MODELING THE RISK FACTORS OF MISCARRIAGE USING ADVANCED SURVIVAL ANALYSIS TECHNIQUES: CENSORED QUANTILE REGRESSION AND CURE MODEL

Abstract

Miscarriage is the involuntary termination of pregnancy before the fetus can survive outside the maternal uterus. The risk factors associated with the transition from a normal pregnancy to a complete miscarriage before 28 weeks of gestational age have not been exhaustively established. Use of logistic regression to assess the factors associated with spontaneous abortion excludes the longitudinal and incompleteness aspects of miscarriage data. Cox's model and the accelerated failure time model have the following drawbacks: Assumes a constant hazard ratio, model the hazard rate rather than the duration of survival, not robust in handling statistical outliers, cannot handle a data with cure fraction and lack flexibility. The main objective of the study was to model the risk factors associated with miscarriage using advanced survival analysis techniques; quantile regression and cure models. The study used secondary data from western Kenya's Kakamega County Teaching and Referral Hospital. The data collection period was January 1, 2019–December 31, 2020. The study used Kaplan-Meier, chi-squared and log-rank tests for independent analysis. The Cox proportional hazards (PH) model and parametric models were used to evaluate miscarriage data based on Akaike Information Criterion (AIC). A semi-parametric shared frailty model was used to examine unobserved variability among expecting women by residence. This study used cure rate and censored quantile regression models. Of the total sample of 6077, 248 mothers (4.1%) miscarried, while 5829 (95.9%) were censored. The significant factors identified by log rank test were ethnicity (𝑝 = 0.000), levels of education (𝑝 = 0.048), place of residence (𝑝 = 0.000), employment status (𝑝 = 0.004), malaria status (𝑝 = 0.000) and urinary tract infection (UTI) status (𝑝 = 0.000). The covariates in categorized form found significant by log rank were number of previous stillbirths (𝑝 = 0.000) and number of antenatal care (ANC) visits. The factors ethnicity, place of residence, malaria status, number of previous miscarriages, number of previous stillbirths and number of ANC visits were identified as the risk factors associated with miscarriages using cox model, parametric proportional hazards model and accelerated failure time models. The study found equivalent hazard ratios for Cox (PH) and accelerated failure time models but log-likely hood and the (AIC) showed that the Gompertz PH model had the best fit of the data. In comparison between the Cox and the censored quantile regression (CQR) models the factors; ethnicity, previous number of miscarriages, previous number of stillbirths, occupation status, and malaria infection, exhibited a statistically significant adverse effect on the duration of survival during the initial phases of pregnancy in the CQR model and not in cox model. The cure model showed that place of residency (𝑝 = 0.0034), ethnicity: kalenjin (𝑝 = 0.0008), kikuyu (𝑝 = 0.014) and luo (𝑝 = 0.04), number of prior miscarriages (𝑝 = 0.000), number of previous stillbirths (𝑝 = 0.000) and number of ANC visits (𝑝 = 0.000) statistically affect cure fraction. However these factors did not affect survival time, apart from the number of ANC visits (𝑝 = 0.001). The number of previous miscarriages, stillbirths, ANC visits, site of residency, and ethnic groups (Kalenjin, Kikuyu, and Luo) had cure probabilities of 54.88%, 47.09%, 76.09%, 87.16%, 39.33%, 44.78%, and 58.25%, respectively. The study concludes that there is association between certain explanatory variables and the time to miscarriages, the Gompertz (PH) regression model best fits this data set. Censored quantile regression model can show that certain variables had a significant effect on survival time in some point in pregnancy and reveal the dynamics of prognostic risk factors and their impact on patient survival over time. The cure model showed that these factors had effect on long-term survivors, on short-term trends there were little changes. The findings of this study will aid the government and healthcare authorities to plan and provide interventions to reduce miscarriages in Kenya.