Exploring two cost-adjustment methods for selection bias in a small sample: using a fetal cardiology dataset

Area of Science:

• Health economics
• Medical decision making
• Biostatistics

Background:

• Economic evaluations in healthcare often face challenges with non-randomized data and small sample sizes, complicating cost estimations.
• Accurate cost data is crucial for evaluating healthcare technologies, especially in specialized areas like prenatal screening.

Purpose of the Study:

• To explore and compare regression analyses and propensity scoring methods for obtaining reliable cost estimates.
• To reduce selection bias in the economic evaluation of screening for fetal cardiac anomalies in pregnant women.

Main Methods:

• Applied regression analyses and propensity scoring methods to estimate pregnancy costs.
• Compared costs between telemedicine and direct referral groups for specialist cardiac advice.
• Assessed group balance and cost differences after applying bias reduction techniques.

Main Results:

• Observed pregnancy costs were higher for telemedicine (£4,918) than direct referral (£4,311).
• Regression analysis indicated referral mode was not a significant cost predictor, reducing the difference to £94.
• Propensity scoring methods balanced groups and reduced cost differences, ranging from -£62 to £333.

Conclusions:

• Regression and propensity scoring methods can enhance homogeneity and reduce variance in adjusted healthcare costs.
• These statistical techniques effectively reduce observed selection bias in economic evaluations.
• Propensity scoring demonstrated superior performance in this dataset, achieving better group similarity and smaller adjusted cost differences.