Abstract
Purpose :
The propensity score (PS) weighting method is commonly used in observational studies. Weighting subjects using PS mimics treatment randomization by reducing bias caused by different patient characteristics at baseline. Our goal was to assess the effect of medication, laser and non-laser surgery (NLS) on 1-year intraocular pressure (IOP) reduction in a longitudinal cohort of glaucoma subjects using PS weighted regression.
Methods :
Subjects had IOP measurements before treatment and for the entire follow-up duration. The last IOP measured before treatment was used as baseline. If a subject did not have IOP measured exactly at the end of 1 year, two adjacent measurements before and after the end of 1 year were used for imputation by linear interpolation. One-year change of IOP was used as the outcome. Multinomial PS for treatment assignment was estimated by baseline age, gender, ethnicity, and baseline IOP. A PS weighted regression was conducted to investigate the effect of medication, laser, NLS on 1-year IOP reduction. The interaction between baseline IOP and treatment was also tested to evaluate the impact of baseline IOP on treatment effect.
Results :
122 eyes (90 subjects) were identified to provide 197 treatment segments of 1-year follow-up duration. The five treatment groups included medication, laser, laser/medication, NLS, and NLS/medication. At baseline, IOP was significantly different among groups (p=0.007). After being weighted by multinomial PS, baseline IOP was no longer different (p=0.81). Our PS-weighted multivariable regression showed that compared to laser, NLS and NLS/medication had the biggest IOP reduction (-2.98 mmHg and -3.07 mmHg, p=0.001 and 0.03, respectively). Laser/medication was not significantly different from laser (p=0.77). Medication effect had a significant interaction with baseline IOP (p=0.012). For an typical patient in our sample, i.e. a 64-year-old white woman, NLS and NLS/medication had the biggest IOP reduction effect. Laser and laser/medication caused bigger IOP reduction than medication when baseline IOP was above 21 mmHg.
Conclusions :
The use of PS minimized the IOP selection bias at baseline in our observational study to allow proper comparison of treatment effect. Treatment with surgery or laser resulted in a bigger IOP reduction than treatment with medication only when baseline IOP was high.
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.