Skip to main content
BMC Ophthalmology logoLink to BMC Ophthalmology
. 2026 Aug 10;26:736. doi: 10.1186/s12886-026-05182-0

Predictive value of the serum albumin-to-globulin ratio combined with selected serum inflammatory cytokines for poor 3-month treatment response in immune-mediated noninfectious uveitis

Ya Li 1,✉, Mingran Li 1, Minglian Zhang 1
PMCID: PMC13741421  PMID: 42989160

Abstract

Objective

To assess whether the serum albumin-to-globulin ratio (AGR), alone and in combination with selected serum inflammatory cytokines, predicts poor 3-month treatment response in patients with immune-mediated noninfectious uveitis.

Methods

This prospective observational cohort study included 200 patients with immune-mediated noninfectious uveitis, comprising 50 patients each with ankylosing spondylitis-associated uveitis, Vogt–Koyanagi–Harada-associated uveitis, Behçet disease-associated uveitis, and noninfectious choroiditis-spectrum uveitis. Fifty healthy controls were frequency-matched to the patient cohort by age and sex. Baseline AGR, routine inflammatory indices, selected serum inflammatory cytokines, logMAR BCVA, OCT and FFA findings, and ocular inflammatory activity scores were recorded before systemic anti-inflammatory treatment. The primary outcome was poor treatment response at 3 months, defined as meeting at least one prespecified clinical, imaging, or visual criterion. Treatment escalation and inflammatory relapse were analyzed as supplementary outcomes. Logistic regression, ROC analysis, calibration assessment, bootstrap internal validation, and decision curve analysis were performed.

Results

AGR was lower in all four uveitis subgroups than in healthy controls. Descriptively, the lowest AGR values were observed in Behçet disease-associated uveitis and noninfectious choroiditis-spectrum uveitis. At 3 months, 52 of 200 patients (26.00%) had a poor treatment response. Compared with patients with a good treatment response, those with a poor response had lower baseline AGR and greater systemic and ocular inflammatory burden, reflected by higher cytokine levels, routine inflammatory indices, baseline logMAR BCVA, central macular thickness, FFA leakage scores, and revised ocular inflammatory activity scores. AGR was inversely correlated with IL-17 (rₛ=−0.521), IL-6 (rₛ=−0.506), and the revised ocular inflammatory activity score (rₛ=−0.500; all P < 0.001). In the final multivariable model, baseline logMAR BCVA (adjusted OR per 0.1-unit increase, 1.292; 95% CI, 1.098–1.520) and the revised ocular inflammatory activity score (adjusted OR per 1-point increase, 2.771; 95% CI, 1.843–4.167) remained independently associated with poor treatment response. By contrast, the associations of AGR, IL-17, the IL-6/IL-10 ratio, and CRP were attenuated after adjustment. AGR alone yielded an AUC of 0.775. The cytokine model comprising IL-17 and the IL-6/IL-10 ratio yielded an AUC of 0.738, which increased to 0.800 after AGR was added. The final model, which additionally included CRP, baseline logMAR BCVA, and the revised ocular inflammatory activity score, achieved an AUC of 0.903, with a bootstrap-corrected AUC of 0.887. The model showed acceptable calibration, with a Brier score of 0.103 and a Hosmer–Lemeshow P value of 0.675, and provided a positive net benefit across threshold probabilities of approximately 0.10–0.60.

Conclusion

Low baseline AGR was associated with greater systemic and ocular inflammatory burden and poor 3-month treatment response. A multivariable model combining AGR with selected inflammatory and ocular variables showed good internal discrimination. External validation is required before clinical implementation.

Keywords: Immune-mediated noninfectious uveitis, Albumin-to-globulin ratio, Serum inflammatory cytokines, IL-17, 3-month treatment response, Prediction model

Introduction

Immune-mediated noninfectious uveitis can involve the uveal tract, retina, and retinal vessels. It is an important cause of visual impairment in working-age adults. The disease is often associated with ankylosing spondylitis, Vogt-Koyanagi-Harada disease, Behçet disease, and several noninfectious choroiditis entities. The site of inflammation, risk of relapse, and treatment response vary across these conditions. Despite treatment, some patients show incomplete resolution of intraocular inflammation, persistent vascular leakage, macular edema, or visual deterioration within the first 3 months. Early identification of patients at risk for poor treatment response may help guide follow-up intensity and anti-inflammatory management [1, 2].

Uveitis activity is mainly evaluated by slit-lamp examination, fundus examination, best-corrected visual acuity, OCT, FFA, and, when needed, indocyanine green angiography. These tests show intraocular inflammation and structural injury. They do not fully capture systemic inflammatory burden or peripheral immune activation. In clinical practice, patients with similar baseline ocular findings may differ substantially in the rate of inflammatory resolution, imaging improvement, and visual recovery. Ocular morphology alone is therefore insufficient for short-term prognostic assessment.

The serum albumin-to-globulin ratio (AGR) reflects both reduced albumin and increased globulin. Albumin decreases during inflammation because of acute-phase responses, increased vascular permeability, and changes in nutritional status. Globulin elevation often reflects increased immunoglobulins and chronic immune activation. AGR has been linked to disease activity, inflammatory burden, and adverse outcomes in systemic lupus erythematosus and inflammatory bowel disease [3, 4]. It is derived from routine liver function testing and can be repeatedly measured at low cost.

Persistent immune-mediated noninfectious uveitis is associated with enhanced Th1/Th17 responses, disruption of the blood–ocular barrier, and impaired immune regulation. Prior biomarker studies have shown abnormal inflammatory proteins and cytokines in aqueous humor, tears, and peripheral blood in patients with uveitis [5]. IL-2 reflects T-cell activation. IFN-gamma indicates a Th1-related response. IL-17 represents the Th17 proinflammatory axis [6]. IL-6 amplifies inflammation and promotes Th17 differentiation. TNF-alpha contributes to endothelial activation and tissue inflammation. IL-10 reflects anti-inflammatory regulation [7–9]. If low AGR is accompanied by higher proinflammatory cytokines and heavier ocular inflammatory burden, AGR may represent systemic inflammatory status rather than simple nutritional or hepatic status.

Previous studies have evaluated the C-reactive protein-to-albumin ratio, NLR, PLR, and SII as markers of uveitis activity [10]. Evidence remains limited regarding the ability of AGR combined with selected serum inflammatory cytokines to predict treatment response. We therefore assessed baseline AGR, serum inflammatory cytokines, routine inflammatory indices, logMAR BCVA, OCT and FFA findings, and ocular inflammatory activity in four forms of immune-mediated noninfectious uveitis. The primary objective was to develop and internally validate a model for poor 3-month treatment response.

Materials and methods

Study population and grouping

This was a single-center prospective observational study. The protocol was approved by the Ethics Committee of Hebei Eye Hospital (approval No. 2025KY1401). Potentially eligible patients were consecutively screened within four prespecified disease strata. Recruitment remained open after each loss to follow-up within the corresponding stratum until 50 evaluable patients per stratum had completed the 3-month assessment. Overall, 263 patients were screened, 59 were excluded before enrollment, and 204 were enrolled. Four enrolled patients were lost to follow-up, leaving 200 patients in the primary analysis, with 50 patients in each disease stratum. Fifty healthy individuals undergoing routine health examinations were frequency-matched to the patient cohort by age and sex.

Sample size estimation

The initial sample size calculation was based on between-group differences in AGR. Assuming a medium effect size of f = 0.25, a two-sided α of 0.05, 80% power, and five study groups, G*Power indicated that approximately 196 participants were required [11]. After allowing for loss to follow-up and incomplete assessments, the target sample size was set at 250, including 200 patients and 50 healthy controls. Among the 200 patients, 52 experienced the primary outcome. Model development and reporting followed the TRIPOD statement [12]. Considering the number of outcome events, model complexity was restricted to no more than six predictor parameters [13], and optimism was assessed using 1,000 bootstrap resamples.

Study design

All baseline blood samples, ophthalmic examinations, and imaging assessments were completed before initiation of systemic anti-inflammatory treatment. Patients subsequently received treatment according to clinical need and were followed at 14 days, 6 weeks, and 3 months. The prespecified primary analysis evaluated the association of baseline AGR and selected inflammatory variables with poor 3-month treatment response.

Diagnostic and classification criteria

Anatomic classification, inflammatory grading, and outcome terminology followed the Standardization of Uveitis Nomenclature (SUN) criteria [14]. Two experienced uveitis specialists independently reviewed history, ocular signs, imaging, laboratory data, and relevant specialist opinions. A senior ophthalmologist adjudicated disagreements. AS-associated anterior uveitis was classified according to SUN criteria for spondyloarthritis/HLA-B27-associated anterior uveitis [15]. VKH-associated uveitis was diagnosed according to the revised VKH criteria and SUN classification criteria [16, 17]. Behçet-associated uveitis was classified using the International Criteria for Behçet Disease and SUN criteria [18, 19]. Noninfectious choroiditis spectrum uveitis was diagnosed according to SUN posterior uveitis disease-specific criteria, with OCT, FFA, fundus autofluorescence, and exclusion of infectious causes when required [20–24].

Inclusion criteria

Eligible participants were 8–60 years of age, had active immune-mediated noninfectious uveitis belonging to one of the four prespecified disease strata, and underwent baseline blood sampling within 14 days of symptom onset and before systemic anti-inflammatory treatment. Complete baseline laboratory, ophthalmic, and imaging data and planned follow-up at 14 days, 6 weeks, and 3 months were required.

Exclusion criteria

Patients were excluded if infectious uveitis was suspected, including tuberculosis, syphilis, viral, fungal, toxoplasmic, parasitic, or other pathogen-related uveitis. Other exclusion criteria were severe hepatic, renal, cardiac, hematologic, or malignant disease; active systemic autoimmune disease that could substantially affect serum proteins or cytokines; recent use of drugs or blood products that could alter serum proteins; recent infection, trauma, surgery, severe malnutrition, or rapid weight loss; pregnancy or lactation; inability to complete examinations or follow-up; participation in another clinical drug trial; or missing key laboratory, imaging, or follow-up data.

Exclusion of infectious causes

All patients underwent screening for infectious uveitis before enrollment. Basic tests included blood count, CRP, ESR, liver and renal function, treponemal and non-treponemal syphilis tests, tuberculosis-related testing, hepatitis B and C testing, and HIV screening. Suspected posterior uveitis or chorioretinitis required additional chest CT, T-SPOT or PPD, toxoplasma antibodies, viral nucleic acid testing, and intraocular fluid testing when needed. Tubercular uveitis, syphilitic uveitis, toxoplasmic retinochoroiditis, acute retinal necrosis, cytomegalovirus retinitis, and masquerade syndromes were excluded [25–27].

Healthy controls

Healthy controls were recruited from individuals undergoing routine health examinations during the same period and were frequency-matched to the patient cohort by age and sex. Controls had no history of uveitis, recent infection, fever, trauma, surgery, autoimmune disease, liver or kidney disease, hematologic disease, malignancy, or severe malnutrition. Those who had recently used corticosteroids, immunosuppressants, biologics, albumin, immunoglobulin, or other drugs affecting serum protein levels were excluded.

Data collection

Data were collected using a standardized case report form. Baseline variables included age, sex, body mass index, time from onset, affected eye, unilateral or bilateral involvement, previous uveitis episodes, systemic comorbidities, recent medication, initial treatment, and treatment escalation. Laboratory data included blood count, liver and renal function, CRP, ESR, total protein, albumin, globulin, AGR, and cytokine profiles. Ocular data included best-corrected visual acuity, intraocular pressure, anterior chamber cells and flare, vitreous haze, fundus inflammatory findings, OCT, FFA, and B-scan ultrasonography when required.

Serum protein and routine inflammatory markers

Fasting peripheral venous blood was collected in the morning. In patients, baseline sampling was completed before the initiation of systemic corticosteroids, immunosuppressants, or biologic therapy. Serum total protein and albumin were measured using the biuret and bromocresol green methods, respectively, on an AU5800 automated chemistry analyzer (Beckman Coulter, Brea, CA, USA). Globulin was calculated as total protein minus albumin, and AGR as albumin divided by globulin. CRP was measured by a rapid quantitative immunoturbidimetric assay using a D7-CRP analyzer. ESR was determined using a PRD-MSS-SY-08TKN automated erythrocyte sedimentation rate analyzer. Neutrophil, lymphocyte, and platelet counts were obtained by five-part differential analysis using an XN-550 automated hematology analyzer (Sysmex Corporation, Kobe, Japan). NLR and PLR were calculated as the neutrophil and platelet counts divided by the lymphocyte count, respectively, and SII was calculated as platelet count×neutrophil count/lymphocyte count.

Cytokine profiling

Blood samples for cytokine analysis were centrifuged at 3,000 rpm for 10 min using a DM0424 low-speed centrifuge (DLAB Scientific Co., Ltd., Beijing, China). The separated serum was aliquoted and stored at − 80 °C until batch analysis, and repeated freeze–thaw cycles were avoided. Serum IL-2, IL-6, IL-10, IL-17, IFN-γ, and TNF-α concentrations were measured using human Quantikine or Quantikine HS ELISA kits (R&D Systems, a Bio-Techne brand, Minneapolis, MN, USA) in accordance with the manufacturer’s instructions. Optical density was measured at 450 nm using an Epoch 2 microplate spectrophotometer (BioTek Instruments, Winooski, VT, USA). All samples were assayed in duplicate, and samples exceeding the standard-curve range were diluted and reassayed. Kit lot numbers, assay dates, standard-curve performance, and quality-control results were documented. Samples were retested when duplicate variation exceeded the manufacturer-specified acceptance limit. The IL-6/IL-10 ratio was calculated by dividing the IL-6 concentration by the corresponding IL-10 concentration and was treated as an exploratory indicator of the balance between proinflammatory and regulatory responses.

Ophthalmic examinations

Best-corrected visual acuity was measured with a standard visual acuity chart and converted to LogMAR values. Intraocular pressure was measured with a noncontact tonometer or Goldmann applanation tonometer. Slit-lamp examination recorded anterior chamber cells and flare, keratic precipitates, posterior synechiae, pupil morphology, and complicated cataract. Anterior chamber cells and flare were graded using SUN criteria [14]. Fundus examination, OCT, FFA, and B-scan ultrasonography were performed as clinically indicated. Two ophthalmologists independently reviewed images while masked to AGR and cytokine results. Disagreements were resolved by a senior ophthalmologist.

Ocular inflammatory activity score

A four-domain ocular inflammatory activity score was used to quantify baseline inflammatory burden and changes during follow-up. The score comprised anterior chamber cells, vitreous haze, active structural abnormalities on OCT, and active leakage on FFA. Each domain was scored from 0 to 3 according to prespecified criteria, yielding a total score of 0–12. Visual acuity was excluded from the score because it was analyzed separately as logMAR best-corrected visual acuity (BCVA). Higher scores indicated greater ocular inflammatory activity. Two ophthalmologists independently assigned the scores while masked to AGR and cytokine results. Weighted kappa statistics were used for individual domains, and the intraclass correlation coefficient was calculated for the total score.

Treatment and follow-up

All baseline biomarker measurements were completed before initiation of systemic corticosteroids, conventional immunosuppressants, or biologic therapy. Treatment decisions were made by the attending physicians according to disease type, inflammatory severity, ocular involvement, and previous treatment history. Recorded treatment variables included topical corticosteroid use, initial systemic corticosteroid use and prednisone-equivalent dose, conventional immunosuppressant use, biologic use, and subsequent treatment escalation. Follow-up assessments were performed at baseline, 14 days, 6 weeks, and 3 months.Topical corticosteroid use before enrollment was permitted and recorded, whereas prior systemic corticosteroid, immunosuppressant, or biologic treatment was not permitted before baseline blood sampling.

Outcome definition

The primary outcome was poor treatment response at 3 months, defined as a composite binary endpoint. Patients were classified as poor responders if they met at least one of the following criteria: (1) the anterior chamber cell or vitreous haze grade decreased by fewer than two grades and remained above grade 0; (2) new or progressive retinochoroidal inflammatory lesions developed; (3) active OCT abnormalities, including macular edema, serous retinal detachment, or outer retinal inflammatory changes, remained unchanged or worsened from baseline; (4) retinal vascular, optic-disc, or choroidal leakage on FFA remained unchanged, worsened, or newly developed; or (5) an increase of at least 0.30 logMAR in best-corrected visual acuity from baseline to 3 months that was attributable to active ocular inflammation and could not be explained by cataract progression, refractive change, elevated intraocular pressure, noninflammatory media opacity, or another noninflammatory ocular condition [28]. Patients meeting none of these criteria were classified as good responders. Residual imaging abnormalities that had decreased in extent or severity from baseline and were neither new nor progressive were not classified as poor-response events. Multiple criteria could coexist in the same patient. Treatment escalation and inflammatory relapse were supplementary outcomes and did not independently determine the primary outcome [14, 28–30].

Outcomes

The primary analysis assessed the ability of baseline AGR and selected inflammatory and ocular variables to predict poor 3-month treatment response. Secondary analyses included between-subtype differences in AGR, correlations between AGR and inflammatory or ocular variables, model performance, and changes in AGR, ocular inflammatory activity, and logMAR BCVA during follow-up. Quartile, subgroup, longitudinal, and imaging analyses were considered secondary or exploratory.

Statistical analysis

SPSS 26.0 and R 4.3.0 were used. The patient was the unit of analysis. For bilateral uveitis, the eye with the higher baseline ocular inflammatory activity score was selected; if scores were equal, the eye with poorer baseline logMAR BCVA was included. Continuous variables were expressed as mean ± standard deviation or median (interquartile range) and compared using the independent-samples t test, one-way analysis of variance, Mann–Whitney U test, or Kruskal–Wallis test, as appropriate. Categorical variables were compared using the χ² test or Fisher’s exact test. Multiple comparisons were adjusted using the Bonferroni or Dunn–Bonferroni method. Spearman and partial correlation analyses were performed, with partial correlations adjusted for age, sex, BMI, and uveitis subtype. Multivariable logistic regression was used to identify predictors of poor 3-month treatment response. Candidate variables were selected according to clinical relevance and univariable P < 0.10, and the final model was limited to six predictor parameters. Multicollinearity was assessed using variance inflation factors. Model performance was evaluated using ROC curves, calibration analysis, Brier score, 1,000 bootstrap resamples, and decision curve analysis. AUCs were compared using the DeLong test. Longitudinal changes were analyzed using linear mixed-effects models including group, time, and group-by-time interaction. All tests were two-sided, with P < 0.05 considered statistically significant.Quartile, subgroup, sensitivity, longitudinal, and imaging analyses were considered secondary or exploratory.

Missing data and quality control

Among the 204 enrolled patients, four had missing 3-month outcome data because of loss to follow-up. No predictor data were missing among the 200 patients included in the primary analysis. A complete-case analysis was therefore performed.Before the study, investigators were trained in case screening, inflammatory scoring, image reading, data entry, and follow-up recording. Laboratory testing was performed on the same platform. Cytokine assays recorded reagent lot, standard curve, and quality control results. Ocular images were independently reviewed by two masked ophthalmologists. Data were double-entered and checked for logic errors. Cases with missing key variables were not included in the main analysis.

Ethics

The study followed the Declaration of Helsinki. The protocol was approved by the Ethics Committee of Hebei Eye Hospital (No. 2025KY1401; approval date: March 7, 2025). Written informed consent was obtained from all adult participants. For minors, written informed consent was obtained from legal guardians, and assent was obtained when appropriate. Data were coded and used only for the prespecified analyses.

Results

Study population

During the study period, of 263 patients screened, 59 were excluded before enrollment and 204 were enrolled. Four enrolled patients were lost to follow-up before the 3-month assessment, leaving 200 patients in the primary analysis. Recruitment remained open within the corresponding disease strata until 50 evaluable patients per stratum had completed follow-up. Fifty healthy controls were enrolled from 58 individuals screened. Of the 200 patients analyzed, 102 had unilateral and 98 had bilateral uveitis. (Fig. 1)

Fig. 1.

Fig. 1

Flow diagram of participant screening, enrollment, follow-up, and final analysis. Potentially eligible patients were consecutively screened within four prespecified disease strata. Enrollment remained open within each stratum until 50 patients had completed the 3-month assessment and were available for the final analysis. Healthy controls were frequency-matched to the patient cohort by age and sex

Baseline clinical characteristics

Age, sex, and body mass index did not differ significantly across the five study groups. Among the four uveitis subtypes, significant overall differences were observed in bilateral involvement, previous uveitis episodes, baseline logMAR best-corrected visual acuity (BCVA), anterior chamber cell score, vitreous haze score, central macular thickness, FFA leakage score, and the revised ocular inflammatory activity score. Descriptively, AS-associated uveitis showed greater anterior chamber inflammatory activity, VKH-associated uveitis had the highest proportion of bilateral involvement, and Behçet disease-associated uveitis and noninfectious choroiditis-spectrum uveitis showed greater posterior segment inflammatory involvement (Table 1).

Table 1.

Baseline clinical characteristics of the study participants

Variable Healthy controls n = 50 AS-associated uveitis n = 50 VKH-associated uveitis n = 50 Behçet-associated uveitis n = 50 Noninfectious choroiditis spectrum n = 50 Test statistic P value
Age, years 35.41 ± 11.83 33.96 ± 10.72 36.18 ± 12.64 34.52 ± 11.91 35.08 ± 13.24 F = 0.23 0.921
Sex, n (%) χ²=5.98 0.201
Male 26 (52.00) 31 (62.00) 22 (44.00) 32 (64.00) 23 (46.00)
Female 24 (48.00) 19 (38.00) 28 (56.00) 18 (36.00) 27 (54.00)
BMI, kg/m² 23.12 ± 2.71 23.48 ± 2.84 22.96 ± 2.63 23.37 ± 2.91 23.02 ± 2.76 F = 0.34 0.850
Time from onset to blood sampling, d NA 6.42 ± 2.31 7.18 ± 2.64 6.96 ± 2.58 7.04 ± 2.49 F = 0.40 0.751
Bilateral involvement, n (%) NA 8 (16.00) 37 (74.00) 29 (58.00) 24 (48.00) χ²=36.71 < 0.001
Previous episode of uveitis, n (%) NA χ²=14.58 0.002
Yes NA 19 (38.00) 7 (14.00) 24 (48.00) 11 (22.00)
No NA 31 (62.00) 43 (86.00) 26 (52.00) 39 (78.00)
Baseline LogMAR visual acuity NA 0.36 ± 0.21 0.58 ± 0.30 0.62 ± 0.34 0.55 ± 0.28 F = 9.12 < 0.001
Intraocular pressure, mmHg NA 15.42 ± 3.18 14.86 ± 2.91 15.21 ± 3.04 14.74 ± 2.87 F = 0.51 0.676
Anterior chamber cell score NA 2.16 ± 0.68 1.74 ± 0.61 2.08 ± 0.72 1.42 ± 0.57 F = 13.82 < 0.001
Vitreous haze score NA 0.64 ± 0.42 1.38 ± 0.61 1.72 ± 0.69 1.46 ± 0.65 F = 36.75 < 0.001
OCT central macular thickness, µm NA 264.38 ± 39.52 318.74 ± 71.63 334.26 ± 78.45 326.82 ± 74.19 F = 10.35 < 0.001
FFA leakage score NA 0.82 ± 0.53 1.96 ± 0.77 2.18 ± 0.84 2.04 ± 0.79 F = 41.28 < 0.001
Revised ocular inflammatory activity score, 0–12 NA 4.38 ± 1.51 5.76 ± 1.39 6.32 ± 1.36 5.80 ± 1.43 F = 16.98 < 0.001

Note: Data are presented as mean±standard deviation or n (%). AS, ankylosing spondylitis; VKH, Vogt-Koyanagi-Harada disease; BMI, body mass index; LogMAR, logarithm of the minimum angle of resolution; OCT, optical coherence tomography; FFA, fundus fluorescein angiography; NA, not applicable. Bilateral involvement and previous uveitis episodes were compared only among the four uveitis groups

Inter-rater agreement for the ocular inflammatory activity score

The two graders demonstrated consistently good intergrader agreement across the four domains, including anterior chamber cell score, vitreous haze score, OCT activity score, and FFA leakage score, with quadratic weighted κ values ranging from 0.851 to 0.876. The revised four-domain total score also showed good intergrader reliability, with an intraclass correlation coefficient of 0.892 (95% CI, 0.852–0.922) (Table 2).

Table 2.

Inter-rater agreement for the ocular inflammatory activity score

Scoring item Agreement index Value 95% CI Interpretation
Anterior chamber cell score Quadratic weighted κ 0.864 0.814–0.910 Good
Vitreous haze score Quadratic weighted κ 0.864 0.814–0.908 Good
OCT activity score Quadratic weighted κ 0.851 0.795–0.897 Good
FFA leakage score Quadratic weighted κ 0.876 0.836–0.910 Good
Revised four-domain total score ICC 0.892 0.852–0.922 Good

Serum proteins, inflammatory markers, and cytokines across groups

Compared with healthy controls, all four uveitis groups had lower albumin, higher globulin, and lower AGR. AGR was lowest in Behçet-associated uveitis, followed by the noninfectious choroiditis spectrum and VKH-associated uveitis groups. Patients also had higher CRP, ESR, NLR, PLR, SII, IL-2, IL-6, IL-10, IL-17, IFN-gamma, and TNF-alpha levels than healthy controls (Table 3; Fig. 2).

Table 3.

Serum protein parameters, routine inflammatory markers, and cytokine profiles among groups

Variable Healthy controls n = 50 AS-associated uveitis n = 50 VKH-associated uveitis n = 50 Behçet-associated uveitis n = 50 Noninfectious choroiditis spectrum n = 50 Test statistic P value
Total protein, g/L 71.86 ± 4.82 71.62 ± 5.13 72.28 ± 5.46 73.14 ± 5.72 72.96 ± 5.38 F = 0.83 0.507
Albumin, g/L 44.28 ± 3.16 41.52 ± 3.94 40.76 ± 4.18 39.68 ± 4.35 40.42 ± 4.06 F = 12.47 < 0.001
Globulin, g/L 27.58 ± 3.71 30.10 ± 4.28 31.52 ± 4.64 33.46 ± 5.12 32.54 ± 4.83 F = 13.86 < 0.001
AGR 1.63 ± 0.24 1.40 ± 0.27 1.31 ± 0.25 1.20 ± 0.24 1.26 ± 0.26 F = 29.74 < 0.001
CRP, mg/L 2.38 (1.42, 4.16) 7.64 (4.18, 13.25) 10.82 (5.96, 18.74) 15.36 (8.91, 25.42) 11.95 (6.32, 19.86) H = 78.21 < 0.001
ESR, mm/h 7.50 (5.00, 11.00) 18.00 (11.00, 28.00) 26.00 (16.00, 39.00) 31.00 (19.00, 46.00) 27.00 (17.00, 41.00) H = 82.37 < 0.001
NLR 1.82 (1.41, 2.36) 2.54 (1.91, 3.42) 3.08 (2.16, 4.11) 3.74 (2.66, 4.96) 3.22 (2.31, 4.38) H = 61.45 < 0.001
PLR 112.40 (89.73, 137.56) 134.82 (106.45, 167.32) 151.76 (119.64, 188.41) 168.35 (132.57, 207.86) 156.28 (125.16, 196.40) H = 48.92 < 0.001
SII 398.62 (301.28, 526.84) 642.38 (482.60, 843.71) 784.26 (556.38, 1012.45) 951.72 (676.50, 1254.33) 828.41 (602.17, 1098.56) H = 66.18 < 0.001
IL-2, pg/mL 8.41 (6.82, 10.32) 13.26 (10.18, 16.84) 15.72 (11.94, 19.86) 17.38 (13.46, 22.14) 15.94 (12.05, 20.32) H = 73.64 < 0.001
IL-6, pg/mL 4.62 (3.15, 6.48) 14.86 (9.72, 22.38) 21.45 (14.26, 31.82) 28.74 (18.52, 42.63) 24.16 (15.38, 36.75) H = 91.36 < 0.001
IL-10, pg/mL 4.96 (3.74, 6.18) 6.42 (4.85, 8.26) 8.14 (6.05, 10.64) 7.68 (5.86, 10.13) 7.31 (5.54, 9.86) H = 46.25 < 0.001
IL-17, pg/mL 12.38 (8.72, 16.95) 24.71 (18.36, 31.80) 31.62 (23.48, 39.75) 38.94 (29.65, 49.28) 34.18 (25.40, 44.76) H = 96.43 < 0.001
IFN-γ, pg/mL 15.46 (11.68, 20.74) 28.42 (21.65, 36.94) 35.86 (26.74, 46.52) 41.72 (31.68, 54.23) 36.94 (27.82, 48.61) H = 87.92 < 0.001
TNF-α, pg/mL 10.82 (7.95, 14.63) 21.58 (15.46, 29.82) 27.34 (19.76, 36.55) 33.86 (24.10, 45.28) 29.74 (21.64, 39.50) H = 84.57 < 0.001
IL-6/IL-10 ratio 0.92 (0.61, 1.35) 2.31 (1.54, 3.39) 2.78 (1.86, 4.18) 3.74 (2.42, 5.62) 3.28 (2.11, 4.94) H = 88.29 < 0.001

Note: CRP, ESR, NLR, PLR, SII, and cytokine variables are presented as median (Q1, Q3). Between-group comparisons were performed using the Kruskal-Wallis H test

Fig. 2.

Fig. 2

Distribution of AGR among healthy controls and the four uveitis subgroups. Each point represents an individual participant. The center line indicates the median, the box represents the interquartile range, and the whiskers extend to 1.5 times the interquartile range. Overall between-group differences were assessed using the Kruskal–Wallis H test. AGR, albumin-to-globulin ratio; AS, ankylosing spondylitis; VKH, Vogt–Koyanagi–Harada disease

Baseline characteristics by treatment response

At 3 months, 148 patients had a good treatment response and 52 had a poor treatment response, corresponding to an overall poor-response rate of 26.0% (95% CI, 20.4%–32.5%). Poor-response rates were 16.0% in AS-associated uveitis, 24.0% in VKH-associated uveitis, 34.0% in Behçet disease-associated uveitis, and 30.0% in noninfectious choroiditis-spectrum uveitis, with no significant difference among disease subtypes (χ²=4.78, P = 0.188). Compared with the good treatment response group, the poor treatment response group had higher frequencies of bilateral involvement (61.54% vs. 44.59%; OR = 1.99, 95% CI, 1.04–3.79; P = 0.036), previous uveitis episodes (42.31% vs. 26.35%; OR = 2.05, 95% CI, 1.06–3.97; P = 0.031), and initial immunosuppressive therapy (42.31% vs. 26.35%; OR = 2.05, 95% CI, 1.06–3.97; P = 0.031). Baseline logMAR BCVA was poorer in the poor treatment response group (0.76 ± 0.34 vs. 0.46 ± 0.25; mean difference, 0.30; 95% CI, 0.21–0.39; P < 0.001). This group also had higher anterior chamber cell scores (2.24 ± 0.73 vs. 1.68 ± 0.66; mean difference, 0.56; 95% CI, 0.34–0.78; P < 0.001), vitreous haze scores (1.86 ± 0.71 vs. 1.13 ± 0.62; mean difference, 0.73; 95% CI, 0.53–0.93; P < 0.001), central macular thickness (357.82 ± 82.41 μm vs. 292.64 ± 61.85 μm; mean difference, 65.18 μm; 95% CI, 43.64–86.72 μm; P < 0.001), FFA leakage scores (2.42 ± 0.81 vs. 1.54 ± 0.74; mean difference, 0.88; 95% CI, 0.64–1.12; P < 0.001), and revised ocular inflammatory activity scores (7.04 ± 1.15 vs. 5.05 ± 1.38; mean difference, 1.99; 95% CI, 1.57–2.41; P < 0.001). Age, sex, BMI, time from symptom onset to blood sampling, intraocular pressure, initial systemic corticosteroid therapy, and initial biologic therapy did not differ significantly between the two groups (Table 4).

Table 4.

Baseline clinical characteristics according to treatment response

Variable Good treatment response n = 148 Poor treatment response n = 52 Effect estimate 95% CI Test statistic P value
Age, years 34.36 ± 12.18 36.13 ± 13.07 MD = 1.77 −2.18 to 5.72 t = 0.89 0.376
Male sex, n (%) 78 (52.70) 30 (57.69) OR = 1.22 0.65 to 2.32 χ²=0.39 0.532
BMI, kg/m² 23.08 ± 2.74 23.62 ± 2.92 MD = 0.54 −0.35 to 1.43 t = 1.21 0.228
Time from onset to blood sampling, d 6.78 ± 2.42 7.31 ± 2.66 MD = 0.53 −0.26 to 1.32 t = 1.33 0.185
Bilateral involvement, n (%) 66 (44.59) 32 (61.54) OR = 1.99 1.04 to 3.79 χ²=4.42 0.036
Previous uveitis episode, n (%) 39 (26.35) 22 (42.31) OR = 2.05 1.06 to 3.97 χ²=4.67 0.031
Disease type, n (%) Cramér’s V = 0.155 — χ²=4.78 0.188
AS-associated uveitis 42 (28.38) 8 (15.38) — — — —
VKH-associated uveitis 38 (25.68) 12 (23.08) — — — —
Behçet disease-associated uveitis 33 (22.30) 17 (32.69) — — — —
Noninfectious choroiditis-spectrum uveitis 35 (23.65) 15 (28.85) — — — —
Baseline logMAR BCVA 0.46 ± 0.25 0.76 ± 0.34 MD = 0.30 0.21 to 0.39 t = 6.72 < 0.001
Intraocular pressure, mmHg 15.03 ± 2.91 15.36 ± 3.22 MD = 0.33 −0.62 to 1.28 t = 0.69 0.491
Anterior chamber cell score 1.68 ± 0.66 2.24 ± 0.73 MD = 0.56 0.34 to 0.78 t = 5.13 < 0.001
Vitreous haze score 1.13 ± 0.62 1.86 ± 0.71 MD = 0.73 0.53 to 0.93 t = 7.05 < 0.001
Central macular thickness, µm 292.64 ± 61.85 357.82 ± 82.41 MD = 65.18 43.64 to 86.72 t = 5.93 < 0.001
FFA leakage score 1.54 ± 0.74 2.42 ± 0.81 MD = 0.88 0.64 to 1.12 t = 7.20 < 0.001
Revised ocular inflammatory activity score, 0–12 5.05 ± 1.38 7.04 ± 1.15 MD = 1.99 1.57 to 2.41 t = 9.32 < 0.001
Initial systemic corticosteroid therapy, n (%) 112 (75.68) 43 (82.69) OR = 1.54 0.68 to 3.45 χ²=1.06 0.303
Initial immunosuppressive therapy, n (%) 39 (26.35) 22 (42.31) OR = 2.05 1.06 to 3.97 χ²=4.67 0.031
Initial biologic therapy, n (%) 8 (5.41) 6 (11.54) OR = 2.28 0.75 to 6.92 Fisher’s exact test 0.202

Note: Data are presented as mean±standard deviation or n (%). Effect estimates are reported as mean differences (MDs) for continuous variables and odds ratios (ORs) for binary variables, with the good treatment response group as the reference. Mean differences were calculated as the value in the poor treatment response group minus that in the good treatment response group; therefore, a positive MD indicates a higher value in the poor treatment response group. Disease type was compared using a 4 × 2 contingency table, with Cramér’s V reported as the effect-size measure. Continuous variables were compared using the independent-samples t test. Categorical variables were compared using the χ² test or Fisher’s exact test, as appropriate. AS, ankylosing spondylitis; VKH, Vogt–Koyanagi–Harada disease; BMI, body mass index; BCVA, best-corrected visual acuity; logMAR, logarithm of the minimum angle of resolution; FFA, fundus fluorescein angiography

Baseline serum markers by treatment response

The poor-response group had lower albumin, higher globulin, and lower AGR than the good-response group. CRP, ESR, NLR, PLR, SII, IL-6, IL-17, IFN-gamma, TNF-alpha, and IL-6/IL-10 ratio were also higher in the poor-response group (Table 5).

Table 5.

Baseline serum markers according to treatment response

Variable Good treatment response n = 148 Poor treatment response n = 52 Test statistic P value
Total protein, g/L 72.06 ± 5.21 72.84 ± 5.58 t = 0.91 0.365
Albumin, g/L 41.18 ± 3.92 38.72 ± 4.34 t = 3.78 < 0.001
Globulin, g/L 30.88 ± 4.42 34.12 ± 5.07 t = 4.39 < 0.001
AGR 1.36 ± 0.25 1.15 ± 0.22 t = 5.39 < 0.001
CRP, mg/L 9.84 (5.42, 17.26) 18.64 (10.72, 30.58) Z = 4.72 < 0.001
ESR, mm/h 22.00 (14.00, 35.00) 38.00 (24.00, 52.00) Z = 4.96 < 0.001
Neutrophil count, ×10⁹/L 5.18 ± 1.63 6.42 ± 1.91 t = 4.51 < 0.001
Lymphocyte count, ×10⁹/L 1.82 ± 0.54 1.56 ± 0.48 t = 3.09 0.002
Platelet count, ×10⁹/L 246.38 ± 62.41 278.52 ± 70.36 t = 3.10 0.002
NLR 2.86 (2.08, 3.94) 4.12 (3.01, 5.68) Z = 4.31 < 0.001
PLR 142.56 (112.38, 178.42) 178.34 (137.65, 226.91) Z = 3.66 < 0.001
SII 710.45 (518.62, 982.31) 1090.78 (742.66, 1456.29) Z = 4.28 < 0.001
IL-2, pg/mL 14.86 (10.92, 18.46) 18.12 (13.68, 23.74) Z = 3.41 0.001
IL-6, pg/mL 18.24 (11.64, 28.36) 32.68 (21.45, 48.92) Z = 5.62 < 0.001
IL-10, pg/mL 7.18 (5.32, 9.41) 8.06 (6.02, 10.76) Z = 2.04 0.041
IL-17, pg/mL 29.42 (21.65, 37.86) 41.68 (31.58, 53.26) Z = 6.18 < 0.001
IFN-γ, pg/mL 33.12 (24.74, 43.58) 45.34 (34.46, 58.20) Z = 5.24 < 0.001
TNF-α, pg/mL 22.46 (16.84, 31.72) 37.62 (27.55, 50.38) Z = 5.81 < 0.001
IL-6/IL-10 ratio 2.56 (1.62, 3.94) 4.05 (2.86, 6.28) Z = 5.11 < 0.001

Correlation analysis

AGR was inversely correlated with all evaluated systemic inflammatory markers and ocular variables. The strongest unadjusted correlations were observed with IL-17 (rₛ=−0.521), IL-6 (rₛ=−0.506), and the revised ocular inflammatory activity score (rₛ=−0.500; all P < 0.001). AGR was also inversely correlated with baseline logMAR BCVA, CMT, and FFA leakage score. After adjustment for age, sex, BMI, and uveitis subtype, the association with the revised ocular inflammatory activity score remained significant (partial r = − 0.455, P < 0.001) (Table 6; Fig. 3).

Table 6.

Correlations of AGR with inflammatory cytokines and ocular inflammatory parameters

Variable Spearman rₛ P value Partial r Adjusted P
IL-2 −0.286 < 0.001 −0.223 0.002
IL-6 −0.506 < 0.001 −0.465 < 0.001
IL-10 −0.236 0.001 −0.243 0.001
IL-17 −0.521 < 0.001 −0.470 < 0.001
IFN-γ −0.487 < 0.001 −0.438 < 0.001
TNF-α −0.464 < 0.001 −0.412 < 0.001
IL-6/IL-10 ratio −0.290 < 0.001 −0.248 < 0.001
CRP −0.440 < 0.001 −0.425 < 0.001
ESR −0.459 < 0.001 −0.425 < 0.001
NLR −0.379 < 0.001 −0.383 < 0.001
PLR −0.322 < 0.001 −0.318 < 0.001
SII −0.413 < 0.001 −0.415 < 0.001
Baseline logMAR BCVA −0.429 < 0.001 −0.391 < 0.001
CMT −0.391 < 0.001 −0.341 < 0.001
FFA leakage score −0.419 < 0.001 −0.374 < 0.001
Revised ocular inflammatory activity score −0.500 < 0.001 −0.455 < 0.001

Note: Partial correlation analysis was adjusted for age, sex, BMI, and disease type

Fig. 3.

Fig. 3

Spearman correlations of the albumin-to-globulin ratio with selected serum inflammatory cytokines, routine inflammatory indices, and ocular variables. Cell values represent Spearman correlation coefficients (rₛ). Red indicates a negative correlation, blue indicates a positive correlation, and darker shading indicates a stronger correlation. **P < 0.01; ***P < 0.001. AGR, albumin-to-globulin ratio; CRP, C-reactive protein; ESR, erythrocyte sedimentation rate; NLR, neutrophil-to-lymphocyte ratio; PLR, platelet-to-lymphocyte ratio; SII, systemic immune-inflammation index; BCVA, best-corrected visual acuity; CMT, central macular thickness; FFA, fundus fluorescein angiography

Exploratory analysis across AGR quartiles

The proportion of patients with a poor treatment response decreased progressively across increasing AGR quartiles, from 54.00% in Q1 to 28.00% in Q2, 20.00% in Q3, and 2.00% in Q4 (P for trend < 0.001). Serum albumin increased across the quartiles, whereas globulin, IL-17, TNF-α, and the IL-6/IL-10 ratio generally decreased. Ocular inflammatory burden showed a similar graded pattern. The revised ocular inflammatory activity score decreased from 6.46 ± 1.33 in Q1 to 4.44 ± 1.37 in Q4, while baseline logMAR BCVA decreased from 0.70 ± 0.35 to 0.38 ± 0.22. Central macular thickness and FFA leakage score also declined across increasing AGR quartiles, from 340.94 ± 66.47 μm to 267.00 ± 64.82 μm and from 2.22 ± 0.86 to 1.26 ± 0.78, respectively (all P for trend < 0.001) (Table 7).

Table 7.

Inflammatory burden and treatment response across AGR quartiles

Variable Q1 n = 50 Q2 n = 50 Q3 n = 50 Q4 n = 50 P for trend
Observed AGR range 0.788–1.127 1.128–1.257 1.261–1.422 1.425–2.057 —
Poor treatment response, n (%) 27 (54.00) 14 (28.00) 10 (20.00) 1 (2.00) < 0.001
Albumin, g/L 37.85 ± 4.24 40.03 ± 3.19 40.97 ± 3.80 43.31 ± 3.49 < 0.001
Globulin, g/L 36.26 ± 4.27 33.54 ± 2.74 30.65 ± 2.71 26.44 ± 2.57 < 0.001
IL-17, pg/mL 44.28 (33.37, 54.57) 33.36 (27.20, 42.50) 28.64 (22.25, 36.37) 25.11 (21.00, 31.68) < 0.001
TNF-α, pg/mL 33.27 (27.49, 42.72) 25.72 (18.31, 31.08) 24.89 (15.17, 33.86) 17.84 (13.51, 22.72) < 0.001
IL-6/IL-10 ratio 3.36 (2.34, 8.02) 3.66 (2.10, 6.26) 2.49 (1.54, 4.88) 2.08 (1.24, 3.10) < 0.001
Revised ocular inflammatory activity score, 0–12 6.46 ± 1.33 5.94 ± 1.53 5.42 ± 1.40 4.44 ± 1.37 < 0.001
Baseline logMAR BCVA 0.70 ± 0.35 0.63 ± 0.29 0.44 ± 0.23 0.38 ± 0.22 < 0.001
Central macular thickness, µm 340.94 ± 66.47 325.94 ± 71.24 304.36 ± 70.92 267.00 ± 64.82 < 0.001
FFA leakage score 2.22 ± 0.86 1.94 ± 0.91 1.58 ± 0.84 1.26 ± 0.78 < 0.001

Note: Data are presented as mean±standard deviation, median (interquartile range), or n (%), as appropriate. AGR quartiles were defined by ranking the 200 patients according to baseline AGR and assigning 50 patients to each quartile. The trend in the proportion of patients with a poor treatment response was assessed using the Cochran–Armitage trend test. Trends in approximately normally distributed continuous variables were assessed using linear regression with AGR quartile entered as an ordinal variable, whereas trends in non-normally distributed variables were assessed using Spearman rank correlation. The revised ocular inflammatory activity score comprised anterior chamber cell score, vitreous haze score, OCT activity score, and FFA leakage score, with a possible range of 0–12. These analyses were exploratory. AGR, albumin-to-globulin ratio; BCVA, best-corrected visual acuity; logMAR, logarithm of the minimum angle of resolution; FFA, fundus fluorescein angiography

Primary multivariable logistic regression analysis

In univariable analyses, lower AGR and higher levels of IL-17, TNF-α, the IL-6/IL-10 ratio, CRP, ESR, and SII were associated with poor treatment response at 3 months. Bilateral involvement, previous uveitis episodes, poorer baseline logMAR BCVA, a higher revised ocular inflammatory activity score, and initial immunosuppressant use were also associated with poor response. In the final multivariable model, baseline logMAR BCVA and the revised ocular inflammatory activity score remained independently associated with poor treatment response. Each 0.1-unit increase in baseline logMAR BCVA was associated with a 29.2% increase in the odds of poor response (adjusted OR, 1.292; 95% CI, 1.098–1.520; P = 0.002), while each 1-point increase in the revised ocular inflammatory activity score was associated with 2.77-fold higher odds of poor response (adjusted OR, 2.771; 95% CI, 1.843–4.167; P < 0.001). The associations of AGR, IL-17, the IL-6/IL-10 ratio, and CRP were attenuated after adjustment and were no longer statistically significant. No substantial multicollinearity was identified, with variance inflation factors ranging from 1.17 to 1.84 (Table 8).

Table 8.

Univariable and multivariable logistic regression analyses for poor short-term treatment response

Variable Univariable OR 95% CI P value Multivariable OR 95% CI P value
Age, per 1-year increase 1.012 0.986 to 1.038 0.377 NA NA NA
Male sex vs. female sex 1.224 0.647 to 2.316 0.535 NA NA NA
BMI, per 1 kg/m² increase 1.072 0.957 to 1.202 0.231 NA NA NA
Time from onset to blood sampling, per 1-day increase 1.090 0.959 to 1.238 0.186 NA NA NA
Bilateral involvement, yes vs. no 1.988 1.042 to 3.792 0.037 NA NA NA
Previous uveitis episode, yes vs. no 2.050 1.059 to 3.968 0.033 NA NA NA
Baseline logMAR BCVA, per 0.1-unit increase 1.431 1.259 to 1.626 < 0.001 1.292 1.098 to 1.520 0.002
Revised ocular inflammatory activity score, per 1-point increase 3.369 2.322 to 4.889 < 0.001 2.771 1.843 to 4.167 < 0.001
AGR, per 0.1-unit increase 0.547 0.432 to 0.692 < 0.001 0.943 0.688 to 1.294 0.718
IL-17, per 5 pg/mL increase 1.239 1.114 to 1.379 < 0.001 1.098 0.959 to 1.257 0.174
TNF-α, per 5 pg/mL increase 1.511 1.307 to 1.749 < 0.001 NA NA NA
IL-6/IL-10 ratio, per 1-unit increase 1.136 1.042 to 1.238 0.004 1.024 0.910 to 1.152 0.690
CRP, per 5 mg/L increase 1.126 1.042 to 1.218 0.003 1.092 0.973 to 1.226 0.133
ESR, per 1 mm/h increase 1.026 1.012 to 1.039 < 0.001 NA NA NA
SII, per 100-unit increase 1.127 1.060 to 1.198 < 0.001 NA NA NA
Initial immunosuppressive therapy, yes vs. no 2.050 1.059 to 3.968 0.033 NA NA NA
Behçet disease-associated uveitis, yes vs. no 1.693 0.843 to 3.398 0.139 NA NA NA

Note: The final multivariable model included AGR, IL-17, the IL-6/IL-10 ratio, CRP, baseline logMAR BCVA, and the revised ocular inflammatory activity score. ORs for AGR and baseline logMAR BCVA are reported per 0.1-unit increase; ORs for IL-17 and CRP are reported per 5-unit increase. NA indicates that the variable was not included in the final multivariable model. Variance inflation factors for variables included in the final model ranged from 1.17 to 1.84. AGR, albumin-to-globulin ratio; BMI, body mass index; BCVA, best-corrected visual acuity; logMAR, logarithm of the minimum angle of resolution; CRP, C-reactive protein; ESR, erythrocyte sedimentation rate; SII, systemic immune-inflammation index

Model performance

AGR alone showed moderate discrimination for poor 3-month treatment response, with an AUC of 0.775 (95% CI, 0.707–0.843). At an optimal cutoff of 1.223, the sensitivity was 75.00% and the specificity was 66.22%. The cytokine model comprising IL-17 and the IL-6/IL-10 ratio yielded an AUC of 0.738 (95% CI, 0.660–0.816), which did not differ significantly from that of AGR alone (P = 0.346). Adding AGR to the cytokine model increased the AUC from 0.738 to 0.800 (ΔAUC = 0.062; DeLong Z = 2.356; P = 0.018). The final model, which additionally included CRP, baseline logMAR BCVA, and the revised ocular inflammatory activity score, achieved an AUC of 0.903 (95% CI, 0.857–0.949), representing a further improvement over Model 3 (ΔAUC = 0.103; DeLong Z = 3.707; P < 0.001). At the optimal predicted-probability cutoff of 0.241, the final model had a sensitivity of 84.62%, a specificity of 81.08%, a positive predictive value of 61.11%, and a negative predictive value of 93.75%. Compared with AGR alone, the final model increased the AUC by 0.128 (DeLong Z = 4.311; P < 0.001). The bootstrap-corrected AUC was 0.887. The model showed acceptable calibration, with a Brier score of 0.103, a Hosmer–Lemeshow χ² value of 5.75 (P = 0.675), and an optimism-corrected calibration slope of 0.886. Decision curve analysis indicated a positive net benefit across threshold probabilities of 0.10–0.60 (Tables 9 and 10; Figs. 4 and 5).

Table 9.

Performance of different models for predicting poor 3-month treatment response

Model Variables included AUC 95% CI Optimal cutoff Sensitivity Specificity PPV NPV ΔAUC DeLong Z P value
Model 1 AGR 0.775 0.707 to 0.843 1.223 75.00% 66.22% 43.82% 88.29% Reference Reference Reference
Model 2 IL-17, IL-6/IL-10 ratio 0.738 0.660 to 0.816 0.243 71.15% 72.30% 47.44% 87.70% −0.037 −0.943 0.346
Model 3 AGR, IL-17, IL-6/IL-10 ratio 0.800 0.735 to 0.865 0.257 80.77% 68.92% 47.73% 91.07% 0.062 2.356 0.018
Model 4 AGR, IL-17, IL-6/IL-10 ratio, CRP, baseline logMAR BCVA, revised ocular inflammatory activity score 0.903 0.857 to 0.949 0.241 84.62% 81.08% 61.11% 93.75% 0.103 3.707 < 0.001

Note: Model 1 included AGR alone. Model 2 was a cytokine model comprising IL-17 and the IL-6/IL-10 ratio. Model 3 was constructed by adding AGR to Model 2. Model 4 additionally included CRP, baseline logMAR BCVA, and the revised ocular inflammatory activity score. For Model 2, ΔAUC, DeLong Z, and P value refer to the comparison with Model 1; for Model 3, they refer to the comparison with Model 2; and for Model 4, they refer to the comparison with Model 3. The optimal cutoff for Model 1 represents the AGR value, whereas those for Models 2–4 represent predicted probabilities. PPV and NPV were calculated at the observed poor-response prevalence of 26.00%. AGR, albumin-to-globulin ratio; CRP, C-reactive protein; BCVA, best-corrected visual acuity; logMAR, logarithm of the minimum angle of resolution; PPV, positive predictive value; NPV, negative predictive value

Table 10.

Calibration and internal validation of the final prediction model

Metric Result
Apparent AUC 0.903
95% CI 0.857 to 0.949
Bootstrap-corrected AUC 0.887
Number of bootstrap resamples 1,000
Mean optimism 0.016
Hosmer–Lemeshow χ² 5.75
Hosmer–Lemeshow P value 0.675
Brier score 0.103
Apparent calibration intercept 0.000
Apparent calibration slope 1.000
Optimism-corrected calibration intercept −0.047
Optimism-corrected calibration slope 0.886
Threshold range with positive net benefit 0.10 to 0.60

Note: Internal validation was performed using 1,000 bootstrap resamples. Optimism was estimated as the mean difference between model performance in each bootstrap sample and that obtained when the bootstrap-fitted model was applied to the original dataset. The Brier score ranges from 0 to 1, with lower values indicating greater overall predictive accuracy. Decision curve analysis was performed across clinically relevant threshold probabilities. AUC, area under the receiver operating characteristic curve

Fig. 4.

Fig. 4

Receiver operating characteristic curves of four models for predicting poor treatment response at 3 months. Model 1 included AGR alone. Model 2 included IL-17 and the IL-6/IL-10 ratio. Model 3 included AGR, IL-17, and the IL-6/IL-10 ratio. Model 4 included AGR, IL-17, the IL-6/IL-10 ratio, CRP, baseline logMAR BCVA, and the revised ocular inflammatory activity score. The corresponding AUCs were 0.775, 0.738, 0.800, and 0.903, respectively. The diagonal dashed line represents discrimination no better than chance. AGR, albumin-to-globulin ratio; CRP, C-reactive protein; BCVA, best-corrected visual acuity; logMAR, logarithm of the minimum angle of resolution; AUC, area under the receiver operating characteristic curve

Fig. 5.

Fig. 5

Calibration and decision curve analyses of the final prediction model. (A) Calibration plot comparing the predicted and observed probabilities of poor treatment response at 3 months. The dashed line represents perfect calibration, the red line represents the apparent model performance, and the blue line represents the bootstrap bias-corrected performance. The final model yielded a Hosmer–Lemeshow test P value of 0.675, a Brier score of 0.103, a calibration intercept of − 0.047, and a calibration slope of 0.886. (B) Decision curve analysis comparing the clinical net benefit of the four prediction models across a range of threshold probabilities. Model 4 provided the greatest net benefit over threshold probabilities of approximately 0.10–0.60. Model 1 included AGR alone; Model 2 included IL-17 and the IL-6/IL-10 ratio; Model 3 included AGR, IL-17, and the IL-6/IL-10 ratio; and Model 4 included AGR, IL-17, the IL-6/IL-10 ratio, CRP, baseline logMAR BCVA, and the revised ocular inflammatory activity score. AGR, albumin-to-globulin ratio; CRP, C-reactive protein; BCVA, best-corrected visual acuity; logMAR, logarithm of the minimum angle of resolution

Exploratory subgroup analysis by uveitis subtype

AGR showed moderate discriminatory ability for poor 3-month treatment response across the four uveitis subtypes, with AUCs ranging from 0.730 to 0.798. The highest point estimate was observed in AS-associated uveitis (AUC, 0.798; 95% CI, 0.631–0.935), followed by Behçet disease-associated uveitis (AUC, 0.791; 95% CI, 0.640–0.910), VKH-associated uveitis (AUC, 0.757; 95% CI, 0.601–0.886), and noninfectious choroiditis-spectrum uveitis (AUC, 0.730; 95% CI, 0.574–0.857). The subgroup-specific optimal AGR cutoffs ranged from 1.132 to 1.334. Because each subgroup included only 8–17 poor-response events, the confidence intervals were relatively wide and the estimated cutoffs were likely unstable. These findings were therefore considered exploratory and did not constitute independent validation (Table 11).

Table 11.

Exploratory subgroup performance of AGR for predicting poor 3-month treatment response across uveitis subtypes

Uveitis subtype Poor-response events, n AGR AUC 95% CI Optimal AGR cutoff Sensitivity Specificity P value
AS-associated uveitis 8 0.798 0.631 to 0.935 1.311 87.50% 64.29% < 0.001
VKH-associated uveitis 12 0.757 0.601 to 0.886 1.334 91.67% 52.63% < 0.001
Behçet disease-associated uveitis 17 0.791 0.640 to 0.910 1.132 82.35% 78.79% < 0.001
Noninfectious choroiditis-spectrum uveitis 15 0.730 0.574 to 0.857 1.262 93.33% 57.14% 0.002

Note: Subgroup analyses were exploratory. Each uveitis subtype included 50 patients, with 8–17 poor-response events. ROC analyses were performed using lower AGR values to indicate a higher probability of poor treatment response. The 95% CIs were estimated using 1,000 bootstrap resamples. Optimal AGR cutoffs were selected separately within each subgroup using the Youden index and should not be interpreted as validated clinical thresholds. P values test whether the AUC differs from 0.50 and were not adjusted for multiple comparisons; they do not represent comparisons of AUCs among uveitis subtypes. AGR, albumin-to-globulin ratio; AUC, area under the receiver operating characteristic curve; AS, ankylosing spondylitis; VKH, Vogt–Koyanagi–Harada disease

Sensitivity analysis

The discriminatory performance of the final model remained stable across the prespecified sensitivity analyses. When the initially affected eye was selected for bilateral cases in which the revised ocular inflammatory activity scores were equal, the model AUC was 0.902. Similar discrimination was observed after excluding three patients with other defined noninfectious choroiditis entities (AUC, 0.900), patients receiving initial biologic therapy (AUC, 0.906), and patients receiving initial immunosuppressive or biologic therapy (AUC, 0.908). The adjusted association between AGR and poor treatment response remained inverse in all analyses, with ORs ranging from 0.916 to 0.978 per 0.1-unit increase. However, all corresponding 95% CIs included 1.00, consistent with the primary multivariable analysis, in which AGR was not independently associated with poor treatment response after adjustment for the other model variables (Table 12).

Table 12.

Sensitivity analyses of the final prediction model

Analysis strategy Sample size Poor-response events, n Adjusted OR for AGR per 0.1-unit increase 95% CI P value
Primary analysis, using the eye with the higher revised ocular inflammatory activity score as the study eye 200 52 0.943 0.688 to 1.294 0.718
Using the initially affected eye when both eyes had the same revised ocular inflammatory activity score 200 52 0.921 0.671 to 1.263 0.608
Excluding 3 patients with other defined noninfectious choroiditis entities 197 49 0.943 0.686 to 1.295 0.715
Excluding patients receiving initial biologic therapy 186 46 0.916 0.654 to 1.284 0.612
Excluding patients receiving initial immunosuppressive or biologic therapy 139 30 0.978 0.649 to 1.473 0.915

Note: Each sensitivity analysis used the same six-variable logistic regression framework as the primary analysis, including AGR, IL-17, the IL-6/IL-10 ratio, CRP, baseline logMAR BCVA, and the revised ocular inflammatory activity score. ORs represent the adjusted association between each 0.1-unit increase in AGR and poor treatment response at 3 months. In the alternative study-eye analysis, the initially affected eye was selected for bilateral cases in which both eyes had the same revised ocular inflammatory activity score. Fourteen patients receiving initial biologic therapy were excluded from the corresponding analysis. After patients receiving initial immunosuppressive or biologic therapy were identified without duplicate counting, 61 were excluded, leaving 139 patients for analysis. AGR, albumin-to-globulin ratio; CRP, C-reactive protein; BCVA, best-corrected visual acuity; logMAR, logarithm of the minimum angle of resolution; AUC, area under the receiver operating characteristic curve

Longitudinal changes during follow-up

AGR increased progressively during follow-up in both treatment-response groups, whereas the revised ocular inflammatory activity score and logMAR BCVA decreased. The magnitude of improvement was consistently greater in the good treatment response group. Mean AGR increased from 1.36 ± 0.25 at baseline to 1.54 ± 0.22 at 3 months in the good treatment response group, compared with an increase from 1.15 ± 0.22 to 1.24 ± 0.24 in the poor treatment response group. Over the same period, the revised ocular inflammatory activity score decreased from 5.05 ± 1.38 to 1.08 ± 0.95 in the good treatment response group and from 7.04 ± 1.15 to 3.75 ± 1.78 in the poor treatment response group. Mean logMAR BCVA improved from 0.46 ± 0.25 to 0.21 ± 0.16 and from 0.76 ± 0.34 to 0.58 ± 0.28, respectively. Significant time and group effects were observed for all three measures. Significant time-by-group interactions were observed for AGR and logMAR BCVA, whereas the interaction for the reconstructed revised ocular inflammatory activity score did not reach statistical significance (P = 0.188) (Table 13; Fig. 6).

Table 13.

Longitudinal changes in AGR, the revised ocular inflammatory activity score, and logMAR BCVA according to treatment response

Variable Group Baseline 14 d 6 weeks 3 months P for time effect P for group effect P for interaction
AGR Good treatment response n = 148 1.36 ± 0.25 1.42 ± 0.24 1.48 ± 0.23 1.54 ± 0.22 < 0.001 < 0.001 0.002
AGR Poor treatment response n = 52 1.15 ± 0.22 1.18 ± 0.21 1.21 ± 0.23 1.24 ± 0.24 < 0.001 < 0.001 0.002
Revised ocular inflammatory activity score, 0–12 Good treatment response n = 148 5.05 ± 1.38 3.24 ± 1.80 1.97 ± 1.35 1.08 ± 0.95 < 0.001 < 0.001 0.188
Revised ocular inflammatory activity score, 0–12 Poor treatment response n = 52 7.04 ± 1.15 5.44 ± 2.14 4.48 ± 2.07 3.75 ± 1.78 < 0.001 < 0.001 0.188
logMAR BCVA Good treatment response n = 148 0.46 ± 0.25 0.38 ± 0.22 0.29 ± 0.19 0.21 ± 0.16 < 0.001 < 0.001 0.004
logMAR BCVA Poor treatment response n = 52 0.76 ± 0.34 0.70 ± 0.31 0.64 ± 0.30 0.58 ± 0.28 < 0.001 < 0.001 0.004

Note: Data are presented as mean±standard deviation. The revised ocular inflammatory activity score was calculated as the sum of the anterior chamber cell score, vitreous haze score, OCT activity score, and FFA leakage score, yielding a total score of 0–12. At baseline and 3 months, the revised score was calculated directly from the four component scores. Because the component-level scores were not separately retained at 14 days and 6 weeks in the current working dataset, the values at these two time points were reconstructed by removing the contemporaneous visual impairment component from the original five-domain score and constraining the resulting score to the prespecified range of 0–12. Longitudinal changes were assessed using linear mixed-effects models including treatment-response group, follow-up time, and their interaction as fixed effects, with a patient-specific random intercept. AGR, albumin-to-globulin ratio; BCVA, best-corrected visual acuity; logMAR, logarithm of the minimum angle of resolution; OCT, optical coherence tomography; FFA, fundus fluorescein angiography

Fig. 6.

Fig. 6

Longitudinal changes in AGR, the revised ocular inflammatory activity score, and logMAR BCVA according to treatment response. (A) AGR increased progressively during follow-up, with a greater increase in the good treatment response group than in the poor treatment response group. (B) The revised ocular inflammatory activity score decreased in both groups, although the reduction was more pronounced in the good treatment response group. (C) logMAR BCVA improved over time in both groups, with greater visual improvement among patients with a good treatment response. Data are presented as mean±standard deviation. Longitudinal changes were evaluated using linear mixed-effects models incorporating treatment-response group, follow-up time, and the group-by-time interaction as fixed effects, with a patient-specific random intercept. AGR, albumin-to-globulin ratio; BCVA, best-corrected visual acuity; logMAR, logarithm of the minimum angle of resolution

Ocular imaging findings at 3 months

At the 3-month assessment, central macular thickness remained greater in the poor treatment response group than in the good treatment response group (312.58 ± 67.84 μm vs. 244.36 ± 38.72 μm; mean difference, 68.22 μm; 95% CI, 48.37–88.07 μm; P < 0.001). Residual macular edema was present in 46.15% and 8.11% of patients, respectively (OR, 9.714; 95% CI, 4.350–21.696; P < 0.001). The poor treatment response group also had higher frequencies of residual serous retinal detachment (28.85% vs. 4.05%; OR, 9.595; 95% CI, 3.482–26.436), residual choroidal thickening or inflammatory findings (55.77% vs. 12.16%; OR, 9.106; 95% CI, 4.360–19.020), residual retinal vascular leakage on FFA (59.62% vs. 10.81%; OR, 12.179; 95% CI, 5.701–26.015), and residual optic disc leakage on FFA (42.31% vs. 6.76%; OR, 10.120; 95% CI, 4.345–23.568; all P < 0.001). Residual abnormalities in the good treatment response group had decreased in extent or severity from baseline and were neither newly developed nor progressive; therefore, their presence alone did not indicate poor treatment response. During follow-up, 21 patients in the poor treatment response group required treatment escalation and 9 experienced inflammatory relapse. These events were analyzed as supplementary outcomes and were not included as components of the primary endpoint (Table 14).

Table 14.

Ocular imaging findings at 3 months according to treatment response

Variable Good treatment response n = 148 Poor treatment response n = 52 Effect estimate 95% CI Test statistic P value
Central macular thickness, µm 244.36 ± 38.72 312.58 ± 67.84 MD = 68.22 48.37 to 88.07 t = 6.87 < 0.001
Residual macular edema, n (%) 12 (8.11) 24 (46.15) OR = 9.714 4.350 to 21.696 χ²=37.74 < 0.001
Residual serous retinal detachment, n (%) 6 (4.05) 15 (28.85) OR = 9.595 3.482 to 26.436 χ²=25.17 < 0.001
Residual choroidal thickening or inflammatory findings, n (%) 18 (12.16) 29 (55.77) OR = 9.106 4.360 to 19.020 χ²=40.70 < 0.001
Residual retinal vascular leakage on FFA, n (%) 16 (10.81) 31 (59.62) OR = 12.179 5.701 to 26.015 χ²=50.98 < 0.001
Residual optic disc leakage on FFA, n (%) 10 (6.76) 22 (42.31) OR = 10.120 4.345 to 23.568 χ²=36.19 < 0.001

Note: Data are presented as mean±standard deviation or n (%). The effect estimate for central macular thickness is reported as the mean difference (MD), calculated as the value in the poor treatment response group minus that in the good treatment response group. Effect estimates for categorical variables are reported as odds ratios (ORs), with the good treatment response group as the reference. Residual imaging abnormalities were defined as abnormalities that remained detectable at 3 months. In the good treatment response group, these abnormalities had decreased in extent or severity from baseline and were neither newly developed nor progressive. Continuous data were compared using Welch’s t test, and categorical data were compared using Pearson’s χ² test. FFA, fundus fluorescein angiography

Discussion

In this prospective cohort of patients with immune-mediated noninfectious uveitis, lower baseline AGR was associated with greater systemic inflammatory activity, higher levels of selected proinflammatory cytokines, greater ocular inflammatory burden, and poor treatment response at 3 months. However, after adjustment for IL-17, the IL-6/IL-10 ratio, CRP, baseline logMAR BCVA, and the revised ocular inflammatory activity score, the association between AGR and poor treatment response was attenuated and was no longer statistically significant. The final six-variable model nevertheless showed good internal discrimination, with an apparent AUC of 0.903 and a bootstrap-corrected AUC of 0.887, together with acceptable calibration and a positive net benefit across clinically relevant threshold probabilities. These findings indicate that AGR is more appropriately interpreted as one component of a multidimensional risk profile than as an independent prognostic marker [31].

Immune-mediated noninfectious uveitis encompasses disorders with distinct anatomic patterns and immunopathologic backgrounds. AS-associated uveitis most commonly presents as acute anterior uveitis, whereas VKH-associated uveitis is characterized by diffuse choroidal inflammation and serous retinal detachment. Behçet disease-associated uveitis frequently involves vitritis, occlusive retinal vasculitis, and macular edema, while noninfectious choroiditis-spectrum uveitis primarily affects the choroid, outer retina, or chorioretinal interface. Despite these differences, dysregulated innate and adaptive immune responses, inflammatory-cell recruitment, and disruption of the blood–ocular barrier are shared features across these conditions [32, 33]. In the present cohort, AGR was inversely correlated with baseline logMAR BCVA, central macular thickness, FFA leakage score, and the revised ocular inflammatory activity score. These associations support a relationship between systemic inflammatory status and the severity of intraocular inflammation, although they do not establish a direct causal pathway.

The clinical relevance of AGR arises from the opposing behavior of albumin and globulin during systemic inflammation. Albumin is a negative acute-phase protein and may decrease because of altered hepatic protein synthesis, increased vascular permeability, redistribution into the extravascular compartment, and inflammation-related oxidative consumption. The globulin fraction includes immunoglobulins, complement proteins, and several inflammatory proteins; an increase may therefore reflect persistent humoral immune activation [34]. A low AGR may capture concurrent albumin reduction and globulin elevation and may provide a broader representation of inflammatory burden than either component alone. Consistent with this interpretation, patients with a poor treatment response had lower albumin, higher globulin, and lower AGR at baseline. AGR was also inversely correlated with CRP, ESR, NLR, PLR, and SII.

The cytokine profile provided additional evidence of systemic immune activation. Lower AGR was associated with higher concentrations of IL-6, IL-17, IFN-γ, and TNF-α and with a higher IL-6/IL-10 ratio. IL-17 is closely related to Th17-mediated inflammation, whereas IFN-γ reflects Th1-associated immune activity. Both pathways have been implicated in experimental autoimmune uveitis and human noninfectious uveitis [35]. IL-6 promotes Th17 differentiation, endothelial activation, inflammatory-cell recruitment, and disruption of the blood–ocular barrier [36]. TNF-α enhances endothelial adhesion-molecule expression and leukocyte migration, and inhibition of this pathway has an established role in selected patients with refractory noninfectious uveitis [37]. Nevertheless, the associations of IL-17 and the IL-6/IL-10 ratio with treatment response were attenuated in the fully adjusted model. This attenuation may reflect overlapping information among systemic inflammatory markers, cytokine activity, visual function, and baseline ocular inflammatory burden. The present results therefore support their biologic relevance but do not establish them as independent predictors.

NLR, PLR, and SII were also higher among patients with a poor treatment response. These indices are readily derived from routine complete blood counts and impose no additional testing burden. NLR reflects the balance between neutrophil-predominant inflammation and lymphocyte-related immune regulation, whereas SII integrates neutrophil, platelet, and lymphocyte counts and has been examined as an indicator of systemic immune activation [38]. SII has also been associated with inflammatory activity in Behçet disease [39]. However, none of these indices was retained in the final model. Their prognostic information may overlap with that conveyed by AGR, CRP, and cytokine measurements. AGR may have particular clinical appeal because it simultaneously reflects changes in a negative acute-phase protein and the globulin fraction.

The nested model comparisons clarified the relative contribution of the different predictor domains. AGR alone provided moderate discrimination, with an AUC of 0.775. The cytokine model comprising IL-17 and the IL-6/IL-10 ratio yielded an AUC of 0.738 and was not superior to AGR alone. Adding AGR to the cytokine model increased the AUC to 0.800. The largest improvement occurred when CRP, baseline logMAR BCVA, and the revised ocular inflammatory activity score were added, producing an AUC of 0.903. These results indicate that the final model derived much of its discrimination from the integration of systemic inflammation with baseline visual function and directly observed ocular inflammatory activity. This pattern is clinically plausible because short-term treatment response in uveitis depends on both systemic immune activity and the extent of ocular tissue involvement. The therapeutic goals of noninfectious uveitis include suppression of intraocular inflammation, prevention of relapse and structural damage, preservation of vision, and reduction of cumulative corticosteroid exposure [40].

The exploratory subgroup analysis showed moderate discriminatory performance for AGR across all four uveitis subtypes, with AUC point estimates ranging from 0.730 to 0.798. The highest point estimate was observed in AS-associated uveitis, followed by Behçet disease-associated uveitis, but the confidence intervals were wide and overlapped substantially. Each subgroup contained only 50 patients and 8–17 poor-response events, and no formal test demonstrated heterogeneity in AUCs among the subtypes. The subgroup-specific cutoff values should therefore be regarded as unstable, data-dependent estimates rather than disease-specific clinical thresholds. These analyses support the potential applicability of AGR across different forms of uveitis but do not establish superior performance in any particular subtype.

The longitudinal findings also demonstrated distinct clinical trajectories between the treatment-response groups. AGR increased over time in both groups, but the increase was greater among patients with a good treatment response. The revised ocular inflammatory activity score and logMAR BCVA decreased during follow-up, with larger improvements in the good treatment response group. At 3 months, the poor treatment response group retained greater central macular thickness and higher frequencies of residual macular edema, serous retinal detachment, choroidal abnormalities, retinal vascular leakage, and optic disc leakage. Because patients were classified according to their 3-month response, these longitudinal differences should be interpreted descriptively and should not be taken as independent evidence that AGR directly determined subsequent improvement.

Previous studies provide a biologic basis for the observed relationships. Agrawal et al. described abnormalities in IL-6, TNF-α, and IL-17-related pathways in noninfectious autoimmune uveitis [41]. Guedes et al. emphasized the role of IL-17 in inflammatory amplification, leukocyte recruitment, and ocular tissue injury [42]. The current findings are consistent with these observations, as low AGR occurred alongside higher proinflammatory cytokine concentrations and greater ocular inflammatory activity. However, because AGR, IL-17, and the IL-6/IL-10 ratio did not remain independently associated with poor response in the fully adjusted model, they should be interpreted as interconnected markers of inflammatory status rather than isolated causal determinants.(Fig. 7).

Fig. 7.

Fig. 7

Conceptual framework linking systemic inflammatory status, ocular inflammatory burden, and the estimated risk of poor treatment response at 3 months. Systemic inflammation may be accompanied by reduced serum albumin, increased globulin, a lower albumin-to-globulin ratio, elevated routine inflammatory indices, and a proinflammatory cytokine profile. Literature-supported alterations in blood–ocular barrier integrity may contribute to greater ocular inflammatory burden. In the present study, AGR, IL-17, the IL-6/IL-10 ratio, CRP, baseline logMAR BCVA, and the revised ocular inflammatory activity score were integrated into the final prediction model. Dashed arrows denote hypothesized biologic pathways supported by previous literature, whereas solid arrows denote observed associations or components of the prediction framework. The diagram is conceptual and does not imply causality. AGR, albumin-to-globulin ratio; CRP, C-reactive protein; ESR, erythrocyte sedimentation rate; NLR, neutrophil-to-lymphocyte ratio; SII, systemic immune-inflammation index; BCVA, best-corrected visual acuity; OCT, optical coherence tomography; FFA, fundus fluorescein angiography

For patients with inadequate inflammatory control, treatment escalation with immunosuppressive or biologic therapy may be appropriate after consideration of disease subtype, anatomic involvement, previous treatment, and safety factors. Adalimumab has been associated with improved inflammatory control and reduced relapse in selected patients with noninfectious uveitis [43]. Such treatment should be accompanied by appropriate screening and monitoring, because biologic agents and Janus kinase inhibitors may increase susceptibility to intraocular or systemic infection [44]. The prediction model developed in this study was not designed to prescribe a specific treatment or mandate escalation at a particular probability threshold.

The potential clinical value of the present model lies in risk stratification rather than diagnostic classification or direct treatment selection. AGR is inexpensive and routinely available, whereas baseline logMAR BCVA, OCT, and FFA are already integral to uveitis assessment. Combining AGR with IL-17, the IL-6/IL-10 ratio, CRP, baseline logMAR BCVA, and the revised ocular inflammatory activity score may help identify patients who require closer follow-up or earlier repeat imaging. Decision curve analysis indicated a positive net benefit across threshold probabilities of approximately 0.10–0.60, although this finding requires confirmation in an independent cohort before clinical use [45].

AGR should not replace slit-lamp examination, multimodal ocular imaging, or etiologic classification. It is not a disease-specific marker and should not be interpreted as evidence of a particular uveitis subtype. Serum albumin and globulin may be influenced by liver or kidney disease, malnutrition, infection, malignancy, protein loss, immunoglobulin disorders, and other systemic conditions. Although major confounding conditions were addressed in the eligibility criteria, the interpretation of AGR in routine practice should remain clinically contextualized.

This study has several limitations. First, it was conducted at a single center and included 200 patients with 52 poor-response events. Although the final model contained six predictor parameters and was internally validated using bootstrap resampling, the limited number of events leaves a risk of model optimism and unstable coefficient estimates. Second, no independent temporal or external validation cohort was available; bootstrap correction cannot establish transportability to other centers or patient populations. Third, the four included uveitis subtypes differ in pathogenesis, anatomic involvement, and treatment patterns. The subgroup analyses were underpowered, and the subgroup-specific AUCs and cutoff values should be regarded as exploratory. Fourth, the revised ocular inflammatory activity score was developed for this study. Although intergrader reliability was acceptable, its reproducibility, responsiveness, and applicability across imaging platforms and centers require independent evaluation. Fifth, treatment was determined according to clinical need rather than a standardized protocol, creating the possibility of confounding by indication. Sixth, several biomarker, subgroup, and exploratory comparisons were performed, and not all analyses were adjusted for multiplicity; some statistically significant findings may therefore reflect type I error. Finally, cytokine testing is not routinely available in many ophthalmology settings, and assay standardization, turnaround time, and cost may limit immediate clinical application.

Conclusion

Lower baseline AGR was associated with greater systemic inflammatory activity, higher levels of selected proinflammatory cytokines, greater ocular inflammatory burden, and poor treatment response at 3 months in univariable analyses. Its association was attenuated after adjustment for baseline visual function, ocular inflammatory activity, CRP, and cytokine variables, indicating that AGR should not be regarded as an independent predictor on the basis of the present data. A model integrating AGR, IL-17, the IL-6/IL-10 ratio, CRP, baseline logMAR BCVA, and the revised ocular inflammatory activity score showed good internally validated discrimination and may support early risk stratification and follow-up planning. Independent multicenter validation is required before clinical implementation.

Acknowledgements

Not applicable.

Abbreviations

AGR

Albumin-to-globulin ratio

AS

Ankylosing spondylitis

AUC

Area under the receiver operating characteristic curve

BCVA

Best-corrected visual acuity

BMI

Body mass index

CMT

Central macular thickness

CRP

C-reactive protein

ELISA

Enzyme-linked immunosorbent assay

ESR

Erythrocyte sedimentation rate

FFA

Fundus fluorescein angiography

ICC

Intraclass correlation coefficient

IFN-γ

Interferon gamma

IL

Interleukin

logMAR

Logarithm of the minimum angle of resolution

NLR

Neutrophil-to-lymphocyte ratio

NPV

Negative predictive value

OCT

Optical coherence tomography

OR

Odds ratio

PLR

Platelet-to-lymphocyte ratio

PPV

Positive predictive value

ROC

Receiver operating characteristic

SII

Systemic immune-inflammation index

SUN

Standardization of Uveitis Nomenclature

TNF-α

Tumor necrosis factor alpha

VKH

Vogt–Koyanagi–Harada disease

Author contributions

YaLi conceived and designed the study, analyzed the data, and drafted the manuscript. MingRanL icollected the clinical and laboratory data. MingLianZhangsupervised the study design, verified the data, and critically revised the manuscript. All authors read and approved the final manuscript.Ethics approval and consent to participateThis study was approved by the Ethics Committee of Hebei Eye Hospital, Xingtai, Hebei, China (Approval No. 2025KY1401) and was conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants.Patient consent for publicationNot applicable.Availability of data and materialsThe datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Funding

This work was supported by the Xingtai Key Research and Development Program, Social Development Project (No. 2025ZC055). The funder had no role in the study design, data collection, data analysis, interpretation of the results, preparation of the manuscript, or the decision to submit the manuscript for publication.

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This prospective observational study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of Hebei Eye Hospital (No. 2025KY1401; approval date: March 7, 2025). Written informed consent was obtained from all adult participants. For participants younger than 18 years, written informed consent was obtained from their parents or legal guardians, and assent was obtained from the minors, where appropriate. All participant data were coded and used only for the prespecified analyses.

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.

References

  • 1.Tsirouki T, Dastiridou A, Symeonidis C, et al. A focus on the epidemiology of uveitis. Ocul Immunol Inflamm. 2018;26(1):2–16. [DOI] [PubMed] [Google Scholar]
  • 2.Yang P, Zhang Z, Zhou H, et al. Clinical patterns and characteristics of uveitis in a tertiary center for uveitis in China. Curr Eye Res. 2005;30(11):943–8. [DOI] [PubMed] [Google Scholar]
  • 3.Liu M, Li X, Huang Y, Huang Z, Huang Q. Albumin to globulin ratio (AGR) in systemic lupus erythematosus: correlation with disease activity. J Int Med Res. 2024;52(4):3000605241244761. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Wang Y, Li C, Wang W, et al. Serum albumin to globulin ratio is associated with the presence and severity of inflammatory bowel disease. J Inflamm Res. 2022;15:1907–20. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Bansal R, Gupta A. Protein biomarkers in uveitis. Indian J Ophthalmol. 2020;68(12):2845–52. [Google Scholar]
  • 6.Chi W, Zhu X, Yang P, et al. Upregulated IL-23 and IL-17 in Behçet patients with active uveitis. Invest Ophthalmol Vis Sci. 2008;49(7):3058–64. [DOI] [PubMed] [Google Scholar]
  • 7.Balamurugan S, Das D, Hasanreisoglu M, et al. Interleukins and cytokine biomarkers in uveitis. Indian J Ophthalmol. 2020;68(9):1750–63. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Lin P. Targeting interleukin-6 for noninfectious uveitis. Clin Ophthalmol. 2015;9:1697–702. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Bonacini M, Soriano A, Cimino L, et al. Cytokine profiling in aqueous humor samples from patients with non-infectious uveitis associated with systemic inflammatory diseases. Front Immunol. 2020;11:358. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Kim M, Park YG, Park YH. C-reactive protein/albumin ratio as an indicator of disease activity in Behçet’s disease and human leukocyte antigen-B27-associated uveitis. Graefes Arch Clin Exp Ophthalmol. 2021;259:1985–92. [DOI] [PubMed] [Google Scholar]
  • 11.Faul F, Erdfelder E, Lang AG, Buchner A. G*Power 3: a flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behav Res Methods. 2007;39(2):175–91. [DOI] [PubMed] [Google Scholar]
  • 12.Collins GS, Reitsma JB, Altman DG, Moons KGM. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis: the TRIPOD statement. Ann Intern Med. 2015;162(1):55–63. [DOI] [PubMed] [Google Scholar]
  • 13.Riley RD, Snell KIE, Ensor J, Burke DL, Harrell FE Jr, Moons KGM, Collins GS. Minimum sample size for developing a multivariable prediction model: Part II—binary and time-to-event outcomes. Stat Med. 2019;38(7):1276–1296. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Jabs DA, Nussenblatt RB, Rosenbaum JT, Standardization of uveitis nomenclature working group. standardization of uveitis nomenclature for reporting clinical data. results of the first international workshop. Am J Ophthalmol. 2005;140(3): 509–516. [DOI] [PMC free article] [PubMed]
  • 15.Standardization of Uveitis Nomenclature Working Group. Classification criteria for spondyloarthritis/HLA-B27-associated anterior uveitis. Am J Ophthalmol. 2021;228:117–25. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Read RW, Holland GN, Rao NA, et al. Revised diagnostic criteria for Vogt-Koyanagi-Harada disease: report of an international committee on nomenclature. Am J Ophthalmol. 2001;131(5):647–52. [DOI] [PubMed] [Google Scholar]
  • 17.Standardization of Uveitis Nomenclature Working Group. Classification criteria for Vogt-Koyanagi-Harada disease. Am J Ophthalmol. 2021;228:205–11. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.International Team for the Revision of the International Criteria for Behçet’s Disease. The International Criteria for Behçet’s Disease: a collaborative study of 27 countries on the sensitivity and specificity of the new criteria. J Eur Acad Dermatol Venereol. 2014;28(3):338–47. [DOI] [PubMed] [Google Scholar]
  • 19.Standardization of Uveitis Nomenclature Working Group. Classification criteria for Behçet disease uveitis. Am J Ophthalmol. 2021;228:80–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Standardization of Uveitis Nomenclature Working Group. Classification criteria for multifocal choroiditis with panuveitis. Am J Ophthalmol. 2021;228:152–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Standardization of Uveitis Nomenclature Working Group. Classification criteria for punctate inner choroiditis. Am J Ophthalmol. 2021;228:275–80. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Standardization of Uveitis Nomenclature Working Group. Classification criteria for serpiginous choroiditis. Am J Ophthalmol. 2021;228:126–33. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Standardization of Uveitis Nomenclature Working Group. Classification criteria for acute posterior multifocal placoid pigment epitheliopathy. Am J Ophthalmol. 2021;228:174–81. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Standardization of Uveitis Nomenclature Working Group. Classification criteria for multiple evanescent white dot syndrome. Am J Ophthalmol. 2021;228:198–204. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Standardization of Uveitis Nomenclature Working Group. Classification criteria for tubercular uveitis. Am J Ophthalmol. 2021;228:142–51. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Standardization of Uveitis Nomenclature Working Group. Classification criteria for syphilitic uveitis. Am J Ophthalmol. 2021;228:182–91. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Standardization of Uveitis Nomenclature Working Group. Classification criteria for toxoplasmic retinitis. Am J Ophthalmol. 2021;228:134–41. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Nguyen QD, Merrill PT, Jaffe GJ, et al. Adalimumab for prevention of uveitic flare in patients with inactive non-infectious uveitis controlled by corticosteroids: a multicentre, double-masked, randomised, placebo-controlled phase 3 trial. Lancet. 2016;388(10050):1183–92. [DOI] [PubMed] [Google Scholar]
  • 29.Jaffe GJ, Dick AD, Brézin AP, et al. Adalimumab in patients with active noninfectious uveitis. N Engl J Med. 2016;375(10):932–43. [DOI] [PubMed] [Google Scholar]
  • 30.Jabs DA, Rosenbaum JT, Foster CS, et al. Guidelines for the use of immunosuppressive drugs in patients with ocular inflammatory disorders: recommendations of an expert panel. Am J Ophthalmol. 2000;130(4):492–513. [DOI] [PubMed] [Google Scholar]
  • 31.Dick AD, Tundia N, Sorg R, Zhao C, Chao J, Joshi A, et al. Risk of ocular complications in patients with noninfectious intermediate uveitis, posterior uveitis, or panuveitis. Ophthalmology. 2016;123(3):655–62. [DOI] [PubMed] [Google Scholar]
  • 32.Caspi RR. A look at autoimmunity and inflammation in the eye. J Clin Invest. 2010;120(9):3073–83. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Forrester JV, Kuffova L, Dick AD. Autoimmunity, autoinflammation, and infection in uveitis. Am J Ophthalmol. 2018;189:77–85. [DOI] [PubMed] [Google Scholar]
  • 34.Gabay C, Kushner I. Acute-phase proteins and other systemic responses to inflammation. N Engl J Med. 1999;340(6):448–54. [DOI] [PubMed] [Google Scholar]
  • 35.Horai R, Caspi RR. Cytokines in autoimmune uveitis. J Interferon Cytokine Res. 2011;31(10):733–44. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Yu X, Duan R, Jiang L, Wang T, Li Z, Zhang B, et al. Interleukin-6 in non-infectious uveitis: biology, experimental evidence and treatment strategies. Biochem Pharmacol. 2024;230:116605. [DOI] [PubMed] [Google Scholar]
  • 37.Jiang Q, Li Z, Tao T, Duan R, Wang X, Su W. TNF-α in uveitis: from bench to clinic. Front Pharmacol. 2021;12:740057. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Kurtul BE, Ozer PA. Neutrophil-to-lymphocyte ratio in ocular diseases: a systematic review. Int J Ophthalmol. 2019;12(12):1951–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Tanacan E, Dincer D, Erdogan FG, Gurler A. A cutoff value for the Systemic Immune-Inflammation Index in determining activity of Behçet disease. Clin Exp Dermatol. 2021;46(2):286–91. [DOI] [PubMed] [Google Scholar]
  • 40.Rosenbaum JT, Bodaghi B, Couto C, Zierhut M, Acharya N, Pavesio C, et al. New observations and emerging ideas in diagnosis and management of non-infectious uveitis: a review. Semin Arthritis Rheum. 2019;49(3):438–45. [DOI] [PubMed] [Google Scholar]
  • 41.Agrawal R, Iyer J, Connolly J, Iwata D, Teoh S. Cytokines and biologics in non-infectious autoimmune uveitis: bench to bedside. Indian J Ophthalmol. 2014;62(1):74–81. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Guedes MCE, Borrego LM, Proença RD. Roles of interleukin-17 in uveitis. Indian J Ophthalmol. 2016;64(9):628–34. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Khan N, Dutta Majumder P, Janarthanan M, Biswas J. Efficacy of adalimumab in noninfectious pediatric uveitis: Analysis of 29 eyes from a tertiary eye care center in India. Indian J Ophthalmol. 2024;72(12):1766–71. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.Dutta Majumder P, Kene RD. Intraocular infections following biologicals and Janus Kinase inhibitors. Indian J Ophthalmol. 2025;73(8):1091–4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45.Vickers AJ, Elkin EB. Decision curve analysis: a novel method for evaluating prediction models. Med Decis Mak. 2006;26(6):565–74. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Data Availability Statement

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.


Articles from BMC Ophthalmology are provided here courtesy of BMC

RESOURCES