Development and Internal Validation of a Simplified Clinical Risk Score for Systemic Complications After Spine Surgery Using a Nationwide Japanese Registry: JOANR/JSSR-DB Authors: Shota Ikegami, Hideyuki Arima, Hiroyuki Ishiguro, Shinji Takahashi, Hayato Takei, Satoshi Maki, Chie Shimmura, Gen Inoue, Takashi Kaito, Haruo Kanno, Shota Takenaka, Hiroaki Nakashima, Nobuyuki Fujita, Naobumi Hosogane, Mitsuru Yagi, Kentaro Yamada, Kei Watanabe, Takanori Fujita, Hiroshi Taneichi, Masahiko Watanabe, Katsushi Takeshita, Masaya Nakamura, Yukihiro Matsuyama, Hirotaka Haro, Hirotaka Chikuda, Tokumi Kanemura Published: 2026 ABSTRACT Introduction We developed a simplified scoring system (JSSR-DB RC3S) to predict systemic complications after spine surgery using data from the nationwide Japanese Society for Spine Surgery and Related Research Database (JSSR-DB). Methods This retrospective cohort study included 8,930 spine surgery cases registered between January and March 2024. The target outcome was systemic complications occurring intraoperatively or within 30 days postoperatively. Candidate preoperative variables were screened using least absolute shrinkage and selection operator logistic regression with 5-fold cross-validation, and the five dominant predictors were used to construct the JSSR-DB RC3S. Integer points were assigned using a zero-start scoring approach, followed by score-based recalibration. Internal validation used out-of-fold predictions from 5-fold cross-validation. Results The incidence rate of systemic complications was 1.7% (156/8,930). The five selected predictors were age, performance status, emergency surgery, degenerative disease, and spinal trauma. The final score ranged from 0 to 22 points. Observed incidence rates increased from 0.23% in the low-risk group (0-2 points) to 6.67% in the high-risk group (14-22 points). The model performance was acceptable, with an area under the curve of 0.740, observed-to-expected ratio of 0.997, and Brier score of 0.0167. Conclusions The JSSR-DB RC3S is a transparent, clinically interpretable, simplified score for predicting systemic complications after spine surgery, derived from Japanese nationwide registry data. It may support perioperative risk communication in daily practice, and may serve as an introductory implementation of the broader JSSR-DB risk calculator framework. Introduction Predicting perioperative complications in spine surgery is crucial for obtaining preoperative informed consent and for ensuring appropriate perioperative management. In Europe and the United States, prediction models that use large-scale data, such as the American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP), are widely recognized1,2). Reports have indicated that these large-scale prediction models can also be useful for patients in other countries3). However, because of differences in background factors such as ethnicity and healthcare systems, these models cannot be directly applied to patients in other countries, such as Japan. Reports have indicated that recalibration can enable their use4). Furthermore, some studies have suggested that their applicability is limited to certain fields5). Most existing Japanese studies on spine surgery–related complications are based on single-center or small-scale data, underscoring the need for nationwide evidence6). Since 2021, the Japanese Society for Spine Surgery and Related Research (JSSR) has operated the JSSR Database (JSSR-DB), a large nationwide registry built on the Japanese Orthopaedic Association National Registry (JOANR) framework7,8). Within this infrastructure, the JSSR-DB Committee is developing the JSSR-DB Risk Calculator (JSSR-DB RC) as a broader registry-based risk assessment platform. However, at the bedside, a simpler, manually applicable tool may also be valuable. Therefore, we aimed to extract five key factors from the JSSR-DB RC development process and develop a simplified scoring system (JSSR-DB RC3S) that is easy to use in routine practice. This study, prepared with reference to Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis+Artificial Intelligence (TRIPOD+AI)9), presents a transparent, practical, and simplified score derived from the JSSR-DB RC development process. Materials and Methods Study design and data source This retrospective cohort study utilized 2024 data from the JSSR-DB, a nationwide surgical registry led by the JSSR. This study was conducted in accordance with the TRIPOD+AI reporting guidelines9). In this study, anonymized registry data from the JOANR were used, which was approved by the association ethics committee, and additional ethical review was waived. Participants and missing data Participants included spine surgery cases registered in the JSSR-DB. Detailed information, including comorbidities, was collected for cases registered between January and March 2024 at JSSR board member and council member facilities nationwide. To strictly define the cohort underlying the model, cases with missing data on key basic attributes—age, height, and weight—were excluded. Consequently, the final analysis cohort included 8,930 cases (N=8,930). In the JSSR-DB RC3S model development, missing values for continuous variables, excluding age, height, and weight, were imputed using the mean, whereas those for categorical and ordinal variables were imputed using the mode. To avoid data leakage, all preprocessing steps, including imputation, were performed within each training fold and were subsequently applied to the corresponding validation fold. Outcome measures This study defined “systemic complications,” which have a significant clinical impact among postoperative complications, as the primary predictive outcome. This outcome is strictly defined in the JOANR and JSSR-DB registries. Systemic complications are defined as those requiring additional treatment during surgery or within 30 days postoperatively or requiring specialized treatment in another department. Included conditions are: acute heart failure, myocardial infarction, pulmonary infarction, pulmonary embolism, symptomatic peripheral venous thrombosis, respiratory failure, aspiration pneumonia, pneumonia, atelectasis, pulmonary edema, cerebral infarction, postoperative delirium, dementia, intestinal obstruction, pseudomembranous colitis, gastrointestinal ulcer or gastric perforation, pancreatitis, hepatitis, cholecystitis, acute renal failure, urinary tract infection, disseminated intravascular coagulation, sepsis, and fractures due to falls. Variable selection To construct the simplified score, all preoperative variables were screened using L1-regularized logistic regression least absolute shrinkage and selection operator (LASSO). Hyperparameters were tuned using 5-fold cross-validation (5-fold CV), and the model with the highest area under the curve (AUC) was selected. For bedside simplicity, the five predictors with the largest absolute coefficients were retained as the dominant factors. Age was treated as a single clinical factor and modeled with three dummy variables, with age <60 years as the reference. Top five factors associated with systemic complications 1. Age group: <60, 60-69, 70-79, and ≥80 years 2. Performance status (assessed using the Eastern Cooperative Oncology Group [ECOG] performance status scale, 0-4) 3. Emergency surgery 4. Degenerative disease: disease classification 5. Spinal trauma: disease classification Development of the simplified scoring system Using the five selected predictors, the model was refitted with L2-regularized logistic regression, with balanced class weights. Coefficients were scaled and rounded to integer points according to the Framingham-style method described by Sullivan et al.10). A zero-start approach was used so that all items could be summed as positive integers. To correct the bias introduced by integerization, the total score was recalibrated as a continuous predictor using logistic regression (Platt scaling). Patients were classified into five risk tiers based on percentile scores (<25th, 25th to <50th, 50th to <75th, 75th to <90th, and ≥90th), according to the score distribution. Internal validation used out-of-fold predictions from 5-fold CV. Discrimination was assessed using the AUC, and calibration was assessed using a calibration plot, the observed-to-expected (O/E) ratio, and the Brier score. The final clinical score and lookup table were reconstructed using the full cohort after cross-validation. This analysis was conducted under the approved JOANR/JSSR-DB research framework. Data and code are not publicly available because of registry governance restrictions. Patient/public involvement or fairness-focused analyses were not performed. Computing infrastructure All data preprocessing, model building, and statistical analyses were performed using Python 3.12.7. The scikit-learn library (version 1.5.1) was used for the machine learning pipeline and model evaluation, whereas pandas (version 2.2.2) and NumPy (version 1.26.4) were used for data manipulation. Results Baseline characteristics The final analysis cohort comprised 8,930 participants (N=8,930) from 128 institutions. In that cohort, missing values were observed for albumin and the Geriatric Nutritional Risk Index (5.0%) and the Prognostic Nutritional Index (20.3%) at the time of registry entry. There were no missing values for categorical measures or the Charlson Comorbidity Index, which is calculated based on the presence of comorbidities. Missing values were imputed according to the protocol. The mean age of participants was 65.1±18.6 years (mean±standard deviation), with males accounting for 54.6% (4,880 participants). By age group, patients aged ≥70 years accounted for more than half (52.3%) of the population. Regarding disease classification and surgical procedures, degenerative diseases (61.6%) were the predominant indication, with surgery in the thoracolumbar region (combined decompression and fusion procedures accounting for 55.9%) being the most common approach. The incidence of systemic complications, the primary predictive outcome of this study, was 1.7% (156 cases), remaining below 2%. Details of the occurrence of each outcome and patient characteristics in this cohort are shown in Table 1. Table 1.Baseline Characteristics.FeatureRepresentative valueMissing dataNumber8930Sex (male/female)4880/4050 (54.6%/45.4%)0 (0.0%)Age (years)65.1±18.60 (0.0%)<602515 (28.2%)60–691749 (19.6%)70–792830 (31.7%)≥801836 (20.6%)Height (cm)159.9±10.70 (0.0%)Weight (kg)61.7±15.10 (0.0%)BMI23.9±4.40 (0.0%)Albumin (g/dL)4.1±0.6449 (5.0%)GNRI105.9±12.9449 (5.0%)PNI49.7±9.01811 (20.3%)CCI0.68±1.330 (0.0%)Disease0 (0.0%)Degenerative disease5498 (61.6%)Spinal deformity, malformation859 (9.6%)Spinal trauma819 (9.2%)Spine/spinal cord tumor385 (4.3%)RA, dialysis, osteoporosis364 (4.1%)Ossification of the ligaments330 (3.7%)Spinal infection170 (1.9%)Reoperation, additional surgery319 (3.6%)Others186 (2.1%)Surgery0 (0.0%)Decompression (cervical spine)963 (10.8%)Fusion (cervical spine)584 (6.5%)Decompression (thoracolumbar spine)2003 (22.4%)Fusion (thoracolumbar spine)2991 (33.5%)Endoscopic surgery902 (10.1%)Correction surgery, osteotomy466 (5.2%)Vertebroplasty332 (3.7%)Intradural surgery241 (2.7%)Total en bloc spondylectomy43 (0.5%)Others405 (4.5%)OutcomeSystemic complications156 (1.7%)0 (0.0%)Continuous values represent the mean±standard deviation.Count and categorical values represent cases (rates).BMI: body mass index; CCI: Charlson Comorbidity Index; GNRI: geriatric nutritional risk index; PNI: prognostic nutritional index; RA: rheumatoid arthritis Baseline Characteristics. Continuous values represent the mean±standard deviation. Count and categorical values represent cases (rates). BMI: body mass index; CCI: Charlson Comorbidity Index; GNRI: geriatric nutritional risk index; PNI: prognostic nutritional index; RA: rheumatoid arthritis Development of the scoring system Table 2 shows the top five factors extracted after LASSO logistic regression during the risk calculator design phase. Emergency surgery and spinal trauma were significant predictors of systemic complications. Degenerative diseases (i.e., spinal stenosis and disc herniation) were associated with a reduced risk. Higher age groups were associated with an increased risk of complications, with those aged ≥70 years representing a significant risk factor. Performance status was a significant predictor of systemic complications, with higher levels indicating a greater likelihood. Table 2.Multivariate Analysis of Systemic Complications Associated with the Top Five Factors.FactorOdds ratio (95% CI)p ValueEmergency surgery2.6 (1.7-4.0)<0.001Degenerative disease0.49 (0.33-0.73)<0.001Spinal trauma1.7 (1.1-2.5)0.018Age group (for 60-69 years)1.2 (0.6-2.3)0.582Age group (for 70-79 years)2.5 (1.5-4.2)<0.001Age group (for ≥80 years)3.3 (2.0-5.5)<0.001PS (+1)1.2 (1.1-1.4)0.001For clinical interpretability, odds ratios and 95% CIs were estimated in a conventional multivariable logistic regression model using the five selected predictors, separate from the L2-regularized model used for score derivation.CI: confidence interval; PS: performance status Multivariate Analysis of Systemic Complications Associated with the Top Five Factors. For clinical interpretability, odds ratios and 95% CIs were estimated in a conventional multivariable logistic regression model using the five selected predictors, separate from the L2-regularized model used for score derivation. CI: confidence interval; PS: performance status These top five factors were converted into integer points for each variable using the zero-start scoring method (Table 3). Emergency surgery and nondegenerative diseases were assigned high scores of 5 points, whereas increasing age and performance status were assigned higher scores. Table 3.Point Loading on the Top Five Factors.FeaturePointsEmergency surgery5Non-degenerative disease5Spinal trauma2Age group <60 years0Age group 60-69 years1Age group 70-79 years5Age group ≥80 years6PS (0–4)0–4Because a zero-start scoring approach was used, the negative coefficient for degenerative disease was transformed into positive points for non-degenerative disease.PS: performance status Point Loading on the Top Five Factors. Because a zero-start scoring approach was used, the negative coefficient for degenerative disease was transformed into positive points for non-degenerative disease. PS: performance status Risk stratification and predicted probabilities Based on the calculated total scores, participants were classified into five risk tiers (Low, Average, Moderate, Elevated, and High) (Table 4). The actual incidence rate was 0.23% in the group with scores ranging from 0 to 2 points (low risk) and 6.67% in the group with scores ranging from 14 to 22 points (high risk), indicating an increasing incidence rate with higher scores. Figure 1 provides a visual summary of the JSSR-DB RC3S, including the scoring items, risk category ranges, and ECOG performance status definitions. Table 4.Risk Stratification and Predicted/Observed Probabilities.Risk categoryScore rangeAverage predicted riskObserved riskNLow0-20.40%0.23%2,144Average3-60.87%0.97%2,689Moderate7-81.26%1.68%1,782Elevated9-132.69%2.60%1,460High14-227.30%6.67%855 Risk Stratification and Predicted/Observed Probabilities. Practical one-page scoring aid for the JSSR-DB RC3S. This figure summarizes the score allocation for each predictor, the corresponding risk group thresholds, and the ECOG performance status definitions, allowing the JSSR-DB RC3S score to be calculated using a single visual reference. ECOG: Eastern Cooperative Oncology Group; JSSR-DB RC3S: Japanese Society for Spine Surgery and Related Research Database Risk Calculator simplified scoring system; PS: performance status Model performance The predictive performance of the scoring system was evaluated in the full analysis cohort (N=8,930); a calibration plot is shown in Figure 2. The AUC, which indicates the discrimination ability of the model, was 0.740, which is an acceptable value. After applying score-based recalibration, the O/E ratio, which indicates the fit between the predicted probability and the actual incidence, was 0.997. The Brier score was 0.0167. A low Brier score should be interpreted in the context of the low incidence rate (1.7%). Calibration was generally acceptable across the risk spectrum, although visual interpretation should remain cautious, given the low number of events. Five-fold cross-validation calibration plot. Calibration plot of the JSSR-DB RC3S for systemic complications based on out-of-fold predictions from stratified 5-fold CV. The squares indicate observed incidence rates within quantile-based bins, the solid line represents a LOWESS smoothing curve, and the dashed diagonal indicates perfect calibration. 5-fold CV: 5-fold cross-validation; JSSR-DB RC3S: Japanese Society for Spine Surgery and Related Research Database Risk Calculator simplified scoring system; LOWESS: locally weighted scatterplot smoothing Discussion This study developed and internally validated a simplified scoring system (JSSR-DB RC3S) to predict systemic complications after spine surgery using the nationwide JSSR-DB. The JSSR-DB RC3S was derived from the development process of the JSSR-DB RC and was intentionally designed as a highly transparent, clinically accessible scoring tool based on five key preoperative factors. Despite its simplified structure, the JSSR-DB RC3S demonstrated acceptable discriminatory ability (AUC=0.740) and good calibration (O/E ratio=0.997) in this cohort, indicating that clinically meaningful risk stratification can be achieved using real-world data from a Japanese national registry. The JSSR-DB RC3S may serve as a practical bedside tool, translating the JSSR-DB RC development findings into a simple point-based score that can support routine perioperative risk assessment. Clinical interpretation of selected predictors The top five predictors selected via LASSO-based variable selection were clinically valid and well aligned with perioperative clinical experience. Age and performance status represent patient frailty and baseline physiological reserve, both of which are central determinants of vulnerability to postoperative systemic complications. Notably, the increased risk associated with advancing age was more pronounced in patients aged ≥70 years, supporting the importance of age-related vulnerability in systemic complications following spine surgery. Similarly, worse performance status was associated with a stepwise increase in risk, reaffirming the importance of functional status in preoperative assessment. Emergency surgery and spinal trauma were also selected as risk factors for systemic complications. This aligns with clinical reality, where these cases often involve significant physiological stress and limited opportunities for preoperative optimization. Conversely, degenerative diseases were inversely associated with systemic complications. In the context of Japanese spinal care, this likely reflects the fact that degenerative disease surgeries constitute the majority of cases and are often performed under relatively stable, planned perioperative management. Thus, the selected predictors appear to be both statistically and clinically reasonable in the context of Japanese spine care. Comparison with existing models As shown in Table 511-14), the performance of the JSSR-DB RC3S was compared with that of representative existing risk prediction models for spine surgery, including SpineSage, ACS-NSQIP, and NZSpine. The JSSR-DB RC3S demonstrated an AUC of 0.740, which falls within a clinically meaningful range and was broadly comparable to those of previously reported models. These models were developed in different countries, targeting slightly different outcomes and employing distinct validation frameworks. Therefore, direct conclusions on superiority cannot be drawn solely from simple numerical comparisons. Instead, Table 511-14) indicates that the JSSR-DB RC3S, derived from a Japanese national registry and constructed as a transparent five-factor score, performs within the range reported for existing models. The JSSR-DB RC3S is not intended to replace existing models but to broaden the available options for clinically interpretable risk assessment in spine surgery. Table 5.Performance Comparison of Spine Surgery Complication Prediction Models.ModelCountryOutcomeAUCValidationFeaturesJSSR-DB RC3S (this study)JapanSystemic complications0.740Internal5-factor simplified scoreSpineSage11, 12)USAMajor complications0.77813)ExternalWeb-based, multiple outcomesACS-NSQIP1, 2)USAMajor complications0.67513)ExternalWeb-based, multiple outcomes, spine application of general surgery modelNZSpine14)NZMajor complications0.7614)InternalSpreadsheet-basedAbbreviations: ACS-NSQIP: American College of Surgeons National Surgical Quality Improvement Program; AUC: area under the curve; JSSR-DB RC3S: Japanese Society for Spine Surgery and Related Research Database Risk Calculator simplified scoring system; NZ: New Zealand; USA: United States of America Performance Comparison of Spine Surgery Complication Prediction Models. Abbreviations: ACS-NSQIP: American College of Surgeons National Surgical Quality Improvement Program; AUC: area under the curve; JSSR-DB RC3S: Japanese Society for Spine Surgery and Related Research Database Risk Calculator simplified scoring system; NZ: New Zealand; USA: United States of America An additional strength of the JSSR-DB RC3S is that it was derived from domestic real-world data. Existing international models have contributed significantly to perioperative risk assessment and remain important reference standards. However, the results of this study demonstrate that a model developed from Japanese registry data can achieve comparable discriminatory performance while reflecting the patient characteristics, disease composition, and perioperative management characteristics specific to Japanese spinal care. This supports the value of developing JSSR-DB–based risk assessment tools for Japanese spine practice. The clinical role of the JSSR-DB RC3S The JSSR-DB RC3S was developed as a point-based score that can be readily used in routine perioperative communication. Using a small number of key factors derived from the JSSR-DB RC development process offers a simple framework for estimating the risk of systemic complications and helps translate registry-based analyses into everyday clinical discussion. The gradient in observed incidence rates across risk strata supports this role. In this cohort, observed incidence rates increased from 0.23% in the low-risk group (0-2 points) to 6.67% in the high-risk group (14-22 points), demonstrating that the score provides a clinically recognizable stratification across the entire risk spectrum. Such categorization can support structured informed consent, perioperative deliberation, and shared understanding of the relative risk between clinicians and patients. These score categories are intended to support communication and risk framing rather than to serve as strict decision thresholds. As the RC3S is directly derived from the JSSR-DB RC development process, it may serve as a practical, simplified score and may also help clinicians become familiar with JSSR-DB–based risk estimation. Positioning of the RC3S within the JSSR-DB RC framework The JSSR-DB RC3S should be viewed as an initial simplified form of the JSSR-DB RC, which is being developed as a more comprehensive prediction platform that incorporates a larger number of variables and targets multiple clinically important outcomes. In this context, the JSSR-DB RC3S represents a simplified, transparent form of the same overall development approach. Accordingly, the JSSR-DB RC3S is best viewed not as a replacement for the JSSR-DB RC, but rather as its initial clinical version. By showing that JSSR-DB data can be translated into a practical risk tool, the JSSR-DB RC3S may help promote clinical familiarity with the JSSR-DB RC before wider implementation. Thus, this study supports the clinical usefulness of the simplified score and suggests the potential value of the JSSR-DB RC as a broader registry-based risk assessment tool. Limitations This study has several limitations. First, although this model showed acceptable performance in internal validation, the lack of external validation remains an important limitation. Because this model was developed using cases from participating JSSR-related institutions, its performance may not be fully generalizable to hospitals with different institutional characteristics, such as smaller or non-teaching centers. Future temporal validation using subsequently accumulated JOANR/JSSR-DB data, as well as prospective and external validation in independent cohorts, will be necessary before broader implementation. Depending on the results of these studies, recalibration or model updating may also be required. Second, the incidence rate for systemic complications was low (1.7%), requiring caution when interpreting calibration-related metrics, including the Brier score. Third, as with other registry studies, underreporting of complications may have affected observed incidence frequencies and potentially influenced the relationship between predicted values and actual incidence rates. Furthermore, the JSSR-DB RC3S was constructed as an intentionally simplified point-based score derived from key predictors identified during the JSSR-DB RC development process. Therefore, the JSSR-DB RC3S should not be interpreted as the final or most comprehensive form of JSSR-DB–based risk prediction. Rather, its purpose is to provide an interpretable, clinically accessible, simplified version, while more detailed risk estimation remains the goal of the ongoing JSSR-DB RC development. Conclusions The JSSR-DB RC3S is a transparent, clinically interpretable, simplified scoring system derived from the JSSR-DB RC development process and internally validated using Japanese nationwide registry data. Its performance falls within the range of existing spine surgery risk models, supporting the clinical utility of JSSR-DB–based risk stratification for systemic complications. More importantly, the JSSR-DB RC3S may help support future clinical use of the more comprehensive JSSR-DB RC. Conflicts of Interest: The authors declare that there are no relevant conflicts of interest. Author Contributions: S.I. wrote and prepared the manuscript. All authors participated in the development discussion. All authors have read, reviewed, and approved the manuscript. Ethical Approval: This study was approved by the Japanese Orthopaedic Association Ethics Committee (Approval code: R5-0003). Informed Consent: The requirement for individual informed consent was waived because this study used anonymized registry data collected under prior comprehensive ethical approval. Disclaimer: Nobuyuki Fujita, Satoshi Maki, Hiroaki Nakashima, Shinji Takahashi, and Kei Watanabe are among the Editors of Spine Surgery and Related Research and members of the journal's Editorial Committee. They were not involved in the editorial evaluation or decision to accept this article for publication at all.