Study area and reference population
The study area (Figure 1) comprises eleven municipalities in the central area of the Autonomous Community of the Principality of Asturias, with a total population of 705,968 inhabitants. The 2016 Municipal Register of Inhabitants was used as a reference value. It was the first year of the study sample, obtained from SADEI (Asturias Society of Economic and Industrial Studies). The individuals were categorised by age groups (less than 15; 15 to 39; 40 to 64; 65 to 84; 85 years and over) and sex. The study population corresponded to 67% of the total population of the region (1,042,628 inhabitants in 2016). The four municipalities with more than 50,000 inhabitants in the Autonomous Community were included (Avilés, Gijón, Oviedo and Siero), thus comprising 81.32% of the study population.
Study population
The health data analysed corresponded to unscheduled (urgent) hospital admissions in the hospitals of Avilés (San Agustin University Hospital), Gijón (Cabueñes University Hospital and Jove Hospital), and Oviedo (Asturias Central University Hospital). The study focused on hospital incidence of admissions for AMI and AP. The data were obtained from the Specialised Care Activity Registry RAE-CMBD - Minimum Basic Data Set, corresponding to the period 2016-2018, recorded according to the ICD code (ICD10: I-20 - I-21).
In this registry, each entry corresponded to an admission event, and included a personal identifier for each registry, sex, date of birth, date of admission, and main diagnosis. Through the personal identifier, the residence addresses entered in the SIPRES (Population and Health Resources Identification System) were obtained for their geographical allocations. The ICD10: I20-I21 records were taken from all age groups (less than 15, 15 to 39, 40 to 64, 65 to 84, 85 and over).
Population area and reference cartography
The unit of study was composed of the census tracts (CT) of the municipalities under study (558 CT), obtained from the National Institute of Statistics. Their total reference population was determined through the code of the CT contained in the data set of the Municipal Register of Inhabitants. The CT constituted the most homogeneous units in terms of population, with an average of 1,265 inhabitants for the 558 CT included.
Procedure
The health registry data was geocoded according to the portal level, using the addresses entered in each health event. Geocoding was performed using ArcGis 10.4, based on the cartographies of specific portals of the city councils of Avilés, Gijón and Oviedo, and the cartography of the project ‘CartoCiudad’ of the National Geographic Institute for the rest of the municipalities. Once the health events were geocoded, they were grouped by CT, disease, sex, and age group.
For each geographical unit (CT), we calculated the standardised admission ratio (SAR), i.e., the quotient between the numbers of observed and expected admissions. The expected admissions for the group of municipalities in the study area was calculated for each sex, using the indirect method of standardisation and the specific rates for age groups in the study period of the central area considered.
Due to the variability of the SAR, resulting from areas with little population or with infrequent health events, it was considered necessary to apply spatial smoothing methods. To that end, the smoothed relative risk (SRR) of admission for AMI and AP was calculated using conditional autoregressive models developed by Besag, York and Mollié (32). They are spatial Poisson models with random effects that take into account the spatial adjacency of the geographical units of the area. Its use is simplified using the Laplace approximation technique to perform Bayesian inference, following the integrated nested Laplace approximation procedure (33,34). Both the SAR and the SRR have been expressed in percentages.
In addition, the posterior risk probability (PP) was calculated, i.e., the probability that the smoothed risk was greater than 100. A PP value ≥0.8 indicated a statistically significant admissions excess (not due to chance). The Stata v14 and R version 3.6.1 programmes were used with the INLA library (R-INLA Project) for calculating the standardised admission ratio, SRR, and PP. The analysis of spatial clusters was performed using the Moran’s index, which measures the spatial autocorrelation between the smoothed relative risks throughout the study area, and tries to contrast the null hypothesis of the absence of global spatial autocorrelation (i.e., spatial randomness) versus the alternative hypothesis of the existence of spatial autocorrelation.
In a complementary manner, we calculated local indicators in order to detect a possible spatial autocorrelation in a certain subset of spatial units. In this way, an index could be obtained for each spatial unit studied, which made it possible to assess the degree of individual dependence of each spatial unit with respect to the others. To that end, we used the local Moran’s statistics, proposed by Anselin (35), whose interpretation is similar to that of the Moran’s index, i.e., if it is statistically significant and positive, it allows confirming the presence of a cluster of similar values around the spatial unit ‘i’. On the contrary, if it is statistically significant, but negative, there will be a cluster of different values around the nth spatial unit (spatial outliers). The results of spatial autocorrelation at the local level are presented using the local indicators of spatial association (LISA), which used the local Moran’s indices calculated for all the assessed spatial units (CT), allowing the geographical determination of spatial groupings (which occurs when a spatial unit that registers a high/low value of the variable is surrounded by spatial units that also register high/low values of that variable, i.e., high-high or low-low) and the spatial outliers, those that arise when a spatial unit with a high value of the assessed variable is surrounded by spatial units in which the variable registers small values or vice versa, i.e., high-low or low-high.