Comparison of 2 Studies of Treatment of Invasive Aspergillosis

To the Editor—Congratulations to Cornely et al. [1] for completing another important randomized study of invasive aspergillosis (IA). This is the second largest completed such study, and enrollment occurred remarkably fast, being completed in 18 months. At face value, it would be easy to make 2 conclusions from this study with respect to the treatment of IA: (1) the efficacy of 3 mg/kg of liposomal amphotericin B is essentially equivalent to that of 10 mg/kg liposomal amphotericin B (although the latter is slightly more toxic), and (2) liposomal amphotericin B has the same efficacy as voriconazole. Although the first conclusion is probably valid, the second needs further analysis and should be questioned. I, as well as others, have demonstrated very different response rates among different patient groups with IA [2, 3], and these response rates depended on how early the diagnosis was achieved [4, 5]. Other factors that are important for a successful outcome include the dose of corticosteroids once the diagnosis has been made [6]. For example, it has been known for many years that patients with IA after allogeneic hematopoietic stem cell transplantation (HSCT) generally experience a poor outcome, with an associated mortality rate that is typically 175%, whereas patients with acute leukemia whose neutropenia resolves have a mortality rate that is generally !50% and, in some series, !30%. These differing outcomes are critically important in interpreting therapeutic trials of IA, because different trials enroll different proportions of at-risk patients. This is particularly germane to a comparison of the study by Cornely et al. [1] with the randomized study by Herbrecht et al. [7], in which voriconazole is compared with standard amphotericin B. Many clinicians would argue that patients with acute leukemia are “at high risk” and, therefore, should be considered to be in the same risk group as those who have received an allogeneic HSCT, but a critical distinction needs to be made between the high risk of acquisition of IA and a high risk of dying of IA. These risks differ markedly between patients with acute leukemia (who have a moderate to low risk of death if treated) and patients with allogeneic HSCT (who, even if treated, have high risk of death), as they do between patients with HIV infection and patients with AIDS, in whom the risk of acquisition of IA is low (2%–4%), but the risk of dying of IA exceeds 80% [8, 9]. A similar contrast applies between risks of IA acquisition and IA-related death in liver transplant recipients [9]. In table 1, the different underlying conditions, enrollment characteristics, and outcomes from the 2 trials [1, 7] are summarized. For both trials, modified intentto-treat populations are shown. This table shows considerable differences between the studies. For example, in the study by Cornely et al. [1], 190% of the patients had hematological malignancy (including autologous HSCT), whereas !60% of the patients in the study by Herbrecht et al. [7] had hematological malignancy, with a higher proportion of allogeneic HSCT recipients and patients with AIDS, (P ! ). However, the biggest contrast be.001 tween the studies was with respect to the confirmation of disease. Two-thirds of the patients in the study by Herbrecht et al. [7] had microbiologically confirmed disease, whereas !40% of the patients in the study by Cornely et al. [1] had microbiologically confirmed disease ( ). P ! .001 Antigen testing was not performed as a direct part of the study by Herbrecht et al. [7]; therefore, all microbiological confirmation related to histological examination, culture, and microscopic examination findings. Furthermore, 120% of the cases in the study by Cornely et al. [1] were confirmed by antigen testing only (usually an early diagnostic feature), and ∼15% of cases were otherwise microbiologically confirmed by other methods. This major difference is accounted for by the use of halo signs (an early sign of infection [5, 10]), which were used at enrollment for approximately one-third of the patients in the study by Herbrecht et al. [7], compared with nearly 60% of the patients in the study by Cornely et al. [1] ( ). In accordance with this, the P ! .001 proportion of patients with proven IA, usually achieved later during the course of disease, was much lower in the study by Cornely et al. [1] (9% vs. 39%; ). P ! .001 Thus, the study by Cornely et al. [1] enrolled many more patients with early disease and a higher proportion of patients with a “good prognosis,” compared with the study by Herbrecht et al. [7]. In the study by Herbrecht et al. [7], all evaluations were performed at 12 weeks, regardless of how long the patients had received the initially assigned therapy, although a secondary analysis examined responses at the completion of assigned therapy. In the study by Cornely et al. [1], responses were evaluated at the end of treatment, which had a median duration of 14–15 days but ranged from 1 day to 60 days. Eighty-four–day (12-week) responses were not reported in the study by Cornely et al. [1], although survival was reported. In the study by Herbrecht et al. [7], major efforts were made by the Data Review Committee to distinguish patients who experienced partial responses, who had to have at least 50% improvement in their radiologic abnormalities, from those who experienced less improvement, who were categorized as having a stable response. Such efforts included clearly defined a priori criteria, duplicate assessments, and clinician input into the final