We used a broad search strategy with special emphasis on randomized controlled trials. We used a variety of electronic databases, including MEDLINE via Pub MED, Ovid, and the Cochrane library.
First we identified key terms of the study drugs “liposomal doxorubicin”, “Doxil” “Myocet”, “Doxorubicin”, “Daunorubicin”, “Epirubicin”,“Mitoxantrone”,and “Idarubicin”. The key word “liposomal doxorubicin” compared with “Doxorubicin” was searched using the MeSH subheadings without any language barrier. The searches were combined with the key word search “randomized controlled trials”. 480 related articles were found, but only randomized controlled trials comparing liposomal doxorubicin and conventional anthracycline were selected for the meta-analysis. The selected randomized articles were again searched for related topics on the Pub MED database. We also searched unpublished studies with results through ‘clinical trials.gov’ and the American Society of Clinical Oncology (ASCO) website. We supplemented the searches by reviewing the bibliographies of key papers.
All randomized controlled trials that compared the efficacy of the liposome encapsulated doxorubicin with the conventional anthracyclines on any cancer with or without other chemotherapy were considered eligible for the analysis. The randomized controlled trials on pegylated liposomal doxorubicin (Doxil®/CAELYX™), non-pegylated liposomal doxorubicin (Myocet), and liposomal Daunorubicin (DaunoXome) pooled for the meta-analysis, irrespective of tumor types and stages. The controlled arms must include one of the conventional anthracyclines, epirubicin, daunorubicin, doxorubicin, idarubicin and mitoxantrone.
Non randomized trials were excluded. Randomized controlled trials with two different anthracyclines, but neither of them are a liposomal anthracycline were also excluded. Incomplete ongoing randomized trials with no published results were excluded as well.
The following information about each trial was recorded: first author, journal name, year of publication, number of patients assigned, median age of the patients in each study, diagnosis, drug combinations and dose of treatment and the cumulative dose of anthracyclines. The adverse effects of the liposomal and conventional anthracyclines were analyzed in two arms. The variables for the adverse effects include cardiac toxicity, hematology toxicity, hand-foot syndrome or palmar plantar erythrodysesthesia (PPE), febrile neutropenia, alopecia, nausea and vomiting.
Different criteria were used to define the grade of toxicity. The majority of the trials (6 out of 9) used the National Cancer Institute-Common Toxicity Criteria (NCI-CTC). The remaining two trials used the WHO criteria for toxicity and one trial used South West Oncology Group (SWOG) toxicity scoring system. The differences on the toxicity grading of the variables were essentially unremarkable among the criteria .
The primary end points were the adverse effects: cardiac toxicity with congestive cardiac failure and significant reduction in the left ventricular ejection fraction (LVEF) were entered in separate arms. The information about the cumulative dose and cardiac toxicity was extrapolated in a separate excel sheet. Four out of the nine studies provided the data about the cumulative doses at which cardiac toxicity was developed. Analysis of cardiotoxicity included comparison of the proportion of patients in each treatment group who developed cardiotoxicity (by protocol specified cardiac event) at any time during the study, as well as comparison of the mean percentage change in LVEF from baseline.
The hematology toxicities (anemia, leukopenia, neutropenia and thrombocytopenia) were extracted to four variables representing the grades of toxicity. The information of the different grades of hematology toxicity was not available in two of the trials. We analyzed the number of the incidence of toxicity of any grade giving specific importance to the grade toxicity.
There is difference in the grading of alopecia between the WHO and NCI- CTC criteria. Alopecia, either partial or complete, was analyzed as one variable.
The nausea and vomiting were included as one variable. Two studies did not include the information about nausea and vomiting. Four out of 9 studies did not have grading for toxicity. Toxicity of all grades was therefore considered as a single variable.
We performed the meta- analysis using the statistical software ‘Stata’ 10 version (statistics and Data created by Stata corp.). The existence of heterogeneity was tested using the Chi square statistics. The heterogeneity was quantified using the I-squared. For those variables with high heterogeneity(p<0.05 for I-squared analysis), the data were analyzed using the random effect model. Otherwise, fixed- effect model was used for the data analysis. We calculated the odds ratio (OR) with 95% confidence intervals (CIs) from the data extracted from the original studies separately. The overall ORs were calculated from the pooled data. We eliminated studies if the data were not available for a particular variable.