| المستخلص: |
Time-to-event data or survival data is a type of data that contains two pieces of information. It includes information about whether an event occurred or not, as well as the duration it takes for an event to occur during a well-dehned period of time. This type of data is usually skewed and involves incomplete observations (censored data), and thus not suitable for analysis using standard statistical methods. If we ignore the censoring and treat the censored observations as if they were measurements of survival time, then the resulting estimates become biased. Survival analysis is a special technique that can handle these features of data while analyzing time-to- event data. The broad objective of this study was modeling time-to-first birth data obtained from retrospective survey using different survival analysis techniques. The specific objectives of the study were to identify the best nonparametric and parametric survival model for analyzing marriage-to-first birth data as well as identify the significant predictors of marriage-to-first birth interval. The data for the study come from the 2018 Jordan Population and Family Health Survey (JPFHS). The 2018 JPFHS is a large nationally representative cross sectional survey, covering a random sample of 14,689 ever married women of age 15-49 years of age. This study considered a subgroup of 4,828 women who were married within 10 years before survey date to minimize the memory recall bias in reporting marriage-to-first birth interval. There are three approaches in survival analysis: nonparametric approach, semi-parametric approach and parametric approach. Nonparametric analysis were done using three techniques, namely Kaplan-Meier estimator, Nelson-Aalen estimator, and Life Table estimator. We applied these techniques to estimate the survival functions and the average duration of first birth interval in Jordan. To compare between two or more survival curves, log-rank test was used. The semi-parametric Cox proportional hazard model was used to identify the significant predictors of the outcome variable. Several parametric models such as exponential, Weibull, lognormal, gamma, generalized gamma, and log-logistic model were fitted to evaluate the best fitted model identifying the prognostic factors of marriage to first birth interval. It was estimated that the median time for a woman in Jordan to give birth for the first time after marriage was about 15 months, while the mean was estimated to be about 26 months. It was seen that the Nelson-Aalen estimator performed well when the hazard rate was increasing and the Kaplan-Meier estimator performed well when the hazard rate was decreasing. Cox-proportional hazard model identified age at marriage, education, wealth index, region, work status and contraceptive use as significant predictor of marriage to first birth interval. Among the parametric models, log-logistic, lognormal and generalized gamma provide very similar results and thus better fit the time-to-first birth interval data than widely recommended Weibull distribution.
|
