Study population

We included all patients with culture-positive TB notified in Shinjuku City from September 2002 to December 2013 as the eligible study population in this cross-sectional observational study. This study forms part of a population-based study on DNA fingerprinting surveillance of Mycobacterium tuberculosis in Shinjuku City that was started in 2002. Shinjuku City (18.3 km²) is one of the most populous (342 867 residents in 2018)22 cities in Tokyo, and its TB notification rate in 2016 was 33.7 per 100 000 people,23 which was higher than the rates in Tokyo and the nation (17.2 and 13.9, respectively20). Experienced public health nurses at the Shinjuku Public Health Centre (PHC) interviewed and collected information from all patients with culture-positive TB at the time of registration using a standardised questionnaire to avoid possible interviewer bias. The study variables and definitions are described in table 1.

Patient and public involvement

Neither the patients nor the public were involved in the design of this study.

DNA fingerprinting and genotype cluster

Clinical isolates from each of the enrolled patients with TB were sent to the Research Institute of Tuberculosis (RIT), Tokyo, where the TB strains were subjected to DNA fingerprinting using insertion sequence 6110 by RFLP (IS6110-RFLP) analysis.24 One clinical isolate per person was used for the clustering analysis. IS6110-RFLP and spoligotyping are the standard methods used in the Shinjuku PHC and were available throughout the study period. The Shinjuku PHC switched from RFLP to VNTR a few years ago, but the RFLP profiles of many TB cases were available. Thus, we employed RFLP due to the sufficient sample size. A genotype cluster was defined as a group of patients with TB whose isolates showed either (1) ≥6 identical IS6110 band patterns or (2) <6 identical IS6110 band patterns confirmed by identical spoligotyping patterns. The data collection and genotyping methods were previously described in detail.14

Data analysis

We calculated the genotype clustering rate by the ‘n−1 method’ according to the formula ((n−c)/N), where N is the total number of cases sampled, c is the number of clusters and n is the total number of cases in the clusters.9 We also calculated the cumulative clustering rate by calculating the clustering rate in 2002 and then adding the patients with TB every year up to 2013. The characteristics of clustered cases, which were the cases belonging to any genotype clusters, were compared with those with unique strain patterns through χ² tests. We performed univariate logistic regression to identify risk factors for genotype clustering using ORs and multivariate logistic regression using adjusted ORs (aORs). Any potential interactions were assessed using likelihood ratio tests.

Additionally, we compared the characteristics of the first two cases in each genotype cluster to identify risk factors for the development of a large cluster within 3 years. For this purpose, a cluster episode was defined as a newly arising genotype cluster in or after 2003 without any TB cases of that genotype notified prior to that year. We classified cluster episodes into the following two groups according to a system developed in a previous study10: (1) ‘large clusters within 3 years’ were cluster episodes with five or more cases (large clusters) occurring within 3 years and (2) ‘small clusters and large clusters after 3 years’ were cluster episodes with two to four cases (small clusters) and cluster episodes that became large clusters after 3 years. We identified the first two cases in each cluster episode based on the notification date and compared their characteristics between these two groups. We performed univariate and multivariate logistic regression analyses to identify predictors of the development of large clusters within 3 years.

A p value of 0.05 was set as the level indicating statistical significance. For variables with more than 5% missing values, the multiple imputation method was considered. The variables used for multivariate logistic regressions were selected by the stepwise maximum-likelihood estimation with a significance level of less than 0.2. We used Stata version 12 for the statistical analyses. Written informed consent was waived because DNA fingerprinting analysis forms part of the routine TB control activities in Shinjuku City. However, oral informed consent was obtained after the PHC staff provided a thorough explanation of the study objectives and confidentiality.

Study population and clustering rate

In total, 1885 patients with TB in Shinjuku City were notified during the study period and 1310 were culture-positive cases (figure 1). Of these, 285 patients were excluded from the analysis, mainly due to the unavailability of culture-positive isolates and the lack of implementation of RFLP. As a result, 1025 (78.2%) patients were included in the analysis. The figure 2 shows the cumulative number of patients with TB and the clustering rates from 2002 to 2013. The number of TB cases gradually increased over the tested decade. In contrast, the cumulative clustering rates sharply increased in the first 4 years, from 10% in 2002 to 28% in 2005, with an average per cent change of +43%, and then continued to increase at a slower rate, from 30% in 2006 to 39% in 2013, with an average per cent change of +4.2%.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Number of reported cases of TB, including culture-positive cases, strain-typed cases and genotype clusters, in Shinjuku during 2002–2013. RFLP, restriction fragment length polymorphism; TB, tuberculosis.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Cumulative clustering rate (restriction fragment length polymorphism, Shinjuku 2002–2013).

We identified a total of 113 genotype clusters consisting of 515 patients (figure 1). The genotype clustering rate was 39.2%, and the average cluster size was 4.56 cases (range 2–30). Fifty-seven (50.4%) genotype clusters consisted of only two patients with TB, and 36 (31.9%) genotype clusters had at least five patients with TB. We further investigated the homelessness status and place of birth of the patients in the genotype clusters. Of the 113 genotype clusters, 45 (39.8%) comprised only non-homeless individuals, seven (6.2%) included only homeless individuals and 61 (54.0%) contained both homeless and non-homeless individuals (mixed cluster). We compared the characteristics of the non-homeless patients in the clusters of only non-homeless patients with those in the mixed clusters, and although the finding was not statistically significant (Pearson χ² test, p=0.17), the proportion of non-homeless patients receiving public assistance in the latter group (13.8%) was higher than that in the former group (8.8%). No differences in sex, age and place of birth were found between the two groups. Of the 113 genotype clusters, 94 (83.2%) consisted of only individuals born in Japan, two (1.8%) consisted of only foreign-born individuals, and 17 (15.0%) consisted of both individuals born in Japan and foreign-born individuals.

Factors associated with genotype clustering

The clustered cases were significantly more likely to consist of male individuals (OR=1.62, 95% CI 1.20 to 2.19), Japan-born individuals (OR=3.74, 95% CI 2.25 to 6.44), individuals receiving public assistance (OR=2.25, 95% CI 1.69 to 3.00), homeless individuals (OR=2.45, 95% CI 1.80 to 3.34), individuals who misuse alcohol (OR=1.37, 95% CI 1.02 to 1.83), individuals engaging in full-time work (OR=1.53, 95% CI 1.15 to 2.05) and part-time/daily work (OR=2.29, 95% CI 1.45 to 3.61) and jobless individuals aged 15–59 years (OR=2.05, 95% CI 1.43 to 2.94) (table 2). A significant interaction among the explanatory variables was not detected. The multivariate analysis demonstrated that the factors associated with genotype clustering were age <40 years (aOR=1.73, 95% CI 1.23 to 2.44), born in Japan (aOR=3.90, 95% CI 2.27 to 6.72), working full-time (aOR=1.63, 95% CI 1.17 to 2.27), having part-time/daily work (aOR=2.20, 95% CI 1.35 to 3.58), receiving public assistance (aOR=1.81, 95% CI 1.15 to 2.84) and homelessness (aOR=1.63, 95% CI 1.02 to 2.62) (table 3).

Factors associated with large genotype clustering within 3 years

We identified 104 genotype cluster episodes according to the definition. Of these, 14 were ‘large clusters within 3 years’, which was equivalent to 13.5% (14/104) of all the genotype clusters and 48.3% (14/29) of the large genotype clusters, and 90 clusters were ‘small clusters and large clusters after 3 years’. The univariate analysis indicated that clusters with registration intervals of 0–2 months were 9.51 times more likely to become large genotype clusters within 3 years compared with clusters with registration intervals of ≥12 months (table 4). After selecting variables using the stepwise method, only the ‘registration interval’ variable remained for the multivariate model.

Clustering rate

We identified 515 genotype clustered cases and estimated a clustering rate of 39.2%. The rate was the same as the pooled clustering rate (40.9%) obtained in a previous meta-analysis of population-based studies of countries with a low TB incidence19 but differed from previous estimates obtained in Japanese studies, which were 27.6% in Shinjuku and 24.6% in Osaka.14 25 Because the meta-regression analysis clarified that longer study durations are associated with an increased clustering rate,19 this difference could be due to shorter study durations combined with the smaller sample sizes of the previous studies (388 patients in 5 years and 195 patients in 1 year, respectively). In our study, as expected, the cumulative clustering rate rapidly increased in the first 4 years and increased more slowly thereafter, which is similar to the trend observed in the previous studies.26 27

Factors associated with genotype clustering

Our results indicated that the clustered cases were more likely to have socioeconomic predictive factors, namely, being homeless, receiving public assistance and having an unstable job, at the time of TB diagnosis. Similarly, previous studies suggested that being homeless significantly contributed to clustering in Shinjuku City14 and other counties.19 In our study, more than half of the genotype clusters were mixtures of non-homeless and homeless patients. Moreover, the non-homeless patients in the mixed clusters tended to be financially unstable and a higher proportion of these patients were receiving public assistance compared with the proportion among clusters of only non-homeless cases, which could imply that relatively poor non-homeless patients share activity spaces with homeless patients, such as urban areas around the large train stations that were reported to be significant hotspots for homeless patients in Shinjuku City.28 These findings could suggest that contact investigations of homeless patients with TB need to be actively expanded to possible contact persons who are not homeless, particularly those who are facing financial difficulty.

A meta-analysis based on studies conducted in European countries where foreign-born patients substantially contribute to TB epidemiology found that the proportion of mixed clusters composed of native and foreign-born patients ranged from 0% to 36.5% and concluded that foreign-born patients did not have a significant influence on TB in the native population.29 In our study, the proportion of mixed clusters (15.0%) fell into this range. Thus, the impact of TB transmission between native and foreign-born populations likely remains limited in this urban setting.30 However, considering the recent increase in immigrant patients with TB in urban cities, TB transmission between native and foreign-born populations needs to be closely monitored.

Factors associated with large genotype clustering within 3 years

A shorter registration interval (≤2 months) was identified as a significant predictor of the development of a large genotype cluster within 3 years, which is compatible with findings of previous studies conducted in the Netherlands and London.10 12 Therefore, when patients with TB with identical genotypes have shorter registration intervals, a thorough active case findings need to be performed to investigate the potential infection sources and infected patients in order to prevent further transmission. However, it is difficult to assume that the first patient infected the second patient because a window of 2 months appears too short. Thus, we believe that a true but unidentified first TB case was not identified in our study. A cluster episode was defined as a cluster without any TB patients in 2002 and at least two patients with identical genotypes in and after 2003. Therefore, a possible true first TB case might have been registered before 2002, which was outside of our study period, or registered outside of Shinjuku City.

Limitations

Our study has some limitations. First, the study population consisted only of patients with TB living in Shinjuku City. Considering the large population flow in and out of the city, as mentioned above, we potentially missed patients living outside of the city who shared TB strain types with patients living in the city. In fact, previous Japanese studies reported clusters with patients with TB living across broad geographic areas.31 Consequently, we may have underestimated the identified genotype clusters. Second, even the existence of patients with TB with identical genotyping patterns may not suggest recent transmission if the strain is a nationwide endemic TB strain,32 which could have led to an overestimated clustering rate. Third, IS6110 RFLP has relatively lower discriminatory power compared with VNTR33 and whole-genome sequencing,34 35 which might have led to overestimation. Lastly, information of epidemiological linkage among patients with TB was not available in our study. Therefore, we could not assess and discuss the current practices involving epidemiological investigations done by the public health centre, which could weaken the programmatic implications of our results.