Barrett Formulas: Strategies

to Improve IOL Power Prediction

37
Graham D. Barrett

The search for improvements for more accurate surement of corneal curvature was now less
methods to improve refractive outcomes began dependent on the skill of the user and more
after Harold Ridley’s implantation of the first repeatable. Optical biometers based on swept-­
intraocular lens implant (IOL) in 1949 [1]. There source ocular coherence tomography (SS-OCT)
were many aspects of Ridley’s intraocular lens [5] such as the IOLMaster 700 introduced in
that were appropriate, including the choice of 2014 further improved the accuracy of AL mea-
polymethylmethacrylate (PMMA) as a lens mate- surements with a reduction in the standard devia-
rial, placement in the posterior chamber, and tion from 25μm to 8 μm. Modern biometers can
even the method of storage with 10% sodium measure additional parameters such as central
hydroxide for sterilization neutralized prior to corneal thickness (CCT), lens thickness (LT), and
implantation. The post-op refraction, however, corneal diameter (CD) measurements of the cor-
was −24.00/+6.00 × 300 as the calculation of the neal limbus more accurately, in addition to the
required IOL power based on the curvature of the anterior chamber depth (ACD), available with
implant did not fully consider the refractive index earlier technology.
of the IOL and needed significant refinement. Improvements in technology have played a
Biometry at this stage was also rudimentary. key role, but equally important to refractive out-
The corneal curvature could be measured by comes, are the formulas required to predict the
keratometers based on the Javal–Schiotz kera- required IOL for individual patients with the
tometer introduced in 1880 [2], but measurement available information from modern biometers.
of the axial length (AL) by A scan ultrasound It was not common in the early decades of
was only introduced commercially in 1970 IOL implantation to use a standard IOL power,
(Kretztechnik AG) [3]. Optical biometry greatly e.g., 18.0 D, or adjust this power by adding the
enhanced the ability to measure AL with greater preop refraction multiplied by a factor of
precision with partial coherence interferometry 1.25D. The first formula was derived by Fyodorov
(PCI) available with the first IOLMaster intro- [6] in 1967 based on Gaussian optics/vergence
duced in 1999 [4]. calculation and was followed by formulas devel-
Automated keratometers based on LEDs were oped by CD Binkhorst (1972) [7], Colenbrander
integrated with optical biometers so that the mea- (1973) [8], Hoffer (1974, publ 1981) [9], Thijssen
(1975) [10], Van der Heijde (1975) [11], and the
G. D. Barrett (*) regression-based SRK (1981) [12], which intro-
Lions Eye Institute, Sir Charles Gairdner Hospital, duced the A-constant. These are considered first-­
University of Western Australia, Perth, WA, Australia generation formulas where the calculated ACD in
e-mail: [email protected]

© The Author(s) 2024 577

J. Aramberri et al. (eds.), Intraocular Lens Calculations, Essentials in Ophthalmology,
https://doi.org/10.1007/978-3-031-50666-6_37
578 G. D. Barrett

Fig. 37.1 Chart to display the classification of formulas based on the method of prediction

the vergence calculation was not adjusted by any Unfortunately, even this classification has lim-
other parameter. itations as formulae based on vergence calcula-
Second-generation formulas were introduced tions or even ray tracing require a data-driven
by Hoffer (1982) using the AL to predict the element to refine the effective lens position
ACD, which was soon followed by R Binkhorst (ELP). This component can incorporate artificial
and the SRK II [13] formula. Soon these were intelligence as a strategy to refine the ELP and
followed by the third-generation theoretical for- are therefore, hybrid in nature.
mulas using the Al and the K for ACD prediction; Furthermore, a formula, such as Barrett
the Hoffer Q [14] (1993), Holladay I [15] (1988), Universal II, which is a theoretical formula incor-
and SRK/T [16] (1990). porating paraxial ray tracing for the cornea and
These formulas were the mainstay of formula IOL, uses third-order polynomial regression to
prediction for about 25 years until recently when refine data-driven refinement of the ELP. The
fourth-generation formulas that considered addi- essence of AI whether based on neural networks
tional parameters such as pre-op ACD and LT or similar algorithms relies on the ability of com-
were introduced including Barrett Universal puters to recognize patterns or dependencies,
(1987) [17, 18], Olsen (1987) [19], Haigis (1990) which are not always evident to the individual
[20], and Holladay II (1996) [21]. observer. Some authorities, however, believe that
More recent formulas could be considered the outcome of AI analysis of large datasets is not
fifth-generation formulas as they incorporate addi- distinctive from this statistical method using
tional calculation methods including ray tracing smaller datasets [30].
and artificial intelligence. These included formu-
las such as Okulix (2005) [22], Barrett Universal II
(2014) [23], Olsen C (2014) [24], Evo (2016), Hill Barrett Universal II (BUII)
RBF (2016) [25], Pearl DGS (2020) [26], Kane
(2019) [27], and Hoffer QST (2020) [28]. The lat- The reason that the Barrett Universal formula is
ter list is not exhaustive, and many new formulas based on paraxial ray tracing is that this allows the
have been published in recent years. input of custom parameters for refractive index
Classifying formulas into generations is and radii of curvature. This allowed me to calcu-
always controversial as the distinction is some- late the required IOL power for the hydrophilic
what arbitrary, and the date of introduction and acrylic IOL I first implanted in August 1983,
grouping is not always sequential. A more logical which was a one-piece foldable lens with an asym-
classification was suggested by an editorial in the metrical optic and different radii of curvature to
Journal of Cataract and Refractive Surgery in conventional PMMA IOLs available at the time
June 2017 based on the method of prediction [31]. The prediction of the lens position is based
[29]. The formulas classified according to the on a theoretical model eye I conceived where the
method of prediction are displayed in a chart in ciliary plane is determined as the intersection of an
Fig. 37.1. anterior sphere—related to the radius of the cor-
37 Barrett Formulas: Strategies to Improve IOL Power Prediction 579

nea—and a posterior sphere—related to the radius Future strategies to improve IOL power pre-
of the globe. The lens factor (LF) is the lens con- diction that is worthy of consideration include
stant that indicates the distance from the ciliary modifications to biometry, measurements, and
plane to the location of the IOL and varies with the the inclusion of additional parameters with exist-
lens model characteristics. A relationship between ing formulas.
the LF and an equivalent a constant was derived as
surgeons are more familiar with the latter value for
different IOLs. The radius of the globe (RG) is a Classical vs. Segmental AL
difficult parameter to measure, and initially, this
was determined empirically and later from actual Traditional pathways to improve ELP power
clinical data using polynomial regression in BUII. prediction include collecting large datasets and
The Barrett Universal II is the core of the different methods of interpreting the relation-
Barrett toric calculator [32–34], which incorpo- ship within them. In addition, using the out-
rates a theoretical model to explain the observed come of the first eye undergoing cataract
behavior of the posterior cornea based on the surgery has also proved helpful in refining the
ellipticity of the corneal limbus. As such, it dif- outcome of the second eye undergoing cataract
fers from a population-based method to derive surgery [42].
the posterior cornea, and a unique posterior cor- Recent papers have demonstrated that using
nea is calculated for each eye according to the different refractive indices for each ocular seg-
measured parameters. ment as opposed to using a single refractive index
Similarly, the Barrett True K is based on the can improve the accuracy of traditional formulas
BUII with an additional theoretical model to such as Holladay 1 or SRK/T. Traditional formu-
account for the disrupted relationship of the ante- las tend to have a myopic prediction error for
rior and posterior cornea in eyes that had under- short eyes and a hyperopic prediction error for
gone myopic [35] or hyperopic [36] refractive long eyes [43, 44]..
surgery including RK [37]. Keratoconus is another An optical biometer provides an optical path
example where the relationship of the posterior length (OPL) which needs to be transformed into
and anterior radii is altered, and more recently, a a geometrical path length (GPL) for use in for-
solution for this condition has been added to the mulas. The average refractive index was derived
online True K available at apacrs.org [38]. from the refractive indices of the different seg-
A formula based on paraxial ray tracing treats ments and then weighted in proportion to the seg-
the IOL as a thick lens, unlike many formulas mented ALs in the Gullstrand model eye.
where the optic is regarded as a thin lens. The GPL = OPL / 1.3549
BUII calculates the first and second principal
planes for the predicted IOL power for an indi- The Classical Axial Length (CAL), as listed in
vidual eye, which requires relatively complex cal- the Partial Coherence Interferometer (PCI)
culations and iterative solutions. Traditional IOLMaster, is adjusted from the GPL calculated
formulas can typically be condensed to a single with the group refractive index such that it
line in a spreadsheet, but the BUII requires 750 remains compatible with immersion ultrasound.
lines of code in its simplest form and up to 3000 CAL = GPL −1.3033 / 0.957
lines of code in the more complex formulas incor-
porating toric and post-refractive predictions. The measured optical path length was trans-
Several published studies have compared the formed by Haigis [45] by the regression equation
BUII to other formulas, and it has been shown to to be compatible with the AL measured by
perform well and be equivalent to other top-­ immersion ultrasound. As the latter is in essence
ranking formulas [39, 40] when targeting emme- a segmental calculation, the derived geometrical
tropia as well as ametropia in the context of path length (Classical Axial Length) can also be
modest monovision [41]. regarded as segmental in nature despite using a
580 G. D. Barrett

group refractive index to measure the optical path used for this purpose, and the individual refrac-
length. tive indices of the media are assumed values and
The Segmented Axial Length (SAL) is the may vary with the density of a cataract as well as
sum of the GPL of the individual segments cal- the wavelength of a biometer. Despite these limi-
culated using their respective refractive tations, however, it appears logical for a formula
indices: to be optimized according to the method used to
SAL = CCTGPL + AQDGPL + LTGPL + VDGPL derive the AL from the measured optical path
length.
Where CCT = Central Corneal Thickness, The Argos biometer uses segmented
AQD = Aqueous Depth, LT = Lens Thickness, AL. Arthur Cummings (Dublin, Ireland) col-
and VD = Vitreous Depth. lected a dataset with AL measured by the Lenstar
A geometrical path length whether derived (CAL) and the Argos Biometer (SAL). Using
from optical biometry in the fashion described these data, I determined a linear relationship
above by Haigis for the original IOLMaster and between the two methods of AL measurement:
subsequent biometers (CAL), or by considering ALSAL = AL CAL∗ 0.96 − LT∗ 0.014 + 1.04
the individual refractive indices (SAL), relies on
assumptions. An empirical adjustment will be This is similar but not identical to the modified
impacted by the nature of the original dataset AL determined by Cooke—CMAL.

CMAL = 1.23853 + 0.95855∗ Traditional AL − 0.05467∗ LT

The refractive indices used by the Lenstar are A trend line in a scatter plot graph of the pre-
not identical to those utilized by the Argos device, diction error versus AL showed a left-leaning
which could explain the differences. Unlike the downward slope with a myopic prediction error
Lenstar, the Argos biometer uses Gullstrand for short eyes and a hyperopic error for long eyes.
refractive indices, developed for white light Using SAL, the MAE and MedAE reduced to
(˜550 nm), and does not scale the refractive indi- 0.35 and 0.276, respectively, and the prediction
ces to the wavelength of the instrument (1060 nm). error within ±0.50 D improved to 75.9%. The
The AL calculated using a global refractive trend line in the scatter plot graph of prediction
index (CAL) is similar to that calculated with seg- error versus AL was now quite flat.
mental AL (SAL) for average eyes but tends to be
longer for short eyes and shorter for long eyes [32].
I used the regression formula I derived from CAL vs. SAL BUII
Arthur Cummings’ data to transform the AL
measured by the Lenstar in a series of 5000 eyes The MAE for BUII using CAL was 0.32, and
to compare the prediction accuracy using CAL or MedAE was 0.25. The percentage of cases pre-
SAL with Holladay 1 representing traditional dicted within ±0.50 D was 80.2%.
formulas and BUII. The lens constant was first A trend line in a scatter plot graph of the pre-
optimized for both formulas such that the mean diction error versus AL showed a relatively flat
error (ME) was zero. curve. Using SAL, the MAE was not altered
(remaining 0.32) but the MedAE increased
slightly to 0.26. The prediction error within ±0.50
CAL vs. SAL Holladay 1 D declined to 78.9%. The trend line in the scatter
plot graph of prediction error versus AL sloped
The MAE for Holladay 1 using CAL was 0.37, downward to the right indicating a trend to hyper-
and MedAE was 0.287. The percentage of cases opic outcomes for short eyes and myopic out-
predicted within ±0.50 D was 74.8%. comes for long eyes.
37 Barrett Formulas: Strategies to Improve IOL Power Prediction 581

The comparison confirmed, the previously 2. The standard BUII formula based on CAL
published data by Cooke et al. and Li Wang et al., was then compared to the new Barrett True
that while the use of SAL improved the predic- AL Formula based on SAL including sub-
tion error and removed AL prediction bias for group analysis of short (<= 22.5 mm) and
traditional formulas, it actually diminished the long eyes (> = 25.5 mm).
prediction accuracy for more modern formulas
such as BUII and Olsen This is because the mod- The IOL implanted in all cases was the Alcon
ern formula that performs well has been opti- SN60WF. An optimized constant was calculated
mized for CAL and the algorithms correct for AL for each formula such that the ME was 0.00
bias. D. IOL constants for all formulas were optimized
This poses a quandary for a surgeon’s selec- in this analysis. The constant for this dataset is
tion of formulas when using a biometer such as somewhat higher for all formulas, e.g., the opti-
Argos, which utilizes SAL. I, therefore, derived a mized a constant for SRK/T was 119.24. This
version of BUII optimized for this Sum of may indicate a shorter refracting lane than 6.0 m,
Segments method. which is not common in the USA, but the refrac-
The EyeSuite software on the Lenstar OLCR tion was not adjusted in this analysis.
machine has research export file capabilities, The error in prediction for each formula was
which can provide the optical path length for the calculated, and the ME, SD, MAE, MedAE, as
segments as an “air” value. The formula was well as the percentage of cases within ±0.25 D,
derived from 17,000 eyes with this data, and the ±0.50 D, ±0.75 D, and ± 1.00 D determined using
segmented AL was calculated from the optical an excel spreadsheet. The results are listed in
path length using the same refractive indices as Tables 37.1, 37.2, 37.3, 37.4, and 37.5 for BUII
the Argos device. In order to maintain consis- (CAL), Haigis, Hoffer Q, Holladay 1, and SRK/T
tency with conventional IOL constants, the SAL formulas, respectively.
AL was offset so the average SAL and CAL were A scatter plot of prediction error vs. AL was
equal—the difference in short and long ALs constructed for each formula with a linear trend
between SAL and CAL was preserved by this line to evaluate whether significant bias existed
strategy. The optimization was derived using the
actual radii of the single model SN60WF IOL,
but the derived formula is intended to be used Table 37.1 ME = mean error, SD = standard deviation,
MAE = mean absolute error, MedAE = median absolute
with the default biconvex model used in the exist- error, and percentage of cases within intervals for BUII
ing BUII formula. (CAL) formula
BUII (CAL) % within D ME SD MAE MedAE
<±0.25 D 47.90% 0.0 0.376 0.310 0.260
 alidation of BUII SAL
V <±0.50 D 80.54%
(Barrett True AL Formula) <±0.75 D 96.98%
<±1.00 D 99.50%
The new formula based on SAL (Barrett True
AL) was validated in a dataset of 595 eyes who
had biometry performed with the Argos biometer Table 37.2 ME = mean error, SD = standard deviation,
shared by John Shammas. The Shammas valida- MAE = mean absolute error, MedAE = median absolute
error, and percentage of cases within intervals for Haigis
tion dataset was not used in any fashion in the formula
derivation or optimization of the Barrett True AL
Haigis % within
formula. (CAL) D ME SD MAE MedAE
<±0.25 D 42.86% 0.0 0.408 0.330 0.298
1. The standard BUII formula based on CAL <±0.50 D 77.82%
was compared to four traditional formulas— <±0.75 D 93.45%
Haigis, Hoffer Q, Holladay 1, and SRK/T. <±1.00 D 99.16%
582 G. D. Barrett

Table 37.3 ME = mean error, SD = standard deviation, which is atypical for this formula when analyzing
MAE = mean absolute error, MedAE = median absolute
datasets based on CAL.
error, and percentage of cases within intervals for Hoffer
Q formula The scatter plot is similar to the Haigis for-
Hoffer Q % within
mula. The trend line for prediction error vs. AL is
(CAL) D ME SD MAE MedAE typically flatter with the Haigis formula than
<±0.25 D 46.05% 0.0 0.410 0.333 0.287 Hoffer Q, Holladay, and SRK/T formulas when
<±0.50 D 74.79% comparing formulas in a dataset based on CAL.
<±0.75 D 93.11%
<±1.00 D 99.33%
 omparison of Standard BUII
C
Table 37.4 ME = mean error, SD = standard deviation, Formula Based on CAL to the New
MAE = mean absolute error, MedAE = median absolute Barrett True AL Formula Based
error, and percentage of cases within intervals for on SAL
Holladay 1 formula
Holladay 1 % within The prediction accuracy for BUII (SAL) listed in
(CAL) D ME SD MAE MedAE
Tables 37.6 and 37.7 is maintained for long eyes
<±0.25 D 45.04% 0.0 0.388 0.322 0.281
<±0.50 D 77.98%
and improves for short eyes compared to BUII
<±0.75 D 96.30% (CAL) in Tables 37.6 and 37.7—the most impres-
<±1.00 D 99.83% sive feature is the flat trend line in Fig. 37.7,
which suggests the potential for improved accu-
racy with larger datasets.
Table 37.5 ME = mean error, SD = standard deviation, Classical formulas with only basic optimiza-
MAE = mean absolute error, MedAE = median absolute tion such as Holladay 1 improved their prediction
error, and percentage of cases within intervals for SRK/T
formula with SAL as compared to CAL as demonstrated
previously with flattening of the curve in predic-
SRK/T % within
(CAL) D ME SD MAE MedAE tion error vs. AL with SAL.
<±0.25 D 43.03% 0.0 ±0.408 0.337 0.297 Using a biometer based on SAL, however,
<±0.50 D 75.13% could potentially have an adverse impact on more
<±0.75 D 94.79% sophisticated formulas as they already have a
<±1.00 D 99.66% relatively flat curve of prediction error vs. AL
over the range of ALs encountered clinically.
This is evident in a comparison of the out-
between these parameters The graphs are dis- comes in the Shammas dataset comprising eyes
played in Figs. 37.2, 37.3, 37.4, 37.5, and 37.6 measured with the Argos device. The formulas
for BUII (CAL), Haigis, Hoffer Q, Holladay 1, can be refined in the future with actual Argos
and SRK/T formulas, respectively. data, but the present derivation appears to resolve
the issues of using formulas optimized for classi-
cal ALs with a sum of segments-based AL such
 omparison of Standard BUII
C as the Argos device.
Formula Based on CAL to Haigis, The trend line of the Barrett True AL formula
Hoffer Q, Holladay 1, and SRK/T based on SAL (Fig. 37.7) is flat unlike the bias
evident using the BUII formula based on CAL
BUII has the lowest error in prediction in terms (fig. 37.2).
of MAE and MedAE as well as the percentage of The optimized constant for the true AL for-
cases with a prediction error within ±0.50 D. The mula (SAL) was LF = 1.972 versus LF = 1.99 for
trend line, however, for the scatter plot graph of the standard BUII (CAL), indicating that no
prediction error versus AK slopes downwards to change in the IOL constant is required when
the right indicating a significant relationship using the new formula.
37 Barrett Formulas: Strategies to Improve IOL Power Prediction 583

Fig. 37.2 Scatter plot of prediction error vs. AL for BUII (CAL) formula

Fig. 37.3 Scatter plot of prediction error vs. AL for Haigis formula
584 G. D. Barrett

Fig. 37.4 Scatter plot of prediction error vs. AL for Hoffer Q formula

Fig. 37.5 Scatter plot of prediction error vs. AL for Holladay 1 formula
37 Barrett Formulas: Strategies to Improve IOL Power Prediction 585

Fig. 37.6 Scatter plot of prediction error vs. AL for SRK/T formula

Table 37.6 ME, SD, MAE, Med.AE, and percentage of cases within intervals for BUII (CAL) formula grouped
according to axial length
BUII (CAL) Short eyes Average eyes Long eyes
No of eyes 595 <= 22 mm > = 22 mm < = 25 mm > 25 mm
Lens factor = 1.972 All eyes (n = 595) (n = 43) (n = 495) (n = 57)
Mean prediction error 0.01 0.12 0.01 −0.067
Standard deviation 0.380 0.440 0.371 0.341
Mean absolute prediction error 0.310 0.391 0.306 0.275
Median absolute error 0.260 0.380 0.255 0.225
Maximum absolute error 1.135 1.040 1.135 1.090
% < =0.25 D 47.90% 34.88% 48.08% 56.14%
% < =0.50 D 80.50% 67.44% 80.61% 89.47%
% < =0.75 D 96.98% 90.70% 97.58% 96.49%
% < =1.00 D 99.50% 97.64% 99.80% 98.25%

Table 37.7 ME, SD, MAE, Med.AE, and percentage of cases within intervals for BUII (SAL) formula grouped
according to axial length
Barrett true axial length (SAL) Short eyes Average eyes Long eyes
No. of eyes 595 <= 22 mm > = 22 mm < = 25 mm > 25 mm
Lens factor = 1.972 All eyes (n = 595) (n = 43) (n = 495) (n = 57)
Mean prediction error −0.008 −0.077 0.002 −0.048
Standard deviation 0.37 0.41 0.37 0.32
Mean absolute prediction error 0.305 0.361 0.305 0.264
Median absolute error 0.264 0.317 0.261 0.224
Maximum absolute error 0.996 0.863 0.996 0.857
% < =0.25 D 48.40% 37.21% 48.89% 52.63%
% < =0.50 D 80.34% 72.09% 80.00% 89.47%
% < =0.75 D 96.97% 95.35% 97.17% 96.49%
% < =1.00 D 100.0% 100.0% 100.0% 100.0%
586 G. D. Barrett

Fig. 37.7 Scatter plot of prediction error vs. AL for BUII (CAL) formula

Summary Measurements

The prediction accuracy is maintained for long In 2008, Sverker Norrby [46] identified postop-
eyes and improves for short—the most impres- erative intraocular lens (IOL) position, postoper-
sive feature is the flat trend line, which suggests ative refraction determination, and preoperative
improved accuracy with larger datasets. AL as the major sources of error contributing to
Classical formulas with only basic optimiza- errors in prediction after cataract surgery.
tion such as Holladay improved their prediction Improvements in the accuracy of optical
with SAL as compared to CAL as demonstrated biometry more recently with swept-source OCT
previously with flattening of the curve in predic- and improved formulas have reduced the impact
tion error vs. AL with SAL. of these factors although subjective post-­
Using a biometer based on SAL, however, operative refraction remains a confounding factor
could potentially have an adverse impact on in comparing outcomes. Variability in k­ eratometry
more sophisticated formulas as they already remains an important source of error in predict-
have a relatively flat curve of prediction error ing spherical outcomes, particularly astigmatism,
vs. AL over the range of ALs encountered following cataract surgery, and arguably now
clinically. should be listed as the most important factor.
This is evident in a comparison of the out- I compared the repeatability of measuring AL,
comes in the Shammas dataset comprising eyes corneal power, and astigmatism on two separate
measured with the Argos device. The formulas biometers on the same visit in 144 consecutive
can be refined in the future with actual Argos data eyes during routine pre-op biometry on the same
but the present derivation appears to resolve the day.
issues of using formulas optimized for classical The axial difference in mm was converted to
ALs with a sum of segments-based AL such as diopters by multiplying by 2.5 to facilitate a
the Argos device. comparison of the impact compared to keratom-
37 Barrett Formulas: Strategies to Improve IOL Power Prediction 587

etry measured in diopters. The mean difference Additional Parameters

in AL between the two devices was −0.02 D
with a SD of ±0.05 while the MAE was 0.038 Originally formulas utilized AL and K as the pri-
and MedAE was 0.025 D. A scatter radar plot mary measured ocular parameters to predict
superimposed on a target is a useful method to intraocular lens power. These remain the most
demonstrate the repeatability of measurements important parameters whether the formula is
and shows how consistent AL measurements based on vergence calculations, data-driven
have become when measured by two different regression, or artificial intelligence. Pre-op pha-
modern biometers. kic ACD measured from the corneal vertex (epi-
The mean difference in keratometry between thelium) to the anterior surface of the lens is also
the two devices was −0.01 D with an SD of ±0.15 correlated to determine the effective lens posi-
while the MAE was 0.10 and MedAE was 0.07 tion of an IOL and was included in the Haigis
D. The standard deviation of the measurements is formulas and most recent formulas. The so-
greater than AL measurements, but the radar called aqueous depth (AQD) does not include
scatter graph demonstrates that the difference in the corneal thickness and is equally useful as a
mean Ks is within ±0.25 D for the majority of measured parameter to improve outcome predic-
eyes. tion. The contribution of different factors can be
The mean vector difference in magnitude of identified using statistical correlation and pre-
the cylinder between the two devices was −0.56 operative LT, horizontal CD, and CCT all show a
D with an SD of ±0.57 while the MAE difference relationship to prediction error. These parame-
magnitude of the cylinder was 0.55 D and MedAE ters can be included in a formula and the
was 0.41 D. The centroid difference in the mea- Holladay II uses up to 7 parameters. The BUII
sured astigmatism between the two devices was can utilize up to 5 parameters including pre-op
−0.10 D @ 79.2°. The difference between the x ACD, LT, and horizontal CD but can also be
and y values of each vector displayed in a double-­ used with only AL and K [49].
angle plot demonstrates that the differences in The utility of the additional parameters is evi-
corneal astigmatism vary more widely than the dent in the analysis of 287 consecutive eyes by
mean K or AL between different devices. considering the MAE and MedAE as well as the
Measures such as using Warren Hill’s vali- percentage of eyes with a predicted outcome
dation criteria are helpful and are optimizing within ±0.50 D.
the corneal surface, but measuring corneal The error in prediction reduces with the inclu-
astigmatism is not always repeatable. I have sion of additional parameters. A graph of the per-
developed a K calculator, which is an integral centage of eyes with a prediction error within
part of the online Barrett toric calculator for ±0.50 D vs. the number of parameters demon-
deriving a vector mean or median K when mea- strates improved prediction accuracy with ACD
suring corneal astigmatism from different and LT as additional parameters but the trend line
devices for toric IOL calculations. In a study of plateaus indicate less impact with the addition of
128 patients, the median K of three devices pro- horizontal CD.
vided the most accurate prediction as it de- Gender, ethnicity, age, and pre-op refraction
emphasizes outliers. The improvement for are other demographic factors that are corre-
spherical prediction was modest but the lated with the prediction of refractive outcomes
improvement in predicting post-­ op residual that can be considered for inclusion to improve
astigmatism was up to 10% and clinically sig- the prediction of formulas. Gender appears to
nificant [47, 48]. This is why I use the K calcu- be the most relevant as female eyes tend to have
lator within the online Barrett toric calculator a more myopic prediction error than male eyes
using three different devices, IOLMaster, for short ALs and a hyperopic outcome for long
Lenstar, and Pentacam to select the sphere and eyes compared to male eyes is evident in the
toric cylinder recommendation in all cases. analysis of large datasets [50]. Even if the data
588 G. D. Barrett

used for formula refinement is not considered may not be equivalent as there is no standard with
separately, a gender bias may still be evident as regard to values such as the refractive index of
the representation of gender is unequal in the the cornea or aqueous that may be used in these
age group undergoing cataract surgery due to equations. Furthermore, unless the measurement
factors such as the longevity of females over is adjusted to be compatible with the traditional
males. Deriving separate data-driven algo- Gullstrand ratio, the lens constants that users
rithms for male and female eyes is likely to have been accustomed may not be appropriate.
improve outcomes. Formulas utilize corneal power in a variety of
Many formulas use a thin lens model and do ways including the actual vergence calculation as
not take into account the change in the principle well as the prediction of the actual IOL position.
plane that occurs with different IOL powers. Ray A customized formula is required to utilize the
tracing including paraxial ray tracing such as new parameter, and the issue is accounted for
BUII uses a thick lens model and allows the lens within the online Barrett formulas in that it allows
parameters to be calculated for each lens power the user to enter the measured posterior cornea
predicted. Ideally, this calculation could include rather than the total corneal power. The formula
the actual lens parameters such as the radii of is incorporated into biometers such as the
curvature or asphericity as these vary with IOLMaster 700 where it is referred to as the
­different manufacturers. The impact of individual Barrett TK. If the measured PCA option is
IOL parameters will have a greater impact on selected, then the posterior cornea values PK1
shorter eyes and using actual radii should improve and PK2 from the IOLMaster or the equivalent
prediction accuracy in this context. Specific IOL posterior cornea values from the Pentacam can be
parameters are proprietary and are not generally entered. The measured posterior corneal power
known so assumptions such as an equi-biconvex will then be used for the sphere and toric predic-
model can be utilized. In addition, IOL-specific tion, which is equivalent to the Barrett TK on the
regression such as the Haigis triple optimization IOLMaster 700. The online formulas require a
is another route to address this aspect of IOL user to select the instrument by which the poste-
prediction. rior cornea has been measured, and the algorithm
is adjusted accordingly. In addition, the formulas
recognize that not all unexplained astigmatism
New Parameters after cataract surgery is due to the posterior cor-
nea and contains additional algorithms to com-
Improvements in technology have enabled us to pensate for factors such as lens tilt. For unusual
measure anatomical parameters that were not corneas such as keratoconus or post-refractive
feasible with earlier optical biometers and ultra- cases, the improvement in prediction is
sound. Scheimpflug tomographers have been significant.
able to measure the posterior cornea as have more
recent optical biometers based on a swept-source
OCT. Direct measurement of the posterior cornea Post-Refractive Formulas
rather than using an estimate based on an assumed
value of the keratometer index or even the I developed the True K formula, which is based
Gullstrand ratio in the paraxial equation for cor- on BUII in order to improve outcome prediction
neal power may potentially improve spherical in eyes that have had previous refractive surgery.
and astigmatic refractive outcomes following The formula utilizes the history of the refractive
cataract surgery. change due to the procedure but can also be used
Typically, a new total corneal power is pro- if this information is not available. Compensation
vided by devices or biometers that measure the for the double K issue where a different K is
corneal power such as “True Net Corneal Power” required for the vergence calculation than that for
or “Total Keratometry.” These measurements the prediction of the IOL position is incorporated
37 Barrett Formulas: Strategies to Improve IOL Power Prediction 589

within the formula. An algorithm for the change PK2 values or corresponding values from other
in corneal thickness that may occur in certain devices such as Scheimpflug devices that are also
refractive procedures is also included within the able to measure the posterior corneal power or
formula. The True K has proved to be effective radius.
for patients who have undergone myopic LASIK Lawless and co-workers published a relatively
when the refractive history is known and when no large series of their patients consisting of 72 eyes
history is available as published in 2016. In a that had undergone previous myopic or hyper-
publication in the JCRS in 2018, the True K for- opic refractive surgery. Their results confirmed
mula proved to be accurate for patients who had that the True K with the inclusion of the posterior
undergone laser correction for hyperopia, and the cornea provided the most accurate and repeatable
True K has been shown to be accurate when com- option in both myopic and hyperopic patients
pared to other methods for RK as published in undergoing cataract surgery without prior refrac-
ophthalmology in 2019. The online True K has a tive information [51].
distinctive feature that allows the user to enter the An important issue that is not widely appreci-
most recent pre-cataract surgery refraction, ated is the need for a custom toric calculator
which has not been impacted by nuclear sclerosis-­ when selecting a Toric IOL in an eye that has
induced myopia without the preop refraction. undergone previous refractive surgery. The theo-
This is different than PRK or LASIK where both retical assumptions within standard toric calcula-
the pre- and post-refractive procedure refraction tors to predict the posterior cornea or
is required for the entered refractive history to be population-based regression methods are no lon-
taken into consideration. This improves the accu- ger valid in the context of toric IOL prediction
racy in prediction for post-RK eyes as the pro- after cataract surgery.
gressive hyperopia, which may be considerable, The True K Toric Calculator was designed
is taken into account in the prediction of the IOL specifically for toric prediction post-refractive
power required post-RK. surgery and, in version 2.0, can now be used
More recently in version 2.5, a solution for with the predicted posterior cornea or a mea-
keratoconus is provided within the True K of the sured option for posterior corneal astigmatism
online True K formula on the APACRS website. (PCA). In addition, the True K Toric calculator
The cornea is steep and irregular within the kera- now includes the K calculator, which allows the
toconus, which is one of the reasons for impre- user to enter up to three different values for the
cise measurements particularly in relation to the anterior cornea and then calculates a new inte-
pupil and visual axis. The most important reason grated K or median vector, which is used for the
for poor prediction in keratoconus, however, is calculation. This is particularly helpful when
the altered relationship between the posterior and the Ks of different devices vary, which is not
anterior cornea not dissimilar to post-refractive unusual in eyes that have undergone refractive
surgery but in its own unique fashion. This latter surgery.
relationship is addressed in the True K option for The default mode for the True K Toric is the
keratoconus by a predictive algorithm or direct theoretical PCA but I have found that utilizing
measurements of the posterior cornea. the measured posterior cornea from SS-OCT pro-
The most accurate method of prediction vides greater accuracy not only for spherical pre-
within the True K formula appears to be a recent diction as previously mentioned but also for toric
modification that allows the True K to incorpo- prediction particularly when no refractive history
rate the measured posterior cornea, the so-called is available.
True K TK. Similar to the toric, when the mea- In a small series of 28 eyes from my own
sured posterior cornea option is selected, a new patients, the method that provided the greatest
page appears where you select the device used percentage predicted within ±0.50 D was the
and enter the measured posterior cornea values True K Toric calculator utilizing the measured
listed within the IOLMaster 700 as the PK1 and posterior cornea from an SS-OCT device.
590 G. D. Barrett

 ormula for Unexpected Refractive

F Finally look at either the bottom of the IOL
Outcome exchange or piggyback page, and the Rx formula
will display a graph and let you know where to
Managing an unexpected refractive outcome rotate the existing lens for the minimum residual
after cataract surgery can be daunting. Corneal astigmatism.
refractive surgery or a lens-based solution can be The Rx is a comprehensive formula that pro-
considered whether by exchanging the implanted vides the required calculations to manage an
lens, adding a piggyback, or rotating an existing unexpected refractive outcome in terms of IOL
toric IOL. There are several formulas that can Exchange, piggyback lens implantation, or toric
provide some of the required calculations such as IOL rotation with both an ELP and IOL option to
the rule of thumb for spherical power, Holladay R determine the expected ELP [52].
for Lens exchange, Astigmatism Fix or Assort for There are many studies comparing the predic-
lens rotation, and the vergence formula for the tion accuracy of different formulas. One of the
required piggyback IOL but sourcing these dif- most recent comparing 13 formulas was pub-
ferent formulas can be confusing. lished by Savini et al.,
I, therefore, developed the Barrett Rx (Asia-­­ this year in the BJO [53]. All the modern for-
Pacific Association of Cataract and Refractive mulas performed well, and the standard deviation
Surgeons (APACRS)), which can be used to pro- was lowest with BUII. There was certainly no
vide a solution for each of these scenarios in a discernible difference in the accuracy of formulas
single formula. The default mode for the Rx is and the method of derivation whether by Gaussian
the ELP mode. Here, the actual effective lens optics or artificial intelligence.
position or ELP is calculated from the post-op Isaac Newton in his famous book on natural
refraction and used as the basis for the vergence philosophy and mathematics noted that what we
calculation. Alternatively, the IOL mode can be know is a drop and what we do not know is an
selected, and here, the IOL constant for the ocean. I would add that when it comes to modern
implanted lens model is used to determine the IOL calculations, we should use every drop of
ELP. The latter is preferred when the problem is knowledge available.
not the ELP prediction but rather due to an abnor-
mal cornea, for example, post-refractive surgery,
or a suspected case of lens power mislabeling. References
The implanted IOL power and post-op refrac-
tions need to be entered including the actual 1. Ridley NHL. Intraocular acrylic lenses. Trans
Ophthalmol Soc UK. 1951;71:617–21. Oxford
alignment if this is a toric IOL. The lens constant Ophthalmological Congress, 1951.
of the implanted lens and that of the planned IOL 2. Javal LE, Schiotz H. Un Opthalmometre Pratique.
exchange are also required. It is important to note Annales d’Oculistique. 1881;86:5–21.
that the optical ACD and lens thickness are the 3. Ossoinig KC. Proceedings of the 3rd meeting of the
International Society for Ultrasonic Diagnosis in
phakic preop parameters and not the post-op Ophthalmology (SIDUO), Vienna, Austria, 1969.
measurements. 4. Fercher AF, Mengedoht K, Werner W. Eye-length
Once the data are entered, select calculate and measurement by interferometry with partially coher-
then Rx exchange IOL to display the recom- ent light. Opt Lett. 1988;13(3):186–8.
5. Fercher AF, Drexler W, Hitzenberge CK, Lasser
mended spherical IOL, toric cylinder, and align- T. Optical coherence tomography—principles and
ment required for an IOL exchange targeting the applications. Rep. Prog. Phys. 2010;66:239.
desired post-op refraction. 6. Fyodorov SN, Galin MA, Linksz A. Calculation
Selecting Rx piggyback on the top menu to of the optical power of intraocular lenses. Investig
Ophthalmol. 1975;14:625–8.
display the alternative piggyback IOL, once 7. Binkhorst CD. Power of the prepupillary pseudopha-
again both spherical, toric power, and alignment. kos. Br J Ophthalmol. 1972;56(4):332–7.
37 Barrett Formulas: Strategies to Improve IOL Power Prediction 591

8. Colenbrander MC. Calculation of the power of an PEARL-DGS formula: the development of an open-­
iris clip lens for distant vision. Br J Ophthalmol. source machine learning-based thick IOL calculation
1973;57(10):735–40. formula. Am J Ophthalmol. 2021;232:58–69.
9. Hoffer KJ. Lens power calculation and the problem 27. Connell BJ, Kane JX. Comparison of the Kane
of the short eye. Ophthalmic Surg. 1982;13(11):962. formula with existing formulas for intraocu-
10. Thijssen JM. The emmetropic and the iseikonic lar lens power selection. BMJ Open Ophthalmol.
implant lens: computer calculation of the refrac- 2019;4(1):e000251.
tive power and its accuracy. Ophthalmologica. 28. Shammas HJ, Taroni L, Pellegrini M, Shammas MC,
1975;171(6):467–86. Jivrajka RV. Accuracy of newer IOL power formu-
11. Van der Heijde GL. A nomogram for calculating the las in short and long eyes using sum-of-segments
power of the prepupillary lens in the aphakic eye. Bibl biometry. J Cataract Refract Surg. 2022; https://doi.
Ophthalmol. 1975;83:273–5. org/10.1097/j.jcrs.0000000000000958.
12. Sanders DR, Kraff MC. Improvement of intraocular 29. Koch DD, Hill W, Abulafia A, Wang L. Pursuing
lens power calculation using empirical data. J Amer perfection in intraocular lens calculations: I. Logical
Intra-Ocular Implant Soc. 1980;6(3):263–7. approach for classifying IOL calculation formulas. J
13. Sanders DR, Retzlaff J, Kraff MC. Comparison of the Cataract Refract Surg. 2017 Jun;43(6):717–8.
SRK II formula and other second generation formu- 30. Cheng X, Davis D, Matloff N, Davis D, Mohanty
las. J Cataract Refract Surg. 1988;14(2):136–41. P. Polynomial regression as an alternative to neural
14. Hoffer KJ. The Hoffer Q formula: A comparison of nets. arXiv. 2019:1806.06850v3.
theoretic and regression formulas. J Cataract Refract 31. Barrett GD. A new hydrogel intraocular lens design. J
Surg. 1993;19(11):700–12. Errata: 1994;20(6):677 Cataract Refract Surg. 1994;20(1):18–25.
and 2007;33(1):2–3. 32. Barrett GD. Flight of the arrow - Toric IOL predic-
15. Holladay JT, Prager TC, Chandler TY, Musgrove KH, tion. Boston, MA: Film Festival, American Society of
Lewis JW, Ruiz RS. A three-part system for refining Cataract & Refractive Surgeons; 2014.
intraocular lens power calculations. J Cataract Refract 33. Abulafia BGD, Kleinmann G, Ofir S, Levy A,
Surg. 1988;14(1):17–24. Marcovich AL, Michaeli A, Koch DD, Wang L, Assia
16. Sanders DR, Retzlaff JA, Kraff MC, Gimbel HV, E. Prediction of refractive outcomes with toric intra-
Raanan MG. Comparison of the SRK/T formula ocular lens implantation. J Cataract Refract Surg.
and other theoretical and regression formulas. J 2015;41(5):936–44.
Cataract Refract Surg. 1990;16(3):341–6. Erratum 34. Ferreira TB, Ribeiro P, Ribeiro FJ, O'Neill
1990;16(4):528. JG. Comparison of astigmatic prediction errors asso-
17. Barrett GD. Intraocular lens calculation formulas ciated with new calculation methods for toric intraocu-
for new intraocular lens implants. J Cataract Refract lar lenses. J Cataract Refract Surg. 2017;43(3):340–7.
Surg. 1987;13(4):389–96. 35. Abulafia A, Hill WE, Koch DD, Wang L, Barrett
18. Barrett GD. An improved universal theoretical for- GD. Accuracy of the Barrett true-K formula
mula for intraocular lens power prediction. J Cataract for intraocular lens power prediction after laser
Refract Surg. 1993;19(6):713–20. in situ keratomileusis or photorefractive kera-
19. Olsen T. Theoretical approach to intraocular lens tectomy for myopia. J Cataract Refract Surg.
calculation using Gaussian optics. J Cataract Refract 2016;42(3):363–9.
Surg. 1987;13(2):141–5. 36. Vrijman V, Abulafia A, van der Linden JW, van der
20. Haigis W. Strahldurchrechnung in Gauß'scher Optik Meulen IJE, Mourits MP, Lapid-Gortzak R. ASCRS
zur Beschreibung des Systems Brille-Kontaktlinse-­ calculator formula accuracy in multifocal intraocu-
Hornhaut-Augenlinse (IOL). 4. Kongreß d. Deutschen lar lens implantation in hyperopic corneal refrac-
Ges. f. Intraokularlinsen Implant., Essen, Germany tive laser surgery eyes. J Cataract Refract Surg.
1990, hrsg.v. K Schott, KW Jacobi, H Freyler, 2019;45(5):582–6.
Springer Berlin, 1991: 233–246. 37. Turnbull AMJ, Crawford GJ, Barrett GD. Methods
21. Holladay JT, Holladay IOL. Consultant computer pro- for intraocular lens power calculation in cataract
gram. TX, Holladay IOL Consultant: Houston; 1996. surgery after radial keratotomy. Ophthalmology.
22. Preussner PR, Wahl J, Weitzel D. Topography-based 2020;127(1):45–51.
intraocular lens power selection. J Cataract Refract 38. Ton Y, Barrett GD, Kleinmann G, Levy A, Assia
Surg. 2005;31(3):525–33. EI. Toric intraocular lens power calculation in cataract
23. Barrett GD. A formula for all seasons. Supplement to patients with keratoconus. J Cataract Refract Surg.
cataract and refractive surgery Today/Europe. 2014. 2021;47(11):1389–97.
24. Olsen T, Hoffmann P. C constant: new concept for ray 39. Melles RB, Holladay JT, Chang WJ. Accuracy of
tracing-assisted intraocular lens power calculation. J intraocular lens calculation formulas. Ophthalmology.
Cataract Refract Surg. 2014;40(5):764–73. 2018;125(2):169–78.
25. IOL Power Calculator for Cataract Surgery|Hill-RBF 40. Roberts TV, Hodge C, Sutton G. Lawless M; contribu-
Calculator. https://rbfcalculator.com/. tors to the vision eye institute IOL outcomes registry.
26. Debellemanière G, Dubois M, Gauvin M, Wallerstein Comparison of Hill-radial basis function, Barrett uni-
A, Brenner LF, Rampat R, Saad A, Gatinel D. The versal and current third generation formulas for the
592 G. D. Barrett

calculation of intraocular lens power during cataract American Society of Cataract & Refractive Surgeons;
surgery. Clin Exp Ophthalmol. 2018;46(3):240–6. April 2018.
41. Turnbull AMJ, Hill WE, Barrett GD. Accuracy of 48. Graham B. Plotting the right course. Vienna, Austria:
intraocular lens power calculation methods when tar- European Society of Cataract and refractive surgeons
geting low myopia in monovision. J Cataract Refract film festival; 2018.
Surg. 2020;46(6):862–6. 49. Vega Y, Gershoni A, Achiron A, Tuuminen R,
42. Wang L, Cao D, Weikert MP, Koch DD. Calculation Weinberger Y, Livny E, Nahum Y, Bahar I, Elbaz
of axial length using a single group refractive index U. High agreement between Barrett universal II calcu-
versus using different refractive indices for each lations with and without utilization of optional biom-
ocular segment: theoretical study and refractive out- etry parameters. J Clin Med. 2021;10(3):542.
comes. Ophthalmol. 2019;126(5):663–70. 50. Zhang Y, Li T, Reddy A, Nallasamy N. Gender differ-
43. Cooke DL, Cooke TL. A comparison of two meth- ences in refraction prediction error of five formulas for
ods to calculate axial length. J Cataract Refract cataract surgery. BMC Ophthalmol. 2021;21(1):183.
Surg. 2019;45(3):284–92. https://doi.org/10.1016/j. https://doi.org/10.1186/s12886-­021-­01950-­2.
jcrs.2018.10.039. 51. Lawless M, Jiang JY, Hodge C, Sutton G, Roberts TV,
44. Shammas HJ, Shammas MC, Jivrajka RV, Cooke Barrett G. Total keratometry in intraocular lens power
DL, Potvin R. Effects on IOL power calculation and calculations in eyes with previous laser refractive sur-
expected clinical outcomes of axial length measure- gery. Clin Exp Ophthalmol. 2020;48(6):749–56.
ments based on multiple vs. single refractive indices. 52. Savini G, Di Maita M, Hoffer KJ, Næser K, Schiano-­
Clin Ophthalmol. 2020;14(6):1511–9. Lomoriello D, Vagge A, Di Cello L, Traverso
45. Haigis W, Lege B, Miller N, Schneider B. Comparison CE. Comparison of 13 formulas for IOL power cal-
of immersion ultrasound biometry and partial culation with measurements from partial coherence
­coherence interferometry for intraocular lens calcu- interferometry. Br J Ophthalmol. 2021;105(4):484–9.
lation according to Haigis. Graefe's Archive for Clin 53. Barrett GD. The Barrett Rx formula: predicting IOL
and Exp Ophthalmol. 2000;238:765–73. power based on refraction after cataract surgery.
46. Norrby S. Sources of error in intraocular lens Barcelona. Spain: European Society of Cataract &
power calculation. J Cataract Refract Surg. Refractive Surgeons; 2015. https://escrs.conference-
2008;34(3):368–76. 2web.com/#!contentsessions/12537.
47. Barrett GD, Lipsky L. Integrated K to improve
toric IOL prediction. Washington, D.C.: Presented

Open Access This chapter is licensed under the terms of the Creative Commons Attribution 4.0 International License
(http://creativecommons.org/licenses/by/4.0/), which permits use, sharing, adaptation, distribution and reproduction in
any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to
the Creative Commons license and indicate if changes were made.
The images or other third party material in this chapter are included in the chapter's Creative Commons license,
unless indicated otherwise in a credit line to the material. If material is not included in the chapter's Creative Commons
license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to
obtain permission directly from the copyright holder.