From 1997 through 2002, we performed a repeated cross-sectional survey of HIV prevalence among TB patients in the 12 most urbanized districts (districts 1, 3, 4, 5, 6, 8, 10, 11, Phu nhuan, Tan binh, Nha be, and Binh thanh) of Ho Chi Minh City. Until 1998, districts included all patients >15 years of age who had confirmed TB according to WHO criteria and who consented to HIV testing during the same quarter each year. Since 1999, enrollment was restricted to the last quarter of the year for all districts. Enrollment stopped after a quota was reached; the quota was proportional to the annual number of patients treated in the district, set to obtain a total sample size of ≈800 patients each year. Ethical approval was obtained from the Ho Chi Minh City Council Research Board.

We determined HIV status by ELISA (Genolavia Mixt, Sanofi, Paris, France, until 1999; and Genscreen, Sanofi, Paris, France, from 1999 onward) and an independent confirmatory test (Serodia; Fujirebio, Tokyo, Japan, or Vironostika, Organon, Boxtel, the Netherlands) if the first test result was positive. District TB staff collected data for each patient on TB disease (diagnostic category, treatment history), age, sex, marital and employment status, education level, and risk factors for HIV infection. Patients who owned small businesses and seasonal workers were coded as “self-employed”; civil servants and patients under contract (e.g., drivers), as “employed.” All data were entered twice in EpiInfo version 6 (Centers for Disease Control and Prevention, Atlanta, GA, USA), and discrepancies were checked against the raw data. TB reporting data were obtained from NTP quarterly district reports. Sex, age, and distributions of urban and rural population size were interpolated from the results of the 1994, 1999, and 2004 census; standard exponential population growth was assumed.

For HIV trend analyses, we used 2-year blocks, increasing group size and power of the analyses. The Cuzick test for trend was used to identify monovariate time trends in HIV prevalence (13). To compare proportions, we used the χ² test.

Multivariable analysis was performed by logistic regression. After transformation (squaring), time of inclusion could be entered as a continuous variable (χ² for departure linear trend = 0.68 [df = 1], p = 0.41). Variables were included when the likelihood ratio χ² test was significant at the 0.1 level.

We identified multivariable time trends by entering time interaction variables (time of inclusion × variable x) in our logistic model (14,15). Compared with the baseline category, an odds ratio (OR) >1 indicates a faster rise in HIV prevalence in that category, thereby identifying high-risk groups. For the sake of interpretability, we present the multivariable model without interaction terms or time trends.

To describe the combined epidemic outside the known high-risk groups, we ran the final multivariable model after excluding all IDUs. The model goodness-of-fit was assessed by the Hosmer-Lemeshow test of goodness-of-fit and by visual inspection of the distribution of the model residuals (16,17).

We used the formula p × (1 – 1/RR) to estimate the fraction of TB in patients that was attributable to HIV (the population-attributable fraction, PAF) (18). In this formula, p represents the prevalence of HIV in TB patients, and RR represents the relative risk of active TB developing in patients with HIV versus in patients without HIV. The PAF thereby corrects for the fact that some of the TB cases among HIV-infected patients would also have occurred were the patients not HIV infected (18). Because this information cannot be known for individual patients, RR represents the average relative risk for HIV-infected patients. Lacking situational data, we conservatively assumed a constant RR of 5 from 1997 through 2002. We based this estimate on the literature and took into account the early stage and specific characteristics (focused in high-risk groups that overlap with TB risk factors such as injection drug use) of the combined epidemic in Southeast Asia (3). To test our assumption, the RR was also varied between 2 and 10 or gradually increased (from 2 to 10) over the study period to simulate a progressive fraction of HIV-infected patients in which active TB develops as a result of increased immune suppression.

The rate of TB observed without HIV was calculated as [(1 – PAF) × current TB rate]. We restricted this part of the analysis to new smear-positive TB patients from urban districts. The diagnosis of these patients’ condition is highly standardized and the HIV/TB data came from urbanized districts, which ensured that combining the 2 datasets was as valid as possible. Exponential growth rates were estimated by using the least squares method.

In 2002, the Ho Chi Minh City Council started a mandatory rehabilitation program for IDUs. Because these rehabilitation centers were not included in our surveillance, we quantified potential resulting bias by estimating how including 50% more IDUs in 2002 would affect our results.

Analyses were performed by using Stata version 8 (Stata Corp., College Station, TX, USA). Excel 2003 (Microsoft Corp., Redmond, WA, USA) was used to asses the effect of HIV on reporting rates.

A total of 5,701 patients consented to HIV testing (92% of those eligible) and were included in the study. Because of clerical error, 504 patients entered the study from July through December 1996. These were added to the subset analyzed for 1997. Apart from IDUs, patient numbers in the individual risk categories were too low to be analyzed separately and were therefore added to the category of “other.”

HIV prevalence rose exponentially from 1997 through 2002 (Table 1). The mean age of HIV-infected patients decreased from 38 (standard deviation [SD] = 10) to 27 (SD = 8) years; the mean age of the HIV-negative group remained stable at ≈39 (SD = 14). The male:female ratio was higher for HIV-infected (9:1) than for other TB (7:1) patients (χ² = 53.6 [df = 3], p<0.001). HIV prevalence in young (<35 years) men rose to 22.3% (108/484) during 2001–2002.

The 12 districts did not differ significantly in HIV prevalence during the study period (χ² = 38.1 [df = 33], p = 0.25) (data not shown). HIV prevalence in reported IDUs rose to 95% in 2001–2002, which accounted for 28% of all HIV-infected patients.

In the multivariable analysis, when time trends and other interactions were disregarded, HIV infection among TB patients was associated with age <45 years, male sex, not being married or employed, and being an IDU (Table 2). As the time-trend analyses show (Table 3), HIV prevalence increased faster in young (<35 years) patients, most prominently in the youngest age group (15–24 years), and in IDUs. Additional interaction (p value for excluding interaction from model = 0.002) between marital status and sex indicated that the high HIV prevalence in single TB patients was mainly attributable to HIV infection among male patients.

When IDUs were excluded, the multivariable model predicted the data less well (–2 log likelihood with IDUs = –686.7, without = –660.2), but this exclusion affected neither the direction of the ORs nor their size in a relevant way (data not shown). Also, the Hosmer-Lemeshow test of goodness-of-fit remained nonsignificant in both models (p = 0.72 with IDUs, p = 0.69 without). Although IDUs and non-IDU HIV-infected patients did not differ relevantly in age, sex, or marital status (χ² test, p = 0.93, 0.53, and 0.78 respectively), they did differ in their employment status and level of education (χ² test, p = 0.05 and <0.001, respectively).

PAF calculations show that 0.7%, 1.5%, 1.3%, 6.7%, 9.5%, and 9.7% of reported new smear-positive TB cases were attributable to HIV in 1997, 1998, 1999, 2000, 2001, and 2002, respectively. After these cases were excluded from analysis, the rising trend in TB reporting rates reversed to a mild decline (Figure, panel A). When stratified for sex (Figure, panel B) and age (Figure, panel C), this effect seemed limited to men and was most prominent in the younger age groups. Increasing the RR over time did not affect the results (data not shown). Adding 50% more IDU patients to the 2002 population enhanced the effect of HIV, which reduced the corrected annual growth in TB reporting from –0.2% to –1.2% in men and from –1.8% to –2.9% in young (<35 years) persons.

Apart from underreporting of risk factors, other limitations may have affected our results. For patients in the districts included in this surveillance, TB may have been diagnosed outside the surveillance project, e.g., in the city’s TB referral hospital or the private sector, which predominantly diagnose smear-negative and extrapulmonary TB. These diagnoses were reported for 29% of the patients in our study compared with 35% of all patients reported in Ho Chi Minh City over the study period. Our data may therefore underrepresent patients with smear-negative and extrapulmonary TB and may have underestimated or overestimated the HIV infection prevalence among them. Our estimates of the impact of HIV infection on TB reporting rates, however, will not be subject to such bias because these were based on new smear-positive patients only.

We have no data on levels of CD4+ lymphocytes and could not stage immune depletion in HIV-infected patients. Whether a case of TB in an HIV-infected patient was due to advanced immune depletion or would have occurred regardless of HIV infection is thus unknown. We have dealt with this possible bias by applying the PAF, which measures excess cases only (27). The PAF depends on the RR of TB for HIV-infected persons compared with non–HIV-infected persons and thereby on the level of immune depletion. Because no estimates of this RR are known for the Vietnamese setting, we assumed a value of 5, which is in accordance with RRs found in several studies conducted elsewhere (3,28,29). We also applied values of 2 and 10 and increased the RR over the study period, simulating an increasingly vulnerable population. Neither affected our results in any relevant way.

The withdrawal of IDUs from regular surveillance in 2002 may also have caused bias. But as our simulations showed, the absence of 50% of IDUs reduced the size of the effect but not its direction.

We recommend that in Ho Chi Minh City all TB patients be tested for HIV because detection of HIV infections can help prevent some of the excess deaths in this population (cotrimoxazole preventive treatment and antiretroviral therapy) (30,31). To prevent active TB, prophylactic isoniazid treatment for HIV-infected patients could be considered (31,32). However, a recent study from Ho Chi Minh City showed that isoniazid resistance levels might be too high (>25% in new TB patients) for successful implementation (33). Highly active antiretroviral therapy is being introduced and is expected to reduce the risk for TB disease in HIV-infected patients (34–36). Injection drug use clearly remains a potent source of health problems; efforts to reach out to the vulnerable population of IDUs should be sustained and increased. In addition to the interventions mentioned, TB screening for HIV-infected IDUs and TB treatment for those found to have TB disease should be considered.