Hospital level does not influence 30-day in-hospital mortality in road traffic accident hospitalisations - a nationwide registry study utilising Explainable AI (XAI) and the ICD-10 based injury severity score (ICISS)

Hospital level does not influence 30-day in-hospital mortality in road traffic accident hospitalisations - a nationwide registry study utilising Explainable AI (XAI) and the ICD-10 based injury severity score (ICISS) | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Hospital level does not influence 30-day in-hospital mortality in road traffic accident hospitalisations - a nationwide registry study utilising Explainable AI (XAI) and the ICD-10 based injury severity score (ICISS) Viktor Ydenius, Sebastian Djerf, Mats Fredrikson, Robert Larsen, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6628847/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 21 Nov, 2025 Read the published version in Scientific Reports → Version 1 posted 10 You are reading this latest preprint version Abstract Background: Globally, road traffic accidents (RTAs) remain a leading cause of mortality, particularly among individuals aged 15–30 years. Sweden has been at the forefront of traffic safety, but in-hospital care is critical in determining outcomes following RTAs. Guided by North American data demonstrating improved survival rates at trauma centres, the Swedish healthcare system is shifting towards trauma centralisation. However, comprehensive national data specific to Sweden remain underexplored. The unique demographic characteristics of Sweden, including vast, sparsely populated regions, distinguish it from other Western nations, complicating direct comparisons. Methods: The epidemiology and risk factors for 30-day mortality from RTAs in Sweden were investigated for 95,954 hospital admissions from 2008 to 2021. The ICD-based injury severity score (ICISS), age, sex, the Charlson comorbidity index (CCI), the year of the event, and hospital level were examined using an explainable AI (XAI) and Logistic regression. Results: The most important factors for 30-day mortality were, in decreasing importance, ICISS, age, CCI, event year, hospital level, and sex. There was a clear trend towards centralising RTA care, with Level 1 hospitals catering to the most critically injured patients. In parallel, however, the hospital level did not affect risk-adjusted traffic-related mortality. XAI enhances mortality prediction over logistic regression, confirming these findings. Discussion: This study presents the most extensive analysis of in-hospital outcomes for road traffic accidents (RTAs) in Europe to date. Factors such as ICISS, age, sex and the CCI had anticipated effects on outcome. Overall treatment outcome, measured as mortality improved over time. Conclusion: The presumption that external validity for trauma centralisation is applicable in the Scandinavian trauma context is not convincing. These findings make it essential to investigate further the trauma organisation given Scandinavian prerequisites so that time to hospital is not sacrificed for the type of hospital. Health sciences/Health care Health sciences/Risk factors Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 1. Introduction Globally, traffic-related injuries are still a major cause of death, especially among the age group of 15–30 years. It causes human suffering and a societal burden, costing some countries up to 5% of their gross domestic product annually [ 1 ]. The national ”Vision Zero” approach, with improvements in road and vehicle safety, presently puts Sweden in the top echelon of reducing traffic-related mortality internationally, with 2.8 deaths per year and 100,000 inhabitants in comparison to a European average and a world average of 9.3 and 17.4 deaths per year and 100,000 inhabitants, respectively [ 2 ]. The first aim of this study is to examine to what extent the decrease in overall traffic-related mortality is due to fewer accidents, fewer injuries or better treatment outcomes. Age, sex and comorbidity (measured as Charlson Comorbidity Index (CCI)) are shown to be independent factors for injury outcome [ 3 – 5 ]. When performing risk adjustment for the severity of the injury, the ICD-based injury severity score (ICISS), first developed by Osler et al. in 1996 [ 6 ], has been seen as an improvement of the consensus-based Injury Severity Score (ISS), and as of today the most accurate risk adjustment method for trauma research [ 7 , 8 ]. The ICISS takes all the patients’ injuries into account improving the accuracy in predicting injury severity. It is also more accessible in the national context as ICD-coding is enforced by law in each patient´s medical record in comparison to the Abbreviated Injury Scale (AIS) registration for which there is no consistent organization. An adaptation of the original ICISS to use the ICD-10 system have been shown to further improve its estimation [ 7 , 8 ] and its mortality predicament shows robustness in both the European and Swedish trauma setting [ 3 ]. The research group has adopted ICISS in earlier publications [ 3 , 4 , 9 ]. However, basing the ICISS value on a combined trauma cohort without regard to the difference in demography and trauma mechanism (Traffic, Fall and Assault) may lessen the performance of the mortality prediction model when applied to a particular subset, such as the traffic group. Therefore, the second aim of this study is to calculate ICISS directly at the traffic subgroup level, which has not been done before, to enhance the performance of the logistic regression model to refine outcome discussions. The influence of hospital type on traffic survival data is inconclusive. In contrast to both North American [ 10 – 12 ] and Finnish reports [ 13 ], our earlier study [ 4 ] indicate that hospital type does not seem to influence risk-adjusted traffic-related mortality in the Swedish setting. In that study, the data was collected between 2001 and 2011. It is now important to put this finding in a more current context with new data retrieval, especially as trauma centralisation has in later years been brought to the fore also and not least as a survival advantage at Level 1 hospitals was claimed recently [ 14 ]. The method of choice to examine large registry data has been multivariable logistic regression (LR). However, Artificial intelligence (AI), especially neural network methods, has recently been successfully applied in medical data analysis [ 15 ]. XGBoost, an ensemble learning algorithm, has become an important tool in scientific research for its exceptional performance in predictive modelling and classification tasks [ 16 ]. This study’s third and final aim is to examine the potential of XGBoost in improving the ICISS mortality prediction model. 2. Method 2.1 Patient cohort The National Patient Registry (NPR) covers all national hospital admissions from 1987 onward and was used to retrieve trauma-related hospital admissions between 2008 and 2021. By using each patient’s unique personal identification number (PIN), the NPR was linked with the Cause of Death Registry, covering the deaths of all identified Swedish citizens. Based upon the International Classification of Diseases version 10 (ICD-10), trauma diagnoses (S00–T79) and external cause of injury codes (E-codes) (V01–Y98) were selected to generate the original database. ICD-10 codes T360-T790, except T689 hypothermia, were excluded as this code interval includes poisoning, intoxication and unspecific adverse effects outside our trauma scope. The E-codes V01-V89, excluding V80-82 and V88, were used to define the subgroup of traffic-related trauma. The ICD-based injury severity score (ICISS) was calculated on both the original calculation database (including subgroups of assault and fall) and specifically on the subset of traffic-related trauma (defined as ”Traffic database for calculation” in Figure A1 in the Appendix). The ICISS values in the analyses were based on the latter. Hospital admissions due to other traumas or those found in multiple groups were excluded from the analysis. Patients admitted to hospitals that did not meet the hospital categorisation criteria (see below) were also excluded—a total of 95,954 hospital admissions made up the final study population of traffic-related trauma (Figure A1 in Appendix). 2.2 Hospital categorisation The Swedish PeriOperative Registry (SPOR) classification of hospital level, recognised by the Swedish Department of Health and Welfare (Socialstyrelsen), the Swedish Counties and Regions, the Regional Insurance Company and the Swedish Association of Anaesthesia and Intensive Care, was used as the basis for hospital categorisation in this study due to its wide acceptance in the Swedish healthcare system [ 4 , 17 ]. Depending on the intensive care unit ability, the degree of specialised care, access to laboratories and radiology, and whether research and education were part of the hospital’s mission, three categories were used: Level 1 hospitals corresponding to university hospitals, Level 2 hospitals corresponding to regional hospitals and Level 3 hospitals corresponding to county hospitals. See the Appendix for in-depth inclusion criteria. For Level 3 hospitals, the retrieved SPOR classification list could not be fully implemented as it did not cover all Level 3 hospitals found in the database [ 17 ]. Thus, Level 3 hospitals missing in SPOR but located within the trauma database were included through face validity by direct communication with the head of the hospital, the head of operations and county administrators or through an in-depth web search. If emergency room or traffic-related hospital admissions throughout the study period could be subjectively confirmed, the hospital was excluded from the analysis (See Appendix for details). 2.3 Injury severity The traffic database (Figure A1 in Appendix) calculated the diagnosis-specific survival probability (DSP) for each ICD-10 diagnosis, giving the likelihood of survival for each injury. Duplicate ICD-10 codes were removed before the DSP calculation. Using an inclusive approach, an ICISS value was calculated for each hospital admission as the product of the DSPs for the ICD-10 diagnoses. As ICISS was highly skewed, the transformation log 10 (1 − ICISS + ϵ ) was used with ϵ = 0.001 for all regression analyses using ICISS. 2.4 Comorbidities The CCI adapted to the ICD-10 system, according to Glasheen et al., was used [ 18 ]. Translation of ICD-9 codes was done as described by Glasheen et al. [ 18 ], and a direct comparison was made with the Swedish National Board of Health and Welfare [ 19 ] to achieve coherent coding with the ICD-9 codes used in Sweden at the time. 2.5 Statistics We used standard multivariable logistic regression and XAI, combining XGBooost and SHAP as implemented in the R packages XGBoost and SHAP [ 20 – 22 ]. The regression models were evaluated with the area under the curve (AUC), and differences in these were assessed using the DeLong test [ 23 ]. The data was divided into an 80% training set and a 20% test set. A grid search on the training set was carried out to find good hyperparameters of XGBoost. scale_pos_weight was set to 99. The ICISS scores were calculated using the training set. XGBoost was then applied to the training set and evaluated on the test set. The XGBoost was interpreted using SHapley Additive exPlanations, SHAP. 2.6 Ethics The Swedish Ethical Review Authority approved the study protocol (registration nos. 2019–04852 and 2021–04333), and the study was conducted according to the 1964 Declaration of Helsinki. In the ethics application and the formal ethics committee approval informed consent from the patients was waived. 3. Results 3.1 Patient characteristics The final study population comprised 95,954 hospital admissions between 2008 and 2021, with a 30-day mortality of 1.16% and a median ICISS of 0.91 (IQR 0.11). The ICISS distribution over hospital levels is shown in Table 1. The ages of the admitted patients ranged between 18 and 103, with a median and mean of 48 years (IQR 33 years, SD 19 years). Men accounted for 67% of the study population and were overrepresented in all age groups. See Table 1 and Fig. 2. The annual incidence of hospitalised RTAs is shown in Fig. 1. Amongst fatal accidents, over 46% were associated with traumatic brain injury (TBI defined as ICD-10 codes S061-S069), and in the elderly cohort (≥ 65 years), over 40% of the fatalities were associated with TBI. The Appendix shows data over time in Table A1 . 3.2 Regression models Figure A2 in the Appendix shows the ROCs and their pertaining AUCs for the LR and XAI 30-day mortality prediction models (p < 0.05, deLong’s test). Calibration, as measured with Brier score, was 1.00 for LR and 0.09 for XAI ( p-value < 0.05). Odds ratios from the multivariable (ICISS, age, sex, CCI, event year, hospital level) logistic regression models of 30-day mortality are shown in Table 2. The most important factors for 30-day mortality were, in decreasing importance, ICISS, age, CCI, event year, hospital level, and sex, as shown in the summary plot in Fig. 3. Decreasing ICISS independently increases mortality as seen in Table 2 and Figs. 4 and 5. ICISS decreased significantly over time, but the magnitude of this was minor, as seen in Table A1 . Increasing age independently increases mortality as seen in Table 2 and Figs. 4 and 5. As seen in Table A1 , age increased markedly over the study period. Increasing CCI independently increases mortality as seen in Table 2 and Figs. 4 and 5. As seen in Table A1 , CCI varied over the study period. Thirty-day mortality was highest for level 1 (primary admission) hospitals, followed by level 2 and level 3. Hospital level was not a clear factor for mortality for the most severely injured and non-transfer patients. Subgroup analysis for transfer patients only (data not shown) showed no significant difference between hospital levels in the LR model. Similarly, the XAI model shows higher mortality for level 1 hospitals with similar mortality for levels 2 and 3 (for all patients and the most severely injured alike, see Figs. 4 and 5). The use of Level 2 hospitals increased over time, mainly due to a lower fraction of patients treated at Level 3 hospitals. See Table A1 . Female sex was also independently associated with lower 30-day mortality for all hospital admissions and non-transfer cases. See Table 2 and Figs. 4 and 5). The adjusted 30-day mortality decreased over the study period for all hospital admissions and non-transfer cases (see Table 2 and Figs. 4 and 5). 4. Discussion By applying XAI to RTA data and ICD-10-based injury severity scoring (ICISS), we showed that the XAI model outperformed standard multivariable logistic regression in estimating 30-day mortality after RTA hospitalisations. The improved XAI model allowed for better characterisation of the risk factors for RTA-associated 30-day mortality. The XAI model’s most important factors for predicting 30-day mortality were, in decreasing order of influence, ICISS, age, CCI, event year, sex, and hospital level. As anticipated, the primary determinant of 30-day mortality is injury severity. The ICISS values in this study are higher than in Osler et al. [ 6 ] (median 0.93) and Larsen et al. [ 9 ] (median 0.95). This difference may arise from using the RTA cohort rather than all trauma cases to estimate ICISS. Our higher ICISS values are also highlighted by our higher median ICISS of 0.91 for fatal accidents, which is 0.45 in Osler’s study [ 6 ] and 0.72 in Ydenius et al. [ 4 ]. The ageing population [ 25 ] in the RTA cohort is thought to be a contributing factor [ 26 ]. In Osler’s study [ 6 ], and in Ydenius et al. [ 4 ], 92% and 77% were 55 years or younger, respectively. In our study, the median ICISS for the 18-25-year-olds was 0.877 and 0.935 for the ≥ 75-year-olds (data not shown). Table 1 shows that severe cases with ICISS below 0.85 are rare, only accounting for around 3% of the hospital admissions. This raises the question of whether an ICISS cutoff of 0.85 for severe RTA is too strict, as many fatalities occur at higher ICISS levels in this dataset. An ICISS threshold of < 0.941 has been previously suggested for severe injury [ 27 ]. The analysis of demographics over time (see Table A1 in the Appendix) shows that the median age of RTA victims increased over time, that median ICISS decreased over time (more severely injured), CCI varied significantly and that fewer patients were treated in Level 3 hospitals in favour of Level 2. These factors should most likely result in a higher mortality. As seen in Table 2 and the bottom right panels of Figs. 4 and 5, the corrected mortality risk decreased with the year of the trauma. Surprisingly, the values were low and high for 2014 and 2018, respectively, for the most severely injured (Fig. 5). These values agree with the mortality observed in the RTA statistics of Transport Analysis (TRAFA) [ 28 ]. Many elderly and young individuals suffer fatal RTAs [ 29 ]. Young suffering RTAs may be due to risk-prone behaviour [ 30 , 31 ]. This fact is also reflected by the low median age of traffic-related admissions, 47 years old (IQR 32 years) in our data. Figure 2 shows that young men are overrepresented. However, the median age for fatal accidents in our data is high, 72 years old (IQR 31 years). This aligns with our findings that mortality increases with age, as seen in Table 2 and Fig. 4. It has been suggested that younger individuals who are exposed to fatal RTAs die at the trauma site or during hospital transfer. Earlier findings show that females have a survival advantage over men [ 32 ], and comorbidity influences 30-day mortality [ 4 , 9 ]. However, the small mortality effect of sex and comorbidity is not significant in the LR model of the severely injured, possibly due to few observations. In contrast, the XAI model indicates a small sex difference as well as an increasing mortality for CCI 1 as compared with CCI 0 for the severely injured. Notably, very few persons with CCI above 6 (CCI ranges from 0 to 20) are admitted to hospitals with RTA-associated trauma (see Fig. 4), possibly reflecting that persons with many comorbidities do not get exposed to traffic [ 33 ]. The most severely injured patients are admitted to Level 1 hospitals, where they die to a greater extent than at Level 2 and 3 hospitals (see Tables 1 and 2 and Figs. 4 and 5), consistent with Ydenius et al. [ 4 ]. However, the hospital level does not clearly impact overall mortality in the LR model of the most severely injured or the non-transfer patients. The XAI model shows a much clearer effect on the hospital level than the LR method. Earlier criticisms about attributing outcomes to the admitting hospital have been addressed by categorising patients based on transfer status. The notion that the initial trauma care, the so-called ”Golden Hour”, is the primary determinant of trauma outcomes is being questioned [ 34 ]. North American studies advocate trauma centre referrals rather than focusing on early care [ 35 , 36 ]. Our results of a lack of survival advantage of Level 1 centra raise whether Sweden should prioritise trauma centre referral or hospital accessibility. Candefjord et al. [ 14 ] argue that reaching trauma centres within one hour in Sweden is possible if helicopter transports are considered. However, helicopter ambulance resources are currently very limited in Sweden. An alternative is ”bringing the emergency room to the site” with well-equipped ambulances [ 37 ]. Transfers complicate the analysis of hospital-level impact. Therefore, patients were categorised as having no transfers or having transfers (based on the first transfer only). The majority of hospital admissions (91%) are never transferred. Early deaths preclude transfers, introducing a selection bias. Future work should include detailed analysis with formal SHAP score statistics to understand the importance of the hospital level. 4.1 Improving Methods Previous studies suggest that XGBoost improves trauma mortality prediction models based on logistic regression on the Injury Severity Score (ISS), the Trauma Mortality Prediction Model TMPM-ICD10 [ 38 ], and the Trauma and Injury Severity Score (TRISS) [ 39 ]. The novel approach of calculating ICISS solely on RTAs and using a machine learning model improves the risk adjustment and mortality prediction (AUC increases from 0.90 for LR to 0.92 for XAI). To our knowledge, the use of XAI for ICISS data is new and gives a better understanding of the complex relationships found in our data. 4.2 Limitations Although the SweTrau registry was considered a data source, SweTrau was only initiated after the start of this study. Consequently, the NPR was chosen. Legal requirements for ICD coding minimise the risk of missing data. While incomplete or inaccurate coding remains a concern, economic incentives for correct coding and NPR validation ensure good precision for ICD codes up to the fourth position [ 3 , 9 ]. By extension, the use of the International Classification of Diseases Injury Severity Score (ICISS) is a stalwart method of risk adjustment in extensive data sets [ 7 ], preferred over the traditional ISS method [ 40 ]. Ydenius et al. have successfully used the ICISS method and are confirming it in a more current setting. The dataset’s representation of the ”real-world” population and its size and inclusion period is a notable strength. Patients dying at the trauma site or during transport to the hospital were not included in this study. TRAFA [ 28 ] suggests a higher survival in the metropolitan areas Stockholm, Gothenburg and Malmö, and conversely, high mortality in the less densely populated northern parts of the country, implying that transport time is a factor for traffic-related mortality. However, it is essential to remember where in the care chain this study is set and as the factual hospital admissions are set for inclusion, transport time as a parameter opts for a different study. Using an anatomical risk-adjustment tool, dynamics in physiological variables affecting outcomes may be overlooked [ 41 – 43 ]. Physiological parameters upon emergency room arrival are, in turn, influenced by pre-hospital stabilisation. Nevertheless, DSP calculation retrospectively includes the impact of physiological parameters, given severe injuries often coincide with deranged vital signs. The hospital-level analysis is complicated by a selection bias, as patients who pass away quickly (the most severely injured) may not be fit for a secondary transport up the hospital-level chain. The group of transported patients, around 9% of the entire data set, is also very diverse as the time of the transport to another hospital is not defined. Thus, transportation may take place in almost immediate connection to the trauma ictus or weeks after the initial admission. One must tread carefully when making conclusions from this group for obvious reasons. The possibility of separating this subgroup of patients was used to benefit the study. The hospital level can be studied independently as a risk factor for 30-day mortality by studying the non-transfer cases. XAI further helps us to visualise the data, whereas traditional statistics give the confidence to make assertions from the SHAP representation. The fact that the XAI approach does not currently allow for formal statistical testing is indeed a limit but using it in symbiosis with traditional logistic regression still makes it useful. Plans are underway to develop computer-intensive methods for formal statistical testing using XGBoost and SHAP, which are outside the scope of the current paper. 5. Conclusion An explainable AI (XAI) model for 30-day mortality of hospitalised RTA victims based on ICD codes (ICISS), age, CCI, event year, hospital level, and sex (order based on importance) showed better discrimination and calibration than a corresponding LR-based model. The adjusted mortality decreased from 2008 to 2021 despite increasing injury severity and age (median increased from 44 years to 52 years) of RTA victims, suggesting advancements in treatment efficacy over time. Factors such as sex and comorbidities impact 30-day mortality rates but do not achieve statistical significance among severely injured patients (ICISS≤0.85). The trend of centralising treatment at Level 1 hospitals remains prevalent, correlating with a decreasing order of injury severity across hospital levels. Nevertheless, upon adjusting for risk factors, hospital level did not significantly influence risk outcomes, especially considering the most severely injured. A key strength of this study is the stratification of hospitalisations based on transfer status and the stalwart method of ICISS for injury severity risk adjustment. A more comprehensive evaluation of hospital-level impact, particularly the direct utilisation of Level 1 facilities, should incorporate transport time as a variable. The XAI method should also be extended to include a formal statistical testing framework. In conclusion, this research shows that prevailing assumptions regarding the advantages of trauma centralisation are not given to be beneficial for Swedish RTA victims. The datasets used in the current study are available from the corresponding author on reasonable request. Declarations Acknowledgements/Funding Regional research support, Region Skåne #2022-1284; Governmental funding of clinical research within the Swedish National Health Service (ALF) #2022:YF0009 and #2022-0075; Crafoord Foundation grant number #2021-0833; Lions Skåne research grants; Skåne University Hospital grants; Swedish Heart and Lung Foundation (HLF) #2022-0352 and #2022-0458.The Region Ostergötland and Linköping University supported the study.¨ Funding sources were not involved in the manuscript’s writing or the decision to submit it for publication. The article was written without any commission. No pharmaceutical companies or other agencies have been involved in writing the submitted manuscript. References Nations, U.: Improving global road safety. General Assembly (2020) https://doi.org/https://documents.un.org/doc/undoc/gen/n20/226/30/pdf/ n2022630.pdf?OpenElement . Accessed: 2023-08-30 Joseph, A.: Road trauma–an ongoing challenge in injury prevention and postcrash care. Injury 49 (9), 1637–1638 (2018) Larsen, R., B¨ackstr¨om, D., Fredrikson, M., Steinvall, I., Gedeborg, R., Sjoberg, F.: Decreased risk adjusted 30-day mortality for hospital admitted injuries: a multi-centre longitudinal study. Scandinavian journal of trauma, resuscitation and emergency medicine 26 (1), 1–8 (2018) Ydenius, V., Larsen, R., Steinvall, I., B¨ackstr¨om, D., Chew, M., Sj¨oberg, F.: Impact of hospital type on risk-adjusted, traffic-related 30-day mortality: a population-based registry study. Burns & Trauma 9 , 051 (2021) Sundararajan, V., Quan, H., Halfon, P., Fushimi, K., Luthi, J.C., Burnand, B., Ghali, W.A.: Cross-national comparative performance of three versions of the ICD-10 Charlson index. Med Care 45 (12), 1210–1215 (2007) Osler, T., Rutledge, R., Deis, J., Bedrick, E.: Iciss: an international classification of disease-9 based injury severity score. Journal of Trauma and Acute Care Surgery 41 (3), 380–388 (1996) Gedeborg, R., Warner, M., Chen, L.-H., Gulliver, P., Cryer, C., Robitaille, Y., Bauer, R., Ubeda, C., Lauritsen, J., Harrison, J., et al. : Internationally comparable diagnosis-specific survival probabilities for calculation of the ICD-10–based injury severity score. Journal of Trauma and Acute Care Surgery 76 (2), 358–365 (2014) Stephenson, S., Henley, G., Harrison, J.E., Langley, J.D.: Diagnosis based injury severity scaling: investigation of a method using australian and new zealand hospitalisations. Injury prevention 10 (6), 379–383 (2004) Larsen, R.: Risk-Adjustment for Swedish In-Hospital Trauma Mortality Using International Classification of Disease Injury Severity Score (ICISS): Issues with Description and Methods vol. 1660. Link¨oping University Electronic Press, ??? (2019) Demetriades, D., Martin, M., Salim, A., Rhee, P., Brown, C., Chan, L.: The effect of trauma center designation and trauma volume on outcome in specific severe injuries. Annals of surgery 242 (4), 512 (2005) Nathens, A.B., Jurkovich, G.J., Maier, R.V., Grossman, D.C., MacKenzie, E.J., Moore, M., Rivara, F.P.: Relationship between trauma center volume and outcomes. Jama 285 (9), 1164–1171 (2001) MacKenzie, E.J., Rivara, F.P., Jurkovich, G.J., Nathens, A.B., Frey, K.P., Egleston, B.L., Salkever, D.S., Scharfstein, D.O.: A national evaluation of the effect of trauma-center care on mortality. New England Journal of Medicine 354 (4), 366–378 (2006) Ala-Kokko, T., Ohtonen, P., Koskenkari, J., Laurila, J.: Improved outcome after trauma care in university-level intensive care units. Acta anaesthesiologica scandinavica 53 (10), 1251–1256 (2009) Candefjord, S., Asker, L., Caragounis, E.-C.: Mortality of trauma patients treated at trauma centers compared to non-trauma centers in sweden: a retrospective study. European journal of trauma and emergency surgery, 1–12 (2020) Holmgren, G., Andersson, P., Jakobsson, A., Frigyesi, A.: Artificial neural networks improve and simplify intensive care mortality prognostication: a national cohort study of 217,289 first-time intensive care unit admissions. Journal of intensive care 7 (1), 1–8 (2019) Chen, T., Guestrin, C.: Xgboost: A scalable tree boosting system. In: Proceedings of the 22nd Acm Sigkdd International Conference on Knowledge Discovery and Data Mining, pp. 785–794 (2016) Register, S.P.: Sjukhus anslutna till spor. Svenskt PeriOperativt Register (2023) https://doi.org/https://spor.se/om-spor-landingpage/anslutna-kliniker/ . Accessed: 2023-01-17 Glasheen, W.P., Cordier, T., Gumpina, R., Haugh, G., Davis, J., Renda, A.: Charlson comorbidity index: ICD-9 update and ICD-10 translation. American Health & Drug Benefits 12 (4), 188 (2019) Socialstyrelsen: ICD9 klassifikation av sjukdomar 1987 KS87. Socialstyrelsen (2022) https://doi.org/https://www.socialstyrelsen.se/globalassets/ sharepoint-dokument/dokument-webb/klassifikationer-och-koder/ icd-9-klassifikation-av-sjukdomar-1987-ks87.xls . Accessed: 2022-09-02 R Core Team: R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria (2021). R Foundation for Statistical Computing. https://www.R-project.org/ Chen, T., Guestrin, C.: XGBoost: A scalable tree boosting system. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. KDD ’16, pp. 785–794. ACM, New York, NY, USA (2016). https://doi.org/10.1145/2939672.2939785 . http://doi.acm.org/10.1145/2939672. 2939785 Lundberg, S.M., Lee, S.-I.: A unified approach to interpreting model predictions. In: Guyon, I., Luxburg, U.V., Bengio, S., Wallach, H., Fergus, R., Vishwanathan, S., Garnett, R. (eds.) Advances in Neural Information Processing Systems 30, pp. 4765–4774. Curran Associates, Inc., ??? (2017). http://papers.nips.cc/paper/ 7062-a-unified-approach-to-interpreting-model-predictions.pdf DeLong, E.R., DeLong, D.M., Clarke-Pearson, D.L.: Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics 44 (3), 837–845 (1988) SCB: Folkm¨angden den 1 november efter region, ˚alder och k¨on. ˚Ar 2002 - 2023Statistikdatabasen — statistikdatabasen.scb.se. https://www.statistikdatabasen. scb.se/pxweb/sv/ssd/STARTBEBE0101BE0101A/FolkmangdNov/table/ tableViewLayout1/. [Accessed 04-10-2024] Lau, L., Ajzenberg, H., Haas, B., Wong, C.L.: Trauma in the aging population: geriatric trauma pearls. Emergency Medicine Clinics 41 (1), 183–203 (2023) Pitta, L.S.R., Quintas, J.L., Trindade, I.O.A., Belchior, P., Gameiro, K.d.S.D., Gomes, C.M., N´obrega, O.T., Camargos, E.F.: Older drivers are at increased risk of fatal crash involvement: results of a systematic review and meta-analysis. Archives of gerontology and geriatrics 95 , 104414 (2021) Berecki-Gisolf, J., Fernando, D.T., D’Elia, A.: International classification of disease based injury severity score (iciss): A data linkage study of hospital and death data in victoria, australia. Injury 53 (3), 904–911 (2022) Analysis, T.: Road traffic injuries 2018. Transport Analysis (2018) https://doi.org/https://www.trafa.se/globalassets/statistik/vagtrafik/ vagtrafikskador/2018/vagtrafikskador-2018---blad.pdf . Accessed: 2023-10-10 Analysis, T.: Road traffic injuries 2021. Transport Analysis (2021) https://doi.org/https://www.trafa.se/globalassets/statistik/vagtrafik/ vagtrafikskador/2021/vagtrafikskador-2021---korr.-2022-05-15.pdf . Accessed: 2023-10-10 Turner, C., McClure, R.: Age and gender differences in risk-taking behaviour as an explanation for high incidence of motor vehicle crashes as a driver in young males. Injury control and safety promotion 10 (3), 123–130 (2003) Falk, B.: Do drivers become less risk-prone after answering a questionnaire on risky driving behaviour? Accident Analysis & Prevention 42 (1), 235–244 (2010) Larsen, R., B¨ackstr¨om, D., Fredrikson, M., Steinvall, I., Gedeborg, R., Sjoberg, F.: Female risk-adjusted survival advantage after injuries caused by falls, traffic or assault: a nationwide 11-year study. Scandinavian journal of trauma, resuscitation and emergency medicine 27 , 1–7 (2019) Marshall, S.C., Man-Son-Hing, M.: Multiple chronic medical conditions and associated driving risk: a systematic review. Traffic injury prevention 12 (2), 142–148 (2011) Rogers, F.B., Rittenhouse, K.J., Gross, B.W.: The golden hour in trauma: dogma or medical folklore? Injury 46 (4), 525–527 (2015) Newgard, C.D., Schmicker, R.H., Hedges, J.R., Trickett, J.P., Davis, D.P., Bulger, E.M., Aufderheide, T.P., Minei, J.P., Hata, J.S., Gubler, K.D., et al. : Emergency medical services intervals and survival in trauma: assessment of the “golden hour” in a north american prospective cohort. Annals of emergency medicine 55 (3), 235–246 (2010) Newgard, C.D., Meier, E.N., Bulger, E.M., Buick, J., Sheehan, K., Lin, S., Minei, J.P., Barnes-Mackey, R.A., Brasel, K., Investigators, R., et al. : Revisiting the “golden hour”: an evaluation of out-of-hospital time in shock and traumatic brain injury. Annals of emergency medicine 66 (1), 30–41 (2015) Strandqvist, E., Olheden, S., B¨ackman, A., J¨ornvall, H., B¨ackstr¨om, D.: Physician-staffed prehospital units: a retrospective follow-up from an urban area in scandinavia. International Journal of Emergency Medicine 16 (1), 43 (2023) Tran, Z., Zhang, W., Verma, A., Cook, A., Kim, D., Burruss, S., Ramezani, R., Benharash, P.: The derivation of an international classification of diseases, tenth revision–based trauma-related mortality model using machine learning. Journal of Trauma and Acute Care Surgery 92 (3), 561–566 (2022) Tran, Z., Verma, A., Wurdeman, T., Burruss, S., Mukherjee, K., Benharash, P.: ICD-10 based machine learning models outperform the trauma and injury severity score (triss) in survival prediction. Plos one 17 (10), 0276624 (2022) Gagne, M., Moore, L., Beaudoin, C., Kuimi, B.L.B., Sirois, M.-J.: Performance of international classification of diseases–based injury severity measures used to predict in-hospital mortality: a systematic review and meta-analysis. Journal of Trauma and Acute Care Surgery 80 (3), 419–426 (2016) Kang, M.W., Ko, S.Y., Song, S.W., Kim, W.J., Kang, Y.J., Kang, K.W., Park, H.S., Park, C.B., Kang, J.H., Bu, J.H., et al. : Prognostic accuracy of the quick sequential organ failure assessment for outcomes among patients with trauma in the emergency department: a comparison with the modified early warning score, revised trauma score, and injury severity score. Journal of Trauma and Injury 34 (1), 3–12 (2021) Jawa, R.S., Vosswinkel, J.A., McCormack, J.E., Huang, E.C., Thode Jr, H.C., Shapiro, M.J., Singer, A.J.: Risk assessment of the blunt trauma victim: the role of the quick sequential organ failure assessment score (qsofa). The American Journal of Surgery 214 (3), 397–401 (2017) Sadhwani, N., Ambore, V., Bakhshi, G.: Predictive value of quick sequential organ failure assessment (qsofa) score in risk assessment and outcome prediction in blunt trauma patients: A prospective observational study. Annals of Medicine and Surgery 74 , 103265 (2022) Tables Tables 1 and 2 are available in the Supplementary Files section Additional Declarations No competing interests reported. Supplementary Files Table1TrafficXAI.docx Table2TrafficXAI.docx TableA1AppendixATrafficXAI.docx TableA2AppendixATrafficXAI.docx Cite Share Download PDF Status: Published Journal Publication published 21 Nov, 2025 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 19 Jun, 2025 Reviews received at journal 18 Jun, 2025 Reviewers agreed at journal 05 Jun, 2025 Reviews received at journal 30 May, 2025 Reviewers agreed at journal 29 May, 2025 Reviewers invited by journal 29 May, 2025 Editor assigned by journal 29 May, 2025 Editor invited by journal 27 May, 2025 Submission checks completed at journal 24 May, 2025 First submitted to journal 09 May, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6628847","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":464716644,"identity":"828dd72f-ea13-403d-8aa0-3707da9109da","order_by":0,"name":"Viktor Ydenius","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA60lEQVRIiWNgGAWjYDACZjYo4wAY2ciw4VEM08LYgKQljYewFgYkLUBwmIegBoPjbOkPPu5hSOw73nvwMM+f8zx8EgnMHz7g03KY7WDjjGcMiTPPnEs4zNt2m4dNIoFNcgYeLZLN7I3NPAcYEjfcyDE4zNsA0cKMz3kILfffGAAddg6khfnzHzxa+JnZDkJt4QFqYTsA0sIgjc/7QC2JM2cckDCeeSbH4ODctmQeNp6HbZI9eLSw8R8z+PDhgI1s3/Ezxh/e/LGTk29PPvzhBz5rIEDCsQHBYWzApQwF2BOlahSMglEwCkYmAACx002A9Ybv4gAAAABJRU5ErkJggg==","orcid":"","institution":"Linköping University Hospital","correspondingAuthor":true,"prefix":"","firstName":"Viktor","middleName":"","lastName":"Ydenius","suffix":""},{"id":464716645,"identity":"a43f1cdf-148d-4c23-9a73-9f86d34fa3ac","order_by":1,"name":"Sebastian Djerf","email":"","orcid":"","institution":"Lund University","correspondingAuthor":false,"prefix":"","firstName":"Sebastian","middleName":"","lastName":"Djerf","suffix":""},{"id":464716646,"identity":"f898b0fa-cc33-49a1-ae08-769763a70403","order_by":2,"name":"Mats Fredrikson","email":"","orcid":"","institution":"Linköping University Hospital","correspondingAuthor":false,"prefix":"","firstName":"Mats","middleName":"","lastName":"Fredrikson","suffix":""},{"id":464716647,"identity":"528d573f-bc4f-41a3-963a-e2df3725cb94","order_by":3,"name":"Robert Larsen","email":"","orcid":"","institution":"Linköping University Hospital","correspondingAuthor":false,"prefix":"","firstName":"Robert","middleName":"","lastName":"Larsen","suffix":""},{"id":464716648,"identity":"d2f0ba31-b34a-4e42-b007-5057272dd1f8","order_by":4,"name":"Folke Sjöberg","email":"","orcid":"","institution":"Linköping University Hospital","correspondingAuthor":false,"prefix":"","firstName":"Folke","middleName":"","lastName":"Sjöberg","suffix":""},{"id":464716649,"identity":"7f85aeca-07a0-47e6-b03d-9033ca619503","order_by":5,"name":"Attila Frigyesi","email":"","orcid":"","institution":"Lund University","correspondingAuthor":false,"prefix":"","firstName":"Attila","middleName":"","lastName":"Frigyesi","suffix":""}],"badges":[],"createdAt":"2025-05-09 12:53:31","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6628847/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6628847/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-025-26519-7","type":"published","date":"2025-11-21T15:57:08+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":83815780,"identity":"4e27dfa5-61c7-4e1f-b92d-12dcd541d1f4","added_by":"auto","created_at":"2025-06-03 07:39:25","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":76125,"visible":true,"origin":"","legend":"\u003cp\u003eLegend not included with this version.\u003c/p\u003e","description":"","filename":"Figure1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6628847/v1/060a2acfb544cd0e0ca038a9.jpg"},{"id":83815795,"identity":"16c190f3-ef24-487f-a84d-5834fc4745a8","added_by":"auto","created_at":"2025-06-03 07:39:26","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":381211,"visible":true,"origin":"","legend":"\u003cp\u003eLegend not included with this version.\u003c/p\u003e","description":"","filename":"Figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-6628847/v1/0bdb2d43050b5cbba9cb9c35.png"},{"id":83817355,"identity":"7da8ef67-fd06-489a-8f8e-e8be2a06b4b0","added_by":"auto","created_at":"2025-06-03 07:55:25","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":21458,"visible":true,"origin":"","legend":"\u003cp\u003eLegend not included with this version.\u003c/p\u003e","description":"","filename":"Figure3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6628847/v1/b15c5d3185acf7879b12b6cf.jpg"},{"id":83815792,"identity":"89c68443-0e37-4732-b60a-c3cd5b102d8c","added_by":"auto","created_at":"2025-06-03 07:39:26","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":146706,"visible":true,"origin":"","legend":"\u003cp\u003eLegend not included with this version.\u003c/p\u003e","description":"","filename":"Figure4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6628847/v1/2b10821d4199a110b0fe53a6.jpg"},{"id":83815794,"identity":"7fbf03e8-4bd6-43ac-8f50-ee5dbd330bf9","added_by":"auto","created_at":"2025-06-03 07:39:26","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":142274,"visible":true,"origin":"","legend":"\u003cp\u003eLegend not included with this version.\u003c/p\u003e","description":"","filename":"Figure5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6628847/v1/f98c6e7088c56b8c6ca1f921.jpg"},{"id":96650125,"identity":"b0c2a65e-9651-467d-bfa9-79d7866f404a","added_by":"auto","created_at":"2025-11-24 16:08:23","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1269141,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6628847/v1/f5ab9d53-95dd-42ec-8e10-bdcbc40ef76a.pdf"},{"id":83815791,"identity":"27da4803-6559-4a52-adbe-7d98d86c6b20","added_by":"auto","created_at":"2025-06-03 07:39:26","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":232269,"visible":true,"origin":"","legend":"","description":"","filename":"Table1TrafficXAI.docx","url":"https://assets-eu.researchsquare.com/files/rs-6628847/v1/4a93e20fb72ab9e59166cf2e.docx"},{"id":83815784,"identity":"651214ea-90f7-4b1e-ac2c-d90963cb7edf","added_by":"auto","created_at":"2025-06-03 07:39:25","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":48939,"visible":true,"origin":"","legend":"","description":"","filename":"Table2TrafficXAI.docx","url":"https://assets-eu.researchsquare.com/files/rs-6628847/v1/bf19f30335b87e733c184528.docx"},{"id":83815790,"identity":"5ce73d9d-fb41-42ea-ba53-9fe2b832ec88","added_by":"auto","created_at":"2025-06-03 07:39:26","extension":"docx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":16896,"visible":true,"origin":"","legend":"","description":"","filename":"TableA1AppendixATrafficXAI.docx","url":"https://assets-eu.researchsquare.com/files/rs-6628847/v1/5f19305ddaed2192b15a7c2b.docx"},{"id":83816165,"identity":"d3c7d29d-876e-4fc2-b062-83f3584bdbfc","added_by":"auto","created_at":"2025-06-03 07:47:26","extension":"docx","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":42808,"visible":true,"origin":"","legend":"","description":"","filename":"TableA2AppendixATrafficXAI.docx","url":"https://assets-eu.researchsquare.com/files/rs-6628847/v1/f82b19c7bcd5272ea25dd1d7.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Hospital level does not influence 30-day in-hospital mortality in road traffic accident hospitalisations - a nationwide registry study utilising Explainable AI (XAI) and the ICD-10 based injury severity score (ICISS)","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eGlobally, traffic-related injuries are still a major cause of death, especially among the age group of 15\u0026ndash;30 years. It causes human suffering and a societal burden, costing some countries up to 5% of their gross domestic product annually [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. The national \u0026rdquo;Vision Zero\u0026rdquo; approach, with improvements in road and vehicle safety, presently puts Sweden in the top echelon of reducing traffic-related mortality internationally, with 2.8 deaths per year and 100,000 inhabitants in comparison to a European average and a world average of 9.3 and 17.4 deaths per year and 100,000 inhabitants, respectively [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. The first aim of this study is to examine to what extent the decrease in overall traffic-related mortality is due to fewer accidents, fewer injuries or better treatment outcomes.\u003c/p\u003e \u003cp\u003eAge, sex and comorbidity (measured as Charlson Comorbidity Index (CCI)) are shown to be independent factors for injury outcome [\u003cspan additionalcitationids=\"CR4\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. When performing risk adjustment for the severity of the injury, the ICD-based injury severity score (ICISS), first developed by Osler et al. in 1996 [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e], has been seen as an improvement of the consensus-based Injury Severity Score (ISS), and as of today the most accurate risk adjustment method for trauma research [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. The ICISS takes all the patients\u0026rsquo; injuries into account improving the accuracy in predicting injury severity. It is also more accessible in the national context as ICD-coding is enforced by law in each patient\u0026acute;s medical record in comparison to the Abbreviated Injury Scale (AIS) registration for which there is no consistent organization. An adaptation of the original ICISS to use the ICD-10 system have been shown to further improve its estimation [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e] and its mortality predicament shows robustness in both the European and Swedish trauma setting [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe research group has adopted ICISS in earlier publications [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. However, basing the ICISS value on a combined trauma cohort without regard to the difference in demography and trauma mechanism (Traffic, Fall and Assault) may lessen the performance of the mortality prediction model when applied to a particular subset, such as the traffic group. Therefore, the second aim of this study is to calculate ICISS directly at the traffic subgroup level, which has not been done before, to enhance the performance of the logistic regression model to refine outcome discussions.\u003c/p\u003e \u003cp\u003eThe influence of hospital type on traffic survival data is inconclusive. In contrast to both North American [\u003cspan additionalcitationids=\"CR11\" citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e] and Finnish reports [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e], our earlier study [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e] indicate that hospital type does not seem to influence risk-adjusted traffic-related mortality in the Swedish setting. In that study, the data was collected between 2001 and 2011. It is now important to put this finding in a more current context with new data retrieval, especially as trauma centralisation has in later years been brought to the fore also and not least as a survival advantage at Level 1 hospitals was claimed recently [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe method of choice to examine large registry data has been multivariable logistic regression (LR). However, Artificial intelligence (AI), especially neural network methods, has recently been successfully applied in medical data analysis [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. XGBoost, an ensemble learning algorithm, has become an important tool in scientific research for its exceptional performance in predictive modelling and classification tasks [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. This study\u0026rsquo;s third and final aim is to examine the potential of XGBoost in improving the ICISS mortality prediction model.\u003c/p\u003e"},{"header":"2. Method","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Patient cohort\u003c/h2\u003e \u003cp\u003eThe National Patient Registry (NPR) covers all national hospital admissions from 1987 onward and was used to retrieve trauma-related hospital admissions between 2008 and 2021. By using each patient\u0026rsquo;s unique personal identification number (PIN), the NPR was linked with the Cause of Death Registry, covering the deaths of all identified Swedish citizens. Based upon the International Classification of Diseases version 10 (ICD-10), trauma diagnoses (S00\u0026ndash;T79) and external cause of injury codes (E-codes) (V01\u0026ndash;Y98) were selected to generate the original database. ICD-10 codes T360-T790, except T689 hypothermia, were excluded as this code interval includes poisoning, intoxication and unspecific adverse effects outside our trauma scope. The E-codes V01-V89, excluding V80-82 and V88, were used to define the subgroup of traffic-related trauma. The ICD-based injury severity score (ICISS) was calculated on both the original calculation database (including subgroups of assault and fall) and specifically on the subset of traffic-related trauma (defined as \u0026rdquo;Traffic database for calculation\u0026rdquo; in Figure A1 in the Appendix). The ICISS values in the analyses were based on the latter. Hospital admissions due to other traumas or those found in multiple groups were excluded from the analysis. Patients admitted to hospitals that did not meet the hospital categorisation criteria (see below) were also excluded\u0026mdash;a total of 95,954 hospital admissions made up the final study population of traffic-related trauma (Figure A1 in Appendix).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Hospital categorisation\u003c/h2\u003e \u003cp\u003eThe Swedish PeriOperative Registry (SPOR) classification of hospital level, recognised by the Swedish Department of Health and Welfare (Socialstyrelsen), the Swedish Counties and Regions, the Regional Insurance Company and the Swedish Association of Anaesthesia and Intensive Care, was used as the basis for hospital categorisation in this study due to its wide acceptance in the Swedish healthcare system [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Depending on the intensive care unit ability, the degree of specialised care, access to laboratories and radiology, and whether research and education were part of the hospital\u0026rsquo;s mission, three categories were used: Level 1 hospitals corresponding to university hospitals, Level 2 hospitals corresponding to regional hospitals and Level 3 hospitals corresponding to county hospitals. See the Appendix for in-depth inclusion criteria. For Level 3 hospitals, the retrieved SPOR classification list could not be fully implemented as it did not cover all Level 3 hospitals found in the database [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Thus, Level 3 hospitals missing in SPOR but located within the trauma database were included through face validity by direct communication with the head of the hospital, the head of operations and county administrators or through an in-depth web search. If emergency room or traffic-related hospital admissions throughout the study period could be subjectively confirmed, the hospital was excluded from the analysis (See Appendix for details).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Injury severity\u003c/h2\u003e \u003cp\u003eThe traffic database (Figure A1 in Appendix) calculated the diagnosis-specific survival probability (DSP) for each ICD-10 diagnosis, giving the likelihood of survival for each injury. Duplicate ICD-10 codes were removed before the DSP calculation. Using an inclusive approach, an ICISS value was calculated for each hospital admission as the product of the DSPs for the ICD-10 diagnoses. As ICISS was highly skewed, the transformation log\u003csub\u003e10\u003c/sub\u003e(1\u0026thinsp;\u0026minus;\u0026thinsp;\u003cem\u003eICISS\u003c/em\u003e+\u003cem\u003eϵ\u003c/em\u003e) was used with \u003cem\u003eϵ\u003c/em\u003e = 0.001 for all regression analyses using ICISS.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Comorbidities\u003c/h2\u003e \u003cp\u003eThe CCI adapted to the ICD-10 system, according to Glasheen et al., was used [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. Translation of ICD-9 codes was done as described by Glasheen et al. [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e], and a direct comparison was made with the Swedish National Board of Health and Welfare [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e] to achieve coherent coding with the ICD-9 codes used in Sweden at the time.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.5 Statistics\u003c/h2\u003e \u003cp\u003eWe used standard multivariable logistic regression and XAI, combining XGBooost and SHAP as implemented in the R packages XGBoost and SHAP [\u003cspan additionalcitationids=\"CR21\" citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. The regression models were evaluated with the area under the curve (AUC), and differences in these were assessed using the DeLong test [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. The data was divided into an 80% training set and a 20% test set. A grid search on the training set was carried out to find good hyperparameters of XGBoost. scale_pos_weight was set to 99. The ICISS scores were calculated using the training set. XGBoost was then applied to the training set and evaluated on the test set. The XGBoost was interpreted using SHapley Additive exPlanations, SHAP.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e2.6 Ethics\u003c/h2\u003e \u003cp\u003eThe Swedish Ethical Review Authority approved the study protocol (registration nos. 2019\u0026ndash;04852 and 2021\u0026ndash;04333), and the study was conducted according to the 1964 Declaration of Helsinki. In the ethics application and the formal ethics committee approval informed consent from the patients was waived.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Patient characteristics\u003c/h2\u003e \u003cp\u003eThe final study population comprised 95,954 hospital admissions between 2008 and 2021, with a 30-day mortality of 1.16% and a median ICISS of 0.91 (IQR 0.11). The ICISS distribution over hospital levels is shown in Table\u0026nbsp;1. The ages of the admitted patients ranged between 18 and 103, with a median and mean of 48 years (IQR 33 years, SD 19 years). Men accounted for 67% of the study population and were overrepresented in all age groups. See Table\u0026nbsp;1 and Fig.\u0026nbsp;2. The annual incidence of hospitalised RTAs is shown in Fig.\u0026nbsp;1. Amongst fatal accidents, over 46% were associated with traumatic brain injury (TBI defined as ICD-10 codes S061-S069), and in the elderly cohort (\u0026ge;\u0026thinsp;65 years), over 40% of the fatalities were associated with TBI. The Appendix shows data over time in Table \u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003eA1\u003c/span\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Regression models\u003c/h2\u003e \u003cp\u003eFigure A2 in the Appendix shows the ROCs and their pertaining AUCs for the LR and XAI 30-day mortality prediction models (p\u0026thinsp;\u003cem\u003e\u0026lt;\u003c/em\u003e\u0026thinsp;0.05, deLong\u0026rsquo;s test). Calibration, as measured with Brier score, was 1.00 for LR and 0.09 for XAI (\u003cem\u003ep-value\u0026thinsp;\u0026lt;\u003c/em\u003e\u0026thinsp;0.05).\u003c/p\u003e \u003cp\u003eOdds ratios from the multivariable (ICISS, age, sex, CCI, event year, hospital level) logistic regression models of 30-day mortality are shown in Table\u0026nbsp;2. The most important factors for 30-day mortality were, in decreasing importance, ICISS, age, CCI, event year, hospital level, and sex, as shown in the summary plot in Fig.\u0026nbsp;3.\u003c/p\u003e \u003cp\u003eDecreasing ICISS independently increases mortality as seen in Table\u0026nbsp;2 and Figs.\u0026nbsp;4 and 5. ICISS decreased significantly over time, but the magnitude of this was minor, as seen in Table \u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003eA1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eIncreasing age independently increases mortality as seen in Table\u0026nbsp;2 and Figs.\u0026nbsp;4 and 5. As seen in Table \u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003eA1\u003c/span\u003e, age increased markedly over the study period.\u003c/p\u003e \u003cp\u003eIncreasing CCI independently increases mortality as seen in Table\u0026nbsp;2 and Figs.\u0026nbsp;4 and 5. As seen in Table \u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003eA1\u003c/span\u003e, CCI varied over the study period.\u003c/p\u003e \u003cp\u003eThirty-day mortality was highest for level 1 (primary admission) hospitals, followed by level 2 and level 3. Hospital level was not a clear factor for mortality for the most severely injured and non-transfer patients. Subgroup analysis for transfer patients only (data not shown) showed no significant difference between hospital levels in the LR model. Similarly, the XAI model shows higher mortality for level 1 hospitals with similar mortality for levels 2 and 3 (for all patients and the most severely injured alike, see Figs.\u0026nbsp;4 and 5).\u003c/p\u003e \u003cp\u003eThe use of Level 2 hospitals increased over time, mainly due to a lower fraction of patients treated at Level 3 hospitals. See Table \u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003eA1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eFemale sex was also independently associated with lower 30-day mortality for all hospital admissions and non-transfer cases. See Table\u0026nbsp;2 and Figs.\u0026nbsp;4 and 5).\u003c/p\u003e \u003cp\u003eThe adjusted 30-day mortality decreased over the study period for all hospital admissions and non-transfer cases (see Table\u0026nbsp;2 and Figs.\u0026nbsp;4 and 5).\u003c/p\u003e \u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eBy applying XAI to RTA data and ICD-10-based injury severity scoring (ICISS), we showed that the XAI model outperformed standard multivariable logistic regression in estimating 30-day mortality after RTA hospitalisations. The improved XAI model allowed for better characterisation of the risk factors for RTA-associated 30-day mortality.\u003c/p\u003e \u003cp\u003eThe XAI model\u0026rsquo;s most important factors for predicting 30-day mortality were, in decreasing order of influence, ICISS, age, CCI, event year, sex, and hospital level. As anticipated, the primary determinant of 30-day mortality is injury severity. The ICISS values in this study are higher than in Osler et al. [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e] (median 0.93) and Larsen et al. [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e] (median 0.95). This difference may arise from using the RTA cohort rather than all trauma cases to estimate ICISS. Our higher ICISS values are also highlighted by our higher median ICISS of 0.91 for fatal accidents, which is 0.45 in Osler\u0026rsquo;s study [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e] and 0.72 in Ydenius et al. [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. The ageing population [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e] in the RTA cohort is thought to be a contributing factor [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. In Osler\u0026rsquo;s study [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e], and in Ydenius et al. [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e], 92% and 77% were 55 years or younger, respectively. In our study, the median ICISS for the 18-25-year-olds was 0.877 and 0.935 for the \u0026ge;\u0026thinsp;75-year-olds (data not shown). Table\u0026nbsp;1 shows that severe cases with ICISS below 0.85 are rare, only accounting for around 3% of the hospital admissions. This raises the question of whether an ICISS cutoff of 0.85 for severe RTA is too strict, as many fatalities occur at higher ICISS levels in this dataset. An ICISS threshold of \u003cem\u003e\u0026lt;\u003c/em\u003e\u0026thinsp;0.941 has been previously suggested for severe injury [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe analysis of demographics over time (see Table \u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003eA1\u003c/span\u003e in the Appendix) shows that the median age of RTA victims increased over time, that median ICISS decreased over time (more severely injured), CCI varied significantly and that fewer patients were treated in Level 3 hospitals in favour of Level 2. These factors should most likely result in a higher mortality.\u003c/p\u003e \u003cp\u003eAs seen in Table\u0026nbsp;2 and the bottom right panels of Figs.\u0026nbsp;4 and 5, the corrected mortality risk decreased with the year of the trauma. Surprisingly, the values were low and high for 2014 and 2018, respectively, for the most severely injured (Fig.\u0026nbsp;5). These values agree with the mortality observed in the RTA statistics of Transport Analysis (TRAFA) [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eMany elderly and young individuals suffer fatal RTAs [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. Young suffering RTAs may be due to risk-prone behaviour [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. This fact is also reflected by the low median age of traffic-related admissions, 47 years old (IQR 32 years) in our data. Figure\u0026nbsp;2 shows that young men are overrepresented. However, the median age for fatal accidents in our data is high, 72 years old (IQR 31 years). This aligns with our findings that mortality increases with age, as seen in Table\u0026nbsp;2 and Fig.\u0026nbsp;4. It has been suggested that younger individuals who are exposed to fatal RTAs die at the trauma site or during hospital transfer.\u003c/p\u003e \u003cp\u003eEarlier findings show that females have a survival advantage over men [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e], and comorbidity influences 30-day mortality [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. However, the small mortality effect of sex and comorbidity is not significant in the LR model of the severely injured, possibly due to few observations. In contrast, the XAI model indicates a small sex difference as well as an increasing mortality for CCI 1 as compared with CCI 0 for the severely injured.\u003c/p\u003e \u003cp\u003eNotably, very few persons with CCI above 6 (CCI ranges from 0 to 20) are admitted to hospitals with RTA-associated trauma (see Fig.\u0026nbsp;4), possibly reflecting that persons with many comorbidities do not get exposed to traffic [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe most severely injured patients are admitted to Level 1 hospitals, where they die to a greater extent than at Level 2 and 3 hospitals (see Tables\u0026nbsp;1 and 2 and Figs.\u0026nbsp;4 and 5), consistent with Ydenius et al. [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. However, the hospital level does not clearly impact overall mortality in the LR model of the most severely injured or the non-transfer patients. The XAI model shows a much clearer effect on the hospital level than the LR method. Earlier criticisms about attributing outcomes to the admitting hospital have been addressed by categorising patients based on transfer status.\u003c/p\u003e \u003cp\u003eThe notion that the initial trauma care, the so-called \u0026rdquo;Golden Hour\u0026rdquo;, is the primary determinant of trauma outcomes is being questioned [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. North American studies advocate trauma centre referrals rather than focusing on early care [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]. Our results of a lack of survival advantage of Level 1 centra raise whether Sweden should prioritise trauma centre referral or hospital accessibility. Candefjord et al. [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e] argue that reaching trauma centres within one hour in Sweden is possible if helicopter transports are considered. However, helicopter ambulance resources are currently very limited in Sweden. An alternative is \u0026rdquo;bringing the emergency room to the site\u0026rdquo; with well-equipped ambulances [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eTransfers complicate the analysis of hospital-level impact. Therefore, patients were categorised as having no transfers or having transfers (based on the first transfer only). The majority of hospital admissions (91%) are never transferred. Early deaths preclude transfers, introducing a selection bias. Future work should include detailed analysis with formal SHAP score statistics to understand the importance of the hospital level.\u003c/p\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Improving Methods\u003c/h2\u003e \u003cp\u003ePrevious studies suggest that XGBoost improves trauma mortality prediction models based on logistic regression on the Injury Severity Score (ISS), the Trauma Mortality Prediction Model TMPM-ICD10 [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e], and the Trauma and Injury Severity Score (TRISS) [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]. The novel approach of calculating ICISS solely on RTAs and using a machine learning model improves the risk adjustment and mortality prediction (AUC increases from 0.90 for LR to 0.92 for XAI). To our knowledge, the use of XAI for ICISS data is new and gives a better understanding of the complex relationships found in our data.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Limitations\u003c/h2\u003e \u003cp\u003eAlthough the SweTrau registry was considered a data source, SweTrau was only initiated after the start of this study. Consequently, the NPR was chosen. Legal requirements for ICD coding minimise the risk of missing data. While incomplete or inaccurate coding remains a concern, economic incentives for correct coding and NPR validation ensure good precision for ICD codes up to the fourth position [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. By extension, the use of the International Classification of Diseases Injury Severity Score (ICISS) is a stalwart method of risk adjustment in extensive data sets [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e], preferred over the traditional ISS method [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]. Ydenius et al. have successfully used the ICISS method and are confirming it in a more current setting. The dataset\u0026rsquo;s representation of the \u0026rdquo;real-world\u0026rdquo; population and its size and inclusion period is a notable strength.\u003c/p\u003e \u003cp\u003ePatients dying at the trauma site or during transport to the hospital were not included in this study. TRAFA [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e] suggests a higher survival in the metropolitan areas Stockholm, Gothenburg and Malm\u0026ouml;, and conversely, high mortality in the less densely populated northern parts of the country, implying that transport time is a factor for traffic-related mortality. However, it is essential to remember where in the care chain this study is set and as the factual hospital admissions are set for inclusion, transport time as a parameter opts for a different study.\u003c/p\u003e \u003cp\u003eUsing an anatomical risk-adjustment tool, dynamics in physiological variables affecting outcomes may be overlooked [\u003cspan additionalcitationids=\"CR42\" citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]. Physiological parameters upon emergency room arrival are, in turn, influenced by pre-hospital stabilisation. Nevertheless, DSP calculation retrospectively includes the impact of physiological parameters, given severe injuries often coincide with deranged vital signs.\u003c/p\u003e \u003cp\u003eThe hospital-level analysis is complicated by a selection bias, as patients who pass away quickly (the most severely injured) may not be fit for a secondary transport up the hospital-level chain. The group of transported patients, around 9% of the entire data set, is also very diverse as the time of the transport to another hospital is not defined. Thus, transportation may take place in almost immediate connection to the trauma ictus or weeks after the initial admission. One must tread carefully when making conclusions from this group for obvious reasons. The possibility of separating this subgroup of patients was used to benefit the study. The hospital level can be studied independently as a risk factor for 30-day mortality by studying the non-transfer cases.\u003c/p\u003e \u003cp\u003eXAI further helps us to visualise the data, whereas traditional statistics give the confidence to make assertions from the SHAP representation. The fact that the XAI approach does not currently allow for formal statistical testing is indeed a limit but using it in symbiosis with traditional logistic regression still makes it useful. Plans are underway to develop computer-intensive methods for formal statistical testing using XGBoost and SHAP, which are outside the scope of the current paper.\u003c/p\u003e \u003c/div\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eAn explainable AI (XAI) model for 30-day mortality of hospitalised RTA victims based on ICD codes (ICISS), age, CCI, event year, hospital level, and sex (order based on importance) showed better discrimination and calibration than a corresponding LR-based model. The adjusted mortality decreased from 2008 to 2021 despite increasing injury severity and age (median increased from 44 years to 52 years) of RTA victims, suggesting advancements in treatment efficacy over time. Factors such as sex and comorbidities impact 30-day mortality rates but do not achieve statistical significance among severely injured patients (ICISS\u0026le;0.85). The trend of centralising treatment at Level 1 hospitals remains prevalent, correlating with a decreasing order of injury severity across hospital levels. Nevertheless, upon adjusting for risk factors, hospital level did not significantly influence risk outcomes, especially considering the most severely injured. A key strength of this study is the stratification of hospitalisations based on transfer status and the stalwart method of ICISS for injury severity risk adjustment. A more comprehensive evaluation of hospital-level impact, particularly the direct utilisation of Level 1 facilities, should incorporate transport time as a variable. The XAI method should also be extended to include a formal statistical testing framework. In conclusion, this research shows that prevailing assumptions regarding the advantages of trauma centralisation are not given to be beneficial for Swedish RTA victims.\u003c/p\u003e\n\u003cp\u003eThe datasets used in the current study are available from the corresponding author on reasonable request.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements/Funding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eRegional research support, Region Skåne #2022-1284; Governmental funding of clinical research within the Swedish National Health Service (ALF) #2022:YF0009 and #2022-0075; Crafoord Foundation grant number #2021-0833; Lions Skåne research grants; Skåne University Hospital grants; Swedish Heart and Lung Foundation (HLF) #2022-0352 and #2022-0458.The Region Ostergötland and Linköping University supported the study.¨\u003c/p\u003e\n\u003cp\u003eFunding sources were not involved in the manuscript’s writing or the decision to submit it for publication. The article was written without any commission. No pharmaceutical companies or other agencies have been involved in writing the submitted manuscript.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eNations, U.: Improving global road safety. General Assembly (2020) https://doi.org/https://documents.un.org/doc/undoc/gen/n20/226/30/pdf/ n2022630.pdf?OpenElement . Accessed: 2023-08-30\u003c/li\u003e\n \u003cli\u003eJoseph, A.: Road trauma\u0026ndash;an ongoing challenge in injury prevention and postcrash care. Injury \u003cstrong\u003e49\u003c/strong\u003e(9), 1637\u0026ndash;1638 (2018)\u003c/li\u003e\n \u003cli\u003eLarsen, R., B\u0026uml;ackstr\u0026uml;om, D., Fredrikson, M., Steinvall, I., Gedeborg, R., Sjoberg, F.: Decreased risk adjusted 30-day mortality for hospital admitted injuries: a multi-centre longitudinal study. Scandinavian journal of trauma, resuscitation and emergency medicine \u003cstrong\u003e26\u003c/strong\u003e(1), 1\u0026ndash;8 (2018)\u003c/li\u003e\n \u003cli\u003eYdenius, V., Larsen, R., Steinvall, I., B\u0026uml;ackstr\u0026uml;om, D., Chew, M., Sj\u0026uml;oberg, F.: Impact of hospital type on risk-adjusted, traffic-related 30-day mortality: a population-based registry study. Burns \u0026amp; Trauma \u003cstrong\u003e9\u003c/strong\u003e, 051 (2021)\u003c/li\u003e\n \u003cli\u003eSundararajan, V., Quan, H., Halfon, P., Fushimi, K., Luthi, J.C., Burnand, B., Ghali, W.A.: Cross-national comparative performance of three versions of the ICD-10 Charlson index. Med Care \u003cstrong\u003e45\u003c/strong\u003e(12), 1210\u0026ndash;1215 (2007)\u003c/li\u003e\n \u003cli\u003eOsler, T., Rutledge, R., Deis, J., Bedrick, E.: Iciss: an international classification of disease-9 based injury severity score. Journal of Trauma and Acute Care Surgery \u003cstrong\u003e41\u003c/strong\u003e(3), 380\u0026ndash;388 (1996)\u003c/li\u003e\n \u003cli\u003eGedeborg, R., Warner, M., Chen, L.-H., Gulliver, P., Cryer, C., Robitaille, Y., Bauer, R., Ubeda, C., Lauritsen, J., Harrison, J., \u003cem\u003eet al.\u003c/em\u003e: Internationally comparable diagnosis-specific survival probabilities for calculation of the ICD-10\u0026ndash;based injury severity score. Journal of Trauma and Acute Care Surgery \u003cstrong\u003e76\u003c/strong\u003e(2), 358\u0026ndash;365 (2014)\u003c/li\u003e\n \u003cli\u003eStephenson, S., Henley, G., Harrison, J.E., Langley, J.D.: Diagnosis based injury severity scaling: investigation of a method using australian and new zealand hospitalisations. Injury prevention \u003cstrong\u003e10\u003c/strong\u003e(6), 379\u0026ndash;383 (2004)\u003c/li\u003e\n \u003cli\u003eLarsen, R.: Risk-Adjustment for Swedish In-Hospital Trauma Mortality Using International Classification of Disease Injury Severity Score (ICISS): Issues with Description and Methods vol. 1660. Link\u0026uml;oping University Electronic Press, ??? (2019)\u003c/li\u003e\n \u003cli\u003eDemetriades, D., Martin, M., Salim, A., Rhee, P., Brown, C., Chan, L.: The effect of trauma center designation and trauma volume on outcome in specific severe injuries. Annals of surgery \u003cstrong\u003e242\u003c/strong\u003e(4), 512 (2005)\u003c/li\u003e\n \u003cli\u003eNathens, A.B., Jurkovich, G.J., Maier, R.V., Grossman, D.C., MacKenzie, E.J., Moore, M., Rivara, F.P.: Relationship between trauma center volume and outcomes. Jama \u003cstrong\u003e285\u003c/strong\u003e(9), 1164\u0026ndash;1171 (2001)\u003c/li\u003e\n \u003cli\u003eMacKenzie, E.J., Rivara, F.P., Jurkovich, G.J., Nathens, A.B., Frey, K.P., Egleston, B.L., Salkever, D.S., Scharfstein, D.O.: A national evaluation of the effect of trauma-center care on mortality. New England Journal of Medicine \u003cstrong\u003e354\u003c/strong\u003e(4), 366\u0026ndash;378 (2006)\u003c/li\u003e\n \u003cli\u003eAla-Kokko, T., Ohtonen, P., Koskenkari, J., Laurila, J.: Improved outcome after trauma care in university-level intensive care units. Acta anaesthesiologica scandinavica \u003cstrong\u003e53\u003c/strong\u003e(10), 1251\u0026ndash;1256 (2009)\u003c/li\u003e\n \u003cli\u003eCandefjord, S., Asker, L., Caragounis, E.-C.: Mortality of trauma patients treated at trauma centers compared to non-trauma centers in sweden: a retrospective study. European journal of trauma and emergency surgery, 1\u0026ndash;12 (2020)\u003c/li\u003e\n \u003cli\u003eHolmgren, G., Andersson, P., Jakobsson, A., Frigyesi, A.: Artificial neural networks improve and simplify intensive care mortality prognostication: a national cohort study of 217,289 first-time intensive care unit admissions. Journal of intensive care \u003cstrong\u003e7\u003c/strong\u003e(1), 1\u0026ndash;8 (2019)\u003c/li\u003e\n \u003cli\u003eChen, T., Guestrin, C.: Xgboost: A scalable tree boosting system. In: Proceedings of the 22nd Acm Sigkdd International Conference on Knowledge Discovery and Data Mining, pp. 785\u0026ndash;794 (2016)\u003c/li\u003e\n \u003cli\u003eRegister, S.P.: Sjukhus anslutna till spor. Svenskt PeriOperativt Register (2023) https://doi.org/https://spor.se/om-spor-landingpage/anslutna-kliniker/ . Accessed: 2023-01-17\u003c/li\u003e\n \u003cli\u003eGlasheen, W.P., Cordier, T., Gumpina, R., Haugh, G., Davis, J., Renda, A.: Charlson comorbidity index: ICD-9 update and ICD-10 translation. American Health \u0026amp; Drug Benefits \u003cstrong\u003e12\u003c/strong\u003e(4), 188 (2019)\u003c/li\u003e\n \u003cli\u003eSocialstyrelsen: ICD9 klassifikation av sjukdomar 1987 KS87. Socialstyrelsen (2022) https://doi.org/https://www.socialstyrelsen.se/globalassets/ sharepoint-dokument/dokument-webb/klassifikationer-och-koder/ icd-9-klassifikation-av-sjukdomar-1987-ks87.xls . Accessed: 2022-09-02\u003c/li\u003e\n \u003cli\u003eR Core Team: R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria (2021). R Foundation for Statistical Computing. https://www.R-project.org/\u003c/li\u003e\n \u003cli\u003eChen, T., Guestrin, C.: XGBoost: A scalable tree boosting system. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. KDD \u0026rsquo;16, pp. 785\u0026ndash;794. ACM, New York, NY, USA (2016). https://doi.org/10.1145/2939672.2939785 . http://doi.acm.org/10.1145/2939672. 2939785\u003c/li\u003e\n \u003cli\u003eLundberg, S.M., Lee, S.-I.: A unified approach to interpreting model predictions. In: Guyon, I., Luxburg, U.V., Bengio, S., Wallach, H., Fergus, R., Vishwanathan, S., Garnett, R. (eds.) Advances in Neural Information Processing Systems 30, pp. 4765\u0026ndash;4774. Curran Associates, Inc., ??? (2017). http://papers.nips.cc/paper/ 7062-a-unified-approach-to-interpreting-model-predictions.pdf\u003c/li\u003e\n \u003cli\u003eDeLong, E.R., DeLong, D.M., Clarke-Pearson, D.L.: Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics \u003cstrong\u003e44\u003c/strong\u003e(3), 837\u0026ndash;845 (1988)\u003c/li\u003e\n \u003cli\u003eSCB: Folkm\u0026uml;angden den 1 november efter region, ˚alder och k\u0026uml;on. ˚Ar 2002 - 2023Statistikdatabasen \u0026mdash; statistikdatabasen.scb.se. https://www.statistikdatabasen. scb.se/pxweb/sv/ssd/STARTBEBE0101BE0101A/FolkmangdNov/table/ tableViewLayout1/. [Accessed 04-10-2024]\u003c/li\u003e\n \u003cli\u003eLau, L., Ajzenberg, H., Haas, B., Wong, C.L.: Trauma in the aging population: geriatric trauma pearls. Emergency Medicine Clinics \u003cstrong\u003e41\u003c/strong\u003e(1), 183\u0026ndash;203 (2023)\u003c/li\u003e\n \u003cli\u003ePitta, L.S.R., Quintas, J.L., Trindade, I.O.A., Belchior, P., Gameiro, K.d.S.D., Gomes, C.M., N\u0026acute;obrega, O.T., Camargos, E.F.: Older drivers are at increased risk of fatal crash involvement: results of a systematic review and meta-analysis. Archives of gerontology and geriatrics \u003cstrong\u003e95\u003c/strong\u003e, 104414 (2021)\u003c/li\u003e\n \u003cli\u003eBerecki-Gisolf, J., Fernando, D.T., D\u0026rsquo;Elia, A.: International classification of disease based injury severity score (iciss): A data linkage study of hospital and death data in victoria, australia. Injury \u003cstrong\u003e53\u003c/strong\u003e(3), 904\u0026ndash;911 (2022)\u003c/li\u003e\n \u003cli\u003eAnalysis, T.: Road traffic injuries 2018. Transport Analysis (2018) https://doi.org/https://www.trafa.se/globalassets/statistik/vagtrafik/ vagtrafikskador/2018/vagtrafikskador-2018---blad.pdf . Accessed: 2023-10-10\u003c/li\u003e\n \u003cli\u003eAnalysis, T.: Road traffic injuries 2021. Transport Analysis (2021) https://doi.org/https://www.trafa.se/globalassets/statistik/vagtrafik/ vagtrafikskador/2021/vagtrafikskador-2021---korr.-2022-05-15.pdf . Accessed: 2023-10-10\u003c/li\u003e\n \u003cli\u003eTurner, C., McClure, R.: Age and gender differences in risk-taking behaviour as an explanation for high incidence of motor vehicle crashes as a driver in young males. Injury control and safety promotion \u003cstrong\u003e10\u003c/strong\u003e(3), 123\u0026ndash;130 (2003)\u003c/li\u003e\n \u003cli\u003eFalk, B.: Do drivers become less risk-prone after answering a questionnaire on risky driving behaviour? Accident Analysis \u0026amp; Prevention \u003cstrong\u003e42\u003c/strong\u003e(1), 235\u0026ndash;244 (2010)\u003c/li\u003e\n \u003cli\u003eLarsen, R., B\u0026uml;ackstr\u0026uml;om, D., Fredrikson, M., Steinvall, I., Gedeborg, R., Sjoberg, F.: Female risk-adjusted survival advantage after injuries caused by falls, traffic or assault: a nationwide 11-year study. Scandinavian journal of trauma, resuscitation and emergency medicine \u003cstrong\u003e27\u003c/strong\u003e, 1\u0026ndash;7 (2019)\u003c/li\u003e\n \u003cli\u003eMarshall, S.C., Man-Son-Hing, M.: Multiple chronic medical conditions and associated driving risk: a systematic review. Traffic injury prevention \u003cstrong\u003e12\u003c/strong\u003e(2), 142\u0026ndash;148 (2011)\u003c/li\u003e\n \u003cli\u003eRogers, F.B., Rittenhouse, K.J., Gross, B.W.: The golden hour in trauma: dogma or medical folklore? Injury \u003cstrong\u003e46\u003c/strong\u003e(4), 525\u0026ndash;527 (2015)\u003c/li\u003e\n \u003cli\u003eNewgard, C.D., Schmicker, R.H., Hedges, J.R., Trickett, J.P., Davis, D.P., Bulger, E.M., Aufderheide, T.P., Minei, J.P., Hata, J.S., Gubler, K.D., \u003cem\u003eet al.\u003c/em\u003e: Emergency medical services intervals and survival in trauma: assessment of the \u0026ldquo;golden hour\u0026rdquo; in a north american prospective cohort. Annals of emergency medicine \u003cstrong\u003e55\u003c/strong\u003e(3), 235\u0026ndash;246 (2010)\u003c/li\u003e\n \u003cli\u003eNewgard, C.D., Meier, E.N., Bulger, E.M., Buick, J., Sheehan, K., Lin, S., Minei, J.P., Barnes-Mackey, R.A., Brasel, K., Investigators, R., \u003cem\u003eet al.\u003c/em\u003e: Revisiting the \u0026ldquo;golden hour\u0026rdquo;: an evaluation of out-of-hospital time in shock and traumatic brain injury. Annals of emergency medicine \u003cstrong\u003e66\u003c/strong\u003e(1), 30\u0026ndash;41 (2015)\u003c/li\u003e\n \u003cli\u003eStrandqvist, E., Olheden, S., B\u0026uml;ackman, A., J\u0026uml;ornvall, H., B\u0026uml;ackstr\u0026uml;om, D.: Physician-staffed prehospital units: a retrospective follow-up from an urban area in scandinavia. International Journal of Emergency Medicine \u003cstrong\u003e16\u003c/strong\u003e(1), 43 (2023)\u003c/li\u003e\n \u003cli\u003eTran, Z., Zhang, W., Verma, A., Cook, A., Kim, D., Burruss, S., Ramezani, R., Benharash, P.: The derivation of an international classification of diseases, tenth revision\u0026ndash;based trauma-related mortality model using machine learning. Journal of Trauma and Acute Care Surgery \u003cstrong\u003e92\u003c/strong\u003e(3), 561\u0026ndash;566 (2022)\u003c/li\u003e\n \u003cli\u003eTran, Z., Verma, A., Wurdeman, T., Burruss, S., Mukherjee, K., Benharash, P.: ICD-10 based machine learning models outperform the trauma and injury severity score (triss) in survival prediction. Plos one \u003cstrong\u003e17\u003c/strong\u003e(10), 0276624 (2022)\u003c/li\u003e\n \u003cli\u003eGagne, M., Moore, L., Beaudoin, C., Kuimi, B.L.B., Sirois, M.-J.: Performance of international classification of diseases\u0026ndash;based injury severity measures used to predict in-hospital mortality: a systematic review and meta-analysis. Journal of Trauma and Acute Care Surgery \u003cstrong\u003e80\u003c/strong\u003e(3), 419\u0026ndash;426 (2016)\u003c/li\u003e\n \u003cli\u003eKang, M.W., Ko, S.Y., Song, S.W., Kim, W.J., Kang, Y.J., Kang, K.W., Park, H.S., Park, C.B., Kang, J.H., Bu, J.H., \u003cem\u003eet al.\u003c/em\u003e: Prognostic accuracy of the quick sequential organ failure assessment for outcomes among patients with trauma in the emergency department: a comparison with the modified early warning score, revised trauma score, and injury severity score. Journal of Trauma and Injury \u003cstrong\u003e34\u003c/strong\u003e(1), 3\u0026ndash;12 (2021)\u003c/li\u003e\n \u003cli\u003eJawa, R.S., Vosswinkel, J.A., McCormack, J.E., Huang, E.C., Thode Jr, H.C., Shapiro, M.J., Singer, A.J.: Risk assessment of the blunt trauma victim: the role of the quick sequential organ failure assessment score (qsofa). The American Journal of Surgery \u003cstrong\u003e214\u003c/strong\u003e(3), 397\u0026ndash;401 (2017)\u003c/li\u003e\n \u003cli\u003eSadhwani, N., Ambore, V., Bakhshi, G.: Predictive value of quick sequential organ failure assessment (qsofa) score in risk assessment and outcome prediction in blunt trauma patients: A prospective observational study. Annals of Medicine and Surgery \u003cstrong\u003e74\u003c/strong\u003e, 103265 (2022)\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTables 1 and 2 are available in the Supplementary Files section\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-6628847/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6628847/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground: \u003c/strong\u003eGlobally, road traffic accidents (RTAs) remain a leading cause of mortality, particularly among individuals aged 15–30 years. Sweden has been at the forefront of traffic safety, but in-hospital care is critical in determining outcomes following RTAs. Guided by North American data demonstrating improved survival rates at trauma centres, the Swedish healthcare system is shifting towards trauma centralisation. However, comprehensive national data specific to Sweden remain underexplored. The unique demographic characteristics of Sweden, including vast, sparsely populated regions, distinguish it from other Western nations, complicating direct comparisons.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods: \u003c/strong\u003eThe epidemiology and risk factors for 30-day mortality from RTAs in Sweden were investigated for 95,954 hospital admissions from 2008 to 2021. The ICD-based injury severity score (ICISS), age, sex, the Charlson comorbidity index (CCI), the year of the event, and hospital level were examined using an explainable AI (XAI) and Logistic regression.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults: \u003c/strong\u003eThe most important factors for 30-day mortality were, in decreasing importance, ICISS, age, CCI, event year, hospital level, and sex. There was a clear trend towards centralising RTA care, with Level 1 hospitals catering to the most critically injured patients. In parallel, however, the hospital level did not affect risk-adjusted traffic-related mortality. XAI enhances mortality prediction over logistic regression, confirming these findings.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDiscussion: \u003c/strong\u003eThis study presents the most extensive analysis of in-hospital outcomes for road traffic accidents (RTAs) in Europe to date. Factors such as ICISS, age, sex and the CCI had anticipated effects on outcome. Overall treatment outcome, measured as mortality improved over time.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion: \u003c/strong\u003eThe presumption that external validity for trauma centralisation is applicable in the Scandinavian trauma context is not convincing. These findings make it essential to investigate further the trauma organisation given Scandinavian prerequisites so that time to hospital is not sacrificed for the type of hospital.\u003c/p\u003e","manuscriptTitle":"Hospital level does not influence 30-day in-hospital mortality in road traffic accident hospitalisations - a nationwide registry study utilising Explainable AI (XAI) and the ICD-10 based injury severity score (ICISS)","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-06-03 07:39:21","doi":"10.21203/rs.3.rs-6628847/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-06-19T22:46:39+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-06-18T04:41:54+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"147651474088099841012034035460592826303","date":"2025-06-05T10:32:27+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-05-30T07:04:50+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"107209036742610813523553059709616983298","date":"2025-05-29T16:39:53+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-05-29T16:26:27+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-05-29T16:06:21+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-05-27T04:59:55+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-05-24T04:31:47+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2025-05-09T12:46:40+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"fab09e89-0455-4ac3-9f3d-7bf03b7c79b4","owner":[],"postedDate":"June 3rd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":49339367,"name":"Health sciences/Health care"},{"id":49339368,"name":"Health sciences/Risk factors"}],"tags":[],"updatedAt":"2025-11-24T16:01:40+00:00","versionOfRecord":{"articleIdentity":"rs-6628847","link":"https://doi.org/10.1038/s41598-025-26519-7","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2025-11-21 15:57:08","publishedOnDateReadable":"November 21st, 2025"},"versionCreatedAt":"2025-06-03 07:39:21","video":"","vorDoi":"10.1038/s41598-025-26519-7","vorDoiUrl":"https://doi.org/10.1038/s41598-025-26519-7","workflowStages":[]},"version":"v1","identity":"rs-6628847","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6628847","identity":"rs-6628847","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}