Pattern recognition produces more accurate disease detection

Pattern-recognition software has already proven its worth in the analysis of large data sets, ranging from remote geographic sensing to patterns of brain activation. Now it has been successfully applied to newborn screening. Using data on inborn errors of metabolism collected from newborns at 154 sites worldwide, the Regional Genetics and Newborn Screening Collaborative has developed and tested an analytic tool that reduces false-positive test results. The resulting software (available at http://region4genetics.org/msms_data_project/priority1) converts raw mass spectroscopy data into individual profiles and generates a composite score based on the overlap between values in the normal range versus the disease range. The pattern-recognition software differs from traditional statistical analysis in that an abnormal result is not defined by a deviation from normal but instead by an established analyte disease range set by a worldwide database of more than 12,000 patients affected with 60 metabolic disorders and 644 heterozygote carriers for 12 conditions. Retrospective evaluation of Minnesota cases suggests that use of the method would have halved the number of false-positives caused by carrier status for fatty-acid oxidation disorders and likewise would have prevented 88% of known false-negative events. The developers report that the tool can be applied to any clinical data set with sufficient data defining normal and disease populations.

Calls for comparative-effectiveness research in cancer genomics

see Building the evidence base for decision making in cancer genomic medicine using comparative effectiveness research

The methods used to generate and review medical evidence are rapidly evolving. Clinical trials structured to meet the requirements of the US Food and Drug Administration are becoming only the first step to ensuring clinical utility. A broad movement is afoot to use comparative-effectiveness research (CER) to contain health-care costs while maintaining (and potentially improving) quality. These patient-centric methods include not only comparative observational and randomized trials but also mathematical and economic modeling and even first-person reports about how medical treatments work in the real world. These methods, structured to extract valuable information from the volumes of raw data generated within the medical research enterprise, are now being applied in cancer genomics, as reported in the review by Goddard et al. The authors describe recent studies that examine issues such as whether single-nucleotide polymorphism testing adds to the predictive value of breast cancer risk screens and whether genotype testing to assist treatment decisions leads to better outcomes for patients. The authors argue that using CER methods in cancer genomics offers the potential for improving the synthesis of evidence from multiple sources and ultimately better informing clinical guidelines.