Comparative evaluation of two automated immunoassays for prolactin using patient testing data

Introduction

Prolactin is secreted by the anterior pituitary gland and controlled by the hypothalamus, which produces dopamine that inhibits prolactin secretion, while thyrotropin-releasing hormone, pregnancy, breastfeeding, or stress can stimulate prolactin secretion. Excessive amount of prolactin can be secreted by pituitary adenomas or when the pituitary stalk or hypothalamus is pressed by intracranial tumors, causing hyperprolactinemia when blood prolactin level increased above the upper reference limit (URL). Hyperprolactinemia can manifest as ovulatory disorders, menstrual irregularities, galactorrhea, and infertility in women, and hypogonadism, decreased libido, impotence, erectile dysfunction, decreased sperm production, infertility, gynecomastia in men (1,2). Further, prolonged hyperprolactinemia in both sexes can also lead to reduced bone density due to hypoestrogenism or hypoandrogenism. Determination of blood prolactin level is essential for the diagnosis and management of these conditions associated with hyperprolactinemia (1-3). In clinical laboratories, prolactin is usually measured with immunoassays on high-throughput, automated immunochemistry instruments in consideration of cost and operations. The challenge of using immunoassays for prolactin is in the difficulties to standardize assays that results by different manufacturers do not always agree, causing confusions with clinicians. Table 1 gives three such cases, where the values of prolactin measured with the Roche Cobas e801 (herein Cobas) analyzer in our laboratory were elevated, that could not be confirmed by the method on Siemens Centaur (herein Centaur) at a reference laboratory. Macroprolactin was ruled out using a method employing polyethylene glycol (PEG) precipitation at the same reference laboratory, and prolactin monomer was within the normal range. These patients were not pregnant and subsequent imaging studies were unremarkable. Suspecting our results may be falsely elevated, we performed a literature search and found several studies reporting the problem of overestimation of prolactin by the Cobas method (1,3,4). Despite the efforts to standardize prolactin methods using the 3rd World Health Organization (WHO) IRP Reference Standard 84/500 (1,5), prolactin results generated with different immunoassays are still not interchangeable, partially due to the presence of multiple isoforms of prolactin in the blood and different reactivities towards different forms of prolactin by different immunoassay methods (1,5).

The most often encountered isoforms of prolactin in clinical samples include the bioactive monomers and the less-active, high molecular weight (mol. wt.) big- or macro- prolactin. The monomeric prolactin (mol. wt. 23 kDa) accounts for 60–90% of the circulating prolactin and is the most biologically active form and central to its diverse physiological roles (1,6,7). Macroprolactin (mol. wt. 150 kDa), a complex of prolactin with immunoglobulin IgG, has minimal biological activity but can be detected by immunoassays leading to false diagnosis of hyperprolactinemia, resulting in unnecessary investigations and treatments (1,5,7). The prevalence of macroprolactin in patients with hyperprolactinemia was reported to be 10–45% depending on the prolactin method and the patient population studied (1,5,8-10). Immunoassays from different manufacturers react differently to these isoforms, leading to discordant results. All the commonly used immunoassays for prolactin on automated platforms are known to cross-react with macroprolactin to varying degrees, and the Roche method had the highest cross-reactivity (4,5,11). It was reported by Schneider et al. (5) that the Abbott method has a relatively lower cross-reactivity towards macroprolactin than the Roche method, and we do have an Abbott Architect 1000 (herein Architect) instrument in our laboratory, we decided to evaluate the prolactin method on the Architect. After a thorough validation of the Architect prolactin method and the verification of the reference intervals for both males and females, we switched our prolactin method from the Cobas to the Architect. Two years after this method change, we wanted to compare the clinical performance of the two prolactin methods by analyzing patient testing data generated by both methods. Prolactin data were pulled from the laboratory information system (LIS) and four datasets were constructed, each contained more than 7,000 data points generated by the two methods from similar periods. The ordering pattern by different medical specialties who saw these patients were reviewed and compared for two periods. Because our health system serves patients in a large metropolitan area in the United States, which has a broad spectrum of race and ethnicity, race or ethnicity was not a unique factor in our analysis. Student t-test was used to determine the significance of the difference in patient demographic distribution, in medical specialties, in percentile values between datasets. If the distributions of the patient demographics and their medical conditions (reflected by the specialties of the doctors who ordered prolactin) are not statistically different between datasets, then we assume patient demographic distribution and their medical conditions were statistically indifferent.

The traditional laboratory evaluation of a test method focuses on the analytical performance such as the sensitivity and specificity, the precision, the accuracy, the linearity and limit of quantitation, as well as the interferences and the reference intervals and other technical characteristics specific to a particular assay. The clinical performance of a testing method is lacking from analytical evaluations. We hope patient testing data analysis can provide a useful tool to assess the clinical impacts of a laboratory method.

The statistical parameters for each dataset include the median (50^th percentile), the 25^th quartile, and the 75^th quartile values. Most importantly, we want to determine the percentage of the prolactin results that were above or below the reference intervals of each method for males and females, respectively. If the percentage of the abnormally high results between the two methods is significantly different, it will provide the information regarding their clinical impacts. A higher percentage of abnormally high results will certainly lead to higher percentage of presumptive hyperprolactinemia that triggers more follow up testing or unnecessary imaging procedures or improper treatments.

Methods

Measurement of prolactin

The Elecsys Prolactin II assay on the Roche (Indianapolis, Indiana, USA) Cobas e801 analyzer uses electrochemiluminescence immunoassay technology with an analytical measurement range (AMR) of 0.5–470 ng/mL. The ARCHITECT Prolactin assay on the Abbott (Abbott Park, Illinois, USA) Architect 1000 uses chemiluminescent microparticle immunoassay technology, with an AMR of 0.6–200 ng/mL. Both assays are sandwich immunoassays using monoclonal antibodies that react with different epitopes of prolactin. The calibration of both assays can be traced to the 3rd WHO IRP Reference Standard 84/500. For the Cobas prolactin assay, we used apparent healthy donor samples (132 female and 144 male) to verify its reference interval of 4.79–23.3 ng/mL for females and 4.04–15.2 ng/mL for males. The reference intervals for the Architect prolactin method verified using 50 female and 48 male samples from apparent healthy donors are: 4.3–30.3 ng/mL for females and 3.5–19.4 ng/mL for males. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was granted an exemption by the Institutional Review Board of UCLA.

Method comparison using samples that did not contain macroprolactin

The prolactin method on the Cobas was compared with that on the Architect using 187 patient samples (139 female and 48 male) not containing macroprolactin based on normal recovery results (>60%) after PEG precipitation. To select samples free of macroprolactin, 300 µL of patient serum sample was mixed with 300 µL of 25% PEG and prolactin was measured in the supernatant after centrifugation. The recovery of prolactin was determined as the ratio of the prolactin in the supernatant divided by half of the prolactin in the original sample, with a recovery >60% indicating free of macroprolactin.

Data collection

Patient testing data of prolactin produced by both the Cobas and the Architect were pulled from LIS. Four datasets were constructed: the Architect-2024 and the Architect-2025, the Cobas-2022, and the Cobas-2023. The Architect-2024 and Architect-2025 dataset contained all the prolactin results from male and female patients measured with the Architect method in a 7-month period, from August 1, 2024, to February 28, 2025 (Architect-2024), and from August 1, 2025, to February 28, 2026 (Architect-2025). The Cobas-2022 and Cobas-2023 datasets contained all the prolactin results from male and female patients measured with the Cobas method from August 1, 2022, to February 28, 2023 (Cobas-2022) and from August 1, 2023, to February 28, 2024 (Cobas-2023). The Architect-2024 dataset and the Cobas-2023 dataset are to be compared in this study. The Cobas-2022 was used as a within-method comparison for the Cobas method, and the Architect-2025 dataset was used as a within-method comparison for the Architect method. All four datasets were used when the patient demographic distributions were compared. Because the amount of the patients contained in each dataset was very large, it was very difficult to find all the diagnosis. Instead, we analyzed the medical specialties of the doctors who ordered prolactin test on these patients in Cobas-2023 and Architect-2024 datasets, that should indirectly reflect the conditions of the patients who needed blood prolactin evaluations.

Statistical analysis

In each dataset, the values of the 5^th, 10^th, 25^th (Q1), 50^th (median), 75^th (Q3), 80^th, and 90^th percentile for male and female patients were determined, respectively. For each sex, patients were divided in four age groups: younger than 18, 18–45, 46–65, and older than 65 years. The percentage of patients for each age-sex group was determined. For each dataset, the percentage of the results that were higher (Percent High) than the URL or lower (Percent Low) than the lower reference limit (LRL) of their sex-specific reference intervals by each testing method were also determined.

Student’s t-test is used to assess the significance of the difference in the percentage of the patients in each age-sex group, and of the difference in the percentage of the medical specialties that ordered prolactin test between the two datasets, with a P value less than 0.05 indicating statistical significance. Confidence intervals (CIs) of the Percent High and Percent Low for each dataset were obtained with overlapping CI of the two values of Percent High or Percent Low indicating insignificant difference.

Results

Method comparison: Cobas vs. Architect

Figure 1A illustrates the comparison between the Cobas prolactin (y-axis) and the Architect prolactin (x-axis) values in 187 patient samples that did not contain macroprolactin, which produced a slope of 1.08 and an intercept of −1.13 with a correlation coefficient of 0.9863. The Bland-Altman Plot (Figure 1B) revealed that the bias is proportional to the average of the prolactin values from the two methods, indicating that the Cobas method tends to produce higher values than the Architect method in samples with higher concentrations of prolactin. As shown in Table 2, the Cobas method had 49 abnormally high prolactin results, and the Architect had 29 abnormally high results. Among the 49 Cobas’ high results, only 28 were confirmed by the Architect. There was one result by the Architect that was not confirmed by the Cobas. The rest of the results were normal by both methods.

Figure 1 Comparison of the prolactin results measured with the Cobas and the Architect method using 187 patient serum samples that did not contain macroprolactin. (A) Deming regression of the prolactin results (ng/mL) measured with the Cobas (y-axis) and the Architect (x-axis) method. (B) Bland-Altman difference plot. Y-axis: The difference in prolactin values by the Cobas and the Architect (ng/mL), X-axis: The mean of the two prolactin values by the Cobas and the Architect (ng/mL).

Comparability of the datasets

The Cobas-2023 dataset had 7,351 prolactin results with 4,952 results from females and 2,399 results from males. The age range for females was 1–89 years with a median of 35 years (Q1=26 years, Q3=43 years) and the age range for males was 1–89 years with a median of 46 years (Q1=36 years, Q3=61 years). The Architect-2024 dataset had very similar number of results, 7,196 total results with 4,972 from females and 2,224 from males. The age range for females was 1–94 years with a median of 35 years (Q1=27 years, Q3=44 years) and for males the range was 1–99 years with a median of 47 years (Q1=36 years, Q3=62 years). Table 3 summarizes the demographic distribution in percentage for each age-sex group for the Architect-2024 and the Architect-2025, and the Cobas-2022 and the Cobas-2023 dataset. Using the Student t-test to compare the percentage of patients in each of the four age groups between each dataset, the P value was >0.99 (2022 vs. 2023), >0.99 (2023 vs. 2024), 0.53 (2024 vs. 2025) for females, and >0.99 (2022 vs. 2023), >0.99 (2023 vs. 2024), 0.64 (2024 vs. 2025) for males, suggesting that patient demographic distributions were statistically consistent from year to year. We also reviewed the medical specialties of the doctors who ordered prolactin test on these patients. As shown in Table 4, endocrinologists, family physicians, obstetrician/gynecologists, and internists ordered about 26.43%, 16%, 15.48%, and 12.85% of prolactin tests in 2024, compared to 27.8%, 14.41%, 15.52%, and 12.98% in 2023. Comparing the specialties of the doctors who ordered prolactin in 2023 and 2024, the P value was 0.70, suggesting the medical conditions or symptoms of the patients in our health system that needed prolactin evaluations were similar in the two periods. The comparison of the demographic distributions of the patients between datasets, as well as the comparison of the medical specialties who ordered prolactin tests on these patients clearly showed that the patient populations and their symptoms that needed prolactin testing in the two periods were statistically indifferent. Therefore, we believe that it is valid to compare the parameters derived from the analysis of the Cobas-2023 and the Architect-2024 datasets and the difference in the parameters should reflect the difference in assay performance, not the difference in patient demographics or conditions. Patient demographic distributions also showed that the largest male and female cohorts were the 18–45 years age group, followed by the 46–65 years age group, a finding coincided with the need of the patients in their reproductive years for the evaluation of their prolactin.

The median, Q1, and Q3 for different age-sex groups

The Cobas-2023 data ranged from <0.5 to >9,400 ng/mL for both male and female patients. For females, the median prolactin was 14.3 ng/mL (Q1=9.8 ng/mL, and Q3=22.4 ng/mL) for the Cobas-2022, and was 13.5 ng/mL (Q1=9.2 ng/mL, and Q3=20.9 ng/mL) for the Cobas-2023 dataset. For males, the median prolactin was 10.3 ng/mL (Q1=7.3 ng/mL, and Q3=14.8 ng/mL) for the Cobas-2022 and was 9.7 ng/mL (Q1=6.98 ng/mL, and Q3=13.9 ng/mL) for the Cobas-2023 dataset. The Architect-2024 data ranged from 0.5 to 5,301 ng/mL for females and <0.6 to >8,000 ng/mL for males. For females, the median prolactin was 11.8 ng/mL (Q1=8.2 ng/mL, and Q3=18.3 ng/mL) for the Architect-2024 and was 12.8 ng/mL (Q1=8.8 ng/mL, and Q3=20.4 ng/mL) for the Architect-2025 dataset. For males, the median prolactin was 8.7 ng/mL (Q1=6.2 ng/mL, and Q3=12.9 ng/mL) for the Architect-2024 and 9.7 ng/mL (Q1=7.0 ng/mL, and Q3=14.2 ng/mL) for the Architect-2025 dataset. The values of the 5th, 10th, 25th, 50th, 75th, 80th, and 90th centile of the Cobas-2022, Cobas-2023, Architect-2024, and Architect-2025 dataset were obtained and given in Table 5. The percentile values of the Cobas-2023 dataset were plotted against that of the Architect-2024 dataset in Figure 2, which yielded a slope of 1.11 for females and 1.07 for males.

Figure 2 Comparison of the values of the 5^th, 10^th, 25^th, 50^th, 75^th, 90^th, and the 95^th percentile between the Cobas-2023 dataset and the Architect-2024 dataset for females (A) and for males (B). The slope of the Deming regression was 1.11 for females and 1.06 for males.

Figure 3 shows the comparison of the median, the Q1, and the Q3 of the four age groups between the Architect-2024 and the Cobas-2023 dataset. The medians of prolactin did not change significantly with age, which supports our current practice of sex-specific but age-independent reference intervals for adult male and female. In the age group of <18, 18–45, 46–65, and >65 years, the differences between the medians of the Cobas-2023 and the Architect-2024 datasets were rather small, with the Cobas values slightly higher than those of the Architect.

Figure 3 Comparison of the values for the 25^th percentile, the median, and the 75^th percentile of the prolactin results measured with the Cobas (red dot) and the Architect (green dot) method for the four age groups: <18, 18–45, 46–65, and >65 years. (A-C) Females. (D-F) Males.

Percent High and Percent Low

Percent High and Percent Low for each sex in each dataset were determined respectively and listed in Table 5. The Percent High for females was 20.3% (CI: 19.09–21.50%) in the Cobas-2023 dataset and 9.76% (CI: 8.73–10.79%) in the Architect-2024 dataset, indicating that the Cobas method yielded a significantly higher percentage of abnormally high results (more than 2X that of the Architect method). As a comparison, the Percent High for females was 23.07% (CI: 21.97–24.17%) for the Cobas-2022 dataset, showing that the Cobas method consistently produced higher percentage of abnormally elevated results for females. The Percent High for males was 13.25% (CI: 11.70–14.81%) for the Cobas-2023 dataset and 11.21% (CI: 9.59–12.84%) for the Architect-2024 dataset. Because the confidence intervals for males from the two datasets overlapped, the difference in the Percent High was statistically insignificant for males.

Discussion

In this study, we analyzed two datasets of patient prolactin results generated by two prolactin methods (Cobas-2023 and Architect-2024) in greater details, with two other datasets (Cobas-2022 and Architect-2025) as the comparison dataset, in four similar periods each containing more than 7,000 data points. The patient’s demographic distributions with respect to the four age groups (<18, 18–45, 46–65, and >65 years) were statistically indifferent (P value =0.70) across the four datasets for males and females, respectively. The clinical presentations of the patients for which prolactin test was ordered by the clinicians in various specialties across our health system were also statistically indifferent (P value =0.70). The values for the 10^th, 25^th, the 50^th, and the 75^th percentiles represent the values of the majority of the patients and they were very close for the Cobas-2022 and Cobas-2023, indicating that statistically patient populations or medical conditions were similar in those two periods. Similarly, the values for the 10^th, 25^th, the 50^th, and the 75^th percentiles for the Architect-2024 and Architect-2025 were also very close. Therefore, the statistical parameters obtained from the analysis of patient testing data can be compared. The difference in the percentage of the prolactin results that were above the reference interval for females was caused by the difference in the two assays, not by the difference in patient demographics and their clinical conditions. The Cobas method produced more abnormally high results (23.07% in 2022 and 20.3% in 2023) than the Architect method (9.76% in 2024 and 13.3% in 2025). We believe that this was caused by the improperly low URL of the female’s reference interval of the Cobas method with which more results were categorized as abnormally high.

Retrospective reviewing of our validation of the Cobas method performed 12 years ago, we realized that the donors (132 apparently healthy female and 144 apparently healthy male) were relatively young, in their 20s to 30s. These donors may be too “clean” to represent the entire population. The reference interval of the Cobas’ prolactin method has been questioned and investigated by several studies, some claimed it improperly low (1,3,4,11), and others reported it acceptable (9,12). Using an indirect non-parametric method to analyze the prolactin results from 127 male and 125 female patients who did not have hypothyroidism, pregnancy, or macroprolactin, Earll et al. (3) obtained higher reference intervals, 5.3–37.8 ng/mL for females and 4.2–22.8 ng/mL for males, than the ones provided by Roche (4.79–23.3 ng/mL for females, and 4.04–15.2 ng/mL for males). If we were to use the value of 37.8 ng/mL as the URL for females in our analysis of the Cobas-2023 dataset, the Percent High would be 8.52%, closer to that of the Architect-2024 dataset (9.76%). On the other hand, two groups reported that they obtained similar reference intervals to that provided by Roche using samples from healthy volunteer donors (9,12). One group reported 4.5–25.6 ng/mL for females using samples from 120 female donors (8), and the other reported 4.8–23.4 ng/mL for females using 198 non-pregnant female donors (12).

In our laboratory, the reference interval for the Architect method was verified as 4.3–30 ng/mL for females. One study using the Architect method obtained a reference interval of 3–26.46 ng/mL for females using 52 samples from non-pregnant female patients who had normal thyroid, liver, and renal functions (10). If we were to use 26.46 ng/mL as the URL for females in our analysis of the Architect-2024, the Percent High would be 12.09%, still a much lower value than that of the Cobas method (20.3%). Therefore, the relatively higher URL of the Architect method was not the cause for the difference in the Percent High. Our method comparison using samples that did not contain macroprolactin showed that the values of prolactin by the Cobas method were about 8% higher than those by the Architect method. The comparison of the percentile values in Figure 2 showed that the percentile values of the Cobas-2023 dataset were 11% higher than that of the Architect-2024 dataset, wherein the Cobas’ reference intervals were lower than that of the Architect. Therefore, the Cobas’ URL of their reference interval was improperly low, thereby causing a higher Percent High value.

It appeared that the values of the reference intervals for prolactin method varied, depending on the sources of the samples and the method to derive the reference intervals in each study. The URL of the reference interval is lower if a direct method was used to analyze samples from healthy volunteer donors, and higher if indirect method was used to analyze patient samples [when data distribution is that resemble a normal distribution according to Oosterhuis et al. (13)]. When volunteer donors were recruited, the number of samples tend to be limited and the donor pools may be too “clean” to represent the entire population. It seems that more samples from a wide spectrum of “normal subjects” are needed to verify or establish the reference intervals for prolactin. This is challenging for hospital laboratories because we tend to use 40-120 samples from volunteer donors to verify the reference intervals provided by assay manufacturers due to limited resources and time constraints.

Given the important role of prolactin in screening for hyperprolactinemia and knowing that all commonly used prolactin assays have cross-reactivity towards macroprolactin, we are in full support of the recommendations to routinely screen for macroprolactin using PEG precipitation method and to determine the concentration of prolactin monomers in samples with elevated prolactin results (5-9,11,12,14).

The strength of this study is in the statistical analysis of large amount of patient testing data. By analyzing the data, we can show that the Cobas method produced higher number of abnormally high results in female patients. This finding is consistent with previous reports that the Roche method overestimated prolactin (1,4,8,11,12). The limitation of the study is in the lack of definitive diagnosis of these patients. We hope this study can show that analyzing patient testing data is a useful way to obtain information regarding the clinical performance of a laboratory test.

Conclusions

Analyzing the prolactin testing data and comparing the statistical parameters between the Cobas and Architect prolactin methods showed that about 20.3% of the prolactin results produced by the Cobas method were above the URL of the reference interval recommended by the manufacturer in samples from female patients. We believe that the reference interval recommended by Roche is improperly low for females. We suggest laboratories that use the Cobas method for prolactin testing review their patient testing data to determine if their reference intervals are appropriate for their patient populations.

Acknowledgments

None.

Footnote

Data Sharing Statement: Available at https://jlpm.amegroups.com/article/view/10.21037/jlpm-25-50/dss

Peer Review File: Available at https://jlpm.amegroups.com/article/view/10.21037/jlpm-25-50/prf

Funding: None.

Conflicts of Interest: Both authors have completed the ICMJE uniform disclosure form (available at https://jlpm.amegroups.com/article/view/10.21037/jlpm-25-50/coif). L.S. serves as an unpaid editorial board member of Journal of Laboratory and Precision Medicine from January 2025 to December 2026. The other author has no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was granted an exemption by the Institutional Review Board of UCLA.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Saleem M, Martin H, Coates P. Prolactin Biology and Laboratory Measurement: An Update on Physiology and Current Analytical Issues. Clin Biochem Rev 2018;39:3-16.
2. Shimatsu A, Hattori N. Macroprolactinemia: diagnostic, clinical, and pathogenic significance. Clin Dev Immunol 2012;2012:167132. [Crossref] [PubMed]
3. Earll E, Javorsky BR, Sarvaideo J, et al. Clinical Impact of New Intervals for the Roche Prolactin II Immunoassay. J Endocr Soc 2024;8:bvae069. [Crossref] [PubMed]
4. De Sousa SMC, Saleem M, Rankin W, et al. Serum prolactin overestimation and risk of misdiagnosis. Endocrinol Diabetes Metab 2019;2:e00065. [Crossref] [PubMed]
5. Schneider W, Marcovitz S, Al-Shammari S, et al. Reactivity of macroprolactin in common automated immunoassays. Clin Biochem 2001;34:469-73. [Crossref] [PubMed]
6. Richa V, Rahul G, Sarika A. Macroprolactin; a frequent cause of misdiagnosed hyperprolactinemia in clinical practice. J Reprod Infertil 2010;11:161-7.
7. Fahie-Wilson MN, John R, Ellis AR. Macroprolactin; high molecular mass forms of circulating prolactin. Ann Clin Biochem 2005;42:175-92. [Crossref] [PubMed]
8. Beltran L, Fahie-Wilson MN, McKenna TJ, et al. Serum total prolactin and monomeric prolactin reference intervals determined by precipitation with polyethylene glycol: evaluation and validation on common immunoassay platforms. Clin Chem 2008;54:1673-81. [Crossref] [PubMed]
9. Hu Y, Ni J, Zhang B, et al. Establishment of reference intervals of monomeric prolactin to identify macroprolactinemia in Chinese patients with increased total prolactin. BMC Endocr Disord 2021;21:197. [Crossref] [PubMed]
10. Whitehead SJ, Cornes MP, Ford C, et al. Reference ranges for serum total and monomeric prolactin for the current generation Abbott Architect assay. Ann Clin Biochem 2015;52:61-6. [Crossref] [PubMed]
11. Pal S. Evaluation of the Correlation Coefficient of Polyethylene Glycol Treated and Direct Prolactin Results and Comparability with Different Assay System Results. EJIFCC 2017;28:315-27.
12. Fahie-Wilson M, Bieglmayer C, Kratzsch J, et al. Roche Elecsys Prolactin II assay: reactivity with macroprolactin compared with eight commercial assays for prolactin and determination of monomeric prolactin by precipitation with polyethylene glycol. Clin Lab 2007;53:301-7.
13. Oosterhuis WP, Modderman TA, Pronk C. Reference values: Bhattacharya or the method proposed by the IFCC? Ann Clin Biochem 1990;27:359-65.
14. Smith TP, Fahie-Wilson MN. Reporting of post-PEG prolactin concentrations: time to change. Clin Chem 2010;56:484-5. [Crossref] [PubMed]