STij + CSTijk +εijk | {z } | {z } main effects
interaction terms
Intercept: Base risk Main effects: Additional risk of belonging to group k, living in region i and period j respectively. Interaction terms: Contribution to the risk due to a combination of the effects. Extra variability term: Overall effect of other minor factors.
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
5 / 26
Framework: Components of the linear predictor
ηijk = g(rijk ) = Int.+Ck + Si + Tj + CSik + CTjk + STij + CSTijk +εijk | {z } | {z } main effects
interaction terms
Intercept: Base risk Main effects: Additional risk of belonging to group k, living in region i and period j respectively. Interaction terms: Contribution to the risk due to a combination of the effects. Extra variability term: Overall effect of other minor factors.
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
5 / 26
Framework: Components of the linear predictor
ηijk = g(rijk ) = Int.+Ck + Si + Tj + CSik + CTjk + STij + CSTijk +εijk | {z } | {z } main effects
interaction terms
Intercept: Base risk Main effects: Additional risk of belonging to group k, living in region i and period j respectively. Interaction terms: Contribution to the risk due to a combination of the effects. Extra variability term: Overall effect of other minor factors.
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
5 / 26
Framework: Components of the linear predictor
ηijk = g(rijk ) = Int.+Ck + Si + Tj + CSik + CTjk + STij + CSTijk +εijk | {z } | {z } main effects
interaction terms
Intercept: Base risk Main effects: Additional risk of belonging to group k, living in region i and period j respectively. Interaction terms: Contribution to the risk due to a combination of the effects. Extra variability term: Overall effect of other minor factors.
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
5 / 26
Framework: Components of the linear predictor A toy example:
ηijk = Intercept
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
6 / 26
Framework: Components of the linear predictor A toy example:
ηijk = Intercept + Si
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
6 / 26
Framework: Components of the linear predictor A toy example:
ηijk = Intercept + Tj
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
6 / 26
Framework: Components of the linear predictor A toy example:
ηijk = Intercept + Si + Tj
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
6 / 26
Framework: Components of the linear predictor A toy example:
ηijk = Intercept + Si + Tj + STij
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
6 / 26
Framework: Components of the linear predictor A toy example:
ηijk = Int. + Ck + Si + Tj + CSik + CTjk + STij + CSTijk + εijk
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
6 / 26
Framework: Structures of the effects
Lack of data ; Additional hypothesis: Neighbouring regions are more likely to have similar risk levels risk levels are expected to evolve smoothly on time
STRUCTURE Borrowing strength Many alternatives for implementing structure in practice
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
7 / 26
Framework: Structures of the effects
Lack of data ; Additional hypothesis: Neighbouring regions are more likely to have similar risk levels risk levels are expected to evolve smoothly on time
STRUCTURE Borrowing strength Many alternatives for implementing structure in practice
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
7 / 26
Framework: Structures of the effects
Lack of data ; Additional hypothesis: Neighbouring regions are more likely to have similar risk levels risk levels are expected to evolve smoothly on time
STRUCTURE Borrowing strength Many alternatives for implementing structure in practice
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
7 / 26
Framework: Structures of the effects
Lack of data ; Additional hypothesis: Neighbouring regions are more likely to have similar risk levels risk levels are expected to evolve smoothly on time
STRUCTURE Borrowing strength Many alternatives for implementing structure in practice
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
7 / 26
Framework: Structures of the effects
Lack of data ; Additional hypothesis: Neighbouring regions are more likely to have similar risk levels risk levels are expected to evolve smoothly on time
STRUCTURE Borrowing strength Many alternatives for implementing structure in practice
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
7 / 26
Framework: Structures of the effects
Lack of data ; Additional hypothesis: Neighbouring regions are more likely to have similar risk levels risk levels are expected to evolve smoothly on time
STRUCTURE Borrowing strength Many alternatives for implementing structure in practice
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
7 / 26
Framework: Structures of the effects
Spatial effect Si : Random effects Heterogeneity effect Clustering (CAR) effect BYM specification
Two-dimensional splines Penalized splines
Combination of Penalized splines and CAR models (new!) Mar´ıa Durb´ an, Dae-Jin Lee P-spline mixed-models for spatio-temporal data Today 15:30hs
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
8 / 26
Framework: Structures of the effects
Spatial effect Si : Random effects Heterogeneity effect Clustering (CAR) effect BYM specification
Two-dimensional splines Penalized splines
Combination of Penalized splines and CAR models (new!) Mar´ıa Durb´ an, Dae-Jin Lee P-spline mixed-models for spatio-temporal data Today 15:30hs
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
8 / 26
Framework: Structures of the effects
Spatial effect Si : Random effects Heterogeneity effect Clustering (CAR) effect BYM specification
Two-dimensional splines Penalized splines
Combination of Penalized splines and CAR models (new!) Mar´ıa Durb´ an, Dae-Jin Lee P-spline mixed-models for spatio-temporal data Today 15:30hs
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
8 / 26
Framework: Structures of the effects
Temporal effect Tj : Parametric models linear, quadratic, ...
First or second order Random Walks (RW) Autoregressive processes (AR) Splines
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
9 / 26
Framework: Structures of the effects
Spatio-temporal interaction STij : Remember: STij only accounts for individual differences from Int. + Si + Tj Yet another classification (Following Knorr-Held (2000)): Type Type Type Type
(GEeitEma – UV)
I: Independence II: Temporal structure III: Spatial structure IV: Spatio-temporal structure
Modelling and Implementation of STDM
STDM Workshop 2009
10 / 26
Framework: Structures of the effects
Spatio-temporal interaction STij : Remember: STij only accounts for individual differences from Int. + Si + Tj Yet another classification (Following Knorr-Held (2000)): Type Type Type Type
(GEeitEma – UV)
I: Independence II: Temporal structure III: Spatial structure IV: Spatio-temporal structure
Modelling and Implementation of STDM
STDM Workshop 2009
10 / 26
Framework: Structures of the effects Spatio-temporal interaction STij : Type I Unstructured set of values for each combination of region and period
Implementation: iid
Usually white noise: STij ∼ N (0, σ 2 )
Models: Non-persistent local circumstances causing a slight increase or decrease in the rates in a specific region-period
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
11 / 26
Framework: Structures of the effects Spatio-temporal interaction STij : Type I Unstructured set of values for each combination of region and period
Implementation: iid
Usually white noise: STij ∼ N (0, σ 2 )
Models: Non-persistent local circumstances causing a slight increase or decrease in the rates in a specific region-period
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
11 / 26
Framework: Structures of the effects Spatio-temporal interaction STij : Type I Unstructured set of values for each combination of region and period
Implementation: iid
Usually white noise: STij ∼ N (0, σ 2 )
Models: Non-persistent local circumstances causing a slight increase or decrease in the rates in a specific region-period
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
11 / 26
Framework: Structures of the effects Spatio-temporal interaction STij : Type II Region-specific temporal evolution structures independent of that in the neighbouring regions
Implementation: As many forms as the temporal main effect (Parametric, Splines, RW, AR, ...)
Models: Risk determinants affecting specific regions and inducing deviations from the global trend
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
12 / 26
Framework: Structures of the effects Spatio-temporal interaction STij : Type II Region-specific temporal evolution structures independent of that in the neighbouring regions
Implementation: As many forms as the temporal main effect (Parametric, Splines, RW, AR, ...)
Models: Risk determinants affecting specific regions and inducing deviations from the global trend
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
12 / 26
Framework: Structures of the effects Spatio-temporal interaction STij : Type II Region-specific temporal evolution structures independent of that in the neighbouring regions
Implementation: As many forms as the temporal main effect (Parametric, Splines, RW, AR, ...)
Models: Risk determinants affecting specific regions and inducing deviations from the global trend
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
12 / 26
Framework: Structures of the effects Spatio-temporal interaction STij : Type III Period-specific spatial structures independent of adjacent periods
Implementation: Typically CAR
Models: Risk determinants affecting an area larger than the region size, but not persistent in time
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
13 / 26
Framework: Structures of the effects Spatio-temporal interaction STij : Type III Period-specific spatial structures independent of adjacent periods
Implementation: Typically CAR
Models: Risk determinants affecting an area larger than the region size, but not persistent in time
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
13 / 26
Framework: Structures of the effects Spatio-temporal interaction STij : Type III Period-specific spatial structures independent of adjacent periods
Implementation: Typically CAR
Models: Risk determinants affecting an area larger than the region size, but not persistent in time
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
13 / 26
Framework: Structures of the effects Spatio-temporal interaction STij : Type IV Full spatio-temporal structure
Implementation: Parametric function of time, with spatially-structured coefficients Random effect with given structure matrix Non-parametric smoothing function (splines) with spatially structured coefficients
Models: Risk determinants exceeding the limits of one or more regions, and that persist for more than one period of time (GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
14 / 26
Framework: Structures of the effects Spatio-temporal interaction STij : Type IV Full spatio-temporal structure
Implementation: Parametric function of time, with spatially-structured coefficients Random effect with given structure matrix Non-parametric smoothing function (splines) with spatially structured coefficients
Models: Risk determinants exceeding the limits of one or more regions, and that persist for more than one period of time (GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
14 / 26
Framework: Structures of the effects Spatio-temporal interaction STij : Type IV Full spatio-temporal structure
Implementation: Parametric function of time, with spatially-structured coefficients Random effect with given structure matrix Non-parametric smoothing function (splines) with spatially structured coefficients
Models: Risk determinants exceeding the limits of one or more regions, and that persist for more than one period of time (GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
14 / 26
Framework: Structures of the effects
Spatio-temporal interaction STij : The four types are suited to modelling different kinds of phenomena Versatility ; Overmodelling! Usual practice: first explore, then choose the adequate interaction type and use that not automatic
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
15 / 26
Framework: Structures of the effects
Spatio-temporal interaction STij : The four types are suited to modelling different kinds of phenomena Versatility ; Overmodelling! Usual practice: first explore, then choose the adequate interaction type and use that not automatic
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
15 / 26
Framework: Structures of the effects
Spatio-temporal interaction STij : The four types are suited to modelling different kinds of phenomena Versatility ; Overmodelling! Usual practice: first explore, then choose the adequate interaction type and use that not automatic
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
15 / 26
Framework: Structures of the effects
Spatio-temporal interaction STij : The four types are suited to modelling different kinds of phenomena Versatility ; Overmodelling! Usual practice: first explore, then choose the adequate interaction type and use that not automatic
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
15 / 26
Framework: Inference methodology
Bayesian approach + + + −
Unified general approach Greatest modelling flexibility Detailed posterior information on every parameter Specification of priors
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
16 / 26
Framework: Inference methodology
Bayesian approach + + + −
Unified general approach Greatest modelling flexibility Detailed posterior information on every parameter Specification of priors
MCMC − − − −
(GEeitEma – UV)
Very slow! (hours, days, ...) Technical aspects (convergence, mixing, ...) not automatic not suitable for a fast and automatic service
Modelling and Implementation of STDM
STDM Workshop 2009
16 / 26
Framework: Inference methodology
Bayesian approach + + + −
Unified general approach Greatest modelling flexibility Detailed posterior information on every parameter Specification of priors
INLA (Rue et al. (2009)) + Fast + Straightforward − Restricted to latent GMRF + Leonhard Held, Birgit Schr¨ odle Spatio-temporal disease mapping using INLA Today 15:30hs
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
16 / 26
Framework: Inference methodology
Frequentist approach Empirical Bayes techniques + Fast + Feasible in many situations + Much work in progress - Not straightforward (reparameterization, develop specific numerical algorithms, ...) - Variability in point estimates often understated
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
17 / 26
STDMs in the literature Classified according to the structure of the temporal evolution of the estimated risk for each region (Tj + STij ) a Parametric b Temporally independent c Smooth temporal evolution
(a)
(GEeitEma – UV)
(b)
Modelling and Implementation of STDM
(c)
STDM Workshop 2009
18 / 26
STDMs in the literature Parametric models Tj + STij is a linear or quadratic function of time coefficients may or may not be spatially structured + Information is shared in time (and possibly in space) + Straightforward formulation − Inappropriate for long periods of time: too restrictive! References: Bernardinelli et al. (1995) Sun et al. (2000) Assun¸c˜ ao, Reis and Oliveira (2001)
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
19 / 26
STDMs in the literature Temporally independent models Tj + STij is a sequence of spatial models + Information is shared in space + Temporal evolution is not restricted to any specific shape − No sharing of information in time − High dimensional models ; identifiability and overmodelling problems References: Waller et al. (1997) Xia and Carlin (1998) Nobre, Schmidt and Lopes (2005)
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
20 / 26
STDMs in the literature Smooth temporal evolution models Tj + STij is a smooth function of time + Temporal evolution is not restricted to any specific shape + Information is shared in time − Many alternatives for STij ; model selection criteria References: Knorr-Held (2000) Macnab and Dean (2001, 2002) Lagazio, Biggeri and Dreassi (2003) Schmid and Held (2004) Richardson, Abellan and Best (2006) Mart´ınez-Beneito, L´ opez-Qu´ılez and Botella-Rocamora (2008)
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
21 / 26
Discussion
Additional information Versatility ; few additional information Almost everything relies on spatio-temporal structure Flexible models
Standardizing Age as a covariate ; Powerful! Age-specific count data is needed Age standardization ; more versatile
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
22 / 26
Discussion
Additional information Versatility ; few additional information Almost everything relies on spatio-temporal structure Flexible models
Standardizing Age as a covariate ; Powerful! Age-specific count data is needed Age standardization ; more versatile
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
22 / 26
Discussion
Additional information Versatility ; few additional information Almost everything relies on spatio-temporal structure Flexible models
Standardizing Age as a covariate ; Powerful! Age-specific count data is needed Age standardization ; more versatile
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
22 / 26
Discussion
Additional information Versatility ; few additional information Almost everything relies on spatio-temporal structure Flexible models
Standardizing Age as a covariate ; Powerful! Age-specific count data is needed Age standardization ; more versatile
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
22 / 26
Discussion
Additional information Versatility ; few additional information Almost everything relies on spatio-temporal structure Flexible models
Standardizing Age as a covariate ; Powerful! Age-specific count data is needed Age standardization ; more versatile
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
22 / 26
Discussion
Additional information Versatility ; few additional information Almost everything relies on spatio-temporal structure Flexible models
Standardizing Age as a covariate ; Powerful! Age-specific count data is needed Age standardization ; more versatile
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
22 / 26
Discussion
Interactions More interaction terms ; more flexibility Many parameters, unidentifiability and overmodelling problems Keep it simple
Model selection Common practice: trying alternative models Not automatic
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
23 / 26
Discussion
Interactions More interaction terms ; more flexibility Many parameters, unidentifiability and overmodelling problems Keep it simple
Model selection Common practice: trying alternative models Not automatic
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
23 / 26
Discussion
Interactions More interaction terms ; more flexibility Many parameters, unidentifiability and overmodelling problems Keep it simple
Model selection Common practice: trying alternative models Not automatic
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
23 / 26
Discussion
Interactions More interaction terms ; more flexibility Many parameters, unidentifiability and overmodelling problems Keep it simple
Model selection Common practice: trying alternative models Not automatic
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
23 / 26
Discussion
Interactions More interaction terms ; more flexibility Many parameters, unidentifiability and overmodelling problems Keep it simple
Model selection Common practice: trying alternative models Not automatic
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
23 / 26
Discussion
Implementation Specificity of the model complex models ; specific situations simpler models ; more versatility Complexity and scope of the inference method simpler models ; more inference tools available & more likely to work smoothly without fine tuning (automatic fit) Speed Inference method Accuracy EB estimation ; typically yields consistent and nearly unbiased point estimates, but often understates their variability MCMC estimation ; careful diagnosis of simulation outcomes
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
24 / 26
Discussion
Implementation Specificity of the model complex models ; specific situations simpler models ; more versatility Complexity and scope of the inference method simpler models ; more inference tools available & more likely to work smoothly without fine tuning (automatic fit) Speed Inference method Accuracy EB estimation ; typically yields consistent and nearly unbiased point estimates, but often understates their variability MCMC estimation ; careful diagnosis of simulation outcomes
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
24 / 26
Discussion
Implementation Specificity of the model complex models ; specific situations simpler models ; more versatility Complexity and scope of the inference method simpler models ; more inference tools available & more likely to work smoothly without fine tuning (automatic fit) Speed Inference method Accuracy EB estimation ; typically yields consistent and nearly unbiased point estimates, but often understates their variability MCMC estimation ; careful diagnosis of simulation outcomes
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
24 / 26
Discussion
Implementation Specificity of the model complex models ; specific situations simpler models ; more versatility Complexity and scope of the inference method simpler models ; more inference tools available & more likely to work smoothly without fine tuning (automatic fit) Speed Inference method Accuracy EB estimation ; typically yields consistent and nearly unbiased point estimates, but often understates their variability MCMC estimation ; careful diagnosis of simulation outcomes
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
24 / 26
Conclusions
Euroheis 2 project: STDM fast, automatic, user-oriented Framework for STDM: popular alternatives for each of the four stages Classification of models: structure of the temporal trends Discussion: summary of relevant aspects to be considered
We hope this provides all the necessary elements to take a well-based decision
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
25 / 26
Conclusions
Euroheis 2 project: STDM fast, automatic, user-oriented Framework for STDM: popular alternatives for each of the four stages Classification of models: structure of the temporal trends Discussion: summary of relevant aspects to be considered
We hope this provides all the necessary elements to take a well-based decision
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
25 / 26
Conclusions
Euroheis 2 project: STDM fast, automatic, user-oriented Framework for STDM: popular alternatives for each of the four stages Classification of models: structure of the temporal trends Discussion: summary of relevant aspects to be considered
We hope this provides all the necessary elements to take a well-based decision
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
25 / 26
Conclusions
Euroheis 2 project: STDM fast, automatic, user-oriented Framework for STDM: popular alternatives for each of the four stages Classification of models: structure of the temporal trends Discussion: summary of relevant aspects to be considered
We hope this provides all the necessary elements to take a well-based decision
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
25 / 26
Thank you
Modelling and Implementation of spatio-temporal disease mapping
Grup d’Estad´ıstica espacial i temporal en Epidemiologia i Medi Ambient
Departament d’Estad´ıstica i Investigaci´ o Operativa
http://www.euroheis.org/
(GEeitEma – UV)
Modelling and Implementation of STDM
STDM Workshop 2009
26 / 26