Abstract
Purpose:
To use the Chennai glaucoma study (CGS) dataset to calculate biometric cutoffs to detect primary angle-closure disease (PACD) and assess its accuracy in detecting PACD in a separate clinical subset.
Methods:
Clinical subset, prospective recruitment. Categories – normal, PACD (included primary angle-closure suspect-PACS, primary angle closure – PAC, and primary angle-closure glaucoma-PACG), and primary open-angle glaucoma (POAG). Biometric parameters assessed included axial length (AXL), anterior chamber depth (ACD), lens thickness (LT), lens position, relative lens position, and simple crowding value. Additionally, the CGS dataset biometric parameters were used to determine the optimal cutoff values (CVs) for the detection of PACD. These were applied to the clinical dataset.
Results:
A total of 942 eyes (474 patients) – clinical, 1844 eyes (1844 patients) – CGS. Mean age: 55.77+/-9.63 yrs – clinical, 52.04 ± 9.75 yrs – CGS (rural), 53.95 ± 9.57 yrs – CGS (urban). Among all biometric parameters across all datasets, ACD had the highest area under the curve (AUC, 0.86–0.89), accuracy (75.56–82.91%). Assessed the ability of ACD CV determined from CGS to detect PACD in the clinical dataset – highest sensitivity, specificity combination seen with CV from CGS (rural) – ACD ≤ 2.815 mm: sensitivity – 63.59%, specificity – 93.21%. Urban and combined rural and urban (U and C) CGS dataset had a similar ACD CV of ≤ 2.695 mm (44.77% sensitivity, 96.74%specificity). At 10% population prevalence, ACD ≤ 2.695 (U and C) had the highest accuracy (91.54%), followed by CGS (rural) CV-ACD ≤ 2.815 (90.24%). At a prevalence 30% accuracy: 81.15% for the U and C and 84.32% for rural CV.
Conclusion:
ACD determined from CGS had an accuracy of 92% and 84% in detecting PACD in a clinical setting at 10% and 30% prevalence, respectively. Use of these parameters to determine PACD may be considered with due caution in resource-constrained areas, along with cataract screening.
Keywords: Anterior chamber depth, biometry, primary angle-closure disease, primary angle-closure glaucoma, screening
Primary angle-closure glaucoma (PACG) accounts for about 40% of all primary glaucoma in Asia.[1] A large majority of angle-closure disease is undiagnosed or misdiagnosed. Among patients diagnosed to have PACG in the Chennai glaucoma study (CGS), only 14.7% in the urban arm and none of those affected in the rural arm were aware of their eye condition.[2] Additionally, 40% of those who knew they had glaucoma in the CGS urban arm had angle-closure glaucoma and were misdiagnosed as open-angle glaucoma.[3]
PACG is a more aggressive disease compared to primary open-angle glaucoma (POAG), with more than three times higher risk of blindness (8.9 vs. 27%) noted in a meta-analysis of all population-based studies.[4] Biometric parameters have been shown to be able to discriminate between angle-closure disease and normal controls.[5,6,7] Studies have also shown that biometric parameters may be used to stratify patients’ risk of progression in angle-closure disease.[8] Whether these parameters are valid when applied to other datasets remains unclear.
We aimed to detect optimum cutoffs for biometric parameters from the CGS dataset and assess its ability to detect primary angle-closure disease (PACD) in a clinical subset.
Methods
The study adhered to the principles of the Declaration of Helsinki, and ethical committee clearance was obtained. The methodology of this study is illustrated in Fig. 1.
Figure 1.
Flow chart depicting study methodology
A cross-section of patients reporting to the glaucoma department of a tertiary eye care center from January 2020 to June 2023 was prospectively recruited after obtaining signed informed consent. Adult phakic patients with a best-corrected Snellen visual acuity (BCVA) of 6/18 or better were included. We included those belonging to normal, POAG, and PACD (comprised of three subcategories, i.e. primary angle-closure suspect – PACS, primary angle closure – PAC, and primary angle-closure glaucoma – PACG).
Glaucomatous eyes were defined as those with optic nerve head (ONH) and retinal nerve fibre layer (RNFL) findings clinically suggestive of glaucoma with corresponding glaucomatous changes on visual fields or optical coherence tomography (OCT). PACD categories were defined as per established guidelines.[9] We also excluded patients with any ocular comorbidity involving the optic nerve, macula, or retina, including diabetic retinopathy (DR), age-related macular degeneration, retinal venous occlusions (RVOs), degenerative myopia, and any other congenital or acquired disease of the optic nerve or macula other than glaucoma that could potentially interfere with the acquisition or results of biometry or visual fields. Patients with a history of ocular surgeries other than glaucoma surgeries were excluded.
History and comprehensive ocular examination of patients were recorded, including Intraocular pressure (IOP) (Goldmann applanation tonometry, Haag Streit diagnostics), indentation gonioscopy (Volk G4 gonioscopy lens), and anterior and posterior segment evaluation. Refractive error was assessed using Visuref 100 (Carl Zeiss Meditec) and subjective refraction. Ocular biometry was assessed using IOLMaster 700 (Carl Zeiss, Meditec) prior to dilation. Visual fields were assessed using the Humphrey visual field (HVF) 24-2 (HFA3 850, Carl Zeiss Meditec) Swedish Interactive Threshold Algorithm (SITA) standard in glaucoma patients and SITA faster in normal eyes.
The axial length (AXL), anterior chamber depth (ACD), and lens thickness (LT) were noted. We used these parameters to calculate the lens position (LP = ACD + 0.5LT), relative lens position (RLP = (ACD+ (LT/2))/AXL), and simple crowding value (Cs = (LT-ACD)/AXL), which have been used to detect angle closure.[6,10,11]
The CGS is a population-based cross-sectional study conducted in Tamil Nadu. The details of the CGS are described elsewhere.[12] We used the biometry dataset from the rural and urban arms of the CGS.[2,13] Ocular biometry was performed in the CGS using an ultrasonic biometer (Ocuscan, 71 Alcon Laboratories Inc., Fort Worth, TX) on all PACD patients every fifth normal subject. Biometric values of the right eye were used to calculate composite indices as in the clinical dataset. The receiver operating characteristic (ROC) curve was plotted, and the optimum cutoff value (CV) was determined. We used the CV from the CGS biometry parameters and determined its ability to detect PACD in the clinical population using statistical parameters such as sensitivity, specificity, positive, and negative likelihood ratios (PLR, NLR) and positive and negative predictive values (PPV, NPV).
Statistical analyses were performed using SPSS V20. Descriptive statistics were obtained for continuous variables, and frequency distribution was calculated for qualitative variables. The ROC curve was plotted for the detection of PACD, and the optimum cutoff was determined using the sensitivity, specificity parameters and Youden index.[14] The sensitivity, specificity, positive, and negative likelihood ratios of biometric parameters from the CGS cutoff when assessed in the clinical population were determined using MedCalc Statistical Software version 19.2.6.[15] We also used PACD prevalence rates of 10% (population prevalence) and 30% (clinical prevalence) to calculate the PPV, NPV, and accuracy on Medcalc.[2,13] Accuracy was calculated from sensitivity, specificity, and prevalence according to the formula:
Accuracy = Sensitivity x prevalence + specificity x (1- prevalence)[15]
Values adjacent to the ACD CVs determined from all datasets were also used in the clinical dataset to look for test metrics in detecting PACD at 10 and 30% prevalence.
Results
A total of 1902 eyes from 1902 patients were obtained from the CGS biometry dataset. After excluding those with missing biometric parameters for the right eye, we obtained 1844 eyes from 1844 patients, which were used for analysis. The clinical dataset initially included 996 eyes of 498 patients. After excluding those with other ocular diseases (18 eyes of 11 patients), glaucoma suspects or those with diagnostic dilemmas (25 eyes of 16 patients), and those without HVF (11 eyes of six patients), we analyzed 942 eyes of 474 patients.
The patient characteristics of the CGS and clinical data are presented in Table 1. Age, sex, and biometry showed varying mean values with overlapping 95% confidence intervals. The mean age was higher in the clinical dataset (55.77 ± 9.63 yrs, P < 0.001) compared to the CGS dataset (rural – 52.04 ± 9.75 yrs, urban – 53.95 ± 9.57 yrs), and more males were recruited in the clinical dataset compared to the CGS dataset (P < 0.001). The proportion of angle-closure disease and the disease profile of the clinical dataset differed from that of the population-based CGS.
Table 1.
Patient characteristics
| Characteristic | CGS rural | CGS urban | Clinical dataset | |
|---|---|---|---|---|
| 1 | Age (yrs) | 52.04 | 53.95 | 55.77 |
| SD | 9.75 | 9.57 | 9.63 | |
| 2 | M/F | 391/497 | 415/541 | 266/209 |
| 3 | Disease category | |||
| Normal | 614 (69.1%) | 580 (60.7%) | 193 (20.5%) | |
| PACS | 222 (25%) | 248 (25.9%) | 203 (21.5%) | |
| PAC | 24 (2.7%) | 99 (10.4%) | 191 (20.3%) | |
| PACG | 28 (3.1%) | 29 (3.0%) | 180 (19.1%) | |
| POAG | - | - | 175 (18.6%) | |
| 2 | ACD (mm) | 2.84 | 2.73 | 2.93 |
| SD | 0.37 | 0.38 | 0.38 | |
| 3 | AXL (mm) | 22.47 | 22.70 | 23.08 |
| SD | 0.78 | 1.07 | 1.11 | |
| 4 | LT (mm) | 4.35 | 4.39 | 4.47 |
| SD | 0.41 | 0.47 | 0.39 | |
CGS=Chennai glaucoma study, SD=standard deviation, PACS=Primary angle-closure suspect, PAC=Primary angle closure, PACG=Primary angle-closure glaucoma, POAG=Primary open-angle glaucoma, ACD=Anterior chamber depth, AXL=Axial length, LT=Lens thickness
The optimal CV was determined from all biometric parameters assessed in the CGS and clinical dataset after plotting the ROC curves [Table 2]. To determine this, we looked at the area under curve (AUC) and the coordinate values of the ROC curve, which showed the CVs with corresponding sensitivity and 1-specificity values in determining PACD. The ACD had the highest AUC (0.86–0.89) and accuracy (75.56–82.91%) among all the parameters across all datasets. Other parameters with a good combination of test metrics included LP (AUC: 0.803–0.819, Accuracy: 73.32–76.24%) and Cs (AUC: 0.647–0.868, Accuracy: 69.87–80.04%).
Table 2.
Optimal cutoff values in datasets
| Parameter | Cutoff value | Sensitivity | Specificity | AUC | AUC CI | Accuracy | Accuracy CI |
|---|---|---|---|---|---|---|---|
| CGS – all eyes | |||||||
| ACD≤ | 2.695 | 78.77% | 77.22% | 0.861 | 0.844 to 0.879 | 77.77% | 75.80 to 79.65% |
| LP≤ | 5.000 | 79.54% | 69.93% | 0.811 | 0.790 to 0.831 | 73.32% | 71.24 to 75.33% |
| AXL≤ | 22.535 | 69.08% | 62.73% | 0.714 | 0.690 to 0.738 | 64.97% | 62.74 to 67.15% |
| RLP≤ | 0.223 | 71.08% | 60.05% | 0.716 | 0.691 to 0.740 | 63.77% | 61.53 to 65.97% |
| LT≥ | 4.540 | 56.77% | 64.82% | 0.592 | 0.563 to 0.621 | 61.98% | 59.72 to 64.21% |
| Cs≥ | 0.081 | 64.77% | 76.80% | 0.647 | 0.721 to 0.771 | 72.51% | 70.41 to 74.53% |
| CGS – rural | |||||||
| ACD≤ | 2.815 | 87.23% | 70.36% | 0.861 | 0.835 to 0.886 | 75.56% | 72.60 to 78.36% |
| LP≤ | 4.933 | 68.25% | 79.80% | 0.803 | 0.772 to 0.834 | 76.24% | 73.30 to 79.00% |
| AXL≤ | 22.535 | 75.18% | 57.33% | 0.719 | 0.683 to 0.755 | 62.84% | 59.56 to 66.03% |
| RLP≤ | 0.224 | 69.34% | 65.47% | 0.724 | 0.687 to 0.760 | 66.67% | 63.46 to 69.76% |
| LT≥ | 4.540 | 55.11% | 69.54% | 0.610 | 0.567 to 0.654 | 65.09% | 61.85 to 68.23% |
| Cs≥ | 0.072 | 75.18% | 69.54% | 0.762 | 0.725 to 0.798 | 71.17% | 68.07 to 74.13% |
| CGS – urban | |||||||
| ACD≤ | 2.695 | 81.91% | 73.62% | 0.860 | 0.836 to 0.883 | 76.88% | 74.08 to 79.52% |
| LP≤ | 5.000 | 82.71% | 67.24% | 0.813 | 0.786 to 0.841 | 73.33% | 70.40 to 76.11% |
| AXL≤ | 22.425 | 59.31% | 74.31% | 0.726 | 0.693 to 0.758 | 68.41% | 65.36 to 71.35% |
| RLP≤ | 0.215 | 52.13% | 78.45% | 0.701 | 0.667 to 0.736 | 67.57% | 64.50 to 70.53% |
| LT≥ | 4.595 | 46.28% | 72.76% | 0.572 | 0.533 to 0.611 | 62.34% | 59.19 to 65.42% |
| Cs≥ | 0.080 | 67.82% | 71.38% | 0.730 | 0.695 to 0.764 | 69.87% | 66.86 to 72.77% |
| Clinical dataset | |||||||
| ACD≤ | 2.985 | 84.15% | 80.98% | 0.890 | 0.869 to 0.911 | 82.91% | 80.35 to 85.26% |
| LP≤ | 5.213 | 77.18% | 72.55% | 0.819 | 0.792 to 0.846 | 75.37% | 72.49 to 78.09% |
| AXL≤ | 22.975 | 63.76% | 77.99% | 0.773 | 0.742 to 0.803 | 69.32% | 66.27 to 72.25% |
| RLP≤ | 0.227 | 66.20% | 50.54% | 0.601 | 0.564 to 0.638 | 59.66% | 56.45 to 62.81% |
| LT≥ | 4.385 | 73.69% | 64.67% | 0.751 | 0.719 to 0.782 | 70.17% | 67.14 to 73.08% |
| Cs≥ | 0.062 | 81.71% | 77.72% | 0.868 | 0.845 to 0.891 | 80.04% | 77.34 to 82.55% |
AUC=Area under curve, CI=95% confidence interval, CGS=Chennai glaucoma study, ACD=Anterior chamber depth, LP=Lens position, AXL=Axial length, RLP=Relative lens position, LT=Lens thickness, Cs=Simple crowding value
The sensitivity, specificity, and likelihood ratio of a test remain the same with varying prevalence. However, it can change with differences in the spectrum of the disease in different settings.[16,17] We used these three biometric parameters (ACD, LP, and Cs) from CGS to determine its effectiveness in detecting PACD in a clinical setting [Tables 3 and 4]. The highest combination of sensitivity and specificity was seen in the rural dataset cutoff (ACD ≤ 2.815 mm: sensitivity – 63.59%, specificity – 93.21%). At the urban and overall CGS dataset cutoff ACD ≤ 2.695 mm, the sensitivity was 44.77% and specificity 96.74%. We also assessed the effect of varying prevalence on the test metrics. As expected, the test metrics varied when using the CGS CVs of the biometric parameters in the clinical dataset, at a varying prevalence of 10% (prevalence in the population, Table 3) and 30% (prevalence in the clinic, Table 4).
Table 3.
Test parameters obtained using the CGS cutoff in the clinical dataset at 10% prevalence
| Statistics: (Prevalence 10%) | CGS all eyes |
CGS rural |
CGS urban |
|||
|---|---|---|---|---|---|---|
| Value | 95%CI | Value | 95%CI | Value | 95%CI | |
| ACD ≤2.695 | ACD ≤2.815 | ACD ≤2.695 | ||||
| Sensitivity | 44.77% | 40.65 to 48.95% | 63.59% | 59.50 to 67.53% | 44.77% | 40.65 to 48.95% |
| Specificity | 96.74% | 94.37 to 98.30% | 93.21% | 90.13 to 95.56% | 96.74% | 94.37 to 98.30% |
| PLR | 13.730 | 7.81 to 24.13 | 9.360 | 6.38 to 13.74 | 13.730 | 7.81 to 24.13 |
| NLR | 0.570 | 0.53 to 0.62 | 0.390 | 0.35 to 0.44 | 0.570 | 0.53 to 0.62 |
| PPV | 60.41% | 46.47 to 72.84% | 50.98% | 41.48 to 60.41% | 60.41% | 46.47 to 72.84% |
| NPV | 94.04% | 93.59 to 94.45% | 95.84% | 95.37 to 96.26% | 94.04% | 93.59 to 94.45% |
| Accuracy | 91.54% | 89.58 to 93.24% | 90.24% | 88.17 to 92.06% | 91.54% | 89.58 to 93.24% |
| LP ≤5.000 | LP ≤4.933 | LP ≤5.000 | ||||
| Sensitivity | 45.99% | 41.86 to 50.17% | 34.49% | 30.61 to 38.54% | 45.99% | 41.86 to 50.17% |
| Specificity | 91.58% | 88.26 to 94.20% | 96.20% | 93.70 to 97.90% | 91.58% | 88.26 to 94.20% |
| PLR | 5.460 | 3.85 to 7.74 | 9.070 | 5.36 to 15.34 | 5.460 | 3.85 to 7.74 |
| NLR | 0.590 | 0.54 to 0.64 | 0.680 | 0.64 to 0.73 | 0.590 | 0.54 to 0.64 |
| PPV | 37.76% | 29.98 to 46.22% | 50.19% | 37.32 to 63.03% | 37.76% | 29.98 to 46.22% |
| NPV | 93.85% | 93.36 to 94.30% | 92.97% | 92.54 to 93.37% | 93.85% | 93.36 to 94.30% |
| Accuracy | 87.02% | 84.70 to 89.10% | 90.03% | 87.93 to 91.86% | 87.02% | 84.70 to 89.10% |
| Cs ≥0.081 | Cs ≥0.072 | Cs ≥0.080 | ||||
| Sensitivity | 47.56% | 43.41 to 51.74% | 64.98% | 60.92 to 68.89% | 50.52% | 46.35 to 54.69% |
| Specificity | 94.29% | 91.41 to 96.43% | 87.23% | 83.38 to 90.46% | 92.93% | 89.82 to 95.33% |
| PLR | 8.330 | 5.45 to 12.74 | 5.090 | 3.87 to 6.69 | 7.150 | 4.89 to 10.45 |
| NLR | 0.560 | 0.51 to 0.60 | 0.400 | 0.36 to 0.45 | 0.530 | 0.49 to 0.58 |
| PPV | 46.00% | 36.23 to 56.08% | 36.12% | 30.07 to 42.64% | 44.28% | 35.22 to 53.73% |
| NPV | 94.19% | 93.72 to 94.62% | 95.73% | 95.22 to 96.19% | 94.41% | 93.94 to 94.86% |
| Accuracy | 89.17% | 87.00 to 91.08% | 85.00% | 82.56 to 87.22% | 88.69% | 86.50 to 90.65% |
CGS=Chennai glaucoma study, ACD=Anterior chamber depth, PLR=Positive likelihood ratio, NLR=Negative likelihood ratio, LP=Lens position, Cs=Simple crowding value
Table 4.
Test parameters obtained using the CGS cutoff in the clinical dataset at 30% prevalence
| Statistics: (Prevalence 30%) | CGS all eyes |
CGS rural |
CGS urban |
|||
|---|---|---|---|---|---|---|
| Value | 95%CI | Value | 95%CI | Value | 95%CI | |
| ACD ≤2.695 | ACD ≤2.815 | ACD ≤2.695 | ||||
| Sensitivity | 44.77% | 40.65 to 48.95% | 63.59% | 59.50 to 67.53% | 44.77% | 40.65 to 48.95% |
| Specificity | 96.74% | 94.37 to 98.30% | 93.21% | 90.13 to 95.56% | 96.74% | 94.37 to 98.30% |
| PLR | 13.730 | 7.81 to 24.13 | 9.360 | 6.38 to 13.74 | 13.730 | 7.81 to 24.13 |
| NLR | 0.570 | 0.53 to 0.62 | 0.390 | 0.35 to 0.44 | 0.570 | 0.53 to 0.62 |
| PPV | 85.47% | 77.00 to 91.18% | 80.05% | 73.22 to 85.48% | 85.47% | 77.00 to 91.18% |
| NPV | 80.34% | 79.11 to 81.52% | 85.66% | 84.23 to 86.98% | 80.34% | 79.11 to 81.52% |
| Accuracy | 81.15% | 78.50 to 83.60% | 84.32% | 81.84 to 86.59% | 81.15% | 78.50 to 83.60% |
| LP ≤5.000 | LP ≤4.933 | LP ≤5.000 | ||||
| Sensitivity | 45.99% | 41.86 to 50.17% | 34.49% | 30.61 to 38.54% | 45.99% | 41.86 to 50.17% |
| Specificity | 91.58% | 88.26 to 94.20% | 96.20% | 93.70 to 97.90% | 91.58% | 88.26 to 94.20% |
| PLR | 5.460 | 3.85 to 7.74 | 9.070 | 5.36 to 15.34 | 5.460 | 3.85 to 7.74 |
| NLR | 0.590 | 0.54 to 0.64 | 0.680 | 0.64 to 0.73 | 0.590 | 0.54 to 0.64 |
| PPV | 70.06% | 62.29 to 76.83% | 79.53% | 69.66 to 86.80% | 70.06% | 62.29 to 76.83% |
| NPV | 79.82% | 78.48 to 81.11% | 77.41% | 76.29 to 78.49% | 79.82% | 78.48 to 81.11% |
| Accuracy | 77.90% | 75.11 to 80.51% | 77.69% | 74.89 to 80.31% | 77.90% | 75.11 to 80.51% |
| Cs ≥0.081 | Cs ≥0.072 | Cs ≥0.080 | ||||
| Sensitivity | 47.56% | 43.41 to 51.74% | 64.98% | 60.92 to 68.89% | 50.52% | 46.35 to 54.69% |
| Specificity | 94.29% | 91.41 to 96.43% | 87.23% | 83.38 to 90.46% | 92.93% | 89.82 to 95.33% |
| PLR | 8.330 | 5.45 to 12.74 | 5.090 | 3.87 to 6.69 | 7.150 | 4.89 to 10.45 |
| NLR | 0.560 | 0.51 to 0.60 | 0.400 | 0.36 to 0.45 | 0.530 | 0.49 to 0.58 |
| PPV | 76.66% | 68.67 to 83.12% | 68.56% | 62.38 to 74.14% | 75.40% | 67.71 to 81.75% |
| NPV | 80.77% | 79.45 to 82.02% | 85.32% | 83.78 to 86.74% | 81.42% | 80.06 to 82.71% |
| Accuracy | 80.00% | 77.30 to 82.51% | 80.55% | 77.88 to 83.04% | 80.21% | 77.52 to 82.71% |
CGS=Chennai glaucoma study, ACD=Anterior chamber depth, PLR=Positive likelihood ratio, NLR=Negative likelihood ratio, LP=Lens position, Cs=Simple crowding value
At a 10% prevalence, i.e., population prevalence, the cutoff of ACD ≤ 2.695 from urban and all CGS datasets, had the highest ability to detect PACD with an accuracy of 91.54% (PPV – 60.41%, NPV – 94.04%), followed by the CGS rural cutoff of ACD ≤ 2.815 with an accuracy of 90.24% (PPV – 50.98%, NPV – 95.84%, Table 3). A change to a higher prevalence of 30% resulted in a change in accuracy to 81.15% (PPV – 85.47%, NPV – 80.34%) for the urban cutoff of ACD ≤ 2.695 and to 84.32% (PPV – 80.05%, NPV – 85.66%) for the rural cutoff of ACD ≤ 2.815 [Table 4].
We also looked at the ability of adjacent ACD CVs to detect PACD in the clinical dataset. Fig. 2 shows the varying sensitivities and specificities in detecting PACD in the clinical dataset with varying ACD CVs
Figure 2.
ACD cutoff versus sensitivity and specificity in detecting PACD in the clinical dataset
The highest combination of sensitivity and specificity (84.15% and 80.98%) was seen with the cutoff taken from the clinical dataset itself. However, it did not have the highest accuracy at a prevalence of 10% and 30%. The test metrics (sensitivity, specificity, accuracy, and predictive values) for detecting PACD at different ACD CVs in the clinical dataset at 10 and 30% prevalence are shown in Table 5.
Table 5.
Test metrics in detecting Primary angle-closure disease vs anterior chamber depth cutoff in clinical dataset
| ACD (mm) | Sensitivity | Specificity | PLR | NLR | PPV | NPV | Accuracy |
|---|---|---|---|---|---|---|---|
| 10% prevalence | |||||||
| ≤2.50 | 20.38% | 98.91% | 18.75 | 0.8 | 67.57% | 91.79% | 91.06% |
| ≤2.60 | 30.66% | 98.10% | 16.12 | 0.71 | 64.17% | 92.72% | 91.35% |
| ≤2.70* | 44.77% | 96.74% | 13.73 | 0.57 | 60.41% | 94.04% | 91.54% |
| ≤2.82* | 63.59% | 93.21% | 9.36 | 0.39 | 50.98% | 95.84% | 90.24% |
| ≤2.90 | 75.44% | 86.96% | 5.78 | 0.28 | 39.12% | 96.96% | 85.80% |
| ≤2.99* | 84.15% | 80.98% | 4.42 | 0.2 | 32.95% | 97.87% | 81.30% |
| ≤3.10 | 90.77% | 64.13% | 2.53 | 0.14 | 21.95% | 98.43% | 66.79% |
| 30% prevalence | |||||||
| ≤2.50 | 20.38% | 98.91% | 18.75 | 0.8 | 88.93% | 74.35% | 75.35% |
| ≤2.60 | 30.66% | 98.10% | 16.12 | 0.71 | 87.36% | 76.75% | 77.87% |
| ≤2.70* | 44.77% | 96.74% | 13.73 | 0.57 | 85.47% | 80.34% | 81.15% |
| ≤2.82* | 63.59% | 93.21% | 9.36 | 0.39 | 80.05% | 85.66% | 84.32% |
| ≤2.90 | 75.44% | 86.96% | 5.78 | 0.28 | 71.25% | 89.20% | 83.50% |
| ≤2.99* | 84.15% | 80.98% | 4.42 | 0.2 | 65.47% | 92.26% | 81.93% |
| ≤3.10 | 90.77% | 64.13% | 2.53 | 0.14 | 52.03% | 94.19% | 72.12% |
*Cutoff taken from analyzed datasets ≤2.7 mm-from Chennai glaucoma study (CGS) urban, ≤2.82 mm-from CGS rural, and ≤2.99 mm-from clinical dataset; ACD=Anterior chamber depth, PLR=Positive likelihood ratio, NLR=Negative likelihood ratio, PPV=Positive predictive value, NPV=Negative predictive value
At 10% prevalence, the accuracy was highest (91.54%) at the CGS urban ACD cutoff of ≤2.70 mm with a sensitivity of 44.77% and specificity of 96.74%. When analyzing with a lower ACD CV of ≤2.60 the accuracy increased to 91.35%. However, the sensitivity decreased drastically to 30.66%. Using the CGS rural dataset, ACD cutoff of ≤2.82 mm showed a slight drop in accuracy to 90.24% and specificity to 93.21%, but had a better sensitivity of 63.59%.
At 30% prevalence, the highest accuracy (84.32%) in detecting PACD in the clinical dataset was seen with the CGS rural cutoff of ≤2.82 mm, and accuracy dropped both above and below this value.
Considering the overall test metrics and accuracy at 10 and 30% prevalence, the best results were obtained with the CGS rural CVs of ACD ≤ 2.82 mm.
Discussion
We calculated the optimal cutoff of biometric parameters from the CGS dataset overall and separately in the rural and urban arms and in a prospectively recruited clinical cross-section. A good discriminatory capacity was obtained with ACD across all subgroups. This has been noted previously in the literature.[5,6,7,15]
We used the IOLMaster 700 to calculate biometric parameters in the clinical dataset, which differed from those in the CGS, where ultrasonic biometry was used. Though mild changes in the measured parameters may affect both methods, they are used in clinics for biometry for intraocular lens (IOL) power calculation with clinically negligible changes in power.[18,19,20] Even taking the difference in measurements into account, this cannot fully explain the drop in the test metrics to screen angle-closure disease.
The ability of biometric parameters to determine angle-closure disease in a population other than the test population has been reported by a few studies. The sensitivity and specificity of the tests did not vary according to disease prevalence. However, the distribution of disease in the population results in varied specificity and sensitivity.[16,17] Additionally, in populations with varying prevalence, the test accuracy and predictive values can vary. In our clinical dataset, although the maximum sensitivity and specificity were observed at the cutoff chosen from the dataset, the predictive values and accuracy were better for the CGS cutoff at 10 and 30% prevalence. Thus, the biometric parameters cannot be used directly in clinics without prior use to assess the reliability of data from the intended study population.
At a CV of ACD ≤2.815 determined from the CGS rural population, the most optimal results were obtained at 10 and 30% prevalence. The sensitivity in detecting PACD was 63.59%, and the specificity was 93.21% when assessed in the clinical population. The test accuracy in the clinical population was 90.24% at a 10% prevalence and 84.32% at a 30% prevalence. When lowering the ACD CV (i.e., from 2.815 mm to 2.695 mm), a few patients with occludable angles on gonioscopy, but with an AC depth greater than 2.695 mm, were missed. This resulted in a sensitivity drop from 48.95 to 44.77%. However, when a more stringent cutoff of 2.695 mm was used, fewer eyes with ACD ≤ 2.695 mm would have open angles and hence resulting in a slight increase in the specificity from 93.21 to 96.74%.
These tests do not serve as a substitute for a comprehensive clinical examination and opportunistic screening with IOP measurement, gonioscopy, and careful examination of the anterior and posterior segments. In settings where a complete evaluation is not possible due to a lack of availability of resources, as in cataract screening camps where biometry is performed, it may be an opportunity to screen for angle-closure disease. Considering the fact that 8.3 million cataract surgeries were done in the year 2022-2023, and an even larger number of people screened, this represents an untapped potential opportunity for the detection of PACD.[21] Screening of all eyes with biometry and further evaluation of positive patients with gonioscopy may be beneficial in such settings. The similarity in age groups of those screened for cataract and those with glaucoma, the challenge in operating on PACD eyes compared to those with open angles, and the opportunity in using available technology to detect a largely undetected disease, which can cause irreversible blindness, biometric CVs may aid in the detection of a subset of the population that needs further attention. Additionally, biometry can also be used for screening in other settings like DR screening camps, vision screening camps, and other settings where routine gonioscopy is not practiced.
As the predictive values of the screening test vary with prevalence, and the distribution of disease affects the sensitivity and specificity of the tests, these tests cannot be used interchangeably in different populations. Further studies in specific populations of intended use are warranted to determine appropriate CVs according to race and ethnicity, and the use of a combination of tests may help in improving test metrics.
Conclusion
ACD CVs determined from CGS were found to have an accuracy of 92% and 84%, at 10% and 30% prevalence, respectively, to determine the presence of angle-closure disease in a clinical population. These parameters cannot be used interchangeably in different populations but may be used with due caution to detect patients who require further evaluation in resource-constrained areas or where biometry is available but a comprehensive evaluation is not feasible.
Author’s declaration statement of patient consent
Written informed consent was obtained from all patients in the clinical dataset prior to inclusion in the study. A separate consent was not obtained from CGS patients for use of the CGS database in this study, as prior informed consent was taken when the patients were included in the study.
Author contributions
RP: Conceptualization, design, definition of intellectual content, literature search, clinical studies, data acquisition, statistical analysis, manuscript preparation, manuscript editing
RS: Data acquisition, clinical studies, statistical analysis, manuscript review
MB: Reviewing and editing the manuscript
RA: Data acquisition, manuscript review
RG: Conceptualization, design, definition of intellectual content, statistical analysis, supervision, reviewing, and editing the manuscript
LV: Reviewing and editing the manuscript
Conflicts of interest
There are no conflicts of interest.
Funding Statement
Nil.
References
- 1.Tham YC, Li X, Wong TY, Quigley HA, Aung T, Cheng CY. Global prevalence of glaucoma and projections of glaucoma burden through 2040: A systematic review and meta-analysis. Ophthalmology. 2014;121:2081–90. doi: 10.1016/j.ophtha.2014.05.013. [DOI] [PubMed] [Google Scholar]
- 2.Vijaya L, George R, Arvind H, Baskaran M, Ve Ramesh S, Raju P, et al. Prevalence of primary angle-closure disease in an urban south Indian population and comparison with a rural population. The Chennai Glaucoma Study. Ophthalmology. 2008;115:655–60. doi: 10.1016/j.ophtha.2007.05.034. [DOI] [PubMed] [Google Scholar]
- 3.George R, Ve RS, Vijaya L. Glaucoma in India: Estimated burden of disease. J Glaucoma. 2010;19:391–7. doi: 10.1097/IJG.0b013e3181c4ac5b. [DOI] [PubMed] [Google Scholar]
- 4.George R, Panda S, Vijaya L. Blindness in glaucoma: Primary open-angle glaucoma versus primary angle-closure glaucoma—A meta-analysis. Eye (Lond) 2022;36:2099–115. doi: 10.1038/s41433-021-01802-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Nongpiur ME, Khor CC, Cheng CY, Husain R, Boey PY, Chew A, et al. Integration of genetic and biometric risk factors for detection of primary angle closure glaucoma. Am J Ophthalmol. 2019;208:160–5. doi: 10.1016/j.ajo.2019.07.022. [DOI] [PubMed] [Google Scholar]
- 6.Savur S, Kaup S, Dinesh A, Shivalli S, Kondal D. Can ultrasonic biometric indices with optimal cut-offs be a potential screening tool for primary angle closure disease? A case-control study. Eye (Lond) 2023;37:1284–9. doi: 10.1038/s41433-022-02118-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Chen YY, Chen YY, Sheu SJ, Chou P. The biometric study in different stages of primary angle-closure glaucoma. Eye (Lond) 2013;27:1070–6. doi: 10.1038/eye.2013.127. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Xu BY, Friedman DS, Foster PJ, Jiang Y, Porporato N, Pardeshi AA, et al. Ocular biometric risk factors for progression of primary angle closure disease: The Zhongshan angle closure prevention trial. Ophthalmology. 2022;129:267–75. doi: 10.1016/j.ophtha.2021.10.003. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Gedde SJ, Chen PP, Muir KW, Vinod K, Lind JT, Wright MM, et al. Primary angle-closure disease preferred practice pattern®. Ophthalmology. 2021;128:30–70. doi: 10.1016/j.ophtha.2020.10.021. [DOI] [PubMed] [Google Scholar]
- 10.Li S, Shao M, Wan Y, Tang B, Sun X, Cao W. Relationship between ocular biometry and severity of primary angle-closure glaucoma: Relevance for predictive, preventive, and personalized medicine. EPMA J. 2019;10:261–71. doi: 10.1007/s13167-019-00174-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Yong KL, Gong T, Nongpiur ME, How AC, Lee HK, Cheng L, et al. Myopia in Asian subjects with primary angle closure: Implications for glaucoma trends in East Asia. Ophthalmology. 2014;121:1566–71. doi: 10.1016/j.ophtha.2014.02.006. [DOI] [PubMed] [Google Scholar]
- 12.Arvind H, Paul PG, Raju P, Baskaran M, George R, Balu S, et al. Methods and design of the Chennai Glaucoma study. Ophthalmic Epidemiol. 2003;10:337–48. doi: 10.1076/opep.10.5.337.17320. [DOI] [PubMed] [Google Scholar]
- 13.Vijaya L, George R, Arvind H, Baskaran M, Paul PG, Ramesh SV, et al. Prevalence of angle-closure disease in a rural southern Indian population. Arch Ophthalmol. 2006;124:403–9. doi: 10.1001/archopht.124.3.403. [DOI] [PubMed] [Google Scholar]
- 14.Perkins NJ, Schisterman EF. The Youden Index and the optimal cut-point corrected for measurement error. Biom J. 2005;47:428–41. doi: 10.1002/bimj.200410133. [DOI] [PubMed] [Google Scholar]
- 15. Available from: https://www.medcalc.org/calc/diagnostic_test.php . [Last accessed 2023 Sep 21]
- 16.Choudhari NS, George R, Asokan R, Khanna R, Vijaya L, Garudadri CS. Combination of simple diagnostic tests to detect primary angle closure disease in a resource-constrained region. Ophthalmic Epidemiol. 2019;26:430–8. doi: 10.1080/09286586.2019.1650380. [DOI] [PubMed] [Google Scholar]
- 17.Parikh R, Mathai A, Parikh S, Chandra Sekhar G, Thomas R. Understanding and using sensitivity, specificity and predictive values. Indian J Ophthalmol. 2008;56:45–50. doi: 10.4103/0301-4738.37595. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Lavanya R, Teo L, Friedman DS, Aung HT, Baskaran M, Gao H, et al. Comparison of anterior chamber depth measurements using the IOLMaster, scanning peripheral anterior chamber depth analyser, and anterior segment optical coherence tomography. Br J Ophthalmol. 2007;91:1023–6. doi: 10.1136/bjo.2006.113761. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Drexler W, Findl O, Menapace R, Rainer G, Vass C, Hitzenberger CK, et al. Partial coherence interferometry: A novel approach to biometry in cataract surgery. Am J Ophthalmol. 1998;126:524–34. doi: 10.1016/s0002-9394(98)00113-5. [DOI] [PubMed] [Google Scholar]
- 20.Kolega MŠ, Kovačević S, Čanović S, Pavičić AD, Bašić JK. Comparison of IOL--master and ultrasound biometry in preoperative intra ocular lens (IOL) power calculation. Coll Antropol. 2015;39:233–5. [PubMed] [Google Scholar]
- 21. Available from: https://pib.gov.in/PressReleasePage.aspx?PRID=1944598#:~:text=For%20the%20FY%2D2022%2D23,blindness%20to%200.25%25%20by%202025 . [Last accessed 2025 Mar 19]