Clinical Chemistry 45, No.

7, 1999 1087

Pediatric Reference Intervals for Serum Thyroxine, Tri- genital hypothyroidism; postsurgical and postablative
iodothyronine, Thyrotropin, and Free Thyroxine, David hypothyroidism; iodine hypothyroidism; and iatrogenic,
Zurakowski,1* James Di Canzio,1 and Joseph A. Majzoub2 acquired, or unspecified hypothyroidism), hyperthyroid-
(1 Departments of Biostatistics and Medicine and 2 Divi- ism (toxic diffuse goiter; toxic nodular, uninodular, and
sion of Endocrinology, Department of Medicine, Chil- multinodular goiter; thyrotoxicosis from ectopic thyroid
drens Hospital, Harvard Medical School, Boston, MA nodule or other specified origin; thyrotoxicosis with or
02115; * address correspondence to this author at: Depart- without storm; and neonatal thyrotoxicosis), and pituitary
ment of Biostatistics, Childrens Hospital, 300 Longwood disorders (acromegaly and gigantism, other anterior pitu-
Ave., Boston, MA 02115; fax 617-278-9770, e-mail itary hyperfunction, panhypopituitarism, pituitary dwarf-
zurakowski@a1.tch.harvard.edu) ism, diabetes insipidus, disorders of neurohypophysis,
iatrogenic pituitary disorders, syndromes of diencephalo-
Thyroid function tests provide information about hor- hypophyseal origin, and dyspituitarism). Infants less than
mone metabolism and thyroid dysfunction. Reference 1 month of age and premature infants were also excluded
intervals enable clinicians to evaluate thyroid function. from analysis. Results from patients who had multiple
Several pediatric reference intervals for thyroid function measurements on the same thyroid function test (,8% for
tests have been published (1 4 ). Laboratory tests and each analyte) were averaged to obtain a single value for
their nomenclature have been published (5 ), and the each patient in the study. The final sample size consisted
American Thyroid Association has classified thyrotropin of 13 145 test results on 5817 patients (3221 females and
(TSH) as the best single measurement of thyroid status 2596 males) after outliers were removed. The total num-
because of its high sensitivity (6 ). However, reference ber of patients for each thyroid function test was as
intervals that are derived from small numbers of patients follows: T4 (n 5 4551), T3 (n 5 2683), TSH (n 5 5558), and
are not reliable for accurately evaluating test results that free T4 (n 5 353).
are dependent on covariates such as age and sex. The Measurements of T4, T3, and TSH were obtained with
IFCC has recommended a minimum of 120 subjects for the DELFIA immunofluorometric system according to the
nonparametric methods in which subgrouping of data is manufacturers instructions (Wallac Oy). The assay is a
performed (7 ). Virtanen et al. (8 ) have proposed that a solid-phase fluoroimmunoassay that provides a quantita-
smaller sample size may be sufficient for regression-based tive determination of hormone concentrations in human
reference intervals. serum (15). The interassay CV in the euthyroid ranges was
Recently, there has been much interest in using hospital ,10% for T3 and TSH (16) and 11% for T4 and free T4 (17).
databases to extract large volumes of patient data for A variety of statistical techniques have been proposed
clinical research (9 12 ). Hospital databases provide a to obtain reference limits (7, 8, 11, 18 ). Some authors have
sufficient number of subjects for evaluating age and sex recommended a nonparametric approach for removing
differences and for establishing age- and sex-based refer- possible abnormal laboratory test results by discarding
ence intervals. Test results can be affected by medications the top and bottom 10% of the results before calculating
or treatment received by patients for thyroid disease that 2.5% and 97.5% limits on the remaining 80% of the data
alter the physiologic features of the thyroid hormone (19 ). However, this method can distort the shape of the
concentrations during the neonatal period (13, 14 ). There- underlying distribution by cutting off the tails. If outliers
fore, neonates and patients who have thyroid disease or that should be removed are not removed, the resulting
demonstrate abnormal test results should be excluded intervals are too wide and can lead to failure to detect
from the analysis used to establish the reference intervals. patients with thyroid dysfunction. In this study, we
Here we report health-related reference intervals for applied a parametric approach to preserve the shape of
serum thyroxine (T4), triiodothyronine (T3), TSH, and free the distribution. For all four analytes, least-squares regres-
T4 to be used as clinical guidelines for screening patients sion was performed using age as the predictor. Outliers
with suspected thyroid dysfunction. These pediatric were detected by the Tukey interquartile range (IQR)
norms are more accurate than those previously published procedure (20 ) and removed from analysis. Briefly, the
because they are age- and sex-specific and were derived IQR is calculated as the difference between the 75th and
from a large population of children and adolescents. 25th percentiles. Lower and upper limits are calculated as
Between January 1993 and August 1996, test results for the 25th percentile 2 (1.5 3 IQR) and the 75th percentile
T4, T3, TSH, and free T4 were obtained retrospectively 1 (1.5 3 IQR), respectively. This procedure led to the
from outpatient records at Childrens Hospital in Boston removal of 320 outliers (2.4%) before reference intervals
for patients 1 month through 20 years of age. Anonymity were determined. The KolmogorovSmirnov test (21 ) was
of patient data was maintained. Medical record number, used to assess normality of the data for each thyroid test,
date of birth, gender, test code, date of test, test result, and and scatter plots were inspected to evaluate the degree of
International Classification of Diseases (ICD-9) codes skewness. Because of the extreme right-tailed skew of the
were downloaded from the Oracle database on the hos- TSH and free T4 distributions, natural log transformations
pital mainframe computer so that queries could be per- were applied before analysis (22 ). Analysis of covariance
formed using Paradox (Ver. 5.0; Borland International). was used to compare slopes and y-intercepts between
ICD-9 codes were used to select patients for the reference females and males. Because of significant gender differ-
group. Excluded diagnoses were hypothyroidism (con- ences, separate regression equations were calculated for
1088 Technical Briefs

females and males. Residual diagnostics included visual Table 1 presents the pediatric reference intervals, the
checks of normal probability plots and Kolmogorov mean thyroid hormone concentrations, and the number of
Smirnov tests. children analyzed for each age group and sex. Similar
Linear equations were used to predict mean values, numbers of females and males were analyzed at all ages
using the midpoint for each of five age groups: 111 except the 16 20 year category, in which females outnum-
months, 15 years, 6 10 years, 1115 years, and 16 20 bered males. The KolmogorovSmirnov test revealed no
years. Reference intervals were defined as the 95% pre- significant departures from a gaussian distribution for T4
diction intervals around the estimated value correspond- or T3. Serum T4 concentrations decreased significantly
ing to each midpoint. Intervals for TSH and free T4 were with chronological age in both females and males (P
converted back to the original units by calculating the ,0.0001). There was no significant difference in slope
antilog of the logarithmically transformed values. Al- between the sexes for T4 (P 5 0.18), indicating that
though there is a theoretical distinction between predic- females and males shared a common rate of decline.
tion and reference intervals, we used prediction intervals Euthyroid mean values for T4 were estimated to be higher
derived from least-squares regression. This method takes for females than males in the same age group (P ,0.0001).
into account both the residual standard error as well as Earlier studies (1, 24 ) found a similar inverse relationship
the uncertainty of the slope estimate of the regression line. between T4 and age; however, no sex differences were
For large sample sizes, using prediction intervals as reported.
reference intervals is equivalent to the method recom- In this investigation, euthyroid mean values for T3,
mended by Virtanen et al. (23 ). In our study, we have TSH, and free T4 were also inversely correlated with age
results on 5817 children and adolescents. Two-tailed val- (P ,0.0001). These results are consistent with those of
ues of P ,0.05 were considered statistically significant. other investigators who have reported age-related
SAS, Ver. 6.12 (SAS Institute), and SPSS, Ver. 8.0 (SPSS), changes for T3 (25 ) and TSH (26 ). In addition, we found
were used for statistical analysis. significant slope differences between females and males

Table 1. Pediatric reference intervals for T4, T3, TSH, and free T4.
Females Males

Reference Reference
Analyte Age Mean interval n Mean interval n
T4, nmol/La 111 months 122 82162 116 120 79161 135
15 years 120 79160 471 116 75158 589
610 years 115 75154 462 111 69152 600
1115 years 109 69149 799 106 63147 614
1620 years 104 64144 565 99 58142 200
Total 2413 2138

T3, nmol/Lb 111 months 2.46 1.523.39 70 2.46 1.583.35 93

15 years 2.37 1.433.30 262 2.38 1.543.27 340
610 years 2.20 1.623.12 255 2.26 1.373.13 362
1115 years 2.03 1.092.95 483 2.12 1.243.00 341
1620 years 1.84 0.922.78 346 1.98 1.112.86 131
Total 1416 1267

TSH, mIU/L 111 months 2.2 0.86.3 131 2.2 0.86.3 158
15 years 2.0 0.75.9 523 2.1 0.76.0 659
610 years 1.8 0.65.1 562 1.9 0.75.4 698
1115 years 1.5 0.54.4 1057 1.7 0.64.9 738
1620 years 1.3 0.53.9 809 1.6 0.54.4 223
Total 3082 2476
Females and Males
c
Free T4, pmol/L 111 months 19.5 9.539.5 47
15 years 18.4 9.037.2 91
610 years 16.9 8.334.1 57
1115 years 15.5 7.631.5 88
1620 years 14.1 7.028.7 70
Total 353
a
To convert nmol/L to mg/dL, divide by 12.87.
b
To convert nmol/L to ng/dL, multiply by 65.1.
c
To convert pmol/L to ng/dL, divide by 12.87.
 Clinical Chemistry 45, No. 7, 1999 1089

Fig. 1. T4, T3, and TSH data for females and males.
Solid lines represent regression-based upper and lower reference limits.
1090 Technical Briefs

for T3 (P 5 0.022) and TSH (P 5 0.003). Although T3 and producing slightly lower T4 values, especially in males
TSH decreased with age for both sexes, this decline was (29 ). Ideally, reference intervals should be determined
significantly faster for females. Scatter plots showing the using healthy subjects. Because endocrine function tests
empirical data for T4, T3, and TSH with superimposed involve drawing blood, practical and ethical consider-
upper and lower reference limits are provided in Fig. 1. ations require that we rely on the hospital database to
No gender differences were detected for free T4 (P 5 0.57), ensure a sufficient number of subjects. We addressed this
and both sexes demonstrated similar rates of decline with concern by excluding all tests from patients whose ICD-9
age (P 5 0.88). Our reference intervals for free T4 are codes indicated a condition that may have affected their
broader than those reported by others (4 ). thyroid hormone concentrations. Another limitation of
Acquisition of clinical data to establish pediatric refer- this study is that immunofluorometric systems can yield
ence intervals for thyroid function testing is challenging slightly different results than other laboratory assay tech-
because of the difficulty in obtaining blood samples from niques (15 ). However, interassay differences of this mag-
a large number of healthy children. Therefore, the number nitude should not limit the applicability of our reference
of tests used to define pediatric norms is limited. Hospital intervals for most children. We recommend that patients
databases contain large stores of clinical data that can be with borderline values be re-tested and that those with
used to establish reference intervals if the appropriate values falling outside the reference intervals should be
selection criteria are applied (9, 10, 26 ). further evaluated using ultrasensitive TSH measurements.
The large number of patients in this study permitted Our pediatric reference intervals for serum T4, T3, TSH,
evaluation of age and sex as possible factors influencing and free T4 are more accurate than those reported previ-
the concentrations of T4, T3, TSH, and free T4. In our ously because they take into account the significant effects
viewpoint, the determination of age and sex effects is of both age and sex and are based on a very large number
essential for establishing reliable age- and sex-based pe- of children. The other advantage of these intervals is that
diatric reference intervals. Significant age-related declines they are easy to use and provide an improved clinical tool
were found for all three thyroid function tests, suggesting for assessing thyroid function in children and adolescents.
that during childhood, the requirements for these thyroid
hormones decrease. Among postpubertal adolescents,
males exhibited lower serum T4 concentrations and We thank Nader Rifai for valuable comments on this
higher T3 concentrations compared with their female manuscript and Shawn F. OBrien for technical expertise
counterparts. Differences in T4 are consistent with stimu- in database management.
latory and suppressive effects of estrogen and androgen
on thyroid-binding globulin serum concentration (26 ). References
1. Nelson JC, Clark SJ, Borut DL, Tomei RT, Carlton EI. Age-related changes in
Reference intervals for thyroid function tests depend serum free thyroxine during childhood and adolescence. J Pediatr 1993;
somewhat on the particular laboratory methodology 123:899 905.
used. All assays vary in sensitivity (27 ). Our results are in 2. Becker KL, ed. Principles and practice of endocrinology and metabolism,
2nd ed. Philadelphia: JB Lippincott, 1995:1996 8.
fairly close agreement with other published data. For
3. Burtis CA, Ashwood ER, eds. Tietz fundamentals of clinical chemistry, 4th
example, some authors (2 ) provide a T4 reference interval ed. Philadelphia: WB Saunders, 1996:816 9.
of 58 161 nmol/L (4.512.5 mg/dL), encompassing all 4. Soldin SJ, Brugnara C, Gunter KC, Hicks JM. Pediatric reference ranges, 2nd
children. Others (3 ) propose an interval of 100 212 ed. Washington, DC: AACC Press, 1997:139 46.
5. Larsen PR, Alexander NM, Chopra IJ, Hay ID, Hershman JM, Kaplan MM, et
nmol/L (7.8 16.5 mg/dL) for children 4 12 months of al. Revised nomenclature for tests of thyroid hormones and thyroid-related
age, and 82171 nmol/L (6.4 13.3 mg/dL) for children proteins in serum. J Clin Endocrinol Metab 1987;64:1089 94.
510 years of age. Their reference intervals for TSH are 6. Surks MI, Chopra IJ, Mariash CN, Nicoloff JT, Solomon DH. American Thyroid
Association guidelines for use of laboratory tests in thyroid disorders. JAMA
0.4 4.2 mIU/L, which correspond closely to our intervals 1990;263:1529 32.
for older children. Becker (2 ) gives an upper limit for TSH 7. International Federation of Clinical Chemistry Expert Panel on Theory of
of 5.0 mIU/L, and others (4, 11 ) report upper limits of 7.1 Reference Values. Approved recommendation on the theory of reference
values. Part 5. Statistical treatment of collected reference values. Determi-
mIU/L for males and 8.1 mIU/L for females 1 month to 5 nation of reference limits. J Clin Chem Clin Biochem 1987;25:64556.
years of age and 6.0 mIU/L for both sexes 6 18 years of 8. Virtanen A, Kairisto V, Uusipaikka E. Regression-based reference limits:
age. With respect to T3, our reference intervals show a determination of sufficient sample size. Clin Chem 1998;44:2353 8.
9. Harwood SJ, Cole GW. Reference values based on hospital admission
progressive decline with age. Recently published intervals laboratory data. JAMA 1978;240:270 4.
by Soldin et al. (4 ) demonstrate an age effect, although the 10. Kouri T, Kairisto V, Virtanen A, Uusipaikka E, Rajamaki A, Finneman H, et al.
limits are lower and do not show a consistent trend across Reference intervals developed from data for hospitalized patients: comput-
erized method based on combination of laboratory and diagnostic data. Clin
the age groups (28 ). A plausible explanation is the appli- Chem 1994;40:2209 15.
cation of a nonparametric percentile-based technique after 11. Solberg HE. Establishment and use of reference values. In: Burtis CA,
cutting off 20% of the data (10% from each tail). Ashwood ER, eds. Tietz fundamentals of clinical chemistry, 4th ed. Phila-
delphia: WB Saunders, 1996:18291.
A limitation of clinical studies that use hospital data- 12. Iezzoni LI. Assessing quality using administrative data. Ann Intern Med
bases is that the information is retrospective and obtained 1997;127:666 74.
from a population of patients with a variety of illnesses, 13. Erenberg A, Phelps DL, Lam R, Fisher DA. Total and free thyroid hormone
concentrations in the neonatal period. Pediatrics 1974;53:211 6.
some of which may affect the thyroid gland. For example,
14. Surks MI, Sievert R. Drugs and thyroid function. N Engl J Med 1995;333:
patients with inherited deficiency of thyroid-binding 1688 94.
globulin may have been included in the population, 15. Lovgren T, Hemmila I, Pettersson K, Halonen P. Time-resolved fluorometry in
 Clinical Chemistry 45, No. 7, 1999 1091

immunoassays. In: Collins WP, ed. Alternative immunoassays. New York: independent of collection tube type (4 ). Following a
John Wiley & Sons, 1985:20317.
16. Taimela E, Aalto M, Koskinen P, Irjala K. Clinical and laboratory studies of
change in the supplier of collection tubes, however, we
time resolved fluorescence immunoassays of thyrotropin and free triiodo- observed an increase in the frequency of pseudohyperkal-
thyronine. Clin Chem 1993;39:679 82. emia, which prompted a formal evaluation of the serum-
17. Beckett GJ, Wilkinson E, Rae PWH, Gow S, Wu PSC, Toft AD. The utility of a
non-isotopic two-step assay (DELFIA) and an analogue radioimmunoassay plasma difference and the effect of different tube types.
(SimulTRAC) for thyroxine compared. Ann Clin Biochem 1991;28:335 44. Free-flowing blood, taken without a tourniquet to avoid
18. Royston P. Constructing time-specific reference ranges. Stat Med 1991;10: venous stasis, was collected from 40 hospital outpatients.
67590.
19. Soldin SJ, Morales A, Albalos F, Lenherr S, Rifai N. Pediatric reference Paired samples of clotted blood (5 mL) were collected in
ranges on the Abbott IMx for FSH, LH, prolactin, TSH, T4, T3, free T4, free T3, random order into the types of tubes listed in Table 1. One
T-uptake, IgE, and ferritin. Clin Biochem 1995;28:603 6. tube of each pair was selected at random, to prevent
20. Rosner B. Fundamentals of biostatistics, 3rd ed. Belmont, CA: Duxbury
Press, 1989:335. systemic bias, and centrifuged, and the potassium concen-
21. Sokal RR, Rohlf FJ. Biometry, 3rd ed. New York: WH Freeman, 1995:708 tration was determined without delay (time 0). Potassium
14. was measured using an automated discrete analyzer with
22. Draper NR, Smith H. Applied regression analysis, 3rd ed. New York: John
Wiley & Sons, 1998:253 4. an ion-selective electrode (Bayer Diagnostics). The second
23. Virtanen A, Kairisto V, Irjala K, Rajamaki A, Uusipaikka E. Regression-based tube of each pair was left to stand for 3 h at room
reference limits and their reliability: example on hemoglobin during the first temperature before centrifugation and analysis (time13),
year of life. Clin Chem 1998;44:32735.
24. Fisher DA, Sack J, Oddie TH, Pekary AE, Hershman JM, Lam RW, et al. representing a typical delay in sample transport to the
Serum T4, TBG, T3 uptake, T3, reverse T3, and TSH concentrations in laboratory. The analytical imprecision (CV) was ,0.5%
children 1 to 15 years of age. J Clin Endocrinol Metab 1997;45:191 8.
25. Penny R, Spencer CA, Frasier D, Nicoloff JT. Thyroid-stimulating hormone
for both serum and plasma within and between batches
and thyroglobulin levels decrease with chronological age in children and (n 5 30). The significance of the differences between
adolescents. J Clin Endocrinol Metab 1983;56:177 80. serum and plasma was tested using the Student t-test, and
26. Smallridge RC. Thyroid function tests. In: Becker KL, ed. Principles and
practice of endocrinology and metabolism, 2nd ed. Philadelphia: JB Lippin-
the relationship between this difference and the plasma
cott, 1995:299 306. potassium concentration was tested using least-squares
27. Nicoloff JT, Spencer CA. The use and abuse of the sensitive thyrotropin linear regression.
assays. J Clin Endocrinol Metab 1990;71:553 8.
28. Murthy JN, Hicks JM, Soldin SJ. Evaluation of the Technicon Immuno I
The mean plasma potassium concentration was 4.16
Random Access Immunoassay Analyzer and calculation of pediatric refer- mmol/L (SD, 0.33 mmol/L; range, 3.49 4.77 mmol/L).
ence ranges for endocrine tests, T-uptake, and ferritin. Clin Biochem When the blood stood for 3 h before centrifugation
1995;28:1815.
29. Refetoff S. Inherited thyroxine-binding globulin abnormalities in man. Endocr (time13), the mean plasma concentration increased by a
Rev 1989;10:27593. nonsignificant 0.01 mmol/L (95% confidence interval,
20.088 to 0.108 mmol/L). The differences between serum
and plasma at baseline and at (time13) are shown in
Table 1. The mean difference between the serum and
plasma potassium concentrations at baseline was similar
Serum Potassium Is Unreliable as an Estimate of in Vivo for the different tube types (0.324 0.379 mmol/L). How-
Plasma Potassium, Andrew J. Hartland* and Richard H. ever, the variation between patients in the amount of
Neary (Department of Clinical Biochemistry, Central Pa- potassium released during clotting was marked, as shown
thology Laboratory, North Staffordshire Hospital NHS by the wide confidence intervals of the plasma-serum
Trust, Hartshill Rd., Stoke-on-Trent ST4 7PA, UK; * au- difference, which spanned 1.08 mmol/L for one type of
thor for correspondence: fax 44 1782 554646) sample tube (tube E). The 3-h delay in separation of serum
from clot increased this difference, with the 95% confi-
The in vitro release of potassium from cells and platelets dence intervals spanning up to 1.5 mmol/L (tube C),
during blood clotting [particularly in patients with blood although in one type of tube, the mean increased by a
dyscrasias (1, 2 )] increases serum potassium, on average, nonsignificant 0.079 mmol/L (tube D). The serum-plasma
by 0.4 mmol/L (3 ). This difference is considered to be and the baseline to (time13) differences were not related

Table 1. Effect of specimen tube and delay in separation on potassium concentration.

Mean difference (95% confidence interval) in potassium Significance of
concentration from plasma at time 0, mmol/L increase from
baseline to 3-h
Tube Type Time, 0 h Time, 1 3 h sample, P
A Plasma (Greiner) 0.010 (20.09 to 0.11) NSa
B Glass (Labco) 0.324 (0.000.64) 0.469 (0.040.90) 0.0001
C Plastic 1 clotting accelerator 0.370 (20.04 to 0.78) 0.469 (20.32 to 1.28) 0.0002
(Greiner)
D Plastic 1 clotting accelerator 1 gel 0.345 (0.020.67) 0.406 (0.150.67) NS
(Becton Dickinson)
E Plastic 1 clotting accelerator 1 gel 0.326 (20.22 to 0.87) 0.405 (20.01 to 0.87) ,0.0001
(Greiner)
a
NS, not significant.