The premise of the Zero TB Cities Initiative is that concentrating resources in urban centers rather than spreading them over national or regional geographies can “drive sharp reductions in TB death rates and prevalence” (18). This study provides evidence that even within cities, especially larger ones such as Karachi, the risk of TB is likely not distributed uniformly (Fig. 7). While district-level GIS analyses have been previously carried out in Pakistan, this is the first study to investigate spatial variation in a major metropolitan area with detailed coordinates of ACF efforts (19–20). The strength of this study was the use of a very large dataset from a ACF program to achieve greater precision in the identification of TB hotspots in a city with diverse ethnicities and large variations in socioeconomic indicators. The ACF dataset allowed for inclusion of cases from the community that would otherwise been missed, thereby increasing the accuracy of our findings, relative to data captured from routine surveillance.
Our approach may allow for more actionable information that can be utilized to guide programmatic decision-making, further enhancing the Zero TB City approach. A notable finding was that nearly three-quarters of all camp locations yielded no MTB + cases, whereas only 5% of camp locations accounted for over 40% of all MTB + cases detected. These results strongly suggest a targeted approach to ACF in high-risk areas to improve yields and cost-effectiveness. A number of UCs with clustering of MTB yield were identified, particularly in the western and southern parts of the city. Clusters were also identified through the spatial point-pattern analysis using camp GPS coordinates in the western and southern regions, as well as in the peripheries of the city. These areas correspond to densely populated areas near the port, slum-dwellings in the west and peri-urban communities and villages in the outskirts of the city. It is likely that population density as well as social determinants of TB, such as crowded housing, low-income and poor nutrition contributed to higher TB risk. UCs and camp locations with clustering of low values were identified in central and eastern parts of the city suggesting lower numbers of people with active TB in these areas. These consist of locations around major avenues of the city that include commercial properties and planned middle and upper-middle income residential areas.
Previous studies utilizing passive case-finding data from Brazil, South Africa and Zimbabwe have identified areas of high TB notifications in peri-urban and lower-income areas within cities (21–23). A modelling study from Ho Chi Minh City found that four-fifths of index cases had no other reported TB cases within a 50m radius (11). These studies and our findings suggest that a useful strategy for improving cost-effectiveness of ACF may also be an avoidance of “cold-spots” or areas where previously no TB cases are reported. This information can be easily extracted from routine TB registers and involvement of local health authorities. An ACF program can therefore be targeted only in areas from where cases have previously been reported from passive case-finding while avoiding new localities that may diminish overall yields.
Further research will be required to investigate the causes for spatial heterogeneity in TB cases. This can include social determinants such as population density, poverty, household family size and type of housing (24). Health systems determinants will also need to be investigated such as number of medical facilities and number of TB testing and treatment centers. Investigating and addressing these factors would require a multi-disciplinary approach and collaboration with researchers involved in urban planning, housing and development as well as coordination with local city officials. Known clinical determinants of TB disease including nutritional deficiencies, smoking history, diabetes and HIV should also be examined for spatial heterogeneity (25–26). Such analyses can be overlaid with this analysis and modeled as predictors for spatial variation in TB detection.
Our approach can be easily replicated by other programs through the use of simple android mobile-phone applications and collection of GPS coordinates in the field. Free of cost tools such as Google Maps can be utilized to visualize color-coded clusters of camps yielding TB cases if propriety software is not available. Software code for mobile-applications utilized in this study is publicly available to support such data collection for field-teams in other settings. Revisions to the national active case-finding guidelines are also being prepared in collaboration with partners and the NTP to support the wider adoption of these methods. A similar analysis is being carried out for other cities in Pakistan where CHS operates.
There are a number of limitations in our analysis. The location of camp site was taken as a proxy for residence of the participants and this limits the internal validity of the study. While camps were carried out in communities and partnering provider clinics, it is possible that some participants were visiting the area and did not reside near the camp site or in the same UC. A random sampling of households at the UC-level will provide a better estimate of TB prevalence, however, prevalence surveys require even more significant resources than ACF. Given the size of Karachi’s population and diversity of its neighborhoods, our results support investment in a city-level prevalence survey to help identify areas for targeted ACF activities. From a programmatic perspective, however, such selection bias may be less relevant. Identified hot-spots could be marketplaces or clinics near “true” TB hot-spots and this may be sufficiently useful information for program teams if the camps consistently provide high-yields. A very limited number of camps were conducted in military cantonments that include several high-income residential areas. These areas were therefore not adequately assessed for spatial variation in TB risk.
Challenges in participant recruitment included lower representation of females. Due to cultural reasons, women may have been hesitant to take part in screening camps in public locations. Older age groups and those with disabilities may have also not taken part in screening. It is possible that such people with TB were residing near camp sites and were missed, affecting the number of hotspots identified. These constraints were however, applicable to all camps and would therefore have limited the bias towards individual clusters. Sputum expectoration and quality also proved challenging in the field and this may have also reduced the number of hotspots identified in the MTB positivity analysis. Limited sputum quantity, salivary samples, food particles and betel nut contaminants were frequent problems identified at the laboratory. People with abnormal X-rays were encouraged to visit the nearest SZ centers to deposit morning samples and provided phone-call reminders. Analyses for abnormal X-rays were included to adjust for missing testing data and these produced similar results to MTB positivity, again suggesting limited bias towards identified clusters. Additional limitations include the sensitivity of the CAD software for screening for TB and of Xpert as a diagnostic test, particularly for pauci-bacillary disease. It is likely that a number of individuals that were started treatment on clinical basis may have early-stage disease and would have converted to bacteriological positivity in the future. This may have accounted for the large proportion of camps with no positive results and limited the number of hotspots identified. Future studies may consider a low cutoff for the CAD scoring and use of Xpert Ultra to improve sensitivity.