Using the Kulldorff’s scan statistical analysis to detect spatio-temporal clusters of tuberculosis in Qinghai Province, China, 2009–2016

Tuberculosis (TB) is an infectious disease caused by Mycobacterium tuberculosis. Over 80% of the new TB cases, globally, were reported in developing countries. According to a World Health Organization report, the TB burden of China is the second largest in the world [1]. In recent years, although the Chinese Government has paid an increasing amount of attention to the control of TB, prevention measures are still insufficient, especially in areas with inadequate medical resources, such as Qinghai, a province where most of the population suffers a high risk of TB even now [2].

A large number of studies on the spatial and temporal distribution of TB have demonstrated that TB has a highly complex dynamics and is spatially heterogeneous at the provincial, national, and international levels during certain periods of time; however, the variations in small area are always be ignored by using a relatively large scale [3–6]. In our previous study, the Moran’s I spatial autocorrelation analysis method was used to analyze the TB incidence data from 2009 to 2013 in Qinghai and found that the distribution of TB in this province was not random [7]. However the Global Moran’s I spatial autocorrelation analysis only evaluates the distribution characteristics of the disease in several specific time points. Moreover, this method can not estimate the risk level of high-risk cluster areas [8–11]. It is a known fact that time is a critical confounder that might directly bias the determination of the high-risk regions of TB. The Kulldorff’s space-time scan statistical method can detect the distribution characteristics in both the temporal and spatial axes, bringing them closer to real-world conditions [12–15]. Additionally, the relative risk of disease in a cluster area can be estimated by comparing it with the area outside the cluster area. This method has been used wildly in the epidemiology studies of infectious diseases [16–19].

Analyzing and evaluating the spatio-temporal patterns and trends of TB in Qinghai is necessary for TB control and elimination. In this study, our aim was to use the Kulldorff’s scan statistical analysis to explore the spatial, temporal, and space-time dynamics of TB at the county level in Qinghai.

In this study, spatial patterns and the secular trends of TB in Qinghai from 2009 to 2016 were examined using the Kulldorff’s scan statistical analysis. To the best of our knowledge, no other similar study has been done in this area. Our study demonstrated that there was a significantly space-time clustering in distribution of TB incidents in Qinghai province. The high-risk areas were mainly concentrated in the southwest Qinghai, and the temporal clusters were mainly concentrated in spring and early summer.

Kulldorff’s retrospective scan statistics take multiple testing problems into account, which is known as the most powerful method for evaluating geographical and temporal distribution by using routinely collected data [18]. This method has been used worldwide to detect the clusters of diseases [2, 4, 5, 30, 31]. As is known, in the temporal and spatial model, selection of a suitable time window and spatial window was very important for model identification. Currently, there are two methods for selecting the size of spatial window: one is based on the geographical area, and the other method is based on the population size covered by the scanning area [16]. In this study, we used the radius of population coverage, because deviation of population in different counties was very large. Similarly, size of time scanning window is another important parameter for analysis. Generally, the window size of time was set as 50% of the entire time period of the study. However, there exists some evidence with regard to whether or not this window size is reasonable [32]. Yue Ma et al. conducted a simulation study to explore how to choose an appropriate scanning window. They found that the window might be too large to include the low-risk area if the window covered 50% of the population [23]. Therefore, this situation might lead to a high false positive rate. However, the window which covered a smaller population might be too small to detect the real high-risk area, and the high-risk area would be separated. Thus, the high false negative rate would be an issue. Tango and Takahashi suggested that when using the irregular scan statistic to detect the aggregated region, the coverage area of a single region should not be more than 15% of the whole study area [24, 25]. In addition, several studies also suggested choosing an appropriate window which could identify the cluster areas with less overlap [4]. Based on the Tango’s criteria, we analyzed the data for many times by using one window value at a time. Finally, we selected the temporal window covering 30% of whole study period and the spatial window covering 7% of the population at risk. And the overlapping of the identified high-risk clusters was not observed.

Our temporal scanning results indicated that there was a high-risk period for TB incidents every year and during the entire 8 years, which mainly occurred in spring and early summer, from January to May. As is known, during winter, the reduction in exposure to ultraviolet rays from sunlight and the poor ventilation in indoor settings may increase the incidences of TB infections. Additionally, in the case of infectious diseases, time is needed for the symptoms to develop and patients may lack the knowledge on where to seek care for TB. All these factors may delay TB diagnosis and treatment [4]. Some studies reported that the average incubation period of TB infection ranges from 4 to 8 weeks, with a 2-month interval from the appearance of symptoms to medical diagnosis [33]. Therefore, the high-risk periods in spring and early summer complied with the disease characteristics. Such seasonal patterns were consistent with the previous studies done in Yunnan province, where is the registration peak of TB cases during spring [34].

Additionally, spatial scanning results displayed statistically significant 5–7 cluster areas for TB diseases in Qinghai each year, which were similar to the results of our previous study [7]. Compared with the median incidence rate of TB (69/100000) out of the cluster areas, TB incidence in our identified TB cluster was higher than 263/100000. This indirectly suggested a relatively high sensitivity of this scanning method. The spatio-temporal model used in this study simultaneously considered time and space distributions. Compared with the separated spatial scanning model and temporal scanning model, the time-space scanning makes a conclusion more closely to the real-world situation. Using this model to detect the spatio-temporal distribution of TB in Qinghai, from 2009 to 2016, we found that the high-risk counties were concentrated in the southwest Qinghai, from September 2014 to December 2016. During this period, the risk of TB infection in these areas was obviously higher than in other areas, especially in the Zhiduo County. In these areas, the inhabitants have very low income, as well as poorer living conditions and sanitation compared to the eastern region of Qinghai. Many studies have showed that poverty is one of the most important social factors responsible for the high prevalence of TB, and also the socio-economic status may contribute to the high risk of TB [4]. Our result indicated that further prevention and special TB control strategies should be considered in relation with the economical and sanitary level in the clustered areas.

Our study also demonstrated the usefulness of spatial and temporal clustering analysis using the ArcGIS and SaTScan to identify the significant space-time clusters of TB in Qinghai. This could be used to provide strategies for TB prevention at the county level. However, the study had limitations on analysis. First, it is important to note that the data were analyzed at the county level, which is not the smallest unit of administrative regionalization. Thus, we may exclude several critical factors. Second, the influence of weather and socio-economic factors were not included in this study.

Our study analyzed the spatial, temporal, and space-time clusters of TB incidents at the county level in Qinghai, from 2009 to 2016, using the Kulldorff’s retrospective scan statistic methods. The spatial and temporal clusters were statistically significant every year, and the space-time scanning result indicated eight high-risk areas for TB incidents which were predominantly located in the southwest Qinghai. These results suggested that it is urgent to establish the preventive and controlling strategies to decrease the TB incidence in Qinghai by Qinghai government and the Center for Disease Control and Prevention.