Statistics professor Paul Horn uses his expertise on reference intervals to help create guidelines set by the Clinical and Laboratory Standards Institute.

Photos By: Kim Burdett

Most patients who receive blood tests don’t see the report that comes back from the lab describing the healthy or unhealthy levels of chemical analytes in their blood. But if they did, they would notice a range of numbers for each tested analyte that a healthy individual should fall into.

If the patient’s numbers reside within the range, they may get a green light from the physician and be considered healthy. Should the patient fall out of the “normal” range, however, a red flag may be raised for the physician to further investigate.

Paul Horn is considered a renowned expert on statistical problems involving reference intervals. |

But these ranges aren’t static. What determines a “normal” range for these chemical analytes changes not only across countries or states, but laboratories as well. Theoretically, then, a patient can get the same blood test analyzed by two separate labs and get conflicting results.

The reason is because laboratories must determine their own normal ranges—or reference intervals as statisticians and clinical chemists call them—to best fit the population it serves. While your triglyceride count resides within normal limits in Cincinnati, it may be red-flagged in Beijing.

So how does a laboratory determine these intervals exactly?

Paul Horn, a statistics professor in McMicken’s Department of Mathematical Sciences, has dedicated much of his research to the topic. In fact, the work he does on reference intervals is influential enough to have an effect on any individual in the world—all due to his collaboration on the recent guidelines set forth by the Clinical and Laboratory Standards Institute (CLSI).

It was back in 1994 when Horn began working with College of Medicine Professor Amadeo Pesce. When his statistical interests overlapped Pesce’s research in clinical chemistry, they started to collaborate on research needed for reference intervals.

“Professor Pesce was interested in reference intervals and I was interested in quantile—or percentile—estimation, and it turns out that they are the same problem. The basic idea is that we want to figure out what the lower and upper cut offs are for the healthy populations.”

The two even co-wrote a book on reference intervals that was published in 2005.

Today, Horn is considered among colleagues as an expert on the subject, confirmed by his invitation a few years ago to join an advisory board as the sole pure statistician of the CLSI. The board worked together on an updated guideline that sets reference interval standards for laboratories worldwide.

“Defining, Establishing, and Verifying Reference Intervals in the Clinical Laboratory; Approved Guideline—Third Edition” was released in December after years of collaboration.

Horn’s work with the institute involves describing the methods he uses for constructing reference intervals that are robust, meaning they are not as sensitive to outliers in the data, they use the data efficiently, and work well even when based on a relatively small sample.

When a laboratory tries to determine the reference intervals of a healthy population, it aims to do so with a sample size of 120 individuals. This sample size allows for the upper and lower 2.5 percentiles of the distribution to be accurately estimated, thus yielding a 95 percent prediction interval that demonstrates the “normal” range the chemical analytes should fall into.

“The question is what if you only have 80 people in your sample,” Horn said. “It can get very tricky because a lot of times, these measurements can be gender, age or race specific,” Horn explained. Rather than only needing 120 healthy individuals, laboratories might need more specific calculations, like 120 healthy African-American males between the ages of 50 and 60.

“It can be quite a large task for a laboratory.”

By working on robust models, however, Horn is able to lower the sample size while not compromising the confidence interval, thus allowing for a more efficient way of determining reference intervals. While he’s been able to lower the sample size significantly, CLSI still recommends having a sample size of at least 80 healthy individuals to determine accurate reference intervals.

“The method I’ve developed may be used now as an alternative,” Horn said. “The idea is that we will try to extract as much information from the data as possible without letting the outliers have too much influence.”

Horn worked with the CLSI for three years on the guidelines, and felt a sense of accomplishment when he finally received a hard copy of the new edition in the mail. Yet he’ll be the first to admit that his work is nowhere near completion.

“Like all research, there are always more questions,” Horn said. “You have to keep asking new questions to avoid getting tired of the same area.”

The professor said he’ll continue to conduct his research on reference intervals, but plans to shift gears from robust models to focus on other statistical problems, involving a unified approach to data transformation and outlier detection.