Blood Transfusion Rates and Predictive Factors in Posterior-Based Minimally Invasive Scoliosis Surgery for Idiopathic Scoliosis: A LASSO-Based Prediction Model Authors: Hong Jin Kim, Dong Yun Kim, Subin Lim, Hyung-Rae Lee, Jae Hyuk Yang, Seung Woo Suh Published: 2026 ABSTRACT AbstractStudy DesignRetrospective cohort study.ObjectivesTo investigate blood transfusion rates in posterior-based minimally invasive scoliosis surgery (MISS) for idiopathic scoliosis (IS) and to develop a prediction model for perioperative blood transfusion.MethodsWe retrospectively reviewed data of 751 patients with IS who underwent posterior-based MISS between 2015 and 2026. The Least Absolute Shrinkage and Selection Operator (LASSO) regression model was applied to identify predictors of transfusion among demographics, perioperative metrics, radiographic measurements, and transfusion-related profiles. The model was constructed by development cohort March 2015 and February 2024 (development cohort, n = 595) and validated using the temporal data between March 2024 and February 2026 (validation cohort, n = 156).ResultsIn the total cohort, 518 patients (69.0%) required a blood transfusion. Among them, 76 patients received only postoperative transfusions, with the majority administered within 3 days postoperatively. Post-LASSO multivariate regression analysis demonstrated that greater estimated blood loss, lower preoperative hemoglobin (Hb) levels, absence of tranexamic acid use, longer fusion levels, and a larger major preoperative Cobb angle were independently associated with perioperative transfusion. The perioperative model demonstrated good discrimination, with an area under the receiver operating characteristic curve of 0.89 in the development cohort and 0.94 in the validation cohort.ConclusionsPostoperative transfusion was uncommon, and transfusion-free surgery may be feasible in selected lower-risk patients undergoing posterior-based MISS for IS. Based on our prediction model, strategies for blood transfusion-free surgery should be individualized, considering patient-specific profiles such as preoperative Hb levels, fusion level, and Cobb angle.Keywords: Blood transfusion, perioperative bleeding risk, minimally invasive scoliosis surgery, posterior spinal fusion, idiopathic scoliosis Introduction Surgical treatment of adolescent idiopathic scoliosis (AIS) primarily focuses on restoring trunk balance and preventing long-term thoracic sequelae through correction of the severe scoliotic curves. 1 Although motion-preserving surgeries, such as vertebral body tethering, have been introduced, posterior spinal fusion (PSF) remains the mainstay for deformity correction.1-3 Conventional open scoliosis surgery (COSS) is typically performed through a single long posterior skin incision with extensive soft-tissue dissection.1,4 In particular, these procedures require long-level instrumented fusion, placing pediatric patients at substantial risk for perioperative blood loss and subsequent blood transfusion.5,6 Blood transfusion carries inherent risks for transfusion-related complications, prolonged hospital stays, and increased hospital costs in patients with idiopathic scoliosis (IS).6-8 Recent advances in surgical techniques, bleeding control, and postoperative care have enabled a reduction in the volume and overall requirement for blood transfusion.6,7 However, studies of COSS have suggested that transfusion rates remain high, reaching 67.6% according to the pediatric National Surgical Quality Improvement Program (NSQIP) database. 5 This high transfusion requirement may be attributable to surgical complexity, including wide exposure of the posterior elements and highly vascularized paraspinal muscles, compounded by the lower body weight and circulating blood volume of pediatric patients compared with that of adults. 5 In addition to intraoperative bleeding, postoperative bleeding associated with hidden blood loss (HBL) after extensive surgical exposure may be considerable; transfusions are often administered more frequently during the postoperative period than intraoperatively in COSS.9-11 Posterior-based minimally invasive scoliosis surgery (MISS), also referred to as the coin-hole technique, achieves deformity correction and PSF through two or three small skin incisions of 3–4 cm, thereby minimizing soft-tissue dissection and limiting excessive exposure.4,12 In flexible AIS, this approach yields radiographic outcomes comparable to those of COSS, while offering notable advantages in perioperative management and cosmetic satisfaction. 13 Although substantial intraoperative blood loss may still be unavoidable, the limited exposure inherent to posterior-based MISS is expected to reduce HBL and subsequently decrease the overall need for blood transfusion. 4 However, whether transfusion-free surgery can be reliably achieved in this setting remains unclear. Achieving truly transfusion-free surgery for pediatric spinal deformity correction is challenging because transfusion requirements are influenced by multiple patient-specific, surgical, and perioperative factors.7,14,15 Therefore, when planning posterior-based MISS, identifying patients who are likely to undergo surgery without requiring transfusions is clinically important for individualized perioperative blood management. Accordingly, the aim of this study was to investigate the rates and predictors of blood transfusion in a large cohort of patients who underwent posterior-based MISS for IS including AIS and young adult IS. By comprehensively evaluating the relevant patient-related and operative variables, we developed a prediction model to assist surgeons in optimizing perioperative blood management and advancing toward transfusion-free scoliosis surgery. Methods Study design This single-center retrospective cohort study was approved by the Institutional Review Board of our hospital (IRB number: 2026GR0106), conducted in accordance with the ethical principles of the Declaration of Helsinki, and reported following the Strengthening the Reporting of Observational Studies in Epidemiology guidelines. 16 Patients, inclusion and exclusion criteria, and group allocations We reviewed 774 consecutive patients with IS (involving AIS and young adult IS) who underwent posterior-based MISS between March 2015 and February 2026. The inclusion criteria were patients with AIS aged 10–18 years and young adults with IS aged < 40 years. Patients were excluded if they met any of the following criteria: (1) age < 10 years or > 40 years, (2) previous history of spinal surgery, (3) non-idiopathic etiology of scoliosis, (4) incomplete medical records during perioperative period or (5) medical history of coagulation disorders or current usage of antiplatelet or anticoagulation agents. After applying these criteria, 751 patients were included in this study. Patients were categorized according to their perioperative transfusion status into a transfusion-free group (n = 233, 31.0%) and transfusion group (n = 518, 69.0%). To compensate for the lack of an external validation cohort for posterior-based MISS, we constructed a development set and an independent temporal validation cohort. The development set comprised 595 patients (79.2%) who underwent the same posterior-based MISS protocol at our institution between March 2015 and February 2024, and the validation cohort comprised 156 consecutive patients (20.8%) who underwent the same posterior-based MISS protocol at our institution between March 2024 and February 2026 (Figure 1). A study flowchart consisting of development and validation sets Posterior-based MISS All surgical procedures were performed by a single senior surgeon with extensive experience in both the MISS and COSS. All posterior-based MISS was conducted with a curve flexibility of > 30% using preoperative lateral bending or supine radiographs. Under general anesthesia, the patients were placed in a prone position on a hyperextended radiolucent table to increase skin redundancy and facilitate instrument handling. After checking the target fusion levels, two or three midline skin incisions of 3–4cm were made to instrument the six–seven thoracic vertebrae or four–six lumbar vertebrae. After splitting the thoracolumbar fascia, 20- to 24-mm diameter right-angled retractors (or tubular retractors) were placed to expose the lateral side of the facets to approach the pedicle screw entry points. The facet joint capsules were removed, and a hole was created at the pedicle screw entry point using a drill. After inserting a guidewire through the trajectory of the pedicle screw, a specially designed reamer with a cutting head was inserted over the guidewire to prepare the fusion bed. A bone graft (chipped cancellous bone allografts or graft substitute) was applied over the facet using a cut-open 5 mL syringe and impactor. Cannulated polyaxial long-head pedicle screws were inserted over the guidewire in all functional segments by using the freehand technique. After the precontoured rods were inserted in a cranial-to-caudal direction, a rod derotation maneuver was conducted to achieve curve correction. The residual rib hump was resolved by thoracoplasty, which was conducted using four to five rib resections with a 5-mm diamond burr in the most prominent region. Walker ambulation was encouraged 1–2 days postoperatively, depending on the patient’s condition. Patients were discharged after confirming their ability to walk independently, typically 7–8 days postoperatively. All patients were required to wear a brace for 3 months (Figure 2). Case of posterior-based minimally invasive scoliosis surgery (MISS) without blood transfusion. (A) A 13-year-old female patient visited our clinic due to abnormal trunk appearance. (B and C) The patient was diagnosed with adolescent idiopathic scoliosis (AIS) (Lenke type I and Cobb angle of 60°). (D) The patient underwent posterior-based MISS (T4–L1) via two 4 cm-sized mini-open incisions. (E and F) At the 2-year follow-up, the deformity correction for the AIS was well-maintained Perioperative blood management and transfusion criteria Our institutional protocol for perioperative blood management and transfusion was consistently applied throughout the study period. All patients underwent preoperative hematologic screening, including a complete blood count and iron-status parameters. In patients who received tranexamic acid (TXA), it was administered intravenously as a 1500 mg bolus before skin incision, followed by a continuous infusion of 1 mg/kg/hr until wound closure. Ringer’s lactate solution was routinely used for hemodilution during the perioperative period, and controlled hypotension with a mean arterial pressure of 70–80 mmHg was maintained during surgery. Red blood cells were transfused when the hemoglobin level fell below 8 g/dL, or when it was between 8 and 10 g/dL in the presence of hemodynamic instability (i.e. hypotension and tachycardia); this threshold was applied uniformly across the development and validation periods. Hemoglobin and hematocrit were measured serially: preoperatively, on the day of surgery, and on postoperative days 1, 2, 3, 4, 7, and 14 (or until discharge). All patients in the transfused group received allogeneic blood exclusively. Neither autologous blood transfusion nor perioperative cell salvage was applied. Outcome measures Relevant patient data were retrieved from the electronic medical records and standing whole-spine anteroposterior and lateral radiographs. Baseline demographic variables included sex, age, body mass index, and Risser stage. Curve-related and surgical variables included Lenke type, major Cobb angle, curve correction rate, number of fusion levels, thoracoplasty performance, number of rib resection levels, number of skin incisions, and total scar length. Perioperative variables included operative time, length of hospital stay, estimated blood loss (EBL), HBL, and use of TXA. Preoperative laboratory parameters, including hemoglobin (Hb) and hematocrit (Hct) levels, were also retrieved. EBL was calculated using weighting calculation method (total weight of gauze after surgery – total weight of gauze before surgery + amount fluid in suction bucket – weight of flushing fluid). 17 HBL was defined as the difference between the total blood loss (TBL) and EBL. TBL was calculated based on the Gross formula as follows: patient blood volume (PBV) × ([preoperative Hct – the lowest postoperative Hct]/[the average of the preoperative and the lowest postoperative Hct]). 18 Transfused red blood cell volume was added to the TBL in transfused patients. PBV was determined using the Nadler equation: PBV = k1×{height(m)} 3 + k2×{weight(kg)} + k3 (for men, k1=0.366, k2=0.032, and k3=0.604; for women, k1=0.356, k2=0.033, and k3=0.183). 19 Regarding transfusion-related data, we evaluated the overall transfusion rate, transfused volume, and incidence of massive transfusion. Massive transfusion was defined as the replacement of >100% of a PBV within a 24-hour period. 20 Postoperative complications that occurred within 30 days after surgery, including pleural effusion, pneumonia, wound dehiscence, and wound infection, were recorded. The primary outcomes were perioperative blood transfusion rates in the development cohort and the identification of associated risk factors for constructing a predictive model. Secondary outcomes included transfusion-related profiles, serial perioperative laboratory changes in the transfusion-free group, and the validation performance of the developed model. Statistical analysis Statistical analyses were performed using Python (version 3.11.5, Python Software Foundation, Wilmington, DE, USA) and matplotlib (version 3.7.2). Normal distribution was confirmed using the Kolmogorov–Smirnov test. After confirming data homogeneity or heteroscedasticity, Student’s t-test was used for continuous variables, which are presented as means and standard deviation (SDs), and the chi-squared test was used for categorical variables, which are presented as numbers and percentages, to compare the two groups, as appropriate. A sparse Least Absolute Shrinkage and Selection Operator (LASSO) regression model was used to identify the most robust predictors of blood transfusion from the comprehensive variable pool. The preoperative model included only preoperatively determinable variables to assist in surgical planning, whereas the perioperative model incorporated both preoperative and intraoperative factors. Variable selection within the LASSO framework was optimized using 5-fold cross-validation, adhering to the 1-standard error rule to ensure model parsimony and prevent overfitting. 21 The selected predictors were then incorporated into a post-LASSO multivariate logistic regression model to elucidate their independent associations with blood transfusion, reporting odds ratios (ORs), and 95% confidence intervals (CIs). The predictive performance of the final model was evaluated and compared between the development and validation datasets using metrics, including overall accuracy, the area under the receiver operating characteristic curve (AUROC), Brier scores, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). For the calculation of accuracy, sensitivity, specificity, PPV, and NPV, a probability threshold of 0.5 was prespecified and applied identically to the development and validation cohorts for both models. Calibration was assessed using decile-based calibration plots with 95% confidence intervals and quantified by the calibration slope and intercept; clinical utility was evaluated by decision curve analysis comparing the net benefit of the model with treat-all and treat-none. A P-value < 0.05 was considered statistically significant. Results Blood-transfusion profiles and baseline characteristics Of the 751 patients in the total cohort, 233 and 518 patients were classified as transfusion-free and transfused group, respectively. For the transfused group, 442 patients (85.3%) received intraoperative transfusion only and 76 patients (14.7%) received postoperative transfusion only. No patient received both intraoperative and postoperative transfusion (Figure 3A). When postoperative transfusions were necessary, they predominantly occurred within the first 3 days after surgery (Figure 3B). The transfusion volume was significantly higher in patients who received intraoperative transfusions (19.9 mL/kg) than in those who received postoperative transfusions only (9.7 ± 4.3 mL/kg) (P < 0.001) (Figure 3C). In the transfusion-free group, the mean intraoperative Hb level drop (2.9 g/dL) was significantly greater than the postoperative Hb level drop (2.1 g/dL) (P < 0.001) (Figure 3D). Furthermore, the mean Hb and Hct levels reached their lowest points on postoperative day 4 (8.6 g/dL and 25.2%, respectively) and recovered thereafter (Figures 3E and F). Blood profiles following posterior-based minimally invasive scoliosis surgery (MISS) with or without transfusions. (A) Blood transfusion rates, intraoperatively (85.3%, 442 of 518) and postoperatively (14.7%, 76 of 518). (B) Postoperative blood transfusion following posterior-based MISS. (C) Transfusion volume per patient weight in the intraoperative and postoperative transfusion groups. (D) Hemoglobin (Hb) level drop, intraoperatively and postoperatively. (E) Violin plot of hemoglobin level follow-up, ranging from preoperative to postoperative day 7. (F) Violin plot of hematocrit level follow-up, ranging from preoperative to postoperative day 7 While demographic variables showed no statistically significant differences between the two groups, the transfusion group exhibited significantly greater deformity severity and surgical invasiveness. This was characterized by a larger major Cobb angle, a higher number of fusion levels, a higher frequency of thoracoplasty, longer operative times, and greater EBL (all P < 0.001). Furthermore, preoperative Hb levels and prophylactic use of TXA were significantly lower in the transfusion group (all P < 0.001). Although pleural effusion was frequently observed in both groups, largely attributable to the routine use of thoracoplasty, most effusions were mild and resolved with observation alone; intervention was required in only 10.1% and 16.7% of affected patients in the transfusion-free and transfusion groups, respectively (P = 0.042). Superficial wound infection was less frequent in the transfusion-free group than in the transfusion group (0.4% vs. 3.1%, P = 0.030) (Table 1). Table 1.Baseline study patient characteristics of the total cohortVariablesTransfusion-free group (n = 233)Transfusion group (n = 518) P Demographics Sex (male/female, n)26/20740/4780.126Age (mean ± SD, years)17.5 ± 4.717.7 ± 4.80.526BMI (kg/m2)19.3 ± 2.818.9 ± 2.70.083Risser stage (mean ± SD, grade)3.8 ± 1.43.9 ± 1.40.587 Curve-related outcomes Lenke type (1/2/3/4/5/6, n)150/4/34/0/34/11303/25/96/2/56/360.056Major Cobb angle (mean ± SD, °)56.4 ± 9.760.3 ± 10.8< 0.001Correction rate (mean ± SD, %)64.6 ± 10.063.0 ± 11.00.062Fusion levels (mean ± SD, n)10.7 ± 1.511.8 ± 1.4< 0.001Thoracoplasty (yes/no, n)179/54 (76.8%)449/69 (86.7%)< 0.001Rib resection levels (mean ± SD, n)4.2 ± 1.24.2 ± 1.10.946Number of incisions (1/2/3, n)4/219/103/464/510.012Total Scar lengths (mean ± SD, cm)8.4 ± 1.88.9 ± 1.60.001 Perioperative outcomes Operative time (mean ± SD, minutes)240.65 ± 55.26302.51 ± 82.56< 0.001Hospital days (mean ± SD, days)8.7 ± 2.510.0 ± 6.1< 0.001Estimated blood loss (mean ± SD, mL)281.4 ± 342.3972.2 ± 645.1< 0.001Hidden blood loss (mean ± SD, mL)1303.2 ± 483.41334.8 ± 616.60.489Tranexamic acid use (n, %)190 (81.5%)270 (52.1%)< 0.001 Laboratory data Preoperative hemoglobin (mean ± SD, g/dL)13.6 ± 1.113.3 ± 1.2< 0.001Preoperative hematocrit (mean ± SD, %)40.7 ± 2.839.9 ± 3.1< 0.001 Blood transfusion Transfusion volume (mean ± SD, mL)0655.3 ± 543.8N/AIntra- and postoperative transfusion (n, %)0​N/A Intraoperative transfusion only442 Postoperative transfusion only76 Both transfusions0Massive transfusion (n, %)03 (0.6%)N/A Complications Pleural effusion (n, %)188 (80.7%)413 (79.7%)0.922Management for pleural effusion (n, %)10.1%16.7%0.042 Observation/Pigtail or Chest tube insertion169/19344/69Pneumonia (n, %)2 (0.9%)5 (1.0%)> 0.999Wound dehiscence3 (1.3%)17 (3.3%)0.144Wound infection Superficial infection1 (0.4%)16 (3.1%)0.030 Deep infection2 (0.9%)13 (2.5%)0.166N, number; SD, standard deviation; BMI, body mass index. Baseline study patient characteristics of the total cohort N, number; SD, standard deviation; BMI, body mass index. Prediction model for blood transfusion in posterior-based MISS of IS A sparse perioperative LASSO regression model selected six predictors from the development set (n = 595): EBL, preoperative Hb level, TXA use, fusion level, major Cobb angle, and operative time (Figures 4A and B). When the LASSO selection procedure was repeated separately in the AIS (n = 353) and young adult IS (n = 242) subgroups of the development cohort, the same six predictors were selected in both subgroups. The receiver operating characteristic (ROC) curve of the final perioperative model incorporating these six parameters demonstrated an AUROC of 0.89, accuracy of 0.83, and Brier score of 0.12 (Figure 4C). For the preoperative model, the four predictors were selected from the development cohort: preoperative Hb level, TXA use, fusion level, and major Cobb angle (Figures 4D and E). The ROC curve of the final preoperative model showed an AUROC of 0.79, accuracy of 0.75, and Brier score of 0.16, indicating lower performance compared to the perioperative model (Figure 4F). Construction and performance of the perioperative and preoperative LASSO model. (A) Coefficient path plot showing shrinkage of the six selected predictors across penalization levels for perioperative model. (B) Five-fold cross-validation curve for selection of the tuning parameter lambda for perioperative model. The cross-validation curve demonstrated a broad low-error region, and the lambda selected by the 1-standard error rule provided a parsimonious model while maintaining near-minimal prediction error. (C) Receiver operating characteristic curve of the final perioperative model incorporating estimated blood loss, preoperative hemoglobin level, tranexamic acid use, fusion extent, major preoperative Cobb angle, and operative time. (D) Coefficient path plot showing shrinkage of the six selected predictors across penalization levels for preoperative model. (E) Five-fold cross-validation curve for selection of the tuning parameter lambda for preoperative model. (F) Receiver operating characteristic curve of the final preoperative model incorporating preoperative hemoglobin level, tranexamic acid use, fusion extent, and major preoperative Cobb angle A subsequent post-LASSO multivariate logistic regression perioperative model confirmed that greater EBL (OR, 1.41 per 100 mL; 95% CI, 1.33–1.49), preoperative Hb level (OR, 0.57 per 1 g/dL; 95% CI, 0.46–0.70), TXA use (OR, 0.33; 95% CI, 0.20–0.55), extended fusion levels (OR, 1.26 per 1 level; 95% CI, 1.06–1.49), and a larger major preoperative Cobb angle (OR, 1.04 per 1°; 95% CI, 1.01–1.06) were significant independent factors for blood transfusion in posterior-based MISS for IS. Similarly, the preoperative model demonstrated consistent results for the shared variables: preoperative Hb level (OR, 0.66 per 1 g/dL; 95% CI, 0.55–0.81), TXA use (OR, 0.26; 95% CI, 0.17–0.39), fusion levels (OR, 1.55 per 1 level; 95% CI, 1.31–1.81), and major preoperative Cobb angle (OR, 1.03 per 1°; 95% CI, 1.01–1.05) (Table 2). Table 2.Multivariate logistic regression analysis for blood transfusion in posterior-based minimally invasive scoliosis surgery for idiopathic scoliosis in the development cohortVariablesBSEOR95% CI P Perioperative model Estimated blood loss (per 100mL)0.340.031.411.33–1.49< 0.001Preoperative Hb (per 1g/dL)-0.570.100.570.46–0.70< 0.001TXA use (ref: no use)-1.100.260.330.20–0.55< 0.001Fusion level (per 1 level)0.230.091.261.06–1.490.009Major Cobb angle (per 1°)0.040.011.041.01–1.060.001Operative time (per 30 min)0.080.071.080.94–1.230.266 Preoperative model Preoperative Hb (per 1g/dL)-0.410.100.660.55–0.81< 0.001TXA use (ref: no use)-1.360.220.260.17–0.39< 0.001Fusion level (per 1 level)0.440.081.551.32–1.81< 0.001Major Cobb angle (per 1°)0.030.011.031.01–1.050.005Predictors selected by the least absolute shrinkage and selection operation (LASSO) model using the 1-standard error rule were included in the post-LASSO multivariate logistic regression model. Regression coefficients (B) and standard errors (SE) in the table are rounded to two decimal places for presentation. The prediction equations below express each coefficient per the unit of measurement of the corresponding variable, allowing reproduction of the predicted probabilities. Probability for transfusion = 1/(1 + e−logit(p)), logit(p) for perioperative model = 2.629+0.0034×(EBL [mL])−0.562×(preoperative Hb level [g/dL])−1.109×(TXA use [1 = yes, 0 = no])+0.231×(Fusion level [number of levels])+0.039×(Major Cobb angle [°])+0.0027×(Operative time [min]). Logit(p) for preoperative model = 0.499−0.416×(preoperative Hb level [g/dL])−1.347×(TXA use [1 = yes, 0 = no])+0.438×(Fusion level [number of levels])+0.030×(Major Cobb angle [°]). SE, standard error; OR, odds ratio; CI, confidence interval; EBL, estimated blood loss; Hb, hemoglobin; TXA, tranexamic acid. Multivariate logistic regression analysis for blood transfusion in posterior-based minimally invasive scoliosis surgery for idiopathic scoliosis in the development cohort Predictors selected by the least absolute shrinkage and selection operation (LASSO) model using the 1-standard error rule were included in the post-LASSO multivariate logistic regression model. Regression coefficients (B) and standard errors (SE) in the table are rounded to two decimal places for presentation. The prediction equations below express each coefficient per the unit of measurement of the corresponding variable, allowing reproduction of the predicted probabilities. Probability for transfusion = 1/(1 + e−logit(p)), logit(p) for perioperative model = 2.629+0.0034×(EBL [mL])−0.562×(preoperative Hb level [g/dL])−1.109×(TXA use [1 = yes, 0 = no])+0.231×(Fusion level [number of levels])+0.039×(Major Cobb angle [°])+0.0027×(Operative time [min]). Logit(p) for preoperative model = 0.499−0.416×(preoperative Hb level [g/dL])−1.347×(TXA use [1 = yes, 0 = no])+0.438×(Fusion level [number of levels])+0.030×(Major Cobb angle [°]). SE, standard error; OR, odds ratio; CI, confidence interval; EBL, estimated blood loss; Hb, hemoglobin; TXA, tranexamic acid. Validation of the prediction model There was no statistical difference in the age and transfusion event between the development and validation set (all Ps > 0.05). In the temporal validation set (n = 156), the perioperative model maintained good discrimination (AUROC, 0.94; accuracy, 0.87; Brier score, 0.11), with a sensitivity of 0.87, specificity of 0.88, PPV of 0.92, and NPV of 0.80. In contrast, the performance of the preoperative model decreased in the validation cohort compared with the development cohort (AUROC, 0.72 vs. 0.79; accuracy, 0.65 vs. 0.75; NPV, 0.53 vs. 0.60; Brier score, 0.20 vs. 0.16), with low specificity in both cohorts (0.49 and 0.53, respectively) (Table 3). Table 3.Comparison of baseline data and model performance between development and validation setVariablesDevelopment set (n = 595)Validation set (n = 156) P Baseline data (age and transfusion rate)Age (mean ± SD, years)17.8 ± 4.917.0 ± 4.40.074AIS (10–18 years, n)3531030.141Young adult IS (18–40 years, n)24253​Transfusion event (n, %)419 (70.4%)99 (63.5%)0.099for AIS248 (70.3%)63 (61.2%)0.092for young adult IS171 (70.7%)36 (67.9%)0.741Perioperative model performanceOverall accuracy0.830.87​AUROC (95% CI)0.89 (0.87–0.92)0.94 (0.90–0.97)​Brier score0.120.11​Sensitivity0.880.87​Specificity0.710.88​Positive Predictive Value0.870.92​Negative Predictive Value0.730.80​Preoperative model performanceOverall accuracy0.750.65​AUROC (95% CI)0.79 (0.74–0.82)0.72 (0.63–0.80)​Brier score0.160.20​Sensitivity0.860.72​Specificity0.490.53​Positive Predictive Value0.800.71​Negative Predictive Value0.600.53​N, number; SD, standard deviation; AIS, adolescent idiopathic scoliosis; IS, idiopathic scoliosis; AUROC, area under the receiver operating characteristic; CI, confidence interval. Comparison of baseline data and model performance between development and validation set N, number; SD, standard deviation; AIS, adolescent idiopathic scoliosis; IS, idiopathic scoliosis; AUROC, area under the receiver operating characteristic; CI, confidence interval. The calibration plots showed good agreement between predicted and observed transfusion rates in the development cohort for both models. In the temporal validation cohort, the preoperative model remained well calibrated (calibration slope, 1.07; 95% CI, 0.71–1.58), whereas the perioperative model showed substantial miscalibration (calibration slope, 1.80; 95% CI, 1.40–2.53; intercept, 0.27; 95% CI, −0.08 to 0.60). In the decision curve analysis, the models yielded a greater net benefit than both the treat-all and treat-none strategies across threshold probabilities of approximately 0.15–0.95 for the perioperative model and 0.40–0.95 for the preoperative model in the validation cohort; below these ranges, the net benefit of each model converged with that of the treat-all strategy (Supplementary Figure 1). Descriptive subgroup analysis of transfusion-free rates for IS To empirically illustrate the cumulative effects of favorable clinical parameters on surgical outcomes, transfusion-free rates were assessed through a descriptive nested subgroup analysis. Although the baseline blood transfusion-free rate was 31.0% in the total cohort, the use of TXA as a baseline intraoperative strategy increased this rate to 41.3%. Among patients receiving TXA, the transfusion-free rate increased to 44.8% in those with preoperative hemoglobin levels of ≥ 13 g/dL, to 57.0% when the major Cobb angle was < 60°, and to 75.4% when fusion was limited to ≤ 12 levels (Table 4). Table 4.Transfusion-free rate in posterior-based minimally invasive scoliosis surgery for idiopathic scoliosisVariablesTransfusion-free (n)Success rate (%) with 95% CIOverall cohort (n = 751)23331.0% (27.9%–34.5%)TXA use (n = 460)19041.3% (37.0%–46.0%)TXA use + Preop Hb ≥ 13.0 g/dL (n = 321)14444.8% (39.4%–50.2%)TXA use + Preop Hb ≥ 13.0 g/dL + Major Cobb angle < 60° (n = 172)9857.0% (49.5%–64.1%)TXA use + Preop Hb ≥ 13.0 g/dL + Major Cobb angle < 60° + Fusion level ≤ 12 (n = 57)4375.4% (62.9%–84.8%)N, number; CI, confidence interval; TXA, tranexamic acid; Preop, preoperative; Hb, hemoglobin. Transfusion-free rate in posterior-based minimally invasive scoliosis surgery for idiopathic scoliosis N, number; CI, confidence interval; TXA, tranexamic acid; Preop, preoperative; Hb, hemoglobin. Discussion While intraoperative transfusion remains relatively common in posterior-based MISS, postoperative transfusion is remarkably infrequent (10.1%) and accompanied by a relatively low overall transfused volume (18.4 mL/kg). These findings suggest that posterior-based MISS may substantially reduce the perioperative transfusion burden, thereby offering distinct perioperative advantages for idiopathic scoliosis including pediatric and young adult patients. Furthermore, achieving completely transfusion-free surgery appears to be highly feasible when tailored to individual patient profiles, such as EBL, preoperative Hb level, TXA use, fusion level, and curve severity. Therefore, transfusion-free surgery may be feasible in selected lower-risk patients undergoing posterior-based MISS for IS based on our prediction model. In our study, 69.0% of patients with IS (including both AIS and young adults) who underwent posterior-based MISS received perioperative blood transfusions, a rate comparable to the 67.6% reported in the NSQIP database between 2012 and 2013. 5 Minha et al. suggested that blood transfusion rates may be underestimated in scoliosis surgery when considering small-volume transfusions and intraoperative cell salvage volumes.5,22 Furthermore, they indicated that transfused volumes exceeding 20 mL/kg are associated with a higher rate of transfusion-related complications. 5 Although the overall transfusion rate observed in our cohort was comparable to that of COSS, patients undergoing posterior-based MISS exhibited distinct profiles regarding blood transfusions. The mean transfusion volume in our cohort was below the threshold (20 mL/kg) for transfusion-related complications. In addition, the postoperative transfusion rate was low in patients who underwent posterior MISS. Although intraoperative blood loss and anatomical exposure may be inevitable during pedicle screw instrumentation-related bony procedures, limiting extensive exposure with soft tissue preservation may substantially reduce the need for postoperative transfusions. In transfusion-free patients, postoperative Hb and Hct levels are expected to decline during the early postoperative period, largely reflecting EBL, HBL, and hemodilution.6,9,23 Spinal fusion is among the procedures associated with the greatest postoperative Hb drift, likely due to substantial blood loss and perioperative fluid requirements, with postoperative Hb levels generally decreasing until postoperative days 2–4 before gradually recovering. 6 Similarly, van Popta et al. described the blood loss pattern following COSS, demonstrating that a clinically relevant Hb level decrease occurred by postoperative day 2, characterized by a mean Hb level drop of 4.1 g/dL, followed by subsequent stabilization. 14 In the present study, the temporal pattern of Hb decline was broadly comparable, with the nadir observed on postoperative days 3–4 and gradual recovery thereafter. However, the magnitude of the postoperative Hb drop in our transfusion-free cohort was relatively modest, with a mean drop of 2.1 g/dL, indicating the reduced hematologic burden in posterior-based MISS. Given that Hb levels continue to decline until postoperative days 3 to 4, and that postoperative transfusions predominantly occur within the first 72 h, close monitoring remains essential during the early postoperative period following posterior MISS. 14 Given the surgical complexity of scoliosis requiring long-level PSF, optimal perioperative management necessitates careful consideration of both the intrinsic anatomical characteristics of the curve and pediatric patient-specific profiles. Dong et al. demonstrated that a greater number of fusion segments and increased intraoperative blood loss were significant predictors of allogeneic blood transfusion in AIS. 10 Similarly, Minhas et al. identified posterior arthrodesis involving ≥ 13 vertebral segments as an independent risk factor. 5 In the present study, our post-LASSO multivariate logistic regression models comprehensively reflected patient’s baseline hematologic reserves as well as the intrinsic anatomical complexity of the scoliotic curve when anticipating transfusion requirements, specifically identifying EBL, fusion levels, major Cobb angle, and preoperative Hb level as key independent predictors. In addition to patient-specific profiles and curve characteristics, intraoperative strategies play a pivotal role in perioperative blood management. Operative time can also influence the likelihood of requiring blood transfusions. Because a prolonged operative times are inherently associated with greater EBL, increased surgical complexity, and extended fusion levels, our post-LASSO multivariate model showed no statistical significance in operative times after adjustment, despite being initially selected by the LASSO algorithm. Meanwhile, the use of TXA was significantly associated with a reduced risk of transfusion in our model. Considering that TXA administration is a modifiable intraoperative factor, our findings emphasize that active surgical efforts to minimize intraoperative blood loss may associate with a reduced transfusion requirement during posterior-based MISS. 24 In our preliminary experiences, posterior-based MISS was primarily indicated for patients with AIS aged 10–18 years presenting with moderate-to-severe curves (major Cobb angle < 80°) and a curve flexibility of greater than 50%. Recently, the feasibility of posterior MISS has been expanded to include young adults with IS, severe curves exceeding 80°, and a curve flexibility of up to 30%. This expansion of surgical indications has been largely driven by continuous advances in pedicle screw insertion techniques through 3–4-cm mini-open incisions. Consequently, optimal perioperative management of posterior-based MISS has become increasingly critical because of the inherent requirement for longer fusion constructs and complex corrections in patients with severe deformities and in young adults. In line with this evolving complexity, the use of TXA has increased year by year, as its administration was selectively determined by the surgeon to mitigate the higher anticipated blood loss in these advanced cases. 25 While this time-dependent and case-selective application acts as a potential confounder in our analysis, multivariable analysis demonstrated that TXA utilization was independently associated with a significantly lower likelihood of allogeneic blood transfusion; however, this association should not be interpreted as a demonstrated treatment effect. Considering this expanding surgical landscape, we sought to determine the achievable range for transfusion-free surgery in posterior MISS using our perioperative prediction model. Based on our empirical data, when combined with the administration of TXA and favorable baseline characteristics, our findings suggest that posterior MISS can potentially achieve blood transfusion-free management. This can inform preoperative planning and patient counseling, enabling surgeons to discuss the likelihood of transfusion with patients prior to surgery. Based on our descriptive nested subgroup analysis, the observed transfusion-free rate progressively increased with favorable clinical factors, reaching 75.4% in patients receiving TXA with a preoperative Hb ≥ 13.0 g/dL, major Cobb angle < 60°, and fusion ≤ 12 levels. Although this subgroup was defined solely by these four descriptive parameters rather than a model-derived risk threshold, it illustrates scenarios frequently encountered in clinical practice. For example, in an IS patient presenting with a moderate Lenke type 1 curve (Cobb <60°) who is a candidate for selective thoracic fusion and has adequate baseline Hb reserves, aiming for a transfusion-free protocol with posterior MISS can be a reasonable goal. In the temporal validation cohort, the perioperative model maintained good discrimination (AUROC, 0.94), although its individual probability estimates were not well calibrated (calibration slope, 1.80); these two properties of model performance should be distinguished. Considering a substantial intraoperative transfusion, our perioperative model includes information that is contemporaneous with, or becomes complete after, many transfusion decisions, which limits a prospective intraoperative prediction tool. From a practical standpoint, approximately one in five model-negative patients still required transfusion (NPV of 0.80), which requires routine hematologic monitoring. Conversely, a high PPV ensures that appropriate blood conservation strategies are strictly implemented for high-risk patients. Thus, the perioperative risk classification tool should be regarded as useful for identifying patients who require close hematologic surveillance during the early postoperative period. Interestingly, the preoperative prediction model shared four variables (preoperative hemoglobin, TXA use, fusion level, and major Cobb angle), which were the exact preoperatively determinable factors identified in our perioperative model, which carries the potential role of preoperative planning and counseling. For the preoperative model, discrimination and accuracy decreased in the validation cohort (AUROC, 0.72 vs. 0.79; accuracy, 0.65 vs. 0.75; NPV, 0.53 vs. 0.60), with low specificity in both cohorts (0.49 and 0.53, respectively). This low specificity primarily reflects false-positive classification attributable to the omission of EBL. Without this critical intraoperative factor, the preoperative model could not distinguish, among patients with similar preoperative risk profiles, those whose intraoperative course would prove favorable; consequently, patients classified as high risk preoperatively who ultimately did not require transfusion, owing to lower-than-anticipated intraoperative blood loss, were misclassified as positive. Although such conservative over-triage may be clinically preferable to the converse error, it limits the ability of the preoperative model to identify patients in whom a transfusion-free strategy is most likely to succeed. Accordingly, when the preoperative model is used for counseling about the likelihood of avoiding transfusion, its output should be presented as an approximate estimate, and a low predicted risk should not be equated with assurance of transfusion-free surgery. Although this model potentially guides personalized perioperative blood management based on decision curve analysis, external validation involving different surgeons and centers is warranted to ensure its generalizability and reproducibility. Schupper et al. reported a bloodless spinal deformity program in a distinct cohort of adult patients for whom transfusion was not an option, utilizing a comprehensive multidisciplinary bundle that included preoperative erythropoietin, intraoperative cell salvage, and staged procedures. 26 While their approach focused on preventing transfusion in a restricted population, our study emphasizes the perioperative prediction of transfusion risk in idiopathic scoliosis. Although their adult deformity cohort differs from our idiopathic scoliosis population, some of their blood conservation strategies are highly transferable to posterior MISS. 26 In particular, intraoperative cell salvage, which was not employed in our cohort, could be selectively applied to patients predicted to be at high risk on the basis of the preoperatively available components of our model, potentially extending the feasibility of transfusion-free surgery. Our study had certain limitations. First, given the retrospective nature of the study, potential biases may have been introduced. Second, our prediction model did not comprehensively account for comorbidities. The Charlson Comorbidity Index is known to affect postoperative complications; however, the utility of the Pediatric Spinal Frailty Index has been suggested in evaluating surgical risks in pediatric patients with scoliosis.27,28 This index is associated with the complication rate following scoliosis correction; thus, it needs to be considered in future studies along with complication outcomes. Furthermore, although the TXA regimen itself was uniform throughout the study period, the decision to administer TXA was made by the operating surgeon according to anticipated case complexity and increased over the study period, introducing potential confounding by indication. In addition, because TXA dosing regimens differ across institutions, the association observed under our fixed regimen may not be directly generalizable to centers using other protocols. Third, this study did not include clinical outcomes, which may be important for long-term follow-up. Finally, the model was developed using a posterior-based MISS at a single institution. Because this technique is performed at only a few centers, an external cohort was unavailable, and we therefore divided the cohort by time period for temporal validation as a feasible alternative. Such sources of variability, including evolving case complexity and the increasing TXA use, likely contributed to the substantial miscalibration of the perioperative model in the temporal cohort (calibration slope, 1.80; 95% CI, 1.40–2.53), reflecting predicted probabilities that were systematically too moderate relative to the observed outcomes; individual predicted probabilities from this model should therefore be interpreted cautiously, and recalibration will be required before its application in different settings or later case mixes. Its applicability and generalizability to patients managed with COSS or those treated at other centers thus remain limited. Considering these limitations, multicenter, prospective, and comparative studies are essential to ensure the generalizability of blood transfusion prediction. Conclusion Postoperative transfusion was uncommon, and transfusion-free surgery may be feasible in selected lower-risk patients undergoing posterior-based MISS for idiopathic scoliosis. Based on our prediction model, strategies for blood transfusion-free surgery should be individualized, taking into account patient-specific profiles, such as preoperative hemoglobin levels, fusion levels, and Cobb angle. Supplemental Material Supplemental Material - Blood Transfusion Rates and Predictive Factors in Posterior-Based Minimally Invasive Scoliosis Surgery for Idiopathic Scoliosis: A LASSO-Based Prediction Model Supplemental Material for Blood Transfusion Rates and Predictive Factors in Posterior-Based Minimally Invasive Scoliosis Surgery for Idiopathic Scoliosis: A LASSO-Based Prediction Model by Hong Jin Kim, Dong Yun Kim, Subin Lim, Hyung-Rae Lee, Jae Hyuk Yang, Seung Woo Suh in Global Spine Journal Author Contributions: HJK and SWS developed study concept and design; HJK, DYK, HRL, and JHY performed acquisition, analysis, or interpretation of data; HJK and DYK drafted the initial manuscript; All authors contributed to editing and reviewing the manuscript and approved the final version. Funding: The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was partially supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (RS-2024-00352326) The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. Supplemental Material: Supplemental material for this article is available online. ORCID iDs Hong Jin Kim https://orcid.org/0000-0002-8162-9585 Dong Yun Kim https://orcid.org/0000-0002-8863-642X Subin Lim https://orcid.org/0000-0002-8720-5484 Hyung-Rae Lee https://orcid.org/0000-0003-3125-1470 Jae Hyuk Yang https://orcid.org/0000-0001-6949-6954 Seung Woo Suh https://orcid.org/0000-0002-1536-4611 Ethical Considerations This retrospective study was approved by the Institutional Review Board of Korea University Guro Hospital (IRB number: 2026GR0106). Consent to Participate Informed consent for retrospective analysis of de-identified data was waived. Data Availability Statement The datasets generated during and/or analysed during the current study are not publicly available due to ethical issues but are available from the corresponding author on reasonable request.* References Associated Data Supplementary Materials Supplemental Material - Blood Transfusion Rates and Predictive Factors in Posterior-Based Minimally Invasive Scoliosis Surgery for Idiopathic Scoliosis: A LASSO-Based Prediction Model Supplemental Material for Blood Transfusion Rates and Predictive Factors in Posterior-Based Minimally Invasive Scoliosis Surgery for Idiopathic Scoliosis: A LASSO-Based Prediction Model by Hong Jin Kim, Dong Yun Kim, Subin Lim, Hyung-Rae Lee, Jae Hyuk Yang, Seung Woo Suh in Global Spine Journal Data Availability Statement The datasets generated during and/or analysed during the current study are not publicly available due to ethical issues but are available from the corresponding author on reasonable request.*