Use of Data Mining in the Establishment of Reference Intervals for Albumin‐Adjusted Calcium

ABSTRACT

Background

Calcium homeostasis is critical for numerous physiological functions, and while total calcium is commonly measured in clinical practice, albumin‐adjusted calcium is widely used to account for protein binding, despite concerns about its accuracy and lack of standardized reference intervals. This study aims to establish a reliable reference range for albumin‐adjusted calcium using real‐world data from a high‐volume tertiary care laboratory.

Methods

This study analyzed data from 33,159 individuals aged 18–80 years, selected from an initial population of 106,920 patients based on specific biochemical and clinical exclusion criteria. Statistical analysis involved outlier detection, partitioning, and calculation of age‐specific reference intervals using the 2.5th and 97.5th percentiles.

Results

The study established age‐stratified reference intervals for albumin‐adjusted calcium (18–49 years: 2.17–2.55 mmol/L; 50–79 years: 2.20–2.61 mmol/L) and total calcium (18–49 years: 2.17–2.56 mmol/L; 50–79 years: 2.17–2.62 mmol/L).

Conclusions

Our study reveals that albumin‐adjusted calcium reference intervals are age‐dependent, with slightly higher upper limits (≈0.05 mmol/L) than traditionally used total calcium ranges, potentially reducing overdiagnosis of hypercalcemia. Further studies are needed to validate these reference intervals across different automated analyzers and diverse patient populations to ensure broader clinical applicability.

1. Introduction

Calcium is the most abundant mineral that plays vital roles in cellular signaling, bone structure, blood clotting, and neuromuscular activity [1]. Its levels are tightly regulated, and disruptions in calcium metabolism can lead to significant health complications, including increased morbidity and mortality [2, 3].

Calcium levels can be quantified either as the ionized (free) form, which is the biologically active form, or as total calcium, encompassing both free and protein‐bound calcium. Approximately 40%–50% of the circulating calcium is bound to negatively charged proteins, primarily albumin, with a smaller fraction bound to other proteins such as immunoglobulins [4]. Since calcium binds to proteins, mainly albumin, lab results commonly include an albumin‐adjusted calcium value to account for this.

Although ionized calcium directly reflects the physiologically active form, total calcium remains the standard measurement in clinical practice due to its cost‐effectiveness, less strict preanalytical requirements, technical simplicity, and widespread availability. However, growing evidence suggests that relying solely on total calcium may lead to diagnostic inaccuracies, as fluctuations in blood pH and serum protein levels can significantly influence calcium‐protein binding [1], thereby disrupting the expected correlation between total and ionized calcium concentrations.

In clinical practice, albumin‐adjusted total calcium has long been used as a substitute for ionized calcium [5]. The first formula was developed by Payne et al. in 1973 [5]. However, due to its limitations, particularly in critically ill patients [6, 7], those with renal failure [8, 9], and the geriatric population [10], numerous equations have been proposed over the years to overcome its shortcomings [11, 12, 13, 14, 15, 16]. Despite ongoing efforts to improve adjustment formulas, many of these studies suggest that unadjusted total calcium correlates more closely with ionized calcium, yet albumin‐adjusted calculations remain widely used in clinical practice despite concerns about their diagnostic accuracy. The National Institute for Health and Care Excellence (NICE) guidelines recommend utilizing albumin‐adjusted calcium for the assessment of patients with calcium‐related disorders [17]. The most widely used formula for adjusted calcium in clinical practice, including in Turkiye, is: albumin‐adjusted calcium (mmol/L) = total calcium (mmol/L) + [0.02 × (40 − albumin (g/L))] [11]. Nonetheless, it is advisable to employ a derived regression equation, as it reflects the characteristics of the local patient population and the analytical method used in the laboratory [18].

Although clinicians commonly rely on albumin‐adjusted calcium, the reference intervals used in practice—including those in national guidelines—are typically based on total calcium. An accurate reference interval is crucial for distinguishing hypercalcemia and hypocalcemia from normal calcium levels in clinical practice. In our laboratory, the reference range for total calcium is 2.15–2.50 mmol/L for adults aged 18–60 years and 2.20–2.55 mmol/L for adults aged 60–90 years, while no specific reference interval is provided for albumin‐adjusted calcium. Establishing a reliable reference range requires the inclusion of parameters such as parathyroid hormone (PTH), 25‐OH Vitamin D, estimated glomerular filtration rate (eGFR), alkaline phosphatase (ALP), alanine transaminase (ALT), potassium and albumin to ensure equilibrium calcium values, along with data from a broad patient population. This study aims to establish a reference range for albumin‐adjusted calcium using real‐world data from a high‐volume tertiary care central laboratory serving a large and diverse patient population.

2. Materials and Methods

2.1. Study Population

The study included 106,920 patients who visited Istanbul Central Laboratory‐2 (ISLAB‐2), one of the nation's largest laboratory, between January 2021 and January 2025, where over 5,000 total calcium tests are processed daily, serving 14 hospitals and 178 primary care clinics, with concurrent Total Calcium, Albumin, Creatinine, eGFR, PTH, and 25‐OH Vitamin D results. Additionally, among these patients, those with concurrent ALT, ALP, and serum potassium results were included in the analysis, and these laboratory findings were also evaluated, where possible. The proposed exclusion criteria were applied based on laboratory information system data, following a method similar to that described by Barth et al. [18] and recommended by the Association for Clinical Biochemistry and Laboratory Medicine position paper [19].

As provided in Figure 1, patients were excluded from the study if they met any of the following criteria: age under 18 or over 80 years, eGFR < 60 mL/min/1.73 m², PTH levels < 1.6 or > 6.9 pmol/L, 25‐OH vitamin D < 50 nmol/L, albumin < 20 g/L or ≥ 55 g/L, ALP > 150 U/L, ALT > 50 U/L, serum potassium < 3.5 mmol/L, or if they were referred from nephrology, general surgery, critical care, endocrinology, or hematology‐oncology departments and 33,159 patients remained.

Figure 1.

Flow diagram for Inclusion and Exclusion Criteria. Note that several exclusion criteria may apply to a single sample. PTH, parathyroid hormone; ALP, alkaline phosphatase; ALT, alanine aminotransferase; eGFR, estimated Glomerular Filtration Rate.

This study was conducted in accordance with the principles of the Helsinki Declaration and received ethical approval from the Istanbul Medeniyet University Non‐Interventional Research Ethics Committee (GOSEK‐0369).

2.2. Laboratory Analysis

In our study, serum calcium, albumin, ALT, ALP, potassium, creatinine, PTH, and 25‐OH vitamin D were measured using a Cobas 8000 (c702, e801) modular system (Roche Diagnostics, Mannheim, Germany) in strict accordance with the manufacturer's protocols. Albumin was analyzed using the Bromocresol Green (BCG) method, with a reference range of 39.7–49.4 g/L for adults and a coefficient of variation (CV) below 2%. Calcium concentration was measured photometrically through a sequential reaction where Ca²⁺ first binds to NM‐BAPTA at alkaline pH, forming a calcium‐NM‐BAPTA complex, which then reacts with Ethylenediaminetetraacetic acid (EDTA) to release NM‐BAPTA while forming a calcium‐EDTA complex, with the resulting absorbance change being directly proportional to the calcium concentration and demonstrating a CV below 2.5%. According to Clinical Laboratory Improvement Amendments (CLIA) criteria, the total allowable error limits are ±8% for albumin and ±1.0 mg/dL for total calcium [20], and the analytical performance for both parameters in our study remained within these limits. All tests were evaluated using external quality assessment schemes, including RIQAS (Randox International Quality Assessment Scheme, UK) from 2021 to 2024 and the Association of Clinical Biochemistry Experts External Quality Control Program (KBUDEK, Turkey) in 2024, and demonstrated acceptable performance throughout the study period.

2.3. Statistical Analysis

Data normality was evaluated through both visual inspection (Quantile‐Quantile plot, histograms) and the Shapiro‐Wilk test, while Levene's test was employed to examine the homogeneity of variances. Numerical data were expressed both as mean ± standard deviation (SD) and as median with interquartile ranges [Q1–Q3]. Categorical variables were presented as frequencies (n) and proportions (%). Potential outliers were identified using Horn's method with Tukey's interquartile fences (k = 1.5), which classifies a value as an outlier if it falls more than 1.5 times the interquartile range (IQR) above or below the quartiles. Gender and age partitioning was performed using the Harris‐Boyd's z‐test [21] or Lahti's method [22] and separate or merged sex and age‐specific upper and lower reference limits were established, where appropriate. The 2.5th and 97.5th percentile values with 90% confidence intervals (CIs) were calculated for albumin‐adjusted calcium for age‐ and sex‐specific groups. All statistical analysis was performed using Microsoft Excel for Microsoft 365 (Microsoft, Redmond, USA) and MedCalc version 19.2.1 (MedCalc Software Ltd, Ostend, Belgium). A p‐value less than 0.05 was considered statistically significant.

3. Results

3.1. Baseline Characteristics

A total of 33,159 participants who met the eligibility criteria were included in the study, with a mean age of 54.6 ± 13.7 years and 75.88% (n = 25,164) being female. The laboratory findings revealed a mean albumin level of 45.4 ± 2.96 g/L and a mean total calcium level of 2.38 ± 0.11 mmol/L. The demographic and laboratory characteristics of the study population are summarized in Table 1.

Table 1.

The demographic characteristics of the participants and laboratory test results.

Characteristics	Statistics
All cohort	n = 33,159
Demographical
Age (years), mean ± SD (range)	54.6 ± 13.7 (46–65 years)
Sex (female/male), n (%)	25,164 (75.88)/7,995 (24.11)
Laboratory results
Albumin (g/L), mean ± SD; median (IQR)	45.4 ± 2.96; 45.5 (43.5–47.4)
Total calcium (mmol/L), mean ± SD; median (IQR)	2.38 ± 0.11; 2.39 (2.32–2.44)
PTH (pmol/L), mean ± SD; median (IQR)	3.77 ± 1.22; 3.89 (3.61–4.6)
Creatinine (μmol/L), mean ± SD; median (IQR)	66.32 ± 13.2; 65.4 (57.4–75.1)
25‐OH‐Vitamin D (nmol/L), mean ± SD; median (IQR)	82.82 ± 30.05; 75.25 (62.2–94.7)
Alkaline phosphatase (U/L), mean ± SD; median (IQR)	74.1 ± 20; 72 (59.2–85)
Serum potassium (mmol/L), mean ± SD; median (IQR)	4.4 ± 0.32; 4.35 (3.9–4.9)
Alanine aminotransferase (U/L), mean ± SD; median (IQR)	18.9 ± 8.2; 16.1 (13–22.4)

3.2. Albumin‐Adjusted Calcium Equation

The albumin‐adjusted calcium values were calculated using a population‐based regression approach as described [18]. First, linear regression analysis determined the relationship between total calcium and albumin levels, along with the calcium intercept representing non‐protein‐bound calcium. Albumin‐adjusted calcium = total calcium ‐ (slope * albumin) + (mean normal total calcium ‐ intercept calcium) [18]. As shown in Figure 2, after applying the exclusion criteria, a total of 33,159 patients were included in the study, and a linear regression analysis was performed. The intercept was found to be 1.7912, while the slope of the graph was 0.013. Based on this, the following equation was derived:

$$\text{Albumin}‐\text{adjusted}\text{calcium}=\text{Total}\text{calcium}+0.013\times (45.6–\text{albumin}).$$

Figure 2.

Linear regression relationship between total calcium (y‐axis) and albumin (x‐axis) (n = 33,159). The y‐intercept was found to be 1.7912 while the slope of the graph was 0.013.

3.3. Reference Interval for Total Calcium and Albumin‐Adjusted Calcium

Table 2 shows the established total calcium reference interval: 2.17–2.56 mmol/L for individuals aged 18–49 years and 2.17–2.62 mmol/L for those aged 50–79 years.

Table 2.

Estimated age‐specific reference intervals for total calcium (mmol/L).

	Groups	Percentiles
		2.5th (90% CI)	97.5th (90% CI)
Total calcium (mmol/L)	18–49 years (n = 10,552)	2.17 (2.14–2.18)	2.56 (2.56–2.56)
	50–79 years (n = 22,607)	2.17 (2.17–2.19)	2.62 (2.61–2.62)

Using the equation derived above, the albumin‐adjusted calcium was calculated for the entire study population (n = 31,159) who had concomitant calcium, albumin, PTH, 25‐OH‐Vitamin D, and creatinine measurements available, after excluding individuals with predefined laboratory values outside the specified ranges and those followed in certain clinical departments, as defined above. As shown in Table 3, the reference interval for albumin‐adjusted calcium was determined to be 2.19–2.60 mmol/L for the entire study population. Age‐specific reference intervals for albumin‐adjusted calcium were also established: 2.17–2.55 mmol/L for individuals aged 18–49 years, 2.20–2.61 mmol/L for individuals aged 50–79.

Table 3.

Estimated age‐ and sex‐specific reference intervals for albumin‐adjusted calcium (mmol/L).

	Groups	Percentiles
		2.5th (90% CI)	97.5th (90% CI)
Albumin‐adjusted calcium (mmol/L)	All group (n = 33,159)	2.19 (2.19–2.21)	2.60 (2.59–2.6)
	All Females (n = 25,164)	2.19 (2.19–2.19)	2.60 (2.6–2.6)
	All Males (n = 7,995)	2.2 (2.19–2.2)	2.58 (2.57–2.58)
	18‐49 years (n = 10,552)	2.17 (2.17–2.18)	2.55 (2.55–2.56)
	50‐79 years (n = 22,607)	2.2 (2.19–2.2)	2.61 (2.60–2.61)

In Table 4, although there is no specifically established reference range for albumin‐adjusted calcium, the current reference ranges used for adjusted calcium evaluation (2.15–2.50 mmol/L for individuals aged 18–60 years and 2.20–2.55 mmol/L for those aged 60–90 years) were compared with the reference ranges determined in our study, and a total agreement of 92.9% was observed.

Table 4.

Classification of adjusted‐calcium status by the proposed age‐specific Reference Intervals and the current Reference Intervals (2.15–2.50 mmol/L for 18–60 years and 2.20–2.55 mmol/L for 60–90 years). NA, not applicable.

	Classification by the proposed reference interval, No. (% of total)
	Hypocalcemia	Normocalcemia	Hypercalcemia	Total
Classification of the current reference interval
Hypocalcemia	573 (1.73)	0	0	573 (1.73)
Normocalcemia	260 (0.78)	29,406 (88.68)	0	29,666 (89.46)
Hypercalcemia	0	2,090 (6.3)	830 (2.5)	2,920 (8.8)
Total	833 (2.51)	31,496 (94.98)	830 (2.5)	33,159 (100)
Observed agreement	NA	NA	NA	30,809 (92.91)

4. Discussion

In our study, we established the reference intervals for albumin‐adjusted calcium in an 18–80‐year‐old Turkish population on the Roche Cobas platform by applying stringent exclusion criteria to a high‐volume central laboratory data set. Despite the decades‐long clinical use of albumin‐adjusted calcium in diagnosing disorders of calcium metabolism, no reference interval had been defined for the Cobas analyzer or our population before this study. Accurate determination of the reference interval is critical, as even minor shifts in its upper and lower limits can significantly alter the classification of calcium metabolism disorders. A narrower range, for instance, may lead to an increased diagnosis of primary hyperparathyroidism, while a broader range could result in overlooking patients with calcium homeostasis disorders. To our knowledge, only one previous study exists in the literature, which was conducted using Beckman Coulter assays in a UK population; however, that study lacked critical exclusion criteria, such as PTH data, potentially affecting the robustness of their reference range [23]. Our findings thus address a significant gap in the endocrinology and clinical biochemistry of calcium assessment.

Reference intervals play a crucial role in distinguishing healthy individuals from those with disease and in monitoring clinical changes over time and accurate and population‐adjusted reference intervals are essential for the proper interpretation of laboratory results [24]. Since reference intervals are not universally harmonized, laboratories are strongly encouraged to establish their own intervals based on the characteristics of their specific patient population, in line with the International Federation of Clinical Chemistry (IFCC) and Clinical Laboratory Standards Institute (CLSI) guidelines [25, 26]. The challenges in establishing self‐produced reference intervals include difficulties in selecting appropriate reference individuals, which often require extensive health questionnaires, medical record reviews, physical examinations, and additional lab tests. Due to these limitations and advancements in IT tools, laboratories are increasingly turning to indirect methods by utilizing data from Laboratory Information Systems (LIS) to identify healthy individuals and establish personalized reference intervals [24, 27].

In our study, albumin‐adjusted calcium levels varied by age, with higher upper reference limits observed in adults aged 50–79 years. While clinicians often apply the total calcium reference intervals (2.15–2.50 mmol/L for adults aged 18–60 years and 2.20–2.55 mmol/L for those aged 60–90 years) to albumin‐adjusted calcium due to the lack of specific reference ranges, our findings suggest slightly higher upper reference limits, approximately 0.05 mmol/L, which could influence the count of individuals classified as having a calcium metabolism disorder. Moreover,the reference interval for total calcium derived from our cohort revealed an upper limit that exceeded the commonly accepted upper reference limit currently in use. Consequently, many patients who were actually normocalcemic according to our newly proposed reference range might have been diagnosed with hypercalcemia based on the previous criteria. Schini et al. [23] highlighted that the albumin‐adjusted calcium reference range differs by sex and age, with postmenopausal women exhibiting a higher upper reference limit, which may be attributed to the inability to exclude hyperparathyroidism due to missing PTH data in their study, suggesting a potential link to a higher prevalence of primary hyperparathyroidism in this group [28].

Since albumin‐adjusted calcium is calculated based on total calcium and albumin levels, any variability of the total testing process in these two parameters can significantly affect the result. Methodological differences in calcium and albumin assays have long been acknowledged, and emerging evidence suggests that physiological variations in serum albumin—affected by age and gender—further complicate interpretation [29, 30]. Jassam et al. demonstrated that calcium reference intervals from most manufacturers were broadly comparable and potentially harmonizable, except for the Siemens Advia systems, which showed lower reference limits [30]. Variations between BCG and Bromocresol Purple (BCP) methods for albumin measurement have been widely reported, with BCG yielding higher results than BCP across different analytical platforms [30, 31, 32, 33], as reflected in their differing reference intervals [34]. In our study, only the BCG method was used for albumin measurement throughout the study period in our laboratory. While this minimized inter‐method variability, our results primarily reflect the analytical platform and method used, and should be validated with alternative methods before applying harmonized reference intervals.

There remains an ongoing debate concerning the optimal approach to measuring calcium, specifically whether to use total calcium, albumin‐adjusted calcium, or ionized calcium. Although ionized calcium measurement is standard practice in conditions such as primary hyperparathyroidism [35], malignancy‐related hypercalcemia [36], neonatal hypocalcemia [37], and critical illness [7], current guidelines for mineral and bone metabolism disorders do not routinely recommend its use [17, 38, 39]. Ionized calcium is typically measured using blood gas analyzers; however, its use is further limited by the need for strict pre‐analytical control and analytical challenges such as frequent electrolyte errors, the requirement for experienced staff, and issues like clotting.

Our study has several limitations. First, we lacked data on medication use that could influence calcium metabolism, such as vitamin D supplements, which could potentially affect the accuracy of our reference intervals. Although patients were instructed to provide fasting morning blood samples, we could not verify compliance with fasting requirements, introducing a possible preanalytical confounding factor. Since all laboratory parameters were measured using a single analytical platform, our findings are specifically applicable to the Roche Cobas system and the methodology employed. Additionally, albumin‐adjusted calcium is a calculated value derived from total calcium and albumin, and the measurement of both analytes can exhibit significant variability across different methods and platforms. Although the male population in each group was sufficient, the sex imbalance (approximately 75% female predominance) in our study cohort may have introduced potential bias in the results. Lastly, the shortcomings of albumin‐adjusted calcium (e.g., in renal failure, hypoalbuminemia, and critically ill patients) should not be overlooked, and its interpretation should be approached with caution in these populations.

In conclusion, using a large central laboratory data set on the Roche Cobas platform, we established reference intervals for albumin‐adjusted calcium in adults and identified differences based on age. Additionally, we derived a specific equation for calculating albumin‐adjusted calcium tailored to our laboratory population and analytical platform. These findings provide a critical foundation for the accurate diagnosis of calcium metabolism disorders in settings utilizing similar analytical methodologies.