2001
There is no doubt that maternal milk should be the only duration of the speci c feeding of each species are
nourishment that children receive during the rst regulated by physiological mechanisms and common
months of life (1). Nevertheless, the majority of women behaviours in mammals (5). In humans, however,
stop breastfeeding their children early, and the use of breastfeeding habits are determined in every age and
non-human milk in the feeding of infants continues to culture by a combination of multiple factors (6),
be a major public health problem (2), since it is including the attitude of the mother and familial and
associated with higher morbidity and mortality rates social support (7), institutional medical care (8),
(3, 4). professional health advice (9), commercialization stra-
The production of milk, and the maintenance and tegies targeting women and children (10), and health
legislation and governmental policies (11).
The planning of preventive measures to protect
* CLACYD study: Córdoba Lactation, Feeding, Growth and Devel- against premature cessation of breastfeeding requires
opment study. diagnosis with a multidimensional focus, permitting the
capture of all associated factors. The gestation–birth– were excluded because of a lack of information
puerperium process provides an ideal opportunity to regarding breastfeeding, 15 were excluded because
implement promotion and prevention strategies in the breastfeeding was not initiated, and 48 cases were lost
area of institutional childbirth assistance (12). The before 6 mo and therefore the age at which breastfeed-
targeting of high-risk groups for breastfeeding support ing was stopped was unknown. The data analysed
activities is a strategy to be developed further; in corresponded to 632 mother–child observation units.
particular, groups at risk of premature end of breast- The mothers were selected during their admission to
feeding are poorly identi ed, partly because of the lack hospital, immediately after birth, for inclusion in the
of adequate instruments to predict early weaning. study. After informed consent was obtained, mothers
The Córdoba Lactation, Feeding, Growth and Devel- were interviewed in their homes at 1 and 6 mo from
opment (CLACYD) study was a representative cohort baseline at birth. The collection of data was carried out
study of children born in the city of Córdoba, Argentina. using structured questionnaires administered by trained
In this population 98% of mothers began breastfeeding healthcare professionals and students.
in 1993, while at 30 d and 6 mo, only 76% and 32%, To evaluate the model, a new sample, CLACYD II,
respectively, of these mothers continued with partial or was selected using the same inclusion and exclusion
complete breastfeeding (13, 14). criteria as in the previous sample. The sample was
The aim of this study was to develop a model to gathered from the registry of 475 births during
predict cessation of breastfeeding before 6 mo in November 1998. Of the initial sample, 128 were
newborns, and to evaluate its performance in a further excluded (29 cases were discharged within 24 h post-
large cohort. The study was based on the hypothesis that partum; 79 cases were resident outside the metropolitan
the length of breastfeeding is associated with demo- area, had birthweight <2000 g, were stillbirths or
graphic, social and cultural phenomena, as well as the multiple births, or the mother was in an intensive care
institutional process of birth and puerperium. Thus, an unit; and 20 cases refused to participate in the study).
adequate combination of variables of these dimensions Thus, 347 pairs were included in the nal sample.
would provide acceptable levels of prediction. In an interview conducted 24–48 h postpartum, a
questionnaire was administered containing the variables
from the model designed in the previous sample. At 6
Methods mo after the birth, a home interview was conducted to
gather information on weaning, which obtained re-
Data from the CLACYD I study, carried out in 1993 and sponses of 84% (291 interviews) of the initial sample.
1995, were used to construct a model that could be used
to predict premature cessation of breastfeeding. The
model’s predictive capacity was evaluated in a later Variables analysed
study, the CLACYD II, conducted in 1998. A child was considered to be breastfed when it received
maternal milk, regardless of the introduction or non-
Sample selection introduction of additional feeding (15). In the analysis,
Data for the construction of the model to predict the dependent variable was the age at which cessation of
premature end of breastfeeding came from the follow- breastfeeding occurred, registered as the number of
up carried out on 2 samples of women and their months of life at the time of the last feeding. Premature
children, called CLACYD I. The rst sample was end was considered to be a last breastfeeding before 6
selected in 1993 and consisted of 654 children with a mo of age. The variable was recoded as dichotomous for
birthweight ¶2500 g, who were born in any of the urban analysis (cessation of breastfeeding before 6 mo: yes/
institutions with obstetric services (public, social no).
services and private) during May 1993. The second To analyse factors associated with premature cessa-
sample, collected between June and August 1995, was tion of breastfeeding, the following variables were
derived exclusively from children with a low birth- incorporated.
weight (<2500 g). In both samples, families living
outside the urban area, multiple births (2 or more) and Sociodemographi c and cultural factors.. These factors
children with major congenital anomalies were ex- comprised: maternal and paternal age; maternal age at
cluded. Of the total of 167 children of the low birth of rst child; maternal education level (non-
birthweight sample, 46 were randomly selected to be completed primary education or less/completed primary
added to the sample for analysis of duration of education or more); marital status of the mother (single
breastfeeding, and for the construction of the instru- or separated/married or in a stable union); family
ments for prediction of premature cessation of breast- composition (nuclear/extended or monoparental); pri-
feeding. Thus, the sample included 700 children, of mary breadwinner (mother/other person responsible);
whom 6.6% were children of low birthweight (2000– living conditions (good or normal/bad, using housing
2499 g). This proportion of low birthweight births was and environmental hygiene indicators) (13); overcrowd-
similar to that of the city of Córdoba in 1993. Five cases ing index (the relation between the number of indivi-
546 S Berra et al. ACTA PÆDIATR 90 (2001)
duals in the home and the number of rooms in the home, statistic estimates the degree of correspondence be-
excluding bathrooms, kitchen, hallways and garage); tween the probability of premature cessation of breast-
mother’s work before pregnancy (yes/no); mother’s feeding estimated by the model, and premature end of
work during pregnancy (yes/no); health coverage with a breastfeeding observed in the sample, grouped by
health insurance company (yes/no); birth order of the deciles of risk. The model’s ability to discriminate
child ( rst/second or more); intergenerational interval was analysed by calculating the area under the receiver
from the previous child (in months); institution where operating characteristics (ROC) curve (18). The ROC
the birth occurred (public/private); smoking before and curve represents the proportion of pairs in which the
during pregnancy (yes/no); and pregnancy planning weaned children have a greater probability of premature
(planned/unplanned). Social strata were de ned using cessation than the children who continued breastfeed-
the occupation of the principal breadwinner, taking into ing, for all possible pairs formed by 1 child from the
account the activity, occupational categories, job func- sample being breastfed and 1 who was not breastfeeding
tions, and the sector and branch of activity. Initially, 45 at the age of 6 mo.
socioeconomic positions were created, which were later The constructed model was then evaluated in the
regrouped into 6 socioeconomic strata (SES) and CLACYD II sample using the coef cients obtained with
recoded for analysis as upper–middle SES and lower– the CLACYD I sample. The above-mentioned calibra-
very low SES (13). tion and discrimination tests were repeated. To improve
the calibration of the model, it was adapted using a
Perinatal factors.. These factors were: sex; birthweight linear logit transformation, taking the logit [ln(p/1 ¡ p)]
(2000–2499 g/¶2500 g); gestational age (<38 wk/¶ of the probability of premature weaning (p) obtained in
38 wk); APGAR 1 min (<7/¶7); history of abortion the original model as the independent variable. Once the
(yes/no); history of caesarean section (yes/no); history model had been adjusted, its calibration was evaluated
of children born with low birthweight (yes/no); begin- again.
ning of pregnancy care ( rst trimester/second trimester
or third trimester); weight gain during the pregnancy
(in kg); duration of labour (in hours); type of birth Results
(vaginal/caesarean); rooming in maternity ward (room-
ing-in/nursery or intensive care unit); time passed The CLACYD I and CLACYD II samples were similar
between birth and rst contact between mother and in their sociodemographic and cultural characteristics,
child (µ90 min/>90 min); and rst food received and in antecedents of breastfeeding (Table 1). However,
(maternal milk/other food). statistically signi cant differences were found between
CLACYD I and CLACYD II samples in the perinatal
Breastfeeding antecedents.. These comprised: history of variables, which showed a lower percentage of births by
breastfeeding of previous child (¶6 mo/no previous caesarean section, as well as of delay in the rst
children/<6 mo), mother having been breastfed (does mother–child contact, of administration of arti cial
not remember or ¶6 mo/<6 mo). The last variable was milk or glucose syrup as rst food, and of lodging of the
recoded, as the risk of premature cessation of breast- child separately from the mother. The proportion of
feeding was similar in the group that did not remember cessation of breastfeeding at 6 mo was 13% points
and in the basal group that was breastfed for 6 mo or lower in the CLACYD II sample (p < 0.05).
longer. The model has improved with this approach. In the bivariate analysis of the CLACYD I sample,
the proportion of duration of lactation less than 6 mo
was greater in women who had completed primary
Statistical analysis education or more, and in those who not living with a
A logistic regression model (16) was constructed with partner or other family members. Premature cessation
data from the CLACYD I sample to predict premature was also more frequent among women who worked
weaning, whilst controlling for the simultaneous effect during pregnancy, in those who had been breastfed less
of sociodemographic and perinatal factors, and incor- than 6 mo, in those who had breastfed their previous
porating those variables that were statistically signi - child for less than 6 mo and in those who had their rst
cant in the bivariate analysis (at the 5% level), as well as contact with their child or gave them their rst feeding
those that were considered to be of interest (e.g. SES, at more than 90 min after birth (Table 2). The
pregnancy planning). proportion of premature cessation was also greater,
After tting the best model (in terms of goodness-of- although not statistically signi cant, in those children
t and explanatory variables), the probabilities of who received arti cial milk or glucose syrup as their
premature end of breastfeeding were calculated for rst nourishment and in those who did not remain with
each case using the combination of factors included in their mothers after birth.
the nal model, given p = 1/1 ‡ exp¡z . The calibration The predictive factors of cessation of breastfeeding
of the model was evaluated using the Hosmer–Leme- before 6 mo (Table 3) which were statistically
show C statistic (17). The Hosmer–Lemeshow C signi cant in the nal model were: mother having been
ACTA PÆDIATR 90 (2001) Predictive model of premature cessation of breastfeedin g 547
breastfed for less than 6 mo [odds ratio (OR) = 1.84], (OR = 2.75), the rst mother–child contact occurring
previous child being breastfed for less than 6 mo after 90 min of life (OR = 1.88) and having an un-
(OR = 4.01), the condition of the rstborn child planned pregnancy (OR = 1.50).
Table 3. Multiple logistic regression model for predictio n of had a greater probability of cessation of breastfeeding
premature end of breastfeedin g (CLACYD I) than those who did.
Variables OR 95% CI
When the model was applied to the CLACYD II
sample, the area under the ROC curve was 0.68 and the
Breastfeeding of previous child
¶6 mo 1a calibration showed an overestimation of weaning above
No previous child 2.75 1.79–4.21 values of 0.567 (C = 31.9; p = 0.0005) (Table 4B). After
<6 mo 4.01 2.58–6.20 the model was adapted using the logit transformation
Mother having been breastfed (Table 4C), the calibration improved, so that the
¶6 mo/does not remember 1a
<6 mo 1.84 1.26–2.70
difference between expected and observed cessation
First mother–child contact of breastfeeding was not signi cant (C = 11.4; p =
<90 min 1a 0.18).
¶90 min 1.88 1.22–2.91
Pregnancy planning
Planned 1a
Unplanned 1.50 1.05–2.15 Discussion
a Reference category. This is the rst study to examine the possibility of
OR: odds ratio; CI: con dence interval. predicting premature cessation of breastfeeding in
newborns, in which both the initial model development
and the evaluation of the model were performed in
prospective cohorts of newborns. Although this type of
Figure 1 shows the probabilities of premature evaluation is recommended (17) few studies actually
cessation of breastfeeding for each combination of have used it. In addition to conforming with a necessary
factors analysed, calculated with the coef cients ob- methodological exercise, the results may be of practical
tained with the logistic regression model. For example, use. The study design thereby avoided some of the risks
a rst child born to a mother who had been breastfed for associated with cross-sectional and retrospective stu-
less than 6 mo, with a rst contact with his or her mother dies, such as directionality of association (19).
after 90 min of life and the product of an unplanned The study identi ed 4 factors associated with early
pregnancy, has a 0.787 probability of cessation of weaning: mother having been breastfed for only a short
breastfeeding before 6 mo. period, reduced breastfeeding of the previous child,
The model’s calibration was acceptable (C = 6.94; excessive time between birth and rst mother–child
p = 0.54). Some differences were seen between ex- contact, and absence of pregnancy planning. The
pected and observed premature cessation of breastfeed- presence of these 4 factors in a single case implies a
ing for probabilities greater than 0.51 and less than 0.74 probability of cessation of breastfeeding before 6 mo
(Table 4A). The ROC curve showed that in 72% of the which is 4 times higher than in those mothers with
pairs, mothers who did not breastfeed for at least 6 mo previous experience of breastfeeding 6 mo or longer,
Table 4. Breastfeeding and cessation of lactation observed and expected by the model, at 6 mo of age
whose pregnancy was planned and who had contact for longer than 6 mo has bene cial effects on the
with their children within 90 min after birth. Those probability of breastfeeding the later siblings.
women who breastfed their previous child for 6 mo or Birth and puerperium assistance practices have been
more never had a greater than 60% probability of shown to in uence the duration of breastfeeding (22).
premature cessation of breastfeeding. Nevertheless, the independent effect of a delay in rst
Few studies have addressed the association between contact has not been well de ned (8). The strong
the duration of breastfeeding of the mother or of a association found in this study deserves consideration.
previous child with the duration of breastfeeding of a This is the only variable included in the model
new child (20). The in uence of the experience of concerning the practices of healthcare services, with a
breastfeeding with a previous child is consistent in the high prevalence in the present context and unjusti ed
literature on this subject (7, 21), and may be representa- from the point of view of clinical practice (22).
tive of a dif culty in the previous breastfeeding that was The lower educational level of the mother, which
not resolved positively or of a negative predisposition appears to be a protective factor against the premature
that in uenced the conduct of the mother. These factors end of breastfeeding in developing countries (2, 23), did
can be investigated beginning with the rst obstetric visit not present an independent contribution to early
and can be addressed by the professional as well as by weaning in the present study. This may be associated
birth and breastfeeding assistance and preparation with a change in lactation patterns in the urban areas of
groups. Although breastfeeding of the previous child is Argentina or, in other words, an intermediate result or
usually considered to be a “non-modi able” factor, it is evidence of a transition towards the lactation standards
also important to keep in mind that breastfeeding lasting of developed countries.
550 S Berra et al. ACTA PÆDIATR 90 (2001)
Some limitations of the study deserve comment. Acknowledgements.—The authors thank the families whose participation
Some factors, related to the family and cultural made this project possible; institutions with obstetrics services for
allowing us access to mothers; professionals and students from Nutrition
context, such as the opinion of the father and the and Medicine for their co-operation ; Montse Rue (Hospital de Sant Pau,
grandmothers regarding the feeding of the child and Barcelona) and Marõ´a del Pilar Dõ´az (Escuela de Nutrición, Universidad
the use of a paci er, could not be used. They were Nacional de Córdoba) for their assistance with statistical methods; Michel
collected in the CLACYD I sample during the Herdman and Michelle Pollit for their help in preparing the manuscript;
Fabiana Pirán and the staff of the CLACYD study for their support of the
interview conducted at the age of 1 mo, when the project. This study was supported by research grants from the Consejo de
maternal lactation pattern was already established, and Investigacione s Cientõ ´ cas y Tecnológicas de la Provincia de Córdoba
therefore the directionality of the association could (CONICOR), Argentina, and Alberto J. Roemmers Foundation, Argen-
not be established. In the same way, the work of the tina.
mother in the rst months of life (24) and the
incorporation of arti cial milks (25), which have been
shown to be associated with the duration of breast- References
feeding, were not taken into account because they
occurred in phases after the birth. The association 1. World Health Organization. Complementary feeding of young
children in developin g countries : a review of current scienti c
between premature abandonment of complete breast- knowledge . Geneva: World Health Organization, 1998
feeding due to the incorporation of arti cial food and 2. Perez Escamilla R. Breastfeedin g in Africa and Latin American
the shortening of the total period of breastfeeding are and Caribbean region: the potential role of urbanization . J Trop
well documented factors (26, 27). This should be the Pediatr 1994; 40: 137–43
3. World Health Organizatio n Collaborativ e Study Team on the
subject of a speci c study aimed at analysing the Role of Breastfeedin g on the Prevention of Infant Mortality.
factors associated with the introduction of non- Effect of breastfeedin g on infant and child mortality due to
maternal milks. infectious in less develope d countries : a pooled analysis. Lancet
The discrimination power of the model, evaluated 2000; 355: 451–5
with the area under the ROC curve, would improve with 4. Cunningha m AS, Jelliffe DB, Jelliffe EF. Breast-feedin g and
health in the 1980s: a global epidemiologi c review. J Pediatr
the inclusion of other potential indicator variables of 1991; 118: 659–66
cessation at birth. In addition, the overestimation of 5. Dettwyler K. A time to wean: the hominid blue print for the
premature ending of breastfeeding that the model makes natural age of weaning in modern human populations . In: Stuart-
with the CLACYD II sample can be partly explained by Macadam P, Dettwyler KA, editors. Breastfeeding , biocultura l
perspectives . New York: Aldine de Gruyter, 1995: 39–73
the reduction in the prevalence of premature cessation 6. Stuart-Macada m P, Dettwyler KA, editors. Breastfeeding , bio-
in relation to the earlier sample. cultural perspectives . New York: Aldine de Gruyter, 1995
One variable that was not studied, although asso- 7. Picado JI, Olson CM, Rasmussen KM. Metodolog õ´a combinada
ciated in other studies with the duration of breastfeed- para entender la duración del amamantamient o en barrios pobres
ing, was the intention to breastfeed (7, 28). This might de Managua, Nicaragua. Rev Panam Salud Pública 1997; 2: 398–
406
be a useful factor in predicting cessation during the rst 8. Perez Escamilla R, Pollit E, Lönerdall B, Dewey KG. Infant
moments of lactation. It is possible that the history of feeding policies in maternity wards and effect on breast-feedin g
breastfeeding of both the mother and the previous child, success: an analytica l overview. Am J Public Health 1994; 84:
as well as the planning of the pregnancy, could be an 89–97
9. Millard AV. The place of the clock in pediatric advice:
expression of the mother’s attitude and disposition rationales , cultural themes, and impediments to breastfeeding .
towards the future child as these relate to the decision to Soc Sci Med 1990; 31: 211–21
breastfeed. 10. Taylor A. Violations of the internationa l code of marketing of
The rural population was not represented in this breast milk substitutes : prevalence in four countries. Br Med J
study. Nevertheless, the model was constructed in a 1998; 316: 1117–22
11. Van Esterik P. The politics of breastfeeding : an advocacy
representative sample of a city in South America, where perspective . In: Stuart-Macadam P, Dettwyler KA, editors.
the population concentration in cities could be more Breastfeeding , biocultura l perspectives . New York: Aldine de
than 70%. Gruyter, 1995: 145–65
The presentation of a predictive model of premature 12. WHO/UNICEF. Protecting, promoting and supporting breast-
feeding: the special role of maternity services. Geneva: World
cessation of breastfeeding is a rst step in the construc- Health Organization , 1989
tion of new instruments which may allow action to be 13. Sabulsky J, Batrouni L, Carballo R, et al. Alimentación en el
taken against the risk of premature ending of breast- primer mes de vida, por estratos sociales, Córdoba, Argentina.
feeding. Bol O cina Sanit Panam 1995; 119: 15–27
The construction of new models based on data 14. Berra S. Per l de lactancia desde el nacimiento hasta los dos años
de vida y factores asociado s a la duración de la lactancia natural.
collected in different contexts and which cover the In: Estudio CLACYD. Per les epidemiológicos de alimentación,
wide spectrum of determinant factors of cessation of crecimiento y desarroll o en los dos primeros años de vida.
breastfeeding will open the way to improved strategies Publicación No. 2. Córdoba: EUDECOR, 1997: 43–53
for the promotion of breastfeeding at birth. The 15. World Health Organization. Indicators for assessing breast-
feeding practices. Report from an informal meeting 11–12 June
development of models that could be used during 1991. Division of diarrhoea l and acute respirator y disease
pregnancy is also an interesting subject for future control. WHO/CDD/SER/91.14. Geneva: World Health Organi-
studies. zation, 1991
ACTA PÆDIATR 90 (2001) Predictive model of premature cessation of breastfeedin g 551
16. Breslow EN, Day EN. Statistical methods in cancer research . 23. Koçtürk T, Zetterström R. Thoughts about rates of breastfeeding .
Vol. 1. The analysis of case–control studies. Lyon: Internationa l Acta Paediatr 1999; 88: 356–8
Agency on Cancer, 1988 24. Fein SB, Roe B. The effect of work status on initiation and
17. Lemeshow S, Hosmer DW. A review of goodnes s of t statistics duration of breastfeeding . Am J Public Health 1998; 88: 1042–6
for use in the developmen t of logistic regressio n models. Am J 25. Blomquist HK, Jonsbo F, Serenius F, Persson LA. Supplemen-
Epidemiol 1982; 115: 92–106 tary feeding in the maternity ward shortens the duration of breast
18. Swets JA. Measuring the accuracy of diagnosti c system. Science feeding. Acta Paediatr 1994; 83: 1122–6
1988; 240: 1285–93 26. Hill PD, Humenick SS, Brennan ML, Woolley D. Does early
19. Szklo M, Nieto FJ. Epidemiology . Beyond the basics. Gaithers- supplementatio n affect long-term breastfeeding ? Clin Pediatr
1997; 36: 345–50
burg, MD: Aspen, 2000
27. Bergevin Y, Dougherty C, Kramer MS. Do infant formula
20. Riva E, Banderali G, Agostoni C, Silano M, Radaelli G,
samples shorten the duration of breast-feeding ? Lancet 1983; i:
Giovaninni M. Factors associate d with initiation and duration 1148–51
of breastfeedin g in Italy. Acta Paediatr 1999; 88: 411–5 28. Quarles A, Williams PD, Hoyle DA, Brimeyer M, Williams AR.
21. Hörnell A, Aarts C, Kylberg E, Hofvander Y, Gebre Medhin M. Mothers’ intention, age, educatio n and the duration and manage-
Breastfeedin g patterns in exclusivel y breastfed infants: a longi- ment of breastfeeding . Matern Child Nurse J 1995; 22: 102–8
tudinal prospectiv e study in Uppsala, Sweden. Acta Paediatr
1999; 88: 203–11
22. World Health Organization. Evidence for the ten steps to Received Sep. 6, 2000; revision received Dec. 29, 2000; accepted
successfu l breastfeeding . Geneva: World Health Organization , Jan. 23, 2001
1998