Menu
Menu
Menu
Menu
Background: Pica, the craving and purposeful consumption of nonfoods, is poorly understood. We described the prevalence of pica among women on Mfangano Island, Kenya, and examined sociodemographic and health correlates. Methods: Our cross-sectional study included 299 pregnant or postpartum women in 2012. We used a 24-h recall to assess pica, defined as consumption of earth (geophagy), charcoal/ash, or raw starches (amylophagy) and built multivariable logistic regression models to examine sociodemographic and health correlates of pica. Results: Eighty-one women (27.1%) engaged in pica in the previous 24 h, with 59.3% reporting amylophagy and 56.8% reporting geophagy, charcoal, and/or ash consumption. The most common substances consumed were raw cassava (n = 30, 36.6%), odowa, a chalky, soft rock-like earth (n = 21, 25.6%), and soil (n = 17, 20.7%). Geophagy, charcoal, and/or ash consumption was negatively associated with breastfeeding (OR = 0.38, 95% CI: 0.18–0.81), and amylophagy was associated with pregnancy (OR = 4.31, 95% CI: 1.24–14.96). Pica was more common within one of six study regions (OR = 3.64, 95% CI: 1.39–9.51). We found no evidence of an association between food insecurity and pica. Conclusion: Pica was a common behavior among women, and the prevalence underscores the need to uncover its dietary, environmental, and cultural etiologies.
This cross-sectional study used baseline data collected between December 2012–March 2013 from a larger prospective study examining fishing livelihoods, fish consumption, and early child nutrition [39,40]. Details on the larger study have been described elsewhere [41,42]. Briefly, households (N = 303) were selected using stratified random sampling proportional to regional population in which Regions 1–5 were on the main island and Region 6 describes the satellite island: Region 1 = 66; Region 2 = 16; Region 3 = 81; Region 4 = 46; Region 5 = 60; and Region 6 = 34. Given the larger study’s focus on early child nutrition, enrollment criteria were (1) having at least one child less than 2 years of age and (2) living on Mfangano Island. Data on all covariates were available for 299 households. Local enumerators conducted surveys in the local language, Dholuo, and data collection tools were developed from validated measures and locally adapted. Heads of households provided consent to participate in the study, with women providing consent for their own participation. The Committee for Protection of Human Subjects at the University of California, Berkeley and the Ethical Review Committee at the Kenya Medical Research Institute approved the study protocol (CPHS 2010-01-608; SSC 2334). Women self-reported pica during a 24-h recall survey in which they were asked if they had consumed any of the following: ash, charcoal, odowa (a chalky, soft rock-like earth), soil/other, uncooked foods (cassava, rice, etc.), and other nonfood items (see Supplementary File S1). We operationalized pica into three categories: (1) the consumption of earth, soil, or clay (geophagy) and/or charcoal/ash, (2) amylophagy, the consumption of raw starchy foods, and (3) any nonfood craving. We combined geophagy and charcoal/ash consumption due to the small number of women who consumed charcoal (n = 9, 11.1%) or ash (n = 4, 4.9%), and their physical properties being similar to those of earth. Sociodemographic information was collected using standardized questionnaires. Maternal education was quantified as the highest level attained and categorized into none or some primary school, primary school, some secondary school, secondary school or higher. We used principal components analysis to create an asset index that included ownership of items such as electricity, livestock, and type of flooring or roofing materials [43]; however, as eigenvalues did not account for substantial variation, we retained a total count as our asset score. Household food insecurity was measured with the Household Food Insecurity Access Scale, a 9-item questionnaire that generates a food insecurity score from 0–27 [44]. A maternal morbidity score was calculated using the Medical Outcomes Survey–HIV (MOS–HIV), a validated, self-reported measure of health-related quality-of-life, and was centered at 0 and normalized by standard deviation [45,46]. While the MOS–HIV metric is correlated with measures of HIV disease progression (e.g., CD4 count), we did not collect information about HIV status [47]. During the 24-h recall, women were also asked if they were ill in the prior day, if they consumed any tablets, herbs, or medicine and what types, and if they consumed any deworming medications in the last 3 months. Maternal breastfeeding practices and pregnancy status were ascertained using maternal self-report. We also included the six regions of the study site as indicator variables to assess regional differences. We first descriptively characterized the sample and the types of pica substances women reported consuming. We then conducted multivariable logistic regression models for any pica, geophagy/charcoal/ash, and amylophagy based on a priori identification of variables that have been hypothesized to be associated with pica behavior in previous studies and included maternal age, maternal education, current pregnant and breastfeeding status, morbidity score, food insecurity, and number of people in the household. We included region in the models because we hypothesized the potential for cultural transmission of pica behavior and environmental differences in access to common pica materials (e.g., odowa). Maternal health factors (e.g., being ill the day before, and consumption of iron tablets, antimalarial or antiretroviral drugs, or deworming medication) were initially hypothesized to be associated with pica; however, these were omitted to improve estimation as they were not significantly different between groups that did and did not participate in pica, and their omission did not alter the magnitude of coefficients of the final models. We also conducted a sensitivity analysis to include interviewer fixed effects and did not find substantial differences in estimates (Supplementary File S2). All statistical analyses were performed using Stata 14 [48].
