Drug Toxicity, HIV Progression, or Comorbidity of Aging: Does Tipranavir Use Increase the Risk of Intracranial Hemorrhage?

The US Food and Drug Administration black box warning issued for tipranavir states that tipranavir “has been associated with reports of both fatal and non-fatal intracranial hemorrhage”; [1, p. 1]. The warning was based on reported adverse events accounting for 2.6 intracranial hemorrhage (ICH) events per 1000 person-years of surveillance. This rate was 10-fold higher than the rate observed in an older, HIV-uninfected population [2]. Although reports from the era before combination antiretroviral therapy (CART) became available suggested that persons with AIDS demonstrated rates of ICH of 2–10 cases per 1000 person-years [3, 4], the background rate of ICH (i.e., the rate of ICH expected among persons who had not been exposed to the drug) among HIV-infected individuals receiving CART was not known.

Long-term HIV infection appears to accelerate the risk of many comorbidities of aging [5, 6]. Because antiretroviral drug toxicities also imitate age-associated comorbidities [7], the only way to differentiate “background”; events from toxic events is to compare rates among patients exposed to the drug with rates among those who were not exposed but who had similar risks.

We first asked whether the rates of ICH reported among HIV-infected individuals exposed to tipranavir are higher than the rates among HIV-infected, unexposed individuals. In a related analysis, we also asked whether rates of ICH are higher among persons with HIV infection than among HIV-uninfected individuals.

Patients, materials, and methods. We used the US Veterans Health Information System–derived “virtual cohort”; (VA-VC) [8] and the California state Medicaid-derived HIV dataset (HIV Medi-Cal) [9] to provide initial indications of the likely attributable risk and number needed to cause 1 excess ICH-related event among HIV-infected individuals who had been exposed to tipranavir (i.e., the number needed to harm).

In the VA-VC, patients are identified as HIV infected by virtue of having at least 2 outpatient codes or 1 inpatient code for HIV infection; this algorithm has previously been validated [8]. HIV-infected veterans are matched to HIV-uninfected veterans on the basis of year of care, race/ethnicity, and age [8]. HIV Medi-Cal enrollees were identified by the Medical Care Statistical Section of HIV Medi-Cal using case identification by diagnosis codes from HIV and/or drug claims for antiretroviral medications. These data were linked to the California state inpatient discharge database to identify all acute hospitalizations occurring during or outside of HIV Medi-Cal enrollment.

We conduct the analyses in both datasets using the same calendar interval, International Classification of Diseases, Ninth Revision (ICD-9), diagnostic groupings, and incidence definitions. Our sample consisted of patients entering care during the period 1 October 1997 through 31 December 2003. ICH is identified using the following established ICD-9 codes [2, 3]: 430.xx, 431.xx, and 432.xx (except 432.0). We excluded subjects who had codes of 850.xx–854.xx (i.e., trauma codes) at the same encounter. ICH had to meet criteria of having 1 inpatient code or ⩾2 outpatient codes during the observation interval. To avoid the inclusion of prevalent cases of ICH, we excluded patients who had ICH codes during the 12-month interval before the baseline date.

Bivariate associations were examined and statistical significance determined using the χ² test. We used Poisson models to estimate adjusted incidence rate ratios (IRRs) and incidences. Analyses are performed using SAS software, version 9.1.3 (SAS Institute).

Results. From 1 October 1997 through 31 December 2003, a total of 16,541 veterans presented for HIV care and were matched to 34,305 demographically similar, HIV-uninfected veterans receiving care; 28,023 HIV Medi-Cal recipients presented for HIV care (table 1). During the observation period, 33 HIV-uninfected veterans, 33 HIV-infected veterans, and 373 HIV Medi-Cal recipients received incident care for ICH (table 2). The crude event rate for the HIV Medi-Cal cohort was 4 cases per 1000 person-years (95% CI, 3.6–4.5 cases per 1000 person-years). The crude event rate for the HIV-infected VA-VC group was 0.4 cases per 1000 person-years (95% CI, 0.3–0.6 cases per 1000 person-years). The crude event rate for HIV-uninfected VA-VC group was 0.1 cases per 1000 person-years (95% CI, 0.1–0.2 cases per 1000 person-years).

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Table 1

Demographic and clinical characteristics of patients, by database source.

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Table 2

Demographic and clinical characteristics, by incident intracranial hemorrhage (ICH) events.

In bivariate analyses stratified by HIV status and database, age, race/ethnicity, atrial fibrillation, congestive heart failure, cocaine abuse or dependence, liver disease, drug abuse or dependence, alcohol abuse or dependence, hypertension, diabetes, coronary artery disease, and peripheral vascular disease were found to be associated with incident ICH (table 2). Among HIV Medi-Cal recipients, AIDS, Cryptococcus infection, exposure to and duration of CART, and exposure to protease inhibitors were also associated with incident ICH.

We evaluated the independent association of the factors listed above with ICH and tested the following interaction terms: age, >50 years, and HIV infection; age, >50 years, and HIV Medi-Cal use; and black or Hispanic race and HIV Medi-Cal use. There were no significant interactions among model variables, ICH events, and either cohort or HIV status; thus, these interaction terms were omitted from the final model.

In the final models (table 3), overall HIV status was associated with an IRR of 2.48 (95% CI, 1.53–4.02). Other independent associates of incident ICH are age of ⩾50 years (IRR, 1.46; 95% CI, 1.16–1.85), minority status (IRR, 1.38; 95% CI, 1.14–1.67), vascular disease (IRR, 1.92; 95% CI, 1.57–2.35), alcohol abuse or dependence (IRR, 1.53; 95% CI, 1.18–1.97), and liver disease (IRR, 2.30; 95% CI, 1.59–3.33). Of note, subjects from the HIV Medi-Cal database were more likely to have incident ICH than were VA-VC patients (IRR, 6.45; 95% CI, 4.49–9.27).

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Table 3

Multivariable adjusted association of risk factors for intracranial hemorrhage, overall and stratified by database and HIV status.

In the VA-VC database, the unadjusted attributable risk of ICH associated with HIV infection was 0.3 events per 1000 person-years (0.4 events per 1000 person-years among HIV-infected persons and 0.1 events per 1000 person-years among HIV-uninfected persons). When the multivariate models were restricted to persons with HIV infection and were adjusted for HIV factors, CD4 cell count, HIV load, toxoplasmosis, Cryptococcus infection, and CART were not significant and were excluded. Also, in the model that included only HIV-infected patients, AIDS was independently associated with ICH (IRR, 2.13; 95% CI, 1.75–2.60). HIV Medi-Cal patients remained at higher risk of ICH than did veterans in the HIV-restricted model (IRR, 5.30; 95% CI, 3.68–7.64).

To generate estimated rates of ICH that could be compared with the observed rate among persons who had been exposed to tipranavir, we fit the aforementioned variables to a Poisson model restricted to persons with HIV infection. The attributable risk of ICH with use of the VA-VC estimate of background rate is 2.6-0.4 (i.e., 2.2 ICH events per 1000 person-years). The attributable risk using HIV Medi-Cal data is 2.6-2.2 (i.e., 0.4 ICH events per 1000 person-years). With use of a number needed to harm, the VA-VC results suggest that we would need to treat 455 HIV-infected patients with tipranavir for 1 year to see 1 excess ICH event. The HIV Medi-Cal data suggest that we would need to treat 5000 HIV-infected patients for 1 year before seeing a single excess event.

Discussion. Our data, together with the rate of events observed among persons exposed to tipranavir, suggest that we would need to treat 455–5000 HIV-infected patients with tipranavir for 1 year to see a single excess ICH event. Although a major strength of the study was the use of 2 large patient databases, a number of limitations should be mentioned. The VA-VC database has very few women in its sample. The HIV Medi-Cal database did not include appropriate HIV-uninfected control subjects. The VA-VC database may have incomplete “capture”; of acute life-threatening events, such as ICH. Prior studies of health care use among dually enrolled Veterans Affairs and Medicare beneficiaries have demonstrated that veterans who routinely receive care at Veterans Affairs facilities are often initially hospitalized for acute conditions at non–Veterans Affairs institutions [10, 11]. We also assume that tipranavir exposure is unrelated to the likelihood of the VA-VC and HIV Medi-Cal data systems capturing an ICH event. If it were, our estimates may not be accurate. However, because Medicaid covers encounters within its statewide jurisdiction, we think it likely that HIV Medi-Cal achieves a more complete capture of ICH-related hospitalization events occurring among HIV Medi-Cal beneficiaries.

The background rate estimates in both the VC-VA and HIV Medi-Cal databases may be conservative, because they rely on ICD-9 diagnostic codes. The use of ICD-9 codes to identify ICH events may have underdetected events in both HIV Medi-Cal and VA-VC samples, because codes are generally specific but insensitive [6]. However, ICD-9 codes have been shown to be accurate for stroke (90% agreement; k=0.86) [12]. Both databases may underreport rates of ICH, compared with the Boehringer Ingelheim figures, which do not include data for persons who died of stroke without being hospitalized.

Finally, it is worth noting that persons with HIV infection and those with advanced HIV infection (i.e., AIDS) experience a stepwise increased risk of ICH. The reasons for this are likely multiple. First, current cocaine use–a known risk factor for ICH that is poorly captured in medical databases–is likely more common among persons with HIV infection. Furthermore, some of the increased risk of ICH seen in the HIV Medi-Cal sample, compared with the risk among HIV-infected veterans, may be explained by higher unmeasured rates of ongoing cocaine use. In addition, chronic HIV infection may directly increase the risk of ICH through effects on cerebral atrophy, clotting mechanisms, and vascular competence.

Because these data likely underestimate the background rate of ICH, they may overestimate the risk attributable to tipranavir (as expressed in the number needed to harm). Ongoing safety monitoring of tipranavir for ICH events should include gathering standardized and detailed information regarding stage of HIV infection and other comorbid conditions. Once a greater number of patients have been exposed to tipranavir and have had their outcomes measured, methods such as propensity scores [13] or marginal structural modeling [14, 15] may be used to determine whether there is evidence of an increased risk of ICH after direct adjustment for these important factors.

• Received April 15, 2008.
• Accepted July 7, 2008.

• © 2008 by the Infectious Diseases Society of America