METHODS

The island of Gran Canaria in Spain has a population of 713 768, nearly half of whom live in the city of Las Palmas. All people have access to the public health service without charge, including immigrants and the homeless.

The study group consisted of all patients with TB confirmed by culture between 1 January 1993 and 31 December 1996 in either of the two referral hospitals on the island (Hospital Insular (HI) and Hospital de Gran Canaria “Dr Negrín” (HN)).

Medical records were systematically reviewed for demographic data (age, sex, race, country of birth), risk factors for TB (homeless, tobacco and alcohol abusers, intravenous drug addiction, and contact with tuberculous patients), and clinical and microbiological characteristics of TB. HIV tests were performed on all patients accompanied by pre- and post-test counselling.

Isolates of M tuberculosis from every patient included in the study were stored at –40°C in Dubos liquid media before being subcultured on Lowenstein-Jensen media and incubated at 37°C in an aerobic atmosphere with 5–10% CO₂. An isolate from each of the patients was genotyped with RFLP-IS6110³; 85 strains from the period 1991–2 (32.5% of all isolates in that period) were also typed.

Fingerprint patterns of the isolates were analysed with the Gel Compare 4.0 (Applied Math, Kortijk, Belgium) software program and subsequently by visual examination of the cluster patterns. A cluster of M tuberculosis isolates was defined as two or more isolates revealing an identical number and location of IS6110 when five or more copies of IS6110 were present.^6,9,17 We assumed that one patient from each cluster corresponded to the index case origin of the infection,^4,5,9,11 so that the number of patients in a cluster is equal to n – 1. Patients with non-matching RFLP-IS6110 were considered non-clustered. Following RFLP-IS6110 analysis, the data for all patients included in the clusters were further reviewed to look for evidence of epidemiological linkage.

Data were analysed using computer software (EPI INFO 6 and SPSS 7.0 statistical programs). Student’s t test and χ² test were used to assess univariate risk factors for being in a cluster and also to compare patients who were genotyped by RFLP-IS6110 with those in whom this study could not be performed. They were then included in multivariate logistic regression model analysis. Significance was defined as p<0.05. Relative risks were calculated from regression estimates based on the probability test for the likelihood ratio statistic; 95% confidence intervals were based on the estimated variance of the regression coefficients.

RESULTS

On the island of Gran Canaria 231 cases of TB were diagnosed (214 confirmed by culture) in 1993 (incidence rate 32.4 per 100 000 inhabitants), 210 (185 confirmed by culture) in 1994 (incidence rate 29.4), 176 (142 confirmed by culture) in 1995 (incidence rate 24.6), and 198 (178 confirmed by culture) in 1996 (incidence rate 27.7). A total of 719 culture positive isolates were therefore obtained during the 4 year period of the study (527 in NH and 192 in HI). In 129 of these (75 in HN and 54 in HI) subcultures of M tuberculosis were not available for DNA fingerprinting (17.9%). Following RFLP-IS6110 evaluation of the remaining 590 patients, 24 isolates were excluded because they had fewer than five copies of IS6110. The final cases studied thus consisted of 566 patients (437 in HN and 129 in HI):179 in 1993, 148 in 1994, 110 in 1995, and 129 in 1996; 393 were male and the mean (SD) age was 40.3 (17.4) years (range 1–87). Eighty eight patients (15.5%) were HIV seropositive.

Table 1 shows a comparison of those with and without RFLP-IS6110 analysis. There was no association with sex, age group, HIV status, being an immigrant, site or type of disease, being homeless, having been in prison, being an intravenous drug addict, or having contact with a patient with TB. Subjects more likely to have had RFLP-IS6110 analysis included those who were tobacco smokers and those abusing alcohol. Some cases did not have certain variables recorded on their record—age (n=67), smoking habit (n=86), alcohol abuse (n=86), and TB contact after contact tracing procedure (n=275).

Of the 566 strains typed by RFLP-IS6110, 409 (72.3%) were grouped in 78 clusters. The number and size of each cluster is shown in table 2. The largest cluster consisted of 75 patients, the second 49, and the third 20. Two other clusters contained 16 and 15 patients each. The RFLP-IS6110 patterns of the largest clusters are shown in fig 1. Assuming that one patient from each cluster corresponds to the index case origin of the infection, of the 409 patients associated in clusters 78 might be excluded as cases of recent transmission of TB. The remaining 331 patients (58.5%) are therefore those judged to have had recent transmission of TB between 1993 and 1996. Recent transmission was calculated per year and no significant differences were found between years: 131 patients in 31 clusters (73.2%) in 1993, 97 in 20 clusters (65.5%) in 1994, 79 in 17 clusters (71.8%) in 1995, and 102 in 11 clusters (79.1%) in 1996.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Results of RFLP analysis of M tuberculosis strains isolated from three or more patients.

To identify risk factors associated with recent transmission, the 409 clustered patients were compared with the 157 patients not clustered. Complete epidemiological information was available for 399 patients (70.5%). In 167 patients we had some difficulties in obtaining some of the epidemiological characteristics, most of them diagnosed in HI. The comparison of clustered and non-clustered patients is shown in table 3. In univariate analysis only age was associated with clustering (p=0.02), with those less than 45 years of age being significantly more likely to be clustered (relative risk (RR) 1.51, 95% CI 1.01 to 2.25). The association with age disappeared when age was included in a model with other variables in multivariate logistic regression analysis, and none of the other variables was found to be significantly associated with clustering. In spite of the high prevalence of HIV infection in the population with TB (15.5%), there was no greater clustering in the HIV positive population than in the HIV negative population.

Additional studies on the 409 patients included in the 78 clusters were carried out, although epidemiological linkage could only be established in 147 (35.9%). No epidemiological connections were found in patients with a unique RFLP-IS6110 pattern. The two largest clusters differed from all other clustered cases in being more likely to have been in prison (RR 2.79, 95% CI 1.44 to 5.38), to be intravenous drug addicts (RR 2.77, 95% CI 1.68 to 4.55), to be immigrants (RR 2.05, 95% CI 1.01 to 4.18), and to be HIV positive (RR 1.77, 95% CI 1.15 to 2.73). The largest cluster of 75 patients (cluster 1) was the basis of a previous publication¹⁸ and differed from the second largest in being immigrants, intravenous drug addicts (RR 8.36, 95% CI 2.08 to 33.53), and in being HIV positive (RR 3.00, 95% CI 1.23 to 7.30). An epidemiological linkage was found in 68% of the cases.

Cluster 2 comprised 49 patients with an identical RFLP-IS6110 pattern (seven bands of IS6110, fig 1). Seven patients were diagnosed in 1993, nine in 1994, 13 in 1995, and 16 in 1996. In 63% of the patients it was possible to establish an epidemiological linkage. This cluster did not include immigrants and most of the patients were natives of the island. The cluster was spread throughout five different areas of the island, most of them of a low socioeconomic level. This RFLP-IS6110 pattern was also found in three of 85 patients (3.5%) diagnosed during the 1991–2 period, suggesting that these strains were in the island before 1992. The index case of the cluster was probably an alcoholic HIV negative patient from Las Palmas who adhered to a directly observed therapy programme¹⁹ and was cured 6 months later. Epidemiological connections were found in two patients related to the index case. They disseminated the disease to another two areas of the city where the cluster was found.

Cluster 3 included 20 patients with an RFLP pattern of 10 bands of IS6110 (fig 1). Ten strains were isolated in 1993, four in 1994, two in 1995, and four in 1996. An epidemiological linkage could be established in 70% of the patients, most of them belonging to a cluster found in Gáldar (a city on the north coast of the island) and in Las Palmas. This RFLP-IS6110 pattern was found in two of 85 patients (2.3%) studied during the years 1991 and 1992, suggesting that the strains were in the island before 1992. The index case of the cluster might have been an alcoholic (HIV negative) patient with poor compliance with treatment. Originally from Gáldar, he lived for some periods in Las Palmas where most of the other patients of this cluster were found.

There was another cluster containing 16 patients which included two homeless people. No epidemiological connection was established between these patients. The specific RFLP-IS6110 pattern was present in three strains out of the 85 studied from 1991 to 1992. The cluster containing 15 patients included four HIV positive patients (26.7%) and one who was homeless. An epidemiological linkage was found in 13.3% of the patients and the RFLP-IS6110 pattern was shared by 14 isolates from the 85 patients (16.5%) in the period 1991–2.

Of the 566 patients included in the study, only 111 (19.6%) identified previous contact with other TB patients. Eighty one (72.9%) belonged to one of the 78 clusters studied: 13 to cluster 1, 15 to cluster 2, four were found in a cluster of 20 people, and two in a cluster of 15 patients. Forty clusters consisting of two patients each were further studied by conventional contact tracing. An epidemiological linkage could be found in only three (7.5%) of them. There was no association between the size of the cluster and the probability of having a record of TB contact.

DISCUSSION

The development of new molecular methods to genetically differentiate M tuberculosis strains provides a useful epidemiological tool. The first described method for differentiating M tuberculosis isolates used mycobacteriophages^20,21 or enzymatic characterisation.²² Although useful for the study of specific outbreaks,^20–22 these techniques had very low discriminatory power. In the last 10 years DNA fingerprinting of M tuberculosis isolates has increasingly been used in epidemiological studies. These techniques primarily involve RFLP analysis using the insertion sequence IS6110.³ The RFLP-IS6110 methods have proved to be useful for investigating TB outbreaks,²³ and their discriminatory power has raised some questions on the interpretation of TB transmission based on classical epidemiological studies.² For example, molecular epidemiology studies in the US have estimated a recent transmission rate of more than 30% of TB cases,^4,5 considerably higher than the generally accepted figure of 10%.² The percentage of patients found in clusters was 35% in Amsterdam,⁸ 25% in Paris,²⁴ and less than 20% in Zurich.²⁵ More surprising are the results in Denmark⁷ where 49% of patients were associated with a cluster, a figure that reached 79% in Greenland. This contrasts with Norway²⁶ with a substantially lower proportion of cases in clusters. These percentages are similar to those described for developing countries where the epidemiology of TB is not under control.^9–15 In South Africa 30% of cases are clustered,⁹ a rate lower than the 39.2% found in Ethiopia¹⁰ or the 41% in Tunisia,¹¹ with a 10 times lower incidence of TB than in Ethiopia. On the other hand, in developing countries some studies have shown that recent transmission is lower than in industrialised countries.^12,13 These discrepancies between molecular and traditional epidemiological approaches require detailed analysis and evaluation. This study of the total population living on an island with a moderate incidence of TB may help to clarify this controversy.

Our study was performed over one of the longest periods (4 years) described to date,^{4,5,7,25,27–29} together with the study by Bauer et al⁶ in Denmark and Greenland. In our study 72.3% of 566 patients were grouped into 78 clusters. Even if we eliminate one patient per cluster (index case),^4,5,9,11 the clustered patients represent 58.5% of the total number of patients with TB during the period 1993–6—that is, the percentage of TB due to recent transmission—which is similar if we calculate it per year. This is the highest rate recorded to date,^{4,5,7–15,30} including countries with a higher incidence of TB,^9,10 and is similar to that reported by Bauer et al⁶ in the study carried out in Greenland.

The high rate of recent transmission found in Gran Canaria is unexpected since there is free and universal access to the public health service and excellent detection and cure rates,¹⁶ even in marginal population areas.¹⁹ However, we believe that our high percentage of clustering does not provide absolute evidence of recent transmission, as has been suggested by previous studies.^17,27,29 This is supported by the following: (1) our total number of TB patients¹⁶ is against a high rate of recent transmission; (2) the good detection and cure rates in the last years¹⁶; (3) only in 35.9% of the patients associated with clusters was it possible to establish an epidemiological linkage, which is similar to other studies^{4,8,9,25,27–29}; and (4) only in 7.5% of the patients belonging to clusters of two patients was it possible to show an epidemiological connection.

Molecular methods have been convincingly shown to be very useful in outbreak studies (including those concerning multidrug resistant strains²³) as well as in the evaluation of laboratory cross contamination,³¹ contact tracing studies,³² and in the differentiation between exogenous reinfections and endogenous reactivation.^33,34 However, in population based studies the number of variables is higher and the results of clustering should be interpreted with caution. When only conventional epidemiological contact tracing is considered, it is possible to underestimate cases of recent transmission, particularly whether the contact is sporadic or the tracing is non-exhaustive.²⁸ On the other hand, if all the cases in a cluster, as defined by RFLP-IS6110, are assumed to result from recent transmission, then this figure could be overestimated. Yang et al²⁷ compared results from different US states and found that some M tuberculosis strains had an identical RFLP-IS6110 pattern in different states without evidence of recent transmission. This could explain the high rate of clustering found in our study, as in others carried out on islands.⁶ In these closer communities it is possible to find some highly prevalent M tuberculosis strains, which may explain the discrepancies between the results obtained using conventional epidemiological methods and RFLP-IS6110.^{4,8,9,17,25,28,29} It may be necessary to use complementary typing methods to assess recent transmission reliably.³⁵

In our study the larger clusters had a considerable impact, with the largest comprising 13.2% of the patients and the second largest 8.6%. This suggests the presence of prevalent microorganisms with a high capacity for infecting and developing the disease, as has been previously suggested.^4–6,36 Cluster 1 (75 patients) is one of the largest reported so far.^{4,5,7–10,24,30} The only comparable clusters have been described in Greenland,⁶ a region that could have been influenced by similar factors to those in this study. The impact of this single strain on the epidemiology of TB on our island may be important. This highly prevalent strain entered Gran Canaria in 1993 from an African immigrant. It appeared to disseminate among susceptible populations, including HIV positive subjects, drug abusers and immigrants.¹⁸ As long as an epidemiological linkage was demonstrated in 68% of the cases, it could be assumed that this particular cluster could be related to recent transmission. Transmission of particularly infectious strains could therefore be possible after a sporadic contact, as has previously been suggested.²⁸ In other clusters, however, the presence of prevalent strains on the island for a long time makes it difficult to assume that these are related to recent transmission.³⁷

We conclude that the rate of recent transmission in our community is high, which indicates the need to increase control measures, mainly in the immigrant population. To assess compliance with treatment, legal measures may be necessary to help to control the disease.³⁸ RFLP-IS6110 is a good method for studying the behaviour of TB in a community, especially as a complement to the classic epidemiological approach and for identifying highly infectious strains.