Vous êtes sur la page 1sur 7

CLB-08550; No.

of pages: 7; 4C:
Clinical Biochemistry xxx (2013) xxx–xxx

Contents lists available at ScienceDirect

Clinical Biochemistry
journal homepage: www.elsevier.com/locate/clinbiochem

1 Validity of establishing pediatric reference intervals based on hospital


2 patient data: A comparison of the modified Hoffmann approach to

F
3 CALIPER reference intervals obtained in healthy children

O
4Q1 Julie L.V. Shaw a,b,1, Ashley Cohen a, Danijela Konforte a,b, Tina Binesh-Marvasti b,
5 David A. Colantonio a,b, Khosrow Adeli a,b,⁎

O
6Q3 a
Clinical Biochemistry Division, Department of Pediatric Laboratory Medicine, The Hospital for Sick Children, University of Toronto, Toronto, ON, Canada
7 b
Department of Laboratory Medicine and Pathobiology, University of Toronto, Toronto, ON, Canada
8

R
a r t i c l e i n f o a b s t r a c t Q4

P
9
10 Article history: Objectives: To compare pediatric reference intervals calculated using hospital-based patient data with those 25
11 Received 11 April 2013 calculated using samples collected from healthy children in the community as part of the CALIPER study. 26
12 Received in revised form 9 November 2013 Methods: Hospital-based data for 13 analytes (calcium, phosphate, iron, ALP, cholesterol, triglycerides, creat-
D 27
13 Accepted 12 November 2013 inine, direct bilirubin, total bilirubin, ALT, AST, albumin and magnesium), measured on the Vitros 5600, collected 28
14 Available online xxxx
between 2007 and 2011 were obtained. The data for each analyte were partitioned by age and gender as previ- 29
16
15
17
E
ously defined by the CALIPER study. Outliers in each partition were removed using the Tukey method. The cumu- 30
18 Keywords:
19 Children
lative distribution function (cdf) was then determined for each analyte value following which, the inverse cdf 31
20 Biochemical markers values of a standard Gaussian distribution were calculated. The analyte values were plotted against the inverse 32
T

21 Reference intervals cdf of the standard Gaussian distribution. Piece-wise regression determined the linear portion of the resulting 33
22 Pediatric graph using the statistical software R. Linear regression determined an equation for the linear portion in each par- 34
C

23 Serum tition and reference intervals were calculated by extrapolating to identify the 2.5th and 97.5th centiles in each 35
24 CALIPER partition based on the inverse cdf values (which would correspond to the values −1.96 and 1.96 of the Gaussian 36
distribution). Using the 90% confidence intervals for the reference intervals defined by CALIPER and the Reference 37
E

Change Value (RCV) as the criteria, these calculated reference intervals were compared to those reported previ- 38
ously by CALIPER. Reference samples were also measured on the Vitros 5600 analyzer in an attempt to validate 39
the calculated reference intervals. 40
R

Results: In general, the reference intervals calculated from hospital-based data were generally wider than 41
those calculated by CALIPER. None of the reference intervals calculated using the Hoffmann approach fell 42
R

completely within the 90% confidence intervals calculated by CALIPER. 43


Conclusions: These results suggest that calculating pediatric reference intervals from hospital-based data 44
may be useful, as a guide, in some cases but will likely not replace the need to establish reference intervals in 45
O

healthy pediatric populations. 46


© 2013 The Canadian Society of Clinical Chemists. Published by Elsevier Inc. All rights reserved. 47
51 49
48
C

50

52 Introduction patients as well as in costly, and often unnecessary, medical follow- 59


N

ups [2]. 60
53 The majority of clinical decisions in medicine are based on laborato- Reference intervals generally consist of a statistically derived range 61
U

54 ry measurements and their associated reference intervals. In pediatric of values denoting the central 95% of values taken from healthy popula- 62
55 medicine there exists a paucity of reliable and accurate reference inter- tion [3]. Reference intervals are of significant importance to modern 63
56 vals for analyte levels in patients [1]. The use of adult reference intervals medicine. Establishing accurate reference intervals is theoretically 64
57 in pediatric medicine is not appropriate and can lead to the under or sound but, practically, relatively challenging. It is difficult to define an 65
58 over-estimation of an analyte level which can result in misdiagnosis of individual as “normal” or “healthy”, ensuring that there are no sub- 66
clinical issues present. Furthermore, differences in analyte values be- 67
Abbreviations: BMI, body mass index; CALIPER, Canadian Laboratory Initiative in tween certain populations and the use of different laboratory methods 68
Pediatric Reference Intervals; CLSI, Clinical Laboratory Standards Institute. by clinical laboratories hamper the use of standard reference intervals, 69
⁎ Corresponding author at: DPLM, Atrium Room 3652, The Hospital for Sick Children, requiring that individual institutions calculate their own intervals. 70
555 University Avenue, Toronto, Ontario M5G 1X8, Canada. Fax: +1 416 813 6257.
The CALIPER (Canadian Laboratory Initiative in Pediatric Reference 71
E-mail address: khosrow.adeli@sickkids.ca (K. Adeli).
1
Current address: Department of Pathology and Laboratory Medicine, The Ottawa Intervals) initiative is a collaborative project between several pediatric 72
Hospital, Ottawa, Ontario, Canada. hospitals across Canada. This initiative aims to update and fill gaps 73

0009-9120/$ – see front matter © 2013 The Canadian Society of Clinical Chemists. Published by Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.clinbiochem.2013.11.008

Please cite this article as: Shaw JLV, et al, Validity of establishing pediatric reference intervals based on hospital patient data: A comparison of the
modified Hoffmann approach..., Clin Biochem (2013), http://dx.doi.org/10.1016/j.clinbiochem.2013.11.008
2 J.L.V. Shaw et al. / Clinical Biochemistry xxx (2013) xxx–xxx

74 that currently exist for pediatric reference intervals. CALIPER has recent-
75 ly published age and gender-specific pediatric reference intervals for 40
76 general chemistry markers. These reference intervals were established
77 through the recruitment of more than 2000 healthy children, aged 0– Table 1 Q2
t1:1
Percentage of patient analyte results used to calculate reference intervals using the t1:2
78 18 years from the community [3]. modified Hoffmann approach. t1:3
79 Establishing reference intervals through recruitment of healthy indi-
80 viduals can be costly and very time consuming. The recruitment of pedi- Analyte Age Gender Number of Number of % results t1:4
patient patient results included
81 atric reference individuals is particularly challenging due to the dynamic results after outlier in final
82 changes occurring with child growth and development. This often results before removal and analysis
83 in the need for age and gender-specific partitioning of reference intervals outlier piece-wise linear
84 which requires a large number of reference samples. An alternative removal regression
85 method for establishing reference intervals was first proposed by biostat- Albumin 0–14 days Both 4730 10 0.2 t1:5
86 istician, Robert G. Hoffmann. In his original 1963 paper, Hoffmann pro- (g/L) 15 days–1 year Both 28784 15 0.1 t1:6
87 posed an indirect a posteriori method in which reference intervals for 1–8 years Both 67472 25 0.0 t1:7
t1:8

F
8–15 years Both 62610 20 0.0
88 analytes could be calculated using hospital in and out-patient data [4]. 15–19 years Female 20464 16 0.1 t1:9
89 The approach proposed by Hoffmann approach requires two assump- 15–19 years Male 19656 29 0.1 t1:10

O
90 tions: 1) that hospital data for a particular analyte forms a Gaussian distri- t1:11
ALP 0–14 days Both 2873 59 2.1 t1:12
91 bution and 2) that the majority of measurements made in the hospital (U/L) 15 days–b1 year Both 15178 147 1.0 t1:13
92 represent normal individuals. Hoffmann began by plotting the cumulative 1–b10 years Both 58593 130 0.2 t1:14

O
93 frequency of a particular result against the analyte value on normal prob- 10–b13 years Both 17217 108 0.6 t1:15
94 ability paper. He then chose the linear portion of the resulting graph, cen- 13–b15 years Female 6647 43 0.6 t1:16
13–b15 years Male 7366 62 0.8 t1:17
95 tered on the 50th percentile, therefore giving the most weight to these

R
15–b17 years Female 7959 34 0.4 t1:18
96 values. By extrapolating the linear portion of the graph, the 2.5th and 15–b17 years Male 8399 0.0 t1:19
97 97.5th centiles could be calculated, representing the range which should 17–b19 years Female 4376 36 0.8 t1:20

P
98 include only apparently healthy individuals, if the assumptions made 17–b19 years Male 4431 40 0.9 t1:21
99 are valid. Hoffmann used this approach with a relatively small number t1:22
ALT 0–1 year Both 48432 21 0.0 t1:23
100 of patient results (n = 60) for glucose as a proof-of-concept. Today, (U/L)
D 1–13 years Both 130177 20 0.0 t1:24
101 with the use of computers, much of the subjectivity of the Hoffmann ap- 13–19 years Female 35159 15 0.0 t1:25
102 proach can be eliminated and very large numbers of samples can be 13–19 years Male 33541 19 0.1 t1:26
t1:27
103 analyzed. AST 0–14 days Both 9967 30 0.3 t1:28
E
104 The obvious advantage to this approach is that it removes the need to (U/L) 15 days–1 year Both 36865 37 0.1 t1:29
105 recruit healthy individuals, instead taking advantage of hospital data 1–7 years Both 67456 20 0.0 t1:30
7–12 years Both 45030 16 0.0 t1:31
T

106 which has already been collected and is readily available. Here, to test 12–19 years Female 38803 12 0.0 t1:32
107 the validity of this approach, we calculated pediatric reference intervals 12–19 years Male 37161 16 0.0 t1:33
C

108 for 13 biochemical markers (albumin, creatinine, ALP, ALT, AST, HDL, cal- t1:34
Calcium 0–b1 year Both 23403 75 0.3 t1:35
109 cium, magnesium, phosphate, iron, cholesterol, triglyceride, unconjugat- (mmol/L) 1–b19 Both 63189 32 0.1 t1:36
110 ed bilirubin) using a modified version of Hoffmann's original method. t1:37
E

111 Calculated reference intervals were compared to those recently published Cholesterol 0–14 days Female N/A t1:38
(mmol/L) 0–14 days Male N/A t1:39
112 by CALIPER [3], as a gold standard. A validation study was also performed 15 days–b1 year Both 691 71 10.3 t1:40
R

113 to definitively assess the feasibility of this approach in a pediatric setting. 1–b19 years Both 21250 231 1.1 t1:41
t1:42
Creatinine 0–14 days Both 18584 27 0.1 t1:43
114 Methods
R

(mmol/L) 15 days–2 years Both 97184 15 0.0 t1:44


2–5 years Both 65637 12 0.0 t1:45
115 Patient data 5–12 years Both 109483 14 0.0 t1:46
O

12–15 years Both 56246 35 0.1 t1:47


15–19 years Female 34407 29 0.1 t1:48
116 Five years (2007–2011) of hospital-based data from children aged
15–19 years Male 33496 41 0.1 t1:49
117 birth to 18 years was requested for the following analytes: albumin, t1:50
C

118 ALP, ALT, AST, total bilirubin, calcium, creatinine, cholesterol, HDL- HDL-C 0–14 days Both N/A t1:51
(mmol/L) 15 days–b1 year Both 171 69 40.4 t1:52
119 cholesterol, iron, magnesium, phosphate and triglycerides. These analytes
1–b4 years Both 485 111 22.9 t1:53
N

120 were all measured on the Vitros 5600 analyzer at the Hospital for Sick 4–b13 years Both 2605 92 3.5 t1:54
121 Children in Toronto. These data were filtered using a unique identifier 13–b19 years Female 3198 59 1.8 t1:55
122 for each patient so that only one result from an individual patient was 13–b19 years Male 1563 31 2.0 t1:56
U

123 used in the analysis. The total number of results obtained, for each analyte, t1:57
Iron 0–b14 years Both 6794 0.0 t1:58
124 from the hospital database is shown in Table 1. The total number of results (mmol/L) 14–b19 years Female 1500 0.0 t1:59
125 used to calculate each reference interval after partitioning, outlier remov- 14–b19 years Male 1042 0.0 t1:60
126 al and piece-wise regression is also listed in Table 1. t1:61
Magnesium 0–14 days Both 11765 26 0.2 t1:62
(mmol/L) 15 days–1 year Both 47119 53 0.1 t1:63
127 Partitioning 1–19 years Both 200627 37 0.0 t1:64
t1:65
Phosphate 0–14 days Both 1516 137 9.0 t1:66
128 The data were partitioned by age and gender based on the partitions (mmol/L) 15 days–b1 year Both 3900 47 1.2 t1:67
129 that were identified by CALIPER [3]. 1–b5 years Both 6321 64 1.0 t1:68
5–b13 years Both 8657 40 0.5 t1:69
13–b16 years Female 3292 58 1.8 t1:70
130 Statistical approach for reference interval calculation
13–b16 years Male 3031 72 2.4 t1:71
16–b19 years Both 3997 59 1.5 t1:72
131 The steps in the approach taken to calculate reference intervals for t1:73
Triglycerides 0–14 days Both 45 0.0 t1:74
132 each analyte are outlined in Fig. 1.

Please cite this article as: Shaw JLV, et al, Validity of establishing pediatric reference intervals based on hospital patient data: A comparison of the
modified Hoffmann approach..., Clin Biochem (2013), http://dx.doi.org/10.1016/j.clinbiochem.2013.11.008
J.L.V. Shaw et al. / Clinical Biochemistry xxx (2013) xxx–xxx 3

133 Outlier removal regression analysis was used to transform reference intervals calculated 160
for measurements made using the Abbott Architect instrument to refer- 161
134 Outliers were removed using the method developed by Tukey [5]. ence intervals suitable for the Vitros 5600. 162
135 Values more than 3 times the inter-quartile range were removed. This Reference serum samples (n = 100) collected from healthy chil- 163
136 method was applied twice. dren as part of the CALIPER study were then analyzed in order to vali- 164
date the transformed reference intervals. According to CLSI guidelines 165
137 Calculation of the cumulative frequency of each result (EP-28-A3), reference interval transformation is considered valid 166
when less than 10% of the measurements from reference samples fall 167
138 The cumulative frequency of each measured analyte value and the outside the calculated reference interval [6]. 168
139 corresponding inverse cdf values for a standard Gaussian distribution
140 were calculated using EXCEL. A plot of the inverse cdf of a standard Calculation of the Reference Change Value (RCV) 169
141 Gaussian distribution against each measured analyte value was plotted
142 using the statistical software, R (http://www.R-project.org) Piece-wise The RCV was calculated from the following equation: RCV = 170
143 linear regression was also performed using R to identify the linear por- √2 ∗ 1.65 ∗ √CVI2 + CVA2, where 1.65 ensures 95% confidence; 171

F
144 tion of the cumulative frequency distribution which crossed the 50th CVI represents the intra-individual biological variation and CVA rep- 172
145 centile (which corresponds to a value of 0 for the inverse cdf of a stan- resents the analytical variation of the method. 173

O
146 dard Gaussian distribution). This subset of analyte values was used to
147 perform linear regression and the resulting linear equation was used Results 174
148 to extrapolate the 2.5th and 97.5th centiles based on the inverse cdf of

O
149 the standard Gaussian distribution as the lower and upper reference in- Calculation of the inverse cdf of a standard Gaussian distribution from the 175
150 terval limits (refer to Tables 2 and 3). cdf of measured values and determination of the linear portion 176

R
151 Transference of reference intervals calculated by CALIPER using the Abbott The cumulative frequency of each measured value for each partition 177
152 platform to reference intervals suitable for the Vitros 5600 was calculated. From the cumulative frequency, the value of the inverse 178

P
cdf of a standard Gaussian distribution was calculated. Each measured 179
153 Reference interval transference was performed according to CLSI analyte value was plotted against the inverse cdf following which, 180
154 guidelines [6]. Two hundred patient serum samples spanning the ana- piece-wise linear regression was used to determine the linear portion
D 181
155 lytical measuring range of each assay were analyzed on both the Abbott of each graph, crossing the 50th centile value. Representative examples 182
156 Architect (the analyzer used for the CALIPER study) and Vitros 5600 of these graphs showing the linear portions identified by piece-wise lin- 183
157 (the analyzer in use at the Hospital for Sick Children) instruments. Lin- ear regression are shown in Figs. 2 and 3 for calcium (both genders ages 184
E
158 ear regression was performed to determine the agreement and relation- birth to 1 year) and ALP (both genders ages 10–13 years), respectively. 185
159 ship between these methods. The linear equation resulting from the The cumulative frequency distribution for calcium is as expected with a 186
T
C
E
R
R
O
C
N
U

Fig. 1. Schematic outline of the process used to calculate reference intervals from hospital-based data in this study.

Please cite this article as: Shaw JLV, et al, Validity of establishing pediatric reference intervals based on hospital patient data: A comparison of the
modified Hoffmann approach..., Clin Biochem (2013), http://dx.doi.org/10.1016/j.clinbiochem.2013.11.008
4 J.L.V. Shaw et al. / Clinical Biochemistry xxx (2013) xxx–xxx

t2:1 Table 2
t2:2 Comparison of reference intervals calculated using the modified Hoffmann approach with those calculated by CALIPER using the 90% confidence intervals calculated by CALIPER.

Analyte Age Gender CALIPER Transformed CALIPER Calculated RI


Lower Upper Lower 90% CI Upper 90% CI Lower Upper
Albumin 0–14 days Both 28 41 32 29 – 35 47 44 – 50 4 32
(g/L) 15 days–<1year Both 25 46 26 23 – 29 50 47 – 53 6 39
1–<8 years Both 35 45 39 36 – 42 50 47 – 53 11 48
8–<15 years Both 37 47 42 39 – 45 51 48 – 54 11 52
15–<19 years Female 35 49 41 38 – 44 52 49 – 55 11 53
15–<19 years Male 38 50 42 39 – 45 55 52 – 58 12 51

ALP 0–14 days Both 90 273 91 74 – 109 256 238 – 274 13 275
(U/L) 15 days – <1 year Both 134 518 131 113 – 149 476 458 – 493 58 382

F
1–<10 years Both 156 369 151 133 – 169 342 324 – 360 67 305
10–<13 years Both 141 460 137 119 – 155 424 406 – 441 49 346

O
13–<15 years Female 62 280 66 48 – 84 262 244 – 280 42 184
13–<15 years Male 127 517 124 106 – 142 474 457 – 492 16 378
15–<17years Female 54 128 59 41 – 77 126 108 – 143 40 124
15–<17 years Male 89 365 91 73 – 109 339 321 – 356 –14 268

O
17–<19 years Female 48 95 54 36 – 72 96 78 – 114 20 126
17–<19 years Male 59 164 64 46 – 81 158 140 – 176 37 160

R
ALT 0–<1 year both 5 51 n/a n/a 52 44 – 61 –1 65
(U/L) 1–<13 years both 11 30 n/a n/a 44 35 – 52 3 61
13–<19 years female 8 24 n/a n/a 40 32 – 49 7 45

P
13–<19 years male 10 33 n/a n/a 42 34 – 51 3 60

AST 0–<14 days Both 32 162 n/a n/a 184 179 – 188 –10 109
(U/L) 15 days–<1 year Both 20 67 n/a D n/a 77 73 – 82 –50 139
1–<7 years Both 21 44 n/a n/a 52 47 – 56 17 64
7–<12 years Both 18 36 n/a n/a 43 38 – 47 14 53
12–<19 years Female 13 26 n/a n/a 31 27 – 36 7 44
E
12–<19 years Male 14 35 n/a n/a 41 37 – 46 6 54

Calcium 0 – < 1year Both 2.13 2.74 2.08 1.99 – 2.16 2.64 2.55 – 2.73 1.98 2.9
T

(mmol/L) 1 – < 19 Both 2.29 2.63 2.22 2.13 – 2.31 2.54 2.45 – 2.63 2.05 2.63

Cholesterol 0–14 days Female 1.2 3.23 1.29 1.08 – 1.49 3.26 3.05 – 3.47 nc nc
C

(mmol/L) 0–14 days Male 1.1 2.82 1.19 0.98 – 1.40 2.86 2.66 – 3.07 nc nc
15d – < 1 year Both 1.66 6.13 1.73 1.53 – 1.94 6.09 5.88 – 6.30 0.73 6.46
1– < 19 years Both 2.9 5.4 2.94 2.73 – 3.15 5.38 5.17 – 5.58 2.34 5.66
E

Creatinine 0–14 days Both 28.5 81.6 30.2 26.1 – 34.5 77.5 73.4 – 81.6 16.0 94.4
R

(µmol/L) 15 days–<2 yrs Both 8.7 32.2 12.5 8.4 – 16.6 33.5 29.4 – 37.6 13.3 38.4
2–<5 years Both 17.6 37.8 20.5 16.4 – 24.6 38.5 34.4 – 42.6 14.3 43.7
5–<12 years Both 27.1 53.5 28.9 24.8 – 33.0 52.5 48.4 – 56.6 18.0 59.7
R

12–<15 years Both 39.8 71.8 40.3 36.2 – 44.3 68.8 64.7 – 72.9 26.4 73.5
15–<19 years Female 43.3 74.1 43.4 39.3 – 47.5 70.8 66.8 – 74.9 28.9 81.7
15–<19 years Male 55.1 95.5 53.9 49.8 – 58.0 89.9 85.8 – 94.0 31.1 96.5
O

HDL–C 0–14 days Both 0.4 1.08 0.49 0.38 – 0.59 1.09 0.98 – 1.20 nc nc
(mmol/L) 15 days – < 1 year Both 0.3 1.85 0.40 0.29 – 0.50 1.78 1.67 – 1.88 0.01 1.91
1 – < 4 years Both 0.84 1.63 0.88 0.77 – 0.98 1.58 1.47 – 1.69 0.31 1.92
C

4–<13 years Both 0.92 1.88 0.95 0.84 – 1.06 1.80 1.70 – 1.91 0.58 2.08
13 – < 19 yrs Female 0.83 1.86 0.87 0.76 – 0.98 1.79 1.68 – 1.89 0.40 2.08
13 – < 19 yrs Male 0.82 1.77 0.86 0.75 – 0.97 1.71 1.60 – 1.81 0.52 1.87
N

Iron 0–<14 years Both 2.8 22.9 5.2 3.1 – 7.1 26.6 24.5 – 28.6 –2.1 26
(µmol/L) 14–<19 years Female 3.5 29 5.9 3.9 – 7.9 33.0 31.0 – 36.1 –2.6 28
U

14–<19 years Male 5.5 30 8.0 6.0 – 10.1 34.1 32.1 – 36.1 –2.8 31

Phosphate 0–14 days Both 1.8 3.4 1.80 1.70 – 1.90 3.31 3.21 – 3.41 1.07 2.91
(mmol/L) 15 days–<1year Both 1.54 2.72 1.56 1.46 – 1.66 2.67 2.57 – 2.77 1.22 2.47
1–<5 years Both 1.38 2.19 1.41 1.31 – 1.51 2.17 2.07 – 2.27 1.1 2.14
5–<13 years Both 1.33 1.92 1.36 1.26 – 1.46 1.92 1.82 – 2.02 1.09 1.98
13–<16 years Female 1.14 1.99 1.07 0.97 – 1.17 1.79 1.69 – 1.89 0.91 1.8
13–<16 years Male 1.02 1.79 1.18 1.08 – 1.28 1.98 1.88 – 2.08 0.97 1.94
16 –<19 years Both 0.95 1.62 1.00 0.90 – 1.10 1.63 1.53 – 1.73 0.87 1.75

Triglycerides 0–14 days Both 0.93 2.93 0.95 0.80 – 1.11 3.01 2.85 – 3.16 nc nc
(mmol/L) 15 d –<1 year Both 0.6 2.92 0.61 0.46 – 0.77 3.00 2.84 – 3.15 0.16 2.58
1–<19 years Both 0.5 2.23 0.51 0.36 – 0.66 2.29 2.13 – 2.44 0.02 1.88

Magesium 0 – 14 days Both 1.99 3.94 Not transferred due to lack of agreement between low correlation
(mg/dL) 15 days – < 1 year Both 1.97 3.09
1 – < 19 years Both 2.09 2.84

Please cite this article as: Shaw JLV, et al, Validity of establishing pediatric reference intervals based on hospital patient data: A comparison of the
modified Hoffmann approach..., Clin Biochem (2013), http://dx.doi.org/10.1016/j.clinbiochem.2013.11.008
J.L.V. Shaw et al. / Clinical Biochemistry xxx (2013) xxx–xxx 5

t3:1 Table 3 CALIPER, were transformed to reference intervals for the Vitros 5600 198
t3:2 Comparison of reference intervals calculated using the modified Hoffmann approach
platform, according to CLSI guidelines. For one of the 13 analytes in 199
t3:3 with those calculated by CALIPER using the Reference Change Value (RCV) for each
t3:4 analyte. this study, magnesium, reference intervals calculated by CALIPER 200
could not be transformed to the Vitros 5600 platform and could not 201
Lower
Transformed RCV Upper RCV be compared. Reference intervals for the following analytes were suc- 202
Analyte Age Gender CALIPER RCV range range Calculated RI cessfully transformed and validated: albumin, ALP, ALT, AST, calcium, 203
Lower Upper Lower Upper
Albumin 0–14 days Both 32 47 5.95 30–34 44–50 4 32 cholesterol, creatinine, HDL, iron, phosphate, total bilirubin and triglyc- 204
(g/L) 15 days–<1year Both 26 50 24–28 47–53 6 39
erides. Comparisons were made using the 90% confidence intervals cal- 205
1–<8 years Both 39 50 37–41 47–53 11 48
8–<15 years Both 42 51 40–44 48–54 11 52 culated by CALIPER for each reference interval (Table 2). Shaded values 206
15–<19 years Female 41 52 39–43 49–55 11 53
15–<19 years Male 42 55 40–44 52–58 12 51
fell within the 90% confidence interval of the reference intervals calcu- 207
lated by CALIPER. 208
ALP 0–14days Both 91 256 22.4 71–111 199–313 13 275
(U/L) 15days –<1 year Both 131 476 102–160 369–583 58 382
1–<10 year Both 151 342 117–185 265–419 67 305
Comparison based on the Reference Change Value (RCV) 209
10–<13 years Both 137 424 106–168 329–519 49 346
13–<15 years Female 66 262 51–81 203–321 42 184 Reference intervals calculated using the Hoffmann approach were 210

F
13–<15 years Male 124 474 96–152 368–580 16 378
15–<17 years Female 59 126 46–72 98–154 40 124
also compared to those calculated by CALIPER, and transformed using 211
15–<17 years Male 91 339 71–111 263–415 –14 268 the Reference Change Value (RCV; Table 3). Lower and upper reference 212

O
17–<19 years Female 54 96 42–66 74–118 20 126
17–<19 years Male 64 158 50–78 123–193 37 160 interval limits shown as shaded fit within the RCV. Shaded values fell 213
within the RCV. 214
ALT 0–<1 year Both 52 51.8 n/a 25–79 65
(U/L) 1–<13 years Both 44 n/a 21–66 61

O
13–<19 years Female 40 n/a 19–61 45
13–<19 years Male 42 n/a 20–64 60
Validation of calculated reference intervals using reference samples 215
Reference samples from healthy children collected as part of the 216
AST 0–<14 days Both 184 35.8 n/a 118–250 109

R
(U/L) 15 days–<1 year Both 77 n/a 50–105 139 CALIPER study were analyzed for levels of albumin, ALP, ALT, AST, calci- 217
1–<7 years Both 52 n/a 33–70 64
um, cholesterol, creatinine, HDL, iron, phosphate, total bilirubin and tri- 218
7–<12 years Both 43 n/a 27–58 53
12–<19 years Female 31 n/a 20–43 44 glycerides using the Vitros 5600 platform. The numbers of samples 219

P
12–<19 years Male 41 n/a 26–56 54
analyzed for each analyte and the number of analyte values falling with- 220
Calcium 0 –<1year Both 2.08 2.64 7.03 1.93–2.22 2.46–2.83 1.98 2.9 in the calculated reference intervals calculated using hospital-based 221
(mmol/L) 1 –<19 years Both 2.22 2.54 2.07–2.38 2.36–2.72 2.05 2.63
patient data are shown in Table 4. The analytes which passed the valida-
D 222
Cholesterol 0–<14 days Female 1.29 3.26 18.5 1.05–1.52 2.66–3.87 nc nc
tion, according to CLSI guidelines are those for which a minimum of 223
(mmol/L) 0–<14 days Male 1.19 2.86 0.97–1.41 2.33–3.39 nc nc
15days –<1year Both 1.73 6.09 1.41–2.05 4.96–7.21 0.73 6.46 twenty reference samples were analyzed and a minimum of 90% of 224
1–<19 years Both 2.94 5.38 2.40–3.49 4.38–6.37 2.34 5.66
the measured values fell within the calculated reference interval (shad- 225
E
Creatinine 0–14 days Both 30.2 77.5 18 24.7–35.6 63.6–91.5 16.0 94.4 ed in black in Table 4). Those analytes for which a minimum of 10% of 226
(µmol/L) 15 days–<2 years Both 12.5 33.5 10.3–14.8 27.4–39.5 13.3 38.4
2–<5 years Both 20.5 38.5 16.8–24.1 31.5–45.4 14.3 43.7 values fell outside the calculated reference interval but for which less 227
T

5–<12 years Both 28.9 52.5 23.7–34.1 43–61.9 18.0 59.7


than twenty reference samples were analyzed are shaded in gray in 228
12–<15 years Both 40.3 68.8 33–47.5 56.4–81.2 26.4 73.5
15–<19 years Female 43.4 70.8 35.6–51.2 58.1–83.6 28.9 81.7 Table 4. 229
C

15–<19 years Male 53.9 89.9 44.2–63.6 73.7–106.1 31.1 96.5

HDL–C 0–<14 days Both 0.49 1.09 24 0.37–0.6 0.83–1.34 nc nc Discussion 230
(mmol/L) 15days –<1 year Both 0.40 1.78 0.3–0.49 1.35–2.20 0.01 1.91
E

1 –< 4 years Both 0.88 1.58 0.67–1.09 1.20–1.96 0.31 1.92


4–<13 years Both 0.95 1.80 0.72–1.18 1.37–2.24 0.58 2.08
Establishing reference intervals for biochemical measurements can 231
13 –< 19 years Female 0.87 1.79 0.66–1.08 1.36–2.21 0.40 2.08
13 –< 19 years Male 0.86 1.71 0.65–1.07 1.30–2.11 0.52 1.87 be challenging as well as expensive. These challenges are amplified in 232
R

Iron 0–<14 years Both 5.2 26.6 74.4 1.3–9.0 6.8–46.3 –2.1 26
pediatrics due to the changes in analyte values that accompany child 233
(µmol/L) 14–<19 years Female 5.9 33.0 1.5–10.3 8.5–57.6 –2.6 28 growth and development. This leads to the need for age and gender 234
14–<19 years Male 8.0 34.1 2.1–14.0 8.7–59.5 –2.8 31
partitioning resulting in a large number of samples being required for 235
R

Phosphate 0–14 days Both 1.80 3.31 24.1 1.37–2.24 2.51–4.11 1.07 2.91 accurate reference interval calculation. Furthermore, collecting samples 236
(mmol/L) 15days–<1year Both 1.56 2.67 1.18–1.93 2.03–3.31 1.22 2.47
1–<5 years Both 1.41 2.17 1.07–1.75 1.65–2.69 1.1 2.14 from healthy children is difficult, adding a further complication. 237
O

5–<13 years Both 1.36 1.92 1.03–1.69 1.45–2.38 1.09 1.98


13–<16 years Male 1.07 1.79 0.81–1.33 1.36–2.23 0.91 1.8
Given these challenges, the idea of calculating accurate reference in- 238
13–<16 years Female 1.18 1.98 0.90–1.47 1.50–2.46 0.97 1.94 tervals from hospital-based patient data is very attractive to laborato- 239
16 –<19 years Both 1.00 1.63 0.76–1.24 1.24–2.03 0.87 1.75
ries. Hoffmann [4] was the first to introduce this concept which has 240
C

Triglycerides 0–14 days Both 0.95 3.01 58.1 0.72–1.18 2.28–3.73 nc nc been adopted and used for several pediatric reference interval studies 241
(mmol/L) 15 days –<1 year Both 0.61 3.00 0.47–0.76 2.28–3.72 0.16 2.58
1–<19 years Both 0.51 2.29 0.39–0.63 1.74–2.84 0.02 1.88 [7–11]. Here, we have used a modified version the Hoffmann approach 242
N

to calculate reference intervals, in which piece-wise linear regression 243


was used to determine the linear portion of the analyte values plotted 244
against a standard Gaussian distribution crossing the 50th centile. This 245
U

approach was used in order to make this selection more objectively. 246
187 relatively even number of samples at both ends of the distribution. The
The limitation of previous studies of this nature has been the inabil- 247
188 cumulative frequency for ALP appears skewed with a larger number of
ity to compare reference intervals calculated using the Hoffmann ap- 248
189 values on the high end.
proach to those calculated from healthy children, as a validation of the 249
Hoffmann approach for pediatrics. In the current study, we overcame 250
190 Comparison of reference intervals calculated by CALIPER to those calculated this limitation by comparing reference intervals calculated using the 251
191 from hospital-based data Hoffmann approach to those calculated in a large population of healthy 252
children as part of the CALIPER study. 253
192 Comparison based on the 90% confidence intervals calculated by CALIPER Here, reference intervals calculated from hospital-based data, using 254
193 for each reference interval the Hoffmann approach, were compared to reference intervals calculat- 255
194 Reference intervals calculated using the Hoffmann approach were ed by CALIPER by several criteria: 1) whether the hospital-based refer- 256
195 compared to the reference intervals recently calculated by CALIPER ence intervals fell within the 90% confidence intervals calculated by 257
196 [3]. Before these comparisons could be made, reference intervals calcu- CALIPER (Table 2), 2) whether the hospital-based reference intervals 258
197 lated from measurements made using the Abbott Architect platform, by fell within the RCV for each analyte (Table 3) and 3) by completing a 259

Please cite this article as: Shaw JLV, et al, Validity of establishing pediatric reference intervals based on hospital patient data: A comparison of the
modified Hoffmann approach..., Clin Biochem (2013), http://dx.doi.org/10.1016/j.clinbiochem.2013.11.008
6 J.L.V. Shaw et al. / Clinical Biochemistry xxx (2013) xxx–xxx

Table 4 t4:1
Validation of the reference intervals calculated using the modified Hoffmann t4:2
approach according to CLSI guidelines. t4:3

Number Number
samples samples % within
Analyte Age Gender analyzed within RI RI
Albumin 0–14 days both NA NA NA
(g/L) 15 days–1yr both NA NA NA
1–8 yrs both 26 24 92%
8–15 yrs both 43 43 100%
15–19 yrs female 15 15 100%
15–19 yrs male 14 14 100%

ALP 0–14d both NA NA NA


(U/L) 15d –<1 yr both NA NA NA

F
1–<10 yr both 39 38 97%
10–<13 yrs both 19 18 95%

O
13–<15 yrs female 7 7 100%
13–<15 yrs male 6 5 83%
15–<17 yrs female 6 5 83%
15–<17 yrs male

O
17–<19 yrs female 9 9 100%
17–<19 yrs male 6 6 100%

R
ALT 0–1 yr both NA NA NA
(U/L) 1–13 yrs both 58 58 100%
13–19 yrs female 22 22 100%

P
Fig. 2. A plot of the inverse cdf of a standard Gaussian distribution against each measured 13–19 yrs male 20 20 100%
calcium value in hospital patients from birth to 1 year of age.
AST 0–14 days both NA NA NA
(U/L)
D 15 days–1 yr both NA NA NA
260 reference interval validation study using reference samples from 1–7 yrs both 24 23 96%
261 healthy children (Table 4). 7–12 yrs both 25 25 100%
262 None of the reference intervals calculated using the Hoffmann ap- 12–19 yrs female 25 25 100%
E
263 proach fell within the 90% confidence intervals calculated by CALIPER. 12–19 yrs male 26 26 100%

264 When the RCV was used for comparison, some of the Hoffmann-
Calcium 0 – < 1yr both NA NA NA
T

265 calculated reference intervals fell within the RCV range calculated (mmol/L) 1 – < 19 both 97 97 100%
266 (creatinine 15 days–b1 year olds and all phosphate partitions with
C

267 the exception of 0–b 14 days old). The RCV is an index used to help iden- Cholesterol 0–14 days female NA NA NA
268 tify whether differing consecutive analyte measurements in a patient (mmol/L) 0–14 days male NA NA NA
15d – < 1 yr both NA NA NA
269 represent significant changes. The calculation takes into account the
E

1– < 19 yrs both 100 91 91%


270 inherent intra-individual biological variation of the analyte in question
271 as well as the analytical variation of the method used to measure the Creatinine 0–14 days both NA NA NA
R

(µmol/L) 15 days–2 yrs both 4 4 100%


2–5 yrs both 13 13 100%
5–12 yrs both 31 31 100%
R

12–15 yrs both 22 22 100%


15–19 yrs female 15 15 100%
15–19 yrs male 14 12 86%
O

HDL–C 0–14 days both NA NA NA


(mmol/L) 15d – < 1 yr both NA NA NA
C

1 – < 4 yrs both 12 12 100%


4–<13 yrs both 44 42 95%
13 – < 19 yrs female 22 21 95%
N

13 – < 19 yrs male 20 19 95%

Iron 0– < 14 yrs both 52 48 92%


U

(µmol/L) 14–< 19 yrs female 20 17 85%


14–< 19 yrs male 19 19 100%

Magnesium 0–14 days both NA NA NA


(mmol/L) 15 days–1 yr both NA NA NA
1–19 yrs both 97 97 100%

Phosphate 0–14d both NA NA NA


(mmol/L) 15d–< 1yr both NA NA NA
1–<5 yr both 14 14 100%
5–< 13 yr both 41 38 93%
13–< 16 yr female 10 9 90%
13–< 16 yr male 10 9 90%
16 – < 19 yr both 22 22 100%

Fig. 3. A plot of the inverse cdf of a standard Gaussian distribution against each measured Triglycerides 0–14 d both NA NA NA
ALP value in hospital patients from 10 to 13 years of age. (mmol/L) 15 d – < 1 yr both NA NA NA

Please cite this article as: Shaw JLV, et al, Validity of establishing pediatric reference intervals based on hospital patient data: A comparison of the
modified Hoffmann approach..., Clin Biochem (2013), http://dx.doi.org/10.1016/j.clinbiochem.2013.11.008
J.L.V. Shaw et al. / Clinical Biochemistry xxx (2013) xxx–xxx 7

273
288
287
286
285
284
283
282
281
280
279
278
277
276
275
274 analyte. The RCV was used to compare Hoffmann-calculated reference have suggested that reliable reference intervals can be calculated 334
289 intervals to those used in clinical practice by Katayev et al. [12] who using the Hoffmann approach when more than 50% of the data are 335
290 calculated reference intervals for HbA1c and T4 in adults using an from ‘healthy’ individuals. In this study a very small percentage of the 336
291 approach similar to the Hoffmann approach. Given the wide biological original data were used to calculate reference intervals, suggesting a 337
292 variation in some analytes, it is not surprising that more of the proportion of ‘sick’ children greater than 50% in the data set. The data 338
293 Hoffmann-calculated reference intervals fell within the RCV than fell in this study included all testing performed at the hospital for 5 years. 339
294 within the 90% confidence intervals calculated by CALIPER. Unfortu- It could be that the calculation of reference intervals from hospital- 340
295 nately biologic variation data are unavailable for pediatrics which may based data may be more reliable if the data are filtered to include only 341
296 limit the use of the RCV as a comparator. patients from outpatient clinics. In future studies, it would also be inter- 342
297 CLSI recommends verification of a reference interval by the mea- esting to calculate reference intervals based on pediatric data from a 343
298 surement of a minimum of 20 reference samples. Verification is con- community lab, where a higher proportion of ‘healthy’ children would 344
299 firmed when less than 10% of the samples fall outside the reference be expected to be included. 345
300 interval being verified. We performed a validation study in an attempt Overall, results from the current study suggest that the use of 346
301 to verify the reference intervals calculated by the Hoffmann approach. hospital-based data to calculate pediatric reference intervals is not ap- 347

F
302 With the exception of ALP (13–b15 year old male; 15–b17 year old propriate, particularly if the data are from a tertiary care pediatric hos- 348
303 female), creatinine (15–b 19 year old male), iron (14–b19 year old fe- pital. It is likely that this approach would produce more reliable 349

O
304 male) and triglycerides, all reference intervals were validated according reference intervals if the data were limited to those from pediatric out- 350
305 to CLSI guidelines. Validation could not be performed for some age and/ patient clinics or from a community-based hospital not treating a large 351
306 or gender-specific partitions due to insufficient numbers of available number of critically ill pediatric patients. 352

O
307 reference samples, particularly samples for very young children and ne-
308 onates. Further to this, in some instances less than twenty reference References 353

R
309 samples were analyzed for a particular age and/or gender-specific par- [1] Adeli K. Closing the gaps in pediatric reference intervals: the CALIPER initiative. Clin 354
310 tition. In most cases, the reference intervals calculated using the modi- Biochem 2011;44:480–2. 355
311 fied Hoffmann approach outlined here were much wider than those [2] Pediatric reference intervals: critical gap analysis and establishment of a national 356

P
312 calculated by CALIPER. This explains why the reference intervals calcu- initiative. Clin Biochem 2006;39:559–60. 357
[3] Colantonio DA, Kyriakopoulou L, Chan MK, Daly CH, Brinc D, Venner AA, et al. Clos- 358
313 lated here were not validated when compared to those calculated and ing the gaps in pediatric laboratory reference intervals: a CALIPER database of 40 359
314 transformed by CALIPER but were validated when measured in refer- biochemical markers in a healthy and multiethnic population of children. Clin 360
361
315 ence samples according to CLSI guidelines. Based on the criteria for ref-
D Chem 2012;58:854–68.
[4] Hoffmann RG. Statistics in the practice of medicine. JAMA 1963;185:864–73. 362
316 erence interval validation outlined by CLSI guidelines, validation of the [5] Tukey J. Exploratory data analysis. Reading, MA: Addison-Wesley; 1977 [688 pp.]. 363
317 reference intervals calculated using the modified Hoffmann approach [6] Defining, establishing and verifying reference intervals in the clinical laboratory; ap- 364
E
318 outlined here would be expected. In light of this, validation according proved guideline. C28-A3. Clinical and Laboratory Standards Institute (CLSI); 2008. 365
[7] Soldin OP, Bierbower LH, Choi JJ, Choi JJ, Thompson-Hoffman S, Soldin SJ. Serum iron, 366
319 to CLSI guidelines, in this case, is likely not the most appropriate meth- ferritin, transferrin, total iron binding capacity, hs-CRP, LDL cholesterol and magne- 367
T

320 od. Comparison of the calculated reference intervals to those calculated sium in children; new reference intervals using the Dade Dimension Clinical Chem- 368
321 by CALIPER, considered the gold standard, is a more reliable method of istry System. Clin Chim Acta 2004;342:211–7. 369
[8] Soldin OP, Hoffman EG, Waring MA, Soldin SJ. Pediatric reference intervals for FSH, 370
C

322 validation. LH, estradiol, T3, free T3, cortisol, and growth hormone on the DPC IMMULITE 371
323 The results of the current study suggest that use of the Hoffmann ap- 1000. Clin Chim Acta 2005;355:205–10. 372
324 proach to calculating reference intervals may be limited in pediatrics, at [9] Soldin SJ, Soldin OP, Boyajian AJ, Taskier MS. Pediatric brain natriuretic peptide and 373
E

N-terminal pro-brain natriuretic peptide reference intervals. Clin Chim Acta 374
325 least when using data from a tertiary care center. This was particularly 2006;366:304–8. 375
326 evident for certain analytes, for example iron, where the lower refer- [10] Soldin OP, Dahlin JR, Gresham EG, King J, Soldin SJ. IMMULITE 2000 age and 376
R

327 ence interval calculated using the modified Hoffmann approach was a sex-specific reference intervals for alpha fetoprotein, homocysteine, insulin, 377
insulin-like growth factor-1, insulin-like growth factor binding protein-3, C- 378
328 negative number. This type of approach may only be useful for certain
peptide, immunoglobulin E and intact parathyroid hormone. Clin Biochem 379
329 analytes and likely depends on the number of “sick” children included
R

2008;41:937–42. 380
330 in the hospital-based data. A higher proportion of ‘sick’ children would [11] Soldin OP, Sharma H, Husted L, Soldin SJ. Pediatric reference intervals for aldosterone, 381
331 be expected to be included in results obtained from a tertiary care pedi- 17alpha-hydroxyprogesterone, dehydroepiandrosterone, testosterone and 25- 382
hydroxy vitamin D3 using tandem mass spectrometry. Clin Biochem 2009;42:823–7. 383
O

332 atric medical center, such as the Hospital for Sick Children, where data [12] Katayev A, Balciza C, Seccombe DW. Establishing reference intervals for clinical lab- 384
333 were obtained for this study. Previous studies by Soldin et al. [10] oratory test results: is there a better way? Am J Clin Pathol 2010;133:180–6. 385
386
C

387
N
U

Please cite this article as: Shaw JLV, et al, Validity of establishing pediatric reference intervals based on hospital patient data: A comparison of the
modified Hoffmann approach..., Clin Biochem (2013), http://dx.doi.org/10.1016/j.clinbiochem.2013.11.008

Vous aimerez peut-être aussi