Access to this full-text is provided by Springer Nature.
Content available from International Journal of Health Geographics
This content is subject to copyright. Terms and conditions apply.
Alvesetal.
International Journal of Health Geographics (2023) 22:8
https://doi.org/10.1186/s12942-023-00329-4
RESEARCH Open Access
© The Author(s) 2023. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which
permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the
original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or
other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line
to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this
licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/. The Creative Commons Public Domain Dedication waiver (http:// creat iveco
mmons. org/ publi cdoma in/ zero/1. 0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
International Journal of
Health Geographics
Uncovering COVID-19 infection determinants
inPortugal: towardsanevidence-based spatial
susceptibility index tosupport epidemiological
containment policies
André Alves1* , Nuno Marques da Costa1,2 , Paulo Morgado1,2 and Eduarda Marques da Costa1,2
Abstract
Background COVID-19 caused the largest pandemic of the twenty-first century forcing the adoption of containment
policies all over the world. Many studies on COVID-19 health determinants have been conducted, mainly using multi-
variate methods and geographic information systems (GIS), but few attempted to demonstrate how knowing social,
economic, mobility, behavioural, and other spatial determinants and their effects can help to contain the disease. For
example, in mainland Portugal, non-pharmacological interventions (NPI) were primarily dependent on epidemiologi-
cal indicators and ignored the spatial variation of susceptibility to infection.
Methods We present a data-driven GIS-multicriteria analysis to derive a spatial-based susceptibility index to COVID-19
infection in Portugal. The cumulative incidence over 14 days was used in a stepwise multiple linear regression as the
target variable along potential determinants at the municipal scale. To infer the existence of thresholds in the relation-
ships between determinants and incidence the most relevant factors were examined using a bivariate Bayesian change
point analysis. The susceptibility index was mapped based on these thresholds using a weighted linear combination.
Results Regression results support that COVID-19 spread in mainland Portugal had strong associations with factors
related to socio-territorial specificities, namely sociodemographic, economic and mobility. Change point analysis
revealed evidence of nonlinearity, and the susceptibility classes reflect spatial dependency. The spatial index of sus-
ceptibility to infection explains with accuracy previous and posterior infections. Assessing the NPI levels in relation to
the susceptibility map points towards a disagreement between the severity of restrictions and the actual propensity
for transmission, highlighting the need for more tailored interventions.
Conclusions This article argues that NPI to contain COVID-19 spread should consider the spatial variation of the sus-
ceptibility to infection. The findings highlight the importance of customising interventions to specific geographical
contexts due to the uneven distribution of COVID-19 infection determinants. The methodology has the potential for
replication at other geographical scales and regions to better understand the role of health determinants in explain-
ing spatiotemporal patterns of diseases and promoting evidence-based public health policies.
*Correspondence:
André Alves
andrejoelalves@campus.ul.pt
Full list of author information is available at the end of the article
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 2 of 16
Alvesetal. International Journal of Health Geographics (2023) 22:8
Keywords COVID-19, Health determinants, GIS, Multicriteria decision analysis, Non-pharmacological interventions,
Spatial-based policies, Spatiotemporal analysis
Introduction
e severe acute respiratory syndrome coronavirus 2
(SARS-CoV-2), responsible for the new coronavirus
disease (COVID-19), has caused the biggest pandemic
of the twenty-first century. Although the mortality
rate is considerably lower in comparison to previous
coronavirus epidemics, COVID-19 has a higher trans-
mission rate [1] that forced the adoption of restrictive
measures to contain human-to-human transmission,
known as non-pharmacological interventions (NPI) [2].
e distribution of confirmed cases of COVID-19
had an uneven spread because the incidence of new
infections was characterized by spatiotemporal het-
erogeneity at multiple scales [3]. e spatial patterns
can be explained by multiple factors [4–10] that justify
spatial variations in contagion exposure, vulnerability
and susceptibility [11–18]. Different NPI management
strategies, ranging from case isolation to comprehen-
sive measures, are also explanatory of COVID-19 spa-
tiotemporal variability [10, 19, 20]. Furthermore, the
literature highlights the importance of spatial depend-
ence stemming from geographical properties, such as
proximity and contiguity to more-prone outbreak areas
[21, 22].
Although there is already considerable literature
devoted to identifying the determinants of COVID-19
infection and their effect on spatial patterns, with high
methodological diversity [23], many of these studies do
not summarize their evidence in a way that can be use-
ful and integrated with public health measures for pan-
demic control. As stated by van Schalkwyk and McKee
[24] there have been “challenges of translating knowl-
edge into policy”. In public health and disease preven-
tion, the use of spatial models tends to increase, with
the growing availability and accessibility of data on
disease incidence with higher granularity, leveraged by
the need for heterogeneous territorially based public
health policies [25, 26]. erefore, in the current pan-
demic context is of the utmost importance the imple-
mentation of spatiotemporal surveillance systems that
prioritize interventions in areas of higher infection
risk [27] and a better incorporation of social factors
into COVID-19 models can improve predictive accu-
racy for more tailored and effective responses [28]. Due
to the uneven distribution associated with exposure
to SARS-CoV-2 a spatial dimension is crucial [29]. In
this perspective, estimating the spatial susceptibility
and vulnerability in health-related subjects is essential
to prevent disease spread [30–32] since knowledge of
the distribution of susceptible individuals allows for the
assessment of multiple susceptibilitylevels [33].
In Epidemiology, susceptibility (to a disease) is under-
stood as “the dynamic state of being more likely or
liable to be harmed by a health determinant” [34]. Nev-
ertheless, it is often used as a synonym for vulnerability,
although the latter incorporates, beyond the position of
relative disadvantage understood as the propensity to
be adversely affected, the capacity for adaptation and
resilience [34, 35]. Literature about the study of the
unequal spatial propensity to COVID-19 infection can
be found using both terms for the same type of analy-
sis. In this paper, susceptibility was conceptualized in
line with the definition of Porta [34]. From a methodo-
logical perspective our approach measures the relation-
ship between the confirmed cases of the disease and the
effect of indicators—e.g., determinants—in explaining
the incidence patterns. is type of analysis is not only
informative for public health policies targeted to differ-
ent population groups [36, 37] but also essential in epi-
demic contexts to manage early warning systems [38,
39]. e classification of territorial units by their pro-
pensity to infection can be used for equity in pandemic
and public health policies avoiding one-size-fits-all
containment measures in favour of geographically-tai-
lored interventions in areas more prone to diffusion
[40, 41]. In this respect, spatial analysis and GIS have
proved to be essential [23, 42].
In the case of Portugal, evidence-based knowledge
about the existence of geographical contexts that are
more favourable to transmission and outbreaks has
been shown and highlighted by several authors [14,
43–45]. e spread of COVID-19 in the country has
been associated with settlement patterns, transport
networks, mobility behaviours, employment and other
economic and social characteristics [8, 46–48]. How-
ever, indicators regarding the causes of the spread of
the disease have not been properly integrated to serve
as policy guidance in assisting public health decision-
makers. erefore, NPI management in Portugal has
resulted exclusively from epidemiological indicators,
ignoring social, economic and mobility information
useful in differentiating local strategies. is comes of
relevance because the inclusion of auxiliary informa-
tion is crucial to model the disease [28] and identify-
ing viral hotspots where lockdowns are most effective,
or less transmission-prone areas where NPIs can be
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 3 of 16
Alvesetal. International Journal of Health Geographics (2023) 22:8
eased [49]. Similarly, spatial analysis can help analyze
policy effects on transmission spatial dynamics. For the
portuguese territory, Sá Marques etal. [14] suggested
the need for territorial customized NPI, proposing a
geographic mosaic based on a vulnerability risk index,
while Pereira et al. [45] developed a risk conceptual
model to monitor COVID-19 spatiotemporal dynamics.
is work explores the hypothesis of developing a
municipal index of susceptibility to COVID-19 infection,
for mainland Portugal, to serve as a basis for the adoption
of NPI tailored to territorial specificities. e aims of this
article are threefold: (i) identify significant determinants
of COVID-19 infection for the first year of the disease;
(ii) derive an infection susceptibility index that classifies
municipalities from thresholds; (iii) assess the relation-
ship between the susceptibility index and the incidence
rate per population for tailored NPI. Overall, we deliver
a proposal for a NPI spatial-based modelling framework,
based on infection susceptibility and epidemiological
data, to assist policy design and decision-making.
e study area is mainland Portugal at municipal scale.
Worth to say, that the municipality (278 units) is the most
disaggregated spatial unit with official data on COVID-
19. As a southern European country, the continental
territory had about 9.8 million inhabitants in 2021, seem-
ingly peripheral to Europe but in a hub position between
continents, it is an interesting case study because of the
very disparate evolution of the number of cases and the
spatial diffusion patterns. While in the first waves the
timely containment ensured low incidence and low mor-
tality, unlike in nearby countries such as Spain and Italy
[50], in later periods the ineffectiveness of containment
policies led to it becoming the country in the world with
the highest COVID-19 incidence per inhabitant.
Despite vaccination campaigns, NPI remain important
to contain SARS-CoV-2 outbreaks [51–53]. NPI have
been adopted throughout the world to contain COVID-
19 transmission and strategies varied [2, 54]. Considering
as extremes the “China COVID zero policy” [20] on one
hand and the laissez-faire Sweden approach [55] on the
other, containment policies in Portugal can be considered
as an intermediate approach, in balancing public health
and economy. NPI were managed based on epidemiologi-
cal monitoring but followed unclear criteria with contra-
dictory decisions and lack of rationality during the first
months, with a quasi-national scope as a “one size fits
all”. After November 2020, a new paradigm began with
measures depending on a risk threshold classification by
the Directorate-General of Health (DGS), that catego-
rized municipalities from the 14 day-cases per 100,000
inhabitants to define NPI at the municipal scale (Coun-
cil of Ministers Resolution no 92-A/2020, November
2). Each category was associated with a set of NPI with
harshness proportional to incidence. is risk classifica-
tion consisted exclusively of the disease incidence, ignor-
ing mortality and hospitalizations, and did not effectively
represent the epidemiological risk, that is, “the probabil-
ity of an adverse or beneficial event in a defined popula-
tion over a specified time interval” [34].
Even though this later approach relied on known for-
mal criteria and was spatial-based, it did not fit munici-
palities with small populations. For comparison purposes
(Table 1) in Manteigas—a rural municipality with less
than 3000 inhabitants—seven new cases were enough to
exceed the first risk threshold, even though the contact
tracing and isolation were simple. On the opposite way,
the city of Lisbon—the capital of Portugal with more than
500 thousand inhabitants—could have more than 1300
cases, which is already community transmission, and not
yet surpass the first risk threshold. erefore, the lack of
adequacy of this approach undermined timely contain-
ment in some cases while in others was excessively harsh.
Following subsequent readjustments, the risk thresholds
for low-density municipalities were changed. However,
despite this improvement, some municipalities remained
to be subject to a criterion with an excessively high value,
resulting in challenges to containment. is way, the
portuguese NPI approach was characterized by the late
implementation of measures, particularly in more popu-
lated municipalities, in a reactive rather than preventive
way, and although it was spatially based, it ignored the
spatial variation of susceptibility.
Methodological steps
e methodology applied to answer the objectives fol-
lowed several steps (Fig. 1). Briefly, a multiple linear
regression (MLR) was performed to reduce the dimen-
sionality of a set of potential determinants of COVID-19
infection and obtain their influences in explaining pat-
terns. ereafter a Bayesian change point analysis (CPA)
was applied to detect thresholds as changing points in the
relationships between COVID-19 incidence and the most
relevant factors, allowing the classification of municipali-
ties accordingly to the susceptibility associated with each
determinant. Afterwards, a weighted linear combination
Table 1 Comparison of risk thresholds and population numbers
across municipalities
Municipality Population (2021) Number of 14-day cases to
exceed the rst risk threshold
(after readjustment)
Lisbon 545,923 1311
Coimbra 140,838 339
Manteigas 2909 7 (14)
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 4 of 16
Alvesetal. International Journal of Health Geographics (2023) 22:8
ensured the conjugation into a composite susceptibility
index. e DGS risk levels and the susceptibility index
were compared using the Spearman and contingency
coefficients.
Given the need to uncover explanatory variables for
COVID-19 spatiotemporal patterns, we supported the
analysis using regression. Linear, generalized, mixed
multi-level, non-linear and geographically based meth-
ods have been used for regression analysis to understand
COVID-19 spatial dynamics and establish relationships
with factors [5, 6, 8, 9, 46, 56, 57]. e choice of linear
regression over more tuned methods is essentially due
to two reasons. First, urner etal. [58] revealed that at
most periods the COVID-19 infection curves of various
countries entered linear growth phases, due partly to the
effect of containment measures. Second, linearity tends
to be lost (resulting in the famous S-curve) only when
working with accumulated data, which was not the case
because the periods modelled in this work corresponded
to accumulations of 14days which are relatively short
and can be accommodated by a linear curve.
Regarding the use of a method for detecting change
points, it has long been recognized that thresholds play
a crucial role in understanding the spread of infectious
diseases [59, 60]. is type of technique has precedents
in COVID-19 modelling [61], however we are unaware of
studies that rely on it to derive information for a suscep-
tibility index.
Data acquisition andtreatment
e relationship between COVID-19 cases and their
spatial determinants was performed in an aggregated
data structure, i.e., an ecological analysis, whose
explanatory variables were selected based on a litera-
ture review on the determinants of COVID-19 infec-
tion. ese potential factors, ranging from indicators
of urban density, employment by sector, to commuting
patterns, were grouped into dimensions. Environmen-
tal and climatic data, used in some studies [7, 46] were
not considered because defining a value that reflects
the municipality’s reality would always revolve around
simplification and bias. Furthermore, there is no con-
sensus on the significance of these variables as predic-
tors, resulting in conflicting findings in the literature,
and normally less relevant than socioeconomic deter-
minants [62].
e data used has multiple sources. A total of 51
potential determinant factors (Table2) were consid-
ered from Statistics Portugal (https:// www. ine. pt/) and
Social Chart (https:// www. carta social. pt/). e epide-
miological information (number of cases) was obtained
from the COVID-19 situation reports of DGS [63] for
6 periods. e periods under analysis correspond to
14-day blocks of new cases of the disease, representa-
tive of the beginning and the peak of the first three
waves of COVID-19 in Portugal between March 2020
and March 2021 (Fig.2).
To avoid scale effects, the absolute values of the origi-
nal variables were swapped into rates, proportions, and
location quotients. To ensure that the linear regres-
sion’s normality assumption was met, data transfor-
mation [64, 65] was applied to both epidemiological
information and determinant factors using the square
root transformation, a common nonlinear fix used in
epidemiological data analysis [66, 67].
Fig. 1 Methodological framework
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 5 of 16
Alvesetal. International Journal of Health Geographics (2023) 22:8
GIS-multicriteria susceptibility analysis
e approach proposed in this paper to derive a territo-
rial differentiation of susceptibility to COVID-19 infec-
tion used thresholds. Following a multicriteria decision
analysis, we assumed susceptibility conceptually as
the definition of Porta [34] and methodologically as
thelikelihood of confirmed cases occurring in relation
to the determinants, similar to Sarkar [16].
Other studies of susceptibility or vulnerability anal-
ysis to COVID-19 in GIS have favoured the use of
multicriteria analysis based on the analytic hierar-
chy process [16, 68, 69]. In these knowledge-based
approaches, there is a subjective influence on the rela-
tive importance of factors. In contrast, data-driven
approaches based on multivariate models enable
parametrizations that are based on the sensitive analy-
sis of factors without the impact of subjectivity [70].
Identifying determinants ofinfection
e identification of determinant factors explaining the
incidence patterns of COVID-19 was based on an MLR.
For this purpose, the epidemiological data and the 51
potential determinants (Table 2) were considered as
follows:
where
Yi
represents the estimated number of COVID-19
cases for the period
i
,
β0
is the intercept of the regres-
sion line,
Xi
are the explanatory factors,
βp
are the coef-
ficients for each variable and
εi
is the mode’s error term.
Yi
=
β
0+
β
1
Xi
1+
β
2
Xi
2+··· +
βpXip
+
εi
Table 2 Considered potential spatial determinants of COVID-19 cases. Source: Statistics Portugal and Carta Social
Dimension Example of indicator(s) Number
Age dimension • Proportion of population by age group 4
Sociodemographic • Population density, urbanization rate and average household size
• Students enrolled by year of schooling
• Beneficiaries of social and unemployment benefits
• Public housing, average age of buildings and decayed dwellings
18
Mobility • Use of public transport and personal vehicle in daily commute
• Time duration of daily commuting route
• Intermunicipal and interparish commuting
6
Economic • Employment location quotients (LQ) for 16 sectors
• Declared income, export value and gross value added of companies
• Tourism overnight stays
• Housing expenses and owner occupied housing ratio
23
Fig. 2 14-day cumulative incidence (dependent variables): a 1st wave start; b 1st wave peak; c 2nd wave start; d 2nd wave peak; e 3rd wave start; f
3rd wave peak
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 6 of 16
Alvesetal. International Journal of Health Geographics (2023) 22:8
A stepwise algorithm was used to ensure the selection of
significant independent variables. Stepping method crite-
ria used a p-value with an entry value of 0.05 and 0.1 for
removal.
Threshold identication
After reducing the initial set of variables and identifying
the most relevant ones to explain the spatial dynamics of
COVID-19, followed the stage of inventorying the exist-
ence of thresholds as change points in the relationship
between determinants and disease incidence.
In this regard, we used a CPA executed with the R
package ‘bcp’ [71] based onthe work of Barry and Harti-
gan [72, 73]. In statistical analysis, CPA or step detection
methods attempt to identify the moments at which the
probability distribution of a stochastic process or time
series changes, which have been used in epidemiologi-
cal studies [74]. In this case, the probability distribution
was not a time series, but the values of an independ-
ent variable in ascending order for each municipality.
e link between the incidence of COVID-19 and each
determinant was modelled in a bivariate approach using
this method. In the specific case of the package used, a
Bayesian and offline method, the abrupt changes in the
posterior mean of COVID-19 incidence in relation to the
determinants of infection were evaluated obtaining the a
posteriori probability of change points.
Weighted linear combination
e results of MLR and CPA fed the WLC using 3 deter-
minants to create the susceptibility index. e relative
importance of the determinants in the outcome was cal-
culated based on the number of periods in which they
were significant in the MLR. As a result, a variable with
a greater number of significant associations contributed
more to the susceptibility index than one with a lower
frequency of significance. With this data-driven meth-
odology, with reduced human parameterization com-
pared to other strategies (e.g., analytic hierarchy process),
mainland Portugal was classified by susceptibility to
COVID-19 infection at the municipal scale.
Validation
e validation of the susceptibility index was performed
by calculating the area under the curve (AUC). Suc-
cessrate curves were constructed for the first three waves
by using the modelling data. In addition, we determined
predictionrate curves for the peaks of the 4th and 5th
waves that are the validation set, i.e., epidemiological
data unknown to the model.
e accuracy of the classification was measured by the
AUC for all the periods considered as:
where
AUCi
is the area under the curve for the period
i
,
a is the area between the 45-degree line and the success
or prediction curve and b is the area above the curve. A
higher value represents a curve that with a lower cumu-
lative percentage of the study area better captures the
cumulative cases, while a lower index means higher dif-
ficulty in separability.
Results
e results indicate that the factors examined accurately
predicted the spatiotemporal dynamics of COVID-19,
albeit with varying importance through time. e sus-
ceptibility analysis methodology, which combined clas-
sical and Bayesian techniques, classified municipalities
according to their susceptibility to COVID-19 infection.
Clusters of greater infection susceptibility were identi-
fied based on economic, sociodemographic, and mobility
characteristics. In summary, the approach adopted sup-
portedthe hypothesis thatNPI should be specificallytai-
lored to localgeographical contexts.
Determinants
e MLR highlighted that COVID-19 diffusion is a mul-
tifactorial phenomenon with associations varying across
time. From the 51 variables for six moments, 19 were
identified as statistically significant (Table3). e num-
ber of significant factors for each moment of incidence
ranged from 6 to 11, with a mean of 9. e importance
of these variables, in terms of regression coefficients and
statistical significance, had variability depending on the
incidence period. We identified the importance of fac-
tors related to the heterogeneous occupation of the ter-
ritory (population density, average family size, students
enrolled of various levels), economic (income, concentra-
tions of employment in sectors where face-to-face work
is indispensable, such as textile industry and storage and
auxiliary transport activities) and mobility (use of pub-
lic transport, average duration of commuting by public
transport, inter-municipal and interparish commuting).
On the contrary, population age did not turn out to be
a key factor although several indicators associated with
school enrollment and employment (active population
proxies) were significant.
ree factors (Table4) stood out by the number of sig-
nificant moments and the relative weight of their regres-
sion coefficients:
• population density (Pdens)—sociodemographic
dimension;
AUCi=
a
(a
+
b)
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 7 of 16
Alvesetal. International Journal of Health Geographics (2023) 22:8
Table 3 Regression model’s standardized coefficients with 95% confidence interval and variance inflation factor (VIF)
Variables First wave Second wave Third wave
Start Peak Start Peak Start Peak
Sociodemografic Population density 0.580 (0.501–0.659)
VIF = 1.450 0.541 (0.458–0.624)
VIF = 1.544 0.498 (0.400–0.597)
VIF = 2.653 0.487 (0.402–0.573)
VIF = 2.486 0.349 (0.248–0.450)
VIF = 2.930 0.393 (0.299–
0.487)
VIF = 2.511
Average household size 0.170 (0.098–0.242)
VIF = 1.198 0.186 (0.107–0.266)
VIF = 1.416) 0.268 (0.196–0.340)
VIF = 1.424 0.352 (0.280–0.424)
VIF = 1.764
Urbanization rate 0.128 (0.043–0.212)
VIF = 1.579 0.209 (0.128–0.290)
VIF = 1.885 0.133 (0.051–
0.214)
VIF = 1.864
Students enrolled pre-
school − 0.144 (− 0.227 to
− 0.060)
VIF = 1.918
Students enrolled 2nd
cycle − 0.133 (− 0.213 to
− 0.052)
VIF = 1.487
− 0.164 (− 0.248 to
− 0.080)
VIF = 1.571
− 0.236 (− 0.314 to
− 0.159)
VIF = 2.054
− 0.144 (− 0.239 to
− 0.048)
VIF = 2.616
Students enrolled in
higher education 0.254 (0.181–0.328)
VIF = 1.239) 0.122 (0.045–0.200)
VIF = 1.642 0.081 (0.005–0.157)
VIF = 1.945
Age Population aged 0–15 0.227 (0.119–0.336)
VIF = 3.355
Mobility Population commuting
by public transport 0.090 (0.017–0.163)
VIF = 1.467 0.130 (0.047–0.213)
VIF = 1.985 0.209 (0.130–
0.289)
VIF = 1.799
Population working/
studying outside parish 0.178 (0.105–0.251)
VIF = 1.199 0.148 (0.074–0.222)
VIF = 1.867 0.251 (0.176–0.326)
VIF = 1.622 0.198 (0.133–
0.263)
VIF = 1.207
Population working/
studying outside
municipality
− 0.137 (− 0.218 to
0.057)
VIF = 1.776
− 0.174 (− 0.260 to
− 0.087)
VIF = 2.562
Average time duration
of commuting 0.154 (0.076–0.231)
VIF = 2.043 0.029 (-0.055–0.113)
VIF = 2.025
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 8 of 16
Alvesetal. International Journal of Health Geographics (2023) 22:8
Table 3 (continued)
Variables First wave Second wave Third wave
Start Peak Start Peak Start Peak
Economic Declared income 0.168 (0.069–0.267)
VIF = 2.819 0.246 (0.151–
0.341)
VIF = 2.548
Housing expenses 0.105 (0.036–0.174)
VIF = 1.606 0.104 (0.030–0.179)
VIF = 1.583 0.141 (0.070–
0.211)
VIF = 1.423
Owner occupied
housing − 0.277 (− 0.368 to
− 0.186)
VIF = 2.282
− 0.203 (− 0.286 to
− 0.119)
VIF = 2.377
LQ textile industry 0.118 (0.044–0.192)
VIF = 1.210 0.142 (0.081–0.203)
VIF = 1.271 0.072 (0.008–0.136)
VIF = 1.192
LQ vehicle trade and
repair 0.080 (0.016–0.143)
VIF = 1.157
LQ storage and auxiliary
transport activities 0.154 (0.080–0.228)
VIF = 1.269 0.165 (0.088–0.242)
VIF = 1.314 0.170 (0.100–0.239)
VIF = 1.321 0.138 (0.072–0.203)
VIF = 1.471 0.081 (0.007–
0.155)
VIF = 1.553
LQ electrical equipment
manufacturing 0.172 (0.103–0.241)
VIF = 1.108 0.130 (0.060–0.200)
VIF = 1.101
LQ hospitality and
restaurants − 0.088 (− 0.157 to
− 0.019)
VIF = 1.312
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 9 of 16
Alvesetal. International Journal of Health Geographics (2023) 22:8
• proportion of population working outside the parish
of residence (PWOparish)—mobility dimension;
• location quotient of employment in storage and aux-
iliary transport activities (LQstotrans)—economic
dimension.
ese three variables demonstrated a positive signifi-
cant relationship with the number of new cases, e.g., the
higher the variable value, the more cases the municipal-
ity tends to have and accounted for more than 60% of the
variation explained by the stepwise models. An examina-
tion of the factors by dimensions reveal that Pdens was
the only one with significance in all moments under
study and had the highest regression coefficients. LQsto-
trans was the most relevant in the economic dimension
and PWOparish had more frequent associations in the
mobility group. e loss of explanation when eliminating
the remaining variables is minimal and considering their
importance these three indicators were the ones selected
for CPA and later to determine the susceptibility index.
e difference in the number of significant variables
between the start and peak of the waves did not present
a link.
Thresholds
e bivariate Bayesian CPA identified the posterior prob-
ability of changing points. Multiple probable points of
changing relationships have been identified between
incidence and determinants. To decrease the number
of changes a minimum probability threshold of 0.7 was
defined to assume the existence of a change in the series
since this is a reference value in statistics. e trends
were segmented to generalize thresholds for all the ana-
lysed periods (Fig.3).
e Bayesian CPA highlighted that the relation-
ship between factors and incidence depends on various
changing points that trigger the posteriori mean inci-
dence of new cases. is way, the results were suggestive
of not fully linear relationships, corroborating 20 to 30%
of unexplained variability of the MLR models.
us, the susceptibility to infection associated with
each determinant is based on varying gradients that show
that the influence of a determinant on the propensity to
infect is not directly proportional to its value. For exam-
ple, Pdens is practically irrelevant until 300 inhabitants
per km2, while in the case of PWOparish, although non-
linear, it is closer to a trajectory that could be partitioned
into mulitple linear segments.
e combination of these three indicators by a WLC
allowed the calculation of the susceptibility index. Also,
at this stage the weight associated with each determinant
resulted from the available information without subjec-
tive influence, and its importance was defined based on
the proportion of the number of periods in which the
respective variables demonstrated an association with
the target (see Table3). ereby, the Pdens assumedan
importance of 40%, the LQstotrans of 33% and the
PWOparish of 27%.
Spatial susceptibility index
The spatial patterns of the susceptibility associated
with each factor revealed contrasting and heterogene-
ous patterns, even though some municipalities were
classified similarly (Fig. 4). This is reflected in the
Table 4 Adjusted R2 comparing models with all variables versus the 3 most significant
Model First wave Second wave Third wave
Start Peak Start Peak Start Peak
Stepwise variables (number) 0.692 (6) 0.679 (8) 0.741 (9) 0.790 (11) 0.751 (11) 0.748 (7)
Pdens + PWOparish + LQstotrans 0.617 0.600 0.646 0.656 0.669 0.680
Difference − 0.075 − 0.079 − 0.095 − 0.134 − 0.082 − 0.068
Fig. 3 Most relevant changing points between incidence and factors: a Pdens; b LQstotrans; c PWOparish
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 10 of 16
Alvesetal. International Journal of Health Geographics (2023) 22:8
patterns of the final susceptibility index that visually
replicates the influence of population density with a
higher susceptibility in the metropolitan areas of Lis-
bon and Oporto and important urban systems such as
the regional capitals. The Algarve coast in the South,
albeit one of the most populous and economically
dynamic regions, shows low susceptibility because
LQstotrans and PWOparish have low expression in
this region. This is not surprising since in the first
three waves the Algarve region registered low numbers
of COVID-19 infections.
e distribution of the susceptibility classes suggests
the existence of specific geographic contexts influenced
by the considered dimensions: sociodemographic, eco-
nomic and mobility. It is also evident the influence of
communication axes and the spatial dependence of the
classes, i.e., the proximity, in terms of geographical dis-
tance between municipalities, seems to be relevant in
terms of susceptibility. is fact is particularly evident
in the case of the Northwest, where the Oporto met-
ropolitan area demonstrated a gradient of diminishing
susceptibility with increasing distance from Oporto
city, but which is “inflated” by the closest regional capi-
tals, such as Viana do Castelo or Braga. Also in the
interior, the case of Guarda or Viseu is representative
of this phenomenon, with adjacent municipalities clas-
sified with high susceptibility.
Fig. 4 Susceptibility to COVID-19 infection in mainland Portugal: a Pdens; b LQstotrans; c PWOparish; d final susceptibility index
Fig. 5 Susceptibility classes and the monthly incidence of COVID-19
during the first year of the disease in Portugal
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 11 of 16
Alvesetal. International Journal of Health Geographics (2023) 22:8
In terms of accuracy, success and prediction rate curves
revealed that 40% of the municipalities (Very High and
High susceptibility) explained between 80 and 90% of the
new cases of the disease (Fig.5). e AUC were signifi-
cant with values above 0.75 which is a reference for good
discrimination. e validation implies that the threshold-
based modelling process had significance in determining
the areas with a greater propensity to register cases of
COVID-19.
Comparing the susceptibility index with COVID-19
DGS risk classification (14-day incidence rate per 100,000
inhabitants) for the third wave demonstrated why inte-
grating susceptibility and epidemiological monitoring is
relevant for NPI management (Fig.6). e comparison
demonstrated little correspondence between the restric-
tiveness of the NPI and the susceptibility index, resulting
in low contingency and Spearman coefficients.
Most of the municipalities in the highest risk level had
very high susceptibility however, almost 30% had only
very low to moderate. Considering the highest three lev-
els (each had different sets of NPI with growing restric-
tiveness) seem to have existed overly rigid measures for
several geographical contexts whose socio-territorial
characteristics were not determinants of COVID-19
spread. us, these locations had NPI that overestimated
the propensity for transmission. Also, in the first and
second levels (alert levels without specific interventions)
some municipalities with high and very high susceptibil-
ity stood out, presumably indicating an underestimation
of outbreak potential.
Overall we can say that the correlation was low and
that a disagreement between the severity of restrictions
and the actual propensity for transmission was found.
erefore, the susceptibility index can be a viable instru-
ment to support epidemiological containment policies
preventing future uncontrolled transmission by imposing
stricter restrictions in more susceptible areas.
Discussion
is study identified COVID-19 infection determinants
and mapped the susceptibility using a data-driven thresh-
old approach based on only three variables, with the
hypothesis that containment measures should consider
not only epidemiological indicators but also the true pro-
pensity to transmission dynamics by taking geographical
contexts into consideration. e results support a multi-
cause aetiology for COVID-19 transmission dynamics
patterns and the spatial susceptibility index highlights
peculiar situations in which public health authorities may
need tailored interventions.
Specicities ofthemethodology
In methodological terms, some distinctive features can
be mentioned. Considering that COVID-19 is often
Fig. 6 Susceptibility classes and DGS risk levels for the peak of the third wave
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 12 of 16
Alvesetal. International Journal of Health Geographics (2023) 22:8
asymptomatic and under-reported [75], leading to diffi-
culties in the identification by epidemiological monitor-
ing and surveillance systems [76], the use of the Bayesian
CPA method is justified. e uncertainty in the model-
ling process regarding the data itself was addressed by
using posterior probabilities to define the thresholds.
Furthermore, according to Nazia etal. [23] the major-
ity of COVID-19 spatiotemporal analyses used frequen-
tist methods, with only a minority embracing Bayesian
approaches. In this sense, by combining a frequentist
regression with a Bayesian method to infer transition
points, the present study distances itself from more clas-
sical approaches and uses an uncommon method in
susceptibility analysis. Moreover, the same authors men-
tion the prevalence of studies addressing regional scales.
However, our research focused on a local analysis that
took advantage of a finer scale to improve the prediction
of transmission prone municipalities.
Summary ofresults
e time-varying relationships of factors identified by
the MLR introduce uncertainty for an effective quantifi-
cation of their real contribution. is was already iden-
tified in other studies [8, 77] and forces researchers to
analyse longer periods to accurately identify and quantify
the factors explaining COVID-19 diffusion. Nevertheless,
the analysis of this study for a period of 1year allowed
us to unequivocally identify the importance of determi-
nants related to the heterogeneous occupation of the ter-
ritory. Urban population distribution and density, as well
as household size, have a strong association with the spa-
tiotemporal dynamics of COVID-19. Aside from a more
structural view of population distribution, employment
concentrations associated with regional employment spe-
cialisation and agglomeration patterns with strong inter-
action dynamics at regional, national, and international
scales (e.g., [78]), have been linked to infection diffusion.
Still, on the economic side, it is worth mentioning indica-
tors such as income, expenses related to housing and the
proportion of owner occupied dwellings. At the study’s
scale, these results cannot be interpreted as indicative of
socio-spatial inequalities as infection-predisposing fac-
tors, but as proxies of the most populous municipalities
(because the standardized coefficients were positive) and,
therefore, with more active epidemiological dynamics.
Although the variables explicitly related to the age dimen-
sion had little association with the dependent variables,
indicators related to employment and school enrollment
were significant. is suggests that the active population
was an important agent of transmission at certain times,
specifically at the beginning of waves, emphasising the
importance of implementing NPI associated with tele-
working and mobility restrictions [48] to prevent disease
transmission. It is also known that population mobility
patterns are an unequivocal driver of infectious disease
transmission [79] and although the data used was some-
what outdated, it showed how commuting had important
links with COVID-19 transmission.
Since the MLR model’s explanatory power, albeit sig-
nificant, did not exceed 70 to 80% of the variation of the
dependent variable, the relationships between incidence
and their explanatory factors were not completely linear.
is is due to residual heteroscedasticity, which can be
indicative of the need to incorporate other factors. For
example, a behavioural dimension, such as adherence
to NPI, mask use, containment and exposure reduction
practices [80], is of extreme importance in such a study
[28] but was not considered. Moreover, the existence of
multiple thresholds in variables’ relationships demon-
strated the importance of territorial specificities, explain-
ing the inability of linear models to accommodate all the
variations in the number of cases.
Combining the results from the MLR and a CPA,
mainland Portugal municipalities were classified by their
susceptibility to COVID-19 infection. Despite the com-
plexity of infectious diseases, good model accuracy was
achieved with only three variables (Pdens, LQstotrans
and PWOparish). e heterogeneous geography of the
index derives from the fact that the distribution of the
determinants is uneven and anisotropic. e suscepti-
bility spatial patterns resemble the distribution of con-
firmed cases in a trend that is “coastlised” along the most
densely populated coastal areas, polarized around the
country’s two metropolitan areas—Lisbon and Porto—
and anchored in mainland regional capitals. e suc-
cess and prediction curves followed a power distribution
since most cases occurred with a high concentration in a
small number of municipalities (e.g., metropolitan areas).
e power distribution loses strength from the first
wave—when the distribution of cases was more evident
on the coast—to the subsequent waves when the infec-
tion spread to all municipalities.
e importance of economic, socio-demographic, and
mobility determinants reinforces the conclusion of pre-
vious studies [8, 44, 46] for Portugal, even though the
present study focused on a longer period. Municipalities
with higher incidence rates coincided with the highest
susceptibility classes at the peak of the third wave. How-
ever, a relevant number of outliers was identified in all
the risk levels proving the that the rigidity of the restric-
tions was not always adequate considering the propensity
to infection based on the spatial determinants conducive
to COVID-19 infection. In this sense, the relevance of
integrating epidemiological monitoring with susceptibil-
ity emerges as a relevant proposal in the domain of the
management of NPI in Portuguese territory.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 13 of 16
Alvesetal. International Journal of Health Geographics (2023) 22:8
The implications ofsusceptibility andfuture developments
e major contribution of this work is the development
of a municipal susceptibility index for spatial decision-
making in managing containment policies. e territorial
units with higher susceptibility—most of which have spa-
tial proximity—need to have more restrictive NPI, com-
pared to those with lower, or adopted ahead of time to
avoid severe outbreaks, due to the spatial conjugation of
socio-territorial specificities that enhance transmission.
e ability to adjust measures to contain the infection
based on its propensity to spread is particularly impor-
tant since according to Jain etal. [81] controlling an out-
break at the grassroots level has profound repercussions
for the nationwide control of transmission chains. Fur-
thermore, municipalities with higher susceptibility were
geographically close to similar classes. As Duarte et al.
[82] have assessed neighbouring municipalities tended
to share similar behaviour because local effects justify
spatial dependence in COVID-19 diffusion, confirmed in
Portugal [43, 44], and which our modelling process did
not account for. is is not unusual, since one of the most
common processes of infectious disease spatial diffu-
sion—contagious diffusion—is based essentially on spa-
tial contiguity [38, 83] and which was boosted by mobility
movements between municipalities. Considering this
information, the geographical character of COVID-19
transmission is reinforced, strengthening the need for
differentiated measures according to local contexts, e.g.,
spatial-based containment measures should also consider
geographic properties such as proximity and contiguity
(to areas of higher susceptibility).
e proposed index appears adequate for customized
NPI, avoiding harsh approaches where it has no benefits
and soft in contexts of rapid diffusion. Knowing also the
potential negative consequences associated with NPI
and long lockdown periods [84], it is important to adapt
strategies to the contexts in which they fall. Despite the
satisfactory results, further work is needed for a more
robust spatial index considering a second order of fac-
tors and incorporating spatial dependence. Alternative
approaches for a broader classification could be the use
of additional epidemiological indicators such as persons
hospitalized and the positivity of testing rate. Also, the
use of “near real-time” mobility data, such as Google’s
Community Mobility Reports [85], is relevant to fore-
cast future cases [48] which can allow for a time-dynamic
susceptibility classification. It should also be noted that
infection patterns have changed with the progression of
the disease, either by vaccination and/or disease variants
[86], therefore identifying factors may require updating,
which has direct implications for susceptibility maps.
Moreover, in light of the non-linear parameters evi-
denced by the CPA, it is relevant to evaluate whether the
patterns of COVID-19 diffusion are indeed non-linear, or
whether this non-linearity results from spatially varying
processes [87]. Based on this evaluation, it may be appro-
priate to test dummy variables as proxies for certain terri-
torial configurations (e.g., municipalities of metropolitan
areas) and use spatial regressions or non-linear models.
e implications of the results are relevant in the con-
text of prevention and for public health policies evalu-
ation, something not always straightforward during
the pandemic contributing to improved containment
policies.
Limitations
In methodological terms, a lack of information on some
important factors may have hindered the development
of an improved index. Also, DGS COVID-19 data has
several known flaws [3], both in the allocation of cases
to territorial units and temporal distribution, as well as
loss of synchronization over time. It is unknown to what
extent some quality problems with this data, which can-
not be overcome, could have caused biased results. In
addition, there were some periods of higher incidence,
namely severe outbreaks that have no known direct
explanation by the determinants [27], as occurred in
migrant communities working in agricultural areas and
residing in conditions of overcrowding and insalubrity
[88]. e susceptibility index cannot explain these situa-
tions since they are the outcome of accidental outbreaks
under very specific conditions for which there is no avail-
able explanatory data. e static character of the inde-
pendent variables, and their outdated condition, were
also an obstacle to better adjustments since numerous
high-magnitude changes have happened, such as vari-
ations in mobility patterns [48]. Finally, given that the
results stem from aggregated units, there is the influence
of modifiable area unit problems as well as ecological fal-
lacy [89, 90] which means that the results should not be
extrapolated to individual-level.
Conclusions
e present study demonstrated how the integration of
susceptibility to COVID-19 infection, based on the dis-
tribution of the known determinants and their effects, is
relevant for policy guidance and containment strategies
in specific geographic contexts using Portugal as a case
study.
e results shed new light on how knowledge of the
distribution of factors explaining transmission is crucial
to identify locations where higher incidence is expected
by the conjugation of sociodemographic, economic and
mobility characteristics. By using factors with proven
explanatory power in COVID-19 diffusion in mainland
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 14 of 16
Alvesetal. International Journal of Health Geographics (2023) 22:8
Portugal, we proposed a susceptibility index to imple-
ment spatial-based NPI.
Conclusions can be summarised in three points:
1. e MLR results showed that the importance of
determinants to COVID-19 infection had time-vary-
ing contributions, although there are three with con-
sistent relationships over time: population density,
inter-parish commuting and employment in storage
and transport auxiliary activities.
2. e bivariate probabilistic CPA revealed a non-linear
nature of the relationships between infection deter-
minants and observed incidence, allowing the identi-
fication of thresholds as transition points in changing
trends.
3. Comparing the susceptibility classes with the risk
levels for NPI evidenced low correlation, suggesting
the need for considering susceptibility as a criterion
together with epidemiological monitoring.
e findings prove that the portuguese NPI strat-
egy was poorly adjusted to the reality of the propensity
to COVID-19 spread. In summary, the results lay the
groundwork for future models that intersect incidence
rate with the susceptibility to infection for NPI manage-
ment, advocating the need for greater incorporation of
spatial variables in epidemiological containment policies.
It is also noteworthy that, unlike previous studies, this
one followed a data-driven approach based on thresh-
olds, reducing subjectivity when compared to previous
studies using multicriteria analysis. e approach can be
extended to other regions of the world for the current or
future epidemic(s).
Author contributions
Conceptualization: AA, NMC, PM and EMC; Methodology: AA, NMC and PM;
data analysis: AA, NMC and PM; writing—original draft preparation: AA; writ-
ing—review and editing: NMC, PM and EMC. All authors read and approved
the final manuscript.
Funding
This research was funded by Portuguese Foundation for Science and
Technology, I.P. (FCT ) (UIDB/00295/2020 and UIDP/00295/2020), Project
COMPRIME—Get to Know More for Intervention (ID: 596685735) and Project
COMPRI_Mov—Get to Know More for Intervention in the context of mobility
(ID: 613765655).
Availability of data and materials
Not applicable.
Declarations
Ethics approval and consent to participate
Not applicable.
Competing interests
The authors confirm that there is no conflict of interest.
Author details
1 Centre of Geographical Studies, Institute of Geography and Spatial Planning,
University of Lisbon, 1600-276, Lisbon, Portugal. 2 Associate Laboratory TERRA ,
1349-017 Lisbon, Portugal.
Received: 15 February 2023 Accepted: 28 March 2023
References
1. Yang Y, Peng F, Wang R, Guan K, Jiang T, Xu G, et al. The deadly coronavi-
ruses: the 2003 SARS pandemic and the 2020 novel coronavirus epidemic
in China. J Autoimmun. 2020;109(February): 102434. https:// doi. org/ 10.
1016/j. jaut. 2020. 102434.
2. Desvars-Larrive A, Der vic E, Haug N, Niederkrotenthaler T, Chen J, Di
Natale A, et al. A structured open dataset of government interventions in
response to COVID-19. Sci Data. 2020;7(1):285. https:// doi. org/ 10. 1038/
s41597- 020- 00609-9.
3. Marques da Costa N, Mileu N, Alves A. Dashboard comprime_compri_
mov: multiscalar spatio-temporal monitoring of the covid-19 pandemic
in Portugal. Future Internet. 2021;13(2):1–17.
4. Andersen LM, Harden SR, Sugg MM, Runkle JDD, Lundquist TE. Analyzing
the spatial determinants of local Covid-19 transmission in the United
States. Sci Total Environ. 2020. https:// doi. org/ 10. 1016/j. scito tenv. 2020.
142396.
5. Coccia M. Factors determining the diffusion of COVID-19 and suggested
strategy to prevent future accelerated viral infectivity similar to COVID. Sci
Total Environ. 2020;729: 138474.
6. Mollalo A, Vahedi B, Rivera KM. GIS-based spatial modeling of COVID-
19 incidence rate in the continental United States. Sci Total Environ.
2020;728: 138884. https:// doi. org/ 10. 1016/j. scito tenv. 2020. 138884.
7. Murgante B, Borruso G, Balletto G, Castiglia P, Dettori M. Why Italy first?
Health, geographical and planning aspects of the COVID-19 outbreak.
Sustainability. 2020;12(12):5064.
8. Sousa P, Marques da Costa N, Marques da Costa E, Rocha J, Peixoto VR,
Fernandes AC, et al. COMPRIME—COnhecer Mais PaRa Intervir MElhor:
preliminary mapping of municipal level determinants of COVID-19 trans-
mission in Portugal at different moments of the 1st epidemic wave. Port J
Public Health. 2021;38(1):18–25.
9. Sugg MM, Spaulding TJ, Lane SJ, Runkle JD, Harden SR, Hege A, et al.
Mapping community-level determinants of COVID-19 transmission
in nursing homes: a multi-scale approach. Sci Total Environ. 2021;752:
141946. https:// doi. org/ 10. 1016/j. scito tenv. 2020. 141946.
10. Oliveira S, Ribeiro AI, Nogueira P, Rocha J. Simulating the effects of
mobility restrictions in the spread of SARS-CoV-2 in metropolitan areas in
Portugal. PLoS ONE. 2022;17(9 September):1–17. https:// doi. org/ 10. 1371/
journ al. pone. 02742 86.
11. Daras K, Alexiou A, Rose TC, Buchan I, Taylor-Robinson D, Barr B. How does
vulnerability to COVID-19 vary between communities in England? Devel-
oping a small area vulnerability index (SAVI). J Epidemiol Community
Health. 2021;75(8):729–34.
12. Magalhães JPM, Ribeiro AI, Caetano CP, Sá Machado R. Community
socioeconomic deprivation and SARS-CoV-2 infection risk: findings from
Portugal. Eur J Public Health. 2022;32(1):145–50. https:// doi. org/ 10. 1093/
eurpub/ ckab1 92.
13. Murgante B, Balletto G, Borruso G, Saganeiti L, Pilogallo A, Francesco S,
et al. A methodological proposal to evaluate the health hazard scenario
from COVID-19 in Italy. Environ Res. 2022;5:209.
14. Sá Marques T, Santos H, Honório F, Ferreira M, Ribeiro D, Torres M. The
territorial mosaic of contagion and mortality risk by covid-19 in mainland
Portugal. Finisterra. 2020;55(115):19–26.
15. Sarkar A, Chouhan P. COVID-19: district level vulnerability assessment in
India. Clin Epidemiol Glob Health. 2021;9:204–15.
16. Sarkar SK. COVID-19 susceptibility mapping using multicriteria evalua-
tion. Disaster Med Public Health Prep. 2020;14(4):521–37.
17. Macharia PM, Joseph NK, Okiro EA. A vulnerability index for COVID-19:
spatial analysis at the subnational level in Kenya. BMJ Glob Health.
2020;5(8): e003014.
18. Savini L, Candeloro L, Calistri P, Conte A. A municipality-based approach
using commuting census data to characterize the vulnerability to
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 15 of 16
Alvesetal. International Journal of Health Geographics (2023) 22:8
influenza-like epidemic: the COVID-19 application in Italy. Microorgan-
isms. 2020;8(6):1–21.
19. Dowd JB, Andriano L, Brazel DM, Rotondi V, Block P, Ding X, et al. Demo-
graphic science aids in understanding the spread and fatality rates of
COVID-19. Proc Natl Acad Sci. 2020;117(18):9696–8.
20. Gao J, Zhang P. China’s public health policies in response to COVID-19:
from an “authoritarian” perspective. Front Public Health. 2021;15:9.
21. Saffary T, Adegboye OA, Gayawan E, Elfaki F, Kuddus MA, Saffary R. Analy-
sis of COVID-19 cases’ spatial dependence in us counties reveals health
inequalities. Front Public Health. 2020;8:728. https:// doi. org/ 10. 3389/
fpubh. 2020. 579190.
22. Sun F, Matthews SA, Yang TC, Hu MH. A spatial analysis of the COVID-19
period prevalence in US counties through June 28, 2020: where geogra-
phy matters? Ann Epidemiol. 2020;52:54-59.e1.
23. Nazia N, Butt ZA, Bedard ML, Tang WC, Sehar H, Law J. Methods used in
the spatial and spatiotemporal analysis of COVID-19 epidemiology: a
systematic review. Int J Environ Res Public Health. 2022;19(14):8267.
24. van Schalkwyk MCI, McKee M. Research into policy: lessons from the
COVID-19 pandemic. Eur J Public Health. 2021;31(Supplement_4):iv3–8.
https:// doi. org/ 10. 1093/ eurpub/ ckab1 55.
25. Riley S. Large-scale models of infectious disease. Science (80-).
2007;316(June):1298–301.
26. De Lima LMM, De Sá LR, Dos Santos Mac Ambira AFU, De Almeida
Nogueira J, De Toledo Vianna RP, De Moraes RM. A new combination rule
for spatial decision support systems for epidemiology. Int J Health Geogr.
2019;18(1):1–10. https:// doi. org/ 10. 1186/ s12942- 019- 0187-7.
27. Moniz M, Soares P, Nunes C. COVID-19 transmission dynamics: a space -
and-time approach. Port J Public Health. 2021;38(1):4–10.
28. Bedson J, Skrip LA, Pedi D, Abramowitz S, Carter S, Jalloh MF, et al. A
review and agenda for integrated disease models including social and
behavioural factors. Nat Hum Behav. 2021;5(7):834–46. https:// doi. org/ 10.
1038/ s41562- 021- 01136-2.
29. Brinks V, Ibert O. From corona virus to corona crisis: the value of an
analytical and geographical understanding of crisis. Tijdschr voor Econ en
Soc Geogr. 2020;111(3):275–87. https:// doi. org/ 10. 1111/ tesg. 12428.
30. Andrew MK, Mitnitski AB, Rockwood K. Social vulnerability, frailty and
mortality in elderly people. PLoS ONE. 2008;3(5): e2232. https:// doi. org/
10. 1371/ journ al. pone. 00022 32.
31. de Paiva CA, Oliveira APDS, Muniz SS, Calijuri ML, Dos Santos VJ, Alves
SDC. Determination of the spatial susceptibility to yellow fever using a
multicriteria analysis. Mem Inst Oswaldo Cruz. 2019;114: e180509.
32. Dickin SK, Schuster-Wallace CJ, Elliott SJ. Developing a vulnerability
mapping methodology: applying the water-associated disease index to
dengue in Malaysia. PLoS ONE. 2013;8(5): e63584.
33. Kottow MH. The vulnerable and the susceptible. Bioethics. 2003;17(5–
6):460–71. https:// doi. org/ 10. 1111/ 1467- 8519. 00361.
34. Porta M. A dictionary of epidemiology. Oxford: Oxford University Press;
2014. https:// doi. org/ 10. 1093/ acref/ 97801 95314 496. 001. 0001/ acref-
97801 95314 496.
35. Adger WN. Social and ecological resilience: are they related? Prog Hum
Geogr. 2000;24(3):347–64. https:// doi. org/ 10. 1191/ 03091 32007 01540 465.
36. Sá Marques T, Ferreira M, Saraiva M, Forte T, Santinha G. Mapping health
vulnerabilities: exploring territorial profiles to support health policies.
Cienc e Saude Coletiva. 2021;26:2459–70.
37. Malta FS, Marques da Costa E. Socio-environmental vulnerability index:
an application to Rio de Janeiro-Brazil. Int J Public Health. 2021. https://
doi. org/ 10. 3389/ ijph. 2021. 584308.
38. Cliff AD, Haggett P, Ord JK, Versey GR. Spatial diffusion: an historical
geography of epidemics in an island community. New York: Cambridge
University Press; 1981. p. 238.
39. Gianquintieri L, Brovelli MA, Pagliosa A, Dassi G, Brambilla PM, Bonora R,
et al. Generating high-granularity COVID-19 territorial early alerts using
emergency medical services and machine learning. Int J Environ Res
Public Health. 2022;19(15):9012.
40. Mah JC, Andrew MK. Social vulnerability indices: a pragmatic tool for
COVID-19 policy and beyond. Lancet Reg Health Eur. 2022. https:// doi.
org/ 10. 1016/j. lanepe. 2022. 100333.
41. Welsh CE, Sinclair DR, Matthews FE. Static socio-ecological COVID-19
vulnerability index and vaccine hesitancy index for England. Lancet
Reg Health Eur. 2022;14: 100296. https:// doi. org/ 10. 1016/j. lanepe. 2021.
100296.
42. Franch-Pardo I, Napoletano BM, Rosete-Verges F, Billa L. Spatial analysis
and GIS in the study of COVID-19. A review. Sci Total Environ. 2020;739:
140033. https:// doi. org/ 10. 1016/j. scito tenv. 2020. 140033.
43. Alves AJJ. Modelação espácio-temporal da propagação da COVID-19 em
Portugal Continental: evidências da importância de fatores geográficos
[Spatio-temporal modeling of COVID-19 spread in mainland Portugal:
evidence of the importance of geographical factors]. Lisbon: Universi-
dade de Lisboa; 2022.
44. Almendra R, Santana P, Costa C. Spatial inequalities of COVID-19 inci-
dence and associated socioeconomic risk factors in Portugal. Boletín la
Asoc Geógrafos Españoles. 2021. https:// doi. org/ 10. 21138/ bage. 3160.
45. Pereira L, Correia J, Sequeiros J, Santos J, Jerónimo C. Spatial-temporal
monitoring risk analysis and decision-making of COVID-19 distribution
by region. Int J Simul Process Model. 2022;18(1):23–35. https:// doi. org/ 10.
1504/ IJSPM. 2022. 123472.
46. Barbosa B, Silva M, Capinha C, Garcia RAC, Rocha J. Spatial correlates of
COVID-19 first wave across continental Portugal. Geospat Health. 2022.
https:// doi. org/ 10. 4081/ gh. 2022. 1073.
47. Marques da Costa E, Marques da Costa N. O processo pandémico da
Covid-19 em Portugal Continental: Análise geográfica dos primeiros 100
dias [The Covid-19 pandemic process in Mainland Portugal: a geographi-
cal analysis of the first 100 days]. Finisterra. 2020;55(115):11–8.
48. Mileu N, Costa N, Marques da Costa E, Alves A. Mobility and dissemina-
tion of COVID-19 in Portugal: correlations and estimates from Google’s
mobility data. Data. 2022;7:107.
49. Imdad K, Sahana M, Rana MJ, Haque I, Patel PP, Pramanik M. A district-
level susceptibility and vulnerability assessment of the COVID-19
pandemic’s footprint in India. Spat Spatiotemporal Epidemiol. 2021;36:
100390. https:// doi. org/ 10. 1016/j. sste. 2020. 100390.
50. Al-Salem W, Moraga P, Ghazi H, Madad S, Hotez PJ. The emergence
and transmission of COVID-19 in European countries, 2019–2020: a
comprehensive review of timelines, cases and containment. Int Health.
2021;13(5):383–98.
51. Caetano C, Morgado ML, Patrício P, Pereira JF, Nunes B. Mathematical
modelling of the impact of non-pharmacological strategies to control
the COVID-19 epidemic in Portugal. Mathematics. 2021;9(10):1084.
52. Hsiang S, Allen D, Annan-Phan S, Bell K, Bolliger I, Chong T, et al. The effect
of large-scale anti-contagion policies on the COVID-19 pandemic. Nature.
2020;584(7820):262–7.
53. Moore S, Hill EM, Tildesley MJ, Dyson L, Keeling MJ. Vaccination and non-
pharmaceutical interventions for COVID-19: a mathematical modelling
study. Lancet Infect Dis. 2021;21(6):793–802.
54. Altman G, Ahuja J, Monrad JT, Dhaliwal G, Rogers-Smith C, Leech G, et al.
A dataset of non-pharmaceutical interventions on SARS-CoV-2 in Europe.
Sci Data. 2022;9(1):1–9.
55. Pashakhanlou AH. Sweden’s coronavirus strategy: the Public Health
Agency and the sites of controversy. World Med Health policy. 2021.
https:// doi. org/ 10. 1002/ wmh3. 449.
56. Kianfar N, Mesgari MS, Mollalo A, Kaveh M. Spatio-temporal modeling
of COVID-19 prevalence and mortality using artificial neural network
algorithms. Spat Spatiotemporal Epidemiol. 2022;40(June 2021): 100471.
https:// doi. org/ 10. 1016/j. sste. 2021. 100471.
57. Scarpone C, Brinkmann ST, Große T, Sonnenwald D, Fuchs M, Walker BB. A
multimethod approach for county-scale geospatial analysis of emerging
infectious diseases: a cross-sectional case study of COVID-19 incidence
in Germany. Int J Health Geogr. 2020;19(1):1–17. https:// doi. org/ 10. 1186/
s12942- 020- 00225-1.
58. Thurner S, Klimek P, Hanel R. A network-based explanation of why
most COVID-19 infection curves are linear. Proc Natl Acad Sci USA.
2020;117(37):22684–9.
59. Bartlett MS. Measles periodicity and community size. J R Stat Soc Ser A.
1957;120(1):48–60. https:// doi. org/ 10. 2307/ 23425 53.
60. Anderson RM. Discussion: the Kermack-McKendrick epidemic threshold
theorem. Bull Math Biol. 1991;53(1):1. https:// doi. org/ 10. 1007/ BF024
64422.
61. Jiang F, Zhao Z, Shao X. Modelling the COVID-19 infection trajectory: a
piecewise linear quantile trend model. J R Stat Soc Ser B Stat Methodol.
2021;84:1–18.
62. Sera F, Armstrong B, Abbott S, Meakin S, O’Reilly K, von Borries R, et al.
A cross-sectional analysis of meteorological factors and SARS-CoV-2
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 16 of 16
Alvesetal. International Journal of Health Geographics (2023) 22:8
•
fast, convenient online submission
•
thorough peer review by experienced researchers in your field
•
rapid publication on acceptance
•
support for research data, including large and complex data types
•
gold Open Access which fosters wider collaboration and increased citations
maximum visibility for your research: over 100M website views per year
•
At BMC, research is always in progress.
Learn more biomedcentral.com/submissions
Ready to submit your research
Ready to submit your research
? Choose BMC and benefit from:
? Choose BMC and benefit from:
transmission in 409 cities across 26 countries. Nat Commun.
2021;12(1):5968. https:// doi. org/ 10. 1038/ s41467- 021- 25914-8.
63. DGS—Direção-Geral da Saúde. Relatório de Situação—COVID-19. 2020.
https:// covid 19. min- saude. pt/ relat orio- de- situa cao/. Accessed 16 Aug
2021.
64. Kutner MH, Nachtsheim CJ, Neter J, Li W. Applied linear statistical models.
Chicago: Irwin; 1996.
65. Waller LA, Gotway CA. Applied spatial statistics for public health data, vol.
100. New Jersey: Wiley; 2004. p. 702–3.
66. Dias P, Nobre F. Análise dos padrões de difusão espacial dos casos de
AIDS por estados brasileiros. Cad Saude Publica. 2001;17(5):1173–87.
67. VanderWeele TJ. On a square-root transformation of the odds ratio for a
common outcome. Epidemiology. 2017;28(6):e58–60.
68. Soni P, Gupta I, Singh P, Porte DS, Kumar D. GIS-based AHP analysis to
recognize the COVID-19 concern zone in India. GeoJournal. 2022. https://
doi. org/ 10. 1007/ s10708- 022- 10605-8.
69. Gao Z, Jiang Y, He J, Wu J, Xu J, Christakos G. An AHP-based regional
COVID-19 vulnerability model and its application in China. Model
Earth Syst Environ. 2022;8(2):2525–38. https:// doi. org/ 10. 1007/
s40808- 021- 01244-y.
70. Razavi-Termeh SV, Sadeghi-Niaraki A, Farhangi F, Choi SM. Covid-19 risk
mapping with considering socio-economic criteria using machine learn-
ing algorithms. Int J Environ Res Public Health. 2021;18(18):9657.
71. Erdman C, Emerson JW. bcp: an R package for performing a bayesian
analysis of change point problems. J Stat Softw. 2007;23(3):1–13.
72. Barry D, Hartigan JA. A Bayesian analysis for change point problems. J Am
Stat Assoc. 1993;88(421):309.
73. Barry D, Hartigan JA. Product partition models for change point prob-
lems. Ann Stat. 1992;20(1):260–79.
74. Kass-Hout TA, Xu Z, McMurray P, Park S, Buckeridge DL, Brownstein
JS, et al. Application of change point analysis to daily influenza-
like illness emergency department visits. J Am Med Inform Assoc.
2012;19(6):1075–81.
75. Lau H, Khosrawipour T, Kocbach P, Ichii H, Bania J, Khosrawipour V. Evalu-
ating the massive underreporting and undertesting of COVID-19 cases in
multiple global epicenters. Pulmonology. 2021;27(2):110–5.
76. Davis JT, Chinazzi M, Perra N, Mu K, Piontti APY, Ajelli M, et al. Cryptic
transmission of SARS-CoV-2 and the first COVID-19 wave. Nature. 2021.
https:// doi. org/ 10. 1038/ s41586- 021- 04130-w.
77. Tieskens KF, Patil P, Levy JI, Brochu P, Lane KJ, Fabian MP, et al. Time-vary-
ing associations between COVID-19 case incidence and community-level
sociodemographic, occupational, environmental, and mobility risk factors
in Massachusetts. BMC Infect Dis. 2021;21(1):686. https:// doi. org/ 10. 1186/
s12879- 021- 06389-w.
78. Schütz MH, Palan N. Restructuring of the international clothing and
textile trade network: the role of Italy and Portugal. J Econ Struct.
2016;5(1):1–29.
79. Gushulak BD, MacPherson DW. Population mobility and infectious
diseases: the diminishing impact of classical infectious diseases and
new approaches for the 21st century. Clin Infect Dis. 2000;31(3):776–80.
https:// doi. org/ 10. 1086/ 313998.
80. de Noronha N, Moniz M, Gama A, Laires PA, Goes AR, Pedro AR, et al. Non-
adherence to COVID-19 lockdown: who are they? A cross-sectional study
in Portugal. Public Health. 2022;211:5–13.
81. Jain N, Hung IC, Kimura H, Goh YL, Jau W, Huynh KLA, et al. The global
response: how cities and provinces around the globe tackled Covid-19
outbreaks in 2021. Lancet Reg Health Southeast Asia. 2022;4: 100031.
82. Duarte I, Ribeiro MC, Pereira MJ, Leite PP, Peralta-Santos A, Azevedo L.
Spatiotemporal evolution of COVID-19 in Portugal’s Mainland with self-
organizing maps. Int J Health Geogr. 2023;22(1):1–18. https:// doi. org/ 10.
1186/ s12942- 022- 00322-3.
83. Gould P. The slow plague: a geography of the AIDS pandemic. Oxford:
Blackwell Publishers; 1993.
84. Schneiders ML, Naemiratch B, Cheah PK, Cuman G, Poomchaichote T,
Ruangkajorn S, et al. The impact of COVID-19 non-pharmaceutical inter-
ventions on the lived experiences of people living in Thailand, Malaysia,
Italy and the United Kingdom: a cross-country qualitative study. PLoS
ONE. 2022;17(1): e0262421.
85. Google. COVID-19 community mobility reports. 2020. https:// www.
google. com/ covid 19/ mobil ity/. Accessed 17 Aug 2021.
86. Green MA, Hungerford DJ, Hughes DM, Garcia-Finana M, Turtle L, Cheyne
C, et al. Changing patterns of SARS-CoV-2 infection through Delta and
Omicron waves by vaccination status, previous infection and neighbour-
hood deprivation: a cohort analysis of 2.7M people. medRxiv. 2022.
http:// medrx iv. org/ conte nt/ early/ 2022/ 04/ 05/ 2022. 04. 05. 22273 169. abstr
act
87. Sachdeva M, Fotheringham AS, Li Z, Yu H. Are we modelling spa-
tially varying processes or non-linear relationships? Geogr Anal.
2022;54(4):715–38.
88. Neto J, Carvalho C, Letras S. Better communication with migrant commu-
nities during COVID-19 pandemic:a portuguese experience. Eur J Public
Health. 2021;31(Supplement_3): ckab164.249. https:// doi. org/ 10. 1093/
eurpub/ ckab1 64. 249.
89. Barceló MA, Saez M. Methodological limitations in studies assessing the
effects of environmental and socioeconomic variables on the spread of
COVID-19: a systematic review. Environ Sci Eur. 2021;33(1):1–18. https://
doi. org/ 10. 1186/ s12302- 021- 00550-7.
90. Wang Y, Di Q. Modifiable areal unit problem and environmental factors of
COVID-19 outbreak. Sci Total Environ. 2020;740: 139984.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in pub-
lished maps and institutional affiliations.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
1.
2.
3.
4.
5.
6.
Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users (“Users”), for small-
scale personal, non-commercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. By
accessing, sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of use (“Terms”). For these
purposes, Springer Nature considers academic use (by researchers and students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and conditions, a relevant site licence or a personal
subscription. These Terms will prevail over any conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription
(to the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of the Creative Commons license used will
apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may also use these personal data internally within
ResearchGate and Springer Nature and as agreed share it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not
otherwise disclose your personal data outside the ResearchGate or the Springer Nature group of companies unless we have your permission as
detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial use, it is important to note that Users may
not:
use such content for the purpose of providing other users with access on a regular or large scale basis or as a means to circumvent access
control;
use such content where to do so would be considered a criminal or statutory offence in any jurisdiction, or gives rise to civil liability, or is
otherwise unlawful;
falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association unless explicitly agreed to by Springer Nature in
writing;
use bots or other automated methods to access the content or redirect messages
override any security feature or exclusionary protocol; or
share the content in order to create substitute for Springer Nature products or services or a systematic database of Springer Nature journal
content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a product or service that creates revenue,
royalties, rent or income from our content or its inclusion as part of a paid for service or for other commercial gain. Springer Nature journal
content cannot be used for inter-library loans and librarians may not upload Springer Nature journal content on a large scale into their, or any
other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not obligated to publish any information or
content on this website and may remove it or features or functionality at our sole discretion, at any time with or without notice. Springer Nature
may revoke this licence to you at any time and remove access to any copies of the Springer Nature journal content which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or guarantees to Users, either express or implied
with respect to the Springer nature journal content and all parties disclaim and waive any implied warranties or warranties imposed by law,
including merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published by Springer Nature that may be licensed
from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a regular basis or in any other manner not
expressly permitted by these Terms, please contact Springer Nature at
onlineservice@springernature.com
Content uploaded by André Alves
Author content
All content in this area was uploaded by André Alves on Apr 08, 2023
Content may be subject to copyright.
