Hitting target visual outcomes in cataract surgery hinges on the accuracy of the IOL’s refractive power. This depends not only on the accuracy of the preoperative biometric data, including axial length, keratometry and anterior chamber depth, but most critically on the effective lens position (ELP). This estimation is derived by the selection of the most appropriate formula.
Many IOL power calculation formulas have been introduced over the years in an effort to obtain the most precise postoperative refractive results.
Traditionally, these formulas were classified by “generations.” For instance, Hoffer Q, Holladay 1 and SRK/T are examples of third-generation formulas. Haigis, Olsen, Holladay 2 and the Barrett Universal II represent the fourth generation.1 Today, newer formulas have been developed that use more sophisticated algorithms and leverage more information to predict the ELP.
These formulas fall under four main categories, according to Douglas D. Koch, MD. They are based on:
- Standard geometric optics
- Ray tracing
- Artificial intelligence (AI) or machine learning
- Some combination of the previous three.
Here, we’ll focus on the impact of AI and machine learning on the future of IOL calculations.
BIG DATA AND IOLS
“When we say ‘AI-based’ IOL calculation formulas, we mean those that are using large amounts of data and some form of machine learning or AI in order to derive relationships among optical parameters that enable them to select an IOL power for any given patient,” says Dr. Koch, professor and the Allen, Mosbacher and Law Chair in Ophthalmology, Cullen Eye Institute, Baylor College of Medicine, Houston.
All formulas use data, of course, whether it be small data or big data, to perform calculations. “The use of machine learning to train a formula is a newer innovation.” Formulas that use a combination of geometric optics and AI in a variety of ways to refine them include Kane, Ladas, EVO, Hoffer QST, PEARL-DGS and the Zeiss AI IOL calculator.
Although the machine learning aspect of these formulas is proprietary in nature, many formulas are freely available online. “Some of these formulas are incorporated directly into the biometers, and ESCRS, for example, has several of the newest formulas available online,” Dr. Koch says.
“Surgeons can upload their data once and get calculations from all of them.”
Dr. Koch cautioned, however, that manually uploading data to a website can be a source of errors. “That is where a web-based platform like Zeiss’ Veracity Surgical Planner comes into play. It automatically pulls the patient data from the biometer and calculates IOL power based on several of the aforementioned formulas in addition to the company’s own AI IOL calculator. The surgeon can select the desired IOL, print it out and never have to touch pencil to paper.”
The Zeiss AI IOL calculator is trained by paraxial ray tracing generating 500,000 model eyes and uses outcomes data collected mostly from Veracity users, according to the company, “to approach the complexity of the interaction between the human eye and IOL in a new and evolving way.”
Dr. Koch notes that more than 10% of cataract surgeries in the United States are now being done using VERACITY; Bausch + Lomb and Alcon will each be introducing their own proprietary software in the coming months as well, giving surgeons multiple options. “As these become more integrated into clinical practice, surgeons will enhance the accuracy of their outcomes because they can easily compare multiple formulas. It saves time for clinicians and staff, as well as preventing transcription errors, which increases patient safety,” he says.
OUTLIER EYES
Another significant advantage to AI-based calculators is that training a high number of cases in the machine-learning process can be tailored to long eyes, short eyes and specific IOLs. “It will automatically optimize for those various features,” Dr. Koch adds.
These outlier cases are the challenge, notes Nicole Fram, MD, because the assumptions that formulas make about the relationship between the anterior and posterior curvature of the cornea can fall apart.
“We are doing really well in normal eyes, but in atypical eyes such as those with axial length less than 21 mm and over 26 mm or post-corneal refractive surgery, we need help. We all see these patients and in practices like mine, it makes up almost 50% of my patient base,” says Dr. Fram, who is in private practice at Advanced Vision Care in Los Angeles and is a clinical instructor of ophthalmology at John A. Moran Eye Center, University of Utah.
Dr. Koch and colleagues studied the predictive accuracy of newer formulas in short eyes and found that there was a slightly better result with the Zeiss AI IOL calculator compared to Barrett, Pearl-DGS and Kane (P < .05).2 The difference was relatively small, he points out, which means that “most of the new formulas differ by small amounts in their accuracy, regardless of the mechanism used to calculate IOL power. That said, none of the formulas in our short-eye study had >75% accuracy within 0.50 D of target.”
A recent review (which did not include the Zeiss AI IOL calculator) evaluated current formulas’ accuracy using outcomes of mean absolute error and percentage of predictions within ±0.50 D and ±1.00 D.1 The Barrett vergence formula achieved the smallest mean absolute error, while among AI-based formulas studied, PEARL-DGS was associated with the highest percentage of patients with ±0.50 D in short eyes and Kane obtained the highest percentage of patients with ±0.50 D in long eyes.
INTRAOPERATIVE ABERROMETRY: AN ENTRY INTO BIG DATA
Intraoperative wavefront aberrometry, specifically the ORA System’s powered by AnalyzOR Technology (Alcon), was Dr. Fram’s first experience with “big data.”
“I liked using the ORA on every patient because I could track my outcomes according to lens type. The optimization was based on global entries and personalized entries based on the refractions entered post-operatively. The ORA utilizes the Holladay formula with an ELP factor changed to the patient’s aphakic refraction spherical equivalent. There are millions of entries at this point that add to the optimization of outcomes,” she says.
The system also takes into consideration both the anterior and posterior cornea in real time as it measures the aphakic refraction. The entries of axial length, anterior Ks, white to white, power and expected refractive error all contribute to the recommended power and refractive outcome. “ORA was a great way for me to get into big data — looking for trends based on these different variables help making predictions,” Dr. Fram says. “I have found ORA to give me the most significant advantage for my post-LASIK/refractive, small and long eyes, and high toric patients. The formula becomes better with every software update because more data comes in from the refractive outcomes.”
According to Alcon, the ORA System with VerifEye+ Technology draws from more than 2 million cases3 and counting with greater postoperative predictability in 58% of cases.4
“The platform’s digital smart solutions provide automated workflow, planning, calculations with the different formulas, and then you can put your postop refractive data and see outcomes analysis — I can look in the mirror and see how I’m doing. It’s no longer, I think I’m doing well; I know,” Dr. Fram says.
The company’s forthcoming advancement within the digital planning software will pull data from the surgeon’s chosen biometer(s) and develop an AI-based formula based on that information that will appear as an option within the drop down menu.
“The biometers and measurements are getting better and better, but what we can’t predict from a biometer is where the IOL will ultimately sit or ELP — we also don’t know how the capsule is going to fibrose over time,” Dr. Fram says. “AI formulas that account for biometry variables as well as surgeon-specific variables such as incision type and capsulotomy sizes in massive numbers — big data — will weed out the noise and use only the variables that will result in increased predictability. AI will hopefully allow us to see the patterns to find that piece of the puzzle that we’re missing: predicting ELP.”
AS GOOD AS IT GETS?
Regardless of the formula, refractive accuracy depends on the quality of the corneal readings, the accuracy of the refraction and how the IOL shifts in position as it heals postoperatively. “There’s probably also some variability in IOL power labeling that can affect outcomes,”
Dr. Koch says. “Putting it all together, if we consistently hit 90% or a little better within 0.50 D of target, that may be our ceiling for accuracy.”
He eagerly anticipates a future in which doctors never have to touch pencil to paper and everything is done in an automated way. “I’m also looking forward to the day when we have increased options for modifying IOL power postoperatively, which is what we’re going to need to tune up those maybe 9% or 10% of eyes that don’t turn out the way we had calculated.” As
Dr. Koch notes, that will require not only technology such as the Light Adjustable Lens (RxSight) but ideally other approaches for adjusting IOL power postoperatively as well. “That technology is evolving, such as that of Perfect Lens, which is in clinical trials,” he says, adding that the early data are encouraging.
CONCLUSION
One cannot give enough credit to the generous contributions of the current and past IOL calculation community, says Dr. Fram, by surgeons such as Drs. Graham Barrett, Warren Hill, Douglas Koch, Li Wang, Ken Hoffer, Jack Holladay and Jack Kane, among many others “to making us better surgeons.”
“Now, AI in IOL power calculations is going to change everything,” she adds. “The more data we get, the more accurate our outcomes. It’s just math. My partner who founded this practice, Sam Masket, would always say, ‘What’s good for the patient is good for the doctor, which is good for industry.’ If you go in that order, then everybody wins.” OM
Dr. Fram is a consultant to Alcon, Carl Zeiss Meditec, Johnson & Johnson Vision and RxSight. Dr. Koch is a consultant to Alcon, Carl Zeiss Meditec, Johnson & Johnson Vision and Perfect Lens.
References
1. Stopyra W, Langenbucher A, GrzybowskiA. Intraocular lens power calculation formulas—a systematic review. Ophthalmol Ther. 2023;12:2881-2902.
2. Kenny PI, Kozhaya K, Truong P, et al. Efficacy of segmented axial length and artificial intelligence approaches to intraocular lens power calculation in short eyes. J Cataract Refract Surg. 2023;49(7):697-703
3. Alcon Data on File. 2021. https://www.myalcon.com/professional/cataract-surgery/surgical-equipment/ora-system/. Accessed November 10, 2023.
4. Chen M, Reinsbach M, Wilbanks ND et al. Utilizing intraoperative aberrometry and digital eye tracking to develop a novel nomogram for manual astigmatic keratotomy to effectively decrease mild astigmatism during cataract surgery. Taiwan J Ophthalmol. 2019;9(1):27-32.
