Assessing spatial confounding in cancer disease mapping using R

Psychosis prevalence in London neighbourhoods; A case study in spatial confounding

Analysis of impacts of neighbourhood risk factors on mental health outcomes frequently adopts a disease mapping approach, with unknown neighbourhood influences summarised by random effects. However, such effects may show confounding with observed predictors, especially when such predictors have a clear spatial pattern. Here, the standard disease mapping model is compared to methods which account and adjust for spatial confounding in an analysis of psychosis prevalence in London neighbourhoods. Established area risk factors such as area deprivation, non-white ethnicity, greenspace access and social fragmentation are considered as influences on psychosis. The results show evidence of spatial confounding in the standard disease mapping model. Impacts expected on substantive grounds and available evidence are either nullified or reversed in direction. It is argued that the potential for spatial confounding to affect inferences about geographic disease patterns and risk factors should be routinely considered in ecological studies of health based on disease mapping.

Alleviating spatial confounding in frailty models

The confounding between fixed effects and (spatial) random effects in a regression setup is termed spatial confounding. This topic continues to gain attention and has been studied extensively in recent years, given that failure to account for this may lead to a suboptimal inference. To mitigate this, a variety of projection-based approaches under the class of restricted spatial models are available in the context of generalized linear mixed models. However, these projection approaches cannot be directly extended to the spatial survival context via frailty models due to dimension incompatibility between the fixed and spatial random effects. In this work, we introduce a two-step approach to handle this, which involves (i) projecting the design matrix to the dimension of the spatial effect (via dimension reduction) and (ii) assuring that the random effect is orthogonal to this new design matrix (confounding alleviation). Under a fully Bayesian paradigm, we conduct fast estimation and inference using integrated nested Laplace approximation. Both simulation studies and application to a motivating data evaluating respiratory cancer survival in the US state of California reveal the advantages of our proposal in terms of model performance and confounding alleviation, compared to alternatives.

The ethnic density effect as a contextual influence in ecological disease models: Establishing its quantitative expression

Research suggests higher neighbourhood ethnic minority density to be associated with lessened chances of ethnic group illness. We focus on the density effect on psychosis, arguing that (at higher ethnic concentrations) it acts as a contextual influence attenuating the compositional influence whereby minority ethnicity is associated with higher psychosis risk. In terms of ecological disease regression, the ethnic density effect will then be apparent in nonlinear impacts of minority concentration. Contextual effects may also be evident in spatially varying regression coefficient models for psychosis. Nonlinearity or heterogeneity may be associated with other contextual processes where geography modifies demography (e.g. deprivation amplification). We illustrate these issues with an analysis of psychosis prevalence in 4835 London neighbourhoods. The data are collected in primary care (during 2019/20) using clinical diagnosis (e.g. based on referrals to specialists or psychosis hospitalisation), and refer to patients currently under care: such care may extend retrospectively over several years. The data offer a complete population perspective in contrast to survey data, which typically offer limited geographic perspectives. We consider impacts on psychosis prevalence of non-white ethnicity, as well as those of deprivation, social fragmentation and urbanicity. We find evidence suggesting nonlinear impacts of non-white ethnicity on psychosis (essentially flat risk above a threshold concentration), but find no evidence for deprivation amplification.

A joint Bayesian spatiotemporal risk prediction model of COVID-19 incidence, IC admission, and death with application to Sweden

The three closely related COVID-19 outcomes of incidence, intensive care (IC) admission and death, are commonly modelled separately leading to biased estimation of the parameters and relatively poor forecasts. This paper presents a joint spatiotemporal model of the three outcomes based on weekly data that is used for risk prediction and identification of hotspots. The paper applies a pure spatiotemporal model consisting of structured and unstructured spatial and temporal effects and their interaction capturing the effects of the unobserved covariates. The pure spatiotemporal model limits the data requirements to the three outcomes and the population at risk per spatiotemporal unit. The empirical study for the 21 Swedish regions for the period 1 January 2020-4 May 2021 confirms that the joint model predictions outperform the separate model predictions. The fifteen-week-ahead spatiotemporal forecasts (5 May-11 August 2021) show a significant decline in the relative risk of COVID-19 incidence, IC admission, death and number of hotspots.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00168-022-01191-1.

A flexible Bayesian nonconfounding spatial model for analysis of dispersed count data

In employing spatial regression models for counts, we usually meet two issues. First, the possible inherent collinearity between covariates and the spatial effect could lead to misleading inferences. Second, real count data usually reveal over- or under-dispersion where the classical Poisson model is not appropriate to use. We propose a flexible Bayesian hierarchical modeling approach by joining nonconfounding spatial methodology and a newly reconsidered dispersed count modeling from the renewal theory to control the issues. Specifically, we extend the methodology for analyzing spatial count data based on the gamma distribution assumption for waiting times. The model can be formulated as a latent Gaussian model, and consequently, we can carry out the fast computation by using the integrated nested Laplace approximation method. We examine different popular approaches for handling spatial confounding and compare their performances in the presence of dispersion. Two real applications from a crime study against women in India as well as stomach cancer incidences in Slovenia motivate the suggested methods. We also perform a simulation study to understand the proposed approach's merits better. Supplementary Materials for this article are available.

Assessing spatial confounding in cancer disease mapping using R

BACKGROUND

Exploring spatial patterns in the context of cancer disease mapping (DM) is a decisive approach to bring evidence of geographical tendencies in assessing disease status and progression. However, this framework is not insulated from spatial confounding, a topic of significant interest in cancer epidemiology, where the latent correlation between the spatial random effects and fixed effects (such as covariates), often lead to misleading interpretation.

AIMS

To introduce three popular approaches (RHZ, HH and SPOCK; details in paper) often employed to tackle spatial confounding, and illustrate their implementation in cancer research via the popular statistical software R.

METHODS

As a solution to alleviate spatial confounding, restricted spatial regressions are constructed by either projecting the latent effect onto the orthogonal space of covariates, or by displacing the spatial locations. Popular parametric count data models, such as the Poisson, generalized Poisson and negative binomial, were considered for the areal count responses, while the spatial association is quantified via the conditional autoregressive (CAR) model. Our method of inference in Bayesian, sometimes aided by the integrated nested Laplace approximation (INLA) to accelerate computing. The methods are implemented in the R package RASCO available from the first author's GitHub page.

RESULTS

The results reveal that all three methods perform well in alleviating the bias and variance inflation present in the spatial models. The effects of spatial confounding were also explored, which, if ignored in practice, may lead to wrong conclusions.

CONCLUSION

Spatial confounding continues to remain a critical bottleneck in deriving precise inference from spatial DM models. Hence, its effects must be investigated, and mitigated. Several approaches are available in the literature, and they produce trustworthy results. The central contribution of this paper is providing the practitioners the R package RASCO, capable of fitting a large number of spatial models, as well as their restricted versions.