INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The introduction of highly active antiretroviral therapy (HAART) has dramatically changed the clinical prognosis for human immunodeficiency virus (HIV)-infected patients in terms of decreased mortality, morbidity, and need for hospitalization for these patients (1).

In addition, the CD4 cell count at diagnosis of acquired immunodeficiency syndrome (AIDS) has increased as HAART has become more widely available (5), and preliminary data suggest that the incidence of some diseases (i.e., non-Hodgkin lymphoma and wasting syndrome) has decreased less than other AIDS-defining diseases (6).

Mycobacterium tuberculosis (TB) and Mycobacterium avium complex (MAC) are both common opportunistic infections. In the pre-HAART era, the incidence of MAC and TB was approximately three and two cases/100 person years of follow-up, respectively, and both have increased throughout the 1980s and early 1990s (9). The two mycobacterial diseases are generally considered to represent completely different parts of the spectrum of opportunistic infections. In the pre-HAART era, MAC was closely linked to the duration and severity of immunosuppression, and nearly exclusively diagnosed at very low CD4 cell counts. The median CD4 cell count at TB was considerably higher, and TB was diagnosed at a much broader spectrum of immunodeficiency (9).

Both diseases have been reported within the first months after initiating HAART, as part of an immune reactivation syndrome (13, 14). In contrast, the impact of HAART on the incidence of these two mycobacterial diseases over a longer term as well as on their clinical presentation has not been studied in detail. We therefore tried to describe the characteristics of patients with TB and MAC, to assess temporal changes in incidences of the diseases, and to analyze the temporal changes, including an evaluation of the impact of the antiretroviral treatment (ART).

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

EuroSIDA is a prospective, European observational cohort study consisting of HIV-infected patients from 51 clinical centers in 17 European countries (appendix).

A prefixed number of consecutive HIV-infected patients older than 16 years seen in outpatient clinics was enrolled from May 1994 (EuroSIDA I cohort, 3,120 patients), from December 1995 (EuroSIDA II cohort, 1,367 patients), and from February 1997 (EuroSIDA III cohort, 2,844 patients). A CD4 count < 500 cells/mm³ within 4 mo prior to enrollment was mandatory. Information was collected from patients' charts and by patient interviews onto a standardized data collection form at enrollment and at semiannual follow-up visits. Details of data collection in the study has previously been published (15). Time variables were collected as month and year. All forms were checked by the scientific staff at the coordinating office for logical errors, and the centers were monitored to ensure uniform and correct patient enrollment and data transmission from patient charts to the data collection form.

In early 1999, information from nine follow-up visits for cohort I, six for cohort II, and three for cohort III was available. The present study includes follow-up until December 1998 to February 1999.

Analysis

All patients without any history of TB and MAC at the time of enrollment to EuroSIDA were included in the respective analyses. That is, patients with TB prior to recruitment to EuroSIDA were excluded from the analysis of TB, and the same for MAC.

For patients who were diagnosed with TB or MAC under prospective follow-up, characteristics were compared between the two groups using chi-squared test and the Kruskall-Wallis test. For logistical reasons, patients experiencing both diseases were in this comparison classified in the group of the first disease. Data included sex, age at time of diagnosis, latest CD4 cell count before or at time of diagnosis, risk group, ethnicity, AIDS at time of diagnosis, as well as region of EuroSIDA, and use of ART. HAART was broadly defined as treatment with three or more drugs, regardless of type and class of drug.

The regional classification was based on an arbitrary division used in earlier EuroSIDA analyses (South: Greece, Israel, Italy, Portugal, and Spain; Central: Belgium, France, South Germany, Luxembourg, and Switzerland; North: Denmark, Ireland, North Germany, The Netherlands, Norway, Sweden, and United Kingdom) (16). The southern region was further divided into two subgroups (Southwest: Spain and Portugal; Southeast: Greece, Israel, and Italy), as preliminary analyses had documented pronounced variations in the incidences of TB in this region.

Incidences of TB and MAC were assessed overall and in various subgroups, using time from enrollment in EuroSIDA to date of a possible event, last follow-up visit, or death.

In particular, incidences were calculated in three time intervals: before September 1995 (before 9/95), September 1995-March 1997 (9/95-3/97), and after March 1997 (after 3/97). This time division was based on the time of uptake of dual (9/95) and triple therapy (3/97) in Europe (17).

For treatment, the follow-up time without ART (any ART, or HAART) was calculated as the time from recruitment in EuroSIDA until the time of initiation of ART, death, last follow-up, or development of an event, whereas the follow-up time on ART was assessed as time from initiation of ART (or recruitment in EuroSIDA, if ART was initiated prior to recruitment) to development of an event, death, or last follow-up.

The CD4 cell count was divided into four groups: < 50, 50-100, 100-200, and > 200 cells/mm³, and incidences were calculated according to the latest CD4 cell count. This method takes into account the changes in CD4 cell count caused by changing treatment regimens. Patient follow-up in different strata was calculated as patient CD4 cell count changed between the respective strata. Confidence intervals (95% CI) were calculated using a normal approximation, or when appropriate (< 20 cases) a Poisson distribution.

Cox proportional hazards models were established to analyze the temporal changes in an adjusted setting. The calendar time in this part of the study was modeled as a continuous variable, as this provided the best fit of the model. Other factors in the models were demographic factors such as region of enrollment, age, sex, ethnicity, route of HIV infection, hemoglobin and weight, markers for HIV disease progression (CD4 cell count, AIDS-defining event, and prior MAC [TB model]/TB [MAC model]), as well as use of ART. The following parameters were modeled as time-dependent covariates: CD4 cell count, diagnosis of AIDS, diagnosis of TB/MAC, hemoglobin, weight, and initiation of ART.

Models that included various treatment variables were considered, and those providing the best fit were chosen for the final model. The variables were (1) each antiretroviral drug separately, (2) treatment with one, two, and three or more drugs, (3) regimens including a protease inhibitor or a nonnucleoside analogue, and (4) the number of new drugs used at initiation of HAART. The chosen variables of ART were included in the final model, but the estimates of relative hazards for these variables were not shown as they are very difficult to interpret in models that also include markers for disease progression.

All analyses were run in accordance with the principle of intent to treat, so no adjustment for stopping treatment regimen was done. For example, the binary variable for treatment with indinavir took the value 0 until the patient started treatment with indinavir, and switched to 1 thereafter, regardless of whether the patient at a later time point stopped indinavir again.

Inclusion of the demographic factors, such as age, sex, exposure group, ethnicity, hemoglobin, and weight, did not appreciably alter the estimates of the relative hazards of MAC and TB associated with calendar time, region of Europe, CD4 cell count, or diagnosis of AIDS, and these factors were therefore excluded from the final model in order to maintain the focus on the temporal changes in MAC or TB.

Various transformations of the continuous covariates in the model were also considered (logarithm and square root transformation), and the transformation that resulted in the best fit was chosen. In addition, the assumption of proportional hazards in the Cox model was formally tested, and revealed no evidence of nonproportionality.

All analyses were performed using SAS version 6.12 (SAS Institute, Cary, NC). p Values of less than 0.05 were considered significant, and all tests of significance were two sided.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Study Population

Of the 7331 patients enrolled in the EuroSIDA study, 167 patients had experienced MAC (2.3%) and 359 patients TB (4.9%) prior to enrollment, leaving 7,164 and 6,972 patients, respectively, for analyzing the development of MAC and TB during prospective follow-up. Table 1 presents characteristics for patients under active follow-up within the three time intervals. A dramatic increase in use of antiretroviral therapy occurred from the period before September 1995 to the period after March 1997, and the median CD4 cell count also increased substantially from the periods before March 1997 to after this time (Table 1). Other patient characteristics remained in general unchanged throughout the three time intervals (Table 1).

Characteristics of Patients with TB or MAC

Among the patients under prospective follow-up, 121 patients developed TB, and 214 patients were diagnosed with MAC (Table 2). Of these, 12 patients were diagnosed with both diseases. The number of pulmonary and extrapulmonary TB was similar, and as explorative differentiation between the two subgroups did not prove major differences, patients with both types of TB are therefore presented together.

Patients diagnosed with TB and MAC differed substantially from each other. The majority of patients with MAC were diagnosed in North Europe and infected by homosexual contact, whereas the majority of patients with TB were observed in Southwest Europe and the major HIV transmission category was intravenous drug use. At the time of diagnosis, there was no significant difference in use of ART between patients with TB and MAC. In both groups a large proportion of the patients were ART experienced but did not receive any treatment at the time of diagnosis (Table 2).

Patients with MAC had a significantly lower CD4 cell count, and significantly more patients with MAC had a prior diagnosis of AIDS compared with those with TB (Table 2). In addition, the CD4 count at diagnosis of MAC remained extremely low throughout the entire study period (before 9/95: 11 [interquartile range, IQR: 5-30], 9/95-3/97: 16 [6-48], and after 3/97: 5 [2-22] 10⁶ cells/L, p = 0.18), whereas the CD4 cell count at TB increased substantially during the study period (49 [10-120], 66 [21-220], and 180 [71-416] 10⁶ cells/L, p < 0.01).

Incidences

Overall incidences of TB and MAC during prospective follow-up were 0.8 (95% confidence interval: 0.7-0.9) and 1.4 (1.2- 1.6) cases/100 person-years of follow-up (PYF). Over time, the incidence of both diseases dropped markedly, and the steepest slope was observed for MAC (a 20-fold reduction, versus a 5-fold reduction for TB). Therefore, after March 1997, TB has become the most common mycobacterial disease (Figure 1A).

View larger version (10K): [in this window] [in a new window]   View larger version (13K): [in this window] [in a new window]   View larger version (14K): [in this window] [in a new window]   Figure 1.   Changes in incidence per 100 person years of follow-up (100 PYF) of Mycobacterium tuberculosis (TB) and Mycobacterium avium complex (MAC) among patients in the EuroSIDA study: (A) Changes over calendar time, p < 0.001 in a test for trend for both diseases. (B) Changes in the incidence of TB within CD4 strata over calendar time. Within each of the periods before 9/95 and 9/95-3/97: p < 0.0001 in a test for trend, and for the period after 3/97: p = 0.002. (C) Changes in the incidence of MAC within CD4 strata over calendar time, p < 0.001 for each of the three time intervals.

For both diseases, the incidence increased with decreasing CD4 cell count in an overall analysis as well as within each of the three time periods (Figures 1B and 1C). This analysis took into account changes in CD4 cell count during follow-up, and thus accounted for treatment-induced changes in CD4 cell count. However, the difference between the CD4 cell count intervals was far more pronounced for MAC than for TB. That is, the incidence of MAC among patients with CD4 cell count above 100 cells/mm³ was more than 50 times less than among patients with a CD4 cell count < 50 cells/mm³ (0.2 [0.1-0.2] versus 11.6 [9.9-13.3] cases/100 PYF). In contrast, the incidence of TB above 100 CD4 cells/mm³ was only six times lower compared with that below 50 CD4 cells/mm³ (0.5 [0.4- 0.7] versus 3.2 [2.3-4.1]).

Analyzing temporal changes within each CD4 cell count interval, the incidence of both diseases decreased among patients with a low CD4 cell count (for MAC: < 100 cells/mm³, and for TB: < 200 cells/mm³) from the early period to the latest period. Among patients with higher CD4 cell counts, the incidences were already very low in the first time period (Figures 1B and 1C).

Marked regional differences were observed for both diseases. The incidence of TB was four to seven times higher in Southwest Europe relative to the other regions, whereas the incidence of MAC was two to four times higher in North Europe (Table 3).

For both diseases, patients with prior diagnosis of AIDS had a higher incidence than patients without prior diagnosis of AIDS; the difference was most pronounced in the MAC analysis. Further, higher incidences were documented among patients who had started HAART relative to patients who had not, probably reflecting the fact that those who were treated were likely to be in the most advanced stages of disease (Table 3).

Antimycobactrial Chemoprophylaxis

Use of antimycobacterial drugs as primary chemoprophylaxis remained overall nearly constant throughout the study period. However when focusing on the subgroup of patients at highest risk (CD4 count < 50 cells/mm³), the relative use of MAC chemoprophylaxis increased from 8% in 1994 to 14% in 1998, whereas use of TB chemoprophylaxis decreased from 7% to 3% (Figure 2). Further, use of chemoprophylaxis against MAC was more common in the southern regions in 1994, and the increase in use of this type of chemoprophylaxis occurred primarily in North and Central Europe. Consequently, use of this chemoprophylaxis in 1998 varied only slightly among regions of Europe.

View larger version (17K): [in this window] [in a new window]   Figure 2.   Changes in use of primary chemoprophylaxis against Mycobacterium tuberculosis (TB) and Mycobacterium avium complex (MAC) among patients (overall and for patients with a CD4 cell count < 50 cells/mm3) in the EuroSIDA study.

Cox Models of the Temporal Changes in TB and MAC

To further explore the marked decreases in the incidences of TB and MAC from 1994 to 1999, multivariate Cox models were established. We included calendar time as a continuous variable as well as demographic factors, and markers of HIV disease progression and adjusted in addition for changes in use of antiretroviral therapy (Table 4). In these adjusted models, the hazard ratio of development of TB was 0.95, indicating that the risk of developing TB was reduced 5% for each year of follow-up, though it was not significantly different from 0%. In contrast, the corresponding yearly reduction was 42% for MAC.

The predictive value of the time-dependent CD4 cell count should be interpreted as an increase in risk of TB and MAC of 32% and 54%, respectively, per 50% lower CD4 cell count. Further, previous diagnosis of AIDS was associated with a considerably higher risk of developing MAC, but not TB, and in addition, prior mycobacterial infection was also associated with a substantially reduced risk of being diagnosed with TB/ MAC, although this was significant only in the model for MAC (i.e., prior TB was associated with a 58% reduction in risk of MAC).

Marked regional differences remained present when adjusting for potential differences among regions; the risk of TB was approximately 10 times higher in Southwest Europe compared with North Europe, whereas the risk of MAC was two to three times lower in South Europe compared with North Europe (Table 4).

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

In the present study we documented marked decreases in the incidence of TB and to an even larger extent MAC from 1994 to 1999. A large part of these decreases in the incidences could be related to raising CD4 cell count and increasing use of antiretroviral therapy within the same period. However, marked differences between the two mycobacterioses were observed.

For TB we were able to explain most of the temporal changes, as adjustment for ART and CD4 cell count reduced the hazard ratio for the calendar time to an insignificant level.

In contrast, the risk of being diagnosed with MAC decreased 42% per year of follow-up after adjustment for all available parameters. If it is a true decrease, how could this have occurred? Increasing use of chemoprophylaxis among patients with severe immunodeficiency may play a role in the decreasing incidence, although the increase was small and is unlikely to explain the dramatic drop in MAC, which started before the use of chemoprophylaxis increased. Identification of risk factors for MAC, such as exposure to water from hot water recirculation systems, might have allowed some precautions, though clinically it is not likely to have caused the marked decrease in incidence of MAC (18, 19). Increasing use of isolation regimens and case finding procedures might have influenced the incidence of TB, but does not seem likely to explain the reduction in risk of MAC, for which interpersonal contact is not thought to be a major transmission category. Further, the clinical picture of MAC might have changed, making the diagnosis more difficult to make.

The apparently reduced risk could also be due to underadjustment, or confounding, that we could not identify. Other studies have not been able to document a causal relation between the decrease in morbidity and the temporal changes in use of antiretroviral therapy (2, 20). This was apparently due to confounding between calendar time and use of antiretroviral therapy, whereas the decrease in mortality has been convincingly related to the intensification of ART (1, 2). In the present study, adjustment for antiretroviral therapy in addition to the CD4 cell count was able to explain more of the temporal change in the incidence than the CD4 cell count alone (data not shown). Therefore, ART may have an effect over and above the effect mediated by changes in CD4 cell count. It may also be due to ART changing some other value that we are not measuring or for which we are not able to adjust. In particular, changes in viral load might explain the residual effect of MAC and calendar time. We are measuring viral load, but virological measurement has not been in clinical use long enough to allow testing for that in an unbiased way. The viral load is available only in a highly selected group of patients; i.e., at the time of diagnosis of TB or MAC in less than 10% of the patients, and mainly for patients being followed in recent years. However, inclusion of plasma viral load in the multivariate model did not change the results significantly (data not shown).

The modeling of antiretroviral treatment might also be subject to underadjustment. The relative use of the different antiretroviral drugs in Europe has changed and there is probably improvement in use of any given regimen over time, as the relative potency of the separate drugs has been evaluated, and drug combinations optimized (17, 21, 22). Importantly, modeling the antiretroviral drugs separately and including the number of new drugs used at initiation of HAART did not change the overall results (data not shown).

Compared with previous studies, the incidence of TB reported in this study was lower than the incidence of three cases/100 years of follow-up among European AIDS patients in the 1980s (10), but at the same level as in a U.S. study from the period 1988-1994 (23). In contrast, the incidence of MAC among patients followed in 1994-1995 in our study was considerably higher than among AIDS patients in Europe in the 1980s (< 50 cells/mm³: 8.2 cases/100 person-year of follow-up, AIDS in Europe, personal communication), which is in accordance with other reports of an increasing incidence of MAC from the 1980s to the early 1990s in Europe and in the United States (9, 12, 24).

Whereas both mycobacterioses have become very rare compared with the level just a few years ago, TB has now become the most common mycobacteriosis in 1997-1999, in contrast to the pre-HAART era, in which MAC was more than twice as common as TB. This is interesting, as only preliminary data are available on the influence of HAART on the pattern of AIDS diseases and combination therapy prior to the introduction of protease inhibitors did not seem to affect the pattern of AIDS events substantially (6, 25). Of note, decreasing incidences of most AIDS-defining diseases, except MAC and TB, were reported in a retrospective study from 1992 to 1996 (26). The present study indicates that in the future TB will be the most common mycobacterial disease among HIV-infected patients. However, for continuous monitoring of the incidences of AIDS-defining events in the post-HAART era, large size studies such as the EuroSIDA study are needed, given the large reduction in incidence overall.

We also found that the CD4 count at diagnosis of TB increased with calendar time, whereas it remained extremely low for MAC. The latter was in contrast to another study, which found increasing CD4 count at AIDS including the diagnosis of MAC (5). These changes were most likely due to the generally increasing CD4 cell count in the EuroSIDA cohort, resulting in a considerably longer total follow-up time in the higher CD4 cell intervals in the later time periods compared with the period before September 1995. As the incidence of MAC among patients with higher CD4 cell counts was low compared with that of TB, this could explain the different temporal pattern of the CD4 cell count for the two diseases. In line with this, the incidence of MAC and TB among patients with a latest CD4 count less than 100 cells/mm³ decreased from 1994 to 1999, whereas the incidences in the interval above 100 (MAC) or 200 cells/mm³ (TB) were already very low in 1994-1995 and did not change significantly.

Though the Cox models indicate only associations and not causal pathways, the results underscore that causal pathways between HIV infection and development of each of the two mycobacterioses remain extremely different over calendar time. That is, severe immunodeficiency (AIDS diagnosis and low CD4 cell count) remains mandatory for the development of MAC, whereas this is not the case for TB.

Marked regional differences in incidences of TB and MAC in the era of HAART were also documented, MAC being most common in North Europe and TB most common in Southwest Europe. In the pre-HAART era HIV-related TB was also more common in Southern Europe, and the regional differences documented in the present study were resistant to adjustment for all available parameters, in particular differences in use of antiretroviral treatment, which varied considerably across Europe (10, 17). Therefore, these regional differences were likely to reflect either differences in prevalence of the pathogens, in pathogenicity of the prevailing subtypes of the mycobacteria, or in behaviors leading to exposure to the specific microorganism (19, 27). Interestingly, TB is much more common in the non-HIV-infected population on the Iberian peninsula compared with other parts of Western Europe (28), whereas only limited information on the regional distribution of MAC within Europe is available. A high incidence of MAC among HIV-infected patients in Finland has been reported, and geographic differences in the prevalence of MAC infection among HIV-infected patients have been related to various water exposures (19, 27).

Prior TB was associated with a relative protection against MAC (and for prior MAC, the relative hazard for development of TB was of equal size, though in this case not significantly different from 1). This has previously been found in some studies (AIDS in Europe, personal communication) (29), but not all (30). A relative protection of prior TB against development of MAC seems plausible as agents against TB also have an effect on MAC.

In conclusion, we documented a marked decrease in the incidences of TB and MAC among HIV-infected patients from 1994 to 1999. The decrease in TB was associated with the introduction of HAART and changes in CD4 cell count. Although the introduction of HAART also seemed to explain some of the decrease in MAC over time, there remained a significantly lower risk of MAC than expected. This might in part be explained by the increasing use of chemoprophylaxis against MAC among patients with severe immunodeficiency, though it is unlikely to explain the dramatic drop in MAC observed in the present study.