Small world networks and treatment discovery process in cancer.

Abstract

6086 Background: The discovery process associated with clinical research is inherently a social enterprise, which means that the pattern of discovery depends on the extent of sharing of ideas within existing scientific research. Clinical trials are often conducted within the same organizational framework of development of therapeutic advances (e.g., NCI cooperative group system [COG]). We hypothesize that the proportion of treatment discovery in cancer should be a function of the extent of interactions between trials at the level of each key social component in the NCI network (cooperative group, type of disease or treatment). Methods: Social network analysis is utilized to model the role of RCT interactions to treatment success. Our data set consists of 280 RCTs enrolling 91,847 patients conducted by the NCI-COG from 1955 to 2006. Each trial is described by three components: the cooperative group that proposed the trial, the type of disease and the type treatment studied. Interactions between trials occur for trials with common components. Results: The pattern of interactions between trials is not random. Rather, interactions between trials follow the small world network model indicating that all trials are connected through a small number of ties. In comparison to random pattern of social interactions, the information exchanged within the COG improved by 45%. Similarly, the COG network exhibited greater than 10 times tendency to cluster together. We further found that trials with the highest centrality measures (closeness and betweenness) were the ones that tested curative/definitive treatments in solid tumors. The most influential group within the NCI network was the ECOG COG. However, the observed higher level of connectivity in the network did not translate into predictably higher treatment success as measured by improvement in survival. Conclusions: The pattern of therapeutic discovery adheres to the small world social network model. On surface it appears counterintuitive that there is no direct relationship between treatment success and RCT interactions. We explain our findings by the role of social network to maintain equipoise in RCTs, which in turn preserves unpredictability of the results at individual RCT level. No significant financial relationships to disclose.