Issue: July 2020
Source/Disclosures
Disclosures: Devgan reports he owns and operates CataractCoach.com, which is free to all users, and is a principal in Advanced Euclidean Solutions, which owns the Ladas Super Formula AI 2.0 and IOLcalc.com.
March 18, 2019
5 min read
Save

When IOL calculations give different results, which do you choose?

Newer calculation methods evolve with time because they are based on artificial intelligence, neural networks and machine learning.

Issue: July 2020
Source/Disclosures
Disclosures: Devgan reports he owns and operates CataractCoach.com, which is free to all users, and is a principal in Advanced Euclidean Solutions, which owns the Ladas Super Formula AI 2.0 and IOLcalc.com.
You've successfully added to your alerts. You will receive an email when new content is published.

Click Here to Manage Email Alerts

We were unable to process your request. Please try again later. If you continue to have this issue please contact customerservice@slackinc.com.

With multiple different formulae available for IOL calculations, we will often encounter a clinical case in which we get different results, which can make it difficult to choose a lens power. While it is easy to select an IOL power if the formulae agree, which do you select when they produce differing results?

We have all been in this situation: We have a patient scheduled for a seemingly routine cataract surgery, but when we look at the printout from the biometer, we see a different result from each of the four formulae (Figure 1). The key to understanding the differences lies in the evolution of the formulae over the past few decades. While the SRK I (first generation) and SRK II (second generation) produced reasonable results, their methodology is about 40 years old. These regression formulae were soon replaced by theoretical formulae (third generation), with the SRK/T, Holladay 1 and Hoffer Q. These third-generation formulae use the two input variables, the keratometry and the axial length, to determine the effective lens position (ELP) of the eye.

different IOL power calculated by the four formulae
Figure 1. This patient has a different IOL power calculated by the four formulae programmed into this biometer. In this case, the surgeon chose to use a more advanced approach with the Ladas Super Formula Artificial Intelligence 2.0, giving the most accurate result.

Souce: Uday Devgan, MD

A case of steep corneas

The patient has cataracts and a long-standing refraction of about –8 D in the right eye and –6 D in the left eye. There is no recent myopic shift from nuclear sclerosis as the cataract changes are primarily cortical. The keratometry values are quite high at about 50 D in each eye. The printout from the surgeon’s biometer is seen in Figure 2.

The surgeon has the biometer configured to use the SRK/T for all calculations and then has chosen to do a printout with four different IOLs. This will work reasonably well for most average eyes, but I would prefer the printout to show four different formulae for one IOL. It is easy to switch from one IOL to another by virtue of the A-constant.

PAGE BREAK

The IOL power varies 1 to 1 with the A-constant. So, if the IOL power is +20 D at an A-constant of 119.0 (for example, a single-piece acrylic), then:

  • If we change to an IOL with an A-constant of 118.5 (three-piece acrylic), the IOL power must drop by 0.5 D because the A-constant dropped by 0.5 D.
  • If we change to a different single-piece acrylic with A-constant of 119.5, the IOL power must increase by 0.5 D because the A-constant increased by 0.5 D.
  • If we change to an anterior chamber IOL with an A-constant of 115.5, the IOL power must drop by 3.5 D because the A-constant dropped by 3.5 D.
  • If we change to a sulcus IOL, first change to the three-piece acrylic IOL power as shown above and then use the “rule of nines” to determine sulcus IOL power.
biometer configured to use just the SRK/T formula with four different IOLs
Figure 2. This biometer is configured to use just the SRK/T formula with four different IOLs shown. This is not as clinically useful as showing multiple different formulae for the same IOL.

The SRK/T formula is the theoretical formula (hence, the T notation in the name) from three surgeons: Sanders, Retzlaff and Kraff. It is in the same family of third-generation theoretical formulae such as the Holladay 1 and Hoffer Q. All three of these formulae use the K and axial length values to determine the ELP, which is where in the eye (in the anterior-posterior direction) the IOL optic will end up after surgery.

The SRK/T formula, in particular, relies heavily on the K value to determine the ELP, and that does make sense in general: If the K is flat (low K value such as 38), the ELP will likely be more anterior and a lower IOL power will be determined. If the K is steep (high K value such as 50), the ELP will likely be more posterior and a higher IOL power will be determined.

However, when we look at the actual measured anterior chamber depth (ACD) of this patient from the biometer, it is 2.64 mm in the right eye. The third-generation formulae do not use the ACD as an input variable, so this value of 2.64 mm is not taken into account by the SRK/T (or the Holladay 1 or the Hoffer Q, for that matter).

PAGE BREAK

If we use a fourth-generation formula that allows more input variables, then this ACD value helps determine the ELP for more accurate calculations. These formulae include the Holladay 2, the Haigis and Barrett Universal II.

Finally, there are newer methods of IOL power calculation that evolve with time because they are based on artificial intelligence, neural networks and machine learning. These include the Ladas Super Formula AI 2.0 and the Hill-RBF, which continue to improve with time. Any formula that does not evolve will eventually be phased out.

Getting back to this challenging case, our last calculation method is with the Ladas AI 2.0, which is available to all ophthalmologists worldwide at IOLcalc.com for free. This Super Formula has the benefit of many thousands of eyes worth of data and results. The patient underwent successful cataract surgery, the +10.5 D IOL was implanted, and the patient achieved the desired plano outcome. The Barrett Universal II was also quite close and would have left the patient about –0.4 D of myopia. The SRK/T was far off and would have resulted in a postop refraction of about –2.5 D of myopia (Figure 3).

SRK/T produces a result that is far from the values
Figure 3. The SRK/T produces a result that is far from the values given by the Barrett Universal II and the Ladas Super Formulae Artificial Intelligence 2.0 methods.

How powerful is this new artificial intelligence technology? A recent graduate of our UCLA ophthalmology residency program compared the different calculations using a neural network based on the framework of the Ladas Super Formula AI method, and the results were clearly the best, even with just 100 eyes. With 1,000 eyes, it was even better and more pronounced (Figure 4). And in the near future, we will be working with 100,000 eyes and even millions of eyes.

Targeted performance
Figure 4. Using just 100 eyes, research from Shawn Lin, MD, MBA, from Mass Eye & Ear, showed the superiority of a neural net using the Ladas AI 2.0 framework over all other methods of IOL power calculation. With 1,000 eyes, it was even more evident that it is the future for calculations.

Change can be difficult to accept, but you must accept that artificial intelligence methods of IOL calculations will be the future. If you have any doubts about this, just remember that there is not a human on the planet who can beat artificial intelligence computers at chess, the Chinese game of Go or the Japanese game of shogi. And doing accurate IOL calculations is next.

Full videos and further explanation can be found at www.CataractCoach.com.