R. KUMAR
Prenatal Growth Curves Corrected for Certain Genetic and Environmental Factors by Ramesh Kumar, DCH, MD (Paed) Department of Paediatrics, M.L.B. Medical College, Jhansi, India
Introduction Among the various parameters of fetal growth, birth weight is, perhaps, the one that has received the greatest attention, both from clinicians as well as human biologists. The accumulated literature during the past century is full of references on this subject. McKeown and Gibson 1 have cited a few large studies, each involving serveral thousand newborns, which were conducted in the first half of this century. Thereafter, in 1960s and 1970s a handful of more such studies 2 " 5 were added to the list. An enormous number of studies have been conducted which included a relatively small number of subjects. Even then, certain questions have remained, by and large, unanswered. These questions related to the relative importance of parity, age, height, and weight of a mother in determining the birth weight of her newborn. Doubts have been expressed regarding the very independence of some or all of the variables mentioned above. With the introduction of electronic computers in social sciences, multivariate regression analysis has become an easy, yet important tool in the study of individual effect of any variable on birth weight while keeping the effect of all other variables constant. However, one has to make certain assumptions before proceeding on with multivariate regression analysis. These are: additive effect of variables under study, independence of variables from each other, normal distribution of each array in a bivariate distribution. In biological sciences, especially those involving experimentation on human subjects, all these assumptions may not always be met. Thus, in a situation like this, Correspondence: Dr Ramesh Kumar, Department of Paediatrics, M.L.B. Medical College, Jhansi-284 128, U.P., India. 256
© Oxford University Press 1992
multiple regression analysis looks merely a simplistic exercise. What would, perhaps, be more rational to do is to factorize each variable and to quantify the contribution of each factor to birth weight, while allowing for interaction among factors. Thus, in an experiment where a variable can be reduced to two or more factors and where the simultaneous effect of several factors is to be tested on a particular response, without prior assumptions of independence or additivity of variables, a factorial design is called for. However, nonorthogonality of data (unequal sub-class numbers) makes the calculation somewhat difficult and lengthy as against the one where orthogonality is maintained, or else when the design is a balanced one (regular pattern of departures from orthogonality. 6 Algorithm for calculations have been worked out by several authors and some are notable for their simplicity.6"8 In the present study a factorial design was used to investigate the straight and interaction effects of variables, viz., gestational age, sex, parity, socioeconomic status of the family, and mother's height and weight on birth weight. Data showed complete breakdown of orthogonality. Materials and Methods Material for this study consisted of 575 hospital-born singleton babies. Cases of hydramnios, babies having manifest congenital or chromosomal anomalies, and those in moribund condition were not included in the sample. All those cases included in the study were born after uncomplicated normal vaginal delivery and had their gestational ages known. Gestational age was computed on the basis of number of completed weeks from the 1st day of last menstrual period to the date of delivery. Out of 575 babies, 48 were born before term Journal of Tropical Pediatrics
Vol. 38
October 1992
Downloaded from http://tropej.oxfordjournals.org/ at Pennsylvania State University on May 24, 2015
Summary Factor analysis for non-orthogonal data was done to assess the straight and interaction effects of certain genetic and environmental factors on birth weight. While the interaction effects did not assume any significance, straight effects of gestational age, parity, mother's weight, and sex were significant in influencing the birth weight. Mother's height and socio-economic status of the family had no significant effect on birth weight. Step-up multiple regression analysis showed that contribution to the total variance explained (R2 x 100 = 29.7 per cent) in birth weight was about 15, 7, 5, and 2 per cent, respectively, from gestational age, parity, mother's weight, and sex of the baby. Correction of the mean weight curve was done by successive graphic approximation method and residuals were further minimized.
R. KUMAR
TABLE 1
Preliminary analysis of variance Source of variation Between combinations Error (within combinations)
Degrees of freedom
Sum of squares
Mean square
Variance ratio
Significance*
73 501
41.258 81.656
0.565 0.163
3.47
VHS
Mean squares
Variance ratio
Significance'
26.28 1.05
VHS NS
* VHS = very highly significant (P< 0.001). TABLE 2
Final analysis of variance (broad effects) Degrees of freedom
Sum of squares
7 66
29.987 11.271
4.284 0.171
Between combinations Error
73 501
41.258 81.656
0.163
Total
574
122.914
Effect of factors All interactions
VHS = very highly significant (i>>0.2).
(30-36 weeks) and the rest were born following full term (37-45 weeks) gestation. There were 14 cases born after 42 weeks of gestation. The sample consisted on 12 per cent small-for-gestational-age babies, as per western norms. 9 However, such cases numbered only three, by Indian standards. 10 Babies were weighed nude between 24 and 48 hours after birth. Mothers were weighed in overalls of known weight and their heights recorded between the 7th and 10th day of delivery, since by that time Indian mother's weight almost equals to or is slightly higher than her prepregnancy weight.11 Standard technique 12 was used to measure the height. All variables (except the birth weight and sex of the baby) were scored as follows: A. Parity: score 1 = parity 0, score 2 = parity 1 and 2, score 3 = parity 3 above. B. Mother's height: score 1 = height < 155.0 cm, score 2 = height ^ 155.0 cm. C. Mother's weight: score 1 = weight < 50 kg, score 2 = weight ^50.0 kg. D. Socio-economic status: 13 score 1= social class 1 and 2, score 2 = social class 3 and 4. Scoring, done in the present study, was necessary to prevent the over saturation of model. This also brought the frequency distribution of parity to near normal. Observations
A preliminary, one-way analysis of variance (Table 1) showed that birth weight mean-square (MS) between Journal of Tropical Pediatrics
Vol.38
October 1992
combinations 1 was higher (0.57) than MS within combinations. (0.16). Mean-sqaure within combinations,, in fact, is the variance arising out of sampling and measuring errors. Next, an approximate analysis of variance was done to separate the main effects, all two-factors interactions with 20 degrees of freedom (d.f.) and combined 6factor interaction (d.f. = 73) sums-of-square (SS). Analysis showed that mother's height made a significant contribution (0.01
Prenatal Growth Curves Corrected for Certain Genetic and Environmental Factors by Ramesh Kumar, DCH, MD (Paed) Department of Paediatrics, M.L.B. Medical College, Jhansi, India
Introduction Among the various parameters of fetal growth, birth weight is, perhaps, the one that has received the greatest attention, both from clinicians as well as human biologists. The accumulated literature during the past century is full of references on this subject. McKeown and Gibson 1 have cited a few large studies, each involving serveral thousand newborns, which were conducted in the first half of this century. Thereafter, in 1960s and 1970s a handful of more such studies 2 " 5 were added to the list. An enormous number of studies have been conducted which included a relatively small number of subjects. Even then, certain questions have remained, by and large, unanswered. These questions related to the relative importance of parity, age, height, and weight of a mother in determining the birth weight of her newborn. Doubts have been expressed regarding the very independence of some or all of the variables mentioned above. With the introduction of electronic computers in social sciences, multivariate regression analysis has become an easy, yet important tool in the study of individual effect of any variable on birth weight while keeping the effect of all other variables constant. However, one has to make certain assumptions before proceeding on with multivariate regression analysis. These are: additive effect of variables under study, independence of variables from each other, normal distribution of each array in a bivariate distribution. In biological sciences, especially those involving experimentation on human subjects, all these assumptions may not always be met. Thus, in a situation like this, Correspondence: Dr Ramesh Kumar, Department of Paediatrics, M.L.B. Medical College, Jhansi-284 128, U.P., India. 256
© Oxford University Press 1992
multiple regression analysis looks merely a simplistic exercise. What would, perhaps, be more rational to do is to factorize each variable and to quantify the contribution of each factor to birth weight, while allowing for interaction among factors. Thus, in an experiment where a variable can be reduced to two or more factors and where the simultaneous effect of several factors is to be tested on a particular response, without prior assumptions of independence or additivity of variables, a factorial design is called for. However, nonorthogonality of data (unequal sub-class numbers) makes the calculation somewhat difficult and lengthy as against the one where orthogonality is maintained, or else when the design is a balanced one (regular pattern of departures from orthogonality. 6 Algorithm for calculations have been worked out by several authors and some are notable for their simplicity.6"8 In the present study a factorial design was used to investigate the straight and interaction effects of variables, viz., gestational age, sex, parity, socioeconomic status of the family, and mother's height and weight on birth weight. Data showed complete breakdown of orthogonality. Materials and Methods Material for this study consisted of 575 hospital-born singleton babies. Cases of hydramnios, babies having manifest congenital or chromosomal anomalies, and those in moribund condition were not included in the sample. All those cases included in the study were born after uncomplicated normal vaginal delivery and had their gestational ages known. Gestational age was computed on the basis of number of completed weeks from the 1st day of last menstrual period to the date of delivery. Out of 575 babies, 48 were born before term Journal of Tropical Pediatrics
Vol. 38
October 1992
Downloaded from http://tropej.oxfordjournals.org/ at Pennsylvania State University on May 24, 2015
Summary Factor analysis for non-orthogonal data was done to assess the straight and interaction effects of certain genetic and environmental factors on birth weight. While the interaction effects did not assume any significance, straight effects of gestational age, parity, mother's weight, and sex were significant in influencing the birth weight. Mother's height and socio-economic status of the family had no significant effect on birth weight. Step-up multiple regression analysis showed that contribution to the total variance explained (R2 x 100 = 29.7 per cent) in birth weight was about 15, 7, 5, and 2 per cent, respectively, from gestational age, parity, mother's weight, and sex of the baby. Correction of the mean weight curve was done by successive graphic approximation method and residuals were further minimized.
R. KUMAR
TABLE 1
Preliminary analysis of variance Source of variation Between combinations Error (within combinations)
Degrees of freedom
Sum of squares
Mean square
Variance ratio
Significance*
73 501
41.258 81.656
0.565 0.163
3.47
VHS
Mean squares
Variance ratio
Significance'
26.28 1.05
VHS NS
* VHS = very highly significant (P< 0.001). TABLE 2
Final analysis of variance (broad effects) Degrees of freedom
Sum of squares
7 66
29.987 11.271
4.284 0.171
Between combinations Error
73 501
41.258 81.656
0.163
Total
574
122.914
Effect of factors All interactions
VHS = very highly significant (i>>0.2).
(30-36 weeks) and the rest were born following full term (37-45 weeks) gestation. There were 14 cases born after 42 weeks of gestation. The sample consisted on 12 per cent small-for-gestational-age babies, as per western norms. 9 However, such cases numbered only three, by Indian standards. 10 Babies were weighed nude between 24 and 48 hours after birth. Mothers were weighed in overalls of known weight and their heights recorded between the 7th and 10th day of delivery, since by that time Indian mother's weight almost equals to or is slightly higher than her prepregnancy weight.11 Standard technique 12 was used to measure the height. All variables (except the birth weight and sex of the baby) were scored as follows: A. Parity: score 1 = parity 0, score 2 = parity 1 and 2, score 3 = parity 3 above. B. Mother's height: score 1 = height < 155.0 cm, score 2 = height ^ 155.0 cm. C. Mother's weight: score 1 = weight < 50 kg, score 2 = weight ^50.0 kg. D. Socio-economic status: 13 score 1= social class 1 and 2, score 2 = social class 3 and 4. Scoring, done in the present study, was necessary to prevent the over saturation of model. This also brought the frequency distribution of parity to near normal. Observations
A preliminary, one-way analysis of variance (Table 1) showed that birth weight mean-square (MS) between Journal of Tropical Pediatrics
Vol.38
October 1992
combinations 1 was higher (0.57) than MS within combinations. (0.16). Mean-sqaure within combinations,, in fact, is the variance arising out of sampling and measuring errors. Next, an approximate analysis of variance was done to separate the main effects, all two-factors interactions with 20 degrees of freedom (d.f.) and combined 6factor interaction (d.f. = 73) sums-of-square (SS). Analysis showed that mother's height made a significant contribution (0.01
