The differential impact of HIV and antiretroviral therapy on gender-specific tuberculosis rates

Abstract

OBJECTIVE

To assess the impact of the HIV epidemic and the rollout of antiretroviral therapy (ART) from 2004 on the gender-specific TB burden in Cape Town, we investigated temporal changes in TB notification rates, the HIV-associated relative risk of TB and the population attributable risk fraction (PAF) of HIV by gender.

METHODS

Annual TB notifications, mid-year population and HIV prevalence estimates were used to calculate rates per 100 000 population stratified by gender and HIV. Annual rate ratios (RR) of TB associated with HIV and PAF were calculated by gender.

RESULTS

Pre-HIV TB notification rates were lower among women than men (146/100 000 vs. 247/ 100 000). With the onset of the HIV, epidemic rates increased 5.3-fold in women (to 778/100 000) and 3.7-fold in men (to 917/100 000) to a peak in 2008, after which they declined by 25% in women (to 634/100 000) and 18% in men (to 755/100 000) by 2014. The HIV-associated RR of TB was 25% higher in women than in men in 2006 (25 vs. 20), but decreased to the same level in 2014. HIV PAF declined between 2008 and 2014 from 56% to 50% and from 40% to 38% in women and men, respectively.

CONCLUSIONS

The HIV epidemic led to greater relative increases in TB rates among women than men. The increased HIV-associated TB risk in women could be compatible with removal of the biological protection of female gender by HIV infection. The decline in RR and PAF in HIV-positive women could be explained by increasing ART usage reversing female gender-related susceptibility.

Introduction

Tuberculosis (TB) rates are consistently higher among men than women worldwide, although male to female ratios vary by region [1–3]. This male bias is hypothesised to be due to behavioural and biological differences leading to differential risks of both infection and progression to disease [2, 4, 5]. Despite evidence to support greater barriers to TB diagnosis and treatment in women than men [6], a recent meta-analysis of TB prevalence surveys in lower-and middle-income settings found a consistent male excess in both prevalence and prevalence-to-notification ratios, suggesting under-detection in men rather than in women [3, 7].

Globally, TB notification rates are higher for men than women at all ages after adolescence [1]. The HIV epidemic has led to increased TB notification rates in both genders,¹ with age and gender patterns reflecting age-and gender-specific HIV prevalence [4, 8]. However, it is unknown whether increased access to antiretroviral therapy (ART) has changed these patterns of disease.

Cape Town has an extremely high prevalence of TB and HIV, both of which have increased rapidly over the last two decades. TB notification rates increased from 390 per 100 000 population in 1993 to 713 per 100 000 in 2013 [9, 10]. During the same period, antenatal HIV seroprevalence increased from 1.2% to 19.7% [11, 12]. Over the last decade ART coverage has increased dramatically from 0% in 2003 to 64% of the estimated HIV-positive population in 2013 [10], with disproportionately more women initiating ART than men, as elsewhere in sub-Saharan Africa [13–15].

We investigated the changes in gender-and age-specific TB notification rates over time since the onset of the HIV epidemic and the rollout of ART. We also determined the change in the relative risk of TB notification associated with HIV infection and the population attributable risk fraction (PAF) of HIV, both stratified by gender.

Methods

Study setting and population

Cape Town is the capital of the Western Cape province of South Africa, with an estimated population size of 3.7 million in 2011 [16]. TB treatment is provided at primary care level and treated patients are recorded in TB registers, which became electronic in 2003 [17]. We included all newly registered drug-sensitive TB patients in 1993 and from 2003 to 2014. Drug-resistant TB cases have a separate register and were therefore not included in this analysis. To avoid duplication we excluded patients who were transferred between clinics.

Definitions

We followed the case definitions of TB as per the Cape Town TB programme; patients were classified by type of disease (pulmonary smear-positive, smear-negative or extrapulmonary TB) and by microbiological confirmation (yes/no). Diagnostics available included sputum smear microscopy, chest X-ray, abdominal ultrasound and lymph node aspirate microscopy. Xpert MTB/RIF only became available in 2011 and was fully rolled out by 2013 [18,19]; Xpert results were not included in the classification of type of disease (above), but were included in the classification into microbiologically confirmed TB.

We defined the start of the generalised HIV epidemic as 1994, when antenatal HIV seroprevalence reached 1% [11, 20]. The national ART rollout programme began in 2004. We assigned the following four calendar years to reflect four periods in the HIV epidemic: 1993 (minimal HIV), 2003 (early HIV with no ART), 2008 (established HIV with early ART rollout, ART coverage estimated at 29%) and 2013 (established HIV with late ART rollout, ART coverage 63%) [10].

Data sources

We abstracted age-and gender-stratified TB notification data from the Cape Town Medical Officer of Health report for 1993 [21]. These data were not available for the years 1994–2002. Individual-level TB notification data were obtained from the Cape Town metropolitan electronic TB register (ETR) for the years 2003–2014, allowing for a comparison across two decades. For denominators we used mid-year Cape Town population estimates retrieved from the Cape Town Medical Officer of Health report (1993) and from Statistics South Africa (2003 onwards) [21, 22]. Annual HIV prevalence estimates were sourced from the Actuarial Society of South Africa Western Cape AIDS model 2008 [23].

Statistical methods

We report baseline characteristics of TB cases stratified by year and gender from 2003 onwards, when individual-level data became available. TB notification rates were calculated per 100 000 population (with 95% confidence intervals [CIs]) and stratified by gender. We calculated rates stratified by HIV status from 2009 onwards when recording of HIV status among TB notifications reached more than 90% [10]. We estimated HIV-stratified rates for 2006–2008 by extrapolating the HIV prevalence among those tested to those who were not, justified by the stable HIV positivity among those tested from 2003 to 2013 [10]. We did not estimate HIV-stratified rates for 2003–2005 as the number of known HIV-positive cases was too small for a reliable rate calculation, nor for 1993 as it was before the advent of the generalised HIV epidemic. We further calculated TB notification rates stratified by 10-year age group. Changes over time were investigated by comparing age-specific TB notification rates in the four calendar years reflecting the periods in the HIV epidemic (1993, 2003, 2008 and 2013). We decided against performing trend analyses due to the large number of individuals included in the analysis. We also performed a sensitivity analysis restricted to only microbiologically confirmed patients.

We calculated the annual male to female TB ratio by dividing the male TB rate by the female rate, overall and stratified by HIV status. We also calculated these ratios by 10-year age group, overall and by HIV status, using the number of TB cases instead of rates due to the lack of reliable estimates of the age-stratified HIV-positive population. Age-stratified TB notifications by HIV status were only available from 2008 onwards.

Annual gender-specific rate ratios (RR) of TB associated with HIV infection and their 95% CIs from 2006 onwards were calculated (HIV-positive rate/HIV-negative rate). The gender-specific annual PAF of HIV for those years was calculated as follows: ((overall rate – HIV negative rate)/overall rate). The calculations and analyses were performed using STATA 13.0 IC (College Station, TX, USA) and Microsoft Excel 2013 (Redmond, VA, USA).

Ethics

This study was approved by the Human Research Ethics Committee of the University of Cape Town.

Results

We included a total of 323 881 TB patients in our analysis. Their baseline characteristics stratified by gender are summarised in Table 1 (for the four calendar years reflecting the periods of the HIV epidemic) and in the appendix (for all calendar years). The proportion of male TB notifications decreased from 61% in 1993 to 55% in 2013. Throughout the study period, the proportions of patients that had been treated previously and of patients with positive microbiological confirmation were higher among men than women. Known HIV status among TB patients increased in both genders from <4% in 2004 to 99% in 2013, and HIV prevalence among female TB patients was consistently higher than among male cases. The proportion of patients with a reported CD4 count at TB diagnosis increased from <5% prior to 2008 to >90% from 2009 onwards. Baseline median CD4 counts were higher among women than men in all calendar years, with slight increases in both genders over time.

Table 1.

Baseline characteristics of TB patients by gender in the four calendar years reflecting the HIV periods^†

		1993‡: minimal HIV		2003:early HIV, no ART		2008: established HIV, early ART		2013: established HIV, late ART
		Female	Male	Female	Male	Female	Male	Female	Male
Total (n [row %])		1768 (39)	2820 (61)	9585 (42)	13 014 (58)	13 199 (46)	15 424 (54)	11 750 (45)	14 416 (55)
Age (years,mean [SD])		27 (18)	33 (18)	28 (15)	33 (16)	29 (15)	34 (16)	30 (16)	34 (16)
Retreatment (n [%])				2108 (22)	3731 (29)	3029 (23)	4549 (30)	2520 (21)	3723 (26)
Type of disease	Pulmonary			4581 (48)	7618 (59)	5276 (40)	7511 (49)	3704 (32)	5626 (39)
(n [%])	smear-positive
	Pulmonary smear-negative			3092 (32)	3590 (28)	5920 (45)	6132 (40)	6317 (54)	7248 (50)
	Extrapulmonary			1912 (20)	1806 (14)	2003 (15)	1781 (12)	1729 (15)	1542 (11)
Microbiological  confirmation  (n [%])	Test(s) positive			5739 (60)	9011 (69)	7023 (53)	9328 (61)	6992 (60)	9505 (66)
	No or negative test(s)			3846 (40)	4004 (31)	6176 (47)	6096 (40)	4758 (41)	4911 (34)
HIV status known  (n [%])				2 (0)	1 (0)	10 488 (80)	12 029 (78)	11 597 (99)	14 166 (98)
HIV prevalence  among those  tested, %				50	0	58	43	51	38
CD4 count (/mm3)  at TB diagnosis	Unknown			0 (0)	0(0)	1946 (32)	1810 (35)	183 (3)	148 (3)
categories (n  [%])§
	<100			0 (0)	0(0)	1359 (22)	1355 (26)	1679 (28)	1886 (35)
	100–199			1 (100)	0(0)	1196 (20)	906 (18)	1174 (20)	1210 (22)
	200–349			0 (0)	0(0)	865 (14)	611(12)	1315 (22)	1191 (22)
	≥350			0 (0)	0(0)	717 (12)	490 (10)	1563 (26)	948 (18)
CD4 count (/mm3)				n/a	n/a	157 (75, 278)	132 (58, 248)	200 (85, 366)	155 (65, 293)
at TB diagnosis
(median [IQR])§

IQR, interquartile range; n, total; n/a, not available; SD, standard deviation; TB, tuberculosis,

^†Baseline characteristics for all calendar years (2003–2013) are shown in Table SI.

^‡For 1993, no data on characteristics other than gender and age were available.

^§CD4 counts were only summarised from 2008 onwards.

TB notification rates increased in women and men from 1993 to 2008–2010 (from 146 to 786 per 100 000 population in women and from 247 to 922 per 100 000 population in men) (Table 2). This constituted a 5.2-fold increase among women and a 3.7-fold increase among men. No differential increases were seen among the HIV-negative population (2.4 and 2.3 fold, respectively). From 2010 onwards, the rates decreased in both genders: by 26% in women and 16% in men. These declines were greater among HIV-positive (32% and 17%) than HIV-negative (23% and 15%, respectively) TB patients. The male to female TB ratio was higher in HIV-negative than in HIV-positive TB patients in 2003; the HIV-negative ratio declined and the HIV-positive ratio increased over time resulting in no difference by HIV status by 2012.

Table 2.

HIV-stratified TB notification rates per 100 000 population by gender, male to female ratios of TB rates over time

		1993	2003	2004	2005	2006	2007	2008	2009	2010	2011	2012	2013	2014
Female	Total	146	625	659	702	721	717	778	758	786	727	676	634	583
	HIV-negative	146	N/A	N/A	N/A	321	319	344	351	382	351	338	330	296
	HIV-positive	N/A	N/A	N/A	N/A	7886	7483	7880	7197	7016	6394	5717	5135	4804
Male	Total	247	856	867	935	894	878	917	922	898	867	835	786	755
	HIV-negative	247	N/A	N/A	N/A	537	522	545	559	556	539	523	504	472
	HIV-positive	N/A	N/A	N/A	N/A	10444	10220	10497	10219	9628	9243	8806	8008	8028
Ratio†	Total	1.7	1.4	1.3	1.3	1.2	1.2	1.2	1.2	1.1	1.2	1.2	1.2	1.3
	HIV-negative	1.7	N/A	N/A	N/A	1.7	1.6	1.6	1.6	1.5	1.5	1.5	1.5	1.6
	HIV-positive	N/A	N/A	N/A	N/A	1.3	1.4	1.3	1.4	1.4	1.4	1.5	1.6	1.7

HIV–, HIV negative; HIV+, HIV positive; n/a, not applicable (1993) or not available (2003–2005).

^†Male to female ratio of TB rates.

Figure 1a-d shows the changes in age distribution of TB notification rates stratified by gender in the different HIV periods (rates and 95% CIs of these periods are reported in Table S2). In 1993, there was a high rate of disease among infants, decreasing sharply in childhood (Figure 1a). In adolescence, rates started diverging between men and women with male rates surpassing those among infants and reaching a peak between 55 and 64 years (at 391 per 100 000). Rates among women also increased but remained lower than among infants and peaked earlier, between 25 and 34 years (191 per 100 000). After the onset of the HIV epidemic, TB rates increased substantially in both men and women (Figure 1b). Since 1993, the age distribution of TB changed among men with rates peaking at a younger age (35–44 years, increasing fourfold to 1631 per 100 000). Rates among women still peaked at the same age (25–34 years), increasing sixfold to 995 per 100 000, but were much higher than among infants. From 2003 to 2008, rates increased consistently, particularly among women (Figure 1c). By 2013, rates started decreasing, more among women (a decline of 22% from the peak) than among men (a decline of 18% from the peak) (Figure 1d). The pattern of increasing and decreasing rates was also seen in paediatric TB rates, but with no gender difference. The sensitivity analysis restricting the above to only microbiologically confirmed TB patients showed lower rates, in particular among children, but overall the same trend in rate changes (Figure S1a-d).

Figure 1.

(a-d) Changes in the age distribution of TB notification rates per 100 000 population (with 95% confidence intervals) in men and women in the four calendar years reflecting the HIV periods. Note. The estimated ART coverage (of total HIV-positive population) was 29% in 2008 and 63% in 2013, and negligible before that.

The age-stratified male to female ratios of TB disease showed that before the advent of HIV, girls and boys were equally affected by TB until adolescence, after which men had a higher burden of disease (Figure 2a). In the HIV era, women from 15 to 34 years of age had a higher burden of disease, after which the male burden of disease took over. These patterns were more pronounced when stratified by HIV: HIV-positive TB cases were more frequent among men than women at all ages except in 15-to 34-year olds (Figure 2b).

Figure 2.

(a, b) Age-stratified male to female ratios of TB notifications in the four calendar years reflecting the HIV periods: overall (a) and stratified by HIV (b, only 2008 and 2013). Note. TB notifications by HIV status were only available from 2008 onwards. HIV–, HIV negative; HIV+, HIV positive

The RR of TB associated with HIV infection was 26% higher in women than in men in 2006: 24.6 (95% CI 23.7–25.5) vs. 19.5 (95% CI 18.8–20.1) (Figure 3). The declining RR among women led to an equalisation in RRs between the genders in 2013: the female RR of TB associated with HIV decreased to 16.2 (95% CI 15.6–16.9) whereas the male RR fluctuated around 17.

Figure 3.

Male to female TB rate ratios (with 95% confidence intervals) associated with HIV infection over time.

In 2006, 56% of female TB disease was attributable to HIV compared to approximately 40% of male TB disease (Figure 4). Despite a slight decline in the female PAF and increase in the male PAF, the female PAF remained consistently higher over time.

Figure 4.

Population attributable fraction of TB disease attributable to HIV infection by gender over time.

Discussion

The HIV epidemic in Cape Town led to a greater relative increase in TB notification rates among women than men, particularly among young women aged 15–34 years. Men also developed TB disease at a younger age than before. However, male rates still remained higher than female rates throughout the study period. Over time, the TB RR associated with HIV equalized between men and women. Nevertheless, the proportion of disease attributable to HIV remained persistently higher among women.

Our findings show that the HIV epidemic impacted the gender distribution of TB disease in Cape Town. It was associated with very high increases in TB notification rates in both men and women. Although the relative impact of HIV on female TB was greater than on male TB, male rates remained higher throughout the study period. This is in contrast with HIV prevalence, which is known to be higher in women than in men (except in childhood and old age) [23]. In addition to changing the gender distribution of TB disease, the HIV epidemic also changed its age distribution. TB rates in men peaked earlier than in pre-HIV years, and rates in women peaked 10 years earlier than in men (25–34 vs. 35–44 years). These findings were robust to restriction to only microbiologically confirmed patients, with the exception of paediatric TB which largely consists of empirical diagnoses [24]. The pattern we found follows the change in HIV prevalence in the general population with steep increases 5 years earlier in women than in men (at 20–24 compared to 25–29 years of age) [23]. Similar age-and gender-specific distributions of TB disease in the HIV era have been reported from other HIV high prevalence settings in sub-Saharan Africa [4, 8, 25, 26], confirming the strong association between HIV and TB epidemics.

In contrast, rates in children also increased but with no gender differential. To our knowledge, there have not been any reports on the change in childhood TB rates stratified by gender since the advent of the HIV epidemic. The absence of male bias in paediatric TB disease has been known since the last century [4], leading to the hypotheses that it is caused by the biological and behavioural differences in puberty and throughout adolescence [2, 4].

The age distribution of TB disease differed by HIV status. Although more men than women had TB at all ages, among HIV-positive patients women had more TB than men in the 15-to 34-year age group. This pattern is reminiscent of the age distribution of disease in the pre-chemotherapy era, when the only group of women with a higher TB notification rate than men were those aged 15–34 years [4]. The mechanism underlying this phenomenon has never been elucidated; evidence suggests it is due to biological differences in susceptibility to infection or progression to disease rather than to behavioural differences [2, 5, 27]. Whether the higher TB rates in this age group in the HIV era simply represent the increased risk due to higher HIV prevalence or point to an additional underlying immunological mechanism cannot be determined from our findings, nor whether these two mechanisms could combine multiplicatively. In any case, the HIV epidemic provides a ‘perfect storm’ for young women with a high risk of TB at the time when they have the highest risk of acquiring HIV, leading to a very high burden of both diseases [4].

Our results are compatible with a temporal association between increasing ART coverage and decreasing relative risk of TB in HIV-positive women in recent years. It could be hypothesised that the protective effect of female gender was removed in the early years of the HIV epidemic, and that widespread use of ART restored this. Considering that people on ART remain at a higher risk of TB than the general population and male coverage lags behind that of women, it will be interesting to see how these trends continue to develop when ART coverage increases even further [13, 14, 28–30].

We found an almost 40% higher PAF of HIV among women than men. Our paper is one of the few to calculate the gender-stratified PAF of HIV in southern Africa. In Malawi, which had an earlier generalised HIV epidemic than Cape Town, the PAF was higher among women than men in 2001 (64% vs. 48%) [31]. These proportions were slightly higher than the corresponding PAFs in Cape Town in 2003 (53% vs. 35%), which might be explained by a higher adult HIV prevalence in Malawi than Cape Town during those years (13% compared to 6%) [23]. Despite the reducing risk of TB associated with HIV infection during rollout of ART, the gender-specific PAF changed only slightly and remained much higher among women. This may be explained by the increase in HIV prevalence in the population during the period of study; had the HIV prevalence stayed the same as in 2003, the PAF in 2014 would have reduced to 39% in women and 32% in men (data not shown). This highlights the need for continued efforts to prevent HIV infection.

The strengths of our analysis include the longitudinal nature of our data and the length of follow-up, spanning the evolution of the generalised HIV epidemic in our setting. To our knowledge, our paper is the first to document the transition of age-and gender-specific distributions over time from prior to the start of the HIV epidemic through to the ART era. In comparison with other TB and HIV high prevalence settings, this TB programme had high levels of HIV testing among TB patients (from 2008 onwards) and reported large numbers allowing for multiple stratification. Nevertheless, the use of routine data collected over a long period of time has a number of limitations, including the risk of misclassification and changes in definitions, diagnostics and data collection over time. An important limitation was the lack of data availability from 1994 to 2002, the early years of the HIV epidemic, due to major political changes and associated changes in administrative procedures in South Africa at the time. Also, high rates of HIV testing were only reached from 2008 onwards, limiting our ability to stratify by HIV status before then. As we used routine notification data to estimate population rates, our estimates are vulnerable to case detection rates in our population. We have no reason to believe that case detection practices in this population changed during the study period, particularly in regard to any gender differences. This is supported by a recent meta-analysis of prevalence surveys in lower-and middle-income countries which found no change in male to female prevalence-to-notification ratios over time [3]. We estimated the HIV prevalence among TB cases in the years before HIV testing became universal, but did not stratify by age group due to the uncertainty surrounding such estimates. We were therefore unable to calculate RRs and PAFs stratified or adjusted by age. We were also unable to test for an association between ART coverage and the declining RR of TB associated with HIV because of the lack of data points.

Conclusions

The HIV epidemic led to a greater relative increase in TB rates among women than men, in keeping with the higher HIV prevalence among women. Nevertheless, TB rates among men remained higher overall. In the early years of the epidemic, HIV infection was associated with a sub-stantially higher relative increase in TB rates in women than in men, which could be compatible with the hypothesis that HIV infection removes the biological protective effect of female gender. The declining gender risk difference in recent years could be due to ART restoring this protection. Despite this encouraging trend, TB control programmes need to focus their attention on maximising TB prevention and treatment, in particular among HIV-positive women of reproductive age.