A Heterogeneity Analysis of Preference-Based Health-Related Quality of Life Outcomes in Individuals Born Very Preterm and Very Low Birthweight Adults: An Individual Participant Data Meta-Analysis

OBJECTIVES: We explore heterogeneity in the effects of preterm birth on HRQoL in early adulthood using a machine learning approach, Causal Forests (CF). Individual participant data were obtained from five prospective cohorts of individuals born very preterm (VP) or with very low birthweight (VLBW) and a group of normal birthweight or at term matched controls. The combined dataset included over 2,100 adult VP/VLBW survivors aged between 18 and 29 years and controls. The main exposure was defined as birth before 32 weeks’ gestation (VP) and/or birth weight below 1500 grams (VLBW). Outcome measures included multi-attribute utility scores generated by the HUI3.

METHODS: We applied novel machine learning methods to identify the risk factors associated with differences in HRQoL by estimating conditional average treatment effects (CATE). We assess variable importance in explaining preterm birth impact on adult HRQoL. CF and Shrinkage Bayesian Causal Forests were used to estimate the CATEs. We were interested in heterogeneity across the following subgroups: maternal age, maternal education and maternal ethnicity, and interactions between maternal age, ethnicity and maternal education.

RESULTS: VP/VLBW status was associated with a significant difference in the HUI3 multi-attribute utility score of − 0.06 (95% confidence interval − 0.08, − 0.04) in comparison to birth at term or at normal birthweight. We find that effects are fairly homogeneous, albeit there is some evidence that longer term effects of VP/VLBW on HRQoL differ by maternal ethnicity (white vs non-white).

CONCLUSIONS: We find that being VP/VLBW significantly lowers overall HRQoL in early adulthood. However, we cannot be confident that effects of preterm birth are heterogeneous by the subgroups considered, limiting the scope to achieve significant gains by targeting specific subgroups of the affected population defined by the variables considered here. Studies that estimate the effects of VP/VLBW status on HRQoL outcomes across the socioeconomic status are needed.