STRONGER INSTRUMENTS VIA INTEGER PROGRAMMING IN AN OBSERVATIONAL STUDY OF LATE PRETERM BIRTH OUTCOMES

Area of Science:

• Biostatistics
• Health Services Research
• Econometrics

Background:

• Observational studies often use nonbipartite matching to strengthen instrumental variables.
• Traditional network optimization methods for matching have limitations in analytical flexibility.
• Late-preterm infants (34-36 weeks gestation) face risks of readmission, necessitating research into optimal hospital stay durations.

Purpose of the Study:

• To introduce integer programming as a flexible tool for optimal nonbipartite matching.
• To investigate whether extending hospital stays for late-preterm infants reduces subsequent readmission rates.
• To address potential biases in observational studies by employing advanced matching techniques.

Main Methods:

• Utilized integer programming for optimal nonbipartite matching, enabling fine balance on multiple variables.
• Employed hour-of-birth as an instrumental variable for hospital stay duration in a study of late-preterm births.
• Formed 80,600 matched pairs of infants with similar covariates but differing anticipated lengths of stay.

Main Results:

• Matched 80,600 pairs of late-preterm infants based on covariates and instrumental variable (hour-of-birth).
• The study found no statistically significant evidence that encouraging an extra day in the hospital reduces readmissions within two days of discharge.
• Sensitivity analysis explored potential biases from unmeasured covariates affecting the instrumental variable.

Conclusions:

• Integer programming offers enhanced capabilities for optimal nonbipartite matching in complex observational studies.
• Extending hospital stays for late-preterm infants did not demonstrate a significant reduction in early readmission rates in this analysis.
• The methodology highlights the importance of rigorous sensitivity analyses to validate instrumental variables and mitigate bias in causal inference.