Empirical evidence for synchrony in the evolution of TB cases and HIV+ contacts among the San Francisco homeless

Abstract

The re-emergence of tuberculosis (TB) in the mid-1980s in many parts of the world, including the United States, is often attributed to the emergence and rapid spread of human immunodeficiency virus (HIV) and acquired immunodeficiency syndrome (AIDS). Although it is well established that TB transmission is particularly amplified in populations with high HIV prevalence, the epidemiology of interaction between TB and HIV is not well understood. This is partly due to the scarcity of HIV-related data, a consequence of the voluntary nature of HIV status reporting and testing, and partly due to current practices of screening high risk populations through separate surveillance programs for HIV and TB. The San Francisco Department of Public Health, TB Control Program, has been conducting active surveillance among the San Francisco high-risk populations since the early 1990s. We present extensive TB surveillance data on HIV and TB infection among the San Francisco homeless to investigate the association between the TB cases and their HIV+ contacts. We applied wavelet coherence and phase analyses to the TB surveillance data from January 1993 through December 2005, to establish and quantify statistical association and synchrony in the highly non-stationary and ostensibly non-periodic waves of TB cases and their HIV+ contacts in San Francisco. When stratified by homelessness, we found that the evolution of TB cases and their HIV+ contacts is highly coherent over time and locked in phase at a specific periodic scale among the San Francisco homeless, but no significant association was observed for the non-homeless. This study confirms the hypothesis that the dynamics of HIV and TB are significantly intertwined and that HIV is likely a key factor in the sustenance of TB transmission among the San Francisco homeless. The findings of this study underscore the importance of contact tracing in detection of HIV+ individuals that may otherwise remain undetected, and thus highlights the ever-increasing need for HIV-related data and an integrative approach to monitoring high-risk populations with respect to HIV and TB transmission.

Figure 1. Annual number of TB cases in San Francisco, January 1993–December 2005.

(A) Non-homeless: general population, exclusive of homeless. (B) Homeless. The rate of decline over time in the number of TB cases is much faster for the non-homeless than that of the homeless. Red line: linear fit.

Figure 2. Annual percentage of TB cases with missing HIV status, January 1993–December 2005.

(A) Homeless. (B) Non-homeless. While the reporting of HIV status has improved over time among the homeless TB cases (negative slope), it has remained more or less constant among the non-homeless TB cases (small positive slope). Red line: linear fit.

Figure 3. Annual percentage of all contacts and PPD− contacts with missing HIV status, January 1993–December 2005.

(A and B) Homeless. (C and D) Non-homeless. In A and C, contacts include both PPD+ and PPD− individuals. While the reporting of HIV status has improved over time among all contacts and PPD− contacts in the homeless population (negative slopes), it has worsened among the non-homeless (positive slopes). Red line: linear fit.

Figure 4. Wavelet time series analysis of the homeless TB cases and their HIV+/PPD− contacts, January 1993–December 2005.

Both time series were log-transformed and standardized (mean centered with unit variance) before wavelet analyses were performed. (A) Log-transformed time series of the TB cases and their HIV+/PPD−contacts. (B) Wavelet coherence between the homeless TB cases and their HIV+/PPD−contacts. One periodic component at 27–31 months is present at significant level (black lines), computed based on 400 bootstrapped series. The colors code for power values from dark blue, representing low values, to dark red, representing high values; the superimposed parabola is the cone of influence, which measures the extent of edge effects. (C) Wavelet phase evolution of homeless TB cases and their HIV+/PPD− contacts computed at the at the 27–31 months periodic band in radians. (D) Wavelet phase difference of the phases of TB cases and their HIV+/PPD− contacts computed at the 27–31 months periodic band. Waves of TB cases and their HIV+/PPD− contacts are separated, with a mean lag time of 5.4 months over the entire 13-year period, and with a mean lag time of 4.4 months over the four-year period from January 2002 to December 2005, when the HIV related data is most complete.

Figure 5. Wavelet coherence between the non-homeless TB cases and their HIV+/PPD− contacts, January 1993–December 2005.

Except for patches of high coherence at various periodic scales appearing transiently in time, there is no evidence of consistently significant coherence and synchrony.

Figure 6. Robustness of coherence and synchrony between the TB cases and their HIV+/PPD− contacts, January 1993–December 2005.

Wavelet coherence analysis was performed, where each PPD− contact with missing HIV status was randomly assigned a status (HIV+ or HIV−). The experiment was repeated 500 times and the average coherence was recorded. Each time series was log-transformed and standardized before wavelet analysis was performed. (A) Homeless: the two time series manifest high coherence at the same 27–31 months periodic band, as in Figure 4B, with the lag time of 5.2 months. (B) Non-homeless: similar to Figure 5, there is no evidence of a significant periodic component in spite of missing HIV data imputation.

Figure 7. Wavelet time series analysis of the homeless TB cases and their PPD− contacts, January 1993–December 2005.

(A) Log-transformed time series of the TB cases and their PPD− contacts. (B) Wavelet coherence between the homeless TB cases and their PPD− contacts. There is no evidence of a consistent periodic component and coherence between the two time series at any time scale (compare with Figure 4B).