Main Data History
Show Index Toggle 7 comments


Search Criteria:
We searched for clinical trials that investigate acute lymphoblastic leukemia (ALL) in the pediatric population.  PubMed was searched to identify relevant studies. The search strategy is listed in Appendix 1.  The search was performed on Tuesday, November 10, 2015.


Coding, the process by which the published papers in our sample were read, reviewed, and documented, was done by C.C.W., W.D.B., J. W., and T.E.N. Articles used were the result of a PubMed search over the past 10 years. At the onset, 1086 potentially relevant papers were assembled. was used as a means of screening abstracts and titles to remove papers that were not useful. Each article required two separate ‘yes’ votes to be included in our study, and two separate ‘no’ votes to be excluded. Disputes were resolved via group discussion. 885 articles were included in our study of which 285 were randomly sampled and divided evenly four ways. After full text screening, 182 articles were included. Animal studies, follow-up studies of adult survivors of pediatric ALL, exclusively genetic studies, and genome-wide association studies were excluded. Meta-analyses were excluded to avoid redundancy, as we had already coded their constituent studies individually. Studies combining original research with meta-analysis were included so as to not exclude original research. (See Prisma Diagram).

An abstraction manual was formulated by D.H. to standardize the coding process. Each person was partnered with another so that all of the coding process was reviewed once over. The papers were analyzed for nine specific pieces of information: outcome, measurement device, method of aggregation, primacy of outcome, whether the outcome was a harm or side effect of an intervention, study design, study type, metric, & sample size. With regards to metric, when an outcome element was implicitly specified, we considered it specified. For example, because quantifying survival is, by definition, measuring time to event, specific metric for survival analysis outcomes was always coded as time to event. For survival, remission, and relapse, measurement device was coded as “N/A” because, other than a calendar, there is no measurement device. For outcomes reported using scales (e.g., NCI-CTC), metric was coded as "value at a time point" unless otherwise specified within the body of the article. When coding sample size for studies including non-pediatric-ALL research, all study participants were counted, including adult patients and pediatric patients with a cancer other than ALL. For the final step in coding, outcomes were grouped into eight domains for analysis: 1) Survival; 2) Mortality; 3) Remission; 4) Relapse; 5) Response to Treatment; 6) Adverse Event; 7) Cognitive Event; 8) Other.

STATA software was used to analyze frequency of appearance of unique outcomes and the specification of the nine outcome elements outlined in the abstraction manual. Unique outcomes were then placed in the eight broad domains listed above and run through STATA to reveal larger trends in reporting.
In order to structure a visual representation and calculate centrality of clinical outcomes in pediatric leukemia, a matrix was constructed. The foundation of this social network was formed using a basis of frequency of connections across outcomes, termed co-occurrences. Each outcome and the number of times it co-occurred with other specific outcomes were recorded in a spreadsheet. Reviewers C.C.W. and W.D.B. produced the network structure with a symmetrically duplicated matrix, ultimately serving to verify the co-occurrences.
We imported the network matrix onto UCINET and used Netdraw software. Each outcome was uploaded onto the program in the order of total co-occurrences. Thus, each outcome was sized in increasingly larger nodes, the plots of FIGURE **; the larger the size of the node, then the larger number of total co-occurrences this outcome maintains across outcomes in pediatric acute lymphoblastic leukemia. Next, the spring embedding function was applied to group outcomes around the largest nodes. This was accomplished by grouping less connected outcomes around nodes in a pattern of descending number of co-occurrences until the network became too dense for coherency. Next, a superstructure was formed, according to FIGURE **, which represents the social network architecture of outcomes.