Effects of interventions
Comparison one: controlled trials of exclusive versus mixed breastfeeding for four to six months, developing countries
Two studies were found in this category, both from the same group of investigators and involving the same study setting (Honduras). The first of these studies, Cohen 1994a, involved term infants unselected for birthweight but included 29 infants (19.9%) weighing less than 2500 g at birth. The second, Dewey 1999a, was restricted to term infants weighing less than 2500 g at birth. The quality ratings of these two trials were not high for several reasons. First, in both trials, allocation was within clusters defined by weeks, rather than to individual women, yet the results were analyzed with individual women and infants as the units of analysis; any similarities in outcome within weeks (intracluster correlation) would tend to reduce the true effective sample size and thereby overestimate the precision (i.e., underestimate the variance) of the results. Second, the first trial allocated the weeks by alternation, rather than by strict randomization, thereby creating a potential for nonconcealment and uncontrolled confounding bias at enrollment (although there is no evidence that such bias actually occurred). Third, the published results were not based on analysis by intention-to-treat. Most of the babies not analyzed in these two trials were truly lost to follow-up; however, rather than excluded for noncompliance, the latter were restricted to four babies (three in the exclusive breastfeeding (EBF) group, one in the mixed breastfeeding (MBF) group) in the first trial and three babies (all three in the exclusive breastfeeding group) in the second trial. Moreover, the investigators have provided (unpublished) data on weight and length gain on five of the nine dropouts in the second Honduran trial (three of the nine moved away before six months), thereby substantially reducing the potential for selection bias in the analysis of that trial.
Most importantly, despite the above-noted methodological problems, these two trials are the only studies uncovered by our search that used an experimental design to specifically address the four to six months versus 'about six months' debate. Thus, at least with respect to bias due to known and unknown confounding variables, these trials are methodologically superior to any of the observational studies included in this review despite their methodological imperfections. Furthermore, the investigators made a considerable effort to ensure compliance with the assigned allocation and to standardize the training of the observers who performed the anthropometric measurements, thereby reducing the random error (improving the precision) of these measurements. Finally, detailed comparisons between trial participants and eligible nonparticipants demonstrated no differences that would detract from the external validity (generalizability) of the trials' findings, at least for the specific type of setting where the study was conducted (an urban, low-income population in Honduras).
For all analyses, the two mixed breastfeeding groups (one of which was intended to maintain frequency of breastfeeding) in the first trial were combined for the purposes of this analysis. Monthly weight gain from four to six months was nonsignificantly slightly higher among infants whose mothers were assigned to continued exclusive breastfeeding (mean difference (MD) +20.78; 95% confidence interval (CI) -21.99 to +63.54 g/mo) (Analysis 1.1). Thus the 95% CI is statistically compatible with a weight gain only 22 g/mo lower in the EBF group, which represents approximately 5% of the mean and 15% of the standard deviation (SD) for the monthly weight gain. Weight gain from six to 12 months was almost identical in the two groups (MD -2.62; 95% CI -25.85 to 20.62 g/mo) (Analysis 1.2).
For length gain from four to six months, the MD was 1.0 mm/mo (95% CI -0.40 to +2.40 mm/mo) (Analysis 1.3); the lower confidence limit represents only 2% of the mean and 8% of the SD for monthly length gain. As with weight gain, length gain from 6 to 12 months was nearly identical in the two groups (MD -0.04; 95% CI -0.10 to 0.02 cm/mo) (Analysis 1.4).
Weight-for-age, length-for-age, and weight-for-length z-scores at six months were all nonsignificantly higher in the EBF group (MD +0.18; 95% CI -0.06 to +0.41 (Analysis 1.5); MD +0.11; 95% CI -0.11 to +0.33 (Analysis 1.6); and MD +0.09; 95% CI -0.13 to +0.31 (Analysis 1.7), respectively).
The impact of the small sample size of the two Honduran trials is evident when examining the risk of undernutrition, as represented by anthropometric z-scores less than -2 at six months. For weight-for-age, the pooled risk ratio (RR) was 2.14 (95% CI 0.74 to 6.24) (Analysis 1.8), which is statistically compatible with a six-fold increase in risk. The results were somewhat more reassuring for length-for-age (RR 1.18; 95% CI 0.56 to 2.50) (Analysis 1.9) but not for weight-for-length (RR 1.38; 95% CI 0.17 to 10.98) (Analysis 1.10).
All hematologic results (Analysis 1.11 to Analysis 1.19) are based on the first Honduras trial (Cohen 1994a), since in the second trial (Dewey 1999a, restricted to low birthweight infants), infants with low hemoglobin concentrations at two and four months were supplemented with iron. A nonsignificantly higher proportion of infants in the exclusively breastfed group received iron supplements from six to nine months (RR 1.20; 95% CI 0.91 to 1.58) (Analysis 1.11). This is consistent with the significantly lower average hemoglobin concentration at six months in the exclusively breastfed group (difference = -5.00 (95% CI -8.46 to -1.54) g/L) (Analysis 1.12). A nonsignificantly higher proportion of exclusively breastfed infants had a hemoglobin concentration below 110 g/L at six months (RR 1.20; 95% CI 0.91 to 1.58) (Analysis 1.13). Similarly, mean plasma ferritin concentration was significantly lower at six months in the exclusively breastfed infants (difference = -18.90 (95% CI -37.31 to -0.49) mcg/L) (Analysis 1.17), with a RR for a low (less than 15 mcg/L) ferritin concentration of 2.93 (95% CI 1.13 to 7.56) (Analysis 1.19).
In the second trial, no significant effect was seen on the proportion of infants with a low zinc concentration (less than 70 mcg/dL) at six months (RR 0.75; 95% CI 0.43 to 1.33) (Analysis 1.20).
In the pooled results from both Honduran trials, no significant difference was seen between the EBF and MBF groups for the percentage of days with fever (Analysis 1.21), cough (Analysis 1.22), or nasal congestion (Analysis 1.23), nasal discharge (Analysis 1.24), hoarseness (Analysis 1.25), or diarrhea (Analysis 1.26) from four to six months, nor for fever (Analysis 1.27), nasal congestion (Analysis 1.28), or diarrhea from six to 12 months (Analysis 1.29).
Again based on pooled results from both trials, mothers in the exclusively breastfed group reported that their infants crawled at an average of -0.80 (95% CI -1.26 to -0.34) months sooner (Analysis 1.30). No difference was seen, however, in the mean age at which the infants were reported to have first sat from a lying position (average MD -0.22 (95% CI -0.91 to 0.46) months), random-effects (Analysis 1.31). The results from the two Honduras trials (Cohen 1994a; Dewey 1999a) differed with respect to maternal reports of walking by 12 months (Analysis 1.32), with a significantly lower proportion of exclusively breastfed infants reported as not having walked by 12 months in the first trial (RR 0.66; 95% CI 0.45 to 0.98) (Cohen 1994a), but a nonsignificantly higher proportion not having done so in the second trial (RR 1.12; 95% CI 0.90 to 1.38) (Dewey 1999a), with statistically significant (P < .01) heterogeneity between the two trials.
Mothers in the exclusively breastfed group (from the two trials combined) had a statistically significantly larger weight loss from four to six months (MD 0.42; 95% CI 0.02 to 0.82) kg) (Analysis 1.33). Women in the exclusively breastfed group were also nonsignificantly less likely to have resumed menses by six months postpartum (RR 0.58; 95% CI 0.33 to 1.03); the effect was statistically significant in the second Honduras trial when considered alone (RR 0.35; 95% CI 0.14 to 0.91) (Dewey 1999a) (Analysis 1.34).
Comparison two: observational studies of exclusive versus mixed breastfeeding for three to seven months, developing countries
The main concern in using an observational design to compare outcomes with EBF versus MBF is confounding due to differences in socioeconomic status, water and sanitation facilities, parental size (a proxy for genetic potential), and (perhaps most importantly) weight and length at the time complementary foods were first introduced in the mixed breastfeeding group. The latter source of confounding (i.e., by indication) will arise if poorly-growing infants are more likely to receive complementary foods.
Four cohort studies in this category from Peru (Brown 1991a), the Philippines (Adair 1993a), Senegal (Simondon 1997a), and Iran (Khadivzadeh 2004) found no evidence of confounding by indication, Adair 1993a found no confounding by several other potential factors, and (in unpublished data provided by the authors). Simondon 1997a calculated adjusted means for weight and length gain from four to six months. Nonetheless, the inability of observational studies to control for subtle (and unknown) sources of confounding and selection bias suggests the need for cautious interpretation. All four studies reported on monthly weight gain from four to six months (Analysis 2.1). The MD was -10.10 (95% CI -27.68 to +7.48) g/mo, a lower confidence limit compatible with a deficit of only 7% of the mean and less than 15% of the SD for monthly weight gain. The Simondon 1997a study also reported on monthly weight gain from six to nine months (difference = -6.00 (95% CI -54.15 to +42.15) g/mo) (Analysis 2.2). All four studies also reported on monthly length gain from four to six months (Analysis 2.3); the MD was 0.04 (95% CI -0.02 to 0.11) cm/mo, a lower confidence limit statistically compatible with a reduced length gain in the EBF group less than 2% of the mean and 4% of the SD. The Simondon 1997a study also reported on monthly length gain from six to nine months (Analysis 2.4), and again the results excluded all but a small reduction in the exclusively breastfed group (difference = 0.04 (95% CI -0.06 to 0.14) cm/mo).
Onayade 2004 actually reported significantly higher absolute weights at both five and six months in the EBF group but did not analyze weight gains; the absence of control for confounding differences between the EBF and MBF groups, as well as the possibility of reverse causality (i.e., those infants with lower weights may have been more likely to receive complementary feeding) argue for cautious interpretation, however.
The Simondon 1997a study also provided (unpublished) data on anthropometric z-scores and mid-upper arm circumference. EBF was associated with nonsignificantly higher MD z-scores at six to seven and nine to 10 months: +0.13 (95% CI -0.09 to +0.35) (Analysis 2.5) and +0.09 (95% CI -0.15 to +0.33) (Analysis 2.6), respectively, for weight-for-age; +0.04 (95% CI -0.14 to +0.22) (Analysis 2.7) and +0.11 (95% CI -0.09 to +0.31) (Analysis 2.8), respectively, for length-for-age; and +0.11 (95% CI -0.09 to +0.31) (Analysis 2.9) and +0.01 (95% CI -0.21 to +0.23) (Analysis 2.10), respectively, for weight-for-length. The risk ratio for low (less than -2) z-scores at six to seven and nine to 10 months were 0.92 (95% CI 0.54 to 1.58) (Analysis 2.11) and 0.93 (95% CI 0.64 to 1.36) (Analysis 2.12), respectively, for weight-for-age; 1.20 (95% CI 0.57 to 2.53) (Analysis 2.13) and 1.21 (95% CI 0.62 to 2.37) (Analysis 2.14), respectively, for length-for-age; and 0.42 (95% CI 0.12 to 1.50) (Analysis 2.15) and 0.82 (95% CI 0.39 to 1.71) (Analysis 2.16), respectively, for weight-for-length. Mid-upper arm circumference was nonsignificantly higher in the EBF group at both six to seven and nine to 10 months: MD 0.20 (95% CI -0.04 to 0.44) cm (Analysis 2.17) and 0.10 (95% CI 0.16 to 0.36) cm (Analysis 2.18), respectively.
Khadivzadeh 2004 found a lower incidence of both gastrointestinal (11 versus 27%; RR 0.41; 95% CI 0.21 to 0.78) (Analysis 2.19) and respiratory (23 versus 35%; RR 0.68; 95% CI 0.43 to 1.06) Analysis 2.20) infection at four to six months in the EBF group. Onayade 2004 reported corresponding crude ORs of 0.02 (95% CI 0.01 to 0.09) and 0.43 (95% CI 0.17 to 1.00), respectively, but did not provide numerators and denominators and did not control for confounding differences between the EBF and MBF groups.
Huffman 1987 reported a longer median duration of lactational amenorrhea associated with EBF (for at least seven months) versus MBF (16.1 versus 15.3 months, respectively), but means and SDs were not reported. In a multivariate (Cox) regression model adjusting for maternal education, parity, religion, and weight, EBF for at least six months was associated with a significantly longer time to resumption of menses versus EBF for less than one month, but no direct comparison was reported versus MBF. Simondon 1997a reported a lower risk of resumption of menses by six to seven months (Analysis 2.21) in the EBF group: crude RR 0.19 (95% CI 0.05 to 0.79), adjusted odds ratio (OR) 0.19 (95% CI 0.04 to 0.86).
Cross-sectional studies share all of the methodological shortcomings of other observational designs (see above) plus one important additional one: selective loss to follow-up. In particular, children who die, are hospitalized, or are referred to a site other than the one under study, may be more likely to experience morbidity or suboptimal growth. If such (unstudied) infants are more heavily represented in one of the feeding groups, the resulting comparison will be biased.
One large cross-sectional study from Chile (Castillo 1996) reported a similar risk of weight-for-age z-score less than -1 and height-for-age z-score less than -1 from three to five and six to eight months in the two feeding groups, but the prevalences, CIs, and standard errors for the reported prevalence ratios are not published, thus precluding any assessment of sampling variation.
Comparison three: observational studies of prolonged (more than six months) exclusive versus mixed breastfeeding, developing countries
One small cross-sectional study from Pune, India (Rao 1992) permitted analysis only of male infants, since a relatively large fraction of female infants in the MBF group received artificial feeding in the first six months of life. The results (Analysis 3.1) showed a nonsignificant reduction of low (less than 75% of the reference mean) weight-for-age at six to 12 months of age in the exclusively breastfed males (RR 0.61; 95% CI 0.26 to 1.43). The strong possibility of confounding by age, even within the range of six to 12 months (the EBF group is likely to have been younger, on average, and therefore less undernourished), further limits the reported result.
A cohort study from Bangladesh (Khan 1984) reported similar weight and length gains in infants who were exclusively breastfed, those who were breastfed with supplements beginning at six to 11 months, and those who were exclusively breastfed for 12 months and supplemented between 12 and 15 months. Unfortunately, the data are presented only graphically and without standard deviations, thus preventing a quantitative assessment or pooling with data from other studies.
Comparison four: observational studies of exclusive versus mixed breastfeeding for three to seven months, developed countries
A pooled sample of breastfed infants from seven studies carried out in six developed countries (WHO 1994a), a pooled analysis from five countries (two developed, three developing, but in which study women were all literate and of middle to high socioeconomic status) (WHO 1997), a large cohort study nested within a randomized trial in Belarus (Kramer 2000a), and a small study from Sweden (Akeson 1996a) reported on weight gain between three and eight months. WHO 1997 and Kramer 2000a controlled for confounding by indication (size or growth in first three to four months) and other potential confounders using multilevel (mixed) regression analyses. Substantial (I2 = 69%) heterogeneity was observed among the four studies, with considerably larger mean weight gains in both groups from Belarus and a slightly but significantly higher gain in the MBF group (Analysis 4.1). The pooled random-effects MD is -7.95 [-31.84, 15.93] g/mo. Heinig 1993 and Kramer 2000a also reported on weight gain between six and nine months (Analysis 4.2). Again, the results show significant heterogeneity (I2 = 76%) but are dominated by the larger size of the Belarussian study. The pooled random-effects MD is 21.11 [-44.70, 86.91] g/mo. Akeson 1996a, Heinig 1993, and Kramer 2000a reported on weight gain from eight to 12 months (Analysis 4.3); the MD was -1.82 (95% CI -16.72 to +13.08) g/mo, which excludes a reduced length gain in the EBF group of 5% of the mean and 10% of the SD for the Belarusian study.
For length gain at three to eight months (Analysis 4.4), the studies again show significant (I2 = 76%) heterogeneity. Kramer 2000a found a slightly but significantly lower length gain in the EBF group at four to eight months (-0.11 [-0.17, 0.05] mm/mo), whereas the pooled analysis yielded a random-effects average MD of -0.03 [-0.11, 0.06] mm/mo. Heinig 1993 and Kramer 2000a also reported on length gain at six to nine months (MD -0.04; 95% CI -0.10 to 0.01) cm/mo) (Analysis 4.5). For the eight to 12 month period, the results show a slightly but significantly higher length gain in the EBF group (MD +0.09; 95% CI 0.03 to +0.14) cm/mo (Analysis 4.6).
Observational analyses from the Belarusian study (Kramer 2000a) also include data on weight-for-age, length-for-age, and weight-for-length z-scores at six, nine, and 12 months. Means in both the EBF and MBF groups were well above (+0.5 to +0.6) the reference values at all three ages. Nonetheless, the weight-for-age z-score was slightly but significantly lower in the EBF group at all three ages: MD -0.09 (95% CI -0.16 to -0.02) (Analysis 4.7) at six months, -0.10 (95% CI -0.18 to -0.02) (Analysis 4.8) at nine months, and -0.09 (95% CI -0.17 to -0.01) (Analysis 4.9) at 12 months. Length-for-age z-scores were very close to the reference (0) at six and nine months and slightly above the reference (0.15) at 12 months. Again, the EBF group had slightly but significantly (except at 12 months) lower values: MD -0.12 (95% CI -0.20 to -0.04) (Analysis 4.10) at six months, -0.14 (95% CI -0.22 to -0.06) (Analysis 4.11) at nine months, and -0.02 (95% CI -0.10 to +0.06) (Analysis 4.12) at 12 months. Mean weight-for-length z-scores were high and rose (from about 0.65 to 0.80) from six to 12 months, with no significant differences between the EBF and MBF groups at any age: MD +0.02 (95% CI -0.07 to +0.11) (Analysis 4.13) at six months, +0.03 (95% CI -0.06 to +0.12) (Analysis 4.14) at nine months, and -0.08 (95% CI -0.17 to +0.01) (Analysis 4.15 at 12 months.
The prevalence of low (less than -2) z-scores did not differ significantly in the two Belarusian feeding groups for any of the three z-scores at any of the three ages, although the small number of infants with low z-scores provided low statistical power to detect such differences. RRs (and 95% CIs) for low weight-for-age were 0.92 (0.04 to 19.04) (Analysis 4.16) at six months, 1.52 (0.16 to 14.62) (Analysis 4.17) at nine months and 1.15 (0.13 to 10.31) (Analysis 4.18) at 12 months. For length-for-age, the corresponding figures were 1.53 (0.84 to 2.78) at six months (Analysis 4.19), 1.46 (0.80 to 2.64) (Analysis 4.20) at nine months, and 0.66 (0.23 to 1.87) (Analysis 4.21) at 12 months. For weight-for-length, the figures were 0.31 (0.02 to 5.34) (Analysis 4.22) at six months, 1.14 (0.24 to 5.37) (Analysis 4.23) at nine months, and 1.15 (0.13 to 10.31) (Analysis 4.24) at 12 months.
The Belarusian study also provided data on head circumference. No significant differences were observed at six months (difference 0.19 (95% CI 0.06 to 0.32) cm) (Analysis 4.25) or nine months (0.07 (95% CI -0.06 to 0.20) cm) (Analysis 4.26), but the EBF group had a slightly but significantly larger circumference at 12 months (Analysis 4.27): difference = 0.19 (95% CI 0.06 to 0.32) cm.
Heinig 1993 reported nearly identical sleeping time (729 versus 728 minutes/day) in the two groups (Analysis 4.28). Akeson 1996a reported similar total amino acid and essential amino acid concentrations at six months of age in the two feeding groups (Analysis 4.29; Analysis 4.30). Both Kramer 2000a and a cohort study from Finland (Kajosaari 1983) reported on atopic eczema at one year (Analysis 4.31). The two studies showed substantial (I2 = 78%) heterogeneity, with Kajosaari 1983 reporting a significantly reduced risk, but the larger Belarusian study finding a much lower absolute risk in both feeding groups and no risk reduction with EBF; the pooled random-effects average RR was 0.65 (0.27, 1.59) (Analysis 4.31). Although Kajosaari 1983 also reported a reduced risk of a history of food allergy (Analysis 4.32), double food challenges showed no significant risk reduction (RR 0.77; 95% CI 0.25 to 2.41) (Analysis 4.33). Neither Oddy 1999 nor Kramer 2000a found a significant reduction in risk of recurrent (two or more episodes) wheezing in the EBF group (pooled RR 0.79; 95% CI 0.49 to 1.28) (Analysis 4.34).
A small Italian study of hematologic outcomes at 12 months by Pisacane in 1995 reported a statistically significantly higher hemoglobin concentration (117 versus 109 g/L (95% CI for the difference = +4.03 to +11.97 g/L)) (Analysis 4.35), a nonsignificant reduction in anemia (hemoglobin less than 110 g/L) (RR 0.12; 95% CI 0.01 to 1.80) (Analysis 4.36), a nonsignificantly higher ferritin concentration (MD +4.70; 95% CI -6.30 to +15.70 mcg/L) (Analysis 4.37), and a nonsignificant reduction in the risk of low (less than 10 mcg/L) ferritin concentration (RR 0.42; 95% CI 0.12 to 1.54) (Analysis 4.38) among infants in the EBF group. Of note in this study is that the exclusive and mixed breastfeeding continued in both groups until at least 12 months (a criterion for selection into the Pisacane 1995 study).
In the Belarusian study (Kramer 2000a), the EBF group had a significantly reduced risk of one or more episodes of gastrointestinal infection in the first 12 months of life (RR 0.67; 95% CI 0.46 to 0.97) (Analysis 4.39), which was maintained in a multivariate mixed model controlling for geographic origin, urban versus rural location, maternal education, and number of siblings in the household (adjusted OR 0.61; 95% CI 0.41 to 0.93). Importantly, when a mixed-level, multivariate Poisson model was used to estimate the adjusted incidence density ratio (IDR) by age period. From zero to three months (when both groups were exclusively breastfed), the IDR was 0.97 (95% CI 0.46 to 2.04), while at three to six months (when the feeding differed), the protective effect of EBF was strong (IDR 0.35: 95% CI 0.13 to 0.96). No significant reduction in risk was observed for hospitalization for gastrointestinal infection, however (RR 0.79; 95% CI 0.42 to 1.49) (Analysis 4.40). In the above-mentioned Australian cohort study, Oddy 1999 found no significant reduction of risk for one or more episodes of upper respiratory tract infection (Analysis 4.41) in the EBF group (RR 1.07; 95% CI 0.96 to 1.20). Neither Oddy 1999 nor Kramer 2000a found a significantly reduced risk of two or more such episodes (pooled RR 0.91; 95% CI 0.82 to 1.02) (Analysis 4.42). Nor did Oddy 1999 find a significant reduction in risk of four or more episodes of upper respiratory infection (RR 0.82; 95% CI 0.52 to 1.29) (Analysis 4.43) or of one or more episodes of lower respiratory tract infection (RR 1.07; 95% CI 0.86 to 1.33) (Analysis 4.44). Kramer 2000a found a small and nonsignificant reduction in risk of two or more respiratory tract infections (upper and lower combined) (RR 0.90; 95% CI 0.79 to 1.03) (Analysis 4.45). Duijts 2010 reported substantially lower adjusted odds ratios (versus a never-breastfed group) for both upper and lower respiratory tract infection in their EBF group compared with their MBF group in the first six months of life but not for months seven to 12 (data not shown). The combined crude results of Oddy 1999 and Kramer 2000a show a substantial and statistically significant reduction in risk for hospitalization for respiratory tract infection (pooled RR 0.75; 95% CI 0.60 to 0.94) (Analysis 4.46), but the crude risk reduction in Kramer 2000a was nearly abolished and became statistically nonsignificant in a multivariate mixed model controlling for geographic region, urban versus rural location, maternal education and cigarette smoking, and number of siblings in the household (adjusted OR 0.96; 95% CI 0.71 to 1.30). In a study from Tucson, Arizona, (Duncan 1993) reported no difference in the average number of episodes of acute otitis media in the first 12 months of life (Analysis 4.47) in the exclusive versus MBF groups (1.48 versus 1.52 episodes, respectively) (95% CI for the difference -0.49 to +0.41 episodes). Duncan 1993 and Kramer 2000a both found a slightly elevated risk for one or more episodes of otitis media (pooled RR 1.28; 95% CI 1.04 to 1.57) (Analysis 4.48), but Duncan 1993 found a nonsignificant reduction in risk for frequent otitis media (RR 0.81; 95% CI 0.43 to 1.52) (Analysis 4.49). Kramer 2000a recorded only one and two deaths (Analysis 4.50) among the 621 and 2862 Belarusian infants in the EBF and MBF groups, respectively (RR 2.30; 95% CI 0.21 to 25.37).
Reported outcomes beyond infancy have included dental caries, growth and adiposity measures, blood pressure, allergy, cognitive ability, and behaviour. Kramer 2000a reported no difference in decayed, missing, or filled teeth either in the total dentition (Analysis 4.51) or the incisors (Analysis 4.52) at age six years. At 6.5 years, no significant differences were observed for height (Analysis 4.53), leg length (Analysis 4.54), head circumference Analysis 4.55), or waist circumference (Analysis 4.59) between the EBF and MBF groups. Body mass index (BMI, Analysis 4.56), triceps (Analysis 4.57) and subscapular (Analysis 4.58) skinfold thicknesses, hip circumference (Analysis 4.60), and systolic (Analysis 4.61) and diastolic blood pressure (Analysis 4.62) were actually significantly higher in the EBF group, however, although multivariate mixed models with adjustment for clustering and for potential confounding variables yielded nonsignificant adjusted MDs for subscapular skinfold thickness [+0.2 (95% CI -0.02 to +0.5) mm], systolic blood pressure [0.0 (95% CI -1.0 to +0.9) mm Hg], and diastolic blood pressure [-0.3 (95% CI -1.2 to +0.5) mm Hg]. For allergic outcomes at ages five to seven years (Kajosaari 1983, Oddy 1999, and Kramer 2000a), no reduction in risk was observed for atopic eczema (Analysis 4.63), hay fever (Analysis 4.64), asthma (Analysis 4.65), food allergy (Analysis 4.66), allergy to animal dander (Analysis 4.67), or positive skin-prick tests (Analysis 4.68 to Analysis 4.73). Despite higher IQ scores at age 6.5 years observed in intention-to-treat analyses of the breastfeeding promotion intervention in PROBIT (Kramer 2000a), no significant differences were observed in these outcomes in observational comparisons of EBF versus MBF (Analysis 4.74 to Analysis 4.80), except for block designs (Analysis 4.77). The latter difference favouring the EBF group was no longer significant, however, in multivariate mixed models with adjustment for clustering and for potential confounding variables (adjusted MD -0.7; 95% CI -1.6 to 0.3). Teachers' ratings of the PROBIT children's academic performance at age 6.5 years (Analysis 4.81 to Analysis 4.84) were actually higher for all subjects except for mathematics (Analysis 4.83), but the differences all became statistically nonsignificant in multivariate mixed models with adjustment for clustering and for potential confounding variables. Finally, no significant differences were observed in the latter study for parents' or teachers' rating of the children's behaviour at age 6.5 years (Analysis 4.85 to Analysis 4.96).
Comparison five: observational studies of prolonged (more than six months) exclusive versus mixed breastfeeding, developed countries
A small observational cohort study from the Baltimore-Washington area (U.S.) (Ahn 1980) reported "no differences in the overall rates of gain in weight and length" for the first year of life in infants who were exclusively breastfed beyond six months versus those exclusively breastfed for less than six months and mixed breastfed thereafter. The actual data were not reported, however, and thus cannot be assessed quantitatively in this review.
One small Finnish study (Savilahti 1987a) reported no difference in lipid concentrations at nine months among infants exclusively breastfed for nine months versus those exclusively breastfed for six months and mixed breastfed from six to nine months. Similar concentrations were observed for very low density lipoprotein, low density lipoprotein, high-density lipoprotein-2, high-density lipoprotein-3, apoprotein B, and total triglycerides (Analysis 5.1 to Analysis 5.6).