Reference intervals are an integral part of any functioning clinical laboratory. Reference intervals are often the endpoint used by clinicians in the management of patients and in clinical decision-making. It has been estimated that 80% of medical decisions are made from laboratory results.¹ The value of the results produced by a laboratory is largely influenced by the quality of the reference interval used in the interpretation of that result.²

An example of where reference intervals are of diagnostic utility is in the diagnosis of the beta thalassaemia trait. The increase in haemoglobin A₂ (HbA₂) level is the most important laboratory parameter for the identification of carriers of beta thalassaemia, and is considered diagnostic in the appropriate clinical context.³ Accurate quantification of HbA₂ in the haematology laboratory is therefore essential to allow for routine diagnosis of the beta thalassaemia trait.⁴ This highlights the value of reliable reference intervals.

Beta thalassaemia is inherited in an autosomal recessive pattern and the beta thalassaemia trait is estimated to have a prevalence of 1.5% worldwide, affecting approximately 80–90 million people.⁵ Africa has a considerable disease burden in terms of haemoglobinopathies and in particular beta thalassaemia with 1520 conceptions affected annually, Western Africa accounting for the majority of cases.⁶

Reference intervals are established through a validation process with a statistically adequate number, ideally 120, of healthy reference individuals.⁷ This needs to be done for all reagents and instrument combinations. However, even the Clinical and Laboratory Standards Institute guidelines recognise that this is not feasible for many laboratories and finding a cohort of 120 healthy individuals is not feasible for every test.⁸ Alternatively, reference intervals may be verified, with only 20 samples needed, or transferred provided that the analytic system and the test population are comparable.⁸ However, verification and transfer of reference intervals is not ideal and these methods have their own disadvantages.⁸

Every laboratory that performs HbA₂ testing is responsible for establishing its own reference interval. This is done by quantifying the HbA₂ percentage in a cohort of healthy adults who do not have iron deficiency or the thalassaemia trait.³ The reference interval is a range that should be calculated including individuals with characteristics that are comparable to the reference group so that the reference interval can be correctly applied to the population serviced by the laboratory.⁷ Given the importance of HbA₂ reference ranges, considerable work has been done internationally on normal reference intervals for HbA₂. However, there is a paucity of literature from Africa with no published reference intervals for HbA_2.

In keeping with good laboratory practice, a need was identified to determine HbA₂ reference intervals in a local African population for the Department of Haematology of the National Health Laboratory Service (NHLS), Tshwane Academic Division (TAD), Pretoria, South Africa. Using a sufficient number of results from medical records would provide a healthy cohort for establishing a reference interval without the need to recruit healthy individuals. This is particularly useful for a test that is not routinely requested and only offered by specialised laboratories. In order to establish a reference interval for HbA₂ without the limitations inherent to traditional methods of establishing reference intervals, we made use of previously reported normal high-performance liquid chromatography (HPLC) results from the NHLS Department of Haematology.

The mean age of the 144 patients included in this data analysis was 40 years (range 3–84 years). The study population comprised 67 female and 77 male patients. After a single outlier of 4.3% was excluded from the analysis, the mean HbA₂ value was 2.95% with the range between 2.2% and 3.9%. The standard deviation was 0.357% (Figure 2).

Data were normally distributed as indicated in the Kernel density estimation (Figure 3). The HbA₂ reference interval established from this data set was 2.3% – 3.6%.

A sex comparison was performed by t-test to compare the mean HbA₂ of male patients to that of female patients in an attempt to identify possible bias. No significant difference was found (p = 0.1328). The correlation between age and HbA₂ was also assessed. A trend towards lower HbA₂ values with increased age was appreciated (Figure 4), with a Pearson correlation coefficient of –0.2826.

The HbA₂ reference interval established from this data set was 2.3% – 3.6%. Normal reference intervals for HbA₂ have been published for other patient populations. A study performed at the Leiden University in the Netherlands, using the Variant Classic HPLC (Bio-Rad^®) platform, reported a HbA₂ reference interval of 2.3% – 3.5%.³ Despite the difference in study population, the reference interval determined in this current study is comparable to that of the Leiden group which used a similar method, that is, HPLC (Bio-Rad^®) technology. Han et al. reported a reference interval of 2.3% – 3.1% for HbA₂ in a Chinese population of reproductive age.⁹ However, these investigators used a CAPILLARYS2 instrument (Sebia, France) to generate their data.⁹

Most evidence suggests that HbA₂ of > 4% is indicative of beta thalassaemia trait with almost 100% sensitivity and 90% specificity.¹⁰ A grey zone between 3.1 and 3.9 is generally accepted as reported in a comprehensive review.¹¹ This often poses a diagnostic challenge. Studies have been conducted to identify the presence of mutations in these individuals.^{10,11,12,13,14} Giambona et al. found that 80% of patients in this group in an Italian population were negative for molecular defects, and the most significant finding was the presence of beta thalassaemia gene mutations found mostly in patients with HbA₂ in the region of 3.5–3.9 and mean cell volume < 80 fL.¹¹

The upper limit of the normal reference interval established in this current study does fall within the previously described ‘grey zone’. However, in the presence of a normal CBC, the possibility of an underlying carrier state in these patients remains small. This highlights the importance of interpreting HbA₂ within the clinical context, taking into consideration the CBC parameters and iron studies.

There appears to be a weak association between a decreasing HbA₂ value and increasing age; this was true even when looking at subsets of age and when excluding patients aged 70 years or older. Although this trend was seen, when calculating the reference interval by age, the reference interval remained 2.3–3.6 when rounded to one decimal place. Therefore, this was not a significant finding.

In this current study we used data available on the laboratory information system in order to establish a reference interval. This represents a novel approach in our setting. Data mining is emerging as an alternative to the traditional direct a priori method. Data mining makes use of electronic data records and statistical techniques to determine the healthy population within a data set in order to establish a reference interval.¹⁵ The electronic data records may be obtained from insurance claims, electronic health records as well as a variety of other sources.¹⁶ All sources require data capturing platforms which allow for database management in order to deal with the enormous volume of information currently being generated as well as the complexity of analysing and interpreting the data.

The analytical methods that have been employed to establish reference intervals include The Hoffmann method, Bhatacharya method and more recently the Truncated Maximum Likelihood method.¹⁵ The Truncated Maximum Likelihood method employs complex statistical algorithms that make use of maximum likelihood estimation and require 4000 data points in order to establish a robust reference interval.¹⁵ Currently there is still hesitancy regarding the use of indirect methods but it is likely to be used to establish many reference intervals in the future. Indirect methods may be particularly useful for tests that are not routinely performed as screening tests in the healthy population.¹⁷ Data mining has many advantages: it is less costly as the blood results of large cohorts of individuals are readily available, it is faster, and it can even be considered more ethical.¹⁵ A study conducted by Katayev et al. showed that reference intervals could be reliably and reproducibly established using data mining. Reference intervals were calculated for eight analytes and were found to be comparable to already accepted published reference intervals.¹⁷ Although our study was small and the data were captured manually, it does highlight the potential of using laboratory information records to glean valuable data with relatively low cost and fewer limitations than are inherent to establishing reference intervals in the conventional manner by direct population sampling.