Next Article in Journal
Quality-Adjusted Life Years in Erythropoietic Protoporphyria and Other Rare Diseases: A Patient-Initiated EQ-5D Feasibility Study
Next Article in Special Issue
A Systematic Review of Areal Units and Adjacency Used in Bayesian Spatial and Spatio-Temporal Conditional Autoregressive Models in Health Research
Previous Article in Journal
Effects of Cognitive Appraisals on Perceived Self-Efficacy and Distress during the COVID-19 Lockdown: An Empirical Analysis Based on Structural Equation Modeling
Previous Article in Special Issue
Factors Associated with Low Birthweight in Low-and-Middle Income Countries in South Asia
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
IJERPH
Volume 20
Issue 7
10.3390/ijerph20075295
ijerph-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Systematic Review

A Systematic Review of Joint Spatial and Spatiotemporal Models in Health Research

by
Getayeneh Antehunegn Tesema
1,2,*,
Zemenu Tadesse Tessema
1,2,
Stephane Heritier
1,
Rob G. Stirling
3,4 and
Arul Earnest
1
1
School of Public Health and Preventive Medicine, Monash University, Melbourne, VIC 3004, Australia
2
Department of Epidemiology and Biostatistics, Institute of Public Health, College of Medicine and Health Sciences, University of Gondar, Gondar 196, Ethiopia
3
Department of Respiratory Medicine, Alfred Health, Melbourne, VIC 3004, Australia
4
Faculty of Medicine, Nursing and Health Sciences, Central Clinical School, Monash University, Melbourne, VIC 3004, Australia
*
Author to whom correspondence should be addressed.
Int. J. Environ. Res. Public Health 2023, 20(7), 5295; https://doi.org/10.3390/ijerph20075295
Submission received: 16 January 2023 / Revised: 13 March 2023 / Accepted: 24 March 2023 / Published: 28 March 2023
(This article belongs to the Special Issue Application of Statistical Methods in Public Health and Medical Research)
Browse Figures
Review Reports Versions Notes

Abstract

:
With the advancement of spatial analysis approaches, methodological research addressing the technical and statistical issues related to joint spatial and spatiotemporal models has increased. Despite the benefits of spatial modelling of several interrelated outcomes simultaneously, there has been no published systematic review on this topic, specifically when such models would be useful. This systematic review therefore aimed at reviewing health research published using joint spatial and spatiotemporal models. A systematic search of published studies that applied joint spatial and spatiotemporal models was performed using six electronic databases without geographic restriction. A search with the developed search terms yielded 4077 studies, from which 43 studies were included for the systematic review, including 15 studies focused on infectious diseases and 11 on cancer. Most of the studies (81.40%) were performed based on the Bayesian framework. Different joint spatial and spatiotemporal models were applied based on the nature of the data, population size, the incidence of outcomes, and assumptions. This review found that when the outcome is rare or the population is small, joint spatial and spatiotemporal models provide better performance by borrowing strength from related health outcomes which have a higher prevalence. A framework for the design, analysis, and reporting of such studies is also needed.

1. Introduction

Recent advances in geographic information systems in medicine have led to the development of advanced spatial analysis of geocoded data in health research [1]. Disease mapping, also defined as the spatial analysis of disease risk, is an important area of public health research [2]. Commonly georeferenced data used in epidemiological investigations have information about space and perhaps time [3]. In spatial epidemiology, Tobler’s first law of geography is considered as the foundation of spatial statistics, which asserts that everything is related to everything else, although proximate things are more closely related than distant things [4]. Nearby areas are more likely to share similar geographic characteristics linked to the disease and, likewise, the temporal dependence is greater for succeeding years than for years apart from one another [5,6].
Spatial models have been in use in the field of public health research for decades [7], and important progress over decades has enabled the development of complex models to examine a potential correlation between disease patterns and covariates that are geographically and/or temporally varied [8]. In disease risk mapping, the Standardised Incidence Rate (SIR) and Standardised Mortality Ratio (SMR) are frequently used to measure spatial risk [9]. However, SIR and SMR have limitations when the outcome is rare or when the population is small [10]. To overcome these issues, Bayesian spatial models are applied to obtain smoothed risk by considering spatial dependence (structured and unstructured spatial random effects) in the model [11,12]. Any overdispersion or spatial dependency in the data that cannot be accounted for by the covariates is taken into account by the random effects in the model [13].
Conditional Autoregressive (CAR) and Simultaneous Autoregressive (SAR) prior distributions are routinely used to model the spatially structured random effects [13,14]. In the majority of spatial studies, the spatially structured random effect is modelled using the CAR prior distribution. There are four classes of CAR prior distribution including intrinsic, convolution, Cressie, and Leroux [15]. As mentioned above, data sparseness and a small population are the common shortcomings in spatial modelling. To overcome these limitations, the Besag York Models (BYM) were introduced by Besag et al., 1991 [16]. These models borrow strength from nearby locations and apply spatial smoothing to the risks of the disease, which could improve the accuracy of risk estimations in areas with limited cases or small populations [17].
Rapid advancement in geographically indexed data and statistical innovations has contributed to the growth of spatial studies [18]. The univariate disease mapping approaches have recently been extended to joint disease mapping (modelling multiple interrelated diseases simultaneously) in space and/or time [19,20]. When the desired outcome is rare, the joint spatial models can improve the statistical power by borrowing strength from neighbourhood areas, periods, and/or related highly prevalent outcomes [21]. As many diseases are interrelated and many public health interventions are planned at several dimensions, joint spatial and spatiotemporal analyses are essential for better decision-making and evaluation of already implemented initiatives [22].
The single disease has been studied using univariate spatial and spatiotemporal models; however, these models are not capable of borrowing strength from related diseases [23]. Models that take into account the correlation between diseases improve the estimates of disease risk [24]. The joint spatial and spatiotemporal models combine information from different diseases that share similar risk factors [25]. The majority of spatial and spatiotemporal studies to date have been at the univariate level, considering spatial modelling of specific diseases. However, as many diseases share similar risk factors [26,27,28], applying models that can incorporate data from related diseases is useful from both an epidemiological and statistical perspective [29]. A new field of spatial analysis called shared component spatial and spatiotemporal models [30] analyses both the specific temporal and spatial patterns for each outcome as well as shared spatial and temporal patterns common to multiple outcomes [31].
Joint spatial and spatiotemporal models, as opposed to univariate models, concurrently account for specific and common spatial and temporal effects by incorporating shared spatial and temporal terms [32]. The joint spatial and spatiotemporal models were frequently applied to non-communicable diseases (NCDs) including cancer and diabetes mellitus (DM) [32,33,34,35,36,37,38,39,40,41,42]. The development of joint spatial and spatiotemporal methodologies has coincided with a huge advancement in statistical approaches addressing technical and statistical issues related to advanced spatial statistics. For example, authors employed different Bayesian inference techniques such as Markov chain Monte Carlo (MCMC) [32,33,35,36,37,38,39,40,41,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60] or Integrated Nested Laplace Approximation (INLA) methods [42,43,61,62,63,64,65] in Bayesian shared component spatial and spatiotemporal models.
Although spatial analysis of multiple health outcomes simultaneously has increased over the past few years, a systematic review of published research using joint spatial and spatiotemporal approaches has not yet been undertaken. These studies are heterogeneous in types of analytical models, methodological gaps, spatial and temporal structures, methods of inference, etc. For researchers, especially in the area of spatial statistics, a summary and description of joint spatial and spatiotemporal analysis methods, software, methodological gaps, and modelling concerns are essential.
Therefore, we conducted a systematic review of joint spatial and spatiotemporal models applied to health outcomes to provide insightful recommendations for future researchers as to when and how to fit a joint spatial and spatiotemporal model. This systematic review helps to improve the decision-making process through the joint spatial and spatiotemporal modelling of two or more health outcomes. In addition, the results could inform researchers in terms of providing insights about advanced joint spatial and spatiotemporal statistical methods and related issues.

2. Materials and Methods

2.1. Data Source and Search Strategy

We performed a systematic review of peer-reviewed published health research that employed joint spatial and spatiotemporal methods. For the formulation of the systematic review methodology, we used the Preferred Reporting Items for Systematic Reviews and Meta-analysis (PRISMA) checklist [66]. We registered the systematic review on the PROSPERO international prospective register of systematic reviews (registration number: CRD42022365445). A comprehensive search strategy was carried out for joint spatial and spatiotemporal models (joint spatial autocorrelation, joint spatiotemporal autocorrelation, joint spatial model, and joint spatiotemporal models) applied to any health or health-related outcomes with no geographic limits.
In our review, a spatial model incorporates a geo-spatial index, a temporal model includes a time index, a spatiotemporal model involves both a geospatial and time index, and joint spatial and spatiotemporal models accommodate a geospatial and/or time index of two or more health outcomes [67,68]. The search was conducted on 19 September 2022. Databases such as PubMed, Medline, Scopus, PsycINFO, Emcare, and Embase were searched. The reference lists of retrieved studies were further searched on Google Scholar and advanced Google to identify more papers. Search terms for the joint spatial and spatiotemporal studies are detailed in Table S1. Studies published between January 2011 and October 2022 without geographic restrictions were considered in our review.
Retrieved articles from each database were exported to Endnote version 20 reference citation software and stored as a single file name and then exported to Covidence for further processing (Covidence systematic review software (Veritas Health Innovation, Melbourne, Australia. Available at www.covidence.org, accessed on 20 August 2022)). Duplicates were deleted manually during the title and abstract and full-text screening and automatically in Endnote and Covidence software. Searches were conducted using the following terms: “multivariate spatiotemporal” OR “bivariate spatiotemporal” OR “multivariate spatio-temporal” OR “bivariate spatio-temporal” OR “joint shared spatial model*” OR “joint space-time model” OR “multivariate space-time model*” OR “bivariate space-time model” OR “small area analys*” AND “shared component model” OR “disease mapping” AND “shared component model” OR “space-time mixture model” OR “shared component model” OR “spatial analys*” AND “joint model*” OR “joint spatial model*” OR “joint spatial analys*” OR “shared latent component model” OR “joint model*” AND “spatial model*” OR “spatial factor analys*” OR “risk map*” AND “shared component model*” OR “shared spatial model*” OR “multivariate spatial analys*” OR “bivariate spatial analys*” OR “bivariate conditional autoregressive model” OR “multivariate conditional autoregressive model” OR “joint conditional autoregressive model” OR “joint spatial autocorrelation” OR “bivariate spatial autocorrelation” OR “multivariate spatial autocorrelation” OR “spatial co-cluster*” OR “spatio-temporal co-cluster*” (Table S1).

2.2. Inclusion and Exclusion Criteria

This systematic review covered peer-reviewed studies published in the English language between 2011 and 2022 modelled using joint spatial and spatiotemporal models. The year 2011 was chosen as the starting point because the joint spatial and spatiotemporal analysis of two or more health outcomes was widely implemented as a new area of study over the past decades and because of the need for most recent evidence due to the rapid changes in analytical techniques due to the ongoing advancement in science and technology. Studies before that date were either outdated or superseded by newer methods included in our review. Every article retrieved from the databases (Medline, PubMed, PsycINFO, Emcare, Scopus, and Embase) was exported to Endnote. After excluding duplicates from Endnote, we transferred to Covidence for additional article screening and extraction.
Title and abstracts and full-texts were screened by two authors (GAT and ZTT) independently to identify eligible studies based on the inclusion and exclusion criteria. When conflicts emerged over the inclusion or exclusion of studies, a consensus was reached through discussion, and if the conflicts were not resolved a third reviewer (AE) was consulted. Research articles performed their analyses using the joint spatial approaches; joint spatial and spatiotemporal autocorrelations, joint spatial models, and joint spatiotemporal models, and the outcomes analysed (could be on health outcomes among humans (not animal study)) were eligible for this review. No exclusion was made based on geography and types of health outcomes studied. Conference abstracts, reviews, texts published in a language other than English, non-human studies, and those not considering joint spatial and spatiotemporal models were considered as exclusion criteria.
Methods for spatially and temporally modelling two or more health outcomes, or the same health outcome in two or more subsets of the population at risk, are referred to as joint spatial and spatiotemporal methods [69].

2.3. Data Extraction

A data extraction template was developed in Microsoft Excel. The tool was developed considering the review question. Two authors (GAT and ZTT) extracted the data independently. When a disagreement appeared, it was resolved by a third author (AE). The data abstraction tools contained key information such as bibliographic information, research study objectives, the nature and type of data, covariate type, and data analysis methods, i.e., modelling approaches, key findings, and methodological gaps.
For each study, data such as the last name of the author, article title, name of the journal, year of publication, country, data source, spatial data type, outcomes of interest, number of outcomes, incidence/prevalence of outcomes, inference approaches, estimation techniques, study design, spatial unit, number of the spatial unit, temporal unit, number of temporal units, objectives, sample size, spatial model type, spatial structure, temporal structure, space-time interaction term, assigned priors, the reason for using joint spatial modelling, covariates used in the model, variable selection approach, number of covariate considered in the model, standardisation, spatial neighbourhood structure, temporal adjacency, software used, analysis method, model validation, model comparison measures, effect measure reported, key findings, map reported, script provided, and methodological gaps were extracted.

2.4. Risk of Bias Assessment

Two authors carried out a thorough assessment of the included studies’ methodological quality (GAT and ZTT). All included studies’ risk of bias was evaluated using a quality assessment tool that has an 8-point scoring system updated and modified to evaluate each study’s quality based on its aims and objectives, model validity, overall results, and study conclusion [70,71]. A standardised item list was used to grade the quality and risk of bias of included studies (Table S2). The checklist consists of 8 questions with possible answers ranging from 0 to 2, with a maximum overall score of 16. Low-, medium-, high-, and very high-quality levels were used to classify the overall score (low = score < 8, medium = score 8–10, high = score 11–13, and very high >13). Two authors (GAT and ZTT) independently assessed each study to determine its score and to determine the overall quality of the included studies. Discussion between the two authors attempted to settle any differences, and for those that could not be settled, a third reviewer (AE) was engaged.

2.5. Data Synthesis and Analysis

Microsoft Excel and STATA version 17 software were used for data entry and analysis, respectively. The review findings were summarised into texts, tables, and figures. The descriptive analysis was presented using the proportion, mean, medians, and ranges.

3. Results

3.1. Search Results and Characteristics of Included Studies

A comprehensive search of international peer-reviewed journals yielded a total of 4077 published articles. Of these, 4071 articles (PubMed: 2813, Embase: 103, PsycINFO: 24, Emcare: 45, Medline: 82, and Scopus: 1004) were obtained from 6 databases, and 6 more studies were found by manual searches in Google and Google Scholar. All these articles were published from 2011 onwards. During title and abstract screening, about 168 duplicates, 1278 non-relevant studies, and 2544 non-joint spatial studies were discarded, and only 87 studies were left for the full-text screening. A total of 44 studies were excluded from the 87 articles that met the inclusion and exclusion criteria for full-text screening; of these, 32 were not joint spatial models, 4 were methodological reviews, 3 were multivariable analyses, 2 were animal studies, 1 article was a duplication, 1 was a genetic study, and 1 was not a spatial study. The systematic review included 43 research publications that met the eligibility criteria (Table S3 and Figure 1).
Since 2011, the number of publications has fluctuated; it increased in 2016 and reached a peak in 2019 and 2020, then sharply dropped in 2021 before rising again in 2022. More ongoing research in this area is also anticipated with an increase in the number of clinical registries being set up. The majority of studies (n = 30, 69.78%) have been published after 2016 (Figure 2). Nearly one-fifth (18.6%) and 9.3% of the studies were conducted in Iran and the United States of America (USA), respectively (Figure 3). Of the 43 total studies, 15 (34.88%) were applied to infectious diseases, such as HIV/AIDS, herpes simplex virus-2, malaria, Zika, Leishmania, and hookworm, and 11 (25.58%) were applied to cancer, respectively. The International Journal of Environmental Research and Public Health (16.67%) and Spatial and Spatio-temporal Epidemiology (11.63%) were the most common journals where the articles were published (Table 1).

3.2. Data Source, Study Design, and Unit of Analysis

Seven studies (26.6%) used data from a national health survey or the Demographic and Health Survey (DHS) [43,46,51,60,64,72,75], while seven other studies (26.6%) used data from cancer registries. An estimated 11.63% of the studies used data from multiple surveys. More than one-third (41.86%) and 14 articles (32.56%) were ecological and cross-sectional studies, respectively. The majority (55.81%) of the studies analysed two outcomes simultaneously in the model. Twenty-one studies reported the prevalence/incidence of outcomes, and nearly half of them had a prevalence of less than 10%.
Given that spatial analysis can be conducted at different spatial scales, about 11 studies performed analyses at the provincial level, and seven studies performed analyses at the county level. The mean number of spatial units was 428, ranging from 11 to 3577 spatial units. Among 43 joint spatial studies considered for the systematic review, 17 employed a joint spatiotemporal model. Of them, the vast majority (n = 15, 88.23%) used year as the unit of analysis. The mean temporal period was 21.71, ranging from 5 to 260 (Table 2).

3.3. Spatial Data and Modelling Techniques

Only seven studies (n = 7, 16.28%) used point data, while the majority of the studies (n = 36, 83.72%) used areal data aggregated at a specific geographic unit, such as a municipality, province, county, Statistical Local Area (SLA), state, etc. to estimate the diseases’ relative risks [59,61,62,76,77,78,79]. The Bayesian estimation approach was used in more than three-quarters (n = 35, 81.40%) of the included studies [32,33,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,73,79]. For Bayesian inference, the MCMC estimation approach was utilised in 27 studies [32,33,35,36,37,38,39,40,41,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60], and for seven studies (16.28%), INLA was employed [42,43,61,62,63,64,65].
To analyse two or more health outcomes simultaneously, several joint spatial and spatiotemporal models were used. To investigate the relative risk of the study variables and their risk factors, joint spatial models were used in 24 studies [36,38,39,40,42,43,45,46,47,48,49,50,51,52,54,56,57,59,60,61,62,64,73,75], joint spatiotemporal models in 12 studies [32,33,35,37,41,44,55,58,63,65,80], and joint spatial and spatiotemporal autocorrelation methods such as Moran Index statistics, Local Indicator Spatial Analysis (LISA), Getis Ord Gi statistics, or Kulldroff spatial and Spatio-temporal scan statistical tests used in 7 studies [34,72,74,76,77,78,79].
In joint spatial temporal models, the structured and unstructured spatial random effects were considered to account for the spatial dependence and independent effects, respectively. In more than two-thirds (n = 26, 72.2%) of the studies, prior CAR was considered for the spatially structured random effect [32,33,35,36,37,38,39,40,41,43,44,46,47,48,49,50,51,52,53,55,58,60,61,63,64,65], and all assigned Identical and Independent Distribution (IID) for the unstructured spatial random effects. Out of 12 joint spatiotemporal studies, seven (58.23%) considered prior first-order random walk to account for the temporal dependence in the model [32,33,35,41,53,58,65]. The simple exchangeable hierarchical structure was taken into consideration for the spatiotemporal interaction terms in five of the joint spatial and temporal studies [32,33,41,53,65].
Of the 43 studies, 26 studies used R software [34,35,36,42,43,44,45,48,51,52,54,55,56,57,58,59,60,61,62,63,64,65,75,77,79,80], 21 studies used either WinBUGS or OpenBUGS or GeoBUGS [32,33,36,37,38,39,40,41,43,44,46,49,50,51,53,54,56,63,73,78,79], 7 studies used ArcGIS or QGIS [38,41,50,53,60,75,77], 4 studies used GeoDa [34,45,72,74], 3 studies used for SaTScan [76,77,80], and 2 studies used Fortran software [47,78]. Numerous joint spatial and spatiotemporal statistical techniques were used to examine spatial risk factors. Seven studies were joint spatial/spatiotemporal autocorrelation studies [34,48,72,74,76,77,80], the joint Bayesian shared Spatiotemporal model was used in four studies [42,53,60,65], and the multivariate Bayesian Spatiotemporal shared component model with Poisson distribution was used in two studies [33,37]. Other models included the bivariate Bayesian logit spatial model [46,51,63,64], geo-additive mixed models [57,75], and the multivariate negative binomial models with CAR random effects [43,80] (Table 3).

3.4. Covariates, Model Validation, and Goodness of Fit Assessment

Different measures of joint spatial and spatiotemporal model performance were reported. The majority (n = 22, 51.16%) of the studies considered Deviance Information Criteria (DIC) for model comparison [32,33,37,38,39,41,43,44,46,47,48,49,52,53,54,60,61,62,63,64,73,78], and six studies used Root Mean Predictive Squared Error (RMPSE) [35,37,44,58,59,78]. A combination of model comparison measures was used in many of the studies. The common effect measures reported in the included studies were relative risk (n = 17, 47.22%), odds ratio (n = 9, 25%), and coefficients (n = 8, 22.22%). Very limited studies underwent model validation (n = 5, 13.89%) [42,58,59,63,80].
Almost all joint spatial and spatiotemporal studies used maps to present the risk estimates. In the joint spatial and joint spatiotemporal models, socio-economic variables were the predominant variables among the covariates considered in the model to predict the outcomes across space or space-time [34,37,38,43,46,47,48,52,54,55,57,59,60,64,73,75]. Regarding standardisation, only five of the joint spatial and spatiotemporal models applied standardisation for common demographic variables [42,58,59,63,80]. Queen contiguity was the most commonly used method to define the neighbourhood structure (n = 10, 23.26%) [45,48,50,53,54,55,60,65,72,75] (Table 3).

3.5. Key Implications of Applying Joint Spatial Modelling, Findings, and Methodological Gaps

The justifications provided in the included studies for fitting the joint spatial and spatiotemporal model, shared component spatial and spatiotemporal model, or multivariate spatial and spatiotemporal model varied. Fourteen (38.89%) studies applied the shared component spatial and spatiotemporal model to consider the spatial dependence of interrelated outcome variables and to better explore their overlapping epidemiology [43,44,47,52,57,58,60,61,62,65,72,73,75,78]. Of the 36 joint spatial and spatiotemporal studies, 12 (33.33%) studies used the joint model for the ease of interpretation and to improve the precision of estimation [38,39,43,44,49,53,54,55,57,60,65,76]. Different reasons were provided for using the joint spatial and spatiotemporal analysis. Nine studies (25%) applied the joint spatial and spatiotemporal model to borrow strength between diseases and to incorporate data from a more common and related disease when interest was in a relatively rare disease, thereby strengthening the relevant results of the rare disease [36,37,42,44,49,53,55,60,61].
Out of 43 studies, 31 studies (72%) found reasonable patterns in the co-occurrence of health outcomes in geographic prevalence across areas [32,34,35,38,40,41,42,45,46,47,48,49,50,51,52,54,55,56,57,58,59,61,64,65,72,74,75,76,77,79,80]. The joint spatial and spatiotemporal model yields more precise and efficient estimates, especially when the number of observed events is rare [33,43,60,62,73,78]. Besides, the shared component joint spatial model had a better model fit relative to a joint spatial model without the shared component [36,46,53,62,73].
The studies included in this systematic review have self-reported methodological gaps. Seven studies acknowledged that in aggregated data, ecological fallacies are introduced, and some relevant information may be concealed by using large geographical units of study [40,47,48,49,53,65,80]. Therefore, using smaller units of analysis as a methodological gap may be a preferred approach. Four of the studies revealed that a meaningful number of temporal units is required to efficiently detect the temporal effect [38,41,44,80], and assuming the shared and specific components as independent ignores the possibility of interactions between the true covariates [38,44,62,73]. Three of the studies reported that MCMC has computational problems, model fitting, and convergence issues [42,43,56] (Table 4).

3.6. Assessment of Quality

Using the adapted quality assessment tool of modelling study qualities, the quality scores ranged from 0 to 16. The median quality score was 12/16, ranging from 8 to 16. Ten studies were classified as medium quality, twenty-one studies as high quality, and 12 studies as very high quality (Table 5).

4. Discussion

In this study, joint spatial and spatiotemporal models in health research were systematically reviewed. These models were mainly applied to infectious diseases [54,57,59,60,61,62,63,64,65,75,76,77,78,79,80], cancer [32,33,34,35,36,37,38,39,40,41,42], chronic diseases [44,48,51,58,72,73,74], and maternal and child health outcomes [46,52,53]. This showed that infectious diseases, which have a major worldwide burden and have the potential to spread to nearby areas, are currently receiving significant attention from researchers [81,82]. The spatial and spatiotemporal studies of interrelated infectious diseases provide a better understanding of the magnitude, pattern, overlapping epidemiology, and shared individual and area-level risk factors. The majority of infectious diseases co-occur in the same patients such as tuberculosis with HIV, herpes simplex virus-2 with HIV, leishmaniasis with malaria, etc. [83].
Joint spatial and spatiotemporal modelling of two or more cancers has become increasingly frequent over the past few decades [24,29,84] to examine the shared and differing trends of cancers regarding geographic patterns and shared risk factors. Contrary to univariate analysis, joint spatial models include shared components as various groups of cancers share common risk factors [85,86]. The majority of studies in spatial studies were based on single health outcomes, even though diseases such as cancer have common risk factors. Joint spatial and spatiotemporal modelling has recently become popular. It can model rare and common cancers to improve the estimates by borrowing strengths. It is feasible to modify behaviour common to cancers, and this has huge potential for preventing cancer. Major cancer risk factors can be altered by applying behavioural strategies including ceasing smoking, getting more exercise, managing weight, improving diet, limiting alcohol consumption, getting regular cancer screening tests, and limiting sun exposure [87]. Given that many of these cancer-preventive techniques lower the risk of many cancers, these might be supported by generating evidence through shared spatial and spatiotemporal studies.
The number of publications on joint spatial studies decreased dramatically during the coronavirus disease (COVID-19) period. It could be explained by the fact that since 2019, journals have prioritised COVID-19 research since little was known about the disease [88]. Evidence on the route of transmission, typical clinical features, underlying risk factors, and pathogenesis was limited, which is why the spatial analysis of COVID-19 with other infectious diseases was infrequent. In addition, advanced statistical approaches such as joint spatial and spatiotemporal models were not commonly applied as little was known about COVID-19’s clinical manifestation, route of transmission, etiology, and treatment.
More than one-third of the studies used ecological data for analysis. This demonstrates how disease mapping frequently used aggregated data at a specific geographic level to generate area-level estimates to guide healthcare decisions and effective resource allocation [89]. Compared with other studies, ecological data can be accessed or retrieved from reports quite easily. The main data sources were DHS or national health survey data [43,46,51,60,64,72,75] and cancer registry data [35,36,37,38,39,41,42]. This could be because of DHS and other national health surveys having location data (GPS) and geolocated covariates including environmental, pollution, demographic, and socio-economic covariates available in these surveys [90].
Some of the studies were exploratory spatial analyses, including joint spatial autocorrelation techniques such as Moran Index (MI) statistics, Local Indicator Spatial Analysis (LISA), Getis Ord Gi* statistics, or Kulldroff spatial or spatiotemporal scan statistical tests [34,72,74,76,77,78,79]. Unlike univariate spatial autocorrelations, multivariate spatial autocorrelations can determine how interrelated health outcomes such as TB and HIV, HSV-2 and HIV, and cancers influence each other spatially or spatiotemporally. They can explore the overlapping spatial and/or temporal distribution of two or more interrelated diseases. To detect the local clusters (hotspot and cold spot areas), spatial and space–time scanning statistical analysis and Getis-Ord Gi statistic are commonly used for cluster detections [91]. The hotspot and coldspot cluster detection is sensitive to the change in the size of the spatial and temporal units of analysis in which the data are aggregated; thus, analysing at a small spatial scale is preferable to identify hotspot areas efficiently. Multivariate spatial autocorrelation methods can investigate the spatial dependence of two or more interrelated health outcomes. In multivariate spatial autocorrelation analysis, the presence of a disease in a particular area is not only influenced by the prevalence of diseases in neighbouring areas, but also influenced by the presence of diseases that are related to one another in the area. However, they are unable to look at how the existence of one health outcome in one area may affect the spread of someone else in nearby areas, how one health outcome may affect the spread of another in adjacent areas, or, further, how it is affected by the spatial risk factors [92].
When variables were accounted for, the majority of the joint spatial and spatiotemporal models showed model improvement [32,34,37,38,43,45,46,47,48,51,52,53,54,55,57,59,60,61,64,73,75,79,80]. If the covariates were considered, joint spatial and spatiotemporal models performed much better than others and offered more insights than univariate models [93]. When the outcome is rare, the joint spatial models could improve the model performance by borrowing strength from interrelated diseases, neighbourhood areas, and/or time [10]. Besides, they can capture the spatial and spatiotemporal effects unexplained by the observed covariates by introducing random effects. This is well-suited to the data that have limited predictors and models that capture few covariates.
Only five studies underwent standardisation for common demographic variables [38,40,48,55,61]. Demographic covariates such as age, sex, race, etc. are the most obvious risk factors for almost all health and health-related conditions. In the univariate spatial analysis, the difference in incidence or mortality of given diseases because of age, sex, or other variables can be addressed by standardisation, which ignores the effects of these variables in the analysis. Age, sex, and other demographic variables are determinants for a multitude of health-related and other outcomes. The impacts of age, sex, and other common risk factors should be estimated using these variables in the model, as well as their interactions with other factors, such as spatial and temporal effects.
The vast majority of the studies were conducted based on a Bayesian modelling framework [32,33,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,73,79]. This was in line with the advancement of Bayesian statistics in the field of disease mapping. Bayesian spatial models are now commonly applied because of advances in knowledge of advanced statistics, programming, coding skills, and improved computer power resources that overcome computational problems. This methodology can provide more reliable area-level estimates specifically when the cases are rare or the population is small [90]. It can smoothen the observed extreme estimates towards the global mean or the neighbourhood values and provide more robust estimates, specifically when studied areas have sparse populations, by assigning prior distributions to define spatial structure to ensure the closer areas have more contribution than distant areas [22,94].
The majority of the Bayesian spatial and spatiotemporal studies undertook model fitting and inference based on MCMC [32,33,35,36,37,38,39,40,41,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60] followed by INLA [42,43,61,62,63,64,65]. The most popular models for spatial studies were Bayesian hierarchical models with structured and unstructured random effects. The MCMC approach is computationally very demanding and has convergence issues [95]. This is especially true for hierarchical models, which by their very nature make MCMC convergence unpredictable and slow. No matter the model, it is necessary to assess the convergence of posterior samples because there is no guarantee that such models can be easily fitted, in which additional simulations and model simplification will be necessary. In contrast, INLA has recently emerged as a reliable alternative method for fitting Bayesian spatial models that overcomes MCMC’s drawbacks [96,97]. The package uses INLA to estimate Bayesian models without the need for posterior sampling techniques. In practical terms, numerical integration is performed via this approximation; therefore, it does not require a lot of iterative processing [98]. Bayesian estimation utilising the INLA methodology typically takes substantially less time than MCMC, which is the reason why this package was developed for spatial statistics in the first place.
The majority of studies applied joint spatial models [36,38,39,40,42,43,45,46,47,48,49,50,51,52,54,56,57,59,60,61,62,64,73,75] followed by joint spatiotemporal models [32,33,35,37,41,44,55,58,63,65,80]. Unlike spatial autocorrelation studies, the spatial and spatiotemporal models can investigate the covariates influencing the distribution of the outcome across space and/or time [33,37,42,43,53,57,60,65,75,80]. The joint spatial and spatiotemporal models are applied for spatial analysis of two or more health outcomes or one outcome across different population groups. They typically rely on Generalised Linear Mixed Models (GLMMs) that consider shared and specific spatial, temporal, and spatiotemporal random effects. The data type, distribution, nature, and incidence of the outcomes determine the type of joint spatial and spatiotemporal models, e.g., poisson or negative binomial GLMM for count data and logistic regression for categorical outcomes. The dependence between diseases with similar spatial or temporal patterns is captured by prior distributions [13]. The multivariate CAR models smoothen noisy estimates and leverage information from nearby areas and interrelated diseases to predict spatially autocorrelated area-level disease risks [99]. However, multivariate CAR models are unable to show how the correlation of outcomes varies across space. The Copula geoadditive model overcomes this limitation and can demonstrate the change in the association between outcomes across geographic locations [100]. However, it is unable to detect the geographic areas contributing to higher or lower risk of simultaneous occurrence of multiple outcomes. Recently, the majority of studies applied the shared spatial and spatiotemporal models that decompose the spatial effect into shared and disease-specific spatial effects [24].
The majority of studies used a prior CAR for the spatially structured random effects to account for the spatial dependence [32,33,35,36,37,38,39,40,41,43,44,46,47,48,49,50,51,52,53,55,58,60,61,63,64,65], and all assigned IID for the unstructured spatial random effects. MCAR models incorporate both spatially structured and unstructured random effects in the model. The spatial dependence among adjacent areas is accounted for by assuming a CAR process in the random effects [101]. Most studies used first-order random walk prior to temporal random effects [32,33,35,41,53,58,65].
The authors’ reasons for fitting joint spatial and spatiotemporal models over univariate models were the following: to considering the spatial dependence of related outcomes in the model and better explore their overlapping epidemiology [43,44,47,52,57,58,60,61,62,65,72,73,75,78]; to improve estimation precision [38,39,43,44,49,53,54,55,57,60,65,76]; and to strengthen relationships between diseases by borrowing data from a more prevalent and related disease when the disease of interest is relatively rare [36,37,42,44,49,53,55,60,61]. In addition, the studies revealed that the shared component models yield more precise and efficient estimates, especially when the disease is rare or the population is small [33,43,60,62,73,78]. Moreover, incorporating the shared component in the model could improve the model’s performance [36,46,53,62,73].
This review pointed out several recommendations for the development of improved joint spatial and spatiotemporal models. Some studies acknowledged that when data are aggregated, ecological fallacies are introduced, and some relevant information may be concealed by using large geographical units of study [40,47,48,49,53,65,80]. Thus, using smaller units of analysis increases the precision of the estimates. Assuming the shared and specific components as independent denies the possibility of interactions between the true covariates [38,41,44,80], and a relevant number of temporal units is necessary to efficiently identify the temporal effect [38,44,62,73]. Moreover, some of the studies showed that MCMC has computational problems, model fitting, and convergence issues [42,43,56]. INLA is better for developing statistical models to obtain efficient risk estimations and direct the efficient distribution of medical interventions.
DIC was the most commonly used model comparison criteria to measure and compare the model goodness of fit and model complexity [32,33,37,38,39,41,43,44,46,47,48,49,52,53,54,60,61,62,63,64,73,78], followed by RMSPE [35,37,44,58,59,78]. Most of the studies used a combination of goodness-of-fit measures for model assessment. DIC and WAIC are model performance measures that are calculated by combining the model likelihood function (deviance (-2LLR)) and a model complexity term (number of effective parameters) [102,103]. In addition to model performance assessment, model accuracy assessment such as mean absolute prediction error and mean square prediction error are considered for model comparison [104]. Moreover, local measures of fit such as the conditional predictive ordinate are also used for making a model comparison. Apart from three studies [34,44,77], most studies reported maps for the visualisation of risk estimates. Maps offer epidemiologists enough evidence to display spatial risk and/or risk factors across time and/or space. It can give decision-makers motivation, insight, and assist potential health interventions in high-risk areas.
The scientific community may benefit from the epidemiological and statistical insights this systematic study provides in terms of joint spatial and spatiotemporal model applications in health research. First, the utility of joint spatial and spatiotemporal models is more pronounced in large data registries and when multiple interrelated diseases are fitted simultaneously. It is therefore crucial to estimate the smoothed relative risk of lower-prevalence cases through the borrowing of strength from the related cases and neighbourhood areas. Although there were joint spatial and spatiotemporal studies, the systematic review found heterogeneity in methods of estimation technique, statistical models, prior selections, defining adjacencies, and model complexities. This showed that a consistent framework for undertaking joint spatial and spatiotemporal models is needed. This framework is currently a focus of our research program. This systematic review provides insight suggesting that jointly modelling two or more cases that have shared characteristics is better to detect clusters of cases specifically when the number of cases is rare, such as in rare cancers and orphan diseases, or when the population is small. Another important finding was that the most complex models (joint spatial and spatiotemporal models incorporating covariates and interaction) performed very well. Overall, the systematic review identified several areas of improvement in joint spatial studies such as providing data, maps, scripts, and methodological gaps.
This review has some strengths and limitations, including an extensive search of six electronic databases to retrieve studies in an area without a previous systematic review. Careful title/abstract and full-text screening was carried out with predefined inclusion and exclusion criteria. One of the limitations of this review was that the majority of the studies were from a few countries, which might be because of spatial data being limited (GPS data and software packages for joint spatial and spatiotemporal model) or insufficient funding or statistical skills, indicating the possibility of publication bias or a focus of research effort on countries included in research publications. Another limitation was that only articles published in English were considered, so we may have excluded valuable contributions.

5. Conclusions

Multivariate disease mapping is crucial for understanding the burden of interrelated health outcomes over space and/or time. Numerous joint spatial and spatiotemporal methodologies aiming to explore the spatial risk of two or more health outcomes simultaneously were reviewed. The majority of studies used Bayesian methods, which handled a wider range of variance components at different levels in the model and could consider model uncertainties to provide reliable estimates. The most often utilised covariates in joint spatial and spatiotemporal models were socio-economic and demographic. Most of the reviewed studies used shared component spatial and spatiotemporal models with a Poisson-based and negative binomial modelling approach. Relatively few studies have been published on the applications of joint spatial and spatiotemporal models since the COVID-19 pandemic. Reviewed studies have acknowledged that aggregated data are liable to ecological fallacies and some relevant information may be concealed by using large geographical units of study. Therefore, this systematic review highlighted the need for future joint spatial and spatiotemporal models to analyse correlated health outcomes to guide decision-making for effective prevention and control strategies.

Supplementary Materials

The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/ijerph20075295/s1, Table S1: Databases and searching terms; Table S2: Risk of bias tool for assessment; Table S3: List of publications included for systematic review.

Author Contributions

Conceptualisation: G.A.T., Z.T.T., S.H., R.G.S. and A.E. Data curation: G.A.T., Z.T.T., S.H., R.G.S. and A.E. Methodology: G.A.T., Z.T.T., S.H., R.G.S. and A.E. Writing—original draft: G.A.T., Z.T.T., S.H., R.G.S. and A.E. Writing—review and editing: G.A.T., Z.T.T., S.H., R.G.S. and A.E. All authors have read and agreed to the published version of the manuscript.

Funding

This systematic review did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

This study is a systematic review of already published articles and the extracted data are provided as a supplementary file.

Acknowledgments

The authors would like to thank Monash University library services.

Conflicts of Interest

The authors have no conflict of interest with this paper.

References

  1. Krieger, N. Place, space, and health: GIS and epidemiology. Epidemiology 2003, 14, 384–385. [Google Scholar] [CrossRef]
  2. Best, N.; Hansell, A.L. Geographic variations in risk: Adjusting for unmeasured confounders through joint modeling of multiple diseases. Epidemiology 2009, 20, 400. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  3. Werneck, G.L. Georeferenced data in epidemiologic research. Ciência Saúde Coletiva 2008, 13, 1753–1766. [Google Scholar] [CrossRef]
  4. Sui, D.Z. Tobler’s first law of geography: A big idea for a small world? Ann. Assoc. Am. Geogr. 2004, 94, 269–277. [Google Scholar] [CrossRef]
  5. Leitner, M.; Glasner, P.; Kounadi, O. Laws of geography. In Oxford Research Encyclopedia of Criminology and Criminal Justice; Oxford University Press: New York, NY, USA, 2018. [Google Scholar]
  6. Andrienko, G.; Andrienko, N.; Demsar, U.; Dransch, D.; Dykes, J.; Fabrikant, S.I.; Jern, M.; Kraak, M.-J.; Schumann, H.; Tominski, C. Space, time and visual analytics. Int. J. Geogr. Inf. Sci. 2010, 24, 1577–1600. [Google Scholar] [CrossRef] [Green Version]
  7. Pfeiffer, D.U.; Robinson, T.P.; Stevenson, M.; Stevens, K.B.; Rogers, D.J.; Clements, A.C. Spatial Analysis in Epidemiology; OUP: Oxford, UK, 2008. [Google Scholar]
  8. Waller, L.A.; Carlin, B.P.; Xia, H.; Gelfand, A.E. Hierarchical spatio-temporal mapping of disease rates. J. Am. Stat. Assoc. 1997, 92, 607–617. [Google Scholar] [CrossRef]
  9. Lawson, A.B. Hierarchical modeling in spatial epidemiology. Wiley Interdiscip. Rev. Comput. Stat. 2014, 6, 405–417. [Google Scholar] [CrossRef]
  10. Beale, L.; Abellan, J.J.; Hodgson, S.; Jarup, L. Methodologic issues and approaches to spatial epidemiology. Environ. Health Perspect. 2008, 116, 1105–1110. [Google Scholar] [CrossRef] [Green Version]
  11. Lindgren, F.; Rue, H. Bayesian spatial modelling with R-INLA. J. Stat. Softw. 2015, 63, 1–25. [Google Scholar] [CrossRef] [Green Version]
  12. Louzada, F.; do Nascimento, D.C.; Egbon, O.A. Spatial statistical models: An overview under the Bayesian approach. Axioms 2021, 10, 307. [Google Scholar] [CrossRef]
  13. Lee, D. A comparison of conditional autoregressive models used in Bayesian disease mapping. Spat. Spatio-Temporal Epidemiol. 2011, 2, 79–89. [Google Scholar] [CrossRef]
  14. Fix, M.J.; Cooley, D.S.; Thibaud, E. Simultaneous autoregressive models for spatial extremes. Environmetrics 2021, 32, e2656. [Google Scholar] [CrossRef]
  15. De Oliveira, V. Bayesian analysis of conditional autoregressive models. Ann. Inst. Stat. Math. 2012, 64, 107–133. [Google Scholar] [CrossRef]
  16. Besag, J.; York, J.; Mollié, A. Bayesian image restoration, with two applications in spatial statistics. Ann. Inst. Stat. Math. 1991, 43, 1–20. [Google Scholar] [CrossRef]
  17. Leroux, B.G.; Lei, X.; Breslow, N. Estimation of disease rates in small areas: A new mixed model for spatial dependence. In Statistical Models in Epidemiology, the Environment, and Clinical Trials; Springer: Berlin/Heidelberg, Germany, 2000; pp. 179–191. [Google Scholar]
  18. Goodchild, M.F.; Longley, P.A. Geographic information science. In Handbook of Regional Science; Springer: Berlin/Heidelberg, Germany, 2021; pp. 1597–1614. [Google Scholar]
  19. Ancelet, S.; Abellan, J.J.; Del Rio Vilas, V.J.; Birch, C.; Richardson, S. Bayesian shared spatial-component models to combine and borrow strength across sparse disease surveillance sources. Biom. J. 2012, 54, 385–404. [Google Scholar] [CrossRef]
  20. Dabney, A.R.; Wakefield, J.C. Issues in the mapping of two diseases. Stat. Methods Med. Res. 2005, 14, 83–112. [Google Scholar] [CrossRef]
  21. Botella-Rocamora, P.; Martinez-Beneito, M.A.; Banerjee, S. A unifying modeling framework for highly multivariate disease mapping. Stat. Med. 2015, 34, 1548–1559. [Google Scholar] [CrossRef]
  22. Corberán-Vallet, A. Prospective surveillance of multivariate spatial disease data. Stat. Methods Med. Res. 2012, 21, 457–477. [Google Scholar] [CrossRef] [Green Version]
  23. Finley, A.O.; Banerjee, S.; Gelfand, A.E. spBayes for large univariate and multivariate point-referenced spatio-temporal data models. arXiv 2013, arXiv:1310.8192. [Google Scholar]
  24. Knorr-Held, L.; Best, N.G. A shared component model for detecting joint and selective clustering of two diseases. J. R. Stat. Soc. Ser. A 2001, 164, 73–85. [Google Scholar] [CrossRef]
  25. Blangiardo, M.; Cameletti, M. Spatial and Spatio-Temporal Bayesian Models with R-INLA; John Wiley & Sons: Hoboken, NJ, USA, 2015. [Google Scholar]
  26. Chang, C.-D.; Wang, C.-C.; Jiang, B.C. Using data mining techniques for multi-diseases prediction modeling of hypertension and hyperlipidemia by common risk factors. Expert Syst. Appl. 2011, 38, 5507–5513. [Google Scholar] [CrossRef]
  27. Wong, C.W.; Wong, T.Y.; Cheng, C.-Y.; Sabanayagam, C. Kidney and eye diseases: Common risk factors, etiological mechanisms, and pathways. Kidney Int. 2014, 85, 1290–1302. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  28. Meijers, W.C.; de Boer, R.A. Common risk factors for heart failure and cancer. Cardiovasc. Res. 2019, 115, 844–853. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  29. Held, L.; Natário, I.; Fenton, S.E.; Rue, H.; Becker, N. Towards joint disease mapping. Stat. Methods Med. Res. 2005, 14, 61–82. [Google Scholar] [CrossRef] [Green Version]
  30. Cressie, N.; Wikle, C.K. Statistics for Spatio-Temporal Data; John Wiley & Sons: Hoboken, NJ, USA, 2015. [Google Scholar]
  31. Gómez-Rubio, V.; Palmí-Perales, F.; López-Abente, G.; Ramis-Prieto, R.; Fernández-Navarro, P. Bayesian joint spatio-temporal analysis of multiple diseases. SORT-Stat. Oper. Res. Trans. 2019, 43, 51–74. [Google Scholar]
  32. Mahaki, B.; Mehrabi, Y.; Kavousi, A.; Schmid, V.J. Joint spatio-temporal shared component model with an application in Iran Cancer Data. Asian Pac. J. Cancer Prev. APJCP 2018, 19, 1553. [Google Scholar]
  33. Ahmadipanahmehrabadi, V.; Hassanzadeh, A.; Mahaki, B. Bivariate spatio-temporal shared component modeling: Mapping of relative death risk due to colorectal and stomach cancers in Iran provinces. Int. J. Prev. Med. 2019, 10, 39. [Google Scholar]
  34. Bermudi, P.M.M.; Pellini, A.C.G.; Rebolledo, E.A.S.; Diniz, C.S.G.; Aguiar, B.S.d.; Ribeiro, A.G.; Failla, M.A.; Baquero, O.S.; Chiaravalloti-Neto, F. Spatial pattern of mortality from breast and cervical cancer in the city of São Paulo. Rev. De Saúde Pública 2020, 54, 142. [Google Scholar] [CrossRef]
  35. Carroll, R.; Lawson, A.B.; Faes, C.; Kirby, R.S.; Aregay, M.; Watjou, K. Extensions to multivariate space time mixture modeling of small area cancer data. Int. J. Environ. Res. Public Health 2017, 14, 503. [Google Scholar] [CrossRef] [Green Version]
  36. Chamanpara, P.; Moghimbeigi, A.; Faradmal, J.; Poorolajal, J. Joint disease mapping of two digestive cancers in Golestan Province, Iran using a shared component model. Osong Public Health Res. Perspect. 2015, 6, 205–210. [Google Scholar] [CrossRef] [Green Version]
  37. Cramb, S.M.; Baade, P.D.; White, N.M.; Ryan, L.M.; Mengersen, K.L. Inferring lung cancer risk factor patterns through joint Bayesian spatio-temporal analysis. Cancer Epidemiol. 2015, 39, 430–439. [Google Scholar] [CrossRef] [Green Version]
  38. Mahaki, B.; Mehrabi, Y.; Kavousi, A.; Akbari, M.E.; Waldhoer, T.; Schmid, V.J.; Yaseri, M. Multivariate disease mapping of seven prevalent cancers in Iran using a shared component model. Asian. Pac. J. Cancer Prev. 2011, 12, 2353–2358. [Google Scholar]
  39. Nasrazadani, M.; Maracy, M.R.; Dreassi, E.; Mahaki, B. Mapping of stomach, colorectal, and bladder cancers in Iran, 2004–2009: Applying Bayesian polytomous logit model. Int. J. Prev. Med. 2018, 9, 104. [Google Scholar]
  40. Stoppa, G.; Mensi, C.; Fazzo, L.; Minelli, G.; Manno, V.; Consonni, D.; Biggeri, A.; Catelan, D. Spatial analysis of shared risk factors between pleural and ovarian cancer mortality in Lombardy (Italy). Int. J. Environ. Res. Public Health 2022, 19, 3467. [Google Scholar] [CrossRef]
  41. Raei, M.; Schmid, V.J.; Mahaki, B. Bivariate spatiotemporal disease mapping of cancer of the breast and cervix uteri among Iranian women. Geospat. Health 2018, 13, 645. [Google Scholar] [CrossRef]
  42. Asmarian, N.; Ayatollahi, S.M.T.; Sharafi, Z.; Zare, N. Bayesian spatial joint model for disease mapping of zero-inflated data with R-INLA: A simulation study and an application to male breast cancer in Iran. Int. J. Environ. Res. Public Health 2019, 16, 4460. [Google Scholar] [CrossRef] [Green Version]
  43. Adeyemi, R.A.; Zewotir, T.; Ramroop, S. Joint spatial mapping of childhood anemia and malnutrition in sub-Saharan Africa: A cross-sectional study of small-scale geographical disparities. Afr. Health Sci. 2019, 19, 2692–2712. [Google Scholar] [CrossRef]
  44. Baker, J.; White, N.; Mengersen, K.; Rolfe, M.; Morgan, G.G. Joint modelling of potentially avoidable hospitalisation for five diseases accounting for spatiotemporal effects: A case study in New South Wales, Australia. PLoS ONE 2017, 12, e0183653. [Google Scholar] [CrossRef] [Green Version]
  45. Besharati, M.M.; Kashani, A.T.; Li, Z.; Washington, S.; Prato, C.G. A bivariate random effects spatial model of traffic fatalities and injuries across Provinces of Iran. Accid. Anal. Prev. 2020, 136, 105394. [Google Scholar] [CrossRef]
  46. Kazembe, L.N.; Kandala, N.-B. Estimating areas of common risk in low birth weight and infant mortality in Namibia: A joint spatial analysis at sub-regional level. Spat. Spatio-Temporal Epidemiol. 2015, 12, 27–37. [Google Scholar] [CrossRef]
  47. Kline, D.; Hepler, S.; Bonny, A.; McKnight, E. A joint spatial model of opioid-associated deaths and treatment admissions in Ohio. Ann. Epidemiol 2019, 33, 19–23. [Google Scholar] [CrossRef] [PubMed]
  48. Kramer, M.R.; Williamson, R. Multivariate Bayesian spatial model of preterm birth and cardiovascular disease among Georgia women: Evidence for life course social determinants of health. Spat. Spatio-Temporal Epidemiol. 2013, 6, 25–35. [Google Scholar] [CrossRef] [PubMed]
  49. Law, J.; Perlman, C. Exploring geographic variation of mental health risk and service utilization of doctors and hospitals in Toronto: A shared component spatial modeling approach. Int. J. Environ. Res. Public Health 2018, 15, 593. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  50. Law, J.; Quick, M.; Jadavji, A. A Bayesian spatial shared component model for identifying crime-general and crime-specific hotspots. Ann. GIS 2020, 26, 65–79. [Google Scholar] [CrossRef] [Green Version]
  51. Lawson, A.; Schritz, A.; Villarroel, L.; Aguayo, G.A. Multi-Scale Multivariate Models for Small Area Health Survey Data: A Chilean Example. Int. J. Environ. Res. Public Health 2020, 17, 1682. [Google Scholar] [CrossRef] [Green Version]
  52. Neelon, B.; Anthopolos, R.; Miranda, M.L. A spatial bivariate probit model for correlated binary data with application to adverse birth outcomes. Stat. Methods Med. Res. 2014, 23, 119–133. [Google Scholar] [CrossRef]
  53. Odhiambo, J.N.; Sartorius, B. Joint spatio-temporal modelling of adverse pregnancy outcomes sharing common risk factors at sub-county level in Kenya, 2016–2019. BMC Public Health 2021, 21, 1–13. [Google Scholar] [CrossRef]
  54. Okango, E.; Mwambi, H.; Ngesa, O.; Achia, T. Semi-parametric spatial joint modeling of HIV and HSV-2 among women in Kenya. PLoS ONE 2015, 10, e0135212. [Google Scholar] [CrossRef]
  55. Ransome, Y.; Subramanian, S.V.; Duncan, D.T.; Vlahov, D.; Warren, J. Multivariate spatiotemporal modeling of drug- and alcohol-poisoning deaths in New York City, 2009–2014. Spat. Spatio-Temporal Epidemiol. 2020, 32, 100306. [Google Scholar] [CrossRef]
  56. Norwood, T.A.; Encisa, C.; Wang, X.; Seliske, L.; Cunningham, J.; De, P. A Bayesian shared components modeling approach to develop small area indicators of social determinants of health with measures of uncertainty. Can. J. Public Health 2020, 111, 342–357. [Google Scholar] [CrossRef]
  57. Adebayo, S.B.; Gayawan, E.; Heumann, C.; Seiler, C. Joint modeling of Anaemia and Malaria in children under five in Nigeria. Spat. Spatio-Temporal Epidemiol. 2016, 17, 105–115. [Google Scholar] [CrossRef]
  58. Huang, G.; Lee, D.; Scott, E.M. Multivariate space-time modelling of multiple air pollutants and their health effects accounting for exposure uncertainty. Stat. Med. 2018, 37, 1134–1148. [Google Scholar] [CrossRef] [Green Version]
  59. Kang, J.; Zhang, N.; Shi, R. A Bayesian nonparametric model for spatially distributed multivariate binary data with application to a multidrug-resistant tuberculosis (MDR-TB) study. Biometrics 2014, 70, 981–992. [Google Scholar] [CrossRef] [Green Version]
  60. Roberts, D.J.; Zewotir, T. Shared component modelling of early childhood anaemia and malaria in Kenya, Malawi, Tanzania and Uganda. BMC Pediatr. 2022, 22, 1–11. [Google Scholar] [CrossRef]
  61. Carabali, M.; Schmidt, A.M.; Restrepo, B.N.; Kaufman, J.S. A joint spatial marked point process model for dengue and severe dengue in Medellin, Colombia. Spat. Spatio-Temporal Epidemiol. 2022, 41, 100495. [Google Scholar] [CrossRef]
  62. Kinyoki, D.K.; Manda, S.O.; Moloney, G.M.; Odundo, E.O.; Berkley, J.A.; Noor, A.M.; Kandala, N.B. Modelling the ecological comorbidity of acute respiratory infection, diarrhoea and stunting among children under the age of 5 years in Somalia. Int. Stat. Rev. 2017, 85, 164–176. [Google Scholar] [CrossRef] [Green Version]
  63. Lawson, A.B.; Carroll, R.; Castro, M. Joint spatial Bayesian modeling for studies combining longitudinal and cross-sectional data. Stat. Methods Med. Res. 2014, 23, 611–624. [Google Scholar] [CrossRef] [Green Version]
  64. Orunmoluyi, O.S.; Gayawan, E.; Manda, S. Spatial Co-Morbidity of Childhood Acute Respiratory Infection, Diarrhoea and Stunting in Nigeria. Int. J. Environ. Res. Public Health 2022, 19, 1838. [Google Scholar] [CrossRef]
  65. Otiende, V.A.; Achia, T.N.; Mwambi, H.G. Bayesian hierarchical modeling of joint spatiotemporal risk patterns for Human Immunodeficiency Virus (HIV) and Tuberculosis (TB) in Kenya. PLoS ONE 2020, 15, e0234456. [Google Scholar] [CrossRef]
  66. Page, M.J.; McKenzie, J.E.; Bossuyt, P.M.; Boutron, I.; Hoffmann, T.C.; Mulrow, C.D.; Shamseer, L.; Tetzlaff, J.M.; Akl, E.A.; Brennan, S.E. The PRISMA 2020 statement: An updated guideline for reporting systematic reviews. Syst. Rev. 2021, 10, 1–11. [Google Scholar] [CrossRef]
  67. Lawson, A.B. Statistical Methods in Spatial Epidemiology; John Wiley & Sons: Hoboken, NJ, USA, 2013. [Google Scholar]
  68. Elliot, P.; Wakefield, J.C.; Best, N.G.; Briggs, D.J. Spatial Epidemiology: Methods and Applications; Oxford University Press: Oxford, UK, 2000. [Google Scholar]
  69. Rezaeian, M.; Dunn, G.; St Leger, S.; Appleby, L. Geographical epidemiology, spatial analysis and geographical information systems: A multidisciplinary glossary. J. Epidemiol. Community Health 2007, 61, 98–102. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  70. Fone, D.; Hollinghurst, S.; Temple, M.; Round, A.; Lester, N.; Weightman, A.; Roberts, K.; Coyle, E.; Bevan, G.; Palmer, S. Systematic review of the use and value of computer simulation modelling in population health and health care delivery. J. Public Health 2003, 25, 325–335. [Google Scholar] [CrossRef] [PubMed]
  71. Harris, R.C.; Sumner, T.; Knight, G.M.; White, R.G. Systematic review of mathematical models exploring the epidemiological impact of future TB vaccines. Hum. Vaccines Immunother. 2016, 12, 2813–2832. [Google Scholar] [CrossRef] [PubMed]
  72. Darikwa, T.B.; Manda, S.O. Spatial co-clustering of cardiovascular diseases and select risk factors among adults in South Africa. Int. J. Environ. Res. Public Health 2020, 17, 3583. [Google Scholar] [CrossRef] [PubMed]
  73. Roberts, D.J.; Zewotir, T. Copula geoadditive modelling of anaemia and malaria in young children in Kenya, Malawi, Tanzania and Uganda. J. Health Popul. Nutr. 2020, 39, 8. [Google Scholar] [CrossRef]
  74. Desjardins, M.; Whiteman, A.; Casas, I.; Delmelle, E. Space-time clusters and co-occurrence of chikungunya and dengue fever in Colombia from 2015 to 2016. Acta Trop. 2018, 185, 77–85. [Google Scholar] [CrossRef]
  75. Freitas, L.P.; Carabali, M.; Yuan, M.; Jaramillo-Ramirez, G.I.; Balaguera, C.G.; Restrepo, B.N.; Zinszer, K. Spatio-temporal clusters and patterns of spread of dengue, chikungunya, and Zika in Colombia. medRxiv 2022, 16, e0010334. [Google Scholar] [CrossRef]
  76. Schur, N.; Gosoniu, L.; Raso, G.; Utzinger, J.; Vounatsou, P. Modelling the geographical distribution of co-infection risk from single-disease surveys. Stat. Med. 2011, 30, 1761–1776. [Google Scholar] [CrossRef]
  77. Stensgaard, A.-S.; Vounatsou, P.; Onapa, A.W.; Simonsen, P.E.; Pedersen, E.M.; Rahbek, C.; Kristensen, T.K. Bayesian geostatistical modelling of malaria and lymphatic filariasis infections in Uganda: Predictors of risk and geographical patterns of co-endemicity. Malar. J. 2011, 10, 298. [Google Scholar] [CrossRef] [Green Version]
  78. Chidumwa, G.; Maposa, I.; Kowal, P.; Micklesfield, L.K.; Ware, L.J. Bivariate joint spatial modeling to identify shared risk patterns of hypertension and diabetes in south africa: Evidence from WHO SAGE South Africa Wave 2. Int. J. Environ. Res. Public Health 2021, 18, 359. [Google Scholar] [CrossRef]
  79. Adegboye, O.; Al-Saghir, M.; LEUNG, D.H. Joint spatial time-series epidemiological analysis of malaria and cutaneous leishmaniasis infection. Epidemiol. Infect. 2017, 145, 685–700. [Google Scholar] [CrossRef] [Green Version]
  80. Darikwa, T.B.; Manda, S.; Lesaoana, M. Assessing joint spatial autocorrelations between mortality rates due to cardiovascular conditions in South Africa. Geospat. Health 2019, 14. [Google Scholar] [CrossRef]
  81. Bhutta, Z.A.; Sommerfeld, J.; Lassi, Z.S.; Salam, R.A.; Das, J.K. Global burden, distribution, and interventions for infectious diseases of poverty. Infect. Dis. Poverty 2014, 3, 21. [Google Scholar] [CrossRef] [Green Version]
  82. Murray, C.J.; Lopez, A.D. Measuring the global burden of disease. N. Engl. J. Med. 2013, 369, 448–457. [Google Scholar] [CrossRef] [Green Version]
  83. Bruchfeld, J.; Correia-Neves, M.; Källenius, G. Tuberculosis and HIV coinfection. Cold Spring Harb. Perspect. Med. 2015, 5, a017871. [Google Scholar] [CrossRef]
  84. Richardson, S.; Abellan, J.J.; Best, N. Bayesian spatio-temporal analysis of joint patterns of male and female lung cancer risks in Yorkshire (UK). Stat. Methods Med. Res. 2006, 15, 385–407. [Google Scholar] [CrossRef]
  85. Lewandowska, A.M.; Rudzki, M.; Rudzki, S.; Lewandowski, T.; Laskowska, B. Environmental risk factors for cancer-review paper. Ann. Agric. Environ. Med. 2018, 26, 1–7. [Google Scholar] [CrossRef]
  86. Stein, C.; Colditz, G. Modifiable risk factors for cancer. Br. J. Cancer 2004, 90, 299–303. [Google Scholar] [CrossRef] [Green Version]
  87. De Pergola, G.; Silvestris, F. Obesity as a major risk factor for cancer. J. Obes. 2013, 2013, 291546. [Google Scholar] [CrossRef] [Green Version]
  88. Morens, D.M.; Breman, J.G.; Calisher, C.H.; Doherty, P.C.; Hahn, B.H.; Keusch, G.T.; Kramer, L.D.; LeDuc, J.W.; Monath, T.P.; Taubenberger, J.K. The origin of COVID-19 and why it matters. Am. J. Trop. Med. Hyg. 2020, 103, 955. [Google Scholar] [CrossRef]
  89. Best, N.; Richardson, S.; Thomson, A. A comparison of Bayesian spatial models for disease mapping. Stat. Methods Med. Res. 2005, 14, 35–59. [Google Scholar] [CrossRef] [PubMed]
  90. Rutstein, S.O.; Rojas, G. Guide to DHS statistics. Calverton MD ORC Macro 2006, 38, 78. [Google Scholar]
  91. Kulldorff, M.; Mostashari, F.; Duczmal, L.; Katherine Yih, W.; Kleinman, K.; Platt, R. Multivariate scan statistics for disease surveillance. Stat. Med. 2007, 26, 1824–1833. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  92. Scrucca, L. Clustering multivariate spatial data based on local measures of spatial autocorrelation. Quad. Del Dip. Di Econ. Finanz. E Stat. 2005, 20, 11. [Google Scholar]
  93. Vicente, G.; Goicoa, T.; Ugarte, M. Bayesian inference in multivariate spatio-temporal areal models using INLA: Analysis of gender-based violence in small areas. Stoch. Environ. Res. Risk Assess. 2020, 34, 1421–1440. [Google Scholar] [CrossRef]
  94. Besag, J. Spatial interaction and the statistical analysis of lattice systems. J. R. Stat. Soc. Ser. B 1974, 36, 192–225. [Google Scholar] [CrossRef]
  95. Mossel, E.; Vigoda, E. Limitations of Markov chain Monte Carlo algorithms for Bayesian inference of phylogeny. Ann. Appl. Probab. 2006, 16, 2215–2234. [Google Scholar] [CrossRef] [Green Version]
  96. Gómez-Rubio, V. Bayesian Inference with INLA; CRC Press: Boca Raton, FL, USA, 2020. [Google Scholar]
  97. Held, L.; Schrödle, B.; Rue, H. Posterior and cross-validatory predictive checks: A comparison of MCMC and INLA. In Statistical Modelling and Regression Structures; Springer: Berlin/Heidelberg, Germany, 2010; pp. 91–110. [Google Scholar]
  98. Carroll, R.; Lawson, A.; Faes, C.; Kirby, R.S.; Aregay, M.; Watjou, K. Comparing INLA and OpenBUGS for hierarchical Poisson modeling in disease mapping. Spat. Spatio-Temporal Epidemiol. 2015, 14, 45–54. [Google Scholar] [CrossRef] [Green Version]
  99. Lawson, A.B.; Biggeri, A.; Böhning, D.; Lesaffre, E.; Viel, J.-F.; Clark, A.; Schlattmann, P.; Divino, F. Disease mapping models: An empirical evaluation. Disease Mapping Collaborative Group. Stat. Med. 2000, 19, 2217–2241. [Google Scholar]
  100. Li, J. Application of Copulas as a New Geostatistical Tool; Universität Stuttgart: Stuttgart, Germany, 2010. [Google Scholar]
  101. Morris, M.; Wheeler-Martin, K.; Simpson, D.; Mooney, S.J.; Gelman, A.; DiMaggio, C. Bayesian hierarchical spatial models: Implementing the Besag York Mollié model in stan. Spat. Spatio-Temporal Epidemiol. 2019, 31, 100301. [Google Scholar] [CrossRef]
  102. Duncan, E.W.; Mengersen, K.L. Comparing Bayesian spatial models: Goodness-of-smoothing criteria for assessing under-and over-smoothing. PLoS ONE 2020, 15, e0233019. [Google Scholar] [CrossRef]
  103. Spiegelhalter, D.J.; Best, N.G.; Carlin, B.P.; Van der Linde, A. The deviance information criterion: 12 years on. J. R. Stat. Soc. Ser. B 2014, 76, 485–493. [Google Scholar] [CrossRef]
  104. Aregay, M.; Lawson, A.B.; Faes, C.; Kirby, R. Bayesian multiscale modeling for aggregated disease mapping data. In Proceedings of the Third ACM SIGSPATIAL International Workshop on the Use of GIS in Public Health, Dallas, TX, USA, 4 November 2014; pp. 45–48. [Google Scholar]
Ijerph 20 05295 g001 550
Figure 1. Flow chart of selection of studies for the systematic review using PRISMA checklists from 2011–2022.
Figure 1. Flow chart of selection of studies for the systematic review using PRISMA checklists from 2011–2022.
Ijerph 20 05295 g001
Ijerph 20 05295 g002 550
Figure 2. Number of studies based on year of publication from 2011 to 2022.
Figure 2. Number of studies based on year of publication from 2011 to 2022.
Ijerph 20 05295 g002
Ijerph 20 05295 g003 550
Figure 3. Number of included articles by country of study.
Figure 3. Number of included articles by country of study.
Ijerph 20 05295 g003
Table
Table 1. General characteristics of the included studies.
Table 1. General characteristics of the included studies.
Characteristics Frequency Percentage (%)References
Study category
Cancer 1125.58[32,33,34,35,36,37,38,39,40,41,42]
Chronic diseases716.28[44,48,51,58,72,73,74]
Infectious diseases1534.88[54,57,59,60,61,62,63,64,65,75,76,77,78,79,80]
Health service utilisation12.33[49]
Maternal and child health outcomes36.98[46,52,53]
Others *613.95[43,45,47,50,55,56]
Publication journal
International Journal of Environmental Research and Public Health716.67[35,40,49,51,64,72,73]
Spatial and Spatio-temporal Epidemiology511.63[46,48,55,57,61]
PLOS ONE36.98[44,54,65]
Statistics in Medicine26.65[58,78]
Statistical Methods in Medical Research24.65[52,63]
BMC Public Health12.32[53]
Malaria journal12.32[79]
Epidemiology and infection12.32[80]
Annuals of GIS12.32[50]
Geospatial Health12.32[74]
International Journal of Preventive Medicine24.65[33,39]
International Statistical Review12.32[62]
African Health Sciences12.32[43]
Journal of Health, Population, and Nutrition12.32[75]
Others **1227.91[32,34,36,37,38,45,47,56,59,60,76,77]
Others *: Trauma, injury, mental health, drug and substance use: Others **: BMC Paediatrics, Acta Tropica, The International Journal of Cancer Epidemiology, Detection, and Prevention, Osong Public Health and Research Perspectives, Accident Analysis and Prevention, Rev Saude Publica (RSP), Asian Pacific Journal of Cancer Prevention, Canadian Journal of Public Health, Biometrics.
Table
Table 2. Data, study design, and covariates of the reviewed studies.
Table 2. Data, study design, and covariates of the reviewed studies.
Item CategoryNumber Percentage (%)References
Data source(s)DHS or National health survey716.28[43,46,51,60,64,72,75]
Malaria indicator survey36.98[57,63,75]
HMIS/DHIS24.65[53,80]
Death and cause of death registration system 12.33[33]
Multiple surveys511.63[37,38,47,65,80]
Hospital records24.65[44,59]
AIDS indicator survey 12.33[54]
Cancer registry 716.28[35,36,37,38,39,41,42]
Others *2353.49[32,40,46,47,48,49,50,51,52,53,54,55,56,58,61,62,63,73,74,76,77,78,79]
Study design (More than one design was applied in some of the studies)Ecological1841.86[32,33,34,38,39,40,47,48,49,52,55,58,61,73,74,76,77,79]
Cross-sectional 1432.56[43,46,51,54,56,57,60,62,63,64,72,75,78,80]
Retrospective 920.93[35,36,37,41,42,44,45,59,65]
Longitudinal 24.65[51,53]
Others **24.65[49,50]
Number of outcomes of the study 22455.81[33,34,36,37,40,41,45,46,48,49,50,52,54,55,58,60,61,65,72,73,75,76,78,79,80]
3716.28[35,39,43,62,64,74,77]
424.65[51,53]
524.65[44,56]
600----
724.65[32,38]
Prevalence of outcomes of the study All less than 10%1125.58[33,34,35,36,37,41,42,46,52,53,61]
Either of them is less than 10%49.30[48,51,79,80]
All greater than 10%613.95[43,45,60,62,73,75]
Not reported2251.16[32,38,39,40,44,47,49,50,54,55,56,57,58,59,63,65,72,74,76,77,78]
Spatial unit Provinces 1125.58[32,33,35,36,38,39,41,43,45,51,80]
County716.28[42,47,48,52,53,54,65]
Municipalities 49.30[40,74,76,77]
Districts 36.98[60,72,73,75]
Schools/Health facility36.98[34,78,79]
SLAs12.33[37]
Not reported12.33[59]
Others ***1227.91[44,46,49,50,55,56,57,58,61,62,63,64]
Temporal units (n = 17)Year1588.23[32,33,34,35,37,41,44,45,53,58,62,65,74,76]
Month 15.88[80]
Weeks15.88[77]
Others *: Food Security and Nutrition Unit, NASA, ABS, Urban Malaria Control Program (UMCP), Others **: Survey reports, Others ***: residence, constituents, point, cluster, state, ZIP code, dissemination area, intermediate geography, ten cell count, neighbourhood.
Table
Table 3. Details of the types, structures, and methods of joint spatial models of included studies (n = 43).
Table 3. Details of the types, structures, and methods of joint spatial models of included studies (n = 43).
ItemsNumber Percentage (%)References
Types of spatial data
Point716.38[59,61,62,76,77,78,79]
Area3683.72[32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,58,60,63,64,65,72,73,74,75]
Methods of inference
Frequentist 818.60[34,72,74,75,76,77,78,80]
Bayesian 3581.40[32,33,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,73,79]
Estimation techniques (n = 36)
ML24.65[75,80]
MCMC2762.79[32,33,35,36,37,38,39,40,41,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60]
INLA716.28[42,43,61,62,63,64,65]
Joint spatial analysis techniques
Joint spatial autocorrelation analysis716.28[34,72,74,76,77,78,79]
Joint spatial models2455.81[36,38,39,40,42,43,45,46,47,48,49,50,51,52,54,56,57,59,60,61,62,64,73,75]
Joint Spatio-temporal models1227.91[32,33,35,37,41,44,55,58,63,65,80]
Spatial structure (n = 36)
MCAR/BCAR/ ICAR/CAR2672.22[32,33,35,36,37,38,39,40,41,43,44,46,47,48,49,50,51,52,53,55,58,60,61,63,64,65]
SAR12.78[45]
GMRF38.33[42,54,57]
Not reported 616.67[56,59,62,73,75,80]
Temporal structure (n = 12)
Prior first-order random walk 758.33[32,33,35,41,53,58,65]
log-linear structure 18.33[44]
Prior first-order autoregressive 216.66[37,55]
Second-order random walk18.33[63]
Not reported18.33[80]
Spatio-temporal term (n = 12)
Uncorrelated ST interaction term18.33[35]
Simple exchangeable hierarchical
Structure		541.67[32,33,41,53,65]
First order autoregressive18.33[58]
Not reported541.67[37,44,55,63,80]
The software’s used
R/R-studio/R2WinBUGS/R-INLA2660.47[34,35,36,42,43,44,45,48,51,52,54,55,56,57,58,59,60,61,62,63,64,65,75,77,79,80]
ArcGIS/QGIS716.28[38,41,50,53,60,75,77]
WinBUGS/OpenBUGS/GeoBUGS2148.84[32,33,36,37,38,39,40,41,43,44,46,49,50,51,53,54,56,63,73,78,79]
GeoDA49.30[34,45,72,74]
SaTScan36.98[76,77,80]
Fortran/MATLAB24.65[47,78]
Spatial models used (n = 36)
A multivariate negative binomial model with CAR random effects25.56[43,80]
Multivariate Bayesian Spatio-temporal shared component model with Poisson distribution25.56[33,37]
Poisson generalised linear mixed model (GLMM) with a shared spatial component with the log-linear temporal trend12.78[44]
Multivariate spatial autocorrelation and hotspot analysis719.44[34,48,72,74,76,77,80]
Joint spatial marked point processes model with Poisson distribution12.78[61]
Bayesian multivariate ST mixture model12.78[35]
Bivariate bayesian logit spatial model411.11[46,51,63,64]
Bayesian hierarchical geostatistical shared component model/ Bivariate bayesian geostatistical logistic model25.56[62,78]
A bayesian multivariate conditional auto-regressive model with Poisson distribution12.78[48]
Bayesian spatial Polytomous Logit Model12.78[39]
Bayesian spatial biprobit model12.78[52]
Joint bayesian Spatio-temporal shared component binomial model/Bayesian joint hierarchical Spatio-temporal Log-linear model/Bayesian shared component model411.11[42,53,60,65]
Bayesian semi-parametric spatial joint model/Bayesian nonparametric model using Gaussian processes for the analysis of spatially distributed multivariate binary outcome25.56[54,59]
Geoadditive mixed model25.56[57,75]
Bayesian geostatistical shared component multinomial modelling12.78[78]
Bayesian ANOVA12.78[56]
Model validation (n = 36)
No 3186.11[32,33,35,36,37,38,39,40,41,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,60,61,62,64,65,73,75]
Yes 513.89[42,58,59,63,80]
Model comparison metrics (n = 36)
DIC2251.16[32,33,37,38,39,41,43,44,46,47,48,49,52,53,54,60,61,62,63,64,73,78]
WAIC49.30[35,40,55,62]
CPO24.65[37,62]
PIT12.33[62]
RMSPE/Mean absolute error613.95[35,37,44,58,59,78]
KL12.33[40]
Credible interval plot12.33[78]
Bayesian p-value and L-criterion12.33[37]
Others * (AIC, BIC)24.65[75,79]
Effect measure reported (n = 36)
OR925.00[43,46,60,61,62,63,64,73,79]
RR1747.22[32,33,36,37,38,39,40,41,42,47,48,49,50,53,54,55,65]
Coefficient822.22[35,44,45,51,52,57,58,59,80]
Covariates (n = 36)
Demographic1438.89[43,46,47,51,52,54,57,59,60,61,64,73,75,79]
Socio-economical1644.44[34,37,38,43,46,47,48,52,54,55,57,59,60,64,73,75]
Environmental 616.67[37,45,60,75,79,80]
Clinical, health service, and behavioral related 616.67[32,43,46,53,57,73]
Standardisation (n = 36)
No3186.11[32,33,35,36,37,39,41,42,43,44,45,46,47,49,50,51,52,53,54,56,57,58,59,60,62,63,64,65,73,75,80]
Yes513.89[38,40,48,55,61]
Method to define spatial neighbourhood structure
Distance-based neighbourhood matrix12.33[34]
Queen contiguity1023.26[45,48,50,53,54,55,60,65,72,75]
Rook contiguity24.65[46,52]
Non-specified adjacency based 36.98[33,44,49]
Not reported2762.79[32,35,36,37,38,39,40,41,42,43,47,51,56,57,58,59,61,62,63,64,73,74,76,77,78,79,80]
Map reported
No 36.98[34,44,77]
Yes 4093.02[32,33,35,36,37,38,39,40,41,42,43,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,72,73,74,75,76,78,79,80]
Script provided (n = 36)
No 3186.11[32,33,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,55,56,57,58,59,60,61,62,64,73,75,80]
Yes 513.89[35,53,54,63,65]
AIC: Akaike Information Criteria, ANOVA: Analysis of Variance, BIC: Bayesian Information Criteria, CPO: Conditional Predictive Ordinate, DIC: Deviance Information Criteria, KL: Kullback Leibler Divergence, OR: Odds Ratio, PIT: Probability Integral Transform, RMSPE: Root Mean square Predictive Error, RR: Relative Risk, WAIC: Watanabe Akaike Information Criteria.
Table
Table 4. Summary of the purpose of fitting joint spatial model, key findings, and reported methodological gaps of selected studies.
Table 4. Summary of the purpose of fitting joint spatial model, key findings, and reported methodological gaps of selected studies.
Items Number Percentage (%)References
Reasons for using joint modelling (n = 36)
To borrow strength between diseases and to incorporate data from a more common and related disease when interest is in a relatively rare disease strengthens the relevant results of the rare disease925.00[36,37,42,44,49,53,55,60,61]
For ease of interpretation, and to improve the precision of estimation1233.33[38,39,43,44,49,53,54,55,57,60,65,76]
To consider the spatial dependence of interrelated outcome variables and to better understand the overlapping epidemiology1438.89[43,44,47,52,57,58,60,61,62,65,72,73,75,78]
To account for such unmeasured exposures that may be common among the diseases 25.56[37,44]
For estimating the relative weight of each shared component for all related disease616.67[38,41,50,53,65,73]
Key findings
The joint spatial model yields more precise and efficient estimates especially when the number of desired observed cases is low 613.95[33,43,60,62,73,78]
Found reasonable patterns in the co-occurrence in geographic prevalence across areas3172.09[32,34,35,38,40,41,42,45,46,47,48,49,50,51,52,54,55,56,57,58,59,61,64,65,72,74,75,76,77,79,80]
They had shared risk factors.716.28[37,39,44,53,60,72,80]
The shared component joint spatial model had a better model fit relative to a joint spatial model without the shared component511.63[36,46,53,62,73]
Methodological gaps (n = 36)
A meaningful time period is required to detect the temporal effects411.11[38,41,44,80]
Assuming the shared and specific components as independent ignores the possibility of interactions between the true covariates411.11[38,44,62,73]
Edge effects38.33[36,38,77]
The results are biased by the Modifiable Areal Unit Problem (MAUP)25.56[48,55]
Aggregation of the data has the effect of introducing ecological fallacy and large geographical units of analysis may mask some information of interest. Results and efficiency may be improved by having smaller units of analysis719.44[40,47,48,49,53,65,80]
MCMC has a computational problems, model fitting, and convergence issues38.33[42,43,56]
Table
Table 5. Quality assessment of included studies.
Table 5. Quality assessment of included studies.
IDAuthorYearAaO SaP MS MM PRDQoD PoR IDORSumRating
1Freitas et al., 2022 [75]20222211221213High
2Kazembe et al., 2015 [46]20152212222215Very high
3Kinyoki et al., 2017 [62]20172211211212High
4Besharati et al., 2020 [45]20202212211213High
5Kramer et al., 2013 [48]2013111111129Medium
6Law et al., 2018 [49]20182211211212High
7Lawson et al., 2014 [63]20142221202112High
8Lawson et al., 2020 [51]2020111122109Medium
9Mahaki et al., 2011 [38] 20112221222114Very high
10Mahaki et al., 2018 [32] 20182211212112High
11Nasrazadani et al., 2018 [39]20182221222215Very high
12Desjardins et al., 2014 [76]20182211212112High
13Odhiambo et al., 2021 [53] 20212222212215Very high
14Okango et al., 2015 [54]20152211221213High
15Orunmoluyi et al., 2022 [64]20222211221213High
16Otiende et al., 2020 [65]20202222222216Very high
17Raei et al., 2018 [41]20182211111110Medium
18Ransome et al., 2019 [55]20192211121111High
19Roberts et al., 2020 [75]20202222222216Very high
20Schur et al., 2011 [78]20112212221113High
21Stensgaard et al., 2011 [79]20111221111110Medium
22Stoppa et al., 2022 [40]20221222121112High
23Norwood et al., 2020 [56]20202222221114Very high
24Adebayo et al., 2016 [57]20162222222115Very high
25Asmarian et al., 2019 [42]20192212211112High
26Huang et al., 2018 [58]20182112222113High
27Kang et al., 2014 [59]20142222222115Very high
28Law et al., 2020 [50]20202222222115Very high
29Roberts et al., 2022 [60]20222212211112High
30Carabali et al., 2022 [61] 20222222121113High
31Cramb et al., 2015 [37]20152222221114Very high
32Kinyoki et al., 2017 [62]20172211211111High
33Kline et al., 2019 [47]20192211221112High
34Chidumwa et al., 2021 [73] 2021111111118Medium
35Adeyemi et al., 2019 [43] 20192211221112High
36Darikwa et al., 2019 [74]20192211221112High
37Darikwa et al., 2020 [51]2020121111119Medium
38Chamanpara et al., 2015 [36]2015111111118Medium
39Carroll et al., 2017 [35]20172211111110Medium
40Adegboye et al., 2017 [80]2017111111118Medium
41Neelon et al., 2014 [52]2014111111118Medium
42Ahmadipanahmehrabadi et al., 2019 [33]20192212221113High
43Bermudi et al., 2020 [34]20202222222216Very high
Range 1–21–21–21–21–20–21–20–28–16
Median score2211221112High
Mean score1.791.841.351.421.721.531.331.2812.26
AaO, aims and objectives; SaP, setting and population; MS, model structure; MM, modelling methods; PRDS, parameter ranges and data sources; QoD, quality of data; PoR, presentation of results; IDoR, interpretation, and discussion of results.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Tesema, G.A.; Tessema, Z.T.; Heritier, S.; Stirling, R.G.; Earnest, A. A Systematic Review of Joint Spatial and Spatiotemporal Models in Health Research. Int. J. Environ. Res. Public Health 2023, 20, 5295. https://doi.org/10.3390/ijerph20075295

AMA Style

Tesema GA, Tessema ZT, Heritier S, Stirling RG, Earnest A. A Systematic Review of Joint Spatial and Spatiotemporal Models in Health Research. International Journal of Environmental Research and Public Health. 2023; 20(7):5295. https://doi.org/10.3390/ijerph20075295

Chicago/Turabian Style

Tesema, Getayeneh Antehunegn, Zemenu Tadesse Tessema, Stephane Heritier, Rob G. Stirling, and Arul Earnest. 2023. "A Systematic Review of Joint Spatial and Spatiotemporal Models in Health Research" International Journal of Environmental Research and Public Health 20, no. 7: 5295. https://doi.org/10.3390/ijerph20075295

APA Style

Tesema, G. A., Tessema, Z. T., Heritier, S., Stirling, R. G., & Earnest, A. (2023). A Systematic Review of Joint Spatial and Spatiotemporal Models in Health Research. International Journal of Environmental Research and Public Health, 20(7), 5295. https://doi.org/10.3390/ijerph20075295

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop