OBJECTIVES: This is the first systematic review to synthesize all available studies that examine the effect of data mining technologies on improving cancer prediction, detection, prevention, and cost-effectiveness.

METHODS: We conducted a systematic review and meta-analysis. The PRISMA framework guided the conduct of this study. We obtained papers via MEDLINE, Cochrane Library, EMBASE and Google Scholar. Quality appraisal was performed using Downs and Black’s quality checklist. Effect sizes of adverse healthcare utilization were computed using Cohen’s D. positive effect sizes. We also constructed a forest plot to show the range of different effect sizes, and a funnel plot to examine the effect of bias and sample size, by using STATA 12.0.

RESULTS: A total of 252 studies were reviewed. Eleven articles met eligibility criteria and were included into the final analysis. Most of studies appeared to implement techniques including artificial neural networks, Bayesian networks, decision trees, and support vector machines. These techniques were applied in cancer prevention for the development of risk identifiers and predictive models, and thus resulted in accurate and effective decision makings. Overall effect of data mining on cancer early detection was positive and significant (d=0.36, p= 0.041). In terms of cost-effectiveness, our findings showed an overall positive direction, but there were apparent limitations in study quality to perform a meta-analysis.

CONCLUSIONS: The adoption of data mining technologies for cancer prevention is at an early stage. The results showed that it is possible to utilize these techniques to support best-practice of cancer prevention. Evidence shows that the data mining can improve cancer early prediction, detection and cost-effectiveness, but limited numbers of included articles and heterogeneity of those studies implied that more rigorous research is expected to further investigate the techniques’ effects.