Table III.
Treatment effects for 0–6 month continuous and categorical anthropometric outcomes*† among the longitudinal analysis subset‡
| Variables§ |
Arm 1 vs Arm 3 |
Arm 2 vs Arm 3 |
Arm 1 vs Arm 2 |
|||
|---|---|---|---|---|---|---|
| Continuous variables | Adjusted mean difference (95% CI)* | P value* | Adjusted mean difference (95% CI)* | P value* | Adjusted mean difference (95% CI)* | P value* |
| Length, cm | 0.50 (0.30–0.71) | <.001 | 0.33 (0.12–0.53) | .002 | 0.18 (−0.03 to 0.38) | .09 |
| LAZ¶* | ||||||
| DRC | 0.44 (0.24–0.64) | <.001 | 0.27 (0.09–0.46) | .004 | 0.17 (−0.02 to 0.36) | .08 |
| Guatemala | 0.06 (−0.11 to 0.23) | .51 | 0.18 (0.01–0.34) | .03 | −0.12 (−0.30 to 0.05) | .17 |
| India | 0.21 (0.01–0.40) | .04 | 0.15 (−0.05 to 0.35) | .15 | 0.06 (−0.03 to 0.15) | .57 |
| Pakistan | 0.33 (0.12–0.55) | .002 | 0.13 (−0.08 to 0.34) | .22 | 0.2 (0.00–0.40) | .05 |
| Weight, g | 104.6 (45.4–163.8) | <.001 | 61.5 (4.0–119.1) | .04 | 43.1 (−15.1 to 101.2) | .15 |
| WAZ | 0.19 (0.09–0.28) | <.001 | 0.13 (0.04–0.22) | .005 | 0.06 (−0.15 to 0.26) | .20 |
| BMI, kg/m2 | 0.08 (−0.03 to 0.19) | .17 | 0.04 (−0.06 to 0.15) | .44 | 0.03 (−0.07 to 0.14) | .53 |
| BMI-for-age z score | 0.06 (−0.02 to 0.14) | .15 | 0.04 (−0.04 to 0.12) | .33 | 0.02 (−0.06 to 0.10) | .60 |
| Weight/length ratio | 0.12 (0.04–0.20) | .003 | 0.07 (−0.01 to 0.14) | .10 | 0.06 (−0.02 to 0.14) | .15 |
| WLZ | −0.01 (−0.09 to 0.07) | .77 | −0.03 (−0.10 to 0.05) | .48 | 0.02 (−0.06 to 0.09) | .69 |
| HC, cm | 0.08 (−0.04 to 0.20) | .21 | 0.05 (−0.07 to 0.16) | .41 | 0.03 (−0.09 to 0.15) | .63 |
| HC-for-age z score | 0.07 (−0.02 to 0.17) | .12 | 0.06 (−0.03 to 0.15) | .20 | 0.02 (−0.08 to 0.11) | .72 |
| Categorical variables | Relative risk (95% CI)† | P value† | Relative risk (95% CI)† | P value** | Relative risk (95% CI)† | P value† |
| LAZ <–2 | 0.76 (0.66–0.87) | <.001 | 0.77 (0.67–0.88) | <.001 | 0.98 (0.84–1.14) | .82 |
| WAZ <–2** | 0.83 (0.72–0.95) | .008 | 0.84 (0.73–0.97) | .02 | 0.98 (0.84–1.13) | .78 |
| BMI-for-age z score <–2 | 0.89 (0.75–1.04) | .13 | 0.92 (0.78–1.07) | .27 | 0.97 (0.82–1.13) | .68 |
| WLZ <–2†† | 1.00 (0.82–1.23) | .98 | 1.12 (0.92–1.37) | .26 | 0.89 (0.73–1.09) | .27 |
| HCAZ <−2 | 0.99 (0.83–1.17) | .88 | 0.92 (0.78–1.09) | .34 | 1.07 (0.90–1.27) | .42 |
DRC, Democratic Republic of the Congo.
Adjusted mean differences, 95% CI, and P values comparing the means of continuous growth outcomes for the pairwise comparisons of the intervention arms were obtained from general linear models with generalized estimating equations for the outcome of interest, adjusted for country, cluster nested within country, maternal education level (α = 0.10), primiparous/multiparous birth indicator, interaction between intervention x primiparous/multiparous birth indicator (α = 0.10), chronological age at the visit, interaction between intervention × country (α = 0.10), sex and its interaction with intervention (α = 0.10) while modeling the within infant correlations across visits with an auto-regressive order 1 (ie, AR[1]). As these are exploratory analyses, no correction for multiple comparisons were made.
Unless otherwise specified, relative risks, 95% CI, and P values comparing proportions of categorical outcomes for the pairwise comparisons of the intervention arms are obtained from log-binomial generalized linear models with generalized estimating equations to estimate parameters while adjusting for country, cluster nested within country, maternal education level (α = 0.10), primiparous/multiparous birth indicator, interaction between intervention × primiparous/multiparous birth indicator (α = 0.10), chronological age at the visit, interaction between intervention × country (α = 0.10), sex and its interaction with intervention (α = 0.10) while modeling the within infant correlations across visits with an auto-regressive order 1 (ie, AR[1]). The robust Poisson generalized linear model was fitted when the log-binomial model failed to converge. If this model also failed to converge, a model which did not account for the clustering of the data was used. As these are exploratory analyses, no correction for multiple comparisons were made.
After we excluded extreme invalid measurements as determined by expert internal manual review, the longitudinal analysis subset includes all live-born infants with birth length measurements measured by 7 days of age and at least 2 follow-up visits (1-, 3-, or 6-month[s]) completed.
All LAZ, WAZ, WLZ, BMI-for-age z score, and HC-for-age z scores were calculated using the expanded tables of the WHO Child Growth Standards.1 LAZ, WAZ, BMI-for-age z score, WLZ, and HC-for-age z score were within the biologically plausible range according to WHO standards (—6≤ LAZ or WAZ ≤6, 5≤ BMI-for-age z score or WLZ or HC-for-age z score ≤5). If an infant was found to have a biologically implausible LAZ or WAZ at a visit, all growth outcomes at the visit were set to missing. If an infant was found to have a biologically implausible BMI-for-age z score, WLZ, or HC-for-age z score at a visit, only the corresponding measurement and z score at the visit were set to missing. WLZ could not be obtained for infants with a length of less than 45.0 cm at any visit due to limitations in the WHO standards and were set to missing for that visit.1
In the longitudinal analysis of LAZ, the intervention × country interaction was significant at α = 0.10 (P = .07) and thus was controlled for in the model.
For the binary growth outcomes WAZ <–2, the log-binomial generalized linear model failed to converge, and thus the specified robust Poisson model† was fitted.
Due to clusters with no infants having a WLZ <–2 the specified log-binomial generalized linear model† did not converge, and therefore, the specified log-binomial generalized linear model† was fitted without adjusting for clusters.