This is a Geographic Information System (GIS)-based retrospective study. Data will be extracted from the NTUH medical records database, including the date of notification, sex, birth date, residence address, etc of each tuberculosis case. Each case will be geocoded to generate spatial coordinates for mapping purpose based on the street address and postal code. To protect the privacy of these patients, the personal identification data of TB patients, including name and ID numbers, will not be encoded. The encoded data file will also be strictly kept confidential.

Disease mapping is the usual means of presenting descriptive geographic data on disease occurrence and creating accurate maps of disease morbidity and mortality. Maps convey instant visual information on the spatial distribution of disease and can identify subtle patterns which may be missed in tabular presentations. The purpose is to display variations in ill-health (for example, related to the underlying sociodemography), formulate etiologic hypotheses, aid surveillance to detect areas of high disease incidence, and help place specific disease.

In detecting the spatial patterns of the points geocoded from the addresses, point pattern analytical methods are used. Quadrat Analysis is one of them. This method evaluates a case distribution by examining how its density changes over space. The density measured by Quadrat Analysis is then compared with the density of a theoretically constructed random pattern to see if the case distribution in question is more clustered or more dispersed than the random pattern. Another one is Nearest Neighbor Statistic, which is derived from the average distance between cases and each of their nearest neighbors and captures information on cases between quadrats. Using ordered neighbor statistic can evaluate the pattern at different spatial scales. A case pattern is said to be more clustered if its observed average distance between nearest neighbors is found to be less than that of a random pattern. Ripley's K statistic is an extension of the ordered neighbor statistics, which can be used to depict the randomness of a case distribution over different spatial scales and capture the characteristic of local variations.

To measure and test how dispersed/clustered the case locations are with respect to their attribute values, e.g. socioeconomic status, spatial autocorrelation can be performed. If significant positive spatial autocorrelation exists in a case distribution, cases with similar characteristics tend to be near each other.

For polygon data in interval or ratio form, such as the tuberculosis incidence of different districts, Moran's I index, Geary's Ratio and the G-statistic can be used. Moran's I uses the mean of the attribute's data values as the benchmark for comparison when neighboring values are evaluated; Geary's Ratio is based on a direct comparison of neighboring values; G-statistic is based on the concept of spatial association or cross-product statistics and is capable of detecting the presence of hot spots or cold spots. In case of spatial heterogeneity, i.e. the magnitude of spatial autocorrelation varies over space, modified versions of the previous three statistics can be used to evaluate spatial association at the local scale.