Meta-analysis of prophylactic or empirical antifungal treatment versus placebo or no treatment in patients with cancer complicated by neutropenia.

Introduction

Bacterial infections are a major cause of death in patients with cancer,[1] particularly when they have neutropenia.[2] Disseminated fungal infection, most often caused by Candida and Aspergillus species,[3] is also considered to be an important cause of morbidity and mortality.[4 5] The death rate in patients with candida sepsis or deep tissue infection is around 75%,[4 5] and a positive blood culture or histological evidence of invasion was found before death in 37% of patients in one series.[2]

Antifungal agents are often given prophylactically in conjunction with chemotherapy or bone marrow transplantation or empirically to patients without documented fungal infection but with persistent fever despite antibiotics. The rationale is to start treatment before it is too late--that is, before death is inevit able--as it is difficult to diagnose an invasive fungal infection with certainty.[4 6] Studies with historical controls have shown a positive effect of antifungal agents on mortality,[3 7] but such non-randomised comparisons substantially overestimate the effect of cancer treatments.[8] Randomised studies have yielded varying results, and the power of most of them has been very low. We conducted a meta-analysis of trials in which a commonly used antifungal agent was compared with non-treatment in a control group.

Methods

The primary aim was to determine whether commonly used antifungal agents decrease mortality. Secondary variables were the effect on invasive fungal infection and colonisation.

Search strategy

All randomised trials, irrespective of language, of amphotericin B, various lipid soluble formulations of amphotericin B (for example, AmBisome), fluconazole, ketoconazole, miconazole, or itraconazole compared with placebo or no treatment in patients with cancer complicated by neutropenia were eligible. We excluded studies which solely concerned treatment or prevention of oral candidiasis. A Medline search from 1966 on SilverPlatter was developed iteratively. The final strategy could retrieve all relevant trials from whatever source if they were registered in Medline. The search was most recently updated in February 1996.

One or more of random* control* blind* clinical-trial in pt, clinical-trials/all subheadings, placebo* and tg=comparative-study were combined with one or more of amphotericin, AmBisome, fluconazol*, itraconazol*, ketoconazol*, and miconazol* and with one or more of bone ?marrow *, transplant*, cancer*, fungemia, h?ematologic*, malignanc*, neoplasm, neutropeni*, granulocytopeni*, leuk?emi*, lymphom*, sepsis, septic*, intensive-care/all. subheadings, intensive care, and immunodeficiency. Information about trials not registered in Medline, including unpublished trials, were located by contacting the pharmaceutical industry and the authors and by scanning reference lists of articles and reviews. We also scanned selected conference proceedings--namely, the interscience conference of antimicrobial agents and chemotherapy, 1990-5; the general meeting of the American Society for Microbiology, 1990-5; and the seventh European congress of clinical microbiology and infectious diseases, 1995.

Data extraction

Decisions on which trials to include and which variables to use when more options were available for the same outcome were based on the methods sections of the trials only. Details on diagnosis, drug, dose, rules for use of additional (rescue) antifungal agents, average length of treatment with placebo, length of follow up, randomisation[9] and blinding methods, number of randomised patients, number of patients excluded from analysis, deaths, invasive fungal infections, colonisation, and use of rescue drugs were extracted by each of us independently. Disagreements were resolved by discussion.

We defined invasive fungal infection as a positive blood culture, oesophageal candidiasis, lung infection, or microscopically confirmed deep tissue infection. We excluded cases of oropharyngeal and vulvovaginal candidiasis, skin infections, Candida in the urine, and vaguely described infections. To check the robustness of our findings we also analysed fungal infection according to the authors" own definitions.

Authors were asked to confirm the extracted information and answer additional questions. For six trials [27 30 31 35 36 39] we obtained additional outcome data. Numbers in the tables therefore differ from those in the published articles. To increase the response rate we used Medline to obtain authors' most recent addresses. In an attempt to increase the power of the meta-analysis and avoid reporting bias we specifically asked authors for three months' mortality data for all randomised patients, including those subjected to secondary exclusion. We also sought details on the randomisation process, especially whether randomisation was concealed and irreversible so that an allocation could not be known beforehand or changed later. This question, however, seemed not always to have been well understood. We considered randomisation to have been concealed when central randomisation, sealed envelopes, or a code provided by a pharmacy or a company was described. On one occasion what seemed to have been sound randomisation provided by a pharmacy proved on further questioning to be medicine packages labelled A and B, which we would not have expected in a trial published in 1993.[39] With such a procedure, should the code be broken for just one patient it would be possible to predict all future allocations.

Statistics

Outcomes were weighted by inverse variance. As the studies were expected to be heterogeneous because of the various designs, diagnoses, drugs, doses, routes of administration, and criteria for fungal invasion, colonisation, and use of rescue drugs a random effects model was used.[10] When the P value for heterogeneity exceeded 0.10, however, a fixed effects analysis was preferred.[11] We used the Meta-analyst 0.975 program.[12] Odds ratios were calculated with 95% confidence intervals.

Results

We identified 31 [13-42] of which 29 were reported in English and two in Japanese. Seven trials were excluded: one was not truly randomised[41]; in another only 14 of 146 patients had neutropenia and only data on oropharyngeal candidiasis were provided[42]; a third trial was concerned only with a subgroup of 72 of 298 randomised patients who came to necropsy[13]; and four trials were unpublished[29 35 38] (H Brincker, personal communication). Of the 24 trials reviewed, two were published only as abstracts[28 29] and one was published as an interim analysis.[22]

An antifungal agent was given prophylactically in 21 trials and empirically in three (table 1). Acute leukaemia was the most common indication in 14 trials and bone marrow transplantation in 10. The total number of randomised patients was 2758 (table 2). Duration of follow up was given in only 13 trials (54%; median 56 days). Probably it varied for different patients within the same study, as several authors stated explicitly that the trial drugs had been given till the neutropenia had resolved [15-21 24 25 30 35 37 39] The average number of days on placebo was reported in 16 trials (67%), in which the median was 20 days. Use of rescue antifungal treatment was more common in the untreated groups (odds ratio 0.68; 95% …