Effect of Instrument Precision on Estimation of Low Birth Weight Prevalence

Abstract

OBJECTIVE

Reductions in prevalence of low birth weight (LBW) are likely to be achieved only through small increments, amplifying the importance of precision of measurement. This study compared two instruments for measuring birth weight to investigate the effect of instrument precision on estimates of LBW prevalence.

STUDY DESIGN

Analysis was based on 497 infants born to mothers enrolled in an ongoing community-based trial of the effect of umbilical cord and skin antisepsis on neonatal mortality in Nepal. For each child, two birth weight measurements were recorded within 72 hours after delivery using weighing scales of differing precisions (100 vs 2 g).

RESULTS

While continuous measures between the two instruments were similar, the prevalence of LBW among lower precision measurements (30%) was 11.3% lower than for higher precision measurements (34%). The difference in precision between the instruments accounted for 96% of the difference in LBW prevalence estimates.

CONCLUSIONS

Differences in estimated LBW rates may be entirely due to differences in instrument precision. Conclusions concerning programmatic or research intervention impact, or comparisons of rates across populations should consider the effect of instrument precision on estimates of LBW.

INTRODUCTION

Low birth weight (LBW) represents the leading determinant of neonatal mortality in the developing world.¹ Progress on reducing the global prevalence of LBW has been limited, and further reductions are likely to come through small incremental decreases, thus amplifying the importance of precision of measurement in both research and programmatic contexts. Within an ongoing community-based trial of the effect of topical umbilical cord and skin antisepsis on neonatal mortality and morbidity in southern Nepal, birth weights are being collected as a comparative measure of neonatal risk across treatment groups. The availability of two well-accepted digital scales with differing precisions, both manufactured by Seca Corporation, allowed us to evaluate the implications of using scales of differing precisions on birth weight measures. The test scale was the mother–infant solar-powered platform scale manufactured by Seca for UNICEF. This platform scale measures weight to a precision of 100 g and was compared to a gold standard neonatal scale (Seca Digital Baby Scale Model 727) that measures weight to within 2 g.

Newly identified pregnant women in Sarlahi district of Nepal were enrolled into the trial at approximately 6 months into their pregnancy. At the time of enrollment, the study goals were explained, and informed consent was obtained from participants. Soon after hearing of a live birth in their area, local female workers notified upper level staff of the delivery by dropping notification forms in suboffices distributed throughout the district. Between March 28, 2003 and May 1, 2003, 541 birth notices were collected. Upon receipt of the notification form, health workers carried data collection forms and both weighing scales to the newborn’s home. Before each measurement, the scales were standardized with known weights (1000, 1100 g). Five live-born infants died before weight measurement, three mothers withdrew their prior consent to participation, and in three cases, one or both of the weight measurements were not taken due to logistical difficulties. Of the remaining 530 cases, 497 (93.7%) were measured within 72 h after birth, and were included in the analysis. Analysis included paired Student’s t-test for comparison of mean birth weights, Wilcoxon’s sign-rank test for assessing matched-pairs equality of birth weight distribution between the two scales, a sign test to assess the equality of medians, and sensitivity and specificity analysis, using the digital baby scale as the gold standard. The Johns Hopkins University Committee on Human Research and the Nepal Health Research Council approved all study protocols.

There were no significant differences in the means, standard deviations, medians, or ranges of weight between the two scales (Table 1). The distributions of birth weight measures, compared using the Wilcoxon sign-rank test, were also found to be equal (p = 0.14). In contrast, significant differences in LBW rates were found between the two scales. Using the digital baby scale, 169 infants weighed less than 2500 g, yielding a gold standard prevalence of LBW of 34.0%. Among these LBW babies, the solar-powered platform scale identified 149 weighing less than 2500 g (sensitivity = 88.2%, 95% CI: 82.3 to 92.6). Among non-LBW babies (n = 328), the solar-powered platform scale correctly identified 327 weighing 2500 g or more (specificity = 99.7%, 95% CI: 98.3 to 99.9). The estimated prevalence of LBW babies using the solar scale (30.2%) was nearly 4 percentage points lower than the prevalence using the neonatal scale, equivalent to an 11.3% lower LBW rate than the gold standard. Of the 34 newborns weighing <2000 g, the solar platform scale correctly identified 27 (sensitivity = 79.4%, 95% CI: 62.1 to 91.3), while all 463 infants 2000 g or above were correctly classified by the solar scales (specificity = 100%, 95% CI: 99.2 to 100.0). Overall prevalence of birth weight <2000 g differed by 20.6% between the two scales (gold standard scale, 6.8%; solar scale, 5.4%) (see Table 2).

Table 1.

Comparison of Birth Weight Parameters of Solar Platform^* and Neonatal Scales^†

	Solar N = 497	Neonatal N = 497	Test	Result	p-Value
Mean	2678.9 g	2683.2 g
Paired differences	(Range: −482 to 402 g)		Student’s paired t-test	Mean difference: −4.3 g 95% CI (−10.4, 1.7 g)	0.16
Standard deviation	474 g	478 g	Variance ratio test	Ratio = 0.985	0.86
Median	2700 g	2692 g	Sign test (Ho medians equal −50% of differences are positive)	46.7% of differences are positive	0.24
Range	900 to 4700 g	996 to 4646 g

^*Mother–infant solar-powered platform scale — Seca Corporation.

^†Digital baby scale — Model 727 — Seca Corporation.

Table 2.

Sensitivity/Specificity Analysis

	Neonatal Digital Scale (Gold Standard)
Platform Scale	≥2500 g	<2500 g	Total
(a) Low birth weight (<2500 g)
≥2500 g	327	20	347
<2500 g	1	149	150
Total	328	169	497
Sensitivity of platform scale	88.2%
Specificity of solar scale	99.7%
(b) Infants less than 2000 g
≥2000 g	463	7	470
<2000 g	0	27	27
Total	463	34	497
Sensitivity of platform scale	79.4%
Specificity of platform scale	100.0%

The difference in prevalence of LBW categories (<2500, <2000 g) between the two scales was almost entirely due to an interaction between the outcome cutoff points, and the difference in precision of the two scales. When comparing the distribution of the continuous measures displayed by the two scales, the median, mean, range, and distribution shape were not significantly different, yet the sensitivity of the solar platform scale in detecting infants <2500 g resulted in a 11.3% difference in the estimate of LBW and a 20.6% difference in the estimated prevalence of infants weighing <2000 g. The key to understanding these differences lies in an inspection of the cutoff value in the definition of LBW and the subsequent handling of infants with birth weight measures equal to this cutoff value. The definition of LBW explicitly excludes from the LBW group those infants with birth weight exactly equal to the cutoff value. If differing precision between two instruments necessarily results in a greater tendency to record values equal to such an excluded cutoff value, the observed rate of classification of subjects into the LBW class may differ significantly between the instruments due solely to these differences in precision. If solar scale measures of exactly 2500 g (there were 40 such instances) were randomly distributed in even proportion between 2501 and 2499 g, the relative difference in estimated LBW prevalence would be reduced from 11.3 to 0.5%. Likewise, the relative difference in the estimate of infants <2000 g would be reduced from 20.6 to 3.1% if solar platform scale measurements equal to the cutoff (n = 11) were distributed evenly between 2001 and 1999 g. Thus, the difference in precision of the scale was responsible for 96 and 85% of the apparent difference in prevalence of infants weighing <2500 and <2000 g, respectively.

Loss of body weight during the initial days of life is routine among newborns^2–4 although development of precise estimates of the timing and magnitude of this potential weight loss is problematic in developing country settings, where such data are lacking and considerable interneonate variation has been observed.^5–7 Measurements collected in this study ranged in time from within minutes of birth to up to 72 hours after birth; thus, weights recorded may not necessarily represent true birth weight. While this has implications for the absolute estimates of prevalence of LBW in this study, each newborn-specific pair of measurements was recorded within minutes of each other. Therefore, any changes in weight between time of birth and time of measurement do not impact conclusions drawn regarding the effect of instrument precision on categorical birth weight classification.

Given that progress toward reductions in LBW rates is likely to occur in small increments, the precision of instruments used to categorize infant birth weight is of utmost importance. Changes in estimated LBW rates entirely due to differences in instrument precision are possible, resulting in erroneous conclusions concerning the presence or absence of programmatic or research intervention impact. Likewise, comparison of LBW rates across populations or within populations over time may result in false conclusions if measurements are collected using instruments with differing precisions.