The present study used data from the first two visits of a population-based prospective cohort study in Bandar Abbas city entitled Bandar Abbas Pregnancy Cohort (BAPC). Since 2014, BAPC is investigating the effects of lifestyle and environmental factors on maternal wellbeing and child growth and development in Bandar Abbas city, the capital of Hormozgan Province in the south of Iran. Through door-by-door inquiry in the suburban neighborhoods, BAPC has recruited over 1,500 pregnant women. Any pregnant woman aged above 16 years who is residing in the suburban areas was eligible to participate in the study. Those who were unable to communicate in Farsi or unwilling to participate were excluded. Following providing the informed consent, data from subjects and their babies are gathering through four visits in pregnancy, 1, 6, and 12 months after birth. The details of the cohort methodology have been published elsewhere (13).
The present paper used the data of 897 subjects who completed the second visit of BAPC (participation rate: 92%, response rate: 95.42%) during September 2016 to May 2019. For this study, the following subjects were further excluded: miscarriage (n = 24, 2.55%), stillbirth or multiple pregnancies (n = 8, 0.85%), and self-report cigarette smoking during pregnancy (n = 4, 0.44%). Therefore, data of 861 live singleton pregnancies (mean ± SD gestational age at recruitment: 22.62 ± 9.66 weeks) was included in the analysis (Fig. 1). Data of the first visit of BAPC is collected during pregnancy and the second visit is performed during post-partum period (0–42 days post-delivery) by in-person or telephone visit. The main study`s outcome was set as LBW and was defined as birth weight below 2,500 grams (yes/no) (1). The main exposures were water pipe and caffeine intake during pregnancy, which both were measured during the first visit of BAPC. Water pipe during pregnancy (regular/only for leisure/never) was measured by a checklist recommended by the World Health Organization (WHO) (14), and was merged into yes/no categories. The checklist was validated by a group of healthcare professionals and epidemiologists while the reliability was checked on a subset of BAPC subjects (n = 25, Cronbach`s alpha = 0.78). The checklist also contained additional information on age at water pipe initiation, duration of water pipe smoking, number of water pipe sessions per day, and Environmental Tobacco Smoking (ETS). Caffeine was measured by a structured checklist using Bunker categorization as a guideline (15). The checklist was validated by a team of nutritionist, epidemiologist and gynecologist to measure dietary intake of caffeine during pregnancy. The checklist contained questions on daily intake of any type of caffeinated beverages available in the local market (including black coffee, instant coffee, black tea, green tea, hot chocolate, soft drinks) and caffeinated medications (e.g., painkillers). The reliability was further checked on a subset of BAPC subjects (n = 15, Cronbach`s alpha = 0.64). The cumulative daily dose of caffeine was then dichotomized to normal (0–99 mg/day) and high (≥ 100 mg/day) (16). Number of prenatal care visits to healthcare centers and/or gynecologist office was dichotomized into regular (i.e., at least nine visits recommended by the national guideline,) and irregular (i.e., less than nine visits). Monthly expenditure as a proxy of socio-economic status was defined as monthly average of usual household expenses during the last six months on housing, food, clothing, and healthcare.
Confounders were selected using the Change-In-Estimate (CIE) strategy. The CIE selects covariates on the basis of how much their control changes exposure effect estimates, i.e. amount of confounding by the covariate. Suppose RRa and RRu denote the estimated risk ratio with and without adjustment for the covariate; then RRa/RRu is traditionally used to judge change importance. By this strategy, an "important covariate" was determined as whether the change in the exposure effect estimate from adjusting for the covariate falls outside an interval of practical equivalence; e.g., 0.91 < RRa/RRu < 1.1 (which is the 10%-change rule for the risk ratio modified to be proportionally symmetric) (17). Based on the CIE strategy, duration of water pipe smoking, maternal education, intake of iron supplement during pregnancy, infant sex, preterm birth, history of LBW infant, and monthly expenditure were included in the final regression model. Adjusted Relative Risks (ARRs) for the effects of the main exposures on LBW were calculated using Modified Poisson regression models (18). The Miettinen formula was applied to calculate PAFs for caffeine intake and water pipe (11). Accordingly, we estimated the PAFs from the estimated ARRs for the exposures of interest (water pipe and caffeine intake, both as dichotomous variables). The prevalence of exposure among cases (pc) was estimated as 22.3% for water pipe smoking and 67.63% for high caffeine intake. The PAF finally estimated as: PAF = pc(1 − 1/RR)(12).
GIFs were calculated using the following formula:
Where, Pi denoted the proportion of the population in exposure category i (fact) (8.59% for water pipe smoking, 56.98% for high caffeine intake). Pi` denoted the proportion of the population in exposure category after an intervention or other change is implemented (counter fact) (3% as hypothesized prevalence of water pipe). Using the WHO recommended safety threshold of 100 mg/day caffeine in pregnancy, we hypothesized that an effective intervention would successfully decrease the proportion of women with high intake of caffeine to 14.9%. Adjusted relative risks were derived from the modified Poisson regression model (12). GIFs were calculated based on a series of proposed action plans to: A) Decrease caffeine intake to less than 100 mg/day only among subjects with high intake of caffeine; and B) Decrease the prevalence of water pipe to national report of the prevalence of water pipe among Iranian women in reproductive age (i.e. 3%) (5). All the analyses were performed using Stata version 13 (Stata Corp., College Station, TX, USA). P-values less than 0.05 were considered statistically significant for the final model.