COMPARISON OF COX PROPORTIONAL HAZARDS AND ACCELERATED FAILURE TIME MODELS IN EVALUATING DISCONTINUATION TIME OF EXCLUSIVE BREASTFEEDING

Abstract

Despite WHO/UNICEF recommendation, EBF practice continues to be challenging to mothers, as only 35% of children were exclusively breastfed worldwide. In Nigeria, only 17% of infant were exclusively breastfed. Studies have utilized Cox Proportional Hazards in modelling breastfeeding data. This study compares this method with Accelerated Failure Time in modelling and evaluating time to discontinuation of EBF in Nigeria. Main objective of this research is to compare CPH and AFT modelling strategies and evaluate the model that better describes and predicts EBF discontinuation time in Nigeria. Specific objectives are to estimate average EBF duration and measure effect of its influential factors on mothers and infant. A total of 2,163 eligible mothers whose child not more than 2 years old, were selected from mothers 5 years breastfeeding history (2007 - 2012) of NARHS multistage cluster survey that utilized NPC sampling frame in 2012. EBF discontinuation time was the outcome variable and the eleven explanatory variables were; Mother's age, zone, locality, education, occupation, marital status, SES, ANC visit, place of child delivery, BF initiation status and Postnatal visit. SPSS version 20 and STATA version 12 software packages were used for data analysis. Data were described using frequencies and percentages, EBF pattern was examined using its distribution and survival plots. L.R. test suitability of AFT model. Fitted CPH and AFT models determine predictors of EBF discontinuation time and comparison was made using AIC and confidence bounds around their estimates to determine the model with better precision. Not less than 113(5.2%) mothers were censored. EBF median duration was 6 months (5.9250 - 6.0250). Almost 81% of mothers discontinue EBF on/before child sixth month of birth. Hazard of discontinuing EBF in rural area was 1.0414 times more than hazard of urban slum. Mothers with at least primary education (Time ratio = 1.0728) have positive influence on ·survival time. EBF survivorship decreases as discontinuation time increases. CPH and AFT models identified Mother's Locality {p=0.017(0.7242, 0.9689), p=0.003(1.0245, 1.1244)} and place of child delivery {p=0.000(1.0530, 1.1909), p=0.000(0.9266, 0.9666)} respectively as predictors of EBF duration. Distribution of survival time skewed right and Weibull AFT model (p=0.0005<0.001) with smaller AIC and BIC value was more suitable parametric model than the exponential AFT model. Both CPH (p=0.008) and AFT (p=0.000) models adequately fit the data but AFT model has 100% chance of been a better fit than Cox model as its confidence bounds was also narrower. Hence AFT model gives a better precision. This study concluded that 81 % of nursing mothers in Nigeria discontinue EBF on/before child sixth month of birth, average EBF discontinuation time is six months and EBF discontinuation time can be predicted from mothers' locality and place of child delivery. EBF survivorship pattern showed stepwise decrease and follows Weibull model distribution. AFT model estimate gives better precision and therefore fit the data better than CPH model. Thus, Public Health researchers should employ parametric models in analysing censored data.