Table 3. Comparison of truncated Fourier series models for seasonality of antenatal care attendance, maternal dietary intake, and infant birth size, VHEMBE study, South Africa, 2012–2013.
| 1st order (unimodal) | 1st-2nd order (bimodal) | 1st-3rd order (trimodal) | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| vs. Null: | vs. Null: | vs.1st order: | vs. Null: | vs. 1st-2nd order: | |||||||
| Outcome | Nd | F (df) | p-value | F (df) | p-value | F (df) | p-value | F (df) | p-value | F (df) | p-value |
| Antenatal care attendancea | |||||||||||
| ≥ 4 total visits | 612 | 2.63 (2) | 0.07 | 2.40 (4) | 0.05 | 2.47 (2) | 0.09 | 2.10 (6) | 0.05 | 1.26 (2) | 0.29 |
| First visit ≤ 12 weeks | 605 | 1.00 (2) | 0.37 | 0.97 (4) | 0.42 | 0.87 (2) | 0.42 | 0.74 (6) | 0.62 | 0.47 (2) | 0.62 |
| Maternal diet (% of energy)b | |||||||||||
| Carbohydrate | 751 | 30.1 (2) | <0.01 | 16.6 (4) | <0.01 | 2.97 (2) | 0.05 | 11.9 (6) | <0.01 | 2.25 (2) | 0.11 |
| Fat | 751 | 32.1 (2) | <0.01 | 17.8 (4) | <0.01 | 3.24 (2) | 0.04 | 12.7 (6) | <0.01 | 2.32 (2) | 0.10 |
| Protein | 751 | 1.32 (2) | 0.27 | 0.91 (4) | 0.46 | 0.50 (2) | 0.61 | 1.44 (6) | 0.20 | 2.49 (2) | 0.08 |
| Infant birth size (z-score)c | |||||||||||
| Birth weight | 751 | 0.20 (2) | 0.82 | 2.29 (4) | 0.06 | 4.38 (2) | 0.01 | 1.68 (6) | 0.12 | 0.47 (2) | 0.63 |
| Birth length | 746 | 4.33 (2) | 0.01 | 5.01 (4) | <0.01 | 5.64 (2) | <0.01 | 3.65 (6) | <0.01 | 0.92 (2) | 0.40 |
| Head circumference | 746 | 1.24 (2) | 0.29 | 2.57 (4) | 0.04 | 3.88 (2) | 0.02 | 1.92 (6) | 0.07 | 0.65 (2) | 0.52 |
Data are likelihood ratio or F test statistics comparing generalized linear models regressed on Fourier terms for date of birth. Four models were compared for each outcome: (1) intercept plus covariates only (Null model); (2) including the first order Fourier pair (unimodal); (3) including the first and second order Fourier pairs (bimodal); and (4) including the first through third order Fourier pairs (trimodal). F tests were used to jointly test Fourier coefficients for each nested model.
a. Models adjusted for maternal parity, HIV status, education, marital status, and pregnancy desire; father’s supportiveness of the pregnancy; household income and distance to a main road; and duration of pregnancy. Inverse probability of censoring weights applied to adjust for missing data.
b. Outcomes are dietary intake type as a percentage of total energy (kJ) in the month before delivery. Models adjusted for maternal parity, HIV status, height, education, and marital status; father’s supportiveness of the pregnancy; household income; and duration of pregnancy.
c. Outcomes are gestational age- and sex-adjusted z-scores. Models adjusted for maternal parity, HIV status, height, education, and marital status; and household income.
d. Number of participants included in each model, as determined by observed outcome data. Multiple imputation was used to impute missing covariates.