Descriptive analysis
During the study time, from the total 209,935 deliveries performed in public hospitals of KRP, 74,291 deliveries were done by CS (35.39% of the total). The average age of mothers was 32.33 years in CS time, but the age followed some geographical variations (Fig. 2). For example, the highest average age was in Joghatai (36 years), and the lowest average was in Firouzeh county (29.20 years). Among the five age groups (15-19, 20-24, 25-29, 30-34, and >35 years) of mothers who had a CS experience, the highest portion of CS (37.64%) was observed in the age group of 35 years and older. The lowest CS rate (1.48%) was related to the age group of 15 to 19 years.
According to figure 3, out of 30 counties, 11 (36.7%) counties experienced high rates (mean = 4,670 and SD = 4553 per 100,000 women of childbearing age). The rate of CS decreased from 862 per 100,000 women in 2016 to 844 per 100,000 women in 2020. However, the mean rate of CS was higher in 2017 (1,079 per 100,000 women of childbearing age) among the other years. These results confirmed that most counties with high rates (> 5,000 per 100,000 women of childbearing age) were in the east, southeast, and west.
Purely temporal and spatial clusters
Figure 4A indicates that high-rate time clusters of total CS were predominantly distributed (OE=1.18, RR=1.34, LLR=799.97, p<0.05) between 2017 and 2018. Global Moran’s I statistics for CS deliveries based on EBS rates (Moran’s I=0.247, z-score=2.46, p<0.05) revealed that the spatial autocorrelation was significant, and the null hypothesis was rejected as CS incidence rates were spatially clustered (Fig. 4B). Figure 5 shows the purely spatial pattern of CS incidence rates based on the Poisson probability model of scan statistics. Five most likely clusters (RR>1, p<0.05) (hot spots) were identified that were distributed heterogeneously in the study region, and 14 of the 30 counties were located in these 5 clusters.
Spatiotemporal clusters
The study region's statistically significant high-rate CS spatiotemporal clusters were mainly south to north (Fig. 6). Based on the 50% maximum window size, altogether 13 locations (counties) were classified as high-rate areas in different time periods (RR > 1, p < 0.05) (Fig. 6). For example, cluster one formed from 2017 to 2018 in the southeast, while cluster three was in the north between 2016 and 2017.
Spatial variation in temporal trends
According to figure 7, two high-trend CS clusters were found. These clusters included the counties with the highest variations compared to counties inside and outside. The first cluster formed in the north, and the second cluster stood in the southeast of the study area. In the first cluster, one location was in high-trend statistically significant clusters (LRR = 453.01, RR= 1.11, p-value < 0.05). In the second cluster, there were eight statistically significant high-trend clusters (LRR = 31.76, RR= 1.43, p-value < 0.05). It is clear that CS rates were increasing overall, but not in the same way in all areas. For example, cluster one experienced an average annual growth of 61.3%, but this growth was 3.56% in cluster two.
Interaction analysis
Figure 8A shows the optimum distance from the origin (counties centroids) to destination (hospitals) based on closest facility and Thiessen polygons methods. The Thiessen polygons boundaries indicate the estimated coverage of the service area of each hospital according to its capacity. The hub-distance (shown as red lines) indicates the Euclidean linear travel distance (in km) from each county centroid (origin) to the nearest hospital (destination). Ideally, people in each county should go to the nearest hospital for CS. The mean optimal was determined as 48.41±36.2 km (Fig. 8A). The maps show a significant difference between the assumption of optimal access (Fig.8A) and the actual flows (Fig.8B) regarding access to the CS facilities. The actual mean distance is about 153.7±152.4 km, three times more than the optimal distance to access CS-related hospitals.
Figure 9A shows the capacity (number of staff) weighted heat map of hospital care facilities (intensity per sq. km) as CS service providers in the study region. The dark color and low range vales (< 0.001 per sq. km) show the low-density areas in this map. Bright yellow color with a high range value (> 0.005 per sq. km) indicates high-density areas (hotspots) and are located in northeast Mashhad County (the capital of the province). The figure indicates spatial inequality in the distribution of hospital care facilities. Global Moran's I showed the positive spatial autocorrelation to clarify the findings, which indicates a strongly clustered pattern and unequal allocation of CS hospitals in the study region. Figure 9B shows the result of the gravity model as a heat map, according to this map, the same areas form hotspots (light-yellow areas) in CS interaction flows. For example, hospitals 4, 10, 13, 14, and 18 located in Mashhad County with high normalized gravity intensity values (GI > 0.02).
The gravity model shows that the intensity of linkages and the capacity of hospitals (Fig.8A) have a profound effect on the CS flows in the study region. We performed the Pearson test between CS rates and hospital capacities to prove this. Test results showed a strong significant correlation at the 0.01 level (r=.726 and p.<0.05) as a strong relationship between CS rates and hospital capacity. For example, hospital 1 with a capacity of 3,518 personnel located in the hotspot area. The gravity model only reveals the linkages and importance of the facility capacity in the interactions and does not clarify the role of distance in the interaction between origin and destination. The distance between origins and destinations was measured using the hub-distance lines method to respond to this limitation. Figure 10 shows the distance between origins and destinations in km. According to the analysis, the mean distance between county centers and hospitals in the entire study region was 153.7 km. The results show that most hospitals located in the hotspot areas have a shorter mean distance to county centroids (mean distance<153.7 km). For example, hospitals 3,4,13,14,17, and 18, were at an average distance less than 139.32 km from the county (population centers) centroids. With more investigation on the gravity model results, the Pearson test was used in two steps. In the first step, the relationship between the gravity intensity of each county centroid and the closest hospital was measured. The test results showed a significant, strong, and inverse correlation (Sig.<0.05, r = -.604). Therefore, the CS rates increased in the destination facility when distance decreased. In the second step, the same test was applied to measure the association among distance between entire county centroids (origins) GI and all hospital care facilities (destinations). The test results show that the correlation was significant and inverse but not strong (Sig.<0.05, Pearson test value = -.17). It can be concluded that close distance has a direct effect on increasing spatial interactions.