INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Homelessness is a major factor in the resurgence of tuberculosis in the United States (1). Many cities have developed large populations of "true homeless" people living in streets or shelters, and "marginally housed" persons living on public support in single room occupancy (SRO) or residential hotels. Two and a half percent of adult Americans reported being homeless during the last 5 yr, or about five million people (4).

A high proportion of U.S.-born tuberculosis cases are in the homeless. In San Francisco, we found that among 399 U.S.-born San Francisco cases interviewed between 1993 and 1996, 47% had been homeless or unstably housed within the previous two yr (unpublished data). Although tuberculosis has declined in the United States in recent years, the homeless remain a high-risk population (5).

In San Francisco, there are at least 10,000 homeless and marginally housed people. We screened 2,774 of them for HIV and tuberculous infection in the San Francisco General Hospital (SFGH) Homeless Study, 1990-1994; 9% were seropositive for HIV and 32% were tuberculin skin test (TST) positive (6). The latter rate is similar to that among household contacts of newly diagnosed cases of smear-positive tuberculosis (7) and suggests that there was extensive transmission of M. tuberculosis among the homeless. Because of coexisting HIV infection, there is a high potential for outbreaks among the homeless (8, 9), and indeed outbreaks have been reported in shelters in several cities (10, 11). In San Francisco, tuberculous infection is associated with time spent in hotels and homeless shelters, and with total time homeless (6).

We followed the SFGH Homeless Study cohort to examine (1) the incidence of and (2) the risk factors for developing active tuberculosis. We also studied the transmission of M. tuberculosis with restriction fragment length polymorphism (RFLP) analysis (DNA fingerprinting). Clustering of RFLP patterns among tuberculosis cases is thought to represent recent transmission of infection with rapid progression to disease (8, 9, 12). A recent study in the skid row neighborhood of Los Angeles suggested a high degree of RFLP clustering among the homeless (13). We examined the incidence of and the risk factors for clustered cases of tuberculosis, as well as all cases, in the cohort.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Study Population

The SFGH Homeless Study recruited homeless persons in free food lines and shelters, using the methods of Zolopa and colleagues (6) and Koegel and Burnham (14); 89% of a validation sample interviewed on the street had used these services in the previous month: 2,774 subjects were tested for HIV and interviewed; 2,550 had TSTs placed or had a prior positive test; 2,198 returned for TST reading or had a prior positive test. HIV-seropositive persons were referred to the specialty clinics at SFGH. Those with positive TST results ( 10 mm for HIV-negatives,  5 mm for HIV positives) without providers were referred to the Department of Public Health (DPH) Tuberculosis Clinic. Unique identifiers were obtained at screening, and names were obtained for all referred persons. Between June 1992 and April 1994, referred persons were recruited to a randomized controlled trial of methods to improve adherence to INH therapy (15).

Cases of tuberculosis between 1991 and 1996 were ascertained through the DPH and interviewed about risk factors for disease. Interviews were completed with 66% of homeless cases between November 1993 and December 1996. Clinic charts were reviewed for a history of INH therapy. To ascertain incidence we compared unique identifiers obtained at baseline to the same identifiers obtained from interviewed cases between 1993 and 1996. We also compared names taken from persons referred with reported cases between 1991 and 1996. These methods undercount cases that developed between 1991 and 1993 among those HIV-seronegative and TST-negative at baseline and cases reported outside San Francisco. The study was approved by the Committee on Human Research at UCSF.

Alcoholism

Alcoholism was defined by the CAGE criterion, i.e., positive responses to two or more of the following questions: "Have you ever in your life felt you ought to cut down on your drinking?" "Have people ever in your life annoyed you or criticized you about your drinking?" "Have you ever in your life felt bad or guilty about your drinking?" "Have you ever in your life had a drink first thing in the morning to steady your nerves or get rid of a hangover?"

DNA Fingerprinting

IS6110-based RFLP analysis was performed on all available cultures using methods previously described (11, 16). Isolates that had identical IS6110 RFLP patterns ("fingerprints") and more than five bands were considered clonal; those with five or fewer IS6110 bands were analyzed with a second technique using a probe for the polymorphic guanine-cytosine-rich sequence (PGRS) (17). The DNA was digested with Sma1 and probed with a 32 bp probe based on the consensus sequence of the PGRS (ATCGGCAACGGCGGCAACGGCGGCAC- GGCGG). On the basis of visual comparison, isolates with identical numbers and relative intensity and molecular weight of bands were considered clonal.

Statistical Methods

For identifier-matched subjects, follow-up time was 3.2 yr (non-cases) or years from November 1, 1993 to diagnosis (cases). For name-matched subjects, follow-up time was years from baseline to December 31, 1996 (non-cases), or to diagnosis date (cases). Prevalent tuberculosis was defined as that diagnosed prior to screening but after August 1, 1990, or within 3 mo of screening. Hazard ratios, confidence intervals, and tests for association were done using Cox proportional hazards regression. We used a forward stepwise regression with a 0.05 significance level for multivariate analysis.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

There were 2,774 subjects who were seen at baseline. Ten had tuberculosis at baseline (0.36%); 9% of the remainder were infected with HIV at baseline and 32% had positive TST results by test or history (34% of African Americans, 23% of whites, 52% of Hispanics). The cohort was primarily "true homeless," with high rates of mental illness and substance use (Table 1).

There were 2,764 subjects without prevalent tuberculosis who were followed for a median of 3.2 yr. There were 25 incident cases (0.27% per person-year, or 270 per 100,000 per year) (Table 2). Twenty-two were bacteriologically confirmed, 18 were among AfricanAmericans or 0.35% per person-year. Five were among other nonwhites and two among whites, and 10 cases (40%) were HIV-seropositive, six in persons who were also TST positive at baseline. The incidence rate among the latter was 4.5% per person-year (Table 2).

One of 15 persons with incident tuberculosis who were HIV seronegative at baseline was HIV seropositive when tuberculosis was diagnosed. Five of six persons who were TST-negative at baseline converted and one was of unknown TST status at diagnosis. Three of five TST converters were HIV seropositive at baseline.

Baseline HIV status, TST status, age, education, and ethnicity were univariate predictors of tuberculosis (Table 3). Intravenous drug use (IDU), crack cocaine use, alcoholism by the CAGE criteria, and heavy alcohol use were not. Primary residence in the year before screening did not predict tuberculosis, though living more than 6 mo in a shelter was of marginal significance (p = 0.09). In a forward stepwise multivariate analysis of persons with known baseline TST status (excluding one case with unknown status), independent risk factors were: being HIV seropositive, being African American or (marginally) being other nonwhite, being TST positive, less education, and age greater than 50 (Table 4). Substance use and residential risk factors remained not predictive.

M. tuberculosis isolates were available for genotyping for 20 of 22 culture-positive cases. Fifteen (75%) were in 12 separate RFLP clusters (Figure 1). Cluster A in Figure 1 results from an outbreak in an HIV residential facility reported earlier (9). All five cases in G are Filipino. H is an outbreak caused by an INH-resistant organism that included nosocomial cases. K is a point-source outbreak of a streptomycin-resistant strain. No common exposures have been identified in the other clusters. However, 46% of cases in the 12 clusters were reported as homeless at diagnosis; 70% of interviewed cases in these clusters had lived in hotels at diagnosis or at some time during the previous 2 yr.

View larger version (13K): [in this window] [in a new window]   Figure 1.   RFLP fingerprint clusters in San Francisco containing incident tuberculosis cases from the SFGH Homeless Study cohort 1991-1996. Number of cases per quarter (cohort cases in black).

Three of 15 clustered cases were the apparent index cases of clusters (Figures 1f, 1j, and 1l). Assuming these are the result of reactivation of old disease, 12 of 20 cases were the result of recent transmission with rapid progression, or 60%. Isolates from 13 of 18 African American cases were fingerprinted; 12 (92%) were in clusters. Two were apparent index cases for an estimated transmission rate of 10 of 13 (77%). For other nonwhites the rate was two of five. Neither of the white cases was in a cluster. Among HIV-positive cases, seven of eight typed were clustered (including five of six HIV+PPD+ cases); none were index cases for a transmission rate of 88%. Five of 12 HIV-negative cases typed were clustered, with three index cases, for a rate of two of 12 (17%).

Risk factors for tuberculosis because of recent transmission (clustered, nonindex tuberculosis) were: being HIV-seropositive, being TST-positive, and being African American or (marginally) other nonwhite, but not age (Table 5). Marginally significant predictors for developing tuberculosis with a unique or nonindex fingerprint were TST-positivity (HR = 4.4, p = 0.08) and age (HR = 1.06 per year, p = 0.05) (data not shown).

Of six HIV-seropositive, TST-positive cases, three received no INH treatment and three had been started and completed an average five mo of the recommended 12 mo. Of 12 cases among persons PPD-positive at baseline, four had begun INH treatment and completed a mean 2 mo of INH. No case in either category had completed the appropriate course of therapy.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Between 1991 and 1996, the incidence of tuberculosis in a large cohort of homeless people in San Francisco was 270 per 100,000 per year. The rate among homeless African Americans was 350 per 100,000, versus 60 per 100,000 in whites. These rates are based on San Francisco surveillance information only and must be regarded as minimum estimates since the homeless are a mobile population and some subjects will have developed tuberculosis elsewhere. In comparison, the rate in the highest-incidence district in New York City was 221 per 100,000 in the peak year (1992) of the recent resurgence of tuberculosis (18).

Incidence of tuberculosis in San Francisco declined during the years 1993 through 1996 (as in most American cities), probably because of improved tuberculosis control after publicity given to the outbreaks of drug-resistant tuberculosis during 1992 and 1993 (19). Considerable effort was put into case-finding, case-holding via directly observed therapy (DOT), contact tracing, and treatment of latent tuberculous infection (LTBI) in HIV-seropositive persons. These interventions resulted in a 42% decline in reported tuberculosis from 1993 to 1998. TB rates in the homeless declined proportionately during this period. The rates we observed between 1992 and 1996 are an indication of what can be expected in the very high-risk urban homeless population should tuberculosis control efforts not be maintained at the present high level.

Sixty percent of incident cases in the cohort were in nine RFLP fingerprint clusters, suggesting that much of the tuberculosis in the homeless resulted from recent transmission. By comparison, the proportion clustered among 1,548 San Francisco cases analyzable by DNA fingerprinting during the years 1991 through 1997 (88% of 1,761 culture-positive cases reported) fell from 20.3% in 1991 to 12.8% in 1997 (20). The high clustering rate in the homeless suggests that many of these cases should be preventable by improved environmental control, along with effective screening treatment of LTBI. All the excess tuberculosis in African Americans as compared with that in whites was accounted for by recent transmission. So was all the excess in HIV-seropositive persons compared with HIV-seronegative persons.

The 4.6% annual tuberculosis incidence rate in those infected with both HIV and M. tuberculosis is comparable to that reported in cohorts of HIV-seropositive persons (21- 24). None of these cohorts has reported genotyping studies of the incident cases; 88% of incident HIV-seropositive cases in our study were in RFLP clusters, suggesting that the very high rates of active disease observed in HIV-seropositive persons are the result of recent infection with rapid progression to disease.

Reichman and colleagues (25) have suggested that IDU is a risk factor for tuberculosis, and IDU is an indication for treatment of LTBI in the Centers for Disease Control recommendations on high-risk populations. However, homelessness is not (26). In the current study, IDU was not a risk factor for tuberculosis among homeless people. Nor were alcohol use or crack cocaine use. Most of the high rate of tuberculosis in the San Francisco urban indigent appears to be the result of recent infection in HIV positive persons and in the African American and other nonwhite homeless population.

Homeless African Americans had a 34% prevalence of tuberculous infection at baseline versus 23% in whites (4). However, they had an adjusted hazard of 5.5 for developing tuberculosis. The high incidence rates in African Americans and other nonwhite homeless persons may be the result of behavior associated with either increased transmission or rapid progression, or may be the result of differences in genetic susceptibility. If the former, the behavior would be independent of residential exposure history and drug or alcohol use since these were not associated with tuberculosis. Stead and colleagues (27) have suggested that Americans of European descent may be more resistant to tuberculous infection than are African Americans and other nonwhite persons because the long history of tuberculosis in European cities has selected for genetic resistance. Among African Americans, Hispanics, and Native Americans, tuberculosis may be a relatively recent infection. Our data may represent greater susceptibility to infection in the nonwhite homeless, or a higher likelihood of progressing rapidly from infection to active disease. Recently, variation in the NRAMP1 gene was shown to be associated with susceptibility to tuberculosis, probably with rapid progression rather than with increased tuberculous infection. The susceptibility allele is uncommon in Europeans but was present in a quarter of the West African population studied (28).

Why was the incidence of tuberculosis so high in the homeless? First, 26% of homeless cases were classified "moved, lost or uncooperative/refused" in 1993-1994 versus only 8% of all other cases (DPH, unpublished data). Second, treatment of LTBI was not being delivered effectively among the homeless. During the study, TST converters without providers were referred to the TB clinic for treatment. Among those referred from June 1992 to April 1994, only half completed a referral without the provision of incentives or case management (15). Only 20% completed a 6-mo course of INH treatment. However, with incentives, 85% completed referral (15), and a comparable improvement can be effected for 6-mo prophylaxis (29 and J. P. Tulsky, unpublished data). Effective treatment of LTBI can reduce the rate of incident tuberculosis markedly in high-risk cohorts (30, 31). Third, although contact-tracing rates have increased in San Francisco, traditional approaches to contact identification may not be appropriate in shelters or hotels where a whole institutional population is at risk. Finally, environmental control has been addressed in public shelters, but there is no regulation of the hotels, where a much larger population is housed, and where 12% of residents are HIV-seropositive (unpublished data). Nearly three quarters of the cases in the current study had lived in hotels during some part of the 2 yr prior to diagnosis.

The proportion of cases caused by recent transmission we found among the homeless (60%) is comparable to the proportion reported in a 1992 sample of New York City cases (32), and to the proportions reported from many New York City hospitals in that era (33). As in New York, ongoing tuberculous infection among the homeless was amplified by a high rate of coexisting HIV infection and perhaps by an independently high susceptibility to infection or rapid progression among nonwhite homeless persons. Control of tuberculosis in the homeless requires, first, attention to tuberculosis control in shelters and, when possible, in hotels, perhaps by extending TST screening requirements to the latter. HIV-seropositive residents in particular should be screened. Second, it requires effective treatment of LTBI in this population, particularly for HIV-positive persons, possibly with short-course regimens. Urban homelessness should be an indi- cation for treatment of LTBI. Third, it requires continued attention to the difficulties of implementing the classic control strategies of casefinding and caseholding in the urban poor.