Background: For many conditions, several treatment options may exist. Multiple-treatments meta-analyses examine the relative merits of these options. The geometry features of a network of treatments may reflect the wider clinical context of the evidence, and they may be shaped under rational choices for treatment comparators or by specific biases. Objectives: To examine and explain the implications of the geometry of networks of evidence about treatments; their diversity (how many treatments are involved and whether some are more extensively tested than others); and their co-occurrence (whether some treatment comparisons are preferred while others are avoided). Methods: We identified published networks of randomized trials that included at least four interventions, irrespectively of the setting, the condition and patients characteristics. We adopted indices from the ecological literature that measure diversity and co-occurrence. We use the Probability of Interspecific Encounter (PIE) index, the probability that two randomly sampled treatment arms from the network have been allocated to two different treatments to measure diversity. Co-occurrence reflects whether there is a special preference or avoidance for specific pair-wise comparisons of specific treatments. The C-score reflects the overall average tendency of the treatments not to occur together. A permutation procedure was used to obtain the statistical significance of the C-score. Results: We identified 18 eligible treatment networks for diverse diseases. They involved 4 to 16 treatments and included 10 to 84 trials each. The clinical context is discussed for each of the 18 networks, and the evidence is interpreted in light of the observed network geometry. Some networks were starshaped: one option (typically placebo or no treatment) was the typical standard comparator, even though several treatments might have shown effectiveness. Other networks showed significant co-occurrence, and we identified specific avoided head-to-head comparisons. These choices could be justified (e.g. newer treatments not compared against old ones already shown to be inferior) or may have reflected preference biases. Conclusions: Evaluation of the geometry of a treatment network can offer valuable insights for the interpretation of the total evidence when many treatment options are available.