Historically, clinical laboratories have referred to reference intervals (RI) as normal ranges and developed them using samples from a “normal or healthy” population. This approach to RI quickly encounters a stumbling block, however. Namely: How do we define who is normal?
In the modern age, we would define normal as biochemically normal, but this is a weak definition, as it uses circular logic.
Recognizing this shortcoming, the laboratory community adopted the term RI in place of normal range, and in 1987 the International Federation of Clinical Chemistry and Laboratory Medicine (IFCC) released a detailed definition for RI, which states that an RI is “the interval between and including the reference limit” created from a reference population (1). This more inclusive definition allows for the determination of an RI in a reference population composed of individuals who are not healthy, such as those with chronic conditions that seriously affect the RI, or patients who have a chronic elevation that could obscure an acute event. Since the shift from normal ranges to RI occurred, different types of RI—known as continuous, common, and personalized—have also emerged, further fine-tuning the concept of RI as well as labs’ ability to identify pathological test results.
RI take into consideration patient variables that clinicians are aware of, such as age and sex, as well as laboratory variables that clinicians are unlikely to know, such as which vendor and assay is used in a particular lab. RI play a critical role in patient care by enabling clinicians to interpret laboratory data accurately and to link that data to clinical action. Without the accompanying RI, a lab result is just a number. It is therefore essential for clinical laboratories to establish RI that are accurate and appropriate for their local patient populations.
To facilitate this, several professional organizations have published recommendations for performing RI studies, including the Clinical and Laboratory Standards Institute (CLSI) and IFCC. According to CLSI, RI can be determined through two processes: direct methods, which use patient samples and are considered the preferred approach, and indirect methods, which use data from the laboratory informatics system (LIS) or the electronic health record (EHR) (2).
Direct Reference Intervals
Current guidance recommends developing RI that are both specific to the population that a clinical laboratory serves and to the assay that the lab is using. To do this with the direct method, labs should use samples from a “healthy” patient population. These samples can come from patients who were prospectively recruited, or they can be banked and/or residual patient samples. Labs can confirm the health status of RI study participants with direct questionnaires in the case of prospective recruitment and/or through chart review. The goal should be to collect a minimum of 120 data points for each partition, which is necessary for determining the 90% confidence interval around an RI (2). Partitions may include sex, age, Tanner stage, and menstrual cycle day, to name a few.
There are three common roadblocks to determining direct RI that labs typically encounter: Not being able to recruit enough people to fill each partition, the recruitment population not reflecting the population served by the clinical laboratory, and preanalytical factors failing to recapitulate those found in standard operations (2, 3).
The first and second issues tend to be related to advertisement. Often the clinical laboratory advertises for RI study recruits within the confines of the hospital. This limits the possible study population to those who are on campus, however, and those on campus have a high probability of not fitting the inclusion criteria. They also might not reflect the whole population serviced by the clinical laboratory, as the lab might receive a large proportion of samples from ancillary locations in outlying areas. In fact, in my experience with performing RI recruitment, most responders tend to be related to the laboratory, which is a population that is usually very different from the service population in age, sex, and education levels. For pediatric populations, there is an additional barrier to recruitment in that the lab is required to get the consent of both the participating minor and their guardian, while for adults, only the consent of the individual participant is needed.
Since recruitment is so difficult, the clinical laboratory often wants to draw various blood tube types during the one encounter with a study participant. The personnel used for this encounter tend to be well-trained in blood draws and the correct order of blood tube draw. In addition, the samples are processed in the most ideal conditions (e.g., at an optimal temperature, time, etc.). In my experience, this creates the third major roadblock mentioned above and can lead to a deviation between the measured RI based on the collected population and routine samples.
As an example, the samples used for our lab’s chromium RI determination were collected as the last tube during a seven-tube collection, allowing the chromium present on stainless-steel needles to wash out adequately. This yielded a lower RI than was experienced when a single tube was collected for routine chromium determination.
With these inherent disadvantages, why is this the preferred method for establishing RI? The major advantage of this method is the low probability of pathological samples being introduced into the RI data, which in turn means that the statistics required to calculate the RI are relatively straightforward and simple to perform.
Indirect Reference Intervals
By using data from the LIS or EHR, laboratories performing indirect RI determinations can deal with many of the disadvantages of the direct method. Recruitment is simpler with the indirect method because the lab is starting with all available data. In fact, this method can make data available that would otherwise be uncollectable if an additional blood collection were required, especially in the pediatric population. Sample data inherently reflects the laboratory service population, and preanalytical factors represent standard operation. In addition, partition groups can be easier to fill, with some partitions having thousands of data points, though others might still be more limited in scope.
However, the indirect method does come with its own set of drawbacks. A major issue is that data sets from the LIS and EHR include intermixed pathological and normal results. To correct this, indirect methods require robust statistical methods to separate normal results from the pathological. Furthermore, with the indirect method, the time and effort used for sample collection with the direct method is shifted toward elucidating the appropriate exclusion criteria. The exclusion criteria need to incorporate clinical as well as institutional knowledge with considerations that include the source of data, how the assay is used, the instrument, and assay changes. Since the RI should be specific to the assay the lab uses, care should be taken to exclude data from previous assays.
The lab must also decide how much of the data to use—for instance, will the lab use all the data available or out patient data only? Inpatient data typically contains more pathological data than outpatient data, but specialty clinics and practices need to be considered, too. Another factor to consider when deciding what data and which system to use is the type of partition. For example, the LIS is typically not going to contain data on Tanner stages or menstrual cycle day. The EHR is more likely to house these data, but they are often not in discrete fields and will require more sophisticated techniques to extract.
One final consideration is that if an assay is used as a confirmatory test, then it probably isn’t a good target for the indirect method because there is a high suspicion of a pathological process. This will reduce the healthy population signal and cause distortions even in robust statistical algorithms.
Even with all these caveats, the indirect method is useful for determining RI for many analytes in the clinical laboratory. A more detailed comparison of the pros and cons of the direct and indirect methods is available in a Clinical Chemistry and Laboratory Medicine review by Jones et al. (4).
Statistics for Reference Intervals
CLSI’s guidance has recommendations on the statistical methods labs should use in RI studies. These recommendations are mainly targeted toward data collected via direct rather than indirect methods, but CLSI also touches on the robust statistical methodologies recommended for indirect methods (3, 4). For direct methods, parametric or transformed parametric statistical methods are recommended, with the caveat that labs should limit their use to analytes that demonstrate a Gaussian probability distribution. The preferred statistical method for determining the 2.5 and 97.5 percentiles of an RI is a simple nonparametric regression.
As for robust statistical methods, IFCC recommends that more research be done on these, and provides guidance on the data that should be collected for a complete research analysis (4). These robust statistical methods can be achieved through consultation of biostatisticians, specialized statistical software, and/or statistical programming languages such as R (5, 6).
There has been a surge in articles on indirect RI over the past 3 years, and these publications often supply the program code or application for the statistical method they used. Statistical methods for indirect RI are also frequently available in online code repositories like Github or on institutional websites. Thanks to this open-source code, laboratories across the world can determine RI using appropriate statistical methods with relative ease compared to previous years. Additionally, the availability of code for different statistical methods enables laboratorians with limited programming experience to compare them.
Labs should strive to gain a thorough understanding of any statistical method they use and of the assumption that method employs. Consultation with a statistician can also be useful when working with a new statistical method for the first time.
The Future of RI
The three biggest concepts that are shaking up the RI world today are continuous, common, and personalized RI. Continuous RI do away with partitions and represent an incremental change in how RI have been determined historically. Currently, partitions create an artificial break in RI. If an analyte has a partition from 8–9 years old and from 10–11 years old, then the day a child turns 10, they move outside the RI for 8–9-year-olds even though the child hasn’t actually changed much. Continuous RI adjust for this by using an equation that automatically accounts for the true age of the child. The biggest hurdle to implementing this type of RI is in the systems used for storing the data, the LIS and EHR. These systems only accept fixed values for RI rather than the dynamic intervals that would be required with continuous RI.
The next big concept shift is common RI. Common RI require three conditions to be met. First, there must be a reference method for the analyte in question. Secondly, assays must be traceable to the reference method. Finally, a multicenter RI study should be conducted with a common protocol. IFCC has performed studies such as this on a global scale for certain analytes with acceptable results (7).
The last and possibly most interesting new concept in the world of RI is the personalized RI. With standard RI, it’s possible for a patient’s analyte to traverse a large interval without being considered abnormal and getting flagged within the EHR. Personalized RI are therefore needed because there is sometimes less inherent variation in analyte concentration within individuals than in the general population from which standard RI are derived. Personalized RI create a smaller reference range for the individual than the population, which allows the EHR to flag and inform clinicians about subtle changes that may indicate a move toward a pathological state at an earlier stage of disease development (8).
With the advent of new RI concepts and larger datasets, the clinical laboratory community is moving into a new age for RI. Large multicenter RI studies are now available due to advances in computing power, greater adoption of EHRs, and newer robust statistical methods that can adapt to variations in datasets. As RI continue to evolve, however, their central tenet of supplying vital information to clinicians will not change.
Dustin Bunch, PhD, DABCC, is the assistant director of clinical chemistry and laboratory informatics at Nationwide Children's Hospital in Columbus, Ohio. +Email: [email protected]
- Solberg HE. International Federation of Clinical Chemistry (IFCC), Scientific Committee, Clinical Section, Expert Panel on Theory of Reference Values, and International Committee for Standardization in Haematology (ICSH), Standing Committee on Reference Values. Approved recommendation (1986) on the theory of reference values. Part 1. The concept of reference values. J Clin Chem Clin Biochem 1987;25:337-42.
- Defining, establishing, and verifying reference intervals in the clinical laboratory; Third edition [C28-A3]. Wayne, Pa.: Clinical and Laboratory Standards Institute 2008.
- Özcürümez MK, Haeckel R, Gurr E, Streichert T, Sack U. Determination and verification of reference interval limits in clinical chemistry. Recommendations for laboratories on behalf of the working group guide limits of the dgkl with respect to iso standard 15189 and the guideline of the German Medical Association on quality assurance in medical laboratory examinations (rili-baek). J Lab Med 2019;43:127-133.
- Jones GRD, Haeckel R, Loh TP, et al. Indirect methods for reference interval determination - Review and recommendations. Clin Chem Lab Med 2018;57:20-9.
- Team RC. R: A language and environment for statistical computing. https://www.R-project.org/ (Accessed May 2021).
- Haymond S and Master S. Why clinical laboratorians should embrace the R programming language: A case for learning R as a gateway to laboratory medicine’s digital future. Clinical Laboratory News 2020;46:12-7.
- Ichihara K, Ozarda Y, Barth JH, et al. A global multicenter study on reference values: 1. Assessment of methods for derivation and comparison of reference intervals. Clin Chim Acta 2017;467:70-82.
- Cos¸kun A, Sandberg S, Unsal I, et al. Personalized reference intervals in laboratory medicine: A new model based on within-subject biological variation. Clin Chem 2020;67:374-384.