- Email: [email protected]
Clinical laboratory reference intervals in pediatrics: The CALIPER initiative
Available online at www.sciencedirect.com
Clinical Biochemistry 42 (2009) 1589 – 1595
Review
Clinical laboratory reference intervals in pediatrics: The CALIPER initiative Benjamin Jung, Khosrow Adeli * Clinical Biochemistry, Department of Pediatric Laboratory Medicine, The Hospital for Sick Children, University of Toronto, Toronto, Ontario, Canada Received 6 May 2009; received in revised form 14 June 2009; accepted 24 June 2009 Available online 7 July 2009
Abstract Objective and rationale: Reference intervals provided on laboratory reports are essential for appropriate interpretation of test results, and can significantly impact clinical decision-making and the quality of patient care. Careful determination and/or validation of reference intervals by the laboratory for use in the patient population it serves are therefore important to ensure their proper utility. Unfortunately, critical gaps currently exist in accurate and up-to-date pediatric reference intervals for accurate interpretation of laboratory tests performed in children and adolescents. These critical gaps in the available pediatric laboratory reference intervals have the clear potential of contributing to erroneous diagnosis or misdiagnosis of many diseases of childhood and adolescence. Most of the available “normal” ranges for laboratory tests were determined over 2 decades ago on older instruments and technologies, and are no longer relevant considering the current testing technology used by clinical laboratories. It is thus critical and of utmost urgency that a more acceptable and comprehensive database be established. Discussion and conclusion: In the present review, we discuss the considerations and challenges faced when generating and validating reference intervals in accordance to the current guidelines published by the Clinical Laboratory Standards Institute (CLSI). We raise particular attention to the present-day deficiencies in available pediatric reference intervals, and highlight the special issues and unique difficulties that are additionally faced when establishing reference intervals in children. Finally, we highlight a recent Canadian initiative, the CALIPER project, whose mandate is to establish and maintain a database of comprehensive and up-to-date pediatric reference intervals to be eventually made available to all clinical laboratories worldwide. © 2009 The Canadian Society of Clinical Chemists. Published by Elsevier Inc. All rights reserved. Keywords: Reference intervals; Reference individuals; Pediatrics; Children; Partitioning
Contents The importance of proper reference intervals . . . . . . . . . . . . . . . . . Pediatric reference intervals: gaps and challenges . . . . . . . . . . . . . . Establishing population-based reference intervals. . . . . . . . . . . . . . . Reference individuals . . . . . . . . . . . . . . . . . . . . . . . . . . . Pre-analytical and analytical variables . . . . . . . . . . . . . . . . . . Exclusion of outlying observations (outliers) using statistical techniques Reference interval determination . . . . . . . . . . . . . . . . . . . . . Validation of partitioned reference intervals . . . . . . . . . . . . . . . Validation of reference intervals by transference . . . . . . . . . . . . . The CALIPER initiative . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
⁎ Corresponding author. Division of Clinical Biochemistry, DPLM, Hospital for Sick Children, University of Toronto, Toronto, Ontario, Canada M5G 1X8. Fax: +1 416 813 6257. E-mail address: [email protected] (K. Adeli).
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
1589 1591 1591 1592 1592 1593 1593 1593 1594 1594 1595
The importance of proper reference intervals Reference intervals provide valuable information to medical practitioners in their interpretation of quantitative laboratory test
0009-9120/$ - see front matter © 2009 The Canadian Society of Clinical Chemists. Published by Elsevier Inc. All rights reserved. doi:10.1016/j.clinbiochem.2009.06.025
1590
B. Jung, K. Adeli / Clinical Biochemistry 42 (2009) 1589–1595
results, and therefore are critical in the assessment of patient health and in clinical decision-making. The reference interval for an analyte denotes the statistically derived range of values determined from a reference interval study that encompasses the central 95% of values from an apparently healthy reference population. It follows then, that a test result that lies outside of the reference interval may represent an abnormal result. The reference interval, therefore, serves as a health-associated benchmark with which to compare an individual test result. While the concept of reference intervals and their utility appear straightforward, the process of establishing accurate and reliable reference intervals is considerably complex and involved. Each clinical laboratory is responsible for assuring the validity of reference intervals issued with their test results. Accrediting and licensing organizations, as well as regulatory bodies governing medical laboratory best practices, require that individual laboratories establish or verify reference intervals for all quantitative test methods offered by the laboratory service, the exception being for tests that employ decision cut-off limits (e.g. troponin, cholesterol, glucose, hemoglobin A1C, vitamins). Furthermore, the establishment or verification of the reference intervals must be carried out in a systematic and scientific manner using well-defined protocols with proper documentation in accordance to internationally established guidelines. A single reference interval study performed de novo by the traditional method requires the recruitment of at least 120 healthy reference individuals in order to achieve an acceptably high enough statistical confidence. One can easily appreciate the significant undertaking in terms of time, resources, and costs to the laboratory in order to abide by this requirement for each and every analyte measured. Although adopting reference intervals determined at another laboratory (or supplied by the manufacturer of the test) may seem like a sound approach that would simplify the task greatly and be of minimal cost, it cannot be done so blindly without performing a proper validation. Validation of an externally determined reference interval, or “transference of a reference interval”, is still required because of the potential consequence of employing a reference interval that would incorrectly classify a test result as abnormal when it is healthy, or healthy when it is abnormal. The reason a reference interval from one laboratory (though properly determined) may not be applicable to another laboratory may be due to 1) differences in the reference population of the donor laboratory compared to the patient population served by the receiving laboratory, or 2) differences in the analytical methods used between the laboratories. Measurements may be affected, for example, by inter-related variations in ethnic composition, geographic factors (i.e. climate, seasonal changes), diet preferences and food availability, and lifestyle factors (e.g. exercise). In addition, measurements made on different analytical methods may not be equivalent due to a number of factors, including differences in the measurement principle, in calibration, and reagent formulation; measurements made on the same analytical method may also differ due to differences in the operating environment, or lot-to-lot variation in reagents and calibrators, for instance. At minimum, a careful and thorough assessment of the available
reference interval must be done before deeming the reference interval appropriate for use by one's own laboratory. All pertinent information from the original reference interval study, such as the demographics and geographics of the reference individuals, pre-analytical and analytical details, and statistical methods used must be evaluated for validity and equivalence. However, as this is a subjective approach, with some degree of inherent uncertainty, there exists a risk that the reference interval is not applicable to the receiving population. Performing a transference of reference interval then is the best means of validation, and, while resource-consuming (a minimum number of 20 reference individuals is required), is still far less laborious and expensive than performing a full-scale reference interval study. In practice, clinical laboratories use transference to validate the majority of reference intervals they implement, and rarely perform a reference interval anew, which is normally reserved for new or emerging analytes for which no reference interval exists or for which a clearly different analytical method principle is being used. To lessen the formidable task of performing a full-scale reference interval study and to address the confusion of having different laboratory-specific reference intervals, a number of groups have recently tested the validity of establishing “common reference intervals” using a multicentre study design [1–5]. These reference intervals are determined through the collaboration of laboratories from different regions or countries that use analytical methods that are metrologically traceable to a reference method. Each participating laboratory contributes reference values to the determination of a common reference interval that can be utilized by the participating laboratories but also by other laboratories that use the same method. The rationale for the validity of common reference intervals determined this way is that potential differences due to analytical variables between laboratories are minimized by the use of reference materials traceable to a higher order reference method. Thus, this leaves only the potential differences due to differences in the reference population between laboratories to be considered, which can be addressed by statistical means. The concept of common reference intervals has been touted to represent, at least in part, “the way forward” for reference intervals [6] because a) they reduce costs in establishing the reference interval by spreading the burden of the number of reference individuals among the laboratories, b) the reference intervals can be widely adopted, at least in theory, by laboratories using the same metrologically traceable method (provided the reference population is similar), and c) they may better serve patients and physicians by having a single universal reference interval with which to interpret a test result. The major obstacles to the common reference interval approach are the lack of harmonization of methods by manufacturers and the lack of availability of reference materials and reference methods for the majority of analytes. Furthermore, at present, there exists no official guideline for setting and monitoring the minimum analytical quality goals that a laboratory must achieve in order to contribute reference values or to use the reference intervals over time. It also remains to be tested whether common reference intervals are robust when applied to ethnically diverse
B. Jung, K. Adeli / Clinical Biochemistry 42 (2009) 1589–1595
populations. The approach of common reference intervals is promising and rapidly evolving, but until these problems are characterized and resolved, establishing and validating laboratory-specific reference intervals as mandated will remain the status quo for the foreseeable future. Pediatric reference intervals: gaps and challenges Additional issues and challenges come into play when establishing reference intervals for use in a pediatric patient population. These issues and challenges (described below) largely underlie the reason for the critical gaps that exist today in accurate and up-to-date reference intervals for laboratory tests performed in children and adolescents. Children should not be viewed as small adults in the context of medical practice. Differences in physical size, organ maturity, body fluid compartments, (rates of) growth and development, immune and hormone responsiveness, nutrition and metabolism, are among the many factors that can influence normal analyte levels in children. Hence, the application of adult reference intervals is often not valid in a pediatric setting. Also, there are diseases that children are susceptible, or more susceptible, in acquiring than adults are, as well as genetically inherited disorders that are screened for in the newborn period or diagnosed during childhood. For these conditions, there may be unique or new biomarkers for which there is no alternative but to perform the reference interval study in a pediatric reference population. Furthermore, separate reference intervals, also called partitions, may be necessary for children of different age groups and/or genders, as well as for neonates, and premature babies. Despite the recognized need for pediatric reference intervals, however, pediatric reference intervals for many analytes and analytical methods remain inadequate or unavailable. Appropriate reference intervals touch all areas of pediatric laboratory testing, including in endocrinology, chemistry, serology, coagulation and haematology. It is well known that levels of sex hormones, growth hormones and bone alkaline phosphatase vary with a child's stage of growth and development; sex hormones, for example, are present at substantially lower concentration in infants and children as compared to adolescents or adults. Serology test results are influenced by the transplacental passage of maternal IgG and by immunization responses in infants and children. Pediatric reference intervals for a number of coagulation assays have been shown to be different [7]. Also, coagulation tests are optimized for anticoagulation monitoring and not for childhood genetic diseases such as haemophilia. In haematology, the automated differential uses algorithms in children that differ from those used in adults. Currently, sources for pediatric reference intervals include published sources, such as scientific journals and textbooks, and unpublished sources, such as may be found from (online) resources available from hospital laboratories, private laboratory centres, and reference laboratories (e.g. ARUP Laboratories, Mayo Medical Laboratories, Quest Diagnostics) and from manufacturer inserts. Unfortunately, most of the available reference intervals are incomplete, cover a limited pediatric age interval, and/or do not always cover both genders. To illustrate,
1591
critical gaps in reference intervals in four sub-specialties of clinical pediatric biochemistry were reviewed recently: bone markers [8], risk markers for cardiovascular disease and metabolic syndrome [9,10], hormones of the thyroid and growth hormone axes [11], and markers of inborn errors of metabolism [12]. Comprehensive pediatric reference intervals actually exist for only a limited number of analytes. Moreover, those available were largely determined over two decades ago, using outdated, less accurate laboratory instruments that employed outdated, less precise methodologies, and performed using protocols that were not in accordance with currently recommended guidelines, which therefore may make them inappropriate for use in the interpretation of test results generated by newer advanced methodologies and instrumentation currently in use. Frequently, changes in instrumentation and methodology are often not accompanied by reference interval validation, and many laboratories have been guilty of adopting published reference intervals or manufacturer supplied reference intervals without critical appraisal. Manufacturer supplied reference intervals often lack pertinent information regarding partitioning factors, sub-class differences, sample size, age, gender, ethnicity, race, percentiles used or traceability. Also, many pediatric reference intervals in use today were derived from samples collected on hospitalized infants and children, which is a departure from the classical requirement of values from healthy reference individuals. These gaps and deficiencies in accurate reference intervals can pose a serious risk to patient care. The interpretation of test results using inappropriate reference intervals may subject infants and children unnecessarily to further blood collection, infection risk, pain and anxiety, lengthier stays, and unpleasant or invasive diagnostic procedures. Worse, it could potentially result in an incorrect or delayed diagnosis leading to harmful inappropriate treatment. Determining age- and gender-specific reference intervals, therefore, is not only essential to the screening, diagnosis and monitoring of many pediatric disorders, but is crucial to ensuring patient safety. Establishing population-based reference intervals The Clinical and Laboratory Standards Institute (CLSI) and the International Federation of Clinical Chemistry and Laboratory Medicine (IFCC) have maintained an ongoing collaboration to provide up-to-date guidelines to the diagnostic laboratory community on how to define, establish, and verify reference intervals so that quality reference intervals may be achieved and implemented. The latest guideline (version C28A3) includes protocols with examples on how to determine reference intervals for new analytes or for analytes measured using a new analytical method, as well as on how to validate an existing reference interval established elsewhere for use in a different laboratory [13]. The guideline covers issues related to the selection of, and minimum number of reference individuals, and describes various statistical procedures to use in generating and transferring reference intervals, excluding outliers, and in partitioning intervals. Additionally, it discusses pre-analytical and analytical considerations, alternative statistical approaches,
1592
B. Jung, K. Adeli / Clinical Biochemistry 42 (2009) 1589–1595
and other practical aspects (e.g. how to present reference intervals on lab reports) and ongoing developments in reference interval methodology. The basic approaches contained in this guideline are described in the next section and special aspects to be considered in pediatric reference intervals are described under the appropriate heading. Some of the topics found in the guideline are beyond the scope of this review and are not expanded on. Reference individuals The quality of a reference interval is dependent on the proper recruitment of reference individuals. Reference intervals should be determined by the direct sampling of a healthy population of the age group(s) of interest, whereby reference individuals are selected using well-defined criteria either before (a priori) or after (a posteriori) samples are collected. The CLSI guideline does not endorse the use of indirect sampling, also referred to as “warehousing”, which involves the application of statistical methods to analytical values previously measured on a patient population. The problem with indirect techniques concerns the uncertainty of including values from patients that have diseases or conditions that have uncharacterized potential to affect the analyte of interest. The inclusion of a sufficient number of such values may potentially skew the reference interval, or lead to a much broader and therefore less sensitive reference interval. It is acknowledged, however, that the indirect technique is often used out of necessity when collection of sufficient numbers of reference samples may be difficult, such as in pediatrics and the elderly population. In direct sampling, what constitutes a “healthy” individual, and inclusion and exclusion criteria, must be well-defined at the outset, so that the inclusion of “non-healthy” reference individuals does not jeopardize the validity of the reference interval. Known sources of biological variation for the analyte, if available in the literature, should be used to develop these criteria and may also help identify what partitioning factors may be needed. Compliance with these criteria is ascertained using questionnaires filled out by reference individuals. Participants must also provide informed consent, and because the intention of the samples is not for medical investigation of the individual, ethical approval of the study is required. The guideline stipulates that for each reference interval or partition, 120 is the minimum number of reference individuals required to determine the 95th percentile reference limits with a confidence of 90%, if using the classical nonparametric protocol. Recruitment of more than 120 individuals is advised, however, to allow for the exclusion of outlying observations. Applying this approach to a pediatric reference interval study with 5 age groupings for each gender, then, would mean 1200 reference values for a single analyte! Convenient sources of healthy volunteers for adult reference intervals include the blood bank, hospital employees and university students. In pediatrics, however, finding sources for healthy volunteers from the different age groups and genders is more difficult. While the student population in primary and secondary schools represents a large base from which to recruit volunteers for most of the age groups 5 years old and above,
launching a campaign and collection of the specimens require significant organization and resources, not to mention the support by school boards. Also, collection from children under the age of 18 requires obtaining parental permission also. The age groups covering newborns to children under five years of age who have not started school, represent the hardest groups to obtain samples for. Collection from children of laboratory staff or their friends, and left-over specimens from routine testing of healthy newborns represent small potential sources for such samples. It should also be mentioned that it is often cost-effective to determine reference intervals for several analytes in batches using the same specimens. This holds true for pediatrics also, but due to the much smaller sample volumes that can be obtained, the number of analytes for which reference intervals can be derived from the same specimens is limited. A special consideration in reference individual selection relates to laboratories that service an ethnically diverse population. Reference intervals for major ethnic populations may need to be determined and evaluated to assess whether a single reference interval is valid or whether ethnic-specific reference intervals are necessary. Pre-analytical and analytical variables Pre-analytical and analytical variables need to be carefully considered. Depending on the analyte, individuals may be required to fast (food and/or fluids), abstain from drugs and other substances (e.g. caffeine, tobacco, alcohol, vitamins), and/ or limit their physical activity or stress. Also, levels of some analytes show variation with circadian rhythms, or stage of the menstrual cycle, in the case of females. Specimen collection considerations include environmental conditions during collection, specimen type, posture, time-of-day, sample volume, anticoagulants, additives, collection site preparation, blood flow, equipment and tourniquet time and technique. Specimen handling and processing variables include time and temperature for transport, clotting, plasma/serum separation, storage and sample preparation for analysis. In the pediatric population, patient age is obviously a significant pre-analytical factor and procedures for specimen collection and handling differ markedly from that for adults. Collection of specimens from infants and children is technically challenging, due to their smaller anatomy and their sensitivity to pain and medical procedures in general, and phlebotomists with special training and skill are required. The difficulties of phlebotomy performed on children, and especially on newborns and toddlers, are well appreciated; hemolyzed samples occur with greater frequency in blood samples collected from children than in adults, which is a problem as hemolysis is a known interferent in many analytical methods. Skin puncture specimens (capillary blood) are more commonly obtained from children and consist of a mixture of blood from arterioles, venules, and capillaries with interstitial and intracellular fluids. Also, liquid anticoagulant in collection tubes may potentially dilute the specimen when sample volumes are small. In regard to analytical variables, measurement of reference values should be performed on the assay method in the identical
B. Jung, K. Adeli / Clinical Biochemistry 42 (2009) 1589–1595
manner as that for routine patient samples. Normal instrument operation, maintenance and quality control procedures for reagents, calibrators, controls and calculation methods must be equivalent. Analytical performance, including traceability, precision, minimum detection limit, reportable range, recovery, and interference characteristics, need to be clearly stated. As mentioned above, many samples collected from children exhibit various degrees of hemolysis. Additionally, high concentrations of bilirubin and lipids are also present in many neonatal specimens. The presence of these substances may interfere with many spectrophotometric and nephelometric methods, causing spurious test results in both chemistry and haematology laboratories. Therefore, for methods that are susceptible to these interferences, samples should be screened for hemolysis, icterus, and lipemia. Exclusion of outlying observations (outliers) using statistical techniques Prior to determination of a reference interval, the data should be examined for outlying observations or outliers. A visual review of the distribution of the data, using a frequency histogram or box plot, may assist in identifying outliers. Excluding outliers from the reference interval calculation, however, requires statistical techniques: outliers must satisfy either the Dixon rule for outlier exclusion or the outlier exclusion method by Tukey. Otherwise, the outlier should be retained (unless an error relating to its measurement is uncovered). In Dixon's test, if the calculated absolute difference (D) between the suspected outlier and the next closest observation is more than one-third the absolute difference between the outlier and the furthest observation from the outlier (i.e. the range of the data, R), i.e. D/R N 1/3, then the outlier can be rejected [14]. If two or more outliers are suspected, Dixon's rule is applied to the least extreme outlier, and the entire group of outliers may be excluded. Tukey's method compares the outlier value against computed upper and lower limits for exclusion, called “fences” [15]. The calculation of the fences is based on the interquartile range for the dataset (i.e. the difference between the values at the 25th and 75th percentiles). The lower fence is given by the 25th percentile value minus 1.5 times the interquartile range, while the upper fence is given by the 75th percentile plus 1.5 times the interquartile range. Any value outside these fences is considered an outlier and may be excluded. In order to be valid, however, Tukey's method should be applied only to data that display a Gaussian distribution or to data that have been transformed to be so. Reference interval determination Following the exclusion of outliers, the reference interval may be statistically determined. The CLSI guideline prefers the use of the non-parametric method over the parametric method to generate reference intervals because the non-parametric method requires no assumption to be made about the shape of the underlying frequency distribution. This is appropriate, as most
1593
analytes do not display a Gaussian distribution. In the nonparametric method, all the reference values in the dataset are first rank ordered from smallest to largest. The conventional 95% reference limits are then determined by calculating the rank numbers for the 2.5th and 97.5th percentiles. The rank number for the 2.5th percentile is given by the product of 0.025 times the total number of reference values plus 1. Similarly, the 97.5th percentile is given by the product of 0.975 times the total number of reference values plus 1. For 120 reference values, the rank numbers for the 2.5th and 97.5th percentiles work out simply to be the 3rd and 118th reference values in the order. The guideline does not recommend the non-parametric method when the number of reference values is less than 120. However, for such instances, a different method is offered as a potential alternative. The “robust” method, developed by Horn and Pesce [16], is a more complicated statistical method both conceptually and computationally. First, each reference value is assigned a statistically computed “weight” that is proportional to the value's distance from the median, with more weight given to central values than to distant ones. These weighted values represent a new dataset whose distribution is a closer approximation of the underlying distribution of the data. Repeating the algorithm again and again but on the latest set of weighted values (referred to as “bootstrapping”) will generate successive distributions that are closer still. After sufficient iterations, an estimate of the 95th percentile reference interval can be generated. The use of as few as 20 reference values can yield reference intervals surprisingly close to the traditional method, as demonstrated in the guideline. However, to achieve a 90% confidence level, 80 is the recommended minimum number. The major advantage of the robust method over the traditional method is obviously the smaller sample size requirement. Additionally, it is also less sensitive to the presence of outliers. Recently, work in our laboratory succeeded in establishing pediatric reference intervals for 24 chemistries and 15 immunoassays on the Abbott Architect ci8200 platform using the Horn–Pesce robust method [17]. Validation of partitioned reference intervals Separate reference intervals for different age groups or sexes should exist only when there is clinical utility to have them, as unnecessary partitioning can create confusion, not to mention the significant effort and expense wasted in establishing them. The biological variability of the analyte, if known, should support the creation of partitions at the planning stages of a reference interval study. A large proportion of measured analytes in children, however, are known to have significant biological variability because during the childhood years, there are rapid and significant changes that occur during growth and development. Therefore, when reference interval studies are performed for biomarkers in pediatrics, partitioning of age groups can be justified, even when the evidence is lacking. Once reference intervals have been determined for each partition, the reference intervals need to be statistically tested to decide whether the partitions should remain separate, or whether they should be combined. The guideline proposes a few
1594
B. Jung, K. Adeli / Clinical Biochemistry 42 (2009) 1589–1595
statistical approaches but acknowledges none are ideal and that more research is needed, especially for testing multiple partitions. The most widely used of these is the Harris–Boyd method [18]. The method tests for a significant difference between the partition means by calculating the standard deviate test score, which is the difference between the means divided by the square root of the sum of the squares of the standard deviations of each mean. If the score exceeds a critical limit, partitioning is recommended. Partitioning is also recommended if the larger standard deviation is greater than 1.5 times the smaller one. Validation of reference intervals by transference The CLSI guideline provides three approaches for validating the transfer of reference intervals between laboratories. All three approaches require that comparability between analytical methods and comparability of reference populations (including pre-analytical and analytical variables) of the donor laboratory and the receiving laboratory be established before beginning. The first method, the transference using the subjective approach, was discussed above. The second involves measuring samples collected from at least 20 reference individuals of the receiving laboratory (local) population and performance of a formal outlier test. If no more than two results are outside the donor reference interval, the reference interval is acceptable for use. If greater than two lie outside the reference interval, then samples from 20 additional individuals are measured. However, if three or more again fall outside the limits (or if five or more in the original set fall outside the limits), the analytical procedures and the biological characteristics of the donor and receiving laboratory populations should be closely scrutinized. In these circumstances, the receiving laboratory may need to consider performing a full-scale study. The limitation of the second approach is that it will validate even donor reference intervals that are overly broad for the receiving laboratory population, which could potentially reduce the sensitivity of the reference interval of the receiving laboratory population. Hence, the third approach proposed, recommended for analytes for which the reference interval limits are critical for medical interpretation, is to consider the receiving laboratory and donor laboratory reference values as if they may belong to separate partitions; then, the same statistical approaches used for testing for partitions, as described above, may be used. If the results of the statistical analysis indicate the “partitions” may be combined, the receiving laboratory can adopt the donor reference interval. For this approach, 60 reference individuals are recommended in order for the statistical analysis to be sufficiently powerful. Again, with regard to transferring of pediatric reference intervals, these approaches are made more problematic for most clinical laboratories since truly normal pediatric samples are difficult to obtain. The CALIPER initiative As discussed, the unique issues and challenges in determining pediatric reference intervals pose significant roadblocks to
the creation of much needed up-to-date comprehensive pediatric reference intervals. An ambitious Canadian team of investigators from the pediatric focus group of the Canadian Society of Clinical Biochemists has assembled to spearhead the first nation-wide research initiative to tackle the substantial deficiencies in current pediatric reference intervals. The CAnadian Laboratory Initiative on PEdiatric Reference intervals database (CALIPER) project is a collaborative research project of children's hospitals across Canada, led by three principal investigators (K. Adeli, Toronto; N. Lepage, Ottawa; and V.J. Grey, Hamilton) and several other-co-investigators, who hold regular meetings to design, implement and monitor different phases of the project. The overall objective of the CALIPER project is to fill the gaps in pediatric reference intervals through the development and maintenance of a comprehensive database of reference intervals for laboratory tests that span the entire pediatric age range from birth to 18 years old. The long-term aims of the CALIPER project are to: • Establish and maintain a Canada-wide, comprehensive database for both traditional, as well as emerging biomarkers of pediatric disease. • Establish age-specific pediatric reference intervals from neonatal age to adolescence. • Establish gender-specific pediatric reference intervals from neonatal age to adolescence. • Establish pediatric reference intervals in major ethnic groups representing the current Canadian diverse population including native Canadians. CALIPER differs from a pediatric reference interval project being conducted in the United States by Children's Health Improvement through Laboratory Diagnostics (CHILDx; http:// www.childx.org/projects/studies/studies.html) in that the pediatric reference intervals by CALIPER will be determined on several different instrument platforms, resulting in instrumentspecific reference intervals. Phase I of CALIPER is underway, which is the organization of a campaign to recruit healthy volunteers belonging to each age grouping (0–2 months, 2–12 months, 1–5 years, 6– 10 years, 11–14, 16–19) for both boys and girls. Volunteers are recruited from varied sources and sites across Canada, including hospital nurseries, day cares, schools and families of hospital volunteers so as to ensure a representative sampling distribution. The campaign has had a good response since launching. Sample collection is performed by specially trained phlebotomists experienced in collecting pediatric specimens. Collected samples are processed (serum separated) before being aliquoted, archived and frozen until analysis. Developed guidelines are applied for tests where on-site analysis is required to preserve specimen stability, to avoid delays in analyte measurement and to achieve accurate results. In Phase II, analyte levels will be measured to establish reference intervals. Analysis will be carried out at specified centres using a number of different instruments commonly used in clinical laboratories across Canada. The goal is to generate
B. Jung, K. Adeli / Clinical Biochemistry 42 (2009) 1589–1595
instrument-specific reference intervals and determine variations in methods across laboratories. Following data collection, reference intervals will be established using parametric and nonparametric methods in collaboration with a clinical biostatistician. Based on detailed gap analyses mentioned earlier, a series of cardiac/metabolic markers, bone markers, genetic metabolic markers, thyroid hormones, the growth hormone-insulin axis, creatinine and C-reactive protein have been selected for the initial phase of reference interval determination. The CALIPER project has recently finished several pilot studies in collaboration with a number of in vitro diagnostics companies, resulting in pediatric reference intervals for numerous analytes [17,19,20]. While these reference intervals were generated using reference values from surplus specimens obtained from children attending outpatient clinics who were deemed to be metabolically stable, data generated from these pilot projects have already benefited the laboratory community. Phase II will determine these pediatric reference intervals again but using specimens collected from healthy volunteers in the community. The final Phase of the project is to implement standardized age-specific reference ranges in various pediatric healthcare centres across Canada. At present, the CALIPER project includes participants from 8 hospital centres across the country. Data generated by each centre is reviewed by the research team to ensure harmonization across the different study centres. Statistical approaches to data analysis and establishment of reference intervals will be in accordance to the CLSI guidelines on reference intervals. Reference intervals will be published on the CALIPER website so that the results of these efforts will not only benefit the practice of pediatric medicine in healthcare centres across Canada but worldwide. In the long-term, the project seeks to improve the health care of children and to promote academic, clinical and industrial collaborations among pediatric laboratory medicine specialists, pediatricians and others caring for children. References [1] Ferré-Masferrer M, Fuentes-Arderiu X, Alvarez-Funes V, Güell-Miró R, Castiñeiras-Lacambra MJ. Multicentric reference values: shared reference limits. Eur J Clin Chem Clin Biochem 1997;35(9):715–8. [2] Ferré-Masferrer M, Fuentes-Arderiu X, Gomà-Llongueras M, et al. Regional reference values for some quantities measured with the ADVIA Centaur analyser. A model of co-operation between the in vitro
[3]
[4]
[5]
[6] [7]
[8] [9]
[10] [11]
[12]
[13]
[14] [15] [16] [17]
[18] [19]
[20]
1595
diagnostic industry and clinical laboratories. Clin Chem Lab Med 2001;39 (2):166–9. Fuentes-Arderiu X, Alonso-Gregorio E, Alvarez-Funes V, et al. Multicentre physiological reference intervals for serum concentrations of immunoglobulins A, G and M, complement C3c and C4 measured with Tina-Quant reagents systems. Clin Chem Lab Med 2007;45(3):387–90. Fuentes-Arderiu X, Mas-Serra R, de-Nó-Lengaran C, et al. Multicentre physiological reference values for some urinary component-to-creatinine (creatininium) concentration ratios. Clin Chem Lab Med 2005;43(9): 958–62. Rustad P, Felding P, Franzson L, et al. The Nordic Reference Interval Project 2000: recommended reference intervals for 25 common biochemical properties. Scand J Clin Lab Invest 2004;64(4):271–84. Ceriotti F, Hinzmann R, Panteghini M. Reference intervals: the way forward. Ann Clin Biochem 2009;46(Pt 1):8–17. Flanders MM, Crist RA, Roberts WL, Rodgers GM. Pediatric reference intervals for seven common coagulation assays. Clin Chem 2005;51(9): 1738–42. Yang L, Grey V. Pediatric reference intervals for bone markers. Clin Biochem 2006;39(6):561–8. Mansoub S, Chan MK, Adeli K. Gap analysis of pediatric reference intervals for risk biomarkers of cardiovascular disease and the metabolic syndrome. Clin Biochem 2006;39(6):569–87. Davis GK, Bamforth F, Sarpal A, Dicke F, Rabi Y, Lyon ME. B-type natriuretic peptide in pediatrics. Clin Biochem 2006;39(6):600–5. Delvin EE, Laxmi Grey V, Vergee Z, CALIPER Working Group. Gap analysis of pediatric reference intervals related to thyroid hormones and the growth hormone-insulin growth factor axis. Clin Biochem 2006;39(6): 588–94. Lepage N, Li D, Kavsak PA, et al. Incomplete pediatric reference intervals for the management of patients with inborn errors of metabolism. Clin Biochem 2006;39(6):595–9. Clinical and Laboratory Standards Institute (CLSI). Defining, establishing, and verifying reference intervals in the clinical laboratory; approved guideline—third edition CLSI document C28-A3; 2008. Dixon WJ. Processing data for outliers. Biometrics 1953;9:74–89. Tukey JW. Exploratory data analysis. Reading, MA: Addison-Wesley; 1977. p. 688. Horn PS, Pesce AJ. Reference intervals. A user's guide. Washington, DC: AACC Press; 2005. Chan M, Seiden-Long I, Aytekin M, et al. Canadian Laboratory Initiative on Pediatric Reference Interval Database (CALIPER): pediatric reference intervals for an integrated clinical chemistry and immunoassay analyzer, Abbott Architect ci8200. Clin Biochem 2009;42(9):885–91. Harris EK, Boyd JC. On dividing reference data into subgroups to produce separate reference ranges. Clin Chem 1990;36(2):265–70. Chan M, Aytekin M, Adeli K. Paediatric reference intervals for 14 chemistries and immunoassays on the Roche Cobas 6000 System. Clin Chem 2008;54 (Suppl. 6):E115. Chan M, Quinn F, Preston N, Ravalico T, Armbruster D, Adeli K. Paediatric reference intervals for 14 chemistries and immunoassays on the Abbott Architect ci8200 System. Clin Chem 2008;54(Suppl. 6):E114.
Clinical Biochemistry 42 (2009) 1589 – 1595
Review
Clinical laboratory reference intervals in pediatrics: The CALIPER initiative Benjamin Jung, Khosrow Adeli * Clinical Biochemistry, Department of Pediatric Laboratory Medicine, The Hospital for Sick Children, University of Toronto, Toronto, Ontario, Canada Received 6 May 2009; received in revised form 14 June 2009; accepted 24 June 2009 Available online 7 July 2009
Abstract Objective and rationale: Reference intervals provided on laboratory reports are essential for appropriate interpretation of test results, and can significantly impact clinical decision-making and the quality of patient care. Careful determination and/or validation of reference intervals by the laboratory for use in the patient population it serves are therefore important to ensure their proper utility. Unfortunately, critical gaps currently exist in accurate and up-to-date pediatric reference intervals for accurate interpretation of laboratory tests performed in children and adolescents. These critical gaps in the available pediatric laboratory reference intervals have the clear potential of contributing to erroneous diagnosis or misdiagnosis of many diseases of childhood and adolescence. Most of the available “normal” ranges for laboratory tests were determined over 2 decades ago on older instruments and technologies, and are no longer relevant considering the current testing technology used by clinical laboratories. It is thus critical and of utmost urgency that a more acceptable and comprehensive database be established. Discussion and conclusion: In the present review, we discuss the considerations and challenges faced when generating and validating reference intervals in accordance to the current guidelines published by the Clinical Laboratory Standards Institute (CLSI). We raise particular attention to the present-day deficiencies in available pediatric reference intervals, and highlight the special issues and unique difficulties that are additionally faced when establishing reference intervals in children. Finally, we highlight a recent Canadian initiative, the CALIPER project, whose mandate is to establish and maintain a database of comprehensive and up-to-date pediatric reference intervals to be eventually made available to all clinical laboratories worldwide. © 2009 The Canadian Society of Clinical Chemists. Published by Elsevier Inc. All rights reserved. Keywords: Reference intervals; Reference individuals; Pediatrics; Children; Partitioning
Contents The importance of proper reference intervals . . . . . . . . . . . . . . . . . Pediatric reference intervals: gaps and challenges . . . . . . . . . . . . . . Establishing population-based reference intervals. . . . . . . . . . . . . . . Reference individuals . . . . . . . . . . . . . . . . . . . . . . . . . . . Pre-analytical and analytical variables . . . . . . . . . . . . . . . . . . Exclusion of outlying observations (outliers) using statistical techniques Reference interval determination . . . . . . . . . . . . . . . . . . . . . Validation of partitioned reference intervals . . . . . . . . . . . . . . . Validation of reference intervals by transference . . . . . . . . . . . . . The CALIPER initiative . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
⁎ Corresponding author. Division of Clinical Biochemistry, DPLM, Hospital for Sick Children, University of Toronto, Toronto, Ontario, Canada M5G 1X8. Fax: +1 416 813 6257. E-mail address: [email protected] (K. Adeli).
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
1589 1591 1591 1592 1592 1593 1593 1593 1594 1594 1595
The importance of proper reference intervals Reference intervals provide valuable information to medical practitioners in their interpretation of quantitative laboratory test
0009-9120/$ - see front matter © 2009 The Canadian Society of Clinical Chemists. Published by Elsevier Inc. All rights reserved. doi:10.1016/j.clinbiochem.2009.06.025
1590
B. Jung, K. Adeli / Clinical Biochemistry 42 (2009) 1589–1595
results, and therefore are critical in the assessment of patient health and in clinical decision-making. The reference interval for an analyte denotes the statistically derived range of values determined from a reference interval study that encompasses the central 95% of values from an apparently healthy reference population. It follows then, that a test result that lies outside of the reference interval may represent an abnormal result. The reference interval, therefore, serves as a health-associated benchmark with which to compare an individual test result. While the concept of reference intervals and their utility appear straightforward, the process of establishing accurate and reliable reference intervals is considerably complex and involved. Each clinical laboratory is responsible for assuring the validity of reference intervals issued with their test results. Accrediting and licensing organizations, as well as regulatory bodies governing medical laboratory best practices, require that individual laboratories establish or verify reference intervals for all quantitative test methods offered by the laboratory service, the exception being for tests that employ decision cut-off limits (e.g. troponin, cholesterol, glucose, hemoglobin A1C, vitamins). Furthermore, the establishment or verification of the reference intervals must be carried out in a systematic and scientific manner using well-defined protocols with proper documentation in accordance to internationally established guidelines. A single reference interval study performed de novo by the traditional method requires the recruitment of at least 120 healthy reference individuals in order to achieve an acceptably high enough statistical confidence. One can easily appreciate the significant undertaking in terms of time, resources, and costs to the laboratory in order to abide by this requirement for each and every analyte measured. Although adopting reference intervals determined at another laboratory (or supplied by the manufacturer of the test) may seem like a sound approach that would simplify the task greatly and be of minimal cost, it cannot be done so blindly without performing a proper validation. Validation of an externally determined reference interval, or “transference of a reference interval”, is still required because of the potential consequence of employing a reference interval that would incorrectly classify a test result as abnormal when it is healthy, or healthy when it is abnormal. The reason a reference interval from one laboratory (though properly determined) may not be applicable to another laboratory may be due to 1) differences in the reference population of the donor laboratory compared to the patient population served by the receiving laboratory, or 2) differences in the analytical methods used between the laboratories. Measurements may be affected, for example, by inter-related variations in ethnic composition, geographic factors (i.e. climate, seasonal changes), diet preferences and food availability, and lifestyle factors (e.g. exercise). In addition, measurements made on different analytical methods may not be equivalent due to a number of factors, including differences in the measurement principle, in calibration, and reagent formulation; measurements made on the same analytical method may also differ due to differences in the operating environment, or lot-to-lot variation in reagents and calibrators, for instance. At minimum, a careful and thorough assessment of the available
reference interval must be done before deeming the reference interval appropriate for use by one's own laboratory. All pertinent information from the original reference interval study, such as the demographics and geographics of the reference individuals, pre-analytical and analytical details, and statistical methods used must be evaluated for validity and equivalence. However, as this is a subjective approach, with some degree of inherent uncertainty, there exists a risk that the reference interval is not applicable to the receiving population. Performing a transference of reference interval then is the best means of validation, and, while resource-consuming (a minimum number of 20 reference individuals is required), is still far less laborious and expensive than performing a full-scale reference interval study. In practice, clinical laboratories use transference to validate the majority of reference intervals they implement, and rarely perform a reference interval anew, which is normally reserved for new or emerging analytes for which no reference interval exists or for which a clearly different analytical method principle is being used. To lessen the formidable task of performing a full-scale reference interval study and to address the confusion of having different laboratory-specific reference intervals, a number of groups have recently tested the validity of establishing “common reference intervals” using a multicentre study design [1–5]. These reference intervals are determined through the collaboration of laboratories from different regions or countries that use analytical methods that are metrologically traceable to a reference method. Each participating laboratory contributes reference values to the determination of a common reference interval that can be utilized by the participating laboratories but also by other laboratories that use the same method. The rationale for the validity of common reference intervals determined this way is that potential differences due to analytical variables between laboratories are minimized by the use of reference materials traceable to a higher order reference method. Thus, this leaves only the potential differences due to differences in the reference population between laboratories to be considered, which can be addressed by statistical means. The concept of common reference intervals has been touted to represent, at least in part, “the way forward” for reference intervals [6] because a) they reduce costs in establishing the reference interval by spreading the burden of the number of reference individuals among the laboratories, b) the reference intervals can be widely adopted, at least in theory, by laboratories using the same metrologically traceable method (provided the reference population is similar), and c) they may better serve patients and physicians by having a single universal reference interval with which to interpret a test result. The major obstacles to the common reference interval approach are the lack of harmonization of methods by manufacturers and the lack of availability of reference materials and reference methods for the majority of analytes. Furthermore, at present, there exists no official guideline for setting and monitoring the minimum analytical quality goals that a laboratory must achieve in order to contribute reference values or to use the reference intervals over time. It also remains to be tested whether common reference intervals are robust when applied to ethnically diverse
B. Jung, K. Adeli / Clinical Biochemistry 42 (2009) 1589–1595
populations. The approach of common reference intervals is promising and rapidly evolving, but until these problems are characterized and resolved, establishing and validating laboratory-specific reference intervals as mandated will remain the status quo for the foreseeable future. Pediatric reference intervals: gaps and challenges Additional issues and challenges come into play when establishing reference intervals for use in a pediatric patient population. These issues and challenges (described below) largely underlie the reason for the critical gaps that exist today in accurate and up-to-date reference intervals for laboratory tests performed in children and adolescents. Children should not be viewed as small adults in the context of medical practice. Differences in physical size, organ maturity, body fluid compartments, (rates of) growth and development, immune and hormone responsiveness, nutrition and metabolism, are among the many factors that can influence normal analyte levels in children. Hence, the application of adult reference intervals is often not valid in a pediatric setting. Also, there are diseases that children are susceptible, or more susceptible, in acquiring than adults are, as well as genetically inherited disorders that are screened for in the newborn period or diagnosed during childhood. For these conditions, there may be unique or new biomarkers for which there is no alternative but to perform the reference interval study in a pediatric reference population. Furthermore, separate reference intervals, also called partitions, may be necessary for children of different age groups and/or genders, as well as for neonates, and premature babies. Despite the recognized need for pediatric reference intervals, however, pediatric reference intervals for many analytes and analytical methods remain inadequate or unavailable. Appropriate reference intervals touch all areas of pediatric laboratory testing, including in endocrinology, chemistry, serology, coagulation and haematology. It is well known that levels of sex hormones, growth hormones and bone alkaline phosphatase vary with a child's stage of growth and development; sex hormones, for example, are present at substantially lower concentration in infants and children as compared to adolescents or adults. Serology test results are influenced by the transplacental passage of maternal IgG and by immunization responses in infants and children. Pediatric reference intervals for a number of coagulation assays have been shown to be different [7]. Also, coagulation tests are optimized for anticoagulation monitoring and not for childhood genetic diseases such as haemophilia. In haematology, the automated differential uses algorithms in children that differ from those used in adults. Currently, sources for pediatric reference intervals include published sources, such as scientific journals and textbooks, and unpublished sources, such as may be found from (online) resources available from hospital laboratories, private laboratory centres, and reference laboratories (e.g. ARUP Laboratories, Mayo Medical Laboratories, Quest Diagnostics) and from manufacturer inserts. Unfortunately, most of the available reference intervals are incomplete, cover a limited pediatric age interval, and/or do not always cover both genders. To illustrate,
1591
critical gaps in reference intervals in four sub-specialties of clinical pediatric biochemistry were reviewed recently: bone markers [8], risk markers for cardiovascular disease and metabolic syndrome [9,10], hormones of the thyroid and growth hormone axes [11], and markers of inborn errors of metabolism [12]. Comprehensive pediatric reference intervals actually exist for only a limited number of analytes. Moreover, those available were largely determined over two decades ago, using outdated, less accurate laboratory instruments that employed outdated, less precise methodologies, and performed using protocols that were not in accordance with currently recommended guidelines, which therefore may make them inappropriate for use in the interpretation of test results generated by newer advanced methodologies and instrumentation currently in use. Frequently, changes in instrumentation and methodology are often not accompanied by reference interval validation, and many laboratories have been guilty of adopting published reference intervals or manufacturer supplied reference intervals without critical appraisal. Manufacturer supplied reference intervals often lack pertinent information regarding partitioning factors, sub-class differences, sample size, age, gender, ethnicity, race, percentiles used or traceability. Also, many pediatric reference intervals in use today were derived from samples collected on hospitalized infants and children, which is a departure from the classical requirement of values from healthy reference individuals. These gaps and deficiencies in accurate reference intervals can pose a serious risk to patient care. The interpretation of test results using inappropriate reference intervals may subject infants and children unnecessarily to further blood collection, infection risk, pain and anxiety, lengthier stays, and unpleasant or invasive diagnostic procedures. Worse, it could potentially result in an incorrect or delayed diagnosis leading to harmful inappropriate treatment. Determining age- and gender-specific reference intervals, therefore, is not only essential to the screening, diagnosis and monitoring of many pediatric disorders, but is crucial to ensuring patient safety. Establishing population-based reference intervals The Clinical and Laboratory Standards Institute (CLSI) and the International Federation of Clinical Chemistry and Laboratory Medicine (IFCC) have maintained an ongoing collaboration to provide up-to-date guidelines to the diagnostic laboratory community on how to define, establish, and verify reference intervals so that quality reference intervals may be achieved and implemented. The latest guideline (version C28A3) includes protocols with examples on how to determine reference intervals for new analytes or for analytes measured using a new analytical method, as well as on how to validate an existing reference interval established elsewhere for use in a different laboratory [13]. The guideline covers issues related to the selection of, and minimum number of reference individuals, and describes various statistical procedures to use in generating and transferring reference intervals, excluding outliers, and in partitioning intervals. Additionally, it discusses pre-analytical and analytical considerations, alternative statistical approaches,
1592
B. Jung, K. Adeli / Clinical Biochemistry 42 (2009) 1589–1595
and other practical aspects (e.g. how to present reference intervals on lab reports) and ongoing developments in reference interval methodology. The basic approaches contained in this guideline are described in the next section and special aspects to be considered in pediatric reference intervals are described under the appropriate heading. Some of the topics found in the guideline are beyond the scope of this review and are not expanded on. Reference individuals The quality of a reference interval is dependent on the proper recruitment of reference individuals. Reference intervals should be determined by the direct sampling of a healthy population of the age group(s) of interest, whereby reference individuals are selected using well-defined criteria either before (a priori) or after (a posteriori) samples are collected. The CLSI guideline does not endorse the use of indirect sampling, also referred to as “warehousing”, which involves the application of statistical methods to analytical values previously measured on a patient population. The problem with indirect techniques concerns the uncertainty of including values from patients that have diseases or conditions that have uncharacterized potential to affect the analyte of interest. The inclusion of a sufficient number of such values may potentially skew the reference interval, or lead to a much broader and therefore less sensitive reference interval. It is acknowledged, however, that the indirect technique is often used out of necessity when collection of sufficient numbers of reference samples may be difficult, such as in pediatrics and the elderly population. In direct sampling, what constitutes a “healthy” individual, and inclusion and exclusion criteria, must be well-defined at the outset, so that the inclusion of “non-healthy” reference individuals does not jeopardize the validity of the reference interval. Known sources of biological variation for the analyte, if available in the literature, should be used to develop these criteria and may also help identify what partitioning factors may be needed. Compliance with these criteria is ascertained using questionnaires filled out by reference individuals. Participants must also provide informed consent, and because the intention of the samples is not for medical investigation of the individual, ethical approval of the study is required. The guideline stipulates that for each reference interval or partition, 120 is the minimum number of reference individuals required to determine the 95th percentile reference limits with a confidence of 90%, if using the classical nonparametric protocol. Recruitment of more than 120 individuals is advised, however, to allow for the exclusion of outlying observations. Applying this approach to a pediatric reference interval study with 5 age groupings for each gender, then, would mean 1200 reference values for a single analyte! Convenient sources of healthy volunteers for adult reference intervals include the blood bank, hospital employees and university students. In pediatrics, however, finding sources for healthy volunteers from the different age groups and genders is more difficult. While the student population in primary and secondary schools represents a large base from which to recruit volunteers for most of the age groups 5 years old and above,
launching a campaign and collection of the specimens require significant organization and resources, not to mention the support by school boards. Also, collection from children under the age of 18 requires obtaining parental permission also. The age groups covering newborns to children under five years of age who have not started school, represent the hardest groups to obtain samples for. Collection from children of laboratory staff or their friends, and left-over specimens from routine testing of healthy newborns represent small potential sources for such samples. It should also be mentioned that it is often cost-effective to determine reference intervals for several analytes in batches using the same specimens. This holds true for pediatrics also, but due to the much smaller sample volumes that can be obtained, the number of analytes for which reference intervals can be derived from the same specimens is limited. A special consideration in reference individual selection relates to laboratories that service an ethnically diverse population. Reference intervals for major ethnic populations may need to be determined and evaluated to assess whether a single reference interval is valid or whether ethnic-specific reference intervals are necessary. Pre-analytical and analytical variables Pre-analytical and analytical variables need to be carefully considered. Depending on the analyte, individuals may be required to fast (food and/or fluids), abstain from drugs and other substances (e.g. caffeine, tobacco, alcohol, vitamins), and/ or limit their physical activity or stress. Also, levels of some analytes show variation with circadian rhythms, or stage of the menstrual cycle, in the case of females. Specimen collection considerations include environmental conditions during collection, specimen type, posture, time-of-day, sample volume, anticoagulants, additives, collection site preparation, blood flow, equipment and tourniquet time and technique. Specimen handling and processing variables include time and temperature for transport, clotting, plasma/serum separation, storage and sample preparation for analysis. In the pediatric population, patient age is obviously a significant pre-analytical factor and procedures for specimen collection and handling differ markedly from that for adults. Collection of specimens from infants and children is technically challenging, due to their smaller anatomy and their sensitivity to pain and medical procedures in general, and phlebotomists with special training and skill are required. The difficulties of phlebotomy performed on children, and especially on newborns and toddlers, are well appreciated; hemolyzed samples occur with greater frequency in blood samples collected from children than in adults, which is a problem as hemolysis is a known interferent in many analytical methods. Skin puncture specimens (capillary blood) are more commonly obtained from children and consist of a mixture of blood from arterioles, venules, and capillaries with interstitial and intracellular fluids. Also, liquid anticoagulant in collection tubes may potentially dilute the specimen when sample volumes are small. In regard to analytical variables, measurement of reference values should be performed on the assay method in the identical
B. Jung, K. Adeli / Clinical Biochemistry 42 (2009) 1589–1595
manner as that for routine patient samples. Normal instrument operation, maintenance and quality control procedures for reagents, calibrators, controls and calculation methods must be equivalent. Analytical performance, including traceability, precision, minimum detection limit, reportable range, recovery, and interference characteristics, need to be clearly stated. As mentioned above, many samples collected from children exhibit various degrees of hemolysis. Additionally, high concentrations of bilirubin and lipids are also present in many neonatal specimens. The presence of these substances may interfere with many spectrophotometric and nephelometric methods, causing spurious test results in both chemistry and haematology laboratories. Therefore, for methods that are susceptible to these interferences, samples should be screened for hemolysis, icterus, and lipemia. Exclusion of outlying observations (outliers) using statistical techniques Prior to determination of a reference interval, the data should be examined for outlying observations or outliers. A visual review of the distribution of the data, using a frequency histogram or box plot, may assist in identifying outliers. Excluding outliers from the reference interval calculation, however, requires statistical techniques: outliers must satisfy either the Dixon rule for outlier exclusion or the outlier exclusion method by Tukey. Otherwise, the outlier should be retained (unless an error relating to its measurement is uncovered). In Dixon's test, if the calculated absolute difference (D) between the suspected outlier and the next closest observation is more than one-third the absolute difference between the outlier and the furthest observation from the outlier (i.e. the range of the data, R), i.e. D/R N 1/3, then the outlier can be rejected [14]. If two or more outliers are suspected, Dixon's rule is applied to the least extreme outlier, and the entire group of outliers may be excluded. Tukey's method compares the outlier value against computed upper and lower limits for exclusion, called “fences” [15]. The calculation of the fences is based on the interquartile range for the dataset (i.e. the difference between the values at the 25th and 75th percentiles). The lower fence is given by the 25th percentile value minus 1.5 times the interquartile range, while the upper fence is given by the 75th percentile plus 1.5 times the interquartile range. Any value outside these fences is considered an outlier and may be excluded. In order to be valid, however, Tukey's method should be applied only to data that display a Gaussian distribution or to data that have been transformed to be so. Reference interval determination Following the exclusion of outliers, the reference interval may be statistically determined. The CLSI guideline prefers the use of the non-parametric method over the parametric method to generate reference intervals because the non-parametric method requires no assumption to be made about the shape of the underlying frequency distribution. This is appropriate, as most
1593
analytes do not display a Gaussian distribution. In the nonparametric method, all the reference values in the dataset are first rank ordered from smallest to largest. The conventional 95% reference limits are then determined by calculating the rank numbers for the 2.5th and 97.5th percentiles. The rank number for the 2.5th percentile is given by the product of 0.025 times the total number of reference values plus 1. Similarly, the 97.5th percentile is given by the product of 0.975 times the total number of reference values plus 1. For 120 reference values, the rank numbers for the 2.5th and 97.5th percentiles work out simply to be the 3rd and 118th reference values in the order. The guideline does not recommend the non-parametric method when the number of reference values is less than 120. However, for such instances, a different method is offered as a potential alternative. The “robust” method, developed by Horn and Pesce [16], is a more complicated statistical method both conceptually and computationally. First, each reference value is assigned a statistically computed “weight” that is proportional to the value's distance from the median, with more weight given to central values than to distant ones. These weighted values represent a new dataset whose distribution is a closer approximation of the underlying distribution of the data. Repeating the algorithm again and again but on the latest set of weighted values (referred to as “bootstrapping”) will generate successive distributions that are closer still. After sufficient iterations, an estimate of the 95th percentile reference interval can be generated. The use of as few as 20 reference values can yield reference intervals surprisingly close to the traditional method, as demonstrated in the guideline. However, to achieve a 90% confidence level, 80 is the recommended minimum number. The major advantage of the robust method over the traditional method is obviously the smaller sample size requirement. Additionally, it is also less sensitive to the presence of outliers. Recently, work in our laboratory succeeded in establishing pediatric reference intervals for 24 chemistries and 15 immunoassays on the Abbott Architect ci8200 platform using the Horn–Pesce robust method [17]. Validation of partitioned reference intervals Separate reference intervals for different age groups or sexes should exist only when there is clinical utility to have them, as unnecessary partitioning can create confusion, not to mention the significant effort and expense wasted in establishing them. The biological variability of the analyte, if known, should support the creation of partitions at the planning stages of a reference interval study. A large proportion of measured analytes in children, however, are known to have significant biological variability because during the childhood years, there are rapid and significant changes that occur during growth and development. Therefore, when reference interval studies are performed for biomarkers in pediatrics, partitioning of age groups can be justified, even when the evidence is lacking. Once reference intervals have been determined for each partition, the reference intervals need to be statistically tested to decide whether the partitions should remain separate, or whether they should be combined. The guideline proposes a few
1594
B. Jung, K. Adeli / Clinical Biochemistry 42 (2009) 1589–1595
statistical approaches but acknowledges none are ideal and that more research is needed, especially for testing multiple partitions. The most widely used of these is the Harris–Boyd method [18]. The method tests for a significant difference between the partition means by calculating the standard deviate test score, which is the difference between the means divided by the square root of the sum of the squares of the standard deviations of each mean. If the score exceeds a critical limit, partitioning is recommended. Partitioning is also recommended if the larger standard deviation is greater than 1.5 times the smaller one. Validation of reference intervals by transference The CLSI guideline provides three approaches for validating the transfer of reference intervals between laboratories. All three approaches require that comparability between analytical methods and comparability of reference populations (including pre-analytical and analytical variables) of the donor laboratory and the receiving laboratory be established before beginning. The first method, the transference using the subjective approach, was discussed above. The second involves measuring samples collected from at least 20 reference individuals of the receiving laboratory (local) population and performance of a formal outlier test. If no more than two results are outside the donor reference interval, the reference interval is acceptable for use. If greater than two lie outside the reference interval, then samples from 20 additional individuals are measured. However, if three or more again fall outside the limits (or if five or more in the original set fall outside the limits), the analytical procedures and the biological characteristics of the donor and receiving laboratory populations should be closely scrutinized. In these circumstances, the receiving laboratory may need to consider performing a full-scale study. The limitation of the second approach is that it will validate even donor reference intervals that are overly broad for the receiving laboratory population, which could potentially reduce the sensitivity of the reference interval of the receiving laboratory population. Hence, the third approach proposed, recommended for analytes for which the reference interval limits are critical for medical interpretation, is to consider the receiving laboratory and donor laboratory reference values as if they may belong to separate partitions; then, the same statistical approaches used for testing for partitions, as described above, may be used. If the results of the statistical analysis indicate the “partitions” may be combined, the receiving laboratory can adopt the donor reference interval. For this approach, 60 reference individuals are recommended in order for the statistical analysis to be sufficiently powerful. Again, with regard to transferring of pediatric reference intervals, these approaches are made more problematic for most clinical laboratories since truly normal pediatric samples are difficult to obtain. The CALIPER initiative As discussed, the unique issues and challenges in determining pediatric reference intervals pose significant roadblocks to
the creation of much needed up-to-date comprehensive pediatric reference intervals. An ambitious Canadian team of investigators from the pediatric focus group of the Canadian Society of Clinical Biochemists has assembled to spearhead the first nation-wide research initiative to tackle the substantial deficiencies in current pediatric reference intervals. The CAnadian Laboratory Initiative on PEdiatric Reference intervals database (CALIPER) project is a collaborative research project of children's hospitals across Canada, led by three principal investigators (K. Adeli, Toronto; N. Lepage, Ottawa; and V.J. Grey, Hamilton) and several other-co-investigators, who hold regular meetings to design, implement and monitor different phases of the project. The overall objective of the CALIPER project is to fill the gaps in pediatric reference intervals through the development and maintenance of a comprehensive database of reference intervals for laboratory tests that span the entire pediatric age range from birth to 18 years old. The long-term aims of the CALIPER project are to: • Establish and maintain a Canada-wide, comprehensive database for both traditional, as well as emerging biomarkers of pediatric disease. • Establish age-specific pediatric reference intervals from neonatal age to adolescence. • Establish gender-specific pediatric reference intervals from neonatal age to adolescence. • Establish pediatric reference intervals in major ethnic groups representing the current Canadian diverse population including native Canadians. CALIPER differs from a pediatric reference interval project being conducted in the United States by Children's Health Improvement through Laboratory Diagnostics (CHILDx; http:// www.childx.org/projects/studies/studies.html) in that the pediatric reference intervals by CALIPER will be determined on several different instrument platforms, resulting in instrumentspecific reference intervals. Phase I of CALIPER is underway, which is the organization of a campaign to recruit healthy volunteers belonging to each age grouping (0–2 months, 2–12 months, 1–5 years, 6– 10 years, 11–14, 16–19) for both boys and girls. Volunteers are recruited from varied sources and sites across Canada, including hospital nurseries, day cares, schools and families of hospital volunteers so as to ensure a representative sampling distribution. The campaign has had a good response since launching. Sample collection is performed by specially trained phlebotomists experienced in collecting pediatric specimens. Collected samples are processed (serum separated) before being aliquoted, archived and frozen until analysis. Developed guidelines are applied for tests where on-site analysis is required to preserve specimen stability, to avoid delays in analyte measurement and to achieve accurate results. In Phase II, analyte levels will be measured to establish reference intervals. Analysis will be carried out at specified centres using a number of different instruments commonly used in clinical laboratories across Canada. The goal is to generate
B. Jung, K. Adeli / Clinical Biochemistry 42 (2009) 1589–1595
instrument-specific reference intervals and determine variations in methods across laboratories. Following data collection, reference intervals will be established using parametric and nonparametric methods in collaboration with a clinical biostatistician. Based on detailed gap analyses mentioned earlier, a series of cardiac/metabolic markers, bone markers, genetic metabolic markers, thyroid hormones, the growth hormone-insulin axis, creatinine and C-reactive protein have been selected for the initial phase of reference interval determination. The CALIPER project has recently finished several pilot studies in collaboration with a number of in vitro diagnostics companies, resulting in pediatric reference intervals for numerous analytes [17,19,20]. While these reference intervals were generated using reference values from surplus specimens obtained from children attending outpatient clinics who were deemed to be metabolically stable, data generated from these pilot projects have already benefited the laboratory community. Phase II will determine these pediatric reference intervals again but using specimens collected from healthy volunteers in the community. The final Phase of the project is to implement standardized age-specific reference ranges in various pediatric healthcare centres across Canada. At present, the CALIPER project includes participants from 8 hospital centres across the country. Data generated by each centre is reviewed by the research team to ensure harmonization across the different study centres. Statistical approaches to data analysis and establishment of reference intervals will be in accordance to the CLSI guidelines on reference intervals. Reference intervals will be published on the CALIPER website so that the results of these efforts will not only benefit the practice of pediatric medicine in healthcare centres across Canada but worldwide. In the long-term, the project seeks to improve the health care of children and to promote academic, clinical and industrial collaborations among pediatric laboratory medicine specialists, pediatricians and others caring for children. References [1] Ferré-Masferrer M, Fuentes-Arderiu X, Alvarez-Funes V, Güell-Miró R, Castiñeiras-Lacambra MJ. Multicentric reference values: shared reference limits. Eur J Clin Chem Clin Biochem 1997;35(9):715–8. [2] Ferré-Masferrer M, Fuentes-Arderiu X, Gomà-Llongueras M, et al. Regional reference values for some quantities measured with the ADVIA Centaur analyser. A model of co-operation between the in vitro
[3]
[4]
[5]
[6] [7]
[8] [9]
[10] [11]
[12]
[13]
[14] [15] [16] [17]
[18] [19]
[20]
1595
diagnostic industry and clinical laboratories. Clin Chem Lab Med 2001;39 (2):166–9. Fuentes-Arderiu X, Alonso-Gregorio E, Alvarez-Funes V, et al. Multicentre physiological reference intervals for serum concentrations of immunoglobulins A, G and M, complement C3c and C4 measured with Tina-Quant reagents systems. Clin Chem Lab Med 2007;45(3):387–90. Fuentes-Arderiu X, Mas-Serra R, de-Nó-Lengaran C, et al. Multicentre physiological reference values for some urinary component-to-creatinine (creatininium) concentration ratios. Clin Chem Lab Med 2005;43(9): 958–62. Rustad P, Felding P, Franzson L, et al. The Nordic Reference Interval Project 2000: recommended reference intervals for 25 common biochemical properties. Scand J Clin Lab Invest 2004;64(4):271–84. Ceriotti F, Hinzmann R, Panteghini M. Reference intervals: the way forward. Ann Clin Biochem 2009;46(Pt 1):8–17. Flanders MM, Crist RA, Roberts WL, Rodgers GM. Pediatric reference intervals for seven common coagulation assays. Clin Chem 2005;51(9): 1738–42. Yang L, Grey V. Pediatric reference intervals for bone markers. Clin Biochem 2006;39(6):561–8. Mansoub S, Chan MK, Adeli K. Gap analysis of pediatric reference intervals for risk biomarkers of cardiovascular disease and the metabolic syndrome. Clin Biochem 2006;39(6):569–87. Davis GK, Bamforth F, Sarpal A, Dicke F, Rabi Y, Lyon ME. B-type natriuretic peptide in pediatrics. Clin Biochem 2006;39(6):600–5. Delvin EE, Laxmi Grey V, Vergee Z, CALIPER Working Group. Gap analysis of pediatric reference intervals related to thyroid hormones and the growth hormone-insulin growth factor axis. Clin Biochem 2006;39(6): 588–94. Lepage N, Li D, Kavsak PA, et al. Incomplete pediatric reference intervals for the management of patients with inborn errors of metabolism. Clin Biochem 2006;39(6):595–9. Clinical and Laboratory Standards Institute (CLSI). Defining, establishing, and verifying reference intervals in the clinical laboratory; approved guideline—third edition CLSI document C28-A3; 2008. Dixon WJ. Processing data for outliers. Biometrics 1953;9:74–89. Tukey JW. Exploratory data analysis. Reading, MA: Addison-Wesley; 1977. p. 688. Horn PS, Pesce AJ. Reference intervals. A user's guide. Washington, DC: AACC Press; 2005. Chan M, Seiden-Long I, Aytekin M, et al. Canadian Laboratory Initiative on Pediatric Reference Interval Database (CALIPER): pediatric reference intervals for an integrated clinical chemistry and immunoassay analyzer, Abbott Architect ci8200. Clin Biochem 2009;42(9):885–91. Harris EK, Boyd JC. On dividing reference data into subgroups to produce separate reference ranges. Clin Chem 1990;36(2):265–70. Chan M, Aytekin M, Adeli K. Paediatric reference intervals for 14 chemistries and immunoassays on the Roche Cobas 6000 System. Clin Chem 2008;54 (Suppl. 6):E115. Chan M, Quinn F, Preston N, Ravalico T, Armbruster D, Adeli K. Paediatric reference intervals for 14 chemistries and immunoassays on the Abbott Architect ci8200 System. Clin Chem 2008;54(Suppl. 6):E114.