CLN Article

One size does not fit all

Long discussed but not yet implemented in routine care, personalized reference intervals are gaining renewed attention as advances make individualized laboratory interpretation more feasible.

Grace Browne

iStock-2291928290

When a patient visits the doctor to have a blood test, their blood sample is drawn and sent to the lab, and a few results come back: cholesterol, glucose, a white blood cell count. Next to each number on the report sits a reference interval — a range that helps the doctor judge whether the result counts as normal or not. If the number falls inside the interval, it goes without further inspection. If it falls outside, it may prompt a second look, a follow-up test, or a diagnosis.  

For decades, this has been the basic logic behind how blood tests are read. Reference intervals are used to interpret a patient’s laboratory test results, which in turn are used for the diagnosis and monitoring of diseases, as well as to guide treatment. Typically, reference intervals are derived from a sample of an overall healthy population. But the heterogeneity of the human population means that they are not one-size-fits-all. What represents a normal range for one person may not be the same for the next. To improve on this, population reference intervals may be partitioned based on factors such as sex and age. But still, the broad brush nature of the data lacks the nuance needed to work for every individual.  

That means that for some specific markers, a patient’s measurement result may fall within the population-based reference interval but still represent an abnormal result for that particular person. For instance, a patient's white blood cell count could fall within the "normal" range, but their measurement could still carry increased risk for that specific individual. On the other hand, a patient may get a result that lands outside the typical population-based reference interval, but that is actually normal and healthy for that individual.

For decades, experts have pointed to the usefulness of reference intervals that are based on an individual’s own repeated measurement results that were taken when they were healthy. The idea is that their intervals would then reflect their own physiology, rather than the general population, and would be specific for them.  

This is not a wholly novel concept. “People in the ’70s actually proposed personalized reference intervals,” said John Higgins, MD, professor of systems biology at Harvard Medical School and a pathologist at Massachusetts General Hospital. “It’s not a new idea by any means, but  there are some reasons why it’s becoming more interesting.” One of the biggest is that there are a lot more laboratory data available these days, meaning more historical information on people, which makes it easier to devise these more personalized intervals. Recent advancements in medicine are allowing for more and more personalized medicine in general.  

Yet despite decades of interest and a growing recognition of their potential advantages, personalized reference intervals remain largely absent from routine clinical practice. No laboratories have implemented them on a broad scale. The reasons are complex, ranging from scientific and operational challenges to questions about how clinicians would use and interpret personalized results. As researchers continue to explore their promise, they are also working to determine where these intervals could have the greatest impact and what evidence is needed to bring them into everyday care.

Where things currently stand

There have been some areas where this work has progressed more than others; hemoglobin is one. Between different healthy people, hemoglobin levels vary widely, but for any one person, the level tends to stay remarkably stable over time. That means a result that’s still “normal” by population standards could represent a meaningful shift for an individual, said Aasne K. Aarsand, MD, PhD, director of the Norwegian Organization for Quality Improvement of Laboratory Examinations, and a consultant medical biochemist in the Norwegian Porphyria Centre, Haukeland University Hospital.  

Elsewhere, the work has been much slower. And the widespread use of population-based reference intervals comes with various cons. “You can think of a population reference interval as really just saying, here’s what healthy looks like. It’s very well designed to pick up abnormalities, so it captures health if you’re in the middle, but not outside of it,” said Brody Foy, DPhil, assistant professor at the University of Washington department of laboratory medicine and pathology.

Personalized reference intervals also could be helpful not only for detecting a deviation from health, but also in the other direction. Foy and Higgins have done some work to show that personalized reference intervals can help people who are returning to good health figure out what their healthy state looks like with more accuracy, which in turn can help them determine whether they’re really heading towards their healthy baseline. “Even for really sick patients in the hospital, in the intensive care unit, knowing what their healthy baseline really looks like can help physicians figure out when somebody is on a good recovery trajectory and when they’re deviating,” Higgins said.

A data-fueled shift

There is a major reason why now is the time to start thinking about personalized reference intervals: We have access to more health data than ever, particularly in higher-income countries. A major step forward on the path has been the introduction of electronic health records (EHRs), Foy said, which have only become the norm in the last 5−10 years. These records mean that much of a patient’s medical history, including test results, is stored in a centrally located, easily accessible database.  

One technological innovation in particular could speed up the implementation of personalized reference intervals: Artificial intelligence (AI) could be used to integrate relevant individual-specific information from EHRs, refine the selection of appropriate historical data, and potentially improve the robustness and clinical relevance of personalized reference interval estimation.  

At the same time, although Higgins acknowledges that AI could be used to analyze existing data in a helpful way, he thinks pretty basic calculations would help, too. “I don’t think there's evidence we need fancy methods, and to some degree the simpler methods are preferable, because then they can be used everywhere and people can learn how to use them in different environments,” he said.

Hurdles on the road to adoption

The implementation of personalized reference intervals will depend on a number of factors. First, there must be robust, appropriately designed clinical studies that can demonstrate that personalized reference intervals have benefits over population-based reference intervals. These studies have to show that personalized reference intervals not only provide more accurate results, but that they also offer earlier or more accurate detection of certain diseases and improve patient outcomes, as well as being cost-effective. A convincing study would need to show that flagging a result outside a patient’s own reference interval actually corresponds to a clinically significant event.

The other hurdle that personalized reference intervals face is how the clinic will handle them. “Before laboratories implement personalized reference intervals,” Aarsand said, “clinicians need to know what they mean and how they should handle them.”

To convince clinicians, “we should inform them about the limitations of population-based reference intervals: why they may not be sufficient to interpret all individuals’ results, and why we need personalized reference intervals,” said Abdurrahman Coşkun, MD, a professor of clinical biochemistry at Acibadem University in Istanbul, Turkey, who is also currently affiliated with Knappschaft Kliniken University Hospital Bochum, at Ruhr University Bochum in Germany.  

The ongoing work is that, even if you can construct a personalized reference interval, you then have to come up with the evidence for how it can be used best alongside existing standards. “Say you have a personal benchmark for a patient, you still have to work out how best to use it for targeted decisions, and build the evidence that it will actually improve care,” Foy said.  

Thus far, these studies have been limited. As more get underway, Coşkun and Aarsand recommend that they should focus on well-understood, commonly requested tests for widespread diseases and conditions. This could include lipid profiles for cardiovascular disease, complete blood counts for hematological disorders, and glucose and HbA1c for diabetes. Because these tests are routinely and repeatedly requested in clinical practice, it’s more likely that there will be longitudinal measurement data available for a major proportion of individuals in EHRs, Coşkun and Aarsand said.  

Still, barriers remain. To acquire the amount of data needed, it will require repeated testing over a substantial time period when the patient is in a healthy state. In reality, this has limited practicality. “If I could wave a magic wand and say everyone gets a test every six months while they're healthy, with no hassle to them and no cost to the lab, and we have that data and can have this rolling benchmark, that would of course be great,” Foy said. “But it's utterly impractical.”

Although patients usually seek out a clinic visit only when they require treatment, personalized reference intervals would require patients to routinely visit the clinic when they are in a healthy or clinically stable state. Part of Foy and Higgins’ work is building methods that can isolate what is likely to be the healthy part of a patient’s prior testing, so that it can be used to create a benchmark.  

A more realistic route would be to estimate a patient’s personalized reference interval based on their longitudinal laboratory data already available in their EHR. Three repeated measurements would be necessary at the very least, Coşkun and Aarsand said, though five would be preferable.  

Another big question remains: There’s a big difference between what is deemed “normal” and what is deemed “healthy,” and that differentiation is really hard to untangle, Foy said. That means when you're trying to construct a benchmark, whether at the population or the personal level, you have to decide what that benchmark is actually supposed to be.

There are also some concerns that come along with personalized reference intervals. When taking a narrower personalized interval, if a patient’s results go outside the bounds of it, that could raise an alarm, which could be a false positive if it is in fact normal variation. “Having lots of alarms and lots of flags would probably not be helpful — it’s about finding the important signals,” Aarsand said.  

“But since there hasn’t been any widespread implementation, and there are so many reasons to expect benefits, I think we should give it a try and find out how many false positives actually occur and whether they can be mitigated if necessary,” Higgins said.  

Coşkun is optimistic that personalized reference intervals are on the way. “There’s gradually increasing interest, and I’m sure that in the near future it will be necessary, because for personalized medicine, we know the bottleneck is the population-based reference interval data,” he said. “[To get there,] we need many more papers published on personalized reference intervals, scientific activities, and maybe workshops to disseminate this information in the laboratory  
community.”  

Grace Browne is a freelance journalist who lives in London. +Email: [email protected]

Read the full November-December issue of CLN.

Advertisement

Advertisement