Printer Friendly

Continuous age- and sex-adjusted reference intervals of urinary markers for cerebral creatine deficiency syndromes: a novel approach to the definition of reference intervals.

BACKGROUND: Urinary concentrations of creatine and guanidinoacetic acid divided by creatinine are informative markers for cerebral creatine deficiency syndromes (CDSs). The renal excretion of these substances varies substantially with age and sex, challenging the sensitivity and specificity of postanalytical interpretation.

METHODS: Results from 155 patients with CDS and 12507 reference individuals were contributed by 5 diagnostic laboratories. They were binned into 104 adjacent age intervals and renormalized with Box-Cox transforms ([XI]). Estimates for central tendency ([mu]) and dispersion ([sigma]) ([XI]) of were obtained for each bin. Polynomial regression analysis was used to establish the age dependence of both [mu][log(age)] and [sigma][log(age)]. The regression residuals were then calculated as z-scores = {[XI] - [mu][log(age)]}/ [sigma][log(age)]. The process was iterated until all z-scores outside Tukey fences [+ or -] 3.372 were identified and removed. Continuous percentile charts were then calculated and plotted by retransformation.

RESULTS: Statistically significant and biologically relevant subgroups of z-scores were identified. Significantly higher marker values were seen in females than males, necessitating separate reference intervals in both adolescents and adults. Comparison between our reconstructed reference percentiles and current standard age-matched reference intervals highlights an underlying risk of false-positive and false-negative events at certain ages.

CONCLUSIONS: Disease markers depending strongly on covariates such as age and sex require large numbers of reference individuals to establish peripheral percentiles with sufficient precision. This is feasible only through collaborative data sharing and the use of appropriate statistical methods. Broad application of this approach can be implemented through freely available Web-based software.

Testing for creatine deficiency syndromes (CDSs) (10) is required when patients manifest unexplained intellectual disability, delayed expressive speech and language development, autistic behavior, and seizures. The most common screening test for CDSs concurrently measures the urinary concentrations of creatinine (crn), creatine (CRE), and guanidinoacetic acid (GAA). Increased GAA, increased GAA/crn ratio, and decreased CRE/ GAA ratio are informative markers of guanidinoacetate methyltransferase (GAMT) deficiency. Decreased GAA, decreased GAA/crn ratio, and increased CRE/GAA ratio define the biochemical phenotype of arginine:glycine amidinotransferase (AGAT) deficiency. Finally, increased CRE, increased CRE/crn ratio, and increased CRE/GAA ratio are usually but inconsistently found in males with the X-linked creatine transporter (CRT) defect (1-4). In symptomatic CRT female carriers, these 3 markers are at the high end of the respective reference interval but rarely frankly increased. The normalization by crn attempts to correct for different degrees of urine concentration and relies on the assumption that the tubular excretion and reabsorption of these compounds is similar. However, preliminary evidence suggests that the synthesis of GAA within renal epithelial cells and subsequent transmembrane transport of CRE and GAA are not synchronized (5-8). Urinary crn concentration depends on many factors, among which age and sex are important covariates. After birth, mean urine crn declines for approximately the first month before rising steadily in response to increasing muscle mass and anabolism, reaching a maximum and gradually leveling off at approximately 18-20 years of age. This characteristic age-dependent pattern may be quite different from unrelated endogenous compounds that are excreted in urine. Because a disease marker in urine is often expressed as a ratio to the crn concentration, such a ratio will have its own specific age dependence, which needs to be corrected for when calculating reference and medical decision limits. Reference limits that are age dependent are generally calculated in separated age intervals. This procedure may result in abrupt discontinuities at the interval borders, which is a definite disadvantage. Such discontinuities are frequently seen in pediatric medicine (9-10), although they could be avoided with the use of different types of regression methods (11).

Cohorts containing at least 120 individuals are recommended to establish reference limits with 90% CIs for a variable that exhibits a unimodal and uniform distribution (12). When separated age bins are needed to describe the different age ranges properly, the total number of individuals should be 120 multiplied by the number of bins. This requirement can be somewhat lessened by the use of regression methods, which tie the different bins together, but the number requirement must be further expanded if there are other covariates (e.g., sex) that must be corrected for. Because the cumulative prevalence of CDS is very low, routine results deemed to be uninformative can be used retrospectively for the calculation of reference limits. When several thousand results are available, the central part of the distribution constitutes a fairly adequate reference population after cases identified as outliers have been excluded.

The purpose of the present study was to establish continuous age-corrected reference limits for the CRE/ crn and GAA/crn ratios. To achieve this with sufficient precision, a collaborative effort was initiated among 6 reference laboratories from different geographical sites. Combination of the data in 1 pool was based on finding sufficiently similar medians along the age axis between the different sites.



We collected reference data inclusive of age at collection and sex from 5 different laboratories (range of number of samples contributed was 1880-3287) that used similar analytical methods (see Supplemental Table 1, which accompanies the online version of this article at These clinical specimens had been previously defined as uninformative for CDS and provide the basis for the statistical modeling. An additional laboratory contributed a smaller set (n = 438) of children with autistic spectrum disorder who were tested to rule out a possible CDS (Table 1). This group was processed separately for independent verification of the validity of the proposed model. True positive samples were diagnosed with either AGAT (n = 3), GAMT (n = 32), or CRT deficiency (n = 120). At least 84 of these cases have been reported previously in peer-reviewed publications (13-14), and all were confirmed by combinations of molecular analysis, magnetic resonance spectroscopy, and enzymatic analysis. The specimens included in this study were collected before treatment. All laboratories have participated in external quality programs with satisfactory results.


We partitioned the combined data into equally sized bins (n = 120) and used the Horn algorithm (15) within each bin to establish a simple polynomial relationship, preferably a first-order polynomial, i.e., a straight line, between the marker (X or a transform thereof, [XI]) and its corresponding z-score. By this approach, it is possible to calculate estimates for expected value [mu] and SD [sigma] by means of first-order regression analysis, and any measured value can be converted to a z-score [([XI] - [mu])/a] or its corresponding percentile in the reference population, analogous to the approach commonly used with bone density measurements or growth charts in pediatric practice. Both [mu] and [sigma] typically depend on sex, age, and possibly other covariates but can be expressed as continuous functions of these factors, preferably by use of low-order polynomials where the coefficients are obtained by regression analysis.

After calculating z-scores corrected for the covariates, we removed statistical outliers with Tukey fences. A Tukey factor of 2 excluded samples with z-scores outside [+ or -] 3.372. The measured values were sorted in ascending order: [X.sub.i], i = 1,2 ... n, and each value was assigned a percentile value by the formula [p.sub.i] = (i - 3/8)/(120 + 1/4) (16). We found a corresponding z-score by use of the inverse standard normal distribution. To correct for skewness, all [X.sub.i] values were then subjected to a Box-Cox power transformation:


and an optimal [lambda] was sought to make the relation between [Y.sub.i] values and [z.sub.i]-scores as linear as possible. We use the abbreviated notation "[B - C] " when 2 subpopulations with different [lambda] values are combined within the same bin [see Eqs. (2) and (3)]. Estimated expected value ([mu]) and SD ([sigma]) of Box-Cox transformed marker values within each age bin were determined by the Horn method. Polynomial regression (order [less than or equal to] 3) on transformed parameters [mu] and [sigma] vs log (age) (Briggs logarithm to base 10) established age-continuous curves for these estimates in separate age segments, which could be seamlessly connected at their borders. A flow chart of the statistical estimation procedure is shown in online Supplemental Fig. 1. We used ANOVAs with Student-Newman--Keuls post hoc tests to identify group differences on the basis of z-scores. We then applied the Gowans criterion to statistically significant differences to evaluate biological significance (17). The latter considers a subgroup [absolute value of bias] <0.25 z-score units to be less biologically relevant and may thereby allow combination of subpopulations. In addition, this technique was supported by the application of a partition rule (18). To use combined reference limits, this rule requires that all proportions of the subgroup distributions cut off by the reference limits of the combined distribution should lie in the interval 0.9%-4.1%.

Nonparametric estimation of quantiles was performed by bootstrap resampling. This also allowed for calculation of 90% CIs on the estimates.



After exclusion of outliers, 103 complete bins remained (see details below). It was not possible to find a unique Box-Cox [lambda] for CRE/crn or GAA/crn that rendered the z-score plot rectilinear over the major part of its central course (-1.96 < z-score < 1.96). The curve always appeared kinked. This is illustrated in online Supplemental Fig. 2 for various age cohorts when [lambda] = 0.45. However, it was noted that for values below a fixed threshold value = T, [lambda] = 0.10 gave a rectilinear z-score plot in that interval (data points denoted by blue circles in online Supplemental Fig. 3). The T values were 30 and 8 [micro]mol/ mmol for CRE/crn and GAA/crn, respectively. The z-score regression equation could then be written

[z.sub.i] = a + b x [[zeta].sub.i] + [e.sub.i]; i = 1,2,..., k, (2)

where [[zeta].sub.i] = [B - C; 0.1][X.sub.j], and et represents the regression residual.

Above T, [lambda] = 0.45 resulted in a straight line relationship between z-scores and the Box-Cox transformed values. These data points are denoted by black circles in online Supplemental Fig. 3.

[z.sub.j] = c + d x [[xi].sub.i] + [[epsilon].sub.j]; j = k + 1, k + 2,..., 120, (3)

where [[xi].sub.i] = [B - C; 0.45][X.sub.j], and represents the regression residual.

The transformed values at threshold were t = [B - C; 0.10] T and t = [B - C; 0.45] T, whereas Tappeared to be largely invariant with respect to age, as shown in online Supplemental Fig. 2. The 2 optimal Box-Cox [lambda] values 0.10 and 0.45 were chosen to give a maximal Pearson correlation coefficient r between z-score and transformed marker value within each of the 2 selected intervals. To have the same rectilinear relationship within a bin between z-scores and transformed values above and below T, a first-order transformation (linear shift) of the [[zeta].sub.j] values was applied:

[[psi].sub.i] = [alpha] + [beta] x [[zeta].sub.i]. (4)

To preserve continuity at the threshold, [alpha] is uniquely determined by [beta] according the following equation:

[alpha] + [beta] x [tau] = t, by which [alpha] = t - [beta] x r. (5)

From Eqs. (4) and (5),

[[psi].sub.i] = t + [beta] x ([[zeta].sub.j] - [tau]). (6)

The 2 variables are now connected by a single parameter [beta]. In online Supplemental Fig. 3, [[psi].sub.i] is symbolized by filled red circles. To let [psi] have the same statistical properties as [xi] (same expectation value and same SD), [[psi].sub.i] and [[xi].sub.i] should belong to the same linear regression line of z-score values. Because the upper part of the z-score curve (above the threshold) has the highest number of regression points, those coefficients were determined first, before adjusting the regression equation of the lower part to the upper part:


where [??] and [??] are estimates already determined from regression analysis on Eq. (3). Equation (7) may be rearranged for the determination of [beta]:


The best estimate for [beta] by the method of least squares is then:

[??] = [k.summation over (i = 1)] [W.sub.i] x [[phi].sub.i]/[k.summation over (i = 1)][[phi].sup.2.sub.i]. (9)

Beta as a function of log age is shown in online Supplemental Fig. 4, and the coefficients for the estimated polynomials describing the central course are presented in online Supplemental Table 2. Having estimated a function [beta] that combines transformed values below and above the threshold, [[psi].sub.i] can be calculated from Eq. (6). The final, combined regression equation within a specific age bin is

[z.sub.i] = A + B x [[XI].sub.i] + [[eta].sub.i]; i = 1, 2, 3,..., n, (10)

where [[XI].sub.i] = [[psi].sub.i] for i [less than or equal to] k and [[xi].sub.i] elsewhere; [[eta].sub.i] denotes the residuals.


Equation (10) can now be used to calculate estimates for expectation value:


and SD:


within each bin. As with [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] will be a function of the covariate log age.

Having established expressions for [beta], the parameters A and B in Eq. (10) can be estimated for each age bin. The resulting estimates for [mu] and [sigma] are depicted in Fig. 1 and with polynomial regression applied to adjacent age segments:

[mu] or [sigma] = [D.sub.0] + [[D.sub.1] x log([age.sub.i])] + {[D.sub.2] x [[log([age.sub.i])].sup.2]} + {[D.sub.3] X [[log([age.sub.i])].sup.3]} + [F.sub.i], (13)

where [F.sub.i] is the residual term. Estimates for the regression coefficients are shown in online Supplemental Table 3.

The following formula then converts every measured value X to a z-score:

[z.sub.i] = [[XI].sub.i] - [mu][log([age.sub.i])]/[sigma][log([age.sub.i]) i = 1, 2,...,N (14)

The final z-scores were shown to follow a standard normal distribution, by which the grand mean and SD can be found as well as the corresponding subgroup estimates.

In the data set available for this study, the whole estimation process comprising Eqs. (2)-(l4) was repeated twice for CRE/crn and 4 times for GAA/crn, leaving 2 (0.02%) low and 36 (0.29%) high outliers for CRE/crn and 35 (0.28%) low and 56 (0.45%) high outliers for GAA/crn. Hence, after outlier exclusion, the final values for N in Eq. (14) were 12469 and 12416 for CRE/crn and GAA/crn, respectively.


Visual inspection of mean z-scores from each age bin showed an unequivocal sex difference both for CRE/crn and GAA/crn. Even in early childhood, females had higher values than males. The difference was statistically significant, was augmented in the course of adolescence (>13.5 years of age), and persisted throughout the adult period. The differences are shown in Table 2. Despite statistical significance, the Gowans criterion suggests that combined reference limits might be adequate for both sexes during childhood for both markers. This conclusion is also supported by another partitioning rule that also includes the effect of unequal SDs (18). However, both procedures point to the necessity of correction for sex in adolescents and adults.

Because of significant sex z-score differences with ages >13.5 years and the sex ratio varying significantly between sites in this age range (see Table 1), the effect of site was examined only for ages <13.5 years. The results are shown in Table 3. Despite the statistical significance, the Gowans criterion suggests that correction for level (mean differences) between sites should be unnecessary during the whole childhood for CRE/crn and for GAA/ crn, except site 3. However, it is then obvious that it must be the difference in local SD that violates the partitioning rule for sites 3 and 6 and invalidates site-independent reference limits.

One may correct z-scores for sex and site differences with the following formula:

[z-score.sub.corrected] = (z-score - [DELTA]mean)/SD, (15)

where [DELTA]mean denotes the deviation from the combined mean and SD is the locally obtained SD of z-scores. The grand mean and global SD are approximately equal to 0 and 1, respectively.


The algorithms described above were applied for independent verification to the smaller set (n = 438) of patients with autism spectrum disorder and uninformative results for CDS. The z-score plot for age-corrected [B - C CRE/crn appeared without outliers and was visually indistinguishable from a straight line (data not shown).

The z-score mean and SD were -0.134 and 0.719, respectively. The z-score plot for [B - C] GAA/crn was also rectilinear, but contained 1 low and 1 high outlier (z-scores at -3.730 and 4.059, respectively). The z-score mean and SD were -0.002 and 0.814, respectively.


Conversely theoretical percentile curves can be constructed from Eq. (14) with concentration-specific dependent retransformation formulae. The results are illustrated in Fig. 2. The time course is not described by a single steady rise or fall throughout the age range, as in growth charts in children. Instead, there is a trough for CRE/crn and a peak for both CRE/crn and GAA/crn, and the variation in percentile form necessitates division into several age regions to describe them adequately with a low-powered polynomial (order cubic or less). This irregularity is greater for the median than for the dispersion. It is also interesting to note how patient values are located in relation to the 2.5%-97.5% percentile interval. The lack of patient data before 1 year of age must be viewed in the light of the difficulties of identifying characteristic clinical symptoms in that period. Nonparametric estimates seem also to correspond well with the parametric model (see online Supplemental Fig. 5, in which CIs are also represented).


Reference limits on the basis of broad age bins may be misleading and lead to diagnostic misclassification. The present study took advantage of a large number of reference individuals (n = 12507) from laboratory productions to circumvent the problems of obtaining samples from healthy children. This procedure tends to broaden the reference intervals and may therefore not be ideal. However when the prevalence of the actual disease is low (as here for AGAT, GAMT, and CRT), and when positively identified disease cases are removed together with regression outliers, the central part of the z-score distribution of the remnant cases should constitute a relevant reference population.


The total laboratory production was reasonably distributed in a uniform manner along the transformed age axis. Polynomial regression was used to obtain continuous curves for central tendency and dispersion for the transformed markers in discrete age regions. An analogous regression procedure has been proposed (19). Special care was taken to keep the polynomial order as low as possible and to avoid discontinuous jumps when crossing age limits. Examination of the reference material revealed a peculiar property of the CRE/crn and GAA/crn distributions: it was not possible to linearize the whole course of the z-score curve with a single Box-Cox transformation. Two different lambdas, [lambda] = 0.1 and 0.45, were needed, each in 1 of 2 separate sections of the curve, separated by a fixed threshold value of the marker. A linear transformation was then able to connect the 2 sections so that the major part of the z-score curve could be used in the estimation of its parameters [mu] and [sigma]. To our knowledge, such a phenomenon has not been reported previously. The physiological basis for this behavior could be related to the production, reabsorption, and excretion of GAA and transport of CRE in the renal tubular cells.

The age variation is substantial for both the estimated central tendency ([mu]) and dispersion ([sigma]). The latter factor points to a heteroskedasticity that must be adequately corrected for in clinical practice. It is important to note the opposite course for CRE/crn and GAA/crn in the neonatal period, where the former decreases and the latter increases. Clinical awareness of these properties is essential to early diagnosis.

The present study has led to the recognition of age periods where either false-positive or, more concerning, false-negative events are likely to occur when applying published cutoff values for CRE/crn and GAA/crn, which are flat over broad age ranges (20-21). Review of available CRT cases illustrates the narrow demarcation between the upper reference limits and the lowest patient values after the first year of age. It remains to be determined where patient values are located in the neonatal period and during the first living year, and this will be critical information once an effective treatment becomes available for this disorder (22).

After correction for the continuous covariate age, z-scores may easily be used for group comparisons, e.g., between sexes and sites. It is important to check which statistically significant differences are biologically relevant with respect to reference intervals. Where differences in SD are also present, a partitioning rule (18) should be used. Females had significantly higher values of CRN/crn and GAA/crn than males for all ages. Nevertheless, it is known that symptomatic female carriers of the SLC6A8 defect do not show an increase of the CRE/ crn ratio comparable to that in males. The mechanism behind this is not known, although it is reasonable to speculate that it could be attributed to differences in X-inactivation pattern in females. The reason for the sex difference in childhood needs further study, although differences in muscular mass may contribute to this effect in adolescence and adulthood.

The site-specific differences in z-score mean and SD may be related to instrumentation, standardization of sampling, and population properties. The conversion of values to age-corrected z-scores allows correction for such differences as illustrated in Eq. (15). This also applies to sex effects.

It could be argued that the outliers in the present parametric regression model may actually represent affected cases. Even if that were the case, their exclusion would have been the desired outcome for a process designed to determine reference intervals. Further outcome evaluation of these cases could be undertaken, but its completeness will likely be hampered by the necessity to contact ordering physicians at referring institutions, and in some cases by the length of time passed since the testing was done.

Many biomarkers for inborn errors of metabolism are measured in urine and divided by creatinine. So far, we have not encountered a urinary component other than CRE/crn and GAA/crn (as well as the absolute concentrations of CRE and GAA) that exhibits the peculiar kinked appearance of the z-score plots observed in this study. Amino acids, uric acid, and other purines and pyrimidines apparently do not exhibit such behavior (L. Morkrid, unpublished observations). Other disease markers excreted in the urine should also be thoroughly investigated in large populations for similar properties before they are factored into diagnostic algorithms. In addition to the collection of a sufficient number of reference results, the type of analysis described here is labor intensive and requires extensive statistical experience. For this reason, we have incorporated these complex calculations and the resulting adjustments for age and other covariates in postanalytical interpretive tools for CDS in a Web-based multivariate pattern recognition software called Collaborative Laboratory Integrated Reports (CLIR) (; access is password protected and requires user registration). CLIR is modeled after the Region 4 Stork worldwide collaborative project for laboratory quality improvement of newborn screening by tandem mass spectrometry (23-25). Like the newborn screening cluster of applications, our objective is to make them freely available to all interested users willing to share data and extend these applications to cover a broad spectrum of either clinical or research endeavors. This evidence-based approach could add substantial value to patient care by providing an environment conducive to standardization through normalization of z-scores, leading to enhanced interpretation of complex laboratory profiles that is driven by multisite evidence and objective peer comparison.

Author Contributions: All authors confirmed they have contributed to the intellectual content of this paper and have met the following 3 requirements: (a) significant contribution to the conception and design, acquisition of data, or analysis and interpretation of data; (b) drafting or revising the article for intellectual content; and (c) final approval of the published article.

Authors' Disclosures or Potential Conflicts of Interest: Upon manuscript submission, all authors completed the author disclosure form. Disclosures and/or potential conflicts of interest:

Employment or Leadership: P.L. Hall, Emory University; G.S. Salomons, Society of Inborn Errors of Metabolism.

Consultant or Advisory Role: None declared.

Stock Ownership: None declared.

Honoraria: None declared.

Research Funding: P. Rinaldo, the T. Denny Sanford professorship fund, Mayo Clinic.

Expert Testimony: None declared.

Patents: None declared.

Role of Sponsor: The funding organizations played no role in the design of study, choice of enrolled patients, review and interpretation of data, or preparation or approval of manuscript.

Received November17, 2014; accepted February 12, 2015. Previously published online at DOI: 10.1373/clinchem.2014.235564


(1.) Valongo C, Cardoso ML, Domingues P, Almeida L, Verhoeven N, Salomons G, et al. Age related reference values for urine creatine and guanidinoacetic acid concentration in children and adolescents by gas chromatography-mass spectrometry. Clin Chim Acta 2004;348:155-61.

(2.) Arias A, Ormazabal A, Moreno J, Gonzalez B, Vilaseca MA, Garda-VilloriaJ, et al. Methodsforthe diagnosis of creatine deficiency syndromes: a comparative study. J Neurosci Methods 2006;156:305-9.

(3.) Mercimek-Mahmutoglu S, Stockler-Ipsiroglu S, Salomons GS. Creatine deficiency syndromes. In: Pagon RA, Adam MP, Ardinger HH, Bird TD, Dolan CR, Fong CT, Smith RJH, Stephens K, editors. Gene Reviews[R]. Seattle, WA: University of Washington; 1993-2014.

(4.) van de Kamp JM, Mancini GM, Salomons GS.X-linked creatine transporter deficiency: clinical aspects and pathophysiology. J Inherit Metab Dis 2014;37:715-33.

(5.) Sims EA, Seldin DW. Reabsorption of creatine and guanidinoacetic acid by the renal tubules. Am J Physiol 1949;157:14-20.

(6.) Kadono K, Yamaguchi T, Tetsutani T, Yasunaga K. Clinical test of renal guanidinoacetic acid metabolism by oral citrulline and creatine loading. Clin Chim Acta 1992; 209:9-17.

(7.) Edison EE, Brosnan ME, Meyer C, Brosnan JT. Creatine synthesis: production of guanidinoacetate by the rat and human kidney in vivo. Am J Physiol Renal Physiol 2007;293:F1799-804.

(8.) Reznichenko A, Sinkeler SJ, Snieder H, van den Born J, de Borst MH, Damman J et al. SLC22A2 is associated with tubular creatinine secretion and bias of estimated GFR in renal transplantation. Physiol Genomics 2013; 45:201-9.

(9.) Hilsted L Rustad P, Aksgtede L, Sorensen K, Juul A. Recommended Nordic paediatric reference intervals for 21 common biochemical properties. Scand J Clin Lab Invest 2013;73:1-9.

(10.) Colantonio DA, Kyriakopoulou L, Chan MK, Daly CH, Brinc D, Venner AA, et al. Closing the gap in pediatric laboratory reference intervals: a CALIPER database of 40 biochemical markers in a healthy and multiethnic population of children. Clin Chem 2012;58:854-68.

(11.) Koenker R, Hallock KF. Quantile regression. J Econom Persp 2001;15:143-56.

(12.) Solberg HE. The IFCC recommendation on estimation of reference intervals. The RefVal program. Clin Chem Lab Med 2004;42:710-4.

(13.) van de Kamp JM, Betsalel OT, Mercimek-Mahmutoglu S, Abulhoul L, Grunewald S et al. Phenotype and genotype in 101 males with X-linked creatine transporter deficiency. J Med Genet 2013;50:463-72.

(14.) Comeaux MS, Wang J, Wang G, Kleppe S, Zhang VW, Schmitt ES et al. Biochemical, molecular, and clinical diagnoses of patients with cerebral creatine deficiency syndromes. Mol Gen Metab 2013;109:260-8.

(15.) Horn PS, Feng L, Li Y, Pesce AJ. Effect of outliers and nonhealthy individuals on reference interval estimation. Clin Chem 2001;47:2137-45.

(16.) Altman DG. Practical statistics for medical research. London: Chapman & Hall; 2006.

(17.) Gowans EM, Pedersen PH, Blaaberg O, Horder M. Analytical goals for the acceptance of common reference intervals for laboratories throughout a geographical area. Scand J Clin Invest 1988;48:757-64.

(18.) Lahti A, Petersen PH, Boyd JC, Rustad P, Laake P, Solberg HE. Partioning of nongaussian-distributed biochemical reference data into subgroups. Clin Chem 2004;50:891-900.

(19.) Cole TJ. Using the LMS method to measure skewness in the NCHC and Dutch national height standards. Ann Hum Biol. 1989;16:407-19.

(20.) Haas D, Gan-Schreier H, Langhans CD, Anninos A, Haege G, Burgard P et al. Diagnosis and therapeutic monitoring of inborn errors of creatine metabolism and transport using liquid chromatography-tandem mass spectrometry in urine, plasma and CSF. Gene 2014;538:188-94.

(21.) Joncquel-Chevalier Curt M, Cheillan D, Briand G, Salomons GS, Mention-Muliez K, Dobbelaere D et al. Creatine and guanidinoacetate reference values in a French population. Mol Genet Metab 2013;110:263-7.

(22.) Dunbar M, Jaggumantri S, Sargent M, Stockler-Ipsiroglu S, van Karnebeek CD. Treatment of X-linked creatine transporter (SLC6A8)deficiency: systematic review of the literature and three new cases. Mol Genet Metab 2014;112:259-74.

(23.) McHugh DMS, Cameron CA, Abdenur JE, Abdulrahman M, Adair O, Al Nuaimi SA et al. Clinical validation of cutoff target ranges in newborn screening of metabolic disorders by tandem mass spectrometry: a worldwide collaborative project. Genet Med 2011;13:230-54.

(24.) Marquardt G, Currier R, McHugh DM, Gavrilov D, Magera MJ, et al. Enhanced interpretation of newborn screening results without analyte cutoff values. Genet Med 2012;14:648-54.

(25.) Hall PL, Marquardt G, McHugh DM, Currier RJ, Tang H, Stoway SD, Rinaldo P. Post-analytical tools improve performance of newborn screening by tandem mass spectrometry. Genet Med 2014;12:889-95.

Lars M0rkrid, (1) * Alexander D. Rowe, (1) Katja B.P. Elgstoen, (1) Jess H. Olesen, (2) George Ruijter, (3) Patricia L. Hall, (4) Silvia Tortorelli, (5) Andreas Schulze, (6) Lianna Kyriakopoulou, (7) Mirjam M.C. Wamelink, (8) Jiddeke M. van de Kamp, (9) Gajja S. Salomons, (8) Piero Rinaldo (5)

(1) Department of Medical Biochemistry, Oslo University Hospital, Rikshospitalet, Oslo, Norway; (2) Rigshospitalet, University of Copenhagen, Copenhagen, Denmark; (3) Department of Clinical Genetics, Erasmus Medical Center, Rotterdam, The Netherlands; (4) Department of Human Genetics, Emory University, Atlanta, GA; (5) Department of Laboratory Medicine and Pathology, Mayo Clinic, Rochester, MN; (6) Division of Clinical and Metabolic Genetics and (7) Division of Clinical Biochemistry, Hospital for Sick Children, University of Toronto, Toronto, Ontario, Canada; (8) Department of Clinical Chemistry and (9) Department of Clinical Genetics, VU University Medical Center, Amsterdam, The Netherlands.

* Address correspondence to this author at: Department of Medical Biochemistry, Oslo University Hospital, Rikshospitalet, Institute of Clinical Chemistry, PO Box 4950 Ny dalen, Oslo, 0424, Norway. Fax +47-2-307-0902; e-mail

Part of this work has been presented at the 2014 annual symposium of the Society for the Study of Inborn Errors of Metabolism (SSIEM), Innsbruk, Austria, September 2-5, 2014.

(10) Nonstandard abbreviations: CDS, cerebral creatine deficiency syndrome; crn, creatinine; CRE, creatine; GAA, guanidinoacetic acid; GAMT, guanidinoacetate methyltransferase; AGAT, arginine:glycine amidinotransferase; CRT, creatine transporter; CLIR, Collaborative Laboratory Integrated Reports.
Table 1. Age distributions of reference individuals provided by
different sites.

                                        Age range
                           Months       Years

Site                 0-3   3-6   6-12   1-4    4-10   10-18   18-40
i                    82    55    i48    1063   1223   458     178
2                    502   i58   211    558    424    272     40
3                    39    29    77     509    510    269     158
4                    288   i06   176    585    642    363     231
5                    56    67    164    707    944    549     101
Subtotal             967   4i5   776    3422   3743   1911    708
6 (verification)     1     0     0      113    219    104     1
Total                968   4i5   776    3535   3962   2015    709
Male prevalence, %   60    55    56     65     70     64      58

                     Age range

Site                 40-55   55-70   >70   Total
i                    62      13      5     3287
2                    24      7       3     2199
3                    102     95      92    1880
4                    89      37      2     2519
5                    20      13      1     2622
Subtotal             297     165     103   12507
6 (verification)     0       0       0     438
Total                297     165     103   12945
Male prevalence, %   54      53      49    64

Table 2. Combined and sex-specific estimates of z-scores for Box-Cox
transformed marker values. (a)

Age, years           Both     Female   Male

CRE/crn ratio
  n                  10374    3679     6695
  Mean               0.012    0.102    -0.037
  SD                 0.991    0.960    1.005
  Less than 2.5th             2.0      2.8
    percentile (b)
  More than 97.5th            2.4      2.6
    percentile (b)
  n                  2095     880      1215
  Mean               -0.035   0.244    -0.237
  SD                 0.996    0.934    0.992
  Less than 2.5th             1.0      4.4
    percentile (b)
  More than 97.5th            3.0      1.8
    percentile (b)
GAA/crn ratio
  n                  10343    3664     6679
  Mean               0.006    0.147    -0.071
  SD                 0.988    0.987    0.981
  Less than 2.5th             2.2      2.7
    percentile (b)
  More than 97.5th            1.9      2.3
    percentile (b)
  n                  2073     868      1205
  Mean               0.003    0.232    -0.161
  SD                 1.028    1.090    0.947
  Less than 2.5th             2.8      2.2
    percentile (b)
  More than 97.5th            5.8      1.9
    percentile (b)

Age, years           Sex          t        P

CRE/crn ratio
  Mean               0.139        6.89     6.0 x [10.sup.-12]
  Less than 2.5th
    percentile (b)
  More than 97.5th
    percentile (b)
  Mean               0.481        11.23    1.9 x [10.sup.-28]
  Less than 2.5th
    percentile (b)
  More than 97.5th
    percentile (b)
GAA/crn ratio
  Mean               0.218        10.766   7.8 x [10.sup.-17]
  Less than 2.5th
    percentile (b)
  More than 97.5th
    percentile (b)
  Mean               0.393        8.540    2.9 x [10.sup.-17]
  Less than 2.5th
    percentile (b)
  More than 97.5th
    percentile (b)

(a) Values in bold exceed the bias criteria.

(b) Portions in subpopulations lying outside the combined reference
limits (%).

Table 3. Combined and site-specific estimates of z-score distribution
for transformed values (age <13.5 years).3

Ratio                       All     Site 1   Site 2   Site 3   Site 4

n                           10374   2842     1989     1290     2258
Mean                        0.000   -0.027   0.058    0.121    -0.030
SD                          1.000   0.940    1.048    1.211    0.893
Lessthan combined 2.5th     2.50    1.70     3.60     6.20     1.10
  percentile (c)
More than combined 97.5th   2.50    2.10     2.80     4.90     1.50
  percentile (c)
n                           10343   2831     1982     1289     2256
Mean                        0.000   -0.070   -0.065   0.292    -0.006
SD                          1.000   0.954    0.998    1.064    0.941
Less than combined 2.5th    2.50    2.50     3.70     2.20     1.60
  percentile (c)
More than combined 97.5th   2.50    2.50     1.70     5.60     2.20
  percentile (c)

Ratio                       Site 5   Fb       Pb

n                           1995
Mean                        0.001    7.125    1.00 x [10.sup.-5]
SD                          0.943
Lessthan combined 2.5th     1.80
  percentile (c)
More than combined 97.5th   2.40
  percentile (c)
n                           1985
Mean                        0.015    34.322   1.67 x [10.sup.-28]
SD                          0.996
Less than combined 2.5th    2.60
  percentile (c)
More than combined 97.5th   2.40
  percentile (c)

Ratio                       Site 6

n                           394
Mean                        -0.134
SD                          0.719
Lessthan combined 2.5th     0.50
  percentile (c)
More than combined 97.5th   0.00
  percentile (c)
n                           392
Mean                        -0.002
SD                          0.814
Less than combined 2.5th    0.80
  percentile (c)
More than combined 97.5th   0.80
  percentile (c)

(a) Values in bold indicate sites that require either separate
reference limits or a site-specific adjustment.

(b) Site group comparison (ANOVA).

(c) Portions in subpopulations lying outside the combined reference
limits (%).
COPYRIGHT 2015 American Association for Clinical Chemistry, Inc.
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2015 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Title Annotation:Informatics and Statistics
Author:Morkrid, Lars; Rowe, Alexander D.; Elgstoen, Katja B.P.; Olesen, Jess H.; Ruijter, George; Hall, Pat
Publication:Clinical Chemistry
Article Type:Report
Geographic Code:4EUNE
Date:May 1, 2015
Previous Article:Investigation of 2 models to set and evaluate quality targets for Hb [A.sub.1c]: biological variation and sigma-metrics.
Next Article:Xenografts as models of clonal selection and acquired resistance to therapy.

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters