Patients and Methods

The Center for Special Studies of the New York Presbyterian Hospital–Weill Cornell Center comprises 2 clinics in Manhattan that serve a diverse population of ∼1600 HIV-infected patients. Osteonecrosis was identified on the basis of radiologic evidence of osteonecrosis at any stage and a compatible clinical history. Patients who met these criteria (case patients) were identified by searching the clinics’ electronic medical record database for the pertinent diagnostic code and by querying clinic providers

Two matched sets of control patients were selected randomly from the Center’s database. In the first set, 2 control patients per case patient were matched on (1) date of the initial clinic visit ±3 months, with follow-up for the control patient continuing at least until the time of diagnosis of osteonecrosis in the corresponding case patient, and (2) initial CD4 cell count by stratum (<0.050, 0.051–0.100, 0.101–0.200, 0.201–0.350, 0.351–0.500, and >0.500 ×10⁹ cells/L). In the second set, 2 control patients per case patient were matched only on date of the initial clinic visit ±3 months. The second set of control patients was used only for comparison of baseline CD4 cell counts

Medical records of case and control patients were reviewed systematically to ascertain demographic and clinical characteristics. Characteristics obtained at the time of presentation with osteonecrosis in case patients were obtained at similar calendar time points in control patients

Conditional logistic regression calculated matched univariate odds ratios (ORs) with 95% confidence intervals (CIs). HIV RNA levels below the level of detection (<400 copies/mL) were assigned values of 400 copies/mL before log transformation. In general, continuous variables were found to have skewness close to normality (between −2 and 2) and were analyzed by Student’s t test. Wilcoxon&amp;rank sum tests, which are valid even if variables have extreme distributions, were used to confirm all univariate findings. Multivariate analysis was done using conditional logistic regression with forward, stepwise selection. As recommended by some authors, the criterion for entry into the model was P⩽.2 [10]. In an additional model, history of protease inhibitor use—a covariate of particular interest—was forced into the model chosen by forward selection. For the multivariate analyses, missing HIV RNA values were imputed by using the overall mean values of the case patients plus control patients. All data were analyzed using Stata software (version 5.0; Stata)

Results

We identified 17 cases of osteonecrosis diagnosed between 1992 and 2000, 15 of which were identified after 1996. All case patients had osteonecrosis of the femoral head, and 10 (59%) had bilateral involvement. One patient had osteonecrosis of the humeral head, in addition to both femoral heads. The median CD4 cell count of case patients at the initial clinic visit was 0.090×10⁹ cells/L (interquartile range [IQR], 0.029–0.352×10⁹ cells/L), compared with 0.291×10⁹ cells/L (IQR, 0.109–0.451×10⁹ cells/L) in control patients matched on date of initial clinic visit only (P=.10; Wilcoxon&amp;rank sum test)

The characteristics of case patients and CD4 cell count–matched control patients are listed in table 1. In univariate analyses, case patients were significantly more likely than were control patients to have received corticosteroids (47.1% vs. 8.8%; matched OR, 13.1; 95% CI, 1.6–106; P=.016) and to have had an increase in CD4 cell count from nadir >0.050×10⁹ cells/L (64.7% vs. 35.3%; matched OR, 4.9; 95% CI, 1.0–24; P=.049). Case patients also were more likely than control patients to have had Pneumocystis carinii pneumonia (PCP; 52.9% vs. 11.8%; matched OR, 7.6; 95% CI, 1.6–36; P=.010) but not other opportunistic infections. Though not statistically significant, case patients tended to have a longer duration of antiretroviral therapy (median duration, 34 months; IQR, 21–66 months, vs. median duration, 18 months; IQR, 4–41 months; P=.08). There was also a trend for greater use of protease inhibitors among case patients than among control patients, both at the time of diagnosis of osteonecrosis (58.8% vs. 38.2%; matched OR, 4.0; 95% CI, 0.77–20.8; P=.10) and at any time before diagnosis (76.5% vs. 58.8%; matched OR, 3.7; 95% CI, 0.68–20; P=.13)

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Figure 1

Multivariate analysis of factors associated with osteonecrosis in patients infected with human immunodeficiency virus. Multivariate analysis was done using conditional logistic regression. ♦, Odds ratio (OR) of the covariates, adjusted for the other covariates depicted in the figure; vertical bars 95% confidence intervals of the ORs. ART, antiretroviral therapy; m, month

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Table 1

Characteristics of human immunodeficiency virus (HIV)–infected case patients with osteonecrosis (ON) and HIV-infected, CD4 cell count–matched control patients

For each case patient and control patient, we counted the number of risk factors traditionally associated with osteonecrosis in the non–HIV-infected population, specifically, corticosteroid use, radiation therapy, alcoholism, pancreatitis, hypertriglyceridemia (random triglycerides >4.52 mmol/L [400 mg/dL]), and hypercholesterolemia (random cholesterol >6.21 mmol/L [240 mg/dL]). Case patients had more traditional risk factors than did control patients (median, 2; IQR, 1–2, vs. median, 1; IQR, 0–1; P<.001). To determine whether the difference in number of risk factors between case and control patients was due only to differences in corticosteroid use, we counted risk factors other than corticosteroid use. The number of traditional factors other than corticosteroid use remained associated with osteonecrosis (OR, 2.2 per risk factor; 95% CI, 1.0–4.8; P=.05). This association strengthened after adjusting for corticosteroid use by conditional logistic regression (OR, 3.9 per risk factor; 95% CI, 1.1–13.6; P=.034)

Figure 1 illustrates the results of multivariate analyses. Of the variables in the final model selected by stepwise regression, only prior corticosteroid use had a strong and statistically significant association with osteonecrosis. The addition to this model of a covariate coding for prior use of protease inhibitors had no impact on the other covariates in the model, and the covariate was not statistically significant itself (figure 1)

Both case and control patients tended to use corticosteroids for brief courses (<4 weeks), primarily as adjunctive therapy for PCP. Among the 8 case patients with known corticosteroid use, 3 were being treated for PCP, 3 for esophageal ulcers, 1 for thrombocytopenia, and 1 for asthma. Among the 3 control patients with known corticosteroid use, 2 were being treated for PCP and 1 for a presumed drug rash. In addition, 3 case patients and 2 control patients had conditions that were previously diagnosed as PCP, but records were insufficient to determine if corticosteroids had been used. In sensitivity analyses, classifying the 5 patients with indeterminate use of corticosteroids as corticosteroid users had no significant impact on the results of univariate (OR for corticosteroid use, 8.9) or multivariate (adjusted OR for corticosteroid use, 24) analyses. The exact dates of corticosteroid use or PCP diagnosis were not known for all patients, although they were known to precede the diagnosis of osteonecrosis; however, for the 9 case patients with known dates of corticosteroid use or PCP, the median time from this date to presentation with osteonecrosis was 35 months (IQR, 29–39 months)

Discussion

In this case-control study, we found a strong association between prior systemic corticosteroid use and osteonecrosis of the femoral head in HIV-infected patients. There were also trends for associations between osteonecrosis and duration of antiretroviral therapy and CD4 cell response to antiretroviral therapy. However, use of protease inhibitors as a class was not associated independently with osteonecrosis. The lower initial CD4 cell count in case patients than in control patients not matched on CD4 cell count and the trend toward a more robust increase in CD4 cell count from nadir in case patients than in CD4 cell count–matched control patients suggest that osteonecrosis occurred primarily in patients with severe immunosuppression who responded successfully to antiretroviral therapy, most of whom had other established risk factors for the disorder

Previous reports have postulated a number of specific factors in HIV-infected patients that may lead to osteonecrosis. Some authors have speculated that a hypercoagulable state due to anticardiolipin antibodies [2] or protein S deficiency [9] is responsible for cases of HIV-associated osteonecrosis. Although biologically plausible, these states are quite prevalent in HIV-infected patients and have not been demonstrated to be more common among those with osteonecrosis. Other investigators have postulated that protease inhibitors may lead to osteonecrosis, perhaps indirectly by causing hyperlipidemia [4]. Osteonecrosis also has been linked to antiretroviral therapy–induced viral suppression and increases in CD4 cell count [9] and to use of megestrol acetate [11]. Although these reports are intriguing, they lack controlled data to support the proposed associations and must be interpreted with caution

Our results indicate that many cases of osteonecrosis in HIV-infected patients occur in association with risk factors previously identified for non–HIV-infected patients. Furthermore, our data suggest that part of the increased occurrence of osteonecrosis since 1996 could be due to improved survival of patients with advanced immunosuppression who received corticosteroids for treatment of opportunistic conditions. There may be a latency period after short-term exposure to corticosteroids until osteonecrosis manifests clinically, as has been described in non–HIV-infected patients [12]. For example, in our study, the median time from PCP or corticosteroid use to onset of symptoms consistent with osteonecrosis was ∼3 years. Before the advent of potent antiretroviral therapy, few HIV-infected patients who were treated with corticosteroids for PCP would have survived this long (i.e., 3 years), since the median survival after an initial AIDS-defining case of PCP was only 1.5 years [13]

Compared with control patients, case patients had a more robust increase in CD4 cell count from nadir. Although not statistically significant, case patients also had lower median nadir CD4 cell counts than did control patients, a higher prevalence of prior opportunistic infections other than PCP, and a trend for longer duration of antiretroviral therapy. We cannot rule out the possibility that corticosteroid use may have been a marker for advanced HIV disease and that recovery from advanced HIV disease or advanced disease itself is a major predictor of osteonecrosis. However, on the basis of self-reported dates of HIV diagnosis and the matching on baseline CD4 cell count and date of the initial clinic visit, case and control patients likely had similar durations of HIV infection

Our findings are consistent with those of 2 recent reports. In a case-control study, Scribner and colleagues found that 22 of 25 HIV-infected patients with osteonecrosis had ⩾1 established risk factor, most commonly hyperlipidemia, alcoholism, pancreatitis, corticosteroid use, and hypercoaguability [14]. In contrast to our results, however, these investigators found no association between CD4 cell counts and the development of osteonecrosis. In a cross-sectional study, Masur and colleagues found osteonecrosis in 4.4% of 339 asymptomatic HIV-infected patients who underwent magnetic resonance imaging of the hips [15]. Osteonecrosis was associated with use of corticosteroids but not protease inhibitors in this study

Our study has several noteworthy limitations. We could evaluate only information captured in the medical records as part of routine care and thus could not examine potential factors such as the presence of anticardiolipin antibodies. Since some patients with PCP had been hospitalized at other institutions, it was not always possible to determine if corticosteroids were used by these patients. However, in a sensitivity analysis that considered all such patients to have used corticosteroids, our findings did not change. Since osteonecrosis is uncommon, our sample size was small. The wide ranges of the 95% CIs of the ORs reflect limited power to rule out associations with many of the covariates under study. We aimed to increase the power of the study by performing matched analyses, and to minimize confounding by stage of disease by matching on CD4 cell count stratum and adjusting for a history of opportunistic infections other than PCP

In summary, we found an association between corticosteroid use and osteonecrosis that was independent of HIV disease stage and protease inhibitor therapy. Because not all case patients in this study and in other series had reported exposure to corticosteroids, the pathogenesis of osteonecrosis in HIV-infected patients may be multifactorial. Since adjunctive corticosteroids are indicated for moderate-to-severe PCP, HIV-infected patients who have received them for this or other indications may be at particularly high risk for developing osteonecrosis. Awareness of this association should lead to earlier recognition of the signs and symptoms of osteonecrosis in such patients and to improved clinical outcomes