We searched several databases such as PubMed/Medline, the Excerpta Medica Database (EMBASE), the Cochrane Library, database of DIMDI (Deutsches Institut für Medizinische Dokumentation und Information) and CAMbase for clinical studies focusing on survival of cancer patients using Iscador extracts. Separate search terms were "Iscador" and "study", "mistletoe" and "study", and "Viscum" and "study". Finally we asked several experts for gray literature not listed in the above mentioned databases, checked the reference lists of relevant articles and authors, and contacted the manufacturer of mistletoe preparation. The complete search was performed between February and April 2008.
Inclusion criteria were controlled clinical studies (at least historic or literature) on parameters associated with survival in cancer patients treated with the VA-E Iscador, published in English or German language journals. We excluded field reports, case series, case reports, studies without any control group, abstracts which proceeded a full length publication, translations of already published manuscripts, double publication of similar data (exception is the presentation of further data), internal reports and unpublished manuscripts. In a few cases we had to exclude studies because of the simultaneous usage of the fermented extract Iscador and the aqueous extract Helixor.
Analysis of data
Two review authors independently assessed trials for inclusion in the review. They took part in the extraction of data and assessment of study quality and clinical relevance. Disagreements were resolved by consensus. We graded the methodological quality of the studies by the following checklist (rater assessment): Adequate description of the study design (retrospective, prospective, retrolective, multicenter study etc.), subject assembly process (randomization, matched pairs, etc.), comparability of groups, description of drop outs, allocation concealment (analysis of concealed treatment allocation was difficult because most studies did not provide sufficient data to judge - either there were no statements or information are at least unclear), description of the intervention (dosage and duration of VA-E application), description of statistical analysis, external validity (representative patients, relevant therapeutic concepts, generalization of results). However, we did not explicitly refer to rating scores such as the JADAD, because blinding of VA-E application is difficult and, due to ethical reasons, rejected by several medical doctors. Thus, 2 out of 5 criteria of the JADAD score were not applicable for these studies; nevertheless, randomization as a criterion was assessed, also dropouts.
The reporting of the results adhered, if possible and appropriate, to the MOOSE guidelines , which involve recommendation for the description of study selection, presentation of results (including a table with descriptive information for each included study), and discussion of biases, consideration of alternative explanations for observed results etc.
If a trial was found to be eligible, assessments of its methodological quality were done independently by two reviewers (AB, TO) and recorded on a pre-especially designed data form together with the basic trial data and the extracted results. Allocation concealment was assessed in accordance with the Cochrane guidelines:
A = adequate (telephone randomization or using consecutively numbered, sealed, opaque envelopes)
B = uncertainty about the concealment (method of concealment is not known).
C = inadequate (e.g. alternate days, odd/even date of birth, hospital number)
Disagreements on methodological quality ratings were discussed by both assessors until they reached a consensus.
Data were independently extracted by two persons (AB, TO) and independently entered into a data form which was especially designed for trials on VA-E by a third person (CR). If the data entries differed, both reviewers were contacted to recheck the publications and were forced to come to a consensus, which could be reached in all cases.
Data on the following topics were:
* Details of the publication (first author, country, year, journal)
* Details on the dosage and application of Iscador
* Type, name, dosage and application of the control therapy/alternative therapies
* Grading and location of cancer
* Age and gender distribution of patients
* Methodological quality of the study (see above)
* Outcome(s): Survival (median survival, overall survival, 3-, 5- or 10-year survival etc.)
All data were separately analyzed for (a) placebo controlled trials, (b) actively controlled trials, and (c) trials where patients of the control group received only standard care but no extra treatment. Control groups where patients were "insufficiently treated" with the VA-E (i.e., < 3 packages within several months/years) were counted as having received no extra treatment. In one study, the patients received a glycopeptide preparation from sheep spleens (Polyerga Neu) versus Iscador versus a complex of B-vitamins which was regarded as a placebo, but was regarded as an "alternative therapy" in our analysis. Active controls used in the studies were interferon alpha2b or interferon gamma, Bacillus Calmette-Guérin (BCG), Vitamin B complex, or radiation.
Outcome data were extracted as hazard ratios (HR), their logarithm, and the respective standard errors. If HRs were not given, we assumed that the overall survival was exponentially distributed. This allowed us to estimate hazard ratios, even if only median survival times or survival rates were given in the publication. HR < 1 indicate superiority of Iscador, HR > 1 indicate superiority of the control condition.
The association between study size and trial results was graphically displayed in funnel plots, by plotting HRs on the horizontal axis (in a logarithmic scale) against their standard errors - or against the total patient numbers - on the vertical axis . Funnel plots are adequate instruments to detect small study size effects, including publication bias. In the absence of bias, results from small studies should scatter widely at the bottom of the graph, with the spread narrowing among larger studies. Publication bias may lead to asymmetrical funnel plots. Moreover, the asymmetry of the funnel plot was further explored by a weighted linear regression analysis (meta-regression) which modelled the log HR as a function of its standard error . Weights were chosen inversely to the squared standard error. From this model, the asymmetry coefficient (AC) was estimated as the slope of the regression line.
Heterogeneity between trials was assessed by standard χ2-tests and the I2 coefficient  which measures the percentage of total variation across studies due to true heterogeneity rather than chance.
Overall estimates of the treatment effect were obtained from random effects meta-analysis . Additionally, from meta-regression a predicted HR was obtained for trials with a standard error as small as the smallest observed standard error of all included trials.
The extent to which study-level variables were associated with log HRs was investigated by fitting multivariable meta-regression models. The following variables were considered: standard error of log HR, tumour localization (breast, stomach, lung, colon, ovary, corpus, skin yes/no), randomization (yes/no), matched-pair comparison (yes/no) - due to the fact that all matched-pair studies were from the same source.