Since 1986, as part of the Karonga Prevention Study in northern Malawi, patients in whom TB was suspected have been identified by using enhanced passive surveillance. Project staff are based at peripheral clinics and the district hospital to examine anyone with chronic cough or enlarged lymph nodes. Patients in whom TB is suspected are also identified in the course of other studies, including household visits to TB patients, although in practice most patients come to the clinic or hospital. Sputum is taken for smear microscopic examination and culture, and material from lymph node biopsy specimens, ascites, and pleural fluid is also cultured when available (11). DNA fingerprinting has been carried out on cultures from all TB patients since late 1995 (12). Cultures that macroscopically resemble M. tuberculosis are sent to the Health Protection Agency Mycobacterium Reference Laboratory, London, United Kingdom, for species identification and drug sensitivity testing. M. tuberculosis specimens are fingerprinted by using IS6110 restriction fragment length polymorphism (RFLP), following standard procedures (13). Spoligotyping (14) is performed on strains with <5 bands on the RFLP pattern. Treatment follows Malawi National TB Control Programme guidelines. TB patients were tested for HIV after counseling and if consent was given. No antiretroviral treatment was available at the time of the study (11).

Since 1997, at the time of diagnosis, all TB patients have been asked about persons they knew who had had TB, either in their family or household (at any time) or among other acquaintances (in the last 5 years) (15). Details gathered on these putative source contacts (PSCs) allowed them to be identified within the project database: ≈90% of named PSCs who were said to have had TB in the district within the previous 5 years were confirmed as having been treated for TB. In addition, all persons seen in the district during long-standing epidemiologic studies are asked about their current and past residences and their parents, allowing genetic linkages and household histories to be constructed. PSCs identified only from the epidemiologic database were included in this study if they were first-degree relatives or half siblings of the patient or if they were documented as having lived in the same household as the patient at the time that the PSC had TB.

DNA fingerprints of cultures from all case-PSC pairs were compared by computer (Gelcompar 4.1, Applied Maths, Kortrijk, Belgium) and checked visually (12). Transmission was "confirmed" if the pair had identical strains, or if the RFLP patterns differed by 1 to 4 bands and the later strain was the first or only example of the new pattern in the dataset (12,16). Since >1 PSC was identified for some patients, the analysis was repeated excluding PSCs with smear-negative results or extrapulmonary disease, and, if >1 PSC was smear-positive, choosing the most likely source of the infection by selecting the most closely matched strains and the closest contacts (e.g., contacts within the household were considered closer than nonhousehold contacts). We have previously explored RFLP pattern evolution among the first 80 such pairs with smear-positive PSCs (to 2001) (12). Here, in a larger dataset, we explored risk factors associated with a named PSC being the confirmed source of transmission and estimate the proportion of TB in the population attributable to contact with a smear-positive household or other close family member and the relative contribution of HIV-positive and HIV-negative patients to transmission.

In addition, we estimated the proportion of RFLP-defined links that can be explained epidemiologically. Strains were defined as clustered if the RFLP pattern was shared by >2 patients. The proportion of patients in RFLP-defined clusters for whom epidemiologic links were known was calculated, and any variation with cluster size or band number was investigated. Assuming 1 index case per cluster, we calculated the proportion of secondary cases within clusters for which an epidemiologically linked source could be identified (17).

Comparison of proportions used χ² tests, or exact tests when numbers were small. Odds ratios were calculated by using logistic regression. To calculate the proportion of cases in the population attributable to different types of contact (the population attributable fraction [PAF]), we adapted the formula PAF = p´(RR-1)/RR where p´ is the prevalence of the exposure (history of contact) in the cases, and RR is the relative risk of TB in those who are exposed compared to those who are not exposed (18). The expression (RR-1)/RR is algebraically equivalent to the risk difference percent, (r₁–r₀)/r₁, where r₁ is the risk in the exposed, and r₀ is the risk in the unexposed. We do not know RR, r₁, or r₀, but the risk difference percent is equivalent to the proportion of epidemiologically linked cases for which transmission from a PSC is confirmed (since this represents the proportion of cases in the exposed group that were caused by the exposure). PAF is thus calculated as the proportion of case-PSC pairs for which transmission was confirmed, multiplied by the prevalence of exposure (having a PSC) among the cases. To estimate the contribution of HIV-positive patients to onward transmission, we adjusted the relative probabilities of transmission being confirmed from HIV-positive and HIV-negative PSCs, by the proportion of smear-positive TB patients in the population who were HIV positive.

From late 1995 to early 2003, a total of 1,248 culture-positive TB patients were identified in Karonga District. Successful RFLP fingerprints were available on 1,194 isolates from 1,044 patients. After excluding 25 fingerprints because laboratory error was suspected (12), 1,029 patients had RFLP results: 74% were clustered (19). The isolates from 81 persons had <5 bands, and spoligotypes were available on 64 of these. HIV results were available for 61%, of whom 65% were positive.

In Africa, case finding for TB is generally passive. Although being a household contact of a TB patient is a strong risk factor, in Africa as elsewhere (15,20), in high-incidence settings, most cases of TB are not attributable to household contact. This finding has been demonstrated in traditional epidemiologic studies (21) and more recently by using molecular techniques (22). The apparent importance of casual contact in TB transmission is not surprising since many people exposed to a small risk can account for more disease than a few exposed to a large risk (23).

DNA fingerprinting allows direct measurement of the proportion of cases with known exposure who acquired TB from that exposure. This proportion has varied from 95% in the Netherlands (10), 70% in San Francisco (8) and 71% elsewhere in the United States (9), to <50% in Cape Town, South Africa (24), and in our study. The studies varied in whether they included smear-negative PSCs, in the way contact was defined, and in whether similar but not identical RFLP patterns were included. Smear-negative PSCs were associated with a lower likelihood of confirmed transmission in our study and in the United States (9). Workplace contacts in the United States (9) and contacts outside the family and household in our study were much less likely to be confirmed as sources of infection. The inclusion of similar RFLP patterns that are the first example of their type will increase the proportion of confirmed transmissions, though with an increased risk of false attributions of the source of infection. Even identical strains may have other origins, particularly if the strain is common. On the other hand, actual transmission may not be recognized if different strains are seen in the PSC and case because of cross-contamination or other laboratory error or because the infection in the PSC was a mixture of strains.

This analysis, like all analyses to date of M. tuberculosis transmission based upon IS6110 RFLP patterns, is based on the assumption that multiple infections are infrequent and thus that a single RFLP-defined strain reflects the infection status of a patient. A recent study from South Africa has hinted that multiple infections may be more frequent than previously assumed (25). If this is the case, then we (and all previous RFLP-based studies) have underestimated the proportions of transmission occurring within households. This assumption should not affect our estimates of relative transmission from HIV-positive and HIV-negative patients.

A lower proportion of confirmed transmission from identifiable PSCs is expected in high-incidence settings. As the incidence of TB decreases in a population, patients are increasingly concentrated in high-risk groups with particular risk factors for disease. Close contacts of patients may share many risk factors. As the risk of a close contact of a TB patient having had TB increases relative to the risk of a casual contact having had TB, the proportion of TB due to transmission from a close contact will also increase. In our study, 44% of patients with a smear-positive PSC in their household within the previous few years appear to have acquired their infection from that person. The source of the infections in persons without confirmed transmission is unknown, but two thirds of patients were part of existing clusters so the infections probably were recently acquired locally.

In Cape Town, the proportion of identifiable household contacts with confirmed transmission was similarly low (24). In that study, the proportion of TB in the community attributable to transmission within the household (from smear-positive or smear-negative contacts) was estimated at 19%. In our study, by using different methods, we estimated that 9%–13% of TB cases were attributable to transmission from smear-positive PSCs within the household or close family, and 3% from other named sources. The low proportion of cases with identified sources of infection is corroborated by conventional epidemiologic studies in this and other populations (15,21).

The low proportion of TB attributable to identifiable links is also supported by the similarly low proportion of persons in clusters who can be linked epidemiologically, both in our study and studies in South Africa (22) and India (26). In our study, the epidemiologic links were established independently of the molecular data; further links might have been found by detailed investigations of particular clusters (10), but many of these more-difficult-to-define links represent casual contact. Identifiable links have been found for a higher proportion of clustered patients in low-incidence settings (10,27,28), but, excluding the casual links, the proportion is still <50%.

One reason for the particularly low proportion of confirmed transmission in our study is the high prevalence of HIV and the effect of HIV on transmission. Our study is the first to demonstrate lower infectiousness of HIV-infected TB patients by DNA fingerprinting of epidemiologically linked case-contact pairs. The lower rate of transmission persisted after adjusting for degree of smear positivity. Although HIV status was not available for all patients, this factor should not bias this estimate.

Extrapolating the results from case-PSC pairs to the community assumes similar relative transmission patterns, but is reasonable since HIV-positive patients had similarly reduced transmission within families and households and outside them. The lower infectiousness of HIV-positive patients does not mean that they have a minor role in TB transmission, since nearly two thirds of TB patients are HIV positive. It does, however, help limit the extent of the HIV-related increase in TB in the population (1,11,29).