Pediatric hematology normal ranges derived from pediatric primary care patients

Abstract

To the Editor: Determination of hematological normal ranges presents particular challenges in pediatric patients due to the difficulty in finding normal samples in a hospital setting, and the challenge of obtaining consent and drawing blood in normal infants. We have previously published normal ranges for complete blood counts (CBC) based on inpatients (excluding hematology/oncology and nephrology patients), outpatients and ER patients seen at Boston Children's Hospital, Boston, MA and Children's National Medical Center, Washington DC, respectively.1 Despite removal of outliers with the Chauvenet's Criterion and/or Hoffmann's transformation, our data suffered from an insufficient contribution of samples originating from pediatric primary care providers. Previously published data from our lab were also generated using the Siemens ADVIA 120, an instrument line that is currently in very limited use in the USA. To generate these new pediatric reference ranges values, we obtained CBC data for the period from January 2016 to March 2020 from the Boston Children's Hospital Hematology Laboratory, focusing on outpatient samples from 132 primary care pediatric practices of the Boston metropolitan area, which use our laboratory as their reference laboratory. Since only a small number of samples are available for patients younger than 60 days in these primary care settings, we expanded our criteria. We collected all CBC data (first encounter only) which were obtained in the laboratory for all patients below 60 days of age from January 2016 to March 2020. Samples were collected in K2-EDTA tubes, and no data were collected about preanalytical processes and timing from blood draw to analysis. The CBC and reticulocyte analyses were performed on Sysmex XN-3000 Hematology analyzers (Sysmex Corporation, Kobe, Japan). Study protocol was approved by Boston Children's Hospital IRB (IRB-P00031454). We have also reanalyzed the CBC and reticulocyte data obtained by the Intermountain Health Care group in infants 0 to 90 days old and published before, mostly in graphic form.2-4 These neonatal data were obtained using the Sysmex XE-5000 and XT-4000i analyzers. Upper and lower reference limits to obtain reference ranges were determined using empirical 95% credible intervals with a Hoffmann-based technique, using the middle 95% of the data after excluding the top and bottom 10% of the data.5 The Hoffmann-based trimming technique was used in order to account for the influence of outliers and potential skewness of the data. The top and bottom 10% of the data within each age and sex strata were trimmed and excluded in order to identify the center of the distribution. Then, the data were rank-ordered and the 2.fifth and 97.fifth percentiles were determined as the lower and upper reference limits for the reference ranges for each parameter.6 All statistical analyses and the determination of normal ranges were done using Stata software (version 16.0, StataCorp LLC, College Station, Texas). The approach of determining reference ranges using a Hoffmann-based non-parametric empirical percentile method does not rely upon the parametric assumptions of regression modeling. The standard assumptions required for regression modeling are linearity, no multicollinearity, homoscedasticity (constant variance), independence, and normality. The key assumption of regression analysis in the context of the construction of reference ranges is normality.7 Several statistical methods were implemented to determine whether the data conformed to a Gaussian distribution within each sex and age category. The methods included the Kolmogorov-Smirnov test and Shapiro-Wilk test, the visual inspection of distributions using histograms and quantile-normal plots (Q-norm) and quantile-quantile plots (QQ-plots). These methods showed that the data were skewed, and therefore an empirical percentile-based approach is preferable to determine reference ranges, as this approach does not have any distributional requirements. Furthermore, the empirical method is not sensitive to outliers while the parametric method may be influenced by high-leverage data points. While a transformation of the data, such as a log transformation, may be appropriate to normalize the data, an empirical method using percentiles following Hoffmann trimming was chosen, because a transformation may not adequately normalize the data for all lab parameters measured.8, 9 By utilizing Hoffmann's technique in combination with empirical normal range determination, we also avoid incorrect parameter estimation due to invalid linear assumptions.10 Data were stratified by gender and age category and are presented in supplementary on-line material for both Boston Children's Hospital and Intermountain Health Care. The CBC data were obtained from the pediatric practices for 23 026 samples from patients age 60 days or greater, 8838 from males and 14 188 from females. We collected CBC data from 4729 patients below 60 days of age, 2704 males and 2025 females. The number of CBC samples collected for each variable within each age category and stratified by sex is included in the supplementary on-line data tables. Results are displayed in Table 1 for hemoglobin values, using the empirical normal ranges, with and without Hoffmann trimming, and the regression model-based approach using 95% prediction intervals, where each model included sex and age category as dependent variables. We also include data with a one-sided trimming for the lower part of the range and no trimming for the upper one, give the virtual absence of pathologies resulting in increased Hb values in the pediatric primary care settings. We found that the empirical nonparametric approach with and without Hoffmann-based trimming and the parametric model-based approach with 95% prediction intervals find similar normal ranges. But, the Hoffmann-based technique produces more precise ranges as compared with the empirical approach without Hoffmann-trimming, and as compared with the model-based 95% prediction intervals. Of the three methods, the parametric approach is the most susceptible to the influence of skewness of the data and outlying observations. Based on NHANES data, anemia for children age 12 to 35 months is usually defined by a Hb value <11.0 g/dL, although this threshold does not take into account the lower values observed in African-American children.11 The lower limit of the Hb normal range for these age categories varies in our data set between 10.2-10.7 g/dL in males and 10.3-10.7 g/dL in females, and is substantially closer to the 11 g/dL value than our previous data. For children age 6 years and older, our lower limit of the Hb normal range is at 110 g/L, or slightly above for both males and females (Table 1). Adeli et al. reported in 2015 pediatric reference ranges, ages 3-19 years, for the Beckman Coulter HmX analyzer, using a nonparametric rank method, after removal of outliers.12 In their dataset, the lower limit for Hb is consistently above 11 g/dL for both males and females. This is also the case for the data generated by Fulgoni et al. with the MAXM instrument from Beckman Coulter, using piecewise regression.13 Limitations of this study are the lack of preanalytical details about sample collection and storage, the inability to verify clinical status with medical chart review for each of the collected samples, the lack of ethnicity information, and the small number of outpatient data for children younger than 60 days of age. We have provided in this report numerical values for the Intermountain Health care data which have been previously published and reported in graphical format. These data provide the largest so far hematological characterization of term newborn and preterm infants, which are exceptionally difficult to replicate in other settings. These new pediatric reference ranges for CBC and reticulocytes, derived from pediatric outpatients age 60 days or older more accurately reflect our patient population, and allow us to improve on our definition of normal hematology values for a pediatric population. Dr. Brugnara wishes to disclose a consulting agreement with Sysmex Diagnostics. Dr. Brugnara is the Editor-in-Chief of the American Journal of Hematology. Dr. Brugnara was not involved in the handling of this manuscript; Dr. Ayalew Tefferi acted as Editor-in-Chief for this manuscript. Appendix S1: Supporting Information Appendix S2: Supporting Information Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.

Topics & Concepts