Cataract Conversion Assessment using Lens Opacity Classification System III and Wisconsin Cataract Grading System

Abstract

Purpose.: To propose a general conversion approximation algorithm and illustrate its application to two cataract classification systems, the Lens Opacity Classification System III (LOCS III) and Wisconsin Cataract Grading System (Wisconsin system).

Methods.: Lens opacity of 3265 participants (78.7% response rate) aged 40 to 80 years from the Singapore Malay Eye study were assessed both clinically during slit-lamp examination using LOCS III, and via slit-lamp and retro-illumination photographic grading using the Wisconsin system. Collapsed contingency tables were used to convert between the LOCS III and Wisconsin System.

Results.: The conversion between the two cataract classification systems is affected by the direction of transformation. The conversion algorithm was validated and explained with an application example.

Conclusions.: An approximate conversion algorithm for any two cataract grading systems was proposed and applied to the LOCS III and Wisconsin system. These findings provide general ways to pool and compare cataract prevalence using different grading systems in epidemiological studies.

Introduction

Age-related cataracts remain the leading cause of blindness worldwide, posing a big challenge to rapidly aging populations. ^1–3 Prevalence and incidence of cataracts have been well examined in many population-based epidemiologic studies. ^4–31 Understanding the burden of cataracts from these prevalence and incidence data is important in the planning of eye care services, particularly health service delivery for cataract surgery. ³²  

Several cataract classification systems have been developed and used to measure the presence and extent of cataracts including the Lens Opacities Classification System (LOCS), ^{6,8,10,11,13–21,27,30} Wisconsin Cataract Grading System (Wisconsin system), ^7,9,23–26 Wilmer, ^12,33 Age-Related Eye Disease Study Grading System (AREDS), ^22,31 World Health Organization Simplified Cataract Grading System (WHOSCGS), ⁴ and Oxford Clinical Cataract Classification System (OCCGS) ⁵ resulting in several arbitrary cutoffs derived within and across the various systems (Tables 1 and 2). It is important to note that direct comparison of prevalence of cataracts between different studies is hampered by the diversity of classification systems using various assessment methods. ³⁴ In fact, some studies adopted more than one classification system simultaneously to assess different cataract subtypes. ¹³ Some other studies attempted to pool or compare prevalence of cataracts between studies despite having used different grading systems. ^19,35 Few studies have developed conversion scores between cataract classification systems, such as the calibration performed for LOCS III and OCCGS. ³⁶ There is a gap in general formulas or conversion algorithm that enables one system's grading to be converted into another to allow for reasonable comparison across centers or studies. 

Table 1. 

 

Prevalence of Nuclear Opalescence, Cortical, and PSC with Various Cutoffs used from Population-Based Studies

Study	Conduct Years	Country/ City	Ethnicity	Rural/Urban	Age Range	Grading Method	Nuclear (Cutoff)	Cortical (Cutoff)	PSC (Cutoff)
Beaver Dam Eye Study7	1988–1990	Wisconsin, US	Caucasians (nonHispanic)	Rural	43–84	Wisconsin	17.3% (≥4)	16.3% (≥5%)	6% (≥5%)
Blue Mountains Eye Study9	1992–1994	Sydney, Australia	Caucasians	Urban	49–96	Wisconsin	51.7% (≥4)	23.8% (≥5%)	6.3% (≥5%)
Tanjong Pagar Survey11	1997–1998	Singapore	Chinese	Urban	40–79	LOCS III	40.11% (≥4)	38.55% (≥2)	12.13% (≥2)
Aravind Comprehensive Eye Study10	1995–1997	South India	Indian	Rural	40+	LOCS III	44.7% (≥3)	27.1% (≥3)	22.9% (≥2)
Shihpai Eye Study19	1999–2000	Taiwan	Chinese	Urban	65+	LOCS III	38.9% (≥2)	21.9% (≥2)	9.2% (≥2)
Indonesia Eye Study17	2003	Indonesia	Malay	Rural	21+	LOCS III	16.89% (≥4)	15.68% (≥2)	7.35% (≥2)
Skovde Cataract Study18	2001	Sweden	Caucasians	Urban	70–84	LOCS III	14.37% (≥4)	6.69% (>3)	9.74% (>1)
Meiktila Eye Study14	2005	Myanmar	Burmese	Rural	40+	LOCS III	27.35% (≥4)	20.91% (≥2)	11.34% (≥2)
Kandy Eye Study15	2006–2007	Sri Lanka	Sinhalese, Tamils, Moors	Rural	40+	LOCS III	4.5% (≥4)	26.0% (≥2)	7.9% (≥2)
India Study of Age-related Eye Disease21	2005–2007	North India	Indian	Rural and Urban	60+	LOCS III	48% (≥4)	7.6% (≥3)	21% (≥2)
India Study of Age-related Eye Disease21	2005–2007	South India	Indian	Rural and Urban	60+	LOCS III	38% (≥4)	10.2% (≥3)	17% (≥2)
Handan Eye Study16	2006–2007	Hebei, China	Chinese	Rural	30+	LOCS III	5.1% (≥4)	18.3% (≥2)	1.5% (≥2)
Casteldaccia Eye Study6	1992	Italy	Caucasians	Rural	40–99	LOCS II	18.5% (≥2)	12.9% (≥2)	10.8% (≥2)
Barbados Eye Study8	1987–1992	Barbados	Blacks	Urban	40–84	LOCS II	19% (≥2)	34% (≥2)	4% (≥2)
			Mixed (Blacks and Whites)			LOCS II	20% (≥2)	30% (≥2)	5% (≥2)
			Whites			LOCS II	23% (≥2)	15% (≥2)	5% (≥2)
Los Angeles Latino Eye Study20	2000–2003	California, US	Latinos (Hispanics)	Urban	40+	LOCS II	9.0% (≥2)	13.4% (≥2)	3.1% (≥2)
Andhra Pradesh Eye Disease Study13	1996–2000	South India	Indian	Rural and Urban	16+	LOCS III & Wilmer*	12.4% (≥3)	7.4% (≥2)	8.1% (≥1)
Salisbury Eye Evaluation Project12	1993–1995	Maryland, US	Blacks	Rural and Urban	65–84	Wilmer	31.0% (≥2)	54.5% (≥1/8)	2.6% (Present)
			Caucasians		65–84	Wilmer	46.3% (≥2)	23.9% (≥1/8)	5.4% (Present)
Beijing Eye Study22	2001	Beijing, China	Chinese	Rural and Urban	40–101	AREDS	82% (≥2)	10.3% (≥5%)	4.3% (≥1%)
Kongwa Eye Project4	1996	Tanzania	Blacks	Rural	40+	WHOSCGS	15.6% (≥1)	8.8% (≥1)	1.9% (≥1)

*   LOCS III was used for nuclear opalescence while Wilmer was used for cortical and PSC cataract.

Table 2. 

 

Incidence Rate of Nuclear Opalescence, Cortical, and PSC with Various Cutoffs used from Population-Based Studies

Study	Start Year (Duration Years)	Country/City	Ethnicity	Rural/Urban	Age	Grading Method	Nuclear (Cutoff)	Cortical (Cutoff)	PSC (Cutoff)
Blue Mountains Eye Study23	1992 (10)	Sydney, Australia	Caucasians	Urban	49–97	Wisconsin	36% (≥4)	28% (≥25%)	9.1% (>0%)
Beaver Dam Eye Study25	1988 (15)	Wisconsin, US	NonHispanic Caucasian Americans	Rural and urban	43–84	Wisconsin	29.7% (≥4)	22.9% (≥5%)	8.4% (≥5%)
Barbados Eye Study27	1987 (9)	Barbados	African Barbadian	Urban	40–84	LOCS II	42% (≥2)	33.8% (≥2)	6.3% (≥2)
			Mixed (black and white)		40–84	LOCS II	42.2% (≥2)	22.4% (≥2)	3.6% (≥2)
			White Barbadian		40–84	LOCS II	36.5% (≥2)	14.2% (≥2)	7.1% (≥2)
Los Angeles Latino Eye Study30	2000 (4)	California, US	Latinos (Hispanics)	Urban	40	LOCS II	10.2% (≥2)	7.5% (≥2)	2.5% (≥2)
Beijing Eye Study31	2001 (5)	Beijing, China	Chinese	Rural and Urban	40–101	AREDS	5.98% (≥4)	11.14% (≥5%)	5.47% (≥1%)

In this study, we aim to develop a general algorithm that approximates conversion between any two cataract systems and illustrate the application in two major cataract classification systems, the LOCS III and Wisconsin system. 

Methods

Study Population

The Singapore Malay Eye Study (SiMES) investigated the prevalence, causes, and risk factors of blindness and visual impairment in the urban Malay community. SiMES is a population-based, cross-sectional epidemiological study of Asian Malays aged between 40 and 80 years old living in south western Singapore. Comprehensive details of the study have been reported and published elsewhere. ^37–39 Between August 2004 and June 2006, 3280 (78.7% participation rate from a total of 4168 eligible) Malays were examined in our study clinic. In total, 6530 eyes from 3265 SiMES participants were graded for cataract using the LOCS III and Wisconsin system; data we used to derive the conversion algorithm. The SiMES study conducted adhered to the Declaration of Helsinki and ethics approval was obtained by the Singapore Eye Research Institute institutional review board. The conversion algorithm was further validated in the Singapore Indian Eye Study ⁴⁰ (SINDI); a population-based, cross-sectional study of 3400 (75.6% participation rate) Indian adults aged 40 and above using the same study protocol as in SiMES. 

Cataract Classification Systems and Grading Procedures

Lens opacity was assessed using both the LOCS III and Wisconsin system, as described previously. ³⁴  

Table 3 summarizes the main characteristics of the LOCS III ⁴¹ and Wisconsin grading system. ⁴²  

Table 3. 

 

Characteristics of LOCS III and Wisconsin Cataract Grading System, WHOSCGS

System	LOCS III40		Wisconsin System41
	Clinical assessment with slit lamp or grading performed on retro-illumination photographs		Slit lamp photos for nuclear cataract; Retro-illumination photographs for cortical and PSC cataract
	NC	0–6.9	NO	0–5
	NO	0–6.9	Cortical	% involved
	Cortical	0–5.9	PSC	% involved
	PSC	0–5.9
	Yes		No

 NO, nuclear opalescence; NC, nuclear color.

Five study ophthalmologists examined all participants for cataracts using slit lamp biomicroscopy with a Haag-Streit slit-lamp microscope (model BQ-900; Haag-Streit, Köniz, Switzerland) in accordance to the LOCS III, ⁴¹ comparing with standard photographic slides for nuclear opalescence, nuclear color, cortical, and posterior subcapsular (PSC) cataract. Prior to the study, all study ophthalmologists were trained for the standardized examination that includes documentation of clinical diseases according to written IOP and assessment of lens opacity using LOCS III grading scale. Interrate reliability was assessed in a set of 30 patients with moderately high interrater reliability with intraclass correlation coefficients ranging from 0.70 to 0.85 between the study ophthalmologists. 

Additionally, lens photographs were taken by digital slit-lamp (Topcon model DC-1 with FD-21 flash attachment; Topcon, Tokyo, Japan) and retro-illumination (Nidek EAS-1000; Nidek, Gamagori, Japan) cameras during the examination, and lens opacity was assessed using the Wisconsin system. The slit beam was adjusted to completely fill the pupil and to vertically bisect the lens at a 45° angle focusing on the sulcus of the lens. All photographs were graded by a single trained grader at the University of Sydney who also graded cataracts for the Blue Mountains Eye Study. 14,378 photos were taken in total with at least two photos for each eye and only the best quality photo based on grader's judgment was evaluated. These photographs were compared against a set of four standards to determine degree of nuclear opacity. Cortical and PSC cataracts were assessed from retro-illumination photographs using an overlying grid to determine the location and percentage of lens involved by the opacity. Percentage of lens area involvement by cortical and PSC cataract were estimated for each segment of the grid in order to calculate the total percentage area of involvement. Adjudication was provided for images with positive nuclear cataract by a senior researcher and PSC cases were confirmed by a senior ophthalmologist. The intragrader reliability was high, with an intraclass correlation coefficient (ICC) of 0.95 (95% confidence interval: 0.93–0.97) in a random sample of 100 photographs regraded by the same grader. 

Statistical Analyses

All statistical analyses were performed using R version 2.14.2 (http://www.R-project.org/, provided in the public domain by R Development Core Team, 2012). ⁴³ Box plots were provided for graphical display of relationship between the LOCS III and Wisconsin system for each cataract subtype. 

Conversion Algorithm.

A contingency table was used to record and analyze the multivariate frequency distribution of modified categorical LOCS III and Wisconsin system. In order to increase the power and also the counts in each cell of the contingency table, a collapsing method was performed as in the following algorithm: 

1.  
   Calculate the conditional frequency distribution;
2.  
   Compare conditional frequency distribution of any two contiguous categories;
3.  
   Collapse the two contiguous categories with smallest distance or similar distribution; and
4.  
   Repeat the first three steps until desired number of contiguous categories is achieved.

In this algorithm, the smallest distance was defined as the least absolute shrinkage in terms of L1-norm. 

Conversion Application using LOCS III and Wisconsin System.

For the nuclear opalescence score, LOCS III ranged from 0.1 to 6.9 with one decimal, while the Wisconsin system used a five-point scale by comparing participant photographs of the eye with the set of four Wisconsin standard photographs. Also, a decimalized system of nuclear grading of one decimal place was estimated on a continuous scale between each standard by the grader (e.g., 3.8). We first start by collapsing the grading scheme used for LOCS III to encompass only half-unit steps: 0.1 to 0.4, 0.5 to 0.9, 1.0 to 1.4, 1.5 to 1.9 … 6.5 to 6.9 before applying our algorithm in the contingency table with Wisconsin system using a five-point scale. LOCS III cortical and PSC scores ranged from 0.1 to 5.9, while Wisconsin system measured the percentage of area involved. LOCS III cortical and PSC grading schemes were collapsed similarly as for nuclear score into half-unit steps. Continuous Wisconsin system for cortical was recoded into fewer categories of 5% incremental steps from 0 to 100 while PSC was initially collapsed as 0 (since 0 may potentially be used as cutoffs), 1 to 4, and 5% increment steps thereafter from 5 to 100. Conversion algorithm was then applied to collapse the LOCS III and Wisconsin system for all three cataract subtypes until we obtain five contiguous categories (i.e., 5 by 5 contingency table). The order of collapsing has little influence on the results. 

The final collapsed contingency tables of nuclear opalescence, cortical, and PSC cataract were used as guidelines for conversion between the LOCS III and Wisconsin system. This method works as a classification approach that tries to minimize the difference within each category and at the same time maximize the variability between any categories. We attempted to maximize the likelihood of the multivariate frequency distribution while restricting to five categories for each classification system. The conversion between the LOCS III and Wisconsin system was based on conditional probabilities, which reflect the likelihood of falling into corresponding LOCS III or Wisconsin system categories. 

Data Cleaning.

We started with the initial contingency table having maximal contiguous categories collapsed arbitrarily. Data points in any cells of the final collapsed contingency table that have conditional probabilities of lesser than 10% were regarded as noise data (may be due to measurement errors) and were removed. Conditional probabilities were then standardized to ensure they summed up to one. 

Validation.

The conversion was validated using the cross-validation method. ⁴⁴ Our data were randomly divided into 10 subsets. Nine subsets of data were used to construct the collapsed contingency table to propose the conversion guideline. The remaining subset of data was used to test the conversion trained by the nine parts of data. We further validated our conversion algorithm with the SINDI population. 

Results

Figure 1 shows the distributions of individual cataract subtypes in SiMES. We observed possible quadratic or nonlinear relationship. 

Figure 1. 

 

Scatter plots and box plots for each cataract subtypes (nuclear opalescence, cortical, and PSC) using the Wisconsin system (Wisconsin) and LOCS III; red lines were estimated by using general additive model.

As the relationship between the LOCS III and Wisconsin system is not one-to-one due to the differences in score ranges and interval sizes, the conversion from the LOCS III to Wisconsin system was different from the reverse direction of the Wisconsin system to LOCS III (i.e., not vice versa). We therefore performed collapsed algorithm for the conversion between the two systems, and the results of our final collapsed contingency table was tabulated in Supplementary Table S1 (see Supplementary Material and Supplementary Table S1). Figure 2 illustrates our proposed conversion approximation results for nuclear, cortical, and PSC cataracts individually from the LOCS III to Wisconsin system and the reverse order of the Wisconsin system to LOCS III. Figure 3 shows our validation analysis performed in 10% test SiMES data and in SINDI data. Relative frequencies of subjects in corresponding collapsed categories after conversion is almost identical to that of original scale for moderate to severe cataract. 

Figure 2. 

 

Conversion between LOCS III and Wisconsin system. Nuclear: nuclear opalescence; bold lines and numbers in plots were estimated from the conditional probabilities given either LOCS III or Wisconsin (conditional on the grading system to be converted).

Figure 3. 

 

Validation of Conversion Algorithm on Relative Subject Frequency in 10% Test data and SINDI data. Columns A, C: Original: percent of subjects in original LOCS III grade; Converted: percent of subjects using converted LOCS III grade (i.e., after conversion from Wisconsin to LOCS III scale). Columns B, D: Original: percent of subjects in original Wisconsin grade; Converted: percent of subjects using converted Wisconsin grade (i.e., after conversion from LOCS III to Wisconsin scale).

The guide to use Figure 2 is as follows: for example, to find the corresponding LOCS III score from the Wisconsin system scale for nuclear opalescence, refer to the top left graph in Figure 2. Assuming the scale in Wisconsin system is 1, Figure 2 shows that two corresponding LOCS III categories, “0.1–2.9” and “3.0–3.9” are most likely to match this Wisconsin system scale of 1 with conditional probability of 0.65 and 0.35, respectively. We may infer that it is 65% likely to be in the “0.1–2.9” category and 35% likely to be in the “3.0–3.9” category. Hence, in the consideration of prevalence, this subject with Wisconsin scale 1 reading contributes 0.65 headcount to “0.1–2.9” and 0.35 headcount to “3.0–3.9.” 

Discussion

We applied our proposed algorithm to approximate the conversion between two major cataract grading systems, the LOCS III and Wisconsin system by collapsing the multivariate frequency distribution contingency table. Our conversion algorithm can be extended and applied to other cataract grading systems. There is a need for such a method of conversion, as prevalence and incidence of cataract cannot be compared directly between studies that were assessed using different classification systems. 

The estimation of prevalence of cataract varies substantially with different grading protocols, ³⁴ which may be often neglected in the pooling and comparison of estimates seen in a few studies. ^19,35 The lack of universal epidemiologic definition of cataract cutoffs added to the inaccurate comparison of cataract prevalence even within studies using common grading systems. Pooling of cataract prevalence in the United States ³⁵ was performed to include Barbados Eye Study (BES) with LOCS II, ⁸ Beaver Dam Eye Study (BDES) with the Wisconsin system, ⁷ Blue Mountain Eye Study (BMES) with the Wisconsin system, ⁹ Salisbury Eye Evaluation Project with Wilmer, ¹² and Melbourne Visual Impairment Project with Wilmer. ⁴⁵ Differences in grading methods, definitions of lens opacities, and examination techniques limit the accuracy of conclusion with regards to pooled prevalence of cataract. 

The application of our proposed conversion method provided a conversion approximation to transform between the LOCS III and Wisconsin system. For example, converting nuclear opalescence from the Wisconsin system to LOCS III in our data gave estimated prevalence of 24.8% (based on the traditional or optimal cutoff of ≥4) compared with prevalence of 26.7% from direct use of the LOCS III with the same cutoff. The reverse conversion of nuclear opalescence from the LOCS III to Wisconsin system gave estimated prevalence of 17.8% (based on the traditional or optimal cutoff of ≥4) compared with prevalence of 16.8% from direct use of the Wisconsin system. The small difference in the converted prevalence and original prevalence on the same scale suggests good approximation derived from our conversion algorithm. Conversion for cortical and PSC had similar performance. Our conversion algorithm allows fairer and more accurate inferences based on the same scale than current naïve comparisons of prevalence and pooling analysis performed directly across studies using different systems. In addition, the cross-validation analysis (Fig. 3) demonstrated that our method was very robust and that our conversion algorithm may be extended to other population data for all three cataract subtypes. 

Our findings have important insights and implications. We provided a general conversion algorithm and its application to approximate the conversion between the LOCS III and Wisconsin system to improve the pooling or comparison of prevalence of cataracts. The Wisconsin system for assessment of cataracts was also used in BMES and BDES, the two landmark epidemiological studies in eye research. At present, there continues to be important new papers on epidemiology of cataract from these two studies using the Wisconsin system, ^46,47 while more recent studies have used the LOCS III. Our study is therefore important by being the first study to directly compare the two systems. 

Large overlaps observed in early cataract scores between grading systems suggests difficulty in the discrimination of subtle lens opacity changes and the detection of early cataracts. Newer methods under development such as Quasielastic or Dynamic Light Scattering (QELS or DLS) with the Scheimpflug imaging system may be more objective and promising as such methods in clinical use have shown that a growing cataract can be detected at the molecular level using the technique of dynamic light scattering. ⁴⁸  

The strengths of our study include a large sample size from two population-based samples, and having performed two standardized cataract grading protocols in the same population. Our main limitation in the study application is the grading variability between clinical grading at the slit-lamp compared with grading lens photos. However, in many clinical and epidemiological studies and multicenter trials, clinical grading at the slit-lamp may be the only feasible (and less expensive and complex) approach, particularly when cataract is important but of secondary interest (e.g., many landmark trials on anti-VEGF injection treatment for age-related macular degeneration (AMD) had clinical LOCS grading for cataract such as Comparison of AMD Treatments Trials [CATT], ⁴⁹ Anti-VEGF antibody for the treatment of predominantly classic choroidal neovascularisation in AMD [ANCHOR], ⁵⁰ Minimally Classic/Occult Trial of the Anti-VEGF Antibody Ranibizumab in the Treatment of Neovascular AMD [MARINA], ⁵¹ and a Study of rhuFAB V2 [Ranibizumab] in Subjects with Subfoveal Choroidal Neovascularization Secondary to AMD [PIER] ^52,53 ). Our conversion algorithm application is therefore more practically relevant and allows comparison and conversion of clinical LOCS III grading performed at the slit lamp with grading of photographs using the Wisconsin system. Further investigation needs to be conducted to ensure our conversion algorithm is widely applicable in other population data. Conversion based on original protocols of grading systems or lens images should be further explored. 

In conclusion, we proposed a general algorithm that approximates conversion between any two cataract systems and illustrated its application in two major cataract classification systems, the LOCS III and Wisconsin system. The transformation is not one-to-one and is validated by using cross-validation. The results of our study suggest that prevalence rates of cataract need to be converted to the same scale before comparison between different grading systems, and ideally with standardized universal epidemiologic definition of cutoffs for cataract subtypes. 

Supplementary Materials

doc S1 

Footnotes

 Supported by grants from National Medical Research Council (NMRC/0796/2003).

Footnotes

 Disclosure: W.L. Wong, None; X. Li, None; J. Li, None; C.-Y. Cheng, None; E.L. Lamoureux, None; J.J. Wang, None; C.Y. Cheung, None; T.Y. Wong, None