Predicting return to work after traumatic brain injury using machine learning and administrative data.

BACKGROUND: Accurate patient-specific predictions on return-to-work after traumatic brain injury (TBI) can support both clinical practice and policymaking. The use of machine learning on large administrative data provides interesting opportunities to create such prognostic models. AIM: The current study assesses whether return-to-work one year after TBI can be predicted accurately from administrative data. Additionally, this study explores how model performance and feature importance change depending on whether a distinction is made between mild and moderate-to-severe TBI. METHODS: This study used a population-based dataset that combined discharge, claims and social security data of patients hospitalized with a TBI in Belgium during the year 2016. The prediction of TBI was attempted with three algorithms, elastic net logistic regression, random forest and gradient boosting and compared in their performance by their accuracy, sensitivity, specificity and area under the receiver operator curve (ROC AUC). RESULTS: The distinct modelling algorithms resulted in similar results, with 83% accuracy (ROC AUC 85%) for a binary classification of employed vs. not employed and up to 76% (ROC AUC 82%) for a multiclass operationalization of employment outcome. Modelling mild and moderate-to-severe TBI separately did not result in considerable differences in model performance and feature importance. The features of main importance for return-to-work prediction were related to pre-injury employment. DISCUSSION: While clearly offering some information beneficial for predicting return-to-work, administrative data needs to be supplemented with additional information to allow further improvement of patient-specific prognose.

Traumatic brain injury hospitalizations in Belgium: A brief overview of incidence, population characteristics, and outcomes.

Background: There is a need for complete and accurate epidemiological studies for traumatic brain injury (TBI). Secondary use of administrative data can provide country-specific population data across the full spectrum of disease. Aim: This study aims to provide a population-based overview of Belgian TBI hospital admissions as well as their health-related and employment outcomes. Methods: A combined administrative dataset with deterministic linkage at individual level was used to assess all TBI hospitalizations in Belgium during the year 2016. Discharge data were used for patient selection and description of injuries. Claims data represented the health services used by the patient and health-related follow-up beyond hospitalization. Finally, social security data gave insight in changes to employment situation. Results: A total of 17,086 patients with TBI were identified, with falls as the predominant cause of injury. Diffuse intracranial injury was the most common type of TBI and 53% had injuries to other body regions as well. In-hospital mortality was 6%. The median length of hospital stay was 2 days, with 20% being admitted to intensive care and 28% undergoing surgery. After hospitalization, 23% had inpatient rehabilitation. Among adults in the labor force pre-injury, 72% of patients with mild TBI and 59% with moderate-to-severe TBI returned to work within 1 year post-injury. Discussion: Administrative data are a valuable resource for population research. Some limitations need to be considered, however, which can in part be overcome by enrichment of administrative datasets with other data sources such as from trauma registries.

Quantifying injury severity for traumatic brain injury with routinely collected health data.

BACKGROUND: Routinely collected health data (RCHD) offers many opportunities for traumatic brain injury (TBI) research, in which injury severity is an important factor. OBJECTIVE: The use of clinical injury severity indices in a context of RCHD is explored, as are alternative measures created for this specific purpose. To identify useful scales for full body injury severity and TBI severity this study focuses on their performance in predicting these currently used indices, while accounting for age and comorbidities. DATA: This study utilized an extensive population-based RCHD dataset consisting of all patients with TBI admitted to any Belgian hospital in 2016. METHODS: Full body injury severity is scored based on the (New) Injury Severity Score ((N)ISS) and the ICD-based Injury Severity Score (ICISS). For TBI specifically, the Abbreviated Injury Scale (AIS) Head, Loss of Consciousness and the ICD-based Injury Severity Score for TBI injuries (ICISS) were used in the analysis. These scales were used to predict three outcome variables strongly related to injury severity: in-hospital death, admission to intensive care and length of hospital stay. For the prediction logistic regressions of the different injury severity scales and TBI severity indices were used, and error rates and the area under the receiver operating curve were evaluated visually. RESULTS: In general, the ICISS had the best predictive performance (error rate between 0.06 and 0.23; AUC between 0.82 [0.81;0.83] and 0.86 [0.85;0.86]). A clearly increasing error rate can be noticed with advancing age and accumulating comorbidity. CONCLUSION: Both for full body injury severity and TBI severity, the ICISS tends to outperform other scales. It is therefore the preferred scale for use in research on TBI in the context of RCHD. In their current form, the severity scales are not suitable for use in older populations.

Health care utilization after hospitalization following a road traffic accident.

Purpose: In recent years, there has been an increasing interest in measuring and modeling health care utilization. However, only limited research has been performed in the field of health care utilization following road traffic accidents. This article aims to measure the incremental health care utilization after hospital discharge after a road traffic accident and explore the association between socio-demographic and injury-related variables and health care utilization.Material and methods: Generalized linear models with negative binomial distribution and log-link were executed per type of health care provider (general practitioner, medical specialists, rehabilitation services and outpatient nursing care) and per type of discharge location (discharged to home, discharged to in-hospital rehabilitation). Health care utilization of the 6 months after discharge was compared with the 6 months before the accident (baseline care).Results: Health care utilization six months after discharge is significantly higher than baseline care, except for outpatient nursing care and general practitioners in in-hospital rehabilitation. The increase in visits to medical specialists ranged on average between 1 and 2.2 visits. For general practitioner, there was an increase of 0.4 visits and 0.8 in outpatient nursing care for those who returned home after acute hospitalization. The average increase in rehabilitation services ranged between 3.6 and 20. Associated influential factors differ per health care provider and discharge destination.Conclusion: Evidence of this study suggests higher health care utilization during the first 6 months following hospitalization due to a road traffic injury, compared with baseline care. Associated variables differ per type of health care provider and discharge-destination. More in-depth research on subgroups is needed.Implications for rehabilitationHealth care utilization varies across different patient characteristics and type of injuries which should be considered in the communication with patients on their care trajectory post-discharge.General descriptions of health care utilization in traffic victims at the population level are lacking. Output similar to our study could serve as a reference for post-discharge care planning.The research output can be a starting point for future research on quality indicators of the expected quantity of care.Efforts must be made to estimate suchlike reference tables on post-discharge services in other patient groups and secondary data are a suitable data-source for those analyses.

The Incremental Hospitalization Cost of Significant Transport-Related Traumatic Brain Injury.

AIM: This study aims to determine the incremental cost of acute hospitalization for traumatic brain injury (TBI) compared with matched controls. A second purpose is to identify the factors contributing to this hospital costs. METHODOLOGY: Analyses were performed on administrative data for injured patients, hospitalized in Belgium between 2009 and 2011 following a road traffic accident. Cases were matched to a control with similar injuries but without TBI. The incremental hospitalization cost of TBI and the factors contributing to the hospital costs were determined using multivariable regression modeling with gamma distribution and log link. RESULTS: A descriptive comparison of cases and controls shows clear differences in healthcare utilization and costs. The presence of a TBI increases the cost by a factor between 1.66 (95% confidence interval: 1.52-1.82) and 2.08 (95% confidence interval: 1.72-2.51). Regarding healthcare utilization, the most important determinants of hospital costs are surgical complexity, use of magnetic resonance imaging, intensive care unit admission, and mechanical ventilation. DISCUSSION: To our knowledge, this is the first matched-control study calculating the incremental hospitalization cost of TBI. The insights provided by this study are relevant in the context of prospective payments and can be an incentive for investments in prevention policies and extramural care.

The incremental cost of traumatic brain injury during the first year after a road traffic accident.

This study aims to determine the incremental cost of TBI during the first year after a traffic accident, compared to other patients with similar non-TBI injuries. Secondly, identification of factors associated with medical costs of TBI is pursued. Analyses were performed on administrative data for traffic victims hospitalised in Belgium between 2009 and 2011. Medical costs attributable to the accident are estimated over one year post-injury. Cases with TBI were matched to controls with similar non-TBI injuries to determine the incremental cost of TBI. Both aims of this research were assessed using regression analysis. The incremental cost of TBI is estimated to range between € 10 042 (95%CI [€8198; €11 887]) and €21 715 (95%CI [€13 5889; €29 540]). Age, problems with self-reliance, survival status, the occurrence of acute events and severity of TBI are significant predictors of medical costs. As to healthcare utilisation, MRI usage, inpatient rehabilitation facilities, nursing homes and readmissions to acute hospital stand out as having most influence on costs. This study reveals a considerable incremental cost of TBI. Policy-making bodies should be made aware of this phenomenon and a diversified policy should be considered when financing programs are discussed.

Determinants and Variations of Hospital Costs in Patients With Lumbar Radiculopathy Hospitalized for Spinal Surgery.

STUDY DESIGN: A retrospective study. OBJECTIVE: The aim of this study was to determine hospital costs related to surgery for lumbar radiculopathy and identify determinants of intramural costs based on minimal hospital and claims data. SUMMARY OF BACKGROUND DATA: Costs related to the initial hospitalization of patients undergoing surgery for lumbar radiculopathy make up the major part of direct health care expenditure in this population. Identifying factors influencing intramural costs can be beneficial for health care policy makers, and clinicians working with patients with lumbar radiculopathy. METHODS: The following data were collected from the University Hospital Brussels data warehouse for all patients undergoing surgery for lumbar radiculopathy in 2016 (n = 141): age, sex, primary diagnosis, secondary diagnoses, type of surgery, severity of illness (SOI), admission and discharge date, type of hospital admission, and all claims incurred for the particular hospital stay. Descriptive statistics for total hospital costs were performed. Univariate analyses were executed to explore associations between hospital costs and all other variables. Those showing a significant association (P < 0.05) were included in the multivariate general linear model analysis. RESULTS: Mean total hospital costs were &OV0556; 5016 ±â€Š188 per patient. Costs related to the actual residence (i.e., "hotel costs") comprised 53% of the total hospital costs, whereas 18% of the costs were claimed for the surgical procedure. Patients with moderate/major SOI had 44% higher hospital costs than minor SOI (P = 0.01). Presence of preadmission comorbidities incurred 46% higher costs (P = 0.03). Emergency procedures led to 72% higher costs than elective surgery (P < 0.001). Patients receiving spinal fusion had 211% higher hospital costs than patients not receiving this intervention (P < 0.001). CONCLUSION: Hospital costs in patients receiving surgery for lumbar radiculopathy are influenced by SOI, the presence of preadmission comorbidities, type of hospital admission (emergency vs. elective), and type of surgical procedure. LEVEL OF EVIDENCE: 3.

Return to work following surgery for lumbar radiculopathy: a systematic review.

BACKGROUND CONTEXT: Informing patients about postoperative return to work (RTW) expectations is of utmost importance because of the influence of realistic expectations on RTW outcomes. PURPOSE: We aimed to give an overview of the duration of sick leave and RTW rates after surgery for lumbar radiculopathy and to list predictors of and factors related to RTW. STUDY DESIGN: A systematic review was carried out. METHODS: A systematic literature search was conducted in PubMed, Web of Science, EMBASE, and SCOPUS. Full-text articles on RTW following surgery for lumbar radiculopathy were included through double-blind screening. Risk of bias was assessed using a modified version of the Downs and Black checklist. RESULTS: Sixty-three full-text articles (total sample size: 7,100 patients) were included. Risk of bias was scored low to high. Mean duration of sick leave ranged from 0.8 to 20 weeks. Within 0.1-240 months post surgery, 3%-100% of patients resumed work. Most important predictors for work resumption were preoperative work status, presence of comorbidities, age, sex and duration of preoperative symptoms. Duration of sick leave can be predicted by the preoperative level of pain or disability and presence of symptoms of depression, occupational mental stress, and lateral disc prolapse. Furthermore, less invasive surgical techniques were found to result in better RTW outcomes compared with more invasive techniques. CONCLUSIONS: Diverse results were found for RTW rates and duration of sick leave. Preoperative work status, presence of comorbidities, and several demographic factors were retrieved as predictors of RTW and duration of sick leave.

Variations in hospital costs after traffic injuries: The importance of sociodemographic aspects and comorbidities.

OBJECTIVE: The impact of sociodemographic aspects and comorbidities on the inpatient hospital care costs of traffic victims are not clear. The main goal of this study is to provide insights into the sociodemographic characteristics and clinical conditions (including comorbidities) of the victims that result in higher hospital costs. PARTICIPANTS: For the period 2009-2011, people admitted to a hospital as a result of a road traffic crash (N=64,304) were identified in the national Minimal Hospital Dataset, after which they were linked to their respective claims data from the sickness funds. METHODS: A generalized linear model was used to analyse hospital costs controlling for roadway user categories, demographics (gender, age, individual socioeconomic status (SES)), and clinical factors (the nature, location, and severity of injury, and comorbidities). RESULTS: The median hospital cost was € 2801 (IQR € 1510-€ 7175, 2015 Euros). There was no significant difference between gender. Low SES inpatients incurred 16% (95% CI: 14%-18%) higher hospital costs than inpatients of high SES. The presence of comorbidities was associated with an increased hospital cost, however with varying magnitude. For example traffic victims suffering from dementia incur significantly higher hospital costs than those who were not (49% higher, 95% CI: 44%-53%), whereas diabetes was associated with a smaller increase in costs compared to non-diabetics (13%, 95% CI: 10%-16%). CONCLUSION: Comorbidities and low SES are associated with higher hospital costs for traffic victims, notwithstanding their age, and the nature and the severity of their injury. The broad variability of hospital costs among trauma inpatients should be accounted for when reconsidering financing models. Furthermore, the strong predictive value of some comorbidities and SES on hospital costs should be considered when projections of future health care utilisation in traffic safety scenarios are prepared.

Predicting in-hospital mortality of traffic victims: A comparison between AIS-and ICD-9-CM-related injury severity scales when only ICD-9-CM is reported.

BACKGROUND AND AIM: Injury severity scores are important in the context of developing European and national goals on traffic safety, health-care benchmarking and improving patient communication. Various severity scores are available and are mostly based on Abbreviated Injury Scale (AIS) or International Classification of Diseases (ICD). The aim of this paper is to compare the predictive value for in-hospital mortality between the various severity scores if only International Classification of Diseases, 9th revision, Clinical Modification ICD-9-CM is reported. METHODOLOGY: To estimate severity scores based on the AIS lexicon, ICD-9-CM codes were converted with ICD Programmes for Injury Categorization (ICDPIC) and four AIS-based severity scores were derived: Maximum AIS (MaxAIS), Injury Severity Score (ISS), New Injury Severity Score (NISS) and Exponential Injury Severity Score (EISS). Based on ICD-9-CM, six severity scores were calculated. Determined by the number of injuries taken into account and the means by which survival risk ratios (SRRs) were calculated, four different approaches were used to calculate the ICD-9-based Injury Severity Scores (ICISS). The Trauma Mortality Prediction Model (TMPM) was calculated with the ICD-9-CM-based model averaged regression coefficients (MARC) for both the single worst injury and multiple injuries. Severity scores were compared via model discrimination and calibration. Model comparisons were performed separately for the severity scores based on the single worst injury and multiple injuries. RESULTS: For ICD-9-based scales, estimation of area under the receiver operating characteristic curve (AUROC) ranges between 0.94 and 0.96, while AIS-based scales range between 0.72 and 0.76, respectively. The intercept in the calibration plots is not significantly different from 0 for MaxAIS, ICISS and TMPM. DISCUSSION: When only ICD-9-CM codes are reported, ICD-9-CM-based severity scores perform better than severity scores based on the conversion to AIS.