骨科手术患者深度学习血栓形成风险模型构建与决策曲线分析

沈云霞; 姚国美; 傅悦渊; 陆升华; 高志朝

doi:10.16766/j.cnki.issn.1674-4152.003793

骨科手术患者深度学习血栓形成风险模型构建与决策曲线分析

doi: 10.16766/j.cnki.issn.1674-4152.003793

1.
浙江大学医学院附属第二医院临平院区手术室，浙江杭州 311100
2.
浙江大学医学院附属第二医院临平院区骨科

基金项目:

浙江省医药卫生科技计划项目 2019KY550

详细信息

通讯作者:
高志朝，E-mail：leadyourdream@163.com

中图分类号: R687.3
计量
- 文章访问数: 8
- HTML全文浏览量: 8
- PDF下载量: 2
- 被引次数: 0
出版历程
- 收稿日期: 2024-03-05
- 网络出版日期: 2025-01-20

Risk model construction and decision curve analysis of deep learning thrombosis in orthopedic surgery patients

1.
Department of Operating Room, Linping District, the Second Affiliated Hospital of Zhejiang University School

摘要

摘要: 目的构建骨科手术患者深度学习血栓形成风险模型，并选用决策曲线分析其临床效能。方法回顾性选取2022年2月—2024年2月浙江大学医学院附属第二医院临平院区收治的180例骨科手术患者，根据7∶3比例将其划分为训练集(126例)和验证集(54例)；根据训练集内患者深静脉血栓(DVT)形成与否进一步分为DVT发生组(32例)与DVT未发生组(94例)。通过Python软件构建骨科手术患者DVT风险预测人工神经网络(ANN)模型，绘制决策曲线分析模型的临床效能。结果训练集内DVT发生组年龄、BMI、手术时间及卧床时间均高于DVT未发生组，糖尿病、高血压、脊柱手术史、病情分布(下肢损伤)、全麻占比均高于DVT未发生组(P < 0.05)。年龄大、BMI高、合并糖尿病、高血压、病情分布(下肢损伤)和卧床时间＞5 d均为骨科手术患者发生DVT的独立危险因素(P < 0.05)。训练集、验证集的AUC分别为0.887、0.903。训练集与验证集分别取阈值概率18%~56%、19%~58%时，对骨科手术患者采取有效干预措施可使临床效益最大化。结论年龄大、BMI高、合并糖尿病、高血压、病情分布(下肢损伤)和卧床时间＞5 d是骨科手术患者发生DVT的独立危险因素，由这些影响因素构建的ANN模型对骨科手术患者DVT风险预测效能显著，有助于DVT防治工作的临床效益最大化。
- 骨科手术 /
- 深度学习 /
- 人工神经网络 /
- 深静脉血栓形成 /
- 决策曲线 /
- 模型
Abstract: Objective To establish a deep learning thrombosis risk model for patients undergoing orthopaedic surgery, and to analyze its clinical efficacy by using a decision curve. Methods A total of 180 orthopaedic patients admitted to Linping Campus in School of Medicine, the Second Affiliated Hospital of Zhejiang University from February 2022 to February 2024 were retrospectively selected and divided into training group (n=126) and verification group (n=54) according to the ratio of 7∶3. The patients were subdivided into two groups on the basis of the results of the deep vein thrombosis (DVT) examination. A predictive model for DVT risk in orthopaedic surgery patients was developed using Python software. The clinical efficacy of the model was evaluated through the construction of a decision curve. Results The training group, the age, BMI, operation time and bed rest time of the group with DVT were higher than those of the group without DVT. Furthermore, the proportions of diabetes, hypertension, spinal surgery history, lower limb injury and general anaesthesia were higher than those of the group without DVT (P < 0.05). The independent risk factors for DVT in orthopaedic surgery patients, as identified through statistical analysis, were age, BMI, diabetes, hypertension, disease distribution (lower limb injury) and bed rest time exceeding five days (P < 0.05). The AUC for the training set and verification set is 0.887 and 0.903, respectively. When the threshold probabilities of the training set and verification set are 18%-56% and 19%-58% respectively, the implementation of effective intervention measures can facilitate the optimal clinical benefits for patients undergoing orthopaedic surgery. Conclusion The following factors have been identified as independent risk factors for DVT in patients undergoing orthopaedic surgery: age, BMI, diabetes, hypertension, disease distribution (lower limb injury) and bed rest time exceeding five days. The ANN model constructed using these influencing factors is an effective method for predicting the risk of DVT in orthopedic surgery patients. This is beneficial in order to ensure that the clinical benefits of DVT prevention and treatment are optimized.
- Orthopedic surgery /
- Deep learning /
- Artificial neural network /
- Deep vein thrombosis /
- Decision curve /
- Model

HTML全文

图 1 训练集与验证集骨科手术患者DVT风险的Kaplan-Meier曲线分析

Figure 1. Kaplan-Meier curve analysis of DVT risk of orthopedic surgery patients in training set and verification set

下载: 全尺寸图片幻灯片

图 2 骨科手术患者DVT风险预测ANN模型构建示意图

Figure 2. Schematic diagram of ANN model for DVT risk prediction of orthopedic surgery patients

下载: 全尺寸图片幻灯片

图 3 ANN模型预测骨科手术患者DVT发生的ROC曲线

Figure 3. ROC curve of predicting DVT in orthopedic surgery patients by ANN model

下载: 全尺寸图片幻灯片

表 1 训练集与验证集骨科手术患者人口学及临床资料比较

Table 1. Comparison of demographic and clinical data of orthopedic surgery patients in training set and verification set

项目	训练集(n=126)	验证集(n=54)	统计量	P值	项目	训练集(n=126)	验证集(n=54)	统计量	P值
性别[例(%)]			0.086^a	0.769	术前PLT[例(%)]			0.164^a	0.685
男性	67(53.17)	30(55.56)			＜240×10⁹/L	81(64.29)	33(61.11)
女性	59(46.83)	24(44.44)			≥240×10⁹/L	45(35.71)	21(38.89)
年龄(x±s, 岁)	62.57±8.04	62.38±7.95	0.146^b	0.884	术前RBC[例(%)]			0.635^a	0.426
BMI(x±s)	24.25±2.13	24.16±2.18	0.256^b	0.799	＜4.5×10¹²/L	78(61.90)	30(55.56)
文化程度[例(%)]			0.267^a	0.790	≥4.5×10¹²/L	48(38.10)	24(44.44)
小学及以下	15(11.91)	8(14.81)			术前D-D[例(%)]			0.070^a	0.791
初中及高中	42(33.33)	17(31.48)			＜0.5 mg/L	75(59.52)	31(57.41)
大专及以上	69(54.76)	29(53.71)			≥0.5 mg/L	51(40.48)	23(42.59)
吸烟史[例(%)]			0.046^a	0.830	手术时间[例(%)]			2.854^a	0.091
有	37(29.37)	15(27.78)			≤4 h	80(63.49)	27(50.00)
无	89(70.63)	39(72.22)			＞4 h	46(36.51)	27(50.00)
饮酒史[例(%)]			0.059^a	0.808	麻醉方式[例(%)]			0.088^a	0.767
有	42(33.33)	17(31.48)			局部麻醉	53(42.06)	24(44.44)
无	84(66.67)	37(68.52)			全身麻醉	73(57.94)	30(55.56)
基础疾病[例(%)]					止血带使用情况[例(%)]			3.320^a	0.068
糖尿病	17(13.49)	6(11.11)	0.192^a	0.661	使用	70(55.56)	22(40.41)
高血压	14(11.11)	7(12.96)	0.126^a	0.723	未使用	56(44.44)	32(59.26)
既往史[例(%)]					卧床时间[例(%)]			0.309^a	0.578
骨科手术	13(10.32)	4(7.41)	0.374^a	0.541	≤5 d	71(56.35)	28(51.85)
脊柱手术	12(9.52)	5(9.26)	0.003^a	0.956	＞5 d	55(43.65)	26(48.15)
病情分布[例(%)]			0.670^a	0.873
上肢损伤	24(19.05)	10(18.52)
颈胸腰椎损伤	39(30.95)	15(27.78)
下肢损伤	33(26.19)	13(24.07)
多发性损伤	30(23.81)	16(29.63)
注：^a为χ²值，^b为t值，^c为Z值。

下载: 导出CSV

表 2 DVT发生组与DVT未发生组骨科手术患者人口学与临床资料比较

Table 2. Comparison of demographic and clinical data between DVT-occurrence group and DVT-non-occurrence group in

项目	DVT发生组 (n=32)	DVT未发生组 (n=94)	统计量	P值	项目	DVT发生组 (n=32)	DVT未发生组 (n=94)	统计量	P值
性别[例(%)]			1.498^a	0.221	术前PLT[例(%)]			1.206^a	0.272
男性	20(62.50)	47(50.00)			＜240×10⁹/L	18(56.25)	63(67.02)
女性	12(37.50)	47(50.00)			≥240×10⁹/L	14(43.75)	31(32.98)
年龄(x±s, 岁)	65.03±9.12	58.65±8.67	3.461^b	0.001	术前RBC[例(%)]			0.252^a	0.616
BMI(x±s)	25.30±2.04	22.08±2.14	7.616^b	< 0.001	＜4.5×10¹²/L	21(65.62)	57(60.64)
文化程度[例(%)]			1.441^c	0.150	≥4.5×10¹²/L	11(34.38)	37(39.36)
小学及以下	5(15.63)	10(10.64)			术前D-D[例(%)]			0.191^a	0.662
初中及高中	13(40.63)	29(30.85)			＜0.5 mg/L	18(56.25)	57(60.64)
大专及以上	14(43.75)	55(58.51)			≥0.5 mg/L	14(43.75)	37(39.36)
吸烟史[例(%)]			1.369^a	0.242	手术时间[例(%)]			15.688^a	< 0.001
有	12(37.50)	25(26.60)			≤2 h	11(34.37)	69(73.40)
无	20(62.50)	69(73.40)			＞2 h	21(65.63)	25(26.60)
饮酒史[例(%)]			2.094^a	0.148	麻醉方式[例(%)]			5.125^a	0.024
有	14(43.75)	28(29.79)			局部麻醉	8(25.00)	45(47.87)
无	18(56.25)	66(70.21)			全身麻醉	24(75.00)	49(52.13)
基础疾病[例(%)]					止血带使用情况[例(%)]			0.253^a	0.615
糖尿病	11(34.38)	6(6.38)	16.027^a	< 0.001	使用	19(59.37)	51(54.26)
高血压	10(31.25)	4(4.26)	17.614^a	< 0.001	未使用	13(40.63)	43(45.74)
既往手术史[例(%)]					卧床时间[例(%)]			13.892^a	< 0.001
骨科手术	3(9.38)	10(10.64)	0.041^a	0.839	≤5 d	9(28.13)	62(65.96)
脊柱手术	8(25.00)	4(4.26)	11.923^a	0.001	＞5 d	23(71.87)	32(34.04)
病情分布[例(%)]			16.337^a	< 0.001
上肢损伤	3(9.38)	21(22.34)
颈胸腰椎损伤	7(21.88)	32(34.04)
下肢损伤	17(53.12)	16(17.02)
多发性损伤	5(15.62)	25(26.60)
注：^a为χ²值，^b为t值，^c为Z值。

下载: 导出CSV

表 3 变量赋值情况

Table 3. Variable assignment

变量	赋值方法
DVT发生	发生=1；未发生=0
年龄	以实际值赋值
BMI	以实际值赋值
糖尿病	是=1；否=0
高血压	是=1；否=0
脊柱手术	是=1；否=0
病情分布	多发性损伤=1，0，0；颈胸腰椎损伤=1，0，0；下肢损伤=1，0，0；上肢损伤=1，0，0
手术时间	＞2 h=1；≤2 h=0
麻醉方式	全身麻醉=1；局部麻醉=0
卧床时间	＞5 d=1；≤5 d=0

下载: 导出CSV

表 4 训练集内骨科手术患者DVT发生影响因素的多因素logistic回归系数分析

Table 4. Multivariate logistic regression coefficient analysis of influencing factors of DVT in orthopedic surgery patients in training set

变量	B	SE	Waldχ²	P值	OR值	95% CI
年龄	0.096	0.042	5.341	0.021	1.101	1.015~1.195
BMI	0.472	0.165	8.147	0.004	1.603	1.159~2.217
糖尿病	2.734	1.026	7.097	0.008	15.392	2.060~25.028
高血压	0.992	1.025	0.935	0.034	2.696	1.361~20.116
脊柱手术	2.084	1.165	3.200	0.074	8.034	0.819~78.779
病情分布			8.239	0.041
多发性损伤	1.096	1.299	0.712	0.399	2.991	0.235~38.149
颈胸腰椎损伤	1.171	1.320	0.787	0.375	3.226	0.243~42.870
下肢损伤	3.805	1.493	6.498	0.011	44.936	2.410~97.981
手术时间	1.291	0.753	2.941	0.086	3.636	0.832~15.892
麻醉方式	0.056	0.804	0.005	0.945	1.057	0.219~5.112
卧床时间	2.823	1.024	7.598	0.006	16.820	2.260~25.156

下载: 导出CSV

参考文献(22)

[1]	LIM H J, JEON J Y, JEONG Y M, et al. Triaging deep vein thrombosis using ultrasonography after lower-extremity orthopedic surgery: analysis of a single-center experience[J]. Ultrasonography, 2021, 40(3): 442-448. doi: 10.14366/usg.20171
[2]	杨涛, 王文辉, 姜涛. "筋骨并重"康复理论在Pilon患者术后应用及对深静脉血栓的预防效果[J]. 中华全科医学, 2022, 20(2): 311-314. doi: 10.16766/j.cnki.issn.1674-4152.002340 YANG T, WANG W H, JIANG T. The application of "paying equal attention to both bones and muscles" rehabilitation theory in Pilon patients after operation and its preventive effect on deep vein thrombosis[J]. Chinese Journal of General Practice, 2022, 20(2): 311-314. doi: 10.16766/j.cnki.issn.1674-4152.002340
[3]	耿宜岚. 骨科手术住院病人下肢深静脉血栓形成的影响因素及临床干预[J]. 全科护理, 2020, 18(29): 4049-4051. doi: 10.12104/j.issn.1674-4748.2020.29.043 GENG Y L. Influencing factors and clinical intervention of deep venous thrombosis of lower limbs in inpatients undergoing orthopedic surgery[J]. Chinese General Practice Nursing, 2020, 18(29): 4049-4051. doi: 10.12104/j.issn.1674-4748.2020.29.043
[4]	杜治国, 何龙, 黄进, 等. 下肢创伤骨折后静脉血栓栓塞症相关检测项目分析[J]. 医学信息, 2019, 32(10): 64-66. DU Z G, HE L, HUANG J, et al. Analysis of related detection items of venous thromboembolism after traumatic fracture of lower limbs[J]. Medical Information, 2019, 32(10): 64-66.
[5]	YUEN H, TRAN H, CHUNILAL S. Upper extremity deep vein thrombosis: current knowledge and future directions[J]. Semin Thromb Hemost, 2021, 47(6): 677-691. doi: 10.1055/s-0041-1725116
[6]	YU J, YAN X. Whole process monitoring based on unstable neuron output information in hidden layers of deep belief network[J]. IEEE Trans Cybern, 2020, 50(9): 3998-4007. doi: 10.1109/TCYB.2019.2948202
[7]	SECHOPOULOS I, TEUWEN J, MANN R. Artificial intelligence for breast cancer detection in mammography and digital breast tomosynthesis: state of the art[J]. Semin Cancer Biol, 2021, 72(1): 214-225.
[8]	BYEON S J, PARK J, CHO Y A, et al. Automated histological classification for digital pathology images of colonoscopy specimen via deep learning[J]. Sci Rep, 2022, 12(1): 12804. DOI: 10.1038/s41598-022-16885-x.
[9]	KHOURY H, LYONS R, SANAIHA Y, et al. Deep venous thrombosis and pulmonary embolism in cardiac surgical patients[J]. Ann Thorac Surg, 2020, 109(6): 1804-1810. doi: 10.1016/j.athoracsur.2019.09.055
[10]	蒋伟兴, 丁晓娟, 王萍, 等. 骨科手术患者深静脉血栓的危险因素及改良Padua评分的预测价值[J]. 血管与腔内血管外科杂志, 2023, 9(6): 687-690, 700. JIANG W X, DING X J, WANG P, et al. Risk factors of deep venous thrombosis in orthopedic surgery patients and predictive value of improved Padua score[J]. Journal of Vascular and Endovascular Surgery, 2023, 9(6): 687-690, 700.
[11]	JANG H J, CHO K O. Applications of deep learning for the analysis of medical data[J]. Arch Pharm Res, 2019, 42(6): 492-504. doi: 10.1007/s12272-019-01162-9
[12]	PARK C W, SEO S W, KANG N, et al. Artificial intelligence in health care: current applications and issues[J]. J Korean Med Sci, 2020, 35(42): e379. DOI: 10.3346/jkms.2020.35.e379.
[13]	张鑫宇, 张磊, 隋汝波. 基于Logistic回归和人工神经网络构建老年脑卒中患者衰弱预测模型[J]. 军事护理, 2023, 40(2): 10-14, 19. ZHANG X Y, ZHANG L, SUI R B. Building a prediction model of senile stroke patients based on logistic regression and artificial neural network[J]. Military Nursing, 2023, 40(2): 10-14, 19.
[14]	肖雅, 王漱阳, 凌人, 等. 人工神经网络算法在消化道恶性肿瘤病理诊断及患者预后预测中的应用[J]. 浙江大学学报(医学版), 2023, 52(2): 243-248. XIAO Y, WANG S Y, LING R, et al. Application of artificial neural network algorithm in pathological diagnosis and prognosis prediction of digestive tract malignant tumor[J]. Journal of Zhejiang University(Medical Sciences), 2023, 52(2): 243-248.
[15]	HAO W, LIU X, FENG J, et al. Age-adjusted D-dimer thresholds combined with the modified wells score as a predictor of lower extremity deep venous thrombosis[J]. Clin Appl Thromb Hemost, 2023, 29(1): 10760296231221141. DOI: 10.1177/10760296231221141.
[16]	王志文, 尹宗生. 代谢综合征与全膝关节置换术后深静脉血栓形成相关性研究[J]. 实用骨科杂志, 2022, 28(12): 1074-1078. WANG Z W, YIN Z S. Correlation between metabolic syndrome and deep vein thrombosis after total knee arthroplasty[J]. Journal of Practical Orthopaedics, 2022, 28(12): 1074-1078.
[17]	JIAO X, LI Z, AN S, et al. Does diabetes mellitus increase the incidence of early thrombosis in deep vein following unicompartmental knee arthroplasty: a retrospective cohort study[J]. BMC Geriatr, 2022, 22(1): 448. DOI: 10.1186/s12877-022-03153-w.
[18]	耿倩, 李天华, 鹿青, 等. 骨科住院患者下肢深静脉血栓形成的发生率及危险因素分析[J]. 护士进修杂志, 2021, 36(9): 817-820. GENG Q, LI T H, LU Q, et al. Analysis of the incidence and risk factors of deep venous thrombosis of lower limbs in orthopedic inpatients[J]. Journal of Nurses Training, 2021, 36(9): 817-820.
[19]	QUE S J, CHEN Q Y, ZHONG Q, et al. Application of preoperative artificial neural network based on blood biomarkers and clinicopathological parameters for predicting long-term survival of patients with gastric cancer[J]. World J Gastroenterol, 2019, 25(43): 6451-6464. doi: 10.3748/wjg.v25.i43.6451
[20]	KUWAYAMA N, HOSHINO I, MORI Y, et al. Applying artificial intelligence using routine clinical data for preoperative diagnosis and prognosis evaluation of gastric cancer[J]. Oncol Lett, 2023, 26(5): 499. DOI: 10.3892/ol.2023.14087.
[21]	薛震, 陆俊, 林嘉, 等. Ⅱ~Ⅲ期胃癌患者淋巴结转移的人工神经网络预测模型构建[J]. 中华胃肠外科杂志, 2022, 25(4): 327-335. doi: 10.3760/cma.j.cn441530-20220105-00010 XUE Z, LU J, LIN J, et al. Construction of artificial neural network prediction model for lymph node metastasis in patients with stage Ⅱ~Ⅲ gastric cancer[J]. Chinese Journal of Gastrointestinal Surgery, 2022, 25(4): 327-335. doi: 10.3760/cma.j.cn441530-20220105-00010
[22]	武小娟, 孟舰, 刘红新, 等. 改良急诊早期预警评分用于急诊分诊的决策曲线分析[J]. 中国急救复苏与灾害医学杂志, 2023, 18(7): 910-913. doi: 10.3969/j.issn.1673-6966.2023.07.017 WU X J, MENG J, LIU H X, et al. Decision curve analysis of improved early warning score for emergency triage[J]. China Journal of Emergency Resuscitation and Disaster Medicine, 2023, 18(7): 910-913. doi: 10.3969/j.issn.1673-6966.2023.07.017