应用Lasso-Logistic回归分析玻璃体切割术(PPV)后增殖型糖尿病视网膜病变(PDR)患者短期预后的影响因素并构建预测模型。

选取2021年1月至2024年2月在临汾市人民医院眼科PPV术后的PDR患者179例(179只眼)作为研究对象。其中，男性96例(96只眼)，女性83例(83只眼)；年龄29~81岁，平均年龄(64.3±8.3)岁。按照术后6个月视力较术前减少程度分为预后不良组和预后良好组。收集PDR患者的病史、血压、血脂、血糖、肝肾功能生化检测结果及术后并发症的信息。年龄、糖尿病病程、术前最佳矫正视力(BCVA)、视网膜激光数量、术前眼内压、血压、空腹血糖、糖化血红蛋白、血脂四项、血常规及肝肾功能生化检测结果进行正态性检验，当符合正态分布用±s表示，组间比较采用独立样本t检验，糖尿病病程和尿白蛋白偏态计量资料采用M(P₂₅, P₇₅)表示，组间比较采用秩和检验；性别、眼别、病史、PDR分期、糖尿病治疗类型、黄斑脱离、虹膜红变、玻璃体积血、视网膜前积血、抗血管内皮生长因子治疗、联合白内障手术、玻璃体腔填充物及术后并发症以例数(眼数)和百分比表示，采用皮尔逊或连续性修正卡方检验；采用单因素分析和Lasso筛选影响PDR患者PPV术后短期预后不良的因素；使用多因素Logistic回归分析独立影响因素，并基于影响因素构建列线图预测模型；拟合度采用霍斯默-莱梅肖检验，C指数分析采用区分能力，准确性采用校准曲线分析，临床效益采用决策曲线分析，预测效能采用受试者工作特征曲线分析。

预后不良组患者33例(33只眼)平均糖尿病病程和尿白蛋白中位和四分位数分别为11.00(9.00，13.50)年和42.37(36.41，48.46)mg/L；高脂血症者13例(13只眼)，冠心病者14例(14只眼)，高血压者6例(6只眼)，分别占39.39%、42.42%及18.18%；PDR Ⅳ期者6例(6只眼)，Ⅴ期者15例(15只眼)，Ⅵ期者12例(12只眼)，分别占18.18%、45.45%及36.36%；口服药物治疗糖尿病者6例(6只眼)，注射胰岛素者17例(17只眼)，胰岛素联合口服药物者10例(10只眼)；黄斑脱离者14例(14只眼)，虹膜红变者4例(4只眼)，玻璃体积血者31例(31只眼)，视网膜前积血者13例(13只眼)，分别占42.42%、12.12%、93.94%及39.39%；未使用抗血管内皮生长因子治疗者11例(11只眼)，术前使用者18例(18只眼)，术中使用者4例(4只眼)，分别占33.33%、54.55%及12.12%；联合白内障手术者17例(17只眼)，占51.52%；玻璃体腔填充六氟乙烷者2例(2只眼)，硅油填充者27例(27只眼)，平衡液填充者4例(4只眼)，分别占6.06%、81.82%及12.12%；术后并发症者18例(18只眼)，占54.55%。预后良好组患者146例(146只眼)糖尿病病程和尿白蛋白中位和四分位数分别为9.00(7.00，11.00)年和41.76(37.10，46.09)mg/L；高脂血症者37例(37只眼)，冠心病者42例(42只眼)，高血压者27例(27只眼)，分别占25.34%、28.77%及18.49%；PDR Ⅳ期者44例(44只眼)，Ⅴ期者85例(85只眼)，Ⅵ期者17例(17只眼)，分别占30.14%、58.22%及11.64%；口服药物治疗糖尿病者37例(37只眼)，注射胰岛素者70例(70只眼)，胰岛素联合口服药物者39例(39只眼)，分别占25.34%、47.95%及26.71%；黄斑脱离者16例(16只眼)，虹膜红变者3例(3只眼)，玻璃体积血者127例(127只眼)，视网膜前积血者52例(52只眼)，分别占10.96%、2.05%、86.99%及35.62%；未使用抗血管内皮生长因子治疗者35例(35只眼)，术前使用者94例(94只眼)，术中使用者17例(17只眼)，分别占23.97%、64.38%及11.65%；联合白内障手术者58例(58只眼)，占39.73%；玻璃体腔填充六氟乙烷者38例(38只眼)，硅油填充者90例(90只眼)，平衡液填充者18例(18只眼)，分别占26.03%、61.64%及12.33%；术后并发症者32例(32只眼)，占21.92%。预后不良组患者术眼术前BCVA、视网膜激光光斑数量、术前眼内压、收缩压、舒张压、空腹血糖、糖化血红蛋白、总胆固醇、甘油三酯、高密度脂蛋白胆固醇、低密度脂蛋白胆固醇、中性粒细胞计数、淋巴细胞计数、血小板计数、白细胞计数、丙氨酸转氨酶、天冬氨酸转氨酶、谷氨酰转肽酶及血肌酐分别为(1.97±0.60)最小分辨角对数(logMAR)、(652.06±381.20)个、(19.12±4.30)mmHg(1 mmHg＝0.133 kPa)、(153.97±20.86)mmHg、(93.85±10.03)mmHg、(10.22±1.41)mmol/L、(7.71±1.72)%、(5.33±0.33)mmol/L、(1.84±0.39)mmol/L、(1.05±0.07)mmol/L、(3.20±0.54)mmol/L、(3.90±1.14)×10⁹/L、(1.60±0.45)×10⁹/L、(254.93±54.14)×10⁹/L、(8.47±2.62)×10⁹/L、(26.92±8.18)U/L、(24.34±8.70)U/L、(36.60±11.12)U/L及(123.39±16.96)μmol/L；预后良好组患者分别为(1.62±0.47)logMAR、(1055.26±433.68)个、(17.40±5.08)mmHg、(149.12±20.14)mmHg、(89.84±11.11)mmHg、(9.71±1.35)mmol/L、(6.99±1.40)%、(5.20±0.51)mmol/L、(1.78±0.34)mmol/L、(1.06±0.13)mmol/L、(3.00±0.55)mmol/L、平均(3.72±1.26)×10⁹/L、(1.76±0.59)×10⁹/L、(254.40±69.00)×10⁹/L、(8.33±2.49)×10⁹/L、(24.80±8.82)U/L、(22.52±7.54)U/L、(33.36±10.23)U/L及(119.65±18.23)μmol/L。两组患者糖尿病病程比较，差异有统计学意义(Z＝-3.316，P<0.05)；黄斑脱离、虹膜红变、玻璃体腔填充物及术后并发症组间比较，差异有统计学意义(χ²＝19.102，4.827，6.479，14.234；P<0.05)；术前BCVA、视网膜激光数量及糖化血红蛋白组间比较，差异有统计学意义(t＝3.667，4.926，2.573；P<0.05)。Lasso回归筛选出PDR分期、术前BCVA、黄斑脱离、视网膜激光数量及术后并发症5个具有非零特征的变量。经多因素Logistic回归分析，PDR分期Ⅴ期、Ⅵ期、黄斑脱离及术后并发症是PDR患者PPV术后短期预后不良的独立危险因素(OR＝3.630，10.296，3.082，3.806，2.920；95%CI：1.088~12.110，2.630~40.303，1.074~8.845，1.260~11.499，1.075~7.930；P<0.05)，视网膜激光数量是其独立保护因素(OR＝0.784，95%CI：0.685~0.897；P<0.05)。基于PDR患者PPV术后短期预后不良的独立影响因素构建列线图预测模型Logit(P)＝-3.233＋1.289×PDR分期Ⅴ期＋2.332×PDR分期Ⅵ期＋1.125×术前BCVA＋1.337×黄斑脱离-0.002×视网膜激光数量＋1.071×术后并发症。预测模型拟合较好，C指数为0.885(95%CI：0.876~0.895，P>0.05)。经校准曲线和决策曲线分析，列线图预测模型预测概率接近理想曲线，净收益范围较大。经受试者工作特征曲线分析，列线图预测模型预测PDR患者PPV术后短期预后不良的曲线下面积0.885(95%CI：0.828~0.927)，敏感度84.85%，特异度78.77%，约登指数0.6362。

PDR分期、术前BCVA、黄斑脱离、视网膜激光数量及术后并发症是PPV术后PDR患者短期预后不良的独立影响因素，视网膜激光数量是其独立保护因素，基于上述影响因素构建的列线图预测模型对PDR患者PPV术后短期预后不良具有较高的预测能效。

The aim of this study is to identify factors influencing the short-term prognosis of patients with proliferative diabetic retinopathy (PDR) after pars plana vitrectomy (PPV) using Lasso-Logistic regression analysis and construct a predictive model.

A total of 179 PDR patients (179 eyes) who underwent PPV in the Ophthalmology Department of Linfen People′s Hospital from January 2021 to February 2024 were selected as the study subjects. There were 83 female (83 eyes) and 96 male (96 eyes) with a mean age of (64.3±8.3) years (ranging from 29 to 81 years). Based on the degree of visual acuity after surgery 6 months decline compared to preoperative levels, patients were divided into the poor prognosis group with 33 cases (33 eyes) and the favorable prognosis group with 146 cases (146 eyes). Data collected included gender, age, eye involved, diabetes duration, medical history, PDR stage, diabetes treatment type, preoperative best-corrected visual acuity (BCVA), disease severity, treatment modality, intraocular pressure (IOP), blood pressure, blood glucose, blood lipids, blood cells, liver and kidney biochemistry, and postoperative complications. The age, diabetes duration, preoperative BCVA, number of retinal laser photocoagulations, preoperative IOP, blood pressure, fasting blood glucose, glycated hemoglobin (HbA1c), four lipid items, complete blood count, liver and kidney function indices. Data conforming to a normal distribution were expressed as ±s and compared between groups using independent samples t-test. Skewed measurement data including the diabetes duration and urinary albumin were expressed as M(P₂₅, P₇₅) and compared between groups using the rank-sum test. The gender, eye involved, medical history, PDR stage, diabetes treatment type, macular detachment, iris neovascularization/rubeosis, vitreous hemorrhage, preretinal hemorrhage, anti-vascular endothelial growth factor (anti-VEGF) treatment, combined cataract surgery, vitreous cavity tamponade, postoperative complications were expressed as number and percentage, and compared between groups using Pearson′s or continuity-adjusted chi-square test. Univariate analysis and Lasso regression were used to screen factors affecting short-term poor prognosis after PPV in PDR patients. Independent influencing factors were identified using multivariate Logistic regression analysis, and a nomogram prediction model was constructed based on these factors. The Hosmer-Lemeshow test was used for goodness-of-fit; the C-index was used to analyze discriminative ability; the accuracy was analyzed using calibration curves; the clinical utility was analyzed using decision curve analysis, and predictive efficacy was analyzed using receiver operating characteristic (ROC) curve analysis.

In the poor prognosis group with 33 cases (33 eyes), the diabetes duration median and urinary albumin were 11.00 (9.00, 13.50) years and 42.37 (36.41, 48.46) mg/L, respectively; there were 13 with hyperlipidemia, 14 with coronary heart disease, 6 with hypertension, accounting for 39.39%, 42.42% and 18.18%, respectively; 6 cases (6 eyes) with PDR stage Ⅳ, 15 cases (15 eyes) with stage Ⅴ, 12 cases (12 eyes) with stage Ⅵ, accounting for 18.18%, 45.45% and 36.36%, respectively; 6 with oral medications of diabetes treatment type, 17 with insulin injections, 10 with insulin combined with oral medications, accounting for 18.18%, 51.52% and 30.30%, respectively; 14 with macular detachment, 4 with iris rubeosis, 31 with vitreous hemorrhage, 13 with preretinal hemorrhage, accounting for 42.42%, 12.12%, 93.94% and 39.39% respectively; 11 not used anti-VEGF treatment, 18 used preoperativly, 4 used intraoperativly, accounting for 33.33%, 54.55% and 12.12% respectively; 17 combined with cataract surgery, accounting for 51.52%; 2 with C3F8 for vitreous cavity tamponade, 27 with silicone oil, 4 with balanced saline solution, accounting for 6.06%, 81.82% and 12.12% respectively; 18 with postoperative complications accounting for 54.55%. In the good prognosis groups with 146 cases (146 eyes), the diabetes duration median were 9.00 (7.00, 11.00) years; urinary albumin 41.76 (37.10, 46.09) mg/L; 37 with hyperlipidemia, 42 with coronary heart disease, 27 with hypertension, accounting for 25.34%, 28.77% and 18.49% respectively; PDR stage: 44 cases (44 eyes) with Ⅳ stage, 85 cases (85 eyes) with Ⅴ stage, 17 cases (17 eyes) with Ⅵ stage, accounting for 30.14%, 58.22% and 11.64% respectively; diabetes treatment type: 37 used by oral medications, 70 used by insulin injections, 39 used by insulin combined with oral medications, accounting for 25.34%, 47.95% and 26.71% respectively; 16 with macular detachment, 3 with iris rubeosis, 127 with vitreous hemorrhage, 52 with preretinal hemorrhage, accounting for 10.96%, 2.05%, 86.99% and 35.62% respectively; 35 not used anti-VEGF treatment, 94 preoperative, 17 intraoperative, accounting for 23.97%, 64.38% and 11.65% respectively; 58 with combined cataract surgery, accounting for 39.73%; 38 with C3F8 after vitreous cavity tamponade, 90 with silicone oil, 18 with balanced saline solution, accounting for 90.41%, 61.64% and 12.33% respectively; 32 with postoperative complications, accounting for 21.92%. The mean preoperative BCVA, number of retinal laser spots, preoperative IOP, systolic blood pressure, diastolic blood pressure, fasting blood glucose, HbA1c, total cholesterol, triglycerides, high density lipoprotein, low density lipoprotein, neutrophil count, lymphocyte count, platelet count, white blood cell count, alanine aminotransferase, aspartate transaminase, glutamyl transpeptidase, serum creatinine were (1.97±0.60) logarithm of the minimum angle of resolution (logMAR), (652.06±381.20)spots, (19.12±4.30)mmHg (1 mmHg=0.133 kPa), (153.97±20.86)mmHg, (93.85±10.03)mmHg, (10.22±1.41)mmol/L, (7.71±1.72)%, (5.33±0.33)mmol/L, (1.84±0.39)mmol/L, (1.05±0.07)mmol/L, (3.20±0.54)mmol/L, (3.90±1.14) ×10⁹/L, (1.60±0.45) ×10⁹/L, (254.93±54.14) ×10⁹/L, (8.47±2.62) ×10⁹/L, (26.92±8.18)U/L, (24.34±8.70)U/L, (36.60±11.12)U/L, (123.39±16.96)μmol/L, respectively; while those of good prognosis group were (1.62±0.47) logMAR, (1055.26±433.68) spots, (17.40±5.08) mmHg, (149.12±20.14) mmHg, (89.84±11.11) mmHg, (9.71±1.35) mmol/L, (6.99±1.40)%, (5.20±0.51) mmol/L, (1.78±0.34) mmol/L, (1.06±0.13) mmol/L, (3.00±0.55) mmol/L, (3.72±1.26) ×10⁹/L, (1.76±0.59) ×10⁹/L, (254.40±69.00) ×10⁹/L, (8.33±2.49) ×10⁹/L, (24.80±8.82) U/L, (22.52±7.54) U/L, (33.36±10.23) U/L, (119.65±18.23) μmol/L, respectively. There was a statistically significant difference in diabetes duration(Z=-3.316, P<0.05); macular detachment, iris rubeosis, vitreous cavity tamponade, and postoperative complications (χ²=19.102, 4.827, 6.479, 14.234; P<0.05); preoperative BCVA, number of retinal laser spots, and HbA1c (t=3.667, 4.926, 2.573; P<0.05). Lasso regression identified five variables with non-zero features: PDR stage, preoperative BCVA, macular detachment, number of retinal lasers, and postoperative complications. Multivariate logistic regression analysis showed that PDR Ⅴ, Ⅵ stages, preoperative BCVA, macular detachment, and postoperative complications were independent risk factors for short-term poor prognosis in PDR patients after PPV (OR=3.630, 10.296, 3.082, 3.806, 2.920; 95%CI: 1.088 to 12.110, 2.630 to 40.303, 1.074 to 8.845, 1.260 to 11.499, 1.075 to 7.930; P<0.05), while the number of retinal lasers was an independent protective factor (OR=0.784, 95%CI: 0.685~0.897; P<0.05). A column chart prediction model was constructed based on independent influencing factors of short-term poor prognosis in PDR patients undergoing PPV Logit (P)=-3.233+ 1.289×PDR stage Ⅴ+ 2.332×PDR stage Ⅵ+ 1.125×preoperative BCVA+ 1.337×macular detachment -0.002×number of retinal lasers+ 1.071×postoperative complications. The prediction model was well, and C index of 0.885 (95%CI: 0.876 to 0.895, P>0.05). The predicted probability of the column chart prediction model after the calibration curve and decision curve was close to the ideal curve, and the net profit range was relatively large. ROC curve analysis showed that the area under the curve of the line chart prediction model for predicting short-term poor prognosis in PDR patients after PPV was 0.885 (95%CI: 0.828 to 0.927), with a sensitivity of 84.85%, specificity of 78.77%, and a Youden index of 0.6362.

PDR staging, preoperative BCVA, macular detachment, number of retinal lasers, and postoperative complications are independent factors affecting the short-term poor prognosis of PDR patients after PPV. The column chart prediction model constructed based on these factors has high predictive efficiency for the short-term poor prognosis of PDR patients after PPV.