When a baby is born, a few drops of blood are collected by heel prick and spotted on filter paper. The dried blood spots are submitted to public health laboratories for a series of screening tests, which include the analysis of amino acids and fatty acid esters called acylcarnitines by mass spectrometry for the detection of more than 40 inherited disorders. This test is quite complex as one specimen may translate into more than 100 markers and ratios. Such profiles are routinely interpreted by applying analyte-based cutoff values and sequential algorithms, visual pattern recognition for certain data subsets may be helpful but it is subjective and inconsistent. Overall, the current status quo is of concern because the false positive rate can be as high as 1-2 percent in some programs and false negative events are pervasively frequent.
To improve analytical performance and patient outcomes, there is now a different way...
Since 2005, the analytical results of millions of healthy newborns, as well as profiles of almost 13,000 diagnosed cases of 70 different conditions have been uploaded to a web-based database. There, the cumulative percentiles of both normal ranges and condition-specific disease ranges have been constantly updated. Clinical significance for a condition is assigned to a marker when the median concentration level of true positive cases is beyond either the 1st or 99th percentile of the normal population. This selection process maximizes the clinical utility of every target in a profile, including markers that would have limited discriminating power if taken in isolation. Through concurrent evaluations of all possible markers for any particular conditions, previously unforeseen trends and informative ratios have been recognized and added as unique and discriminating features, resulting in fewer false-positive and -negative outcomes.
The collective information in R4S has become the dynamic training set for a multivariate pattern recognition software that generates post-analytical interpretive tools. Available since November 2011, tools target either a single condition or the differential diagnosis between two or more conditions that have an active tool based on a sufficient number of cases. Another feature, the dual scatter plot, shows the distribution of the scores for the same case calculated with two different tools, which together are designed to better discriminate between the two conditions with overlapping phenotypes. The plot provides an immediate visual report of likelihood of one condition over the other.
All informative results of a patient are merged into a single score. The score is determined by the degree of overlap between normal and disease ranges, the penetration of a given result within the condition-specific disease range, differences between conditions, and weighted correction factors. Scores are expressed as the percentile rank among all cases with the same condition, and are compared to selectable interpretation guidelines (for example: not informative, possible, likely, or very likely) based on thresholds in the percentile ranking of scores calculated for true positive cases.
The novelty of establishing condition-specific disease ranges of clinically validated markers is that they can obsolete strict analyte-based cutoff values in the interpretation of complex laboratory profiles. There is a compelling reason to do so. In a previous report of the collaborative project, 43 percent of more than 5,300 cutoff values used in screening laboratories overlapped significantly with many disease ranges, and therefore had intrinsically poor sensitivity, risking false negative events. But the most striking observation was that almost half of these inappropriately set cutoffs were applied to 37 markers with no overlap between normal population and disease ranges, indeed a situation where selecting a clinically effective cutoff value should be a straightforward process.
In addition to providing more accurate diagnoses of metabolic diseases in newborns, a new initiative is in progress to upgrade the IT infrastructure to host a broad spectrum of clinical and research endeavors that could benefit from this mode of interpretation of complex data sets. This system, called Collaborative Laboratory Integrated Reporting (CLIR), will rely on the same principles of fostering collaboration, data sharing, peer comparison, and widely available, up-to-date tools that can be customized to the needs of clinicians and researchers around the world.
Piero Rinaldo is a co-director of the Biochemical Genetics Laboratory at the Mayo Clinic College of Medicine in Rochester, Minnesota. He is a professor of laboratory medicine and the T. Denny Sanford professor of Pediatrics. For access to the R4S website send requests for password to firstname.lastname@example.org.