Complete details of subject recruitment, data collection, and data analysis have been reported previously (18, 19). Briefly, the study subjects were recruited from among healthy, term infants born at all Albuquerque, New Mexico, hospitals between January 1, 1988, and June 30, 1990. Families were invited to participate in the study if the mother was at least age 18 years, spoke English, did not smoke, and had a telephone in the home. Families were excluded if there were any smokers in the home, if the parents planned to place the infant in full-time day care, or if the family planned to move from Albuquerque within the next 18 months.
Enrolled subjects were withdrawn from further follow-up if the infant was placed in day care for more than 20 hours a week, if any household member began to smoke, if the family failed to comply with study procedures, or if the family wished to discontinue participation.
At an initial home visit, study interviewers obtained information on family composition, household income, sources of medical care for the index subject, ethnicity of the parents, and years of education completed by mother and father. Parents completed the standardized respiratory symptoms questionnaire of the American Thoracic Society (20), which includes questions on history of asthma, hay fever, and desensitization injections.
On enrollment, mothers were trained by a nurse to complete a daily calendar, on which they recorded signs and symptoms exhibited by the infant. Symptoms included runny nose, stuffy nose, fever, dry cough, wet cough, wheezing, difficulty breathing, loss of appetite, rash, and whether the mother thought the baby was sick. Every 2 weeks, up to age 18 months, an interviewer telephoned the mother to review the calendar diary for the subject's symptoms. During this call, the interviewer also determined whether the infant had attended day care during the previous week, and, if so, the number of hours and the number of children present. The interviewer also asked whether the infant was being breastfed, and, if so, whether breast milk was the only source of milk in the diet (full breastfeeding) or whether the infant received other milk in addition to breast milk (partial breastfeeding). Breastfeeding status was determined at each telephone interview until a previously breastfed infant no longer received any breast milk.
Illness events were defined, by using standard criteria (21), as the calendar diary report of at least 2 consecutive days of any of the following: runny or stuffy nose, wet cough, dry cough, wheeze, or trouble breathing. The illness event ended with 2 symptom free days. Illness events were further classified as URI (2 or more consecutive days of runny or stuffy nose, dry cough, or trouble breathing) or LRI (2 or more consecutive days of any upper respiratory symptoms and either wet cough or wheezing or both for at least 1 day by calendar diary report). Duration was calculated for each illness. The comparability of this surveillance system in comparison with conventional clinical evaluation and diagnoses has been reported (22).
Although data were collected for 18 months, this analysis focused on the first 6 months of life, when breastfeeding is expected to have its greatest effect. The breastfeeding status of the infant was classified for each 2-week follow-up interval (see below) as fully breastfed, partially breastfed, or not breastfed. A parent was classified as having asthma on the basis of his or her self-report. Parents were classified as atopic on the basis of a questionnaire report of having had hay fever or desensitization injections. The high-risk season was the 4 months of each year of the study with the largest number of respiratory syncytial virus isolates.
Calculation of incidence rates
Incidence rates of the various types of illness were calculated as the ratio of the number of events to the number of days at risk, excluding days with illness in progress and days when the subject was not under surveillance (18). Days with an LRI in progress were excluded from the days at risk for URI. Rates are expressed on an annualized basis, reflecting 365 days at risk. This approach was used so that the rates could be compared across categories. Confidence intervals for the rates were calculated by using an estimate of the standard error based on the Poisson distribution (23). Because of interdependence of the illness events, standard errors are likely to be underestimated and confidence intervals inappropriately narrow.
Duration of illness
Duration of illness was measured from the onset of symptoms to the last day of symptoms followed by 2 symptom-free days.
Multivariate methods were used to control for potential confounding factors and to test for effect modification. In analyses of determinants of incident illnesses, the outcome variable was the occurrence of illness during 2-week, at-risk intervals. An at-risk interval was defined as a 2-week interval of follow-up time preceded by 2 days free of any respiratory tract illness. Breastfeeding status for each 2-week interval was classified on the basis of the surveillance call covering the first day of the interval.
The independent variables considered in the multivariate analysis included the fixed factors of birth order (firstborn, not firstborn), sex, ethnicity (Hispanic, non-Hispanic), parental asthma, parental atopy, household income (less than $10,000, $10,000-$40,000, greater than $40,000), and maternal education (12 years or less, 13-15 years, 16 years or more). Temporally varying factors were age (0-2 months, 3-4 months, 5-6 months), season as six 2-month groups, day-care attendance (none, 1-4 hours with n
Full details of the analytic strategy have been reported (18). The final analyses were performed b; using the generalized estimating equation approach described by Zeger and Liang (24) that takes into account the correlation structure in estimating regression coefficients and their standard errors. We also used the generalized estimating equation approach to assess the effect of breastfeeding on illness duration after controlling for other factors. The logarithm of the duration minus 1 day was used as the dependent variable. Therefore, the antilogarithm of the model coefficient estimates the ratio of illness duration (less 1day) for two categories, for example, fully breastfed compared with not breastfed.
All analyses were performed using the SAS software (25). The generalized estimating equations were applied using a SAS program supplied by Dr. Scott Zeger, The Johns Hopkins University, Baltimore, Maryland.