Abstract
Purpose :
Transient one-day postoperative intraocular pressure (IOP) spikes following cataract surgery are adverse outcome of cataract surgery and can lead to pain, corneal edema, glaucomatous nerve damage, anterior ischemic optic neuropathy, and vision loss. To our knowledge, no studies have combined known and unknown risk factors to predict IOP spikes. We performed a retrospective, observational cohort study to build a clinical prediction rule for one-day postoperative intraocular pressure spikes based on demographic, clinical, and ophthalmic data using machine learning methods.
Methods :
First eye phacoemulsification cataract surgeries between January 2014 and June 2020 from the University of Colorado Cataract Outcomes Registry were included. Exclusion criteria were the absence of documented postoperative IOP. The presence of a one-day IOP spike was defined as an IOP ≥ 30 mmHg at the first day post-operative visit. The data were randomly split into two independent sets: a training set (80%) for development and tuning of the model and a test set (20%) to determine the final fit of the model. Random forests were used to build the clinical prediction model.
Results :
A total of 7,676 first eye phacoemulsification cataract surgeries were included in the analyses. A random forest model with 75 variables produced the largest area under the receiver operator curve (0.80) in the training set and 0.71 (95% confidence interval 0.66, 0.79) in the test set, where 0.50 is no predictive power. Among the most influential predictors were surgery characteristics (cumulative dissipated energy, length of surgery, complex case, combined surgery), eye characteristics (eye diseases and measurements), baseline IOP, and race and sex of patient. The optimum operating point had a sensitivity of 0.78, specificity of 0.55 and a positive predictive value of 0.84.
Conclusions :
A clinical prediction rule for postoperative IOP spikes can identify at-risk patients immediately after the surgery, allowing the surgeon to mitigate risk by providing prophylactic IOP lowering drops, ultimately reducing short-term pain and long-term sequalae. We also identified several novel risk factors for IOP spikes that warrant follow-up.
This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.