Abstract Background The therapeutic outcome for Type 2 Diabetic Retinopathy (T2DR) following vitrectomy has been unsatisfactory, with no definitive biomarker available to predict treatment response. Identifying a biomarker correlated with treatment efficacy is crucial, as the vitreous—situated between the lens and retina—may influence retinal metabolic perturbations. Methods Vitreous samples were collected during vitrectomy, and their metabolic profiles were analyzed using Ultraperformance Liquid Chromatography coupled with Tandem Mass Spectrometry (UPLC-MS/MS). Statistical analyses were conducted to identify metabolites and metabolic pathways correlated with therapeutic outcomes. Results Patients demonstrating poor therapeutic responses exhibited elevated levels of specific metabolites, including Dodecanoylcarnitine, Linoleylcarnitine, Stearylcarnitine, Decanoic acid, and Proline. Perturbed metabolic pathways included Fatty Acid Biosynthesis, Beta Oxidation of Very Long Chain Fatty Acids, and Mitochondrial Beta-Oxidation of Short Chain Saturated Fatty Acids. These metabolites showed strong discriminatory power for predicting positive outcomes, with Area Under the Curve (AUC) values of 0.925, 0.885, 0.864, 0.811, and 0.808, respectively. Conclusions This study highlights the potential of Dodecanoylcarnitine, Linoleylcarnitine, Stearylcarnitine, Decanoic acid, and Proline as biomarkers for predicting therapeutic outcomes following vitrectomy for T2DR. These findings provide novel insights into the metabolic factors influencing treatment response variability and suggest pathways for future therapeutic interventions. Supplementary Information The online version contains supplementary material available at 10.1186/s12886-025-04283-6. Keywords: Metabolites, Diabetic retinopathy, Vitrectomy Introduction Diabetic retinopathy is the leading cause of blindness among working-age individuals in the United States [[38]1]. Prolonged high blood sugar levels can damage various tissues, such as the kidneys, heart, and retina [[39]2]. Among these, diabetic retinopathy is one of the most serious complications, often resulting in vision impairment [[40]2]. Proliferative diabetic retinopathy (PDR) affects 7% of the 425 million individuals with diabetes mellitus (DM) worldwide. Although treatments like anti-vascular endothelial growth factor (VEGF) therapies and surgery have improved outcomes [[41]2], the growing diabetic epidemic is expected to increase the burden of PDR. PDR often progresses to conditions like tractional retinal detachment in the macular region, non-clearing hemorrhage, and premacular subhyaloid hemorrhage [[42]3, [43]4]. In such cases, vitrectomy is recommended by the American Academy of Ophthalmology. Ostri et al. [[44]5] found that two-thirds of PDR patients achieved a visual acuity of over 0.3 after a vitrectomy over a 10-year period. Factors such as older age, lower best-corrected visual acuity (BCVA), and the presence of tractional retinal detachment were associated with a poorer outcome post-vitrectomy [[45]4]. Given the increasing number of PDR patients undergoing vitrectomy, the outcomes of this therapy vary, highlighting the need to identify biomarkers that can better predict clinical outcomes. Metabolomics, a field that lies downstream of proteomics and gene transcription and deal with monitoring and interpreting the metabolic fluctuation by environmental changes [[46]6]. Also metabolome is closely associated with phenotype and multifactorial diseases. It is used to identify biomarkers for disease diagnosis, prognosis, and treatment selection. Additionally, metabolomics helps enhance our understanding by providing insights into genetic, environmental, and lifestyle factors influencing disease. This study summarizes metabolites changes in diabetic retinopathy and potential pathways to better understand the physiology involved. The vitreous, the largest part of the eye located between the lens and the retina, plays a crucial role in ocular health and diseases, including retinal detachment, diabetic retinopathy, and macular hole formation [[47]7]. It is thought to be involved in normal oxygen metabolism and consumption and affects VEGF-mediated diseases. The research on vitreous samples primarily focuses on the proliferative stage of diabetic retinopathy. Paris et al. used Liquid Chromatograph Mass Spectrometer (LC-MS) to analyze vitreous samples from PDR patients, revealing disruptions in arginine metabolism and the urea cycle, with elevated methionine and arginine linked to impaired glial cell metabolism and neurovascular issues. Excessive arginase activity reduced nitric oxide, contributing to endothelial dysfunction [[48]8]. Previous studies have largely focused on identifying biomarkers and key pathways as potential therapeutic targets for the development of new treatments. In our study, we examined vitreous samples from PDR patients to analyze their relationship with therapy response, aiming to identify biomarkers that can predict treatment outcomes. Meterial and methods Chemicals and regents LC-MS grade methanol, acetonitrile, isopropanol, formic acid, and ammonium formate were purchased from Merck (Darmstadt, Germany). The water used in the experiments was prepared through a Milli-Q pure water system (Merck, Germany). A Q300 Kit which contains stand solutions and derivative reagents [3-nitrophenylhydrazine (3-NPH) and N-(3-(dimethylamino)propyl)-N-ethylcarbodiimide (EDC)·HCl] is provided by Metabo-Profile Biotech Co., LTD (Shanghai, China). All standard compounds and stable isotope-labeled internal standards were acquired from reputable suppliers including Sigma-Aldrich (St. Louis, MO, USA), Steraloids Inc. (Newport, RI, USA), and TRC Chemicals (Toronto, Canada), with a purity exceeding 99.7%. Patients and sample selection This retrospective study was approved by the Institutional Review Board of Wenzhou Medical University Eye Hospital Ethics Committee from January 2020 to December 2023. Informed consent was obtained from each participant before the study began. The sample collection procedure was also approved by the Institutional Review Board, with participants signing informed consent forms for the collection of samples prior to study enrollment. The study was approved by the Biomedical Research Involving humans H2023-012-K-09 by Wenzhou eye hospital. Subjects were recruited from the Retina Department of Wenzhou Medical University Eye Hospital. Patients diagnosed with type 2 diabetes by endocrinologists at the diabetes center were considered for inclusion. All subjects underwent vitrectomy, and when necessary, this was combined with cataract surgery. Indications for surgery, determined by senior retina specialists, included tractional retinal detachment in the macular region, non-clearing hemorrhage, and premacular subhyaloid hemorrhage. A total of 53 PDR patients (32 men and 21 women) were initially included in the study. However, only 40 patients (15 with a good response and 25 with a poor response) were eventually enrolled. Six patients who had bilateral eye surgery within one month, three diagnosed with retinal vein occlusion during surgery, two diagnosed with glaucoma, and two with type 1 diabetes were excluded from the study. BCVA was measured at each visit, with a good response defined as BCVA ≥ 0.5 and a poor response as BCVA < 0.5 at 6 months post-therapy. Inclusion criteria were an age range of 18–70 years. Exclusion criteria included previous vitreoretinal surgery, photocoagulation in the previous six months, glaucoma, macular hole, any ocular trauma, and other immune systemic diseases. Vitreous samples were collected at the beginning of surgery by a retinal specialist using a 23-gauge standard three-port pars plana vitrectomy. As bleeding may affect the vitreous results. The doctor observed the vitreous condition at the beginning, it is advisable to choose a vitreous body with less blood contamination for sampling. If the blood is too much, the vitreous can not be used. To avoid blood cell contamination, samples were collected from the core vitreous at the surgery’s start. After anesthesia, the eyes were prepared, and trocars were inserted in the standard fashion. Before any surgical maneuvers, the vitrector was used to aspirate a 0.3 ml sample into a tuberculin syringe at 7500 cuts per minute. The vitreous samples were immediately cooled and stored at −80°C [[49]9]. Targeted metabolomics analysis Metabolites were characterized using the Q300 Metabolite Array Kit (Metabo-Profile, Shanghai, China) [[50]10], encompassing amino acids, phenols, organic acids, fatty acids, bile acids, and other small molecules. These compounds are widely recognized as ubiquitous in biological matrices, including cellular material, fecal matter, serum, and tissue samples [[51]11]. Vitreous humor (100 µL) was dispensed into a 96-well plate, freeze-dried, and reconstituted in 20 µL of 50% methanol. Subsequently, 120 µL of ice-cold methanol containing partial internal standards was automatically introduced, followed by vigorous vortexing for 5 min. Post-centrifugation at 4000 g for 30 min, 30 µL of the resulting supernatant underwent derivatization using freshly prepared reagents on the Biomek 4000 platform. This chemical reaction was performed at 30 °C for 1 h. The derivatized samples were diluted with 330 µL of 50% methanol, stored at −20 °C for 20 min, and centrifuged under the same conditions. For quantification, an ultra-performance liquid chromatography-tandem mass spectrometry (UPLC-MS/MS) system (ACQUITY UPLC-Xevo TQ-S, Waters Corp., USA) was utilized. Chromatographic separation employed a BEH C18 column under a gradient of water with 0.1% formic acid (mobile phase A) and acetonitrile/IPA (70:30, mobile phase B), with conditions optimized for reproducibility. The flow rate was set at 0.40 mL/min, and the injection volume was 5 µL. Ionization was achieved using electrospray in both positive and negative modes. Internal standards, including isotopically labeled compounds such as L-Arginine-15N2 and Hippuric acid-D5, alongside pooled quality control samples, ensured methodological accuracy and reliability. Data acquisition and metabolite quantification were conducted using QuanMET software (v2.0, Metabo-Profile, Shanghai, China). The standards used included amino acids, phenols, phenyl or benzyl derivatives, indoles, organic acids, fatty acids, and bile acids, sourced from Sigma-Aldrich, Steraloids Inc., and TRC Chemicals. To ensure the quality of the metabolomics data, two types of mixtures were employed: internal standards (e.g., L-Arginine-15N2, Hippuric acid-D5, Taurochenodeoxycholic acid-D9, D-Glucose-D7, Carnitine-D3, Valeric acid-D9, Citric acid-D4, among others) and pooled quality control (QC) samples. Derivatized QC samples were injected at regular intervals among the study samples. All raw data were processed using UPLC-MS/MS with QuanMET software (v2.0, Metabo-Profile, Shanghai, China) for peak integration, calibration, and quantification of each metabolite. Metabolomic features were annotated by comparison with standard metabolites [[52]10]. Statistical analysis The Chi-square test was used to compare categorical data between the good and poor response groups of PDR (proliferative diabetic retinopathy) patients. A two-tailed t-test was applied to assess differences in clinical data. Principal component analysis (PCA), partial least squares discriminant analysis (PLS-DA), and orthogonal partial least squares discriminant analysis (OPLS-DA) were conducted using iMAP (v1.0, Metabo-Profile, Shanghai, China). Variable importance in projection (VIP) scores were derived from the OPLS-DA model. Metabolites with a VIP score ≥ 1 and a p-value < 0.05, determined through univariate analyses based on the normality of data distribution, were considered statistically significant. Metabolite-associated pathway enrichment analysis was conducted using MetaboAnalyst 4.0 (Xia Lab, McGill University, Montreal, Canada; metaboanalyst.ca) with the Kyoto Encyclopedia of Genes and Genomes (KEGG) database. Pathways or functions with a p-value < 0.05 were deemed impactful. To account for multiple comparisons, p-values were adjusted using the Benjamini-Hochberg method. All statistical analyses were performed with SPSS 21.0 (IBM Corp., Armonk, NY, USA). Results Study population This study included two groups of patients with proliferative diabetic retinopathy (PDR), categorized based on their response to treatment: the good response group and the poor response group. The good response group was defined as patients achieving a best-corrected visual acuity (BCVA) of ≥ 0.5 after vitrectomy, while the poor response group comprised those with a BCVA of < 0.5. Demographic and clinical data are summarized in Table [53]1. The mean age of PDR patients did not differ significantly between the two groups (58.27 ± 9.246 vs59.56 ± 9.465, p = 0 0.844). Similarly, no significant differences were observed in fasting blood glucose (FBG), triglycerides (TG), or urea levels. However, creatinine levels were significantly higher in the poor response group compared to the good response group (108.60 ± 57.48 vs. 89.12 ± 37.83, p = 0.038). In terms of preoperative treatment, 10 patients in the good response group and 21 patients in the poor response group received intraocular anti-VEGF therapy 3–4 days before surgery (p = 0.071). The baseline BCVA was 0.09 ± 0.49 in the good response group and 0.15 ± 0.02 in the poor response group (p = 0.045). Three months post-treatment, the BCVA improved to 0.46 ± 0.23 in the good response group, whereas it remained at 0.07 ± 0.07 in the poor response group (p = 0.017). Table 1. Characteristics of study population for vitreous humor metabolomics Characteristic Good Response Poor Response P[a] value Age*,y 58.27 ± 9.246 59.56 ± 9.465 0.844 Gender^☦ (male/female), n 9/6 16/9 0.800 FBG^✝ (mmol/L) 9.29 ± 4.58 9.70 ± 4.28 0.822 Creatinine^✝(µml/L) 108.60 ± 57.48 89.12 ± 37.83 0.038 Urea^✝ (mmol/L) 8.71 ± 3.55 8.09 ± 2.31 0.213 TC^✝ (mmol/L) 4.66 ± 1.35 4.57 ± 1.15 0.979 TG^✝ (mmol/L) 2.09 ± 0.95 1.83 ± 1.11 0.656 HTN^☦ (yes/no), n 10/5 17/8 0.637 DLD^☦ (yes/no), n 13/2 16/9 0.120 BCVA^✝ [(before−treatment)] 0.09 ± 0.49 0.15 ± 0.02 0.045 BCVA^✝ [(post−treatment)] 0.46 ± 0.23 0.07 ± 0.07 0.017 Treatments, n  Anti-VEGF^☦ (yes/no), n 10/5 21/4 0.071  Insulin^☦ (yes/no), n 9/6 6/19 0.023  Oral medicine^☦ (yes/no), n 9/6 17/8 0.813  Antihypertension^☦ (yes/no), n 8/7 12/13 0.898  VH^☦ (yes/no), n 12/3 22/3 0.493  TRD^☦ (yes/no), n 2/12 9/16 0.148  Phakic (yes/no)^☦ 14/1 23/2 0.877  Macular edema^☦ (yes/no), n 5/10 14/11 0.165  Traction(yes/no), n 6/9 14/11 0.327  Axial length* 23.21 ± 0.95 mm 23.13 ± 0.98 mm 0.767 [54]Open in a new tab * Student’s t-test ☦ Fisher’s precision probability test ✝ Mann-Whitney U test Alterations in vitreous metabolome profiles A total of 150 metabolites were accurately quantified in the samples, including amino acids, organic acids, fatty acids, bile acids, carbohydrates, carnitines, and other metabolites. Global differences in the targeted metabolites between groups were identified using multiple statistical analysis methods. PCA score plots demonstrated tight clustering of the QC samples, confirming that the overall sample analysis process was repeatable and stable. Baseline vitreous metabolite data, collected via mass spectrometry (MS), were analyzed using orthogonal projections to latent structures discriminant analysis (OPLS-DA) six months post-surgery. OPLS-DA, a widely employed method for identifying metabolic profiles that differentiate groups, was utilized to distinguish between the good and poor response groups, as previously reported [[55]12]. The OPLS-DA score plot of vitreous samples from these groups is shown in Fig. [56]1. Clear discrimination between the groups was observed, indicating the reliability of the model (R2X = 0.704, Q2Y = 0.382). The OPLS-DA model demonstrated a good fit and acceptable predictive ability, suggesting that the vitreous metabolomic profile effectively differentiates between good and poor response groups. Fig. 1. [57]Fig. 1 [58]Open in a new tab Orthogonal Partial Least-Squares Discriminant Analysis (OPLS-DA) Score Plots of Vitreous Samples Differentiating Good and Poor Response Groups (Reliability: R^2X = 0.704, Q^2Y = 0.382). The OPLSDA plots showed a clear separation between the good and poor response groups in the vitreous samples Metabolomic study of PDR patients with good or poor responses The metabolites were analyzed and compared between the good and poor response groups, as presented in Fig. [59]2 and the Supplementary Material. A total of 45 metabolites were found to be elevated in the poor response group compared to the good response group, as detailed in Supplementary Material 1. Among these, the top 10 metabolites identified were octanoic acid, dodecanoylcarnitine, tetradecanoylcarnitine, linoleylcarnitine, palmitoylcarnitine, 3-methyl-2-oxopentanoic acid, indolelactic acid, stearylcarnitine, oleic acid, and oleylcarnitine. Heatmaps of these metabolites in the two groups were generated using Pearson’s correlation coefficient analysis, as shown in Fig. [60]2. Fig. 2. [61]Fig. 2 [62]Open in a new tab Distinct Metabolite Profiles Between Good and Poor Response Groups. A Volcano plot comparing therapy response groups: poor versus good response. B Heatmap visualization of the two groups based on 45 metabolites. Rows: biomarkers; Columns: samples. Color key indicates metabolite values, dark blue: lowest; dark red: highest. Red denotes the poor response group, while green and blue indicate the good response group. C Bar plot of the metabolite profiles, highlighting metabolites that are upregulated in the poor response group Enrichment metabolite analysis and metabolic pathway assessment The primary metabolic pathways distinguishing good and poor responses in vitreous samples from PDR patients were identified through enrichment analysis, as depicted in Fig. [63]3. The top three pathways identified were fatty acid biosynthesis, beta-oxidation of very long-chain fatty acids, and mitochondrial beta-oxidation of short-chain saturated fatty acids. Additionally, the predicted pathways, shown in Fig. [64]3, include citrate synthase activity, fatty acid omega-hydroxylation, and the cotransport of 2-hydroxybutyrate with a proton. Fig. 3. [65]Fig. 3 [66]Open in a new tab Enriched metabolic pathway according to the different metabolites between good and poor responses group. A stands for the function enriched between the two groups. B indicates the predicted function pathway between two groups. C stands for the enriched metabolic pathway based on the metabolites study, Fatty Acid Biosynthesis, Beta Oxidation of Very Long Chain Fatty Acids, and Mitochondrial Beta-Oxidation of Short Chain Saturated Fatty Acids were the main pathway in PDR patients Metabolite profiling to predict therapy response outcome To identify markers predictive of therapy outcomes in PDR patients post-vitrectomy, we analyzed candidate metabolites that exhibited significant differences between the good and poor response groups in vitreous samples. Metabolites with the highest area under the AUC values were selected for outcome prediction. The top metabolites with high AUC values, reflecting their predictive capabilities, are summarized in Table [67]2; Fig. [68]4. Among these, dodecanoylcarnitine, linoleylcarnitine, stearylcarnitine, decanoic acid, and proline demonstrated exceptionally high AUC values (0.925, 0.885, 0.864, 0.811, and 0.808, respectively), highlighting their effectiveness in predicting a positive response to vitrectomy. Table 2. Estimated AUC value for prediction of therapy outcome after vitrectomy Significantly metabolites AUC value for prediction of outcome p Dodecanoylcarnitine 0.925 0.004 Linoleylcarnitine 0.885 0.014 Stearylcarnitine 0.864 0.025 Decanoic acid 0.811 0.081 Proline 0.808 0.084 2-Butenoic acid 0.765 0.154 Carnitine 0.731 0.234 Pipecolic acid 0.72 0.297 Malonic acid 0.715 0.297 Propionylcarnitine 0.696 0.368 [69]Open in a new tab Fig. 4. [70]Fig. 4 [71]Open in a new tab Receiver operating characteristic (ROC) curve of 10 top metabolites for prediction of the therapy outcomes poor response vs good response after vitrectomy Discussion Proliferative diabetic retinopathy is a severe complication of diabetes. Conditions such as hemorrhage, premacular subhyaloid hemorrhage, and retinal detachment often necessitate vitrectomy. However, responses to this therapy can vary significantly. Our study analyzed vitreous samples during vitrectomy to differentiate between good and poor response groups. We found that a total of 45 metabolites were elevated in the poor response group compared to the good response group. Additionally, we identified the top 8 metabolites that could predict therapy response. Pathway enrichment analysis revealed that Fatty Acid Biosynthesis, Beta Oxidation of Very Long Chain Fatty Acids, and Mitochondrial Beta-Oxidation of Short Chain Saturated Fatty Acids were the primary pathways altered in PDR patients (poor vs. good response). Specifically, Octanoic acid, Decanoic acid, and Dodecanoic acid, which were associated with Fatty acid biosynthesis, were elevated in the poor response group. Octanoic and Decanoic acids could influence metabolic pathways, resulting in increased ketone body production, and Decanoic acid could stimulate fatty acid synthesis in glioblastoma cells [[72]13]. Our study indicated that the enrichment pathway was focused on fatty acid synthesis in poor response compared to good response group. Increased Decanoic acid could stimulate both fatty acid and cholesterol synthesis [[73]13]. Increased lipogenesis was associated with impaired β-oxidation in diabetic kidney disease among American Indians [[74]13, [75]14], and lead to oxidative stress in diabetic kidney disease [[76]15]. Additionally, Octanoic and Decanoic acids could increase the flux of glucose to lactate [[77]16, [78]17], a process known as the Warburg Effect, which represents an incomplete, non-oxidative metabolism of glucose in oxygen-rich conditions [[79]17]. Warburg effect could induce glycine metabolism-related pathways in human retinal endothelial cells. A higher glycine was reported in vitreous humor with PDR patients. Therefore, Warburg effect was regarded as a biomarker for PDR pathogenesis [[80]18]. In our study, we found increased lactic acid levels in the poor response group than good response group (1026.25 ± 187.42 for good response vs. 1203.06 ± 198.32 for poor response, p < 0.005). Therefore, Impaired β-oxidation and increased Warburg effect were the main metabolic alternations in poor response group which might be the reason for the poor responses. In our study, six amino acids with significant differences were identified: proline, alanine, lysine, aspartic acid, and glutamine, along with seven amino acid metabolic byproducts. Proline is essential for maintaining retinal health, particularly through its metabolism in retinal pigment epithelium (RPE) cells. When proline metabolism is impaired, the RPE’s ability to support photoreceptors by providing nutrients and clearing waste is diminished [[81]19]. In addition, Glutamine can be readily converted into glutamate, and excessive extracellular glutamate concentration in retinal ganglion cells can lead to uncontrolled neuronal depolarization, resulting in excitotoxicity, oxidative stress, and apoptosis [[82]20]. Previous study pointed that fatty acid metabolism was altered in PDR patients than NPDR patients [[83]21].Impaired Fatty Acid oxidation was related to diabetic retinopathy [[84]22]. Lower citrate was positively related to diabetic retinopathy [[85]22]. Citrate synthase is an important enzyme in the mitochondrial membrane which utilized acetyl-CoA and oxaloacetate to form citrate in TCA cycle and electron transport chain [[86]23]. α-Ketoglutarate could be converted to citrate in an acetyl-CoA manner, entering lipid synthesis. It could also generate citrate through oxidative metabolism, forming intermediates such as succinate, fumarate, malate, oxaloacetate, and citrate. However, within functional mitochondria, acetyl-CoA could converge from fatty acids β-oxidation via the pyruvate dehydrogenase complex (PDH) [[87]16]. In this study, increased levels of Octanoic acid and Decanoic acid were detected in the poor response group, suggesting an increase in glucose flux to lactate. Citrate serves not only as a primer for lipid synthesis but also as an allosteric modulator for the lipogenic pathway [[88]24]. However, there was no accurate reason to explain the reason for the poor response. Acylcarnitines (ACs) were the top metabolites that can predict the therapy response. ACs were generated by mitochondrial and peroxisomal enzymes, such as carnitine palmitoyltransferase 1 (CPT1) and CPT2, which transported long-chain fatty acids (LCFAs) across the mitochondrial membrane for β-oxidation. They were intermediate oxidative metabolites of altered acid oxidation [[89]25, [90]26]. The study found increased AC levels were in the poor response group. Simultaneously an increased fatty acids metabolism was detected in the poor response [[91]27]. We speculated that the elevated ACs levels in the poor response group might result from the increased fatty acids metabolism. Fatty Acid Biosynthesis alternations might be the main reason for poor responses. There are also some limitations of the present study. First, the relatively small sample size may limit the statistical power to detect subtle metabolic changes and may affect the generalizability of the results. Future investigations involving larger and independent cohorts are necessary to confirm and extend these findings. Secondly, another limitation of this study is the potential metabolic impact of preoperative anti-VEGF therapy. Anti-VEGF therapy is standard practice for patients undergoing vitrectomy for diabetic retinopathy and was administered to 31 of 40 patients in our cohort. The inclusion of patients who received anti-VEGF injections may potentially influence the results of this study; But no studies based on vitreous samples are currently available. In addition, the use of BCVA as the sole primary outcome measure may be insufficient. Although BCVA is a reliable quantitative parameter, it may not fully reflect the postoperative retinal status. Structural factors such as macular condition, peripheral ischemia, and proliferative vitreoretinopathy (PVR) are closely related to prognosis and should also be considered in outcome analyses. Future studies are warranted to integrate these structural parameters with functional outcomes for a more comprehensive evaluation. Finally, an animal model is also essential to explore the hypothesis and investigate its pathological implications further. While these metabolic markers may be useful in predicting therapy outcomes, the underlying mechanisms warrant further investigation. Conclusion In summary, our study offers a detailed analysis of specific metabolic levels between poor and good responses in PDR patients following vitrectomy. The findings indicate that Dodecanoylcarnitine, Linoleylcarnitine, Stearylcarnitine, Decanoic acid, and Proline exhibit exceptionally high AUC values, making them promising predictors of therapeutic outcomes. This research is a standard medical quantitative study based on targeted mass spectrometry. However, verification through clinical trials is necessary to substantiate these results. Supplementary Information [92]12886_2025_4283_MOESM1_ESM.xlsx^ (13.8KB, xlsx) Supplementary Material 1. Metabolites that differed between the good and poor groups. Acknowledgements