Abstract
Background
Central Serous Chorioretinopathy(CSC) is a chorioretinal disorder, predominantly affects young to middle-aged adults, resulting serious vision disorder. This study aimed to develop a Bayesian network model to predict the key factors influencing the early therapeutic efficacy of 577 nm-SML in patients with CSC.
Methods
A total of 159 patients (159 eyes) diagnosed with CSC and treated with 577 nm-SML at the First Affiliated Hospital of Guangxi Medical University from January 2019 to November 2023 were retrospectively analyzed. Baseline data including age, sex, eye side, disease course, and past medical history were collected. Ophthalmic imaging detects central macular thickness (CMT), macular foveal volume (MFV within 1mm, 3mm, 6mm diameter), height and area of subretinal fluid (SRF), structural changes in the outer retinal layers (ORL), type and area of leakage lesions, etc. Influential variables significantly associated with 577nm-SML efficacy were screened using LASSO regression, then construct a Bayesian network model to predict factors that significantly affect the therapeutic effect.
Results
LASSO regression identified 19 significant variables related to treatment outcomes from the 40 possible risk factors included, including disease duration, sex, eye Side, smoking, hormone, macular foveal volumes (3 mm and 6 mm diameters), and the height and area of SRF, ORL integrity, typical PED, location of PED, location of RPE bulging, heterogeneity of NPL, HF of ORL, HF of choroid, leakage type, leakage location, leakage correlate with OCT. The Bayesian network presents complex interactions among these factors, shows that patients with smaller macular foveal volumes (within 3 mm diameter), shorter disease duration, and focal leakage exhibited superior responses to 577nm-SML treatment.
Conclusion
The therapeutic response to 577nm-SML in CSC is influenced by multifactorial dynamics. Bayesian network can well present the complex network relationship between the therapeutic effect of 577nm-SML and related influencing factors, and identify potential risk factors that affect early efficacy.
Keywords: Central serous chorioretinopathy, 577nm-SML, Bayesian network, LASSO regression, Network analysis
Introduction
Central Serous Chorioretinopathy (CSC ) is a chorioretinal disorder characterized by serous retinal detachment, with or without retinal pigment epithelial (RPE) detachment, leading to symptoms such as visual distortion, decreased visual acuity, and visual field defects [1, 2]. Although the exact mechanism remains unclear, increased choroidal vascular permeability and RPE dysfunction are recognized as the key contributors [3, 4]. Several risk factors have been implicated, including type A personality, psychological stress, elevated corticosteroid levels, hypertension, coronary artery disease, psychotropic drug use, allergic respiratory conditions, peptic ulcers, smoking, coagulation abnormalities, platelet aggregation, and hyperopia [5, 6]. CSC predominantly affects middle-aged men. It reported male to femal radio was 6:1 [7].
As to the treatment, there are many different options, including laser photocoagulation, photodynamic therapy (PDT), oral mineralocorticoid receptor antagonists, and anti-VEGF (vascular endothelial growth factor) injections. The conventional retinal laser photocoagulation uses the thermal effect generated by the short-term high temperature to coagulate the lesion-retinal tissue, and then directly seal leakage points or abnormal blood vessels [8]. However, it is easy to damage the surrounding normal retinal tissue and form retinal scars. Subthreshold Micropulse Laser (SML) exerts its therapeutic effect by inducing controlled thermal stress in RPE cells to inhibit apoptosis, downregulate inflammatory mediators, reduce choroidal permeability, and seal RPE leakage sites, thereby facilitating subretinal fluid (SRF) absorption [9–11]; it can directly act on the fovea or parafovea, and has little damage to the structure or function of the retinal-choroid, making it a safe and effective treatment method [12]. The 577 nm yellow laser wavelength can be effectively absorbed by the melanin and hemoglobin in the retinal pigment epithelium, and almost not absorbed by lutein in the macular area, so it has low damage to the macular visual sensitive area. Therefore, 577 nm-SML is widely used in clinical, and many researches have also confirmed its safety and effectiveness [11, 13]. Howere, clinical observations have found that the therapeutic effects vary significantly due to individual patient differences, which may be related to the fact that the pathogenesis of CSC is complex and affected by multiple factors.
Optical coherence tomography (OCT) has become an essential imaging modality for CSC diagnosis and follow-up, enabling visualization of RPE structural defects and SRF dynamics, and has emerged as a predictive tool for treatment outcomes. Bayesian network is a graphical model used to represent and reason about uncertain knowledge. It reflects the potential relationship between variables by constructing directed acyclic graphs and conditional probability tables. It can effectively handle complex nonlinear relationships and multi-factor interactions between variables. It integrates incomplete data through probabilistic reasoning and accurately depicts the correlation between risk factors. It has received more attention in the research and analysis of disease causal effects and is also very useful for prediction and diagnosis [14, 15]. Compared with traditional statistical methods, it can intuitively reveal the causal logic between variables with its directed graph structure, clearly track the risk transmission paths, and provide a reliable decision-making basis for risk prediction and intervention. Then our research aims to systematically analyze the early treatment effect of 577 nm-SML and the relationship between its influencing factors, and use the Bayesian network model to predict the risk factors that affect the early treatment effect, so as to provide early therapeutic intervention to patients.
Materials and methods
Study population
A total of 159 patients from the First Affiliated Hospital of Guangxi Medical University, diagnosed with CSC and deemed to require 577 nm-SML therapy between January 2019 and November 2023, were included for data collection and analysis. The exclusion criteria : (1) any history within the preceding 6 months of intravitreal anti-VEGF injection, photodynamic therapy (PDT), or other intraocular surgical interventions; (2) presence of concomitant ocular disorders or retinal pathologies, such as choroidal neovascularization (CNV), polypoidal choroidal vasculopathy (PCV), diabetic macular edema (DME), or age-related macular degeneration (AMD).
Clinical examination
Comprehensive baseline demographic and clinical information was collected from all participants, including age, sex, eye side, disease course, medical history (hypertension, diabetes, gastrointestinal disorders), smoking and alcohol, mental health status, and prior or current systemic corticosteroid use. Baseline and 2-week post-treatment evaluations were conducted, comprising slit-lamp biomicroscopy, dilated fundus examination, intraocular pressure measurement, and OCT imaging. Fundus fluorescein angiography (FFA) and indocyanine green angiography (ICGA) were performed at baseline to detect leakage, assess the presence of fluorescein leakage at the RPE, characterize leakage type, extent, and anatomical location, and determine any signs of RPE atrophy or hyperplasia. OCT measurements included CMT, MFV (1 mm, 3 mm, and 6 mm diameters), height and area of SRF, quantification of SRF, assessment of serous retinal detachment morphology, and evaluation of neuroepithelial layer structural alterations.
577 nm-subthreshold micropulse laser
Pupil dilation was achieved using compound tropicamide ophthalmic solution, followed by superficial anesthesia with 0.4% oxybuprocaine hydrochloride eye drops. The procedure was performed by the same experienced ophthalmic surgeon. Initial titration was conducted in single-spot micropulse mode using a yellow laser system, the wavelength is fixed at 577 nm, spot diameter 200 μm; exposure time 200 ms, duty cycle 5%. Power titration was performed in single-point micropulse mode, starting with 700 mW in the retinal area outside the vascular arch, and then gradually increasing the power until retinal spots appeared to determine the threshold energy. Finally, half the titration power was applied to achieve full coverage photocoagulation of the SRD area. All patients initially use the same treatment parameters, and the parameters will be adjusted according to different patients.
This study mainly observes the efficacy 2 weeks after treatment. With reference to the study of Altinel et al., complete SRF resolution or a decrease of more than or equal to 10% from the baseline is considered effective, and an increase or reduction of less than 10% in SRF was classified as ineffective [16].
Statistical analysis
All statistical analyses were performed using IBM SPSS Statistics version 20.0, with a p-value < 0.05 considered statistically significant. Continuous variables were expressed as mean ± standard deviation (SD) or as median. Structural learning for the Bayesian network was conducted using R software version 4.1.3, and LASSO regression was implemented via the glmnet package in R 4.1.3 to identify influential variables. Multivariable logistic regression was employed to assess risk factors influencing 577 nm-SML efficacy, using treatment outcome as the dependent variable and LASSO-selected parameters as independent predictors.
Bayesian network and LASSO regression
The Bayesian network, also termed a belief network or directed acyclic graphical model, is a classic machine learning algorithm capable of integrating historical, incomplete, and uncertain datasets [17]. By applying probabilistic inference, it models relationships among variables, enabling identification of the most probable diagnoses, selection of the most informative diagnostic tests, and sensitivity analysis of model conclusions relative to variable changes. Within the network, variables are represented as nodes in a directed acyclic graph (DAG), with edges denoting causal relationships quantified by probability values. This approach facilitates an intuitive depiction of inter-variable dynamics and complex associations between independent and dependent variables, proving highly valuable for elucidating disease-related causal pathways and uncovering novel or latent risk factors.
The LASSO regression technique applies a penalty to regression coefficients proportional to their absolute magnitude, thereby shrinking less informative coefficients toward zero. This facilitates selection of independent variables exerting greater influence on the dependent variable and estimation of corresponding regression coefficients [18]. Its capacity to process extensive sets of potential predictors allows it to identify those most relevant to disease occurrence and progression, making it particularly suitable for screening CSC risk factors and constructing predictive models [19].
Results
Basic information
The 577 nm-SML therapy was administered to 159 eyes from 159 patients (133 males, 26 females), with ages ranging from 15 to 77 years (Tables 1, 2 and 3).
Table 1.
Baseline characteristics of patients
| Total population(n = 159) | Efficacy | χ² | P | ||
|---|---|---|---|---|---|
| ineffective (n = 50) | effective (n = 109) | ||||
| Age | - | 0.064 | |||
| < 45 | 56 (35.22) | 14 (28.00) | 42 (38.53) | ||
| 45–59 | 101 (63.52) | 34 (68.00) | 67 (61.47) | ||
| ≥ 60 | 2 (1.26) | 2 (4.00) | 0 (0.00) | ||
| Sex | χ²=3.12 | 0.077 | |||
| Male | 133 (83.65) | 38 (76.00) | 95 (87.16) | ||
| Female | 26 (16.35) | 12 (24.00) | 14 (12.84) | ||
| Disease course | χ²=6.32 | 0.012 | |||
| < 6 m | 137 (86.16) | 38 (76.00) | 99 (90.83) | ||
| ≥ 6 m | 22 (13.84) | 12 (24.00) | 10 (9.17) | ||
| Eye Side | χ²=1.45 | 0.228 | |||
| Right eye | 81 (50.94) | 29 (58.00) | 52 (47.71) | ||
| Left eye | 78 (49.06) | 21 (42.00) | 57 (52.29) | ||
| Insomnia | χ²=0.03 | 0.861 | |||
| Yes | 113 (71.07) | 36 (72.00) | 77 (70.64) | ||
| No | 46 (28.93) | 14 (28.00) | 32 (29.36) | ||
| Alcohol | χ²=0.61 | 0.435 | |||
| Yes | 38 (23.90) | 10 (20.00) | 28 (25.69) | ||
| No | 121 (76.10) | 40 (80.00) | 81 (74.31) | ||
| Smoking | χ²=1.24 | 0.266 | |||
| Yes | 51 (32.08) | 13 (26.00) | 38 (34.86) | ||
| No | 108 (67.92) | 37 (74.00) | 71 (65.14) | ||
| Hormone | χ²=7.40 | 0.007 | |||
| Yes | 11 (6.92) | 8 (16.00) | 3 (2.75) | ||
| No | 148 (93.08) | 42 (84.00) | 106 (97.25) | ||
| Hypertension | χ²=0.06 | 0.804 | |||
| Yes | 7 (4.40) | 3 (6.00) | 4 (3.67) | ||
| No | 152 (95.60) | 47 (94.00) | 105 (96.33) | ||
| Diabetes | χ²=2.09 | 0.148 | |||
| Yes | 6 (3.77) | 4 (8.00) | 2 (1.83) | ||
| No | 153 (96.23) | 46 (92.00) | 107 (98.17) | ||
| Gastropathy | χ²=0.00 | 1 | |||
| Yes | 8 (5.03) | 3 (6.00) | 5 (4.59) | ||
| No | 151 (94.97) | 47 (94.00) | 104 (95.41) | ||
Note: Variable data are shown as n (%), - Fisher’s exact test, no χ² value
Table 2.
Baseline OCT characteristics
| Total population(n = 159) | Efficacy | χ² | P | ||
|---|---|---|---|---|---|
| ineffective (n = 50) | effective (n = 109) | ||||
| CMT(µm) | χ²=13.13 | 0.04 | |||
| ≤ 377 | 35 (22.01) | 19 (38.00) | 16 (14.68) | ||
| 378–466 | 41 (25.79) | 13 (26.00) | 28 (25.69) | ||
| 467–578 | 41 (25.79) | 11 (22.00) | 30 (27.52) | ||
| ≥ 579 | 42 (26.42) | 7 (14.00) | 35 (32.11) | ||
| MFV-1 mm(µm) | χ²=14.56 | 0.002 | |||
| ≤ 0.29 | 36 (22.64) | 19 (38.00) | 17 (15.60) | ||
| 0.3–0.37 | 41 (25.79) | 14 (28.00) | 27 (24.77) | ||
| 0.38–0.45 | 42 (26.42) | l2 (24.00) | 30 (27.52) | ||
| ≥ 0.46 | 40 (25.16) | 5 (10.00) | 35 (32.11) | ||
| MFV 3 mm(µm) | χ²=11.49 | 0.003 | |||
| ≤ 2.67 | 35 (22.01) | 18 (36.00) | 17 (15.60) | ||
| 2.68–3.05 | 41 (25.79) | 15 (30.00) | 26 (23.85) | ||
| 3.06–3.61 | 42 (26.42) | 12 (24.00) | 30 (27.52) | ||
| ≥ 3.62 | 41 (25.79) | 5 (10.00) | 36 (33.03) | ||
| MFV 6 mm(µm) | χ²=11.88 | 0.008 | |||
| ≤ 9.15 | 35 (22.01) | 16 (32.00) | 19 (17.43) | ||
| 9.16–9.82 | 42 (26.42) | 16 (32.00) | 26 (23.85) | ||
| 9.83–11.11 | 44 (27.67) | 14 (28.00) | 30 (27.52) | ||
| ≥ 11.12 | 38 (23.90) | 4 (8.00) | 34 (31.19) | ||
| SRF height(µm) | χ²=14.40 | 0.002 | |||
| ≤ 153 | 33 (20.75) | 18 (36.00) | 15 (l3.76) | ||
| 154–255 | 42 (26.42) | 15 (30.00) | 27 (24.77) | ||
| 256–348 | 39 (24.53) | 10 (20.00) | 29 (26.61) | ||
| ≥ 349 | 45 (28.30) | 7 (14.00) | 38 (34.86) | ||
| SRF area | χ²=8.45 | 0.015 | |||
| < 2PD | 37 (23.27) | 17 (34.00) | 20 (18.35) | ||
| 2-4PD | 63 (39.62) | 22 (44.00) | 41 (37.61) | ||
| > 4PD | 59 (37.11) | 11(22.00) | 48 (44.04) | ||
| SRD shape | - | 0.537 | |||
| Circular shape | 119 (74.84) | 35 (70.00) | 84 (77.06) | ||
| Irregular shape | 39 (24.53) | 15 (30.00) | 24 (22.02) | ||
| Structural abnormalities beneath SRD | χ²=7.40 | 0.007 | |||
| None | 32 (20.13) | 8 (16.00) | 24 (22.02) | ||
| HF | 51 (32.08) | 17 (34.00) | 34 (31.19) | ||
| HF fibrin exudation | 45 (28.30) | 16 (32.00) | 29 (26.61) | ||
| ONRL depression | 4 (2.52) | 1 (2.00) | 3 (2.75) | ||
| HF & HF fibrin exudation | 17 (10.69) | 7 (14.00) | 10 (9.17) | ||
| HF & ONRL depression | 3 (1.89) | 1 (2.00) | 2 (1.83) | ||
| HF fibrin exudation & ONRL depression | 5 (3.14) | 0 (0.00) | 5 (4.59) | ||
| HF & HF fibrin exudation & ONRL depression | 2 (1.26) | 0 (0.00) | 2 (1.83) | ||
| ORL integrity | χ²=0.30 | 0.583 | |||
| Yes | 153 (96.23) | 47 (94.00) | 106 (97.25) | ||
| No | 6 (3.77) | 3 (6.00) | 3 (2.75) | ||
| Morphology of POSL | χ2 = 3.80 | 0.284 | |||
| Smooth | 19 (11.95) | 4 (8.00) | 15 (13.76) | ||
| Thickening | 37 (23.27) | 16 (32.00) | 21 (19.27) | ||
| Granular | 72 (45.28) | 20 (40.00) | 52 (47.71) | ||
| Thickening & granular | 31 (19.50) | 10 (20.00) | 21 (19.27) | ||
| Typical PED | χ²=0.77 | 0.38 | |||
| Yes | 32 (20.13) | 8 (16.00) | 24 (22.02) | ||
| No | 127 (79.87) | 42 (84.00) | 85 (77.98) | ||
| Location of PED | - | 0.018 | |||
| None | 126 (79.25) | 41 (82.00) | 85 (77.98) | ||
| Within SRD | 26 (16.35) | 4 (8.00) | 22 (20.18) | ||
| Outside SRD | 5 (3.14) | 3 (6.00) | 2 (1.83) | ||
| Inside and outside SRD | 2 (1.26) | 2 (4.00) | 0 (0.00) | ||
| RPE bulging | χ²=0.25 | 0.62 | |||
| Yes | 130 (81.76) | 42 (84.00) | 88 (80.73) | ||
| No | 29 (18.24) | 8 (16.00) | 21 (19.27) | ||
| Location of RPE bulging | - | 0.331 | |||
| None | 29 (18.24) | 8 (16.00) | 21 (19.27) | ||
| Within SRD | 124 (77.99) | 39 (78.00) | 85 (77.98) | ||
| Outside SRD | 4 (2.52) | 1 (2.00) | 3 (2.75) | ||
| Multifocal | 1 (0.63) | 1 (2.00) | 0 (0.00) | ||
| Inside and outside SRD | 1 (0.63) | 1 (2.00) | 0 (0.00) | ||
| RPE atrophy | χ²=0.01 | 0.910 | |||
| Yes | 23 (14.47) | 7 (14.00) | l6 (14.68) | ||
| No | 136 (85.53) | 43 (86.00) | 93 (85.32) | ||
| RPE rough | χ²=0.94 | 0.332 | |||
| Yes | 116 (72.96) | 39 (78.00) | 77 (70.64) | ||
| No | 43 (27.04) | 11(22.00) | 32 (29.36) | ||
| Heterogeneity of NPL | - | 0.051 | |||
| None | 6 (3.77) | 1 (2.00) | 5 (4.59) | ||
| ONL atrophy | 20 (12.58) | 4 (8.00) | 16 (14.68) | ||
| HF | 116 (72.96) | 42 (84.00) | 74 (67.89) | ||
| ONL atrophy & HF | 13 (8.18) | 1 (2.00) | 12 (11.01) | ||
| HF & microcyst | 2 (1.26) | 1 (2.00) | 0 (0.00) | ||
| Edema | 1(0.63) | 1 (2.00) | 0 (0.00) | ||
| HF & edema | 1 (0.63) | 0 (0.00) | 2 (1.83) | ||
| HF of ORL | χ²=0.80 | 0.371 | |||
| Yes | 141 (88.68) | 46 (92.00) | 95 (87.16) | ||
| No | 18 (11.32) | 4 (8.00) | 14 (12.84) | ||
| HF of choroid | χ²=5.61 | 0.018 | |||
| Yes | 126 (79.25) | 34 (68.00) | 92 (84.40) | ||
| No | 33 (20.75) | 16 (32.00) | 17 (15.60) | ||
| HF of choroidal vascular | χ²=0.92 | 0.338 | |||
| Yes | 110 (69.l8) | 32 (64.00) | 78 (71.56) | ||
| No | 49 (30.82) | 18 (36.00) | 31 (28.44) | ||
Note: Variable data are shown as n (%), - Fisher’s exact test, no χ² value
Table 3.
Baseline FFA characteristics
| Total population(n = 159) | Efficacy | χ² value | P | ||
|---|---|---|---|---|---|
| ineffective (n = 50) | effective (n = 109) | ||||
| Leakage type | - | 0.011 | |||
| Multiple lesions | 18 (11.32) | 10 (20.00) | 8 (7.34) | ||
| Inkblot | 110 (69.18) | 36 (72.00) | 74 (67.89) | ||
| Smoke-like lesion | 30 (18.87) | 4 (8.00) | 26 (23.85) | ||
| Multiple & Smoke-like | 1(0.63) | 0 (0.00) | 1 (0.92) | ||
| Leakage area | χ²=1.17 | 0.28 | |||
| ≤l/3PD | 111 (69.81) | 32 (64.00) | 79 (72.48) | ||
| > l/3PD | 48 (30.19) | 18 (36.00) | 30 (27.52) | ||
| Leakage location | χ²=6.33 | 0.176 | |||
| < 0.5 mm | 20 (12.58) | 2 (4.00) | 18 (16.51) | ||
| 0.5–1 mm | 37 (23.27) | 13 (26.00) | 24 (22.02) | ||
| l-l.5 mm | 30 (18.87) | 8 (16.00) | 22 (20.18) | ||
| Outside l.5 mm | 65 (40.88) | 25 (50.00) | 40 (36.70) | ||
| Multiple | 7 (4.40) | 2 (4.00) | 5 (4.59) | ||
| Number of leakage | - | 0.513 | |||
| 1 | 130 (81.76) | 38 (76.00) | 92 (84.40) | ||
| 2 | 6 (3.77) | 2 (4.00) | 4 (3.67) | ||
| 3 | 4 (2.52) | 2 (4.00) | 2 (1.83) | ||
| More than 4 | 19 (11.95) | 8 (16.00) | 11 (10.09) | ||
| RPE leakage | χ²=0.00 | 0.958 | |||
| Yes | 130 (81.76) | 41 (82.00) | 89 (81.65) | ||
| No | 29 (18.24) | 9 (18.00) | 20 (18.35) | ||
| Delineate the SRD area | χ²=3.40 | 0.065 | |||
| Yes | 33 (20.75) | 6 (12.00) | 27 (24.77) | ||
| No | 126 (79.25) | 44 (88.00) | 82 (75.23) | ||
| RPE atrophy | χ²=1.41 | 0.236 | |||
| Yes | 78 (49.06) | 28 (56.00) | 50 (45.87) | ||
| No | 81 (50.94) | 22 (44.00) | 59 (54.13) | ||
| RPE prolification | χ²=0.17 | 0.682 | |||
| Yes | 35 (22.01) | 12 (24.00) | 23 (21.10) | ||
| No | 124 (77.99) | 38 (76.00) | 86 (78.90) | ||
| Leakage correlate with OCT | χ²=0.51 | 0.477 | |||
| Yes | 144 (90.57) | 47 (94.00) | 97 (88.99) | ||
| No | 15 (9.43) | 3 (6.00) | 12 (11.01) | ||
Note: Variable data are shown as n (%), - Fisher’s exact test, no χ² value
LASSO regression analysis screens the influencing factors
Using treatment efficacy as the dependent variable, 40 potential risk factors were incorporated as independent variables in LASSO regression screening. The analysis yielded the following non-zero coefficient predictors: sex, disease course, eye side, smoking history, corticosteroid use, MFV (3 mm and 6 mm diameters), the height and area of SRF, ORL integrity, location of typical PED, RPE bulging, HF of ORL and choroidal, and leakage location, lesion type, leakage correlate with OCT (Table 4; Fig. 1).
Table 4.
Coefficients and λmin of LASSO
| variable | coefficient | λmin |
|---|---|---|
| Sex | -0.03463 | 0.02352 |
| Disease course | -0.14292 | |
| Eye side | 0.07036 | |
| Smoking | -0.01957 | |
| Hormone | 0.25341 | |
| MFV-3 mm(µm) | 0.01299 | |
| MFV 6 mm(µm) | 0.03715 | |
| SRF height(µm) | 0.06999 | |
| SRF area | 0.00616 | |
| ORL integrity | -0.00111 | |
| Typical PED | -0.01966 | |
| Location of PED | -0.01971 | |
| Location of RPE bulging | -0.0198 | |
| Heterogeneity of NPL | -0.04466 | |
| HF of ORL | 0.13886 | |
| HF of choroid | -0.21733 | |
| Lesion type | 0.05725 | |
| Leakage location | -0.04059 | |
| Leakage correlate with OCT | 0.08679 |
Fig. 1.
Variable selection for CSC efficacy prediction model based on LASSO regression. Note: Left picture shows the distribution of LASSO coefficients; Right illustrates optimal parameter (λ) selection through 10-fold cross-validation, with the curve plotting MSE against logλ and vertical dashed lines marking the minimum and 1-standard-error criteria
Multivariate logistic regression
The 19 screened variables were included in multifactor logistic regression, and the results are shown in Table 5 (only the results of variables that are significantly related to the efficacy are listed).
Table 5.
Multivariate logistic regression on efficacy of CSC
| Variable | β | SE | Wald χ2 | P | OR (95% CI) |
|---|---|---|---|---|---|
| Disease course | |||||
| < 6 m | 1 | ||||
| ≥ 6 m | -2.19 | 0.965 | -2.271 | 0.023 | 0.112 (0.014, 0.675) |
| Eye Side | |||||
| Right eye | 1 | ||||
| Left eye | 2.272 | 0.819 | 2.776 | 0.006 | 9.701 (2.203, 57.22) |
| SRF height(µm) | |||||
| ≤ 153 | 1 | ||||
| 154–255 | 1.467 | 0.95 | 1.544 | 0.123 | 4.335 (0.722, 32.509) |
| 256–348 | 1.483 | 1.184 | 1.253 | 0.21 | 4.406 (0.445, 50.023) |
| ≥ 349 | 3.287 | 1.659 | 1.981 | 0.048 | 26.762 (1.191, 936.922) |
| HF of ORL | |||||
| Yes | 1 | ||||
| No | 3.389 | 1.383 | 2.452 | 0.014 | 29.647 (2.401, 607.102) |
| Leakage location | |||||
| < 0.5 mm | 1 | ||||
| 0.5–1 mm | -4.345 | 1.57 | -2.768 | 0.006 | 0.013 (0, 0.199) |
| l-l.5 mm | -4.548 | 1.666 | -2.73 | 0.006 | 0.011 (0, 0.194) |
| Outside l.5 mm | -4.405 | 1.711 | -2.574 | 0.01 | 0.012 (0, 0.235) |
| Multiple | -2.496 | 3.054 | -0.817 | 0.414 | 0.082 (0, 42.107) |
| Leakage correlate with OCT | |||||
| Yes | 1 | ||||
| No | 4.103 | 1.764 | 2.326 | 0.02 | 60.543(2.773, 3505.929) |
Bayesian network construction
By synthesizing diverse clinical and imaging parameters, the Bayesian network model enables dynamic prognosis assessment, inferring probabilities of unknown nodes from known node states and thereby predicting complex system outcomes. In this study, a Bayesian network comprising 19 nodes and 23 directed edges representing CSC efficacy-associated variables was built based on LASSO-selected predictors, with calculated conditional probabilities for each node (Fig. 2).The model classification accuracy is 72.5% and the AUC value is 0.793, indicating that the model prediction performance is good.
Fig. 2.
Bayesian network model for CSC. Note: HF = Hyperreflective Foci; ORL = Outer Retinal Layers; PED = Pigment Epithelium Detachment; ONL = Outer Nuclear Layer; SRD = Serous Retinal Detachment
Bayesian network model diagnostic inference
The Bayesian network model can infer the probability of unknown nodes based on the status of known nodes, and determine the different curative effects of CSC. For men, MFV (within 3 mm) ≤ 2.67 μm, and the effective rate is 76.5%; For men, MFV (within 3 mm) 2.68–3.05 μm, and the effective rate is 57.1%; for men, MFV (within 3 mm) 3.06–3.61 μm, the effective rate is 28.9%; for men, MFV (within 3 mm) 3.06–3.61 μm, the effective rate is 14.3% (Table 6).
Table 6.
Conditional probability table of different efficacy nodes for CSC
| Sex | MFV-3 mm(µm) | Effective |
|---|---|---|
| Male | ≤ 2.67 | 76.5 |
| Male | 2.68–3.05 | 57.1 |
| Male | 3.06–3.61 | 28.9 |
| Male | ≥ 3.62 | 14.3 |
| Female | ≤ 2.67 | 80.0 |
| Female | 2.68–3.05 | 40.0 |
| Female | 3.06–3.61 | 62.5 |
| Female | ≥ 3.62 | 54.5 |
Discussion
In this study, a Bayesian network-based predictive framework was utilized to elucidate factors impacting the therapeutic outcomes of 577 nm-SML, thereby mapping the intricate interplay between efficacy and its determinants. Epidemiological surveys indicate that CSC predominantly affects young to middle-aged adults, with typical onset between ages 20 and 50 [7]. While literature suggests that advanced age correlates with prolonged CSC episodes and increased SRF persistence risk [20, 21], age was not incorporated as a predictive variable in our model. In our study, there are obvious differences between gender and efficacy. Male patients have a higher overall incidence rate, but female patients may recover faster and have lower recurrence rates after treatment. This difference is mainly related to the gender specificity of triggers, hormone levels and lifestyle habits [22–24]. Evidence indicates that female patients tend to display fewer HF on SD-OCT and less pronounced RPE structural alterations, whereas males exhibit greater RPE disruption, with FFA and ICGA revealing more extensive leakage patterns [25, 26].
The duration of the disease is an important time factor that affects the therapeutic effect of 577 nm-SML [20, 27]. Acute CSC (disease duration < 6 months) usually responds better to 577 nm-SML, while chronic cases have higher failure rates [27]. This difference may be related to photoreceptor damage and RPE functional degradation caused by long-term subretinal fluid accumulation. Research shows that patients with a disease duration of more than 12 months experience significant decreases in treatment effectiveness [27]. Extended disease duration has been linked to irreversible injury to both RPE and photoreceptor cells, with longer duration associated with higher rates of treatment failure and suboptimal post-treatment anatomical recovery [28, 29]. Prolonged CSC also correlates with increased macular fovea volume [30]. Clinically, volumetric assessment of the macular fovea offers a metric for disease severity, progression monitoring, and post-intervention efficacy evaluation.
Our findings indicate that macular foveal volume within a 3 mm diameter is directly associated with CSC treatment efficacy. Previous studies have demonstrated that macular foveal volume strongly correlated with retinal sensitivity, and that larger baseline volume may predict early functional recovery; larger volumes reflect more severe SRF accumulation and greater RPE impairment, thereby compromising treatment outcomes [31]. Notably, while macular foveal volume within a 6 mm diameter was not directly linked to efficacy, our Bayesian network analysis revealed indirect influence through MFV area and disease duration, underscoring its complex role in CSC pathophysiology.
Advanced multimodal ophthalmic imaging plays a pivotal role in CSC prognosis prediction. FFA enables detailed visualization of leakage patterns, which commonly include “inkblot,” “smokestack,” and multifocal configurations [32, 33]. Smokestack leakage is often linked to choroidal vascular anomalies, carries a higher recurrence risk, and portends poorer outcomes [34]. Diffuse or multifocal leakage suggests widespread choroidal dysfunction involving multiple RPE regions and typically responds less favorably to treatment [35]. Micropulse laser therapy has been reported to yield favorable results in focal leakage cases [36]. OCT, a non-invasive modality, provides high-resolution retinal microstructure imaging, with baseline parameters serving as key predictors of treatment response [16]. Integrity of the RPE and ellipsoid zone has been correlated with favorable SML outcomes [16]. Given its barrier function and role in SRF clearance to the choroid, RPE integrity is critical to preventing edema and SRD [37]. Diffuse RPE alterations have been associated with suboptimal SML response [38, 39], and Altinel et al. [16] identified baseline RPE status as a potential prognostic marker for SML therapy. SRF represents a hallmark manifestation of CSC, and its persistence may precipitate vision deterioration alongside structural and functional retinal changes. Consequently, a primary therapeutic objective in CSC management is complete SRF resolution [40].
During SML intervention, the RPE constitutes the primary therapeutic target, as enhancing its pump function facilitates SRF absorption [41]. SRF height has been reported as a potent predictor of both anatomical and functional recovery in CSC, with greater heights correlating with poorer SML responsiveness [42]. Impaired RPE barrier function may permit fluid accumulation between the retina and RPE, precipitating PED formation [4]. PEDs, more prevalent in chronic CSC, pose substantial visual risks. Literature indicates that PED presence predicts diminished response to reduced-dose verteporfin PDT [43]. Our Bayesian network model further demonstrated that PED indirectly affects treatment outcomes through its influence on MFNRD area and disease duration. Broader and higher PEDs have been implicated in enlarging neurosensory retinal leakage regions, prolonging CSC, and worsening prognosis; additionally, PED frequency may modulate therapeutic results [44, 45].
Furthermore, the presence of subretinal hyperreflective material is considered a predictor of poor efficacy of 577 nm-SML treatment. The quantity of HF is closely correlated with MFNRD height. Clinical evidence suggests that greater HF counts correspond to higher CSC severity [46, 47]. Lee et al. reported that HF presence serves as a predictor of visual prognosis and CSC recurrence risk [48], while Kiraly et al. observed better SML treatment outcomes in CSC patients with fewer HF [42].
Bayesian network is an intelligent prediction model based on probability theory and graph theory, which is particularly suitable for dealing with the uncertainty and complex correlations that are prevalent in medical data [49]. The application of Bayesian networks in the medical field has achieved remarkable results, especially in disease diagnosis, treatment effect prediction and prognosis assessment. Compared with traditional logistic regression models, Bayesian networks show obvious advantages in handling multi-variable interactions and nonlinear relationships. This study analyzed the risk factors that affect the efficacy of 577 nm-SML in the CSC, and used the Bayesian network model to predict and identify the key factors that affect the efficacy, and timely identify patients with good response to treatment to avoid ineffective treatment. Specific predictive biomarkers can also encourage early treatment, prevent irreversible morphological changes, and reduce visual functional deficits. However, our study still faces many limitations. Due to the high rate of patient loss to follow-up, we can only collect and observe the effects 2 weeks after treatment, and cannot evaluate the long-term effects.
Conclusions
By constructing a Bayesian network prediction model based on 19 key influencing factors related to the therapeutic effect of 577 nm-SML, the Bayesian network can further display the complex network relationships between variables and more intuitively reveal the network connection between the therapeutic effect and related factors.
Acknowledgements
Not applicable.
Abbreviations
- CSC
Central Serous Chorioretinopathy
- OCT
Optical Coherence Tomography
- CMT
Central Macular Thickness
- HF
Hyperreflective
- ORL
Outer Retinal Layer
- RPE
Retinal Pigment Epithelial
- SML
Subthreshold Micropulse Laser
- NPL
Neuroepithelial Layer
- POSL
Photoreceptor Outer Segment Layer
- ONL
Outer Nuclear Layer
- SRF
Subretinal Fluid
- PED
Pigment Epithelium Detachment
- MFV
Macular Foveal Volume
Author contributions
LY Conceptualization, Methodology, Data Curation, Formal Analysis, Writing – Original Draft, Writing – Review & Editing. HZ Conceptualization, Methodology, performed examinations and operations, Writing – Review & Editing. MH Conceptualization, Funding Acquisition, supervision of the research.
Funding
This study was supported by Natural Science Foundation of Guangxi Zhuang Autonomous Region (No.2024GXNSFAA010322) and Science and Technology Plan Project of Nanning Qingxiu District (No.2020016).
Data availability
The data that support the findings of this study are not publicly available due to patient privacy，but are available from the corresponding author on reasonable request.
Declarations
Ethics approval and consent to participate
This study adhered to the principles of the Declaration of Helsinki and was approved by the Medical Ethics Committee of the First Affiliated Hospital of Guangxi Medical University (Ethics Approval Number: 2026-E0192), which waived the requirement for written informed consent.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Liying Fang and Huiyi Zuo contributed equally to this work and share co-first authorship. Both are regarded as joint first authors with equal intellectual and practical input.
References
- 1.Ozdemir I, Eren A, Ersoz G. Outer nuclear layer thickness at the central fovea relation with symptom duration in central serous chorioretinopathy. Int Ophthalmol. 2019;39(6):1323–8. [DOI] [PubMed] [Google Scholar]
- 2.Mirshahi R, et al. Clinical predictive factors and imaging biomarkers of treatment response to half dose PDT in patients with chronic central serous chorioretinopathy. Photodiagnosis Photodyn Ther. 2024;48:104224. [DOI] [PubMed] [Google Scholar]
- 3.van Rijssen TJ, et al. Central serous chorioretinopathy: Towards an evidence-based treatment guideline. Prog Retin Eye Res. 2019;73:100770. [DOI] [PubMed] [Google Scholar]
- 4.Nicholson B, et al. Central serous chorioretinopathy: update on pathophysiology and treatment. Surv Ophthalmol. 2013;58(2):103–26. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Daruich A, et al. Central serous chorioretinopathy: Recent findings and new physiopathology hypothesis. Prog Retin Eye Res. 2015;48:82–118. [DOI] [PubMed] [Google Scholar]
- 6.Ersoz MG, et al. Patient characteristics and risk factors for central serous chorioretinopathy: an analysis of 811 patients. Br J Ophthalmol. 2019;103(6):725–9. [DOI] [PubMed] [Google Scholar]
- 7.Fung AT, Yang Y, Kam AW. Central serous chorioretinopathy: A review. Clin Exp Ophthalmol. 2023;51(3):243–70. [DOI] [PubMed] [Google Scholar]
- 8.Brader HS, Young LH. Subthreshold Diode Micropulse Laser: A Review. Semin Ophthalmol. 2016;31(1–2):30–9. [DOI] [PubMed] [Google Scholar]
- 9.Maruko I, et al. Subthreshold 577 nm micropulse laser treatment for central serous chorioretinopathy. PLoS ONE. 2017;12(8):e0184112. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Gulkas S, Sahin O. Current Therapeutic Approaches to Chronic Central Serous Chorioretinopathy. Turk J Ophthalmol. 2019;49(1):30–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Long H, et al. 577 nm subthreshold micropulse laser treatment for acute central serous chorioretinopathy: a comparative study. BMC Ophthalmol. 2022;22(1):105. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Zhou L, et al. Subthreshold Micropulse Laser vs. Conventional Laser for Central Serous Chorioretinopathy: A Randomized Controlled Clinical Trial. Front Med (Lausanne). 2021;8:682264. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Zhou L, et al. A pilot prospective study of 577-nm yellow subthreshold micropulse laser treatment with two different power settings for acute central serous chorioretinopathy. Lasers Med Sci. 2019;34(7):1345–51. [DOI] [PubMed] [Google Scholar]
- 14.Park SB, et al. Causal Inference Network of Genes Related with Bone Metastasis of Breast Cancer and Osteoblasts Using Causal Bayesian Networks. J Bone Metab. 2018;25(4):251–66. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Karaboga HA, et al. Bayesian network as a decision tool for predicting ALS disease. Brain Sci. 2021;11(2). [DOI] [PMC free article] [PubMed]
- 16.Altinel MG, et al. Optical coherence tomography parameters as predictors of treatment response to a 577-nm subthreshold micropulse laser in chronic central serous chorioretinopathy. Lasers Med Sci. 2021;36(7):1505–14. [DOI] [PubMed] [Google Scholar]
- 17.Ma SX, et al. Bayesian Networks in Radiology. Radiol Artif Intell. 2023;5(6):e210187. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Kang J, et al. LASSO-Based Machine Learning Algorithm for Prediction of Lymph Node Metastasis in T1 Colorectal Cancer. Cancer Res Treat. 2021;53(3):773–83. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Li Y, Lu F, Yin Y. Applying logistic LASSO regression for the diagnosis of atypical Crohn’s disease. Sci Rep. 2022;12(1):11340. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Isik MU, Degirmenci MFK, Saglik A. Efficacy of the subthreshold micropulse yellow wavelength laser photostimulation in the treatment of chronic central serous chorioretinopathy. Int J Ophthalmol. 2020;13(9):1404–10. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Daruich A, Marchionno MA, Acute L, et al. Factors influencing episode duration.pdf. Retina. 2017;37(10):1905–15. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Tsai DC, et al. Epidemiology of idiopathic central serous chorioretinopathy in Taiwan, 2001–2006: a population-based study. PLoS ONE. 2013;8(6):e66858. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Shahin MM. Angiographic characteristics of central serous chorioretinopathy in an Egyptian population. Int J Ophthalmol. 2013;6(3):342–5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Kitzmann AS, et al. The incidence of central serous chorioretinopathy in Olmsted County, Minnesota, 1980–2002. Ophthalmology. 2008;115(1):169–73. [DOI] [PubMed] [Google Scholar]
- 25.Landa G, et al. Quantitative and Qualitative Spectral Domain Optical Coherence Tomography Analysis of Subretinal Deposits in Patients with Acute Central Serous Retinopathy. Ophthalmologica. 2013;230(2):62–8. [DOI] [PubMed] [Google Scholar]
- 26.Hanumunthadu D, et al. Gender variation in central serous chorioretinopathy. Eye (Lond). 2018;32(11):1703–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Uzlu D, et al. The efficacy of subthreshold micropulse yellow laser (577 nm) in chronic central serous chorioretinopathy. Lasers Med Sci. 2021;36(5):981–8. [DOI] [PubMed] [Google Scholar]
- 28.Mrejen S, et al. Long-term Visual Outcomes and Causes of Vision Loss in Chronic Central Serous Chorioretinopathy. Ophthalmology. 2019;126(4):576–88. [DOI] [PubMed] [Google Scholar]
- 29.Liu L, et al. Evaluation of changes in macular structures after subthreshold micropulse laser therapy on chronic central serous chorioretinopathy. Lasers Med Sci. 2024;39(1):83. [DOI] [PubMed] [Google Scholar]
- 30.Gawecki M, Grzybowski A, Swiech A. Biometric Risk Factors for Central Serous Chorioretinopathy. Ophthalmol Ther. 2023;12(2):1327–38. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Fujihara-Mino A, et al. Optical coherence tomography parameters predictive of visual outcome after anti-VEGF therapy for retinal vein occlusion. Clin Ophthalmol. 2016;10:1305–13. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Spitznas M. Number, shape, and topography of leakage points in acute type I central serous retinopathy. Graefes Arch Clin Exp Ophthalmol. 1987;225(6):437–40. [DOI] [PubMed] [Google Scholar]
- 33.Turchetti R, Maia dMHJ. Number, shape, and topography of leakage points in patients with central serous chorioretinopathy. Arq Bras Oftalmol. 2005;68(3):317–20. [DOI] [PubMed] [Google Scholar]
- 34.Pryds A, Sander B, Larsen M. Characterization of subretinal fluid leakage in central serous chorioretinopathy. Invest Ophthalmol Vis Sci. 2010;51(11):5853–7. [DOI] [PubMed] [Google Scholar]
- 35.van Rijssen TJ, et al. Clinical characteristics of chronic central serous chorioretinopathy patients with insufficient response to reduced-settings photodynamic therapy. Graefes Arch Clin Exp Ophthalmol. 2018;256(8):1395–402. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.van Rijssen TJ, et al. Focal and Diffuse Chronic Central Serous Chorioretinopathy Treated With Half-Dose Photodynamic Therapy or Subthreshold Micropulse Laser: PLACE Trial Report 3. Am J Ophthalmol. 2019;205:1–10. [DOI] [PubMed] [Google Scholar]
- 37.Rizzolo LJ, et al. Integration of tight junctions and claudins with the barrier functions of the retinal pigment epithelium. Prog Retin Eye Res. 2011;30(5):296–323. [DOI] [PubMed] [Google Scholar]
- 38.Chen SN, et al. Subthreshold diode micropulse photocoagulation for the treatment of chronic central serous chorioretinopathy with juxtafoveal leakage. Ophthalmology. 2008;115(12):2229–34. [DOI] [PubMed] [Google Scholar]
- 39.Zheng F, et al. OCT biomarkers related to subthreshold micropulse laser treatment effect in central serous chorioretinopathy. BMC Ophthalmol. 2022;22(1):252. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Altinel MG, et al. Comparison of the efficacy and safety between subthreshold micropulse laser, standard-fluence and low-fluence photodynamic therapy for chronic central serous chorioretinopathy. J Fr Ophtalmol. 2021;44(4):499–508. [DOI] [PubMed] [Google Scholar]
- 41.Wu Z, Wang H, An J. Comparison of the efficacy and safety of subthreshold micropulse laser with photodynamic therapy for the treatment of chronic central serous chorioretinopathy: A meta-analysis. Med (Baltim). 2021;100(17):e25722. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42.Kiraly P, Smrekar J, Jaki Mekjavic P. Morphological parameters predicting subthreshold micropulse laser effectiveness in central serous chorioretinopathy. Lasers Med Sci. 2022;37(8):3129–36. [DOI] [PubMed] [Google Scholar]
- 43.Chan WM, Lai LT, Tang RY, Liu EW, Lam DT. Safety enhanced photodynamic therapy for chronic central serous chorioretinopathy: one-year results of a prospective study. Retina. 2008;28(1):85–93. [DOI] [PubMed] [Google Scholar]
- 44.Chronopoulos A, et al. The significance of pigment epithelial detachment in central serous chorioretinopathy. Eur J Ophthalmol. 2021;31(2):556–65. [DOI] [PubMed] [Google Scholar]
- 45.Borrelli E, et al. Optical coherence tomography parameters as predictors of treatment response to eplerenone in central serous chorioretinopathy. J Clin Med. 2019;8(9). [DOI] [PMC free article] [PubMed]
- 46.Yalcinbayir O, Akova-Budak GO, Ozkaya B, Gorkem Cevik G, Yucel S. Correlation of spectral domain optical coherence tomography findings and visual acuity in central serous chorioretinopathy. Retina. 2014;34(4):705–12. [DOI] [PubMed] [Google Scholar]
- 47.Hanumunthadu D, et al. Evaluation of choroidal hyperreflective dots in acute and chronic central serous chorioretinopathy. Indian J Ophthalmol. 2019;67(11):1850–4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 48.Lee H, Chung LJ, Kim H. Baseline spectral domain optical hc, coherence tomographic hyperreflective foci as a predictor of visual outcome and recurrence for central serous chorioretinopathy. Retina. 2016;36(7):1372–80. [DOI] [PubMed] [Google Scholar]
- 49.Sugahara S, Aomi I, Ueno M. Bayesian netw model averaging classifiers subbagging entropy (Basel). 2022;24(5). [DOI] [PMC free article] [PubMed]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
The data that support the findings of this study are not publicly available due to patient privacy，but are available from the corresponding author on reasonable request.