Enormous volumes of data flow through laboratory information systems each day. Orders are received, specimens are tracked, results are entered and verified and reports are sent back to the ordering physicians. Most laboratories realize value from the transactional activity in these databases--the use of laboratory information systems to support daily lab testing and result delivery. Few laboratories, however, have begun to tap into the enormous benefits of aggregating and analyzing that data at other levels. Clinical laboratories can benefit greatly from adopting a data-centric mindset.

Enhanced QC
Michael Lewis' bestseller Moneyball chronicles the Oakland A's baseball team during the 1990s.¹ With one of the lowest payrolls in Major League Baseball, the A's could not competitively recruit star players away from teams such as the New York Yankees (with around five times the total payroll budget). Yet each year the A's managed to have a winning season and make the playoffs. How did they do it? They used analytics to improve decision making.

Though baseball has had a reputation as a statistics-heavy sport, the truth is managers and owners have historically relied primarily on intuition and non-rigorous data for recruitment and other key decisions. Billy Beane, General Manager of the A's, changed all that with the help of statisticians. He first identified the statistics that best correlate with winning games, and then rigorously used those statistics for decision making. For example, he discovered that on-base percentage is a more reliable measure of a batter's efficacy than batting average (because batting average excludes walks). Beane regularly recruited players with high on-base percentage but modest batting averages whom other teams had passed up.

Medicine, likewise, has a history of relying on a mix of human intuition and incomplete evidence, though the proponents of evidence-based medicine are trying to change that.³ An interesting laboratory analogy is quality control (QC). This may seem an odd example, since QC with its Westgard rules represents one of the most statistical areas of laboratory decision making. Yet traditional QC is incomplete in its reliance on artificial controls. In our laboratory we have identified cases where switching to a new reagent lot results in a significant change to patient results without a corresponding change in control values, presumably due to matrix effects. If the controls are consistently "in" yet the assay appears to be drifting, how should a laboratorian decide what to do other than follow intuition? Using human material for controls would be one solution, but is not feasible in most cases. What our lab does instead is monitor median patient result values to detect changes over time. This doesn't replace traditional QC, but rather is complementary to it. On a typical run, patient values vary widely in ways that depend mainly on the patient, not the assay. But suppose you have a high volume assay, say 2,000 results per week. The median of those results should be extremely close to the median of the results last week, or that of the week before. If it's not, then it's more likely due to changes in the assay itself than changes in the patient mix.

Utilization Analysis
In the mid-1990s, bookselling might have seemed an unlikely market sector for advanced informatics. After all, the basic business model was quite simple-offer a broad menu of items, accept orders from customers, fill the orders and collect the corresponding charges (not all that different from a clinical laboratory.) Yet Jeff Bezos saw an opportunity to do more, particularly around data analytics. The result is Amazon.com, which has been one of the enduring business successes of the dot-com era.³

As you browse books on Amazon, the Web site presents a broad range of information ranging from what other customers who viewed this book ultimately bought, to an accurate estimate of shipping times given the different shipping options. This information comes from aggregating and analyzing customer data. The resulting information-rich customer experience is one of the main reasons why Amazon's sales figures continue to grow within an otherwise dismal retail economy.

Like Amazon, laboratories have the opportunity to add additional information value to our customers (patients, physicians and other laboratories) by aggregating and mining the data that we generate. One example in our own laboratory is our ATOP^® (Analyzing Test Ordering Patterns™) program that screens for patterns of inappropriate laboratory utilization by physicians. The largest contributor to excess laboratory costs is inappropriate test ordering by physicians. This represents a huge opportunity for laboratories to steer local physicians toward more appropriate ordering patterns. But first, pathologists and laboratory managers need detailed data on what tests are being inappropriately ordered and by whom.

For example, suppose a laboratory receives an order for an HCV RIBA (recombinant immunoblot). For most patients receiving testing for HCV, the CDC recommends confirmation by viral load rather than by RIBA. But in the absence of other clinical information, it isn't possible for the laboratory to tell whether that isolated RIBA order is appropriate or not. On the other hand, suppose we knew the ratio of RIBAs to viral loads for that physician or clinic over a long time interval. For doctors following the CDC algorithm, that ratio is typically 1:10 or more. A lower ratio, say 1:2, would suggest wasteful overuse of RIBA.

Besides ratios of tests within a single disease category, normalized test volumes across facilities often identify outlier ordering practices. Distributions of patient age, patient sex and/or test results can also be compared to the distributions we would expect if doctors are following available guidelines. Armed with this information, laboratory managers and pathologists can approach local clinician leaders to discuss strategies to improve patient care and save money.

Final Thoughts
Just as in professional baseball and bookselling, the most successful laboratories of the future will be those that embrace data analytics to improve their internal processes and provide more value to their customers. The software tools are widely available; now we just need laboratory professionals with the vision to take advantage of them.

Dr. Jackson is the medical director of Informatics at ARUP Laboratories and an assistant professor of Pathology at the University of Utah School of Medicine.

References

1. Lewis MM. Moneyball: The art of winning an unfair game. New York: W. W. Norton & Company, 2003.

2. Beane B, Gingrich N, Kerry J. How to take American health care from worstto first. New York Times. Oct. 24, 2008.

3. Amazon. Accessed: www.amazon.com.