OBJECTIVES: Higher than expected small for gestational age (SGA) rates and lower than expected large for gestational age (LGA) rates have been observed. A possible explanation is a leftward shift of percentile curves for birth weight due to a systematic error in plotting birth weight values in charts (ie, plotting weekly mean birth weight data at the beginning of the weeks). Our objectives were to assess how common this plotting error is and to analyze the effect of this error on SGA and LGA classification based on data from the German perinatal survey.
METHODS: First, a systematic literature search for birth weight charts was performed, and the charts were analyzed for the plotting error. Second, percentile values (10th, 50th, and 90th) for 25 to 42 completed weeks of gestation were calculated from the data of 1 181 200 male singleton newborns (German perinatal survey, 1995–2000). Birth weight percentile curves were calculated with and without the plotting error, and the resulting SGA and LGA rates were analyzed.
RESULTS: Fourteen of the 16 identified publications contained the systematic error in plotting. Using our calculated percentile curves, a leftward shift caused by the plotting error led to an SGA rate of 12.5% and an LGA rate of 7.7%; ∼5% of newborns were misclassified.
CONCLUSIONS: Percentile charts should be examined for the described systematic error and, if necessary, corrected.
- LGA —
- large for gestational age
- SGA —
- small for gestational age
What’s Known on This Subject:
Percentile charts for birth weight are used to assess the somatic development of neonates (small, appropriate, or large for gestational age).
What This Study Adds:
A systematic error was identified in the majority of birth weight percentile charts. As a consequence, small for gestational age rates are overestimated and large for gestational age rates are underestimated; ∼5% of neonates are misclassified.
Newborn classification according to birth weight is used to identify small and large for gestational age (SGA and LGA, respectively) infants because such infants are at a higher risk for postnatal complications. The 10th and 90th percentiles are used as cutoff values for SGA and LGA classification, which implies that the clinically observed SGA and LGA rates should each be ∼10%. However, in practice, we have observed an SGA rate >10% and an LGA rate <10%. Similarly, a recent neonatal survey reported an SGA rate of 16.1%.1
A systematic error may explain this variation in SGA and LGA rates. Percentile values of birth weight for gestational age are calculated for completed weeks. We observed that, instead of plotting these weekly average percentile values for birth weight at the mean gestational age for the completed week in question, values are commonly plotted at the start of the week (Fig 1). This leads to curves that are shifted leftward along the x-axis (gestational age). Thus, the cutoff values are higher, leading to increased rates of SGA and lower rates of LGA neonates.
We aimed, first, to analyze how common this systematic error in the plotting of birth weight percentile values is and, second, to assess its effect on the SGA and LGA classification of newborns based on data from the German perinatal survey.
Medline (via Ovid), PubMed, and Web of Science were searched for full-text articles published between January 1995 and November 2011 by using the search terms “birth weight percentile,” “birth weight chart,” “growth chart,” and “growth standard.” The references of the retrieved articles were screened for further relevant papers; publications citing the retrieved articles were also assessed (for the Web of Science search). Publications written in English or German that presented birth weight percentile values according to gestational age were included. For inclusion, papers needed to report on infants born in 1980 or later. The publications also needed to present the data in both numerical form and chart form. Alternatively, it was also acceptable if numerical values and charts for the same data set could be obtained from separate publications.
In these papers, the methods used to plot the birth weight percentile values were examined. The systematic error was considered to be present when the birth weight percentile values from the numerical data were not plotted at the average gestational age of each completed week but rather at the beginning of the week.
Analysis Based on Data From the German Perinatal Survey
Day-specific birth weight data of 1 181 200 male singleton infants with gestational ages of 25 to 42 completed weeks were obtained from the German perinatal survey of 1995–1997 (when all German federal states except Baden-Württemberg had kindly contributed data to our database) and of 1998–2000 (when the following German federal states had contributed data: Bavaria, Brandenburg, Hamburg, Lower Saxony, Mecklenburg-Western Pomerania, Saxony, Saxony-Anhalt, and Thuringia).
Weekly mean values for each completed week of gestation were calculated from the day-specific birth weight data. Percentiles (10th, 50th, and 90th) were then computed with the SPSS EXAMINE command by using the HAVERAGE distribution function (SPSS Inc, Chicago, IL). Percentile curves were constructed by using the calculated mean birth weight percentile values for each completed week as the y-coordinate and as the x-coordinate (Fig 2), the following: scenario 1, the mean gestational age (for that week of gestation); or scenario 2, the start of the completed week of gestation in question.
For these 2 scenarios, percentile values specified by the day were computed from the weekly data for the 10th, 50th, and 90th percentiles of birth weight. Calculations were performed for the age range 25 weeks + 0 days (175 days) to 42 weeks + 6 days (300 days) by using the cubic spline method. The sample infant population from the German perinatal survey was then classified by computing day-specific 10th and 90th percentile values generated for the aforementioned scenarios. For an assessment of the effect of the plotting error, SGA and LGA rates obtained by using the 10th and 90th percentile values generated by using the 2 scenarios were compared. The differences in birth weight between the percentiles that were generated by using scenarios 1 and 2 were then calculated for each week of gestation.
The search generated 832 Medline (via Ovid), 868 PubMed, and 59 Web of Science hits. Sixteen publications met our inclusion criteria.2–17 The plotting error was identified in 14 of these 16 publications.2–15
Analysis Based on Data From the German Perinatal Survey
In this simulation, percentile curves of birth weight for gestational age with the systematic plotting error were shifted leftward compared with the percentile curves constructed “correctly” with the mean weekly birth weight plotted at the mean gestational age for each completed week. The average leftward shift of the percentile curves was 0.43 weeks (∼3 days).
Newborn classification using the percentile curves with the systematic plotting error yielded an SGA rate of 12.5% and an LGA rate of 7.7% (Table 1). In comparison, when using the curves generated correctly, the SGA rate was 9.7% and the LGA rate was 9.9%. Our analysis therefore revealed that 2.8% of the newborn population who were really appropriate for gestational age were incorrectly classified as SGA and 2.2% of the population who were really LGA were incorrectly classified as appropriate for gestational age.
The 10th, 50th, and 90th percentile curves of birth weight constructed with and without the systematic error differed by up to 139 g (Table 1).
In the current study, we have identified a systematic plotting error in percentile charts of birth weight for gestational age. In our analysis, plotting weekly percentile values of birth weight at the start of the completed week of gestation resulted in a leftward shift of cutoff values, causing ∼5% of newborns to be misclassified. The percentile curves were displaced to the left by an average of ∼3 days, causing the curves to indicate a higher birth weight reference value for a given age. Leftward shifted percentile curves result in the overestimation of SGA rates and underestimation of LGA rates.
In our simulated birth weight percentile charts, percentile curves that were computed without the systematic error resulted in SGA and LGA rates of 9.7% and 9.9%, respectively, and were close to the expected 10% rate. The slightly lower than expected rates may be attributed to the definition of SGA and LGA to not include infants whose birth weights fall directly on the computed 10th and 90th percentile curves.
Classifying infants according to their birth weight by using completed weeks of gestation decreases the precision of classification to a weekly period. Voigt et al17 demonstrated that with percentile values based on weekly data rather than daily data and by using tabulated values (not charts), SGA rates were overestimated and LGA rates were underestimated on the first days of every week. Conversely, SGA rates were underestimated and LGA rates overestimated on the last days of every week. This is because infants born at the beginning of a week have lower birth weights compared with infants born at the end of the same week. The effect on SGA and LGA rates described there differs from the phenomenon described in the current paper. In the study by Voigt et al, the overestimation and underestimation occurred when tabulated percentile values for completed weeks of gestation were used to classify newborns and were due to using the same average weekly value for every day of the week. This does not occur in plotted growth charts because each day corresponds to a different point on the birth weight percentile curve.
Newborn classification is used in clinical decision-making, rendering the systematic plotting error described here an issue of concern. SGA and LGA infants are at an increased risk for perinatal morbidity, associated health problems (such as neurodevelopmental disorders), and metabolic alterations in later life. Meaningful SGA and LGA classification requires the following: (1) accurate knowledge of gestational age; (2) accurate measurement at birth of weight, length, or head circumference; and (3) cutoff values based on reference data from a relevant population.18
Because the described systematic plotting error causes a horizontal shift of the percentiles, the cutoff values for identifying SGA and LGA infants are affected, and newborn classification may be incorrect. Percentile charts used in clinical practice should be examined for the described error to ensure that weekly percentile values are plotted correctly (ie, at the mean gestational age of each completed week).
We are grateful to the German federal states that contributed data to our perinatal database and to the clinicians who were involved in data collection.
- Accepted April 16, 2012.
- Address correspondence to Niels Rochow, MD, Division of Neonatology, Department of Pediatrics, McMaster University, 1280 Main St West, HSC-4F5, Hamilton, ONT, Canada L8S 4K1. E-mail:
Dr Rochow conceived of the study and was involved in literature searching and data analysis; P. Raja was involved in literature searching; and Dr Voigt was involved in data analysis. All authors were involved in writing the paper.
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
FUNDING: No external funding.
- ↵Annual report 2009. Canadian Neonatal Network. Available at: www.canadianneonatalnetwork.org. Accessed October 10, 2011
- Kramer MS,
- Platt RW,
- Wen SW,
- et al.,
- Fetal/Infant Health Study Group of the Canadian Perinatal Surveillance System
- Olsen IE,
- Groveman SA,
- Lawson ML,
- Clark RH,
- Zemel BS
- Clayton PE,
- Cianfarani S,
- Czernichow P,
- Johannsson G,
- Rapaport R,
- Rogol A
- Copyright © 2012 by the American Academy of Pediatrics