Comment on “Construct validation of machine learning models for predicting surgical site infection risk following ankle fracture surgery”

Abstract

Dear Editor, Ankle fractures account for 10%–15% of all fracture cases, making them a prevalent injury in lower limb trauma[1,2]. Ankle fractures appear to be more common in recent years as a result of fragility fractures caused by aging. The accepted approach to treat ankle fractures with positive clinical outcomes is anatomical reduction, early functional rehabilitation and solid fixation[3]. However, the skin around the ankle and foot is thin and lacks muscle, and one of the main postoperative complications that orthopedic surgeons deal with is surgical site infection (SSI). According to reports, the postoperative infection risk for ankle fractures ranges from 3% to 12%. The growing volume and complexity of biological data have led to the widespread use of machine learning, a new analytical tool, in a variety of domains. Predicting postoperative outcomes using machine learning has shown promising potential. In order to create and test a prediction classification model for SSI risk, Zhang et al[4] conducted a retrospective investigation using cutting-edge machine learning techniques. There were 509 patients in all (Table 1). The results of this study showed that the Gradient Boosting Machine model outperforms the conventional logistic regression model and that contemporary machine learning techniques provide good prediction accuracy for SSI risk. Table 1 - Characteristics of patients in the SSI group and non-SSI group. Total Non-SSI SSI P-value Number of patients 509 458 51 Age (years) < 65 384 (75.4) 347 (75.8) 111 (24.2) 0.732 ≥ 65 125 (24.6) 37 (72.5) 14 (27.5) Gender (%) Female 244 (47.9) 220 (48.0) 24 (47.1) 0.895 Male 265 (52.1) 238 (52.0) 27 (52.9) Total underlying diseases (%) < 3 489 (96.1) 443 (96.7) 46 (90.2) 0.040 ≥ 3 20 (3.9) 5 (3.3) 5 (9.8) BMI (%) < 18.5 11 (2.2) 10 (2.2) 1 (2.0) 0.844 18.5–24.9 297 (58.3) 267 (58.3) 30 (58.8) 25–29.9 175 (34.4) 159 (34.7) 16 (31.4) ≥ 30 26 (5.1) 22 (4.8) 4 (7.8) Stroke (%) No 488 (95.9) 444 (96.9) 44 (86.3) 0.003 Yes 21 (4.1) 14 (3.1) 7 (13.7) Diabetes (%) No 438 (86.1) 405 (88.4) 33 (64.7) <0.001 Yes 71 (13.9) 53 (11.6) 18 (35.3) Hypertension (%) No 407 (80.0) 377 (82.3) 30 (58.8) <0.001 Yes 102 (20.0) 81 (17.7) 21 (41.2) OCS (%) No 299 (58.7) 275 (60.0) 24 (47.1) 0.098 Yes 210 (41.3) 183 (40.0) 27 (52.9) Trauma type (%) Close fracture 423 (83.1) 399 (87.1) 24 (47.1) <0.001 Open fracture 86 (16.9) 59 (12.9) 27 (52.9) ASA (%) I–II 440 (86.4) 402 (87.8) 38 (74.5) 0.012 III–V 69 (13.6) 56 (12.2) 13 (25.5) Hospital stays (%) < 17 277 (54.4) 258 (56.3) 19 (37.3) 0.011 ≥ 17 232 (45.6) 200 (43.7) 32 (62.7) Operation time (%) < 100 305 (59.9) 282 (61.6) 23 (45.1) 0.025 ≥ 100 204 (40.1) 176 (38.4) 28 (54.9) Blood loss (%) < 100 248 (48.7) 225 (49.1) 23 (45.1) 0.658 ≥ 100 261 (51.3) 223 (50.9) 28 (54.9) Osteoporosis (%) No 331 (65.0) 299 (65.3) 32 (62.7) 0.758 Yes 178 (35.0) 159 (34.7) 19 (37.3) Bone grafting (%) No 335 (65.8) 302 (65.9) 33 (64.7) 0.877 Yes 174 (34.2) 156 (34.1) 18 (35.3) Injury-to-surgery time (%) < 4 266 (52.3) 254 (55.5) 12 (23.5) <0.001 ≥ 4 243 (47.7) 204 (44.5) 39 (76.5) Drainage duration (%) < 3 219 (43.0) 201 (43.9) 18 (35.3) 0.297 ≥ 3 290 (57.0) 257 (56.1) 33 (64.7) Pre-HGB (g/L) 125.6 (18.7) 126.5 (17.8) 117.9 (24.2) 0.002 Pre-RBC (*109/L) 4.1 (0.6) 4.2 (0.6) 4.1 (0.8) 0.562 Pre-WBC (*109/L) 8.4 (3.1) 8.3 (2.9) 9.1 (4.2) 0.098 Pre-PLT (*109/L) 191.1 (65.2) 190.8 (64.3) 193.5 (73.6) 0.780 Pre-LYMPH (%) 19.8 (8.4) 19.9 (8.2) 18.5 (9.7) 0.245 Pre-APTT (s) 29.5 (4.2) 29.5 (4.3) 29.3 (3.6) 0.712 Pre-Fbg (g/L) 3.0 (0.9) 2.9 (0.8) 3.2 (1.1) 0.045 Pre-PT (s) 12.1 (0.9) 12.1 (1.0) 12.1 (0.9) 0.824 Pre-TBIL (µmmol/L) 15.9 (7.6) 15.8 (7.5) 16.6 (9.2) 0.495 Pre-TP (g/L) 65.5 (5.6) 66.1 (5.2) 60.8 (6.8) <0.001 Pre-ALB (g/L) 38.4 (3.7) 38.9 (3.2) 33.6 (3.9) <0.001 APTT, activated partial thromboplastin time; ALB, albumin; ASA, American Standards Association; BMI, body mass index; Fbg, fibrinogen; HGB, hemoglobin; LYMPH, lymphocyte; OCS, osteofascial compartment syndrome; PLT, platelet; PT, prothrombin time; RBC, red blood cell; SSI, surgical site infection; TBIL, total bilirubin; TP, total protein; WBC, white blood cell. Preoperative albumin level, trauma type, history of hypertension and diabetes, and length of surgery were shown to be five statistically significant risk factors. When constructing the postoperative SSI risk prediction model, these factors showed significant predictive value. The pathophysiology of postoperative SSI is significantly influenced by the risk factors that have been established. Clinicians can enhance preoperative preparation procedures, create tailored perioperative preventative efforts, and carry out focused treatments by identifying these factors early. This evidence-based risk assessment not only helps reduce the incidence of SSI but also improves patient outcomes and reduces the overall health care burden. Therefore, in order to improve perioperative care for patients undergoing ankle fracture surgery, we advise integrating this risk assessment approach into clinical pathway management (Fig. 1). Figure 1.: Evidence-based clinical pathway for surgical site infection prevention in ankle fracture. Open fractures are major difficulty in orthopedic care and are usually induced by high-energy trauma. When the soft tissue envelope is exposed during trauma, the hematoma at the fracture site becomes contaminated, significantly raising the risk of infection. The occurrence of an open fracture is inevitable and cannot be changed by the surgeon, in contrast to other risk factors for SSI. Therefore, in order to reduce the risk of infection, the risk of SSI linked to open fractures has to be carefully treated, with early I.V. antibiotic administration and thorough debridement. To reduce the risk of osteomyelitis and the necessity for future amputation, external fixation is favored over internal fixation in type IIIB fractures, which frequently entail involve soft tissue destruction and comminuted fractures. Nevertheless, the study is a retrospective analysis with a small sample size, which may introduce potential bias. Future prospective RCTs performed across multiple centers could improve the accuracy of the findings. This study was conducted in compliance with the Transparency in the Reporting of Artificial Intelligence guidelines[5].

Ähnliche Arbeiten