Of the 23,662 live-born infants in the study area who were eligible for inclusion in the parent trial, 23,257 (98.2%) were met at one or more home visits during the neonatal period; 17 did not have any axillary measured temperatures; 23,240 contributed 213,636 axillary temperatures measures (mean 9.2 measures/infant) to the analytic dataset. There were 3134 measures of less than 35.0°C - equivalent to 1.47/100 measures (95% CI: 1.42, 1.52) from among 2342 infants, giving a cumulative risk estimate of 10.7% (95% CI: 10.3, 11.1).
All infant-level variables were strongly associated with axillary measures < 35.0°C, especially birth weight (Table ). Compared to babies ≥ 2500 g, very low birth weight (VLBW; < 1500 g) babies were substantially more likely to be hypothermic [PR = 17.3 (95% CI: 12.6, 23.6)], as were those < 2000 g [PR = 5.34 (95% CI: 4.66, 6.12)]. Females (PR = 1.54), early preterm (< 34 weeks, PR = 4.39) and late preterm births (34-37 weeks, PR = 1.64) were also more likely to be hypothermic. Maternal and paternal education and literacy were associated with a lower prevalence of hypothermia. Hypothermia risk was not associated with either of the chlorhexidine interventions provided in the parent trial.
Individual-factor associations between potential confounders and axillary temperature measures throughout the newborn period
Babies born outside a clinic or hospital (90.2% of all infants in the sample) were at a higher risk of hypothermia [PR = 1.62 (95% CI: 1.27, 2.06)], as were those born to mothers assisted at birth by non-skilled personnel [PR = 1.47 (95% CI: 1.19, 1.82)] and babies of mothers reporting at a > 30 min interval between the birth and delivery of the placenta. Upper caste, ethnic group and variables indicating improved socioeconomic status provided a greater protection from hypothermia.
Hypothermia was almost 50% more prevalent among babies for whom breastfeeding was delayed beyond 24 h [PR = 1.49 (95% CI: 1.37, 1.62)) and 75% more prevalent among babies for whom the cord was not cut until after the placenta was delivered [PR = 1.74 (95% CI: 1.31, 2.33)]. Risk was marginally higher among babies bathed after birth, an almost universal practice in this setting [24
] There was little evidence that the timing of the bath within the first day of life was associated with hypothermia risk, although post-bath oil massage was slightly protective. Babies of mothers who reported warming the room, giving the infant a cap and providing skin-to-skin contact during the first 14 days were at similar risk relative to babies of mothers not reporting these practices.
In a multivariate model adjusting for neonatal-level factors (Model 1, Table ), the association between birth weight and hypothermia attenuated only slightly, while the magnitude of the comparative risk estimates for early (< 34 weeks) and late (34-37 weeks) preterm categories were reduced on the log-linear scale by about 55%, but remained significant. Female sex remained strongly associated with hypothermia. Among delivery and newborn care practices added to the model, only delayed initiation of breastfeeding and cutting the cord after placental delivery were associated statistically with hypothermia, while all neonatal-level factors remained strongly associated. Incorporating more distal socioeconomic and parental variables (Model 3) did not change the magnitude of the associations among neonatal-level factors: females remained at 1.49 (95% CI: 1.37, 1.63) times higher risk, VLBW infants were at 11.63 (95% CI: 8.10, 16.70) times higher risk, and infants between 1500-2000 g were at a 4.32 (95% CI: 3.73, 5.00) times greater risk compared to infants weighing ≥ 2500 g. None of the socioeconomic factors was an important predictor of risk; only maternal hypothermia at first study contact after delivery was associated with higher prevalence of hypothermic measures (PR = 1.50 [95% CI: 1.32, 1.71]).
Multivariate analysis of risk factors associated with axillary temperature measures < 35.0°C among newborns of Sarlahi
Hypothermia risk associated with incremental weight differences
Given the strength of the association with infant weight, the risk associated with incremental (50 g) changes in weight was examined (Figure ). After modelling the risk associated with weight as a continuous measure in a spline Poisson regression model, there was only weak evidence that weight of between 3500-3000 g was associated with hypothermia risk (P = 0.09). However, each 100 g decrement in weight between 3000 g and 2500 g was associated with a 7.4% (95% CI: 4.4%, 10.5%) increase in risk of hypothermia and the association became substantially stronger and of greater magnitude between 2500 g and 2000 g [13.5% (9.7%, 17.5%)] and < 2000 g [31.3% (26.3%, 36.5%)].
Adjusted prevalence ratios for axillary temperature < 35.0°C, by weight interval. NB: Ratio adjusted for age at measurement, ambient temperature at measurement and gestational age at birth; reference weight 3500 g.
Influence of seasonality on risk factor analysis
The season during which a child was born was examined for interaction terms for birth weight category, delivery location, birth attendance with skilled personnel, hat-wearing during the first 14 days and caste and ethnicity. There was evidence that the season during which babes were born modified the association between hypothermia and the two lowest birth weight categories and hat wearing by the infant (interaction P-values: < 0.001, 0.06 and 0.003, respectively). In the warm season, relative to newborns ≥ 2500 g, the adjusted prevalence of hypothermia was 23.46 (95% CI: 16.49, 33.37) and 5.05 (95% CI: 4.14, 6.16) times greater in the lowest (< 1500 g) and second-lowest (2000 g-2499 g) weight categories, respectively. The parallel ratios of adjusted prevalence (Adj PR) during the cold season were less: 4.04 (95% CI: 2.34, 7.00) and 3.80 (95% CI: 3.07, 4.71), respectively. Similarly, wearing a hat was protective only during the cold season [Adj PR = 0.71 (0.52 - 0.96)].