Abstract
Background
Machine learning (ML), a subset of artificial intelligence (AI), uses algorithms to analyze data and predict outcomes without extensive human intervention. In healthcare, ML is gaining attention for enhancing patient outcomes. This study focuses on predicting additional hospital days (AHD) for patients with cervical spondylosis (CS), a condition affecting the cervical spine. The research aims to develop an ML-based nomogram model analyzing clinical and demographic factors to estimate hospital length of stay (LOS). Accurate AHD predictions enable efficient resource allocation, improved patient care, and potential cost reduction in healthcare.
Methods
The study selected CS patients undergoing cervical spine surgery and investigated their medical data. A total of 945 patients were recruited, with 570 males and 375 females. The mean number of LOS calculated for the total sample was 8.64 ± 3.7 days. A LOS equal to or <8.64 days was categorized as the AHD-negative group (n = 539), and a LOS > 8.64 days comprised the AHD-positive group (n = 406). The collected data was randomly divided into training and validation cohorts using a 7:3 ratio. The parameters included their general conditions, chronic diseases, preoperative clinical scores, and preoperative radiographic data including ossification of the anterior longitudinal ligament (OALL), ossification of the posterior longitudinal ligament (OPLL), cervical instability and magnetic resonance imaging T2-weighted imaging high signal (MRI T2WIHS), operative indicators and complications. ML-based models like Lasso regression, random forest (RF), and support vector machine (SVM) recursive feature elimination (SVM-RFE) were developed for predicting AHD-related risk factors. The intersections of the variables screened by the aforementioned algorithms were utilized to construct a nomogram model for predicting AHD in patients. The area under the curve (AUC) of the receiver operating characteristic (ROC) curve and C-index were used to evaluate the performance of the nomogram. Calibration curve and decision curve analysis (DCA) were performed to test the calibration performance and clinical utility.
Results
For these participants, 25 statistically significant parameters were identified as risk factors for AHD. Among these, nine factors were obtained as the intersection factors of these three ML algorithms and were used to develop a nomogram model. These factors were gender, age, body mass index (BMI), American Spinal Injury Association (ASIA) scores, magnetic resonance imaging T2-weighted imaging high signal (MRI T2WIHS), operated segment, intraoperative bleeding volume, the volume of drainage, and diabetes. After model validation, the AUC was 0.753 in the training cohort and 0.777 in the validation cohort. The calibration curve exhibited a satisfactory agreement between the nomogram predictions and actual probabilities. The C-index was 0.788 (95% confidence interval: 0.73214–0.84386). On the decision curve analysis (DCA), the threshold probability of the nomogram ranged from 1 to 99% (training cohort) and 1 to 75% (validation cohort).
Conclusion
We successfully developed an ML model for predicting AHD in patients undergoing cervical spine surgery, showcasing its potential to support clinicians in AHD identification and enhance perioperative treatment strategies.
Introduction
Within an aging population, cervical spondylosis (CS) is characterized by age-related changes within the cervical column. The annual prevalence of CS increases with a rise in long-term desk work and needs more attention [Citation1,Citation2]. CS causes altered motor and sensory functions due to one or more nerve roots and/or spinal cord compression. Its detrimental impact on patients includes neck-shoulder pain and/or atraumatic quadriplegia [Citation3,Citation4]. Nonsurgical medical treatment is widely employed for CS, but surgery is indicated when conservative treatments have failed [Citation5].
In clinical practice, hospital length of stay (LOS) is a measure of the success of surgery, but it is inadequate [Citation6]. Southern et al. labeled patients based on their LOS at the hospital and investigated the association between LOS and patient outcomes [Citation7]. LOS is a crucial indicator in assessing in-patient management, as prolonged hospitalization can lead to additional hospital days (AHD) for patients, potentially resulting in further complications [Citation8]. Several factors, such as medical severity and psychosocial and institutional factors, influence AHD. Early prediction and early intervention of factors influencing AHD can help patients recover quicker, alleviate doctor-patient conflicts, improve management efficiency, and ease the economic burden for both patients and governments [Citation9,Citation10]. In recent years, the healthcare industry has witnessed a growing interest in leveraging advanced data analytics techniques to enhance patient care and management. The research direction is almost focused on the relationship between ADH and cost [Citation11–13]. There are also some studies looking at the relationship between ADH and complications of certain diseases, such as post-cardiac infection, urinary tract infection, etc. [Citation14,Citation15]. However, studies on the analysis of disease-related factors and modeling to predict ADH are rarely reported.
Machine Learning (ML) is a branch of artificial intelligence (AI) involving algorithms that learn to make predictions from data [Citation16]. The main objective of ML is to make decisions rapidly and with minimum human intervention [Citation17]. ML algorithms are increasingly being used in the medical field for some clinical determinations [Citation18]. However, few studies predict AHD in patients undergoing cervical spine surgery. The current study used ML algorithms to construct prediction models based on common risk factors of patients undergoing cervical spine surgery. By employing sophisticated algorithms, we aim to identify subtle patterns and relationships within patient data that may escape traditional analytical approaches. This study helps to provide a nuanced understanding of the factors that influence longer hospital stays, thereby facilitating more targeted and effective patient care strategies.
Materials and methods
Patient population, inclusion, and exclusion criteria
A total of 945 patients were recruited. The mean number of LOS calculated for the total sample was 8.64 ± 3.7 days. The study protocol was authorized and supervised by the Ethics Committee of the First Affiliated Hospital of Guangxi Medical University (Nanning, China). The Approval Number: 2023-E018-01. We identified and collected clinical data from patients who were diagnosed with CS and underwent cervical spine surgery at the Department of Spinal Osteopathology in the First Affiliated Hospital of Guangxi Medical University. The data collection spanned from June 2012 to June 2021. The inclusion criteria for the patients were as follows: (1) Severe pain of cervical spondylotic radiculopathy which affected sleep; (2) Moderate or severe cervical spondylotic myelopathy; (3) With neurological deficit, muscular atrophy; (4) Conservative treatment for three months with poor effect; (5) Completion of all preoperative examinations; (6) Successful completion of the operation of cervical decompression and internal fixation. The exclusion criteria are involved in the following aspects: (1) Patients with cervical tuberculosis, cervical tumor, and cervical medullary space-occupying lesions; (2) Patients who have not completed cervical decompression and internal fixation; (3) Patients with cervical spinal cord injury due to trauma; (4) Patients with incomplete clinical records.
Clinical data
The clinical data of patients with CS and having undergone cervical spine surgery was collected in this study. A LOS equal to or <8.64 days was categorized as the AHD-negative group (n = 539), and a LOS > 8.64 days comprised the AHD-positive group (n = 406). The collected data was randomly divided into training and validation cohorts using a 7:3 ratio. The training cohort included 285 patients with AHD and 378 patients without AHD. The validation cohort included 121 patients with AHD and 161 patients without AHD. The parameters evaluated included general conditions, such as age, gender, body mass index (BMI), smoking, history of cervical surgery, other infections, and emergency admission. Chronic diseases, such as hypertension, diabetes, chronic obstructive pulmonary disease (COPD), coronary heart disease, liver and kidney dysfunction, cerebrovascular disease, peptic ulcer, history of malignant tumor, osteoporosis, ankylosing spondylitis, and rheumatoid arthritis, were also assessed. Preoperative clinical scores, including the visual analog score (VAS), American Spinal Injury Association (ASIA) scores, and Japanese Orthopedic Association (JOA) improvement rate, were recorded. Preoperative radiographic data included ossification of the anterior longitudinal ligament (OALL), ossification of the posterior longitudinal ligament (OPLL), cervical instability, and magnetic resonance imaging T2-weighted imaging high signal (MRI T2WIHS). The primary indicators were operative approach, operated segment, operation time, intraoperative bleeding volume, the volume of drainage, blood transfusion, and LOS. The postoperative complications were collected, such as dyspnea, cerebrovascular accident, heart failure, postoperative peptic ulcer, pneumonia, sepsis, mental disorder, deep venous thrombosis (DVT), pneumoderma, dysphagia, hoarseness, cerebrospinal fluid leak (CSFL), palsy of C5, axial pain, esophagostomy, local hematoma formation, sense of girdle, ischemia-reperfusion injury (IRI), urinary tract infection, surgical site infection, additional antibiotics, were recorded. In total, we collected 53 features.
Statistical analysis
Statistical analyses were performed by using SPSS Statistics 23 (IBM Corporation) and R software (version 4.2.2; https://www.R-project.org). Quantitative data were presented as mean ± standard deviation. Quantitative data between the two groups were compared by independent sample t-test and Mann–Whitney U test. Categorical variables were grouped and compared using Chi-square test. A probability <0.05 was considered statistically significant.
The ML operations would become very difficult upon including all 53 features. Hence, all components with p < 0.05 were first screened for the training cohort. Further screening was then continued using ML algorithms.
Model development and validation
We adopted 10-fold cross-validation (CV) on the training cohort to avoid overfitting and find the optimal hyperparameters. This study developed ML-based models like the Lasso regression [Citation19], random forest (RF) [Citation18], and support vector machine (SVM) recursive feature elimination (SVM-RFE) [Citation20] for predicting AHD-related risk factors in patients undergoing cervical spine surgery. The intersections of the variables screened by the above algorithms were used to build the nomogram model to predict AHD.
The prediction model ability was verified by using the area under the curve (AUC) of the receiver operating characteristic curve (ROC) and C-index (10,000 bootstraps resamples) to evaluate the performance of the nomogram. Calibration curve and decision curve analysis (DCA) were performed to test the calibration performance and clinical utility.
Results
Characteristics of the patients
In total, 25 statistically significant risk factors were identified for AHD. They included age, gender, BMI, history of cervical surgery, other infections, diabetes, coronary heart disease, osteoporosis, ASIA scores, OALL, MRI T2WIHS, operative approach, operated segment, operation time, intraoperative bleeding volume, the volume of drainage, blood transfusion, dyspnea, pneumonia, mental disorder, CSFL, palsy of C5, IRI, surgical site infection, and additional antibiotics ().
Table 1. Statistically significant risk factors for AHD between the two groups.
Lasso regression
The LASSO regression analysis screened out 22 risk factors that exhibited significant differences between the patients with AHD and without AHD (). These factors were gender, age, BMI, ASIA scores, history of cervical surgery, OALL, MRI T2WIHS, operative approach, operated segment, intraoperative bleeding volume, the volume of drainage, diabetes, coronary heart disease, osteoporosis, other infections, dyspnea, pneumonia, mental disorder, palsy of C5, IRI, surgical site infection, and additional antibiotics.
Figure 1. The 22 risk factors screened out by LASSO regression with significant differences between the patients with and without additional hospital days.
Random forest
The random forest algorithms ‘%IncMSE’ and ‘IncNodePurity’ screened the ten most important factors. The ideal regression effect was obtained by retaining these factors after 10-fold cross-validation (; ).
Figure 2. The importance of the top ten variables in the Random Forest (RF) model. BMI: body mass index; MRI T2WIHS: magnetic resonance imaging T2-weighted imaging high signal; ASIA: American Spinal Injury Association.
Table 2. The final selection of random forest (RF) regression.
SVM-RFE
The SVM-RFE calculation identified a collection of 19 risk factors. They included the volume of drainage, intraoperative bleeding volume, MRI T2WIHS, operation time, BMI, age, operated segment, OALL, blood transfusion, operative approach, ASIA scores, diabetes, additional antibiotics, gender, osteoporosis, IRI, history of cervical surgery, coronary heart disease, and other infections. The error rate was the lowest in SVM-RFE when these 19 factors were selected as diagnostic models. Thus, all the factors included were meaningful for diagnosis ().
Figure 3. Selection of 19 factors by the support vector machine recursive feature elimination (SVM-RFE).
The intersection of factors obtained from Lasso regression, RF, and SVM-RFE analyses
Finally, nine factors were screened out as the intersection of the above three ML algorithms above, and a Venn diagram was generated (). These nine factors comprised gender, age, BMI, ASIA scores, MRI T2WIHS, operated segment, intraoperative bleeding volume, the volume of drainage, and diabetes.
Figure 4. The intersection of nine factors screened using Lasso regression, random forest (RF), support vector machine recursive feature elimination (SVM-RFE).
The prediction model
A nomogram was based on the intersection of factors, such as gender, age, BMI, ASIA scores, MRI T2WIHS, operated segment, intraoperative bleeding volume, the volume of drainage, and diabetes (). The AUC of the ROC of the nomogram model was 0.753 (). The calibration curve exhibited a satisfactory agreement between the nomogram predictions and actual probabilities (). The C-index was 0.788 (95% confidence interval: 0.73214–0.84386). In the DCA (), the threshold probability of the nomogram ranged from 1 to 99%.
Figure 5. The nomogram model to predict AHD in patients undergoing cervical spine surgery. AHD: additional hospital day; BMI: body mass index; MRI T2WIHS: magnetic resonance imaging T2-weighted imaging high signal; ASIA: American Spinal Injury Association.
Figure 6. Receiver operating characteristic curve analysis (A), calibration curves (B), and decision curve analysis (C) of the nomogram prediction in the training cohort. AUC: area under the curve; AHD: additional hospital day.
Model validation
Finally, the model was validated in the validation cohort. The AUC of the ROC of the nomogram model was 0.777 (). In the DCA, the threshold probability of the nomogram ranged from 1 to 75% ().
Figure 7. Receiver operating characteristic curve analysis (A) and decision curve analysis (B) of the nomogram prediction in the validation cohort. AUC: area under the curve; AHD: additional hospital day.
Discussion
The management of LOS is an important index to judge the capacity of hospital’s comprehensive ability. However, the AHD is an important indicator of LOS management and should be studied more frequently in routine medical practice. AHD may cause additional complications for hospitalized patients, increase doctor-patient conflicts, reduce management efficiency, and elevate the economic burden for both patients and government [Citation9,Citation10]. The prevalence of CS has been increasing annually, especially in the aging population. This frequency can be detrimental to the patients or burden the families severely [Citation3,Citation4].
This study focused on identifying AHD in patients undergoing cervical spine surgery. AHD is a multifactorial condition affected by medical severity and psychosocial and institutional factors. It is important to consider as many risk factors as possible in the analysis of AHD, such as general conditions, chronic diseases, preoperative clinical scores, preoperative radiographic data, operative indicators, and complications. Too many risk factors may increase the sensitivity [Citation21]. However, studying numerous factors appears impractical and may result in a complex model with little predictive power [Citation22]. Therefore, the current prediction model was built by considering the intersection of factors from three ML algorithms (Lasso Regression, RF, and SVM-RFE).
Nine factors were finally included in the prediction model (). The general conditions studied were gender, age, and BMI. Men were more likely to develop AHD than women, consistent with several previous studies [Citation23–25]. Increased age was also associated with AHD [Citation26]. Müller et al. [Citation27] found that LOS increased stepwise with higher BMI values. However, opposite results were observed in the present study, wherein a negative correlation was observed between BMI and LOS [Citation28]. Perhaps the cause lies in a severely low BMI caused a decrease in activities of daily living [Citation29]. Lower ASIA scores and MRI T2WIHS were related to AHD in this study. The operated segment, intraoperative bleeding volume, and the volume of drainage were associated with surgical trauma, and they showed a positive correlation with AHD. For the operated segment, the score of three levels was lower than that of one level. This phenomenon may be caused by selection bias due to single-center study and geographical differences; other reasons need to be studied further. In addition, diabetes was the only chronic disease to be included in the final prediction model. Diabetes causes changes in multiple organ systems [Citation30] and increases the risk of perioperative complications [Citation31].
The machine learning method is divided into three approaches: supervised learning, unsupervised learning, and semi-supervised learning [Citation32]. Unsupervised methods are advantageous when the labeled data is limited. On the other hand, supervised learning is preferred if a large, representative, and correctly labeled training set is available [Citation33]. This was the case for the data characteristics in the present study. Hence, supervised learning was chosen. The LASSO method avoids compressing variables with greater parameter estimations while precisely compressing them with smaller parameter estimates [Citation34]. RF is widely used for feature selection, and the importance of each variable can be computed on the decision tree using their features [Citation35]. Furthermore, SVM-RFE is an effective method for reducing feature dimensions to reach the highest accuracy with the least number of channels [Citation36]. Therefore, Lasso regression, RF, and SVM-RFE were preferred for feature selection.
The random forest algorithms, specifically the ‘%IncMSE’ and ‘IncNodePurity’, were used to screen and identify the 10 most important factors. These factors were selected based on their impact on the mean squared error (‘%IncMSE’) and node purity (‘IncNodePurity’). To achieve the optimal regression effect, these ten important factors were retained after conducting a 10-fold cross-validation. This cross-validation process helps to ensure the robustness and generalizability of the regression model. By retaining these factors, we aim to capture the most influential variables in the dataset and improve the accuracy and performance of the regression analysis.
The prediction model encountered no postoperative complications. Postoperative complications were important for AHD but were not included in the intersection of factors from the three ML algorithms. This was unexpected, but that was exactly the characteristic of ML to make decisions with minimum intervention from humankind. This action is also the advantage of the prediction model, which predicts AHD in the preoperative and immediate postoperative periods.
The AUC of the ROC in the prediction model was 0.753 () in the training cohort and 0.777 () in the validation cohort. Thus, the model possessed excellent discriminative power. The calibration curve revealed the effectivity of the predictive model (). In addition, the prediction model had strong prognostic ability in patients with AHD (C-index, 0.788). The DCA suggested a tremendous potential clinical application value of this nomogram ().
The present study has a few limitations. Firstly, this is a retrospective study. Secondly, the research objects were from the same medical center in West China, which could create data selection bias. Thirdly, the external verification was lacking in this study.
Conclusion
In this study, we successfully developed an ML-based model for predicting AHD in patients undergoing cervical spine surgery. The results demonstrate the potential of this model to assist clinicians in identifying AHD and optimizing perioperative treatment strategies.
Acknowledgments
We thank the reviewers and editors for their helpful comments on this article.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Additional information
Funding
References
- Liu B, Chu Y, Ma J, et al. Analysis of risk factors for C5 nerve root paralysis after posterior cervical decompression. BMC Musculoskelet Disord. 2021;22(1):614. doi:10.1186/s12891-021-04434-y.
- Marie-Hardy L, Pascal-Moussellard H. Degenerative cervical myelopathy. Rev Neurol. 2021;177(5):1–10. doi:10.1016/j.neurol.2020.11.015.
- Luyao H, Xiaoxiao Y, Tianxiao F, Yuandong L, Ping W. Management of cervical spondylotic radiculopathy: a systematic review. Global Spine J. 2022;12(8):1912–1924, doi:10.1177/21925682221075290.
- Yin M, Xu C, Ma J, et al. A bibliometric analysis and visualization of current research trends in the treatment of cervical spondylotic myelopathy. Global Spine J. 2021;11(6):988–998. doi:10.1177/2192568220948832.
- Zhang B, Chen G, Chen X, et al. Cervical ossification of ligamentum flavum: elaborating an underappreciated but occasional contributor to myeloradiculopathy in aging population based on synthesis of individual participant data. Clin Interv Aging. 2021;16:897–908. doi:10.2147/CIA.S313357.
- Mutair AA, Al Mutairi A, Schwebius D. The retention effect of staff education programme: sustaining a decrease in hospital-acquired pressure ulcers via culture of care integration. Int Wound J. 2021;18(6):843–849. doi:10.1111/iwj.13586.
- Southern W, Arnsten J. Increased risk of mortality among patients cared for by physicians with short length-of-stay tendencies. J Gen Intern Med. 2015;30(6):712–718. doi:10.1007/s11606-014-3155-8.
- Bell M, Eriksson L, Svensson T, et al. Days at home after surgery: an integrated and efficient outcome measure for clinical trials and quality assurance. EClinicalMedicine. 2019;11:18–26. doi:10.1016/j.eclinm.2019.04.011.
- Xiao H, Huang W, Qin X, et al. Day ward glaucoma patients have lower depression levels and higher glaucoma knowledge levels than inpatients. J Ophthalmol. 2019;2019:4182030–4182037. doi:10.1155/2019/4182030.
- Schweiger C, Manica D. Ongoing laryngeal stenosis: conservative management and alternatives to tracheostomy. Front Pediatr. 2020;8:161. doi:10.3389/fped.2020.00161.
- Majumder MAA, Rahman S, Cohall D, et al. Antimicrobial stewardship: fighting antimicrobial resistance and protecting global public health. Infect Drug Resist. 2020;13:4713–4738. doi:10.2147/IDR.S290835.
- Dumville JC, McFarlane E, Edwards P, et al. Preoperative skin antiseptics for preventing surgical wound infections after clean surgery. Cochrane Database Syst Rev. 2015;2015(4):CD003949. doi:10.1002/14651858.CD003949.pub4.
- Doupnik SK, Lawlor J, Zima BT, et al. Mental health conditions and medical and surgical hospital utilization. Pediatrics. 2016;138(6):e20162416. doi:10.1542/peds.2016-2416.
- Algado-Sellés N, Mira-Bernabeu J, Gras-Valentí P, et al. Estimated costs associated with surgical site infections in patients undergoing cholecystectomy. Int J Environ Res Public Health. 2022;19(2):764. doi:10.3390/ijerph19020764.
- Crintea A, Carpa R, Mitre AO, et al. Nanotechnology involved in treating urinary tract infections: an overview. Nanomaterials. 2023;13(3):555. doi:10.3390/nano13030555.
- Thakkar HK, Liao WW, Wu CY, et al. Predicting clinically significant motor function improvement after contemporary task-oriented interventions using machine learning approaches. J NeuroEngineering Rehabil. 2020;17(1):131. doi:10.1186/s12984-020-00758-3.
- Giambagli L, Buffoni L, Carletti T, et al. Machine learning in spectral domain. Nat Commun. 2021;12(1):1330. doi:10.1038/s41467-021-21481-0.
- Yang B, Gao L, Wang X, et al. Application of supervised machine learning algorithms to predict the risk of hidden blood loss during the perioperative period in thoracolumbar burst fracture patients complicated with neurological compromise. Front Public Health. 2022;10:969919. doi:10.3389/fpubh.2022.969919.
- Wang S, Su W, Zhong C, et al. An eight-circRNA assessment model for predicting biochemical recurrence in prostate cancer. Front Cell Dev Biol. 2020;8:599494. doi:10.3389/fcell.2020.599494.
- Ivanoska I, Trivodaliev K, Kalajdziski S, et al. Statistical and machine learning link selection methods for brain functional networks: review and comparison. Brain Sci. 2021;11(6):735. doi:10.3390/brainsci11060735.
- de Assis RR, Jain A, Nakajima R, et al. Analysis of SARS-CoV-2 antibodies in COVID-19 convalescent blood using a coronavirus antigen microarray. Nat Commun. 2021;12(1):6. doi:10.1038/s41467-020-20095-2.
- Kahm M, Navarrete C, Llopis-Torregrosa V, et al. Potassium starvation in yeast: mechanisms of homeostasis revealed by mathematical modeling. PLOS Comput Biol. 2012;8(6):e1002548. doi:10.1371/journal.pcbi.1002548.
- Zhou Q, Fan L, Lai X, et al. Estimating extra length of stay and risk factors of mortality attributable to healthcare-associated infection at a Chinese university hospital: a multi-state model. BMC Infect Dis. 2019;19(1):975. doi:10.1186/s12879-019-4474-5.
- Wang QH, Wang X, Bu XL, et al. Comorbidity burden of dementia: a hospital-based retrospective study from 2003 to 2012 in seven cities in China. Neurosci Bull. 2017;33(6):703–710. doi:10.1007/s12264-017-0193-3.
- Sultana R, Luby SP, Gurley ES, et al. Cost of illness for severe and non-severe diarrhea borne by households in a low-income urban community of Bangladesh: a cross-sectional study. PLOS Negl Trop Dis. 2021;15(6):e0009439. doi:10.1371/journal.pntd.0009439.
- Chan KF, Kwok WC, Ma TF, et al. Territory-wide study on hospital admissions for asthma exacerbations in the COVID-19 pandemic. Ann Am Thorac Soc. 2021;18(10):1624–1633. doi:10.1513/AnnalsATS.202010-1247OC.
- Müller M, Gutwerk A, Greve F, et al. The association between high body mass index and early clinical outcomes in patients with proximal femur fractures. J Clin Med. 2020;9(7):2076. doi:10.3390/jcm9072076.
- Mekal D, Czerw A, Deptala A. Dietary behaviour and nutrition in patients with COPD treated with long-term oxygen therapy. Int J Environ Res Public Health. 2021;18(23):12793. doi:10.3390/ijerph182312793.
- Shirai Y, Momosaki R, Kokura Y, et al. Validation of Asian body mass index cutoff values for the classification of malnutrition severity according to the global leadership initiative on malnutrition criteria in patients with chronic obstructive pulmonary disease exacerbations. Nutrients. 2022;14(22):4746. doi:10.3390/nu14224746.
- Dham D, Roy B, Gowda A, et al. 4-Hydroxy-2-nonenal, a lipid peroxidation product, as a biomarker in diabetes and its complications: challenges and opportunities. Free Radic Res. 2021;55(5):510–524. doi:10.1080/10715762.2020.1866756.
- Grant B, Chowdhury TA. New guidance on the perioperative management of diabetes. Clin Med. 2022;22(1):41–44. doi:10.7861/clinmed.2021-0355.
- Razali NAM, Malizan NA, Hasbullah NA, et al. Opinion mining for national security: techniques, domain applications, challenges and research opportunities. J Big Data. 2021;8(1):150. doi:10.1186/s40537-021-00536-5.
- Ionita-Laza I, McCallum K, Xu B, et al. A spectral approach integrating functional genomic annotations for coding and noncoding variants. Nat Genet. 2016;48(2):214–220. doi:10.1038/ng.3477.
- Chen X, Wang H, Huang K, et al. CT-based radiomics signature with machine learning predicts MYCN amplification in pediatric abdominal neuroblastoma. Front Oncol. 2021;11:687884. doi:10.3389/fonc.2021.687884.
- Sujeeun LY, Goonoo N, Ramphul H, et al. Correlating in vitro performance with physico-chemical characteristics of nanofibrous scaffolds for skin tissue engineering using supervised machine learning algorithms. R Soc Open Sci. 2020;7(12):201293. doi:10.1098/rsos.201293.
- Yi W, Qiu S, Wang K, et al. EEG oscillatory patterns and classification of sequential compound limb motor imagery. J NeuroEngineering Rehabil. 2016;13(1):11. doi:10.1186/s12984-016-0119-8.