Programming techniques for improving rule readability for rule-based information extraction natural language processing pipelines of unstructured and semi-structured medical texts.

BACKGROUND: Extraction of medical terms and their corresponding values from semi-structured and unstructured texts of medical reports can be a time-consuming and error-prone process. Methods of natural language processing (NLP) can help define an extraction pipeline for accomplishing a structured format transformation strategy. OBJECTIVES: In this paper, we build an NLP pipeline to extract values of the classification of malignant tumors (TNM) from unstructured and semi-structured pathology reports and import them further to a structured data source for a clinical study. Our research interest is not focused on standard performance metrics like precision, recall, and F-measure on the test and validation data. We discuss how with the help of software programming techniques the readability of rule-based (RB) information extraction (IE) pipelines can be improved, and therefore minimize the time to correct or update the rules, and efficiently import them to another programming language. METHODS: The extract rules were manually programmed with training data of TNM classification and tested in two separate pipelines based on design specifications from domain experts and data curators. Firstly we implemented each rule directly in one line for each extraction item. Secondly, we reprogrammed them in a readable fashion through decomposition and intention-revealing names for the variable declaration. To measure the impact of both methods we measure the time for the fine-tuning and programming of the extractions through test data of semi-structured and unstructured texts. RESULTS: We analyze the benefits of improving through readability of the writing of rules, through parallel programming with regular expressions (REGEX), and the Apache Uima Ruta language (AURL). The time for correcting the readable rules in AURL and REGEX was significantly reduced. Complicated rules in REGEX are decomposed and intention-revealing declarations were reprogrammed in AURL in 5 min. CONCLUSION: We discuss the importance of factor readability and how can it be improved when programming RB text IE pipelines. Independent of the features of the programming language and the tools applied, a readable coding strategy can be proven beneficial for future maintenance and offer an interpretable solution for understanding the extraction and for transferring the rules to other domains and NLP pipelines.

Pharmacotherapy, drug-drug interactions and potentially inappropriate medication in depressive disorders.

INTRODUCTION: The aim of this study was to describe the number and type of drugs used to treat depressive disorders in inpatient psychiatry and to analyse the determinants of potential drug-drug interactions (pDDI) and potentially inappropriate medication (PIM). METHODS: Our study was part of a larger pharmacovigilance project funded by the German Innovation Funds. It included all inpatients with a main diagnosis in the group of depressive episodes (F32, ICD-10) or recurrent depressive disorders (F33) discharged from eight psychiatric hospitals in Germany between 1 October 2017 and 30 September 2018 or between 1 January and 31 December 2019. RESULTS: The study included 14,418 inpatient cases. The mean number of drugs per day was 3.7 (psychotropic drugs = 1.7; others = 2.0). Thirty-one percent of cases received at least five drugs simultaneously (polypharmacy). Almost one half of all cases received a combination of multiple antidepressant drugs (24.8%, 95% CI 24.1%-25.5%) or a treatment with antidepressant drugs augmented by antipsychotic drugs (21.9%, 95% CI 21.3%-22.6%). The most frequently used antidepressants were selective serotonin reuptake inhibitors, followed by serotonin and norepinephrine reuptake inhibitors and tetracyclic antidepressants. In multivariate analyses, cases with recurrent depressive disorders and cases with severe depression were more likely to receive a combination of multiple antidepressant drugs (Odds ratio recurrent depressive disorder: 1.56, 95% CI 1.41-1.70, severe depression 1.33, 95% CI 1.18-1.48). The risk of any pDDI and PIM in elderly patients increased substantially with each additional drug (Odds Ratio: pDDI 1.32, 95% CI: 1.27-1.38, PIM 1.18, 95% CI: 1.14-1.22) and severity of disease (Odds Ratio per point on CGI-Scale: pDDI 1.29, 95% CI: 1.11-1.46, PIM 1.27, 95% CI: 1.11-1.44), respectively. CONCLUSION: This study identified potential sources and determinants of safety risks in pharmacotherapy of depressive disorders and provided additional data which were previously unavailable. Most inpatients with depressive disorders receive multiple psychotropic and non-psychotropic drugs and pDDI and PIM are relatively frequent. Patients with a high number of different drugs must be intensively monitored in the management of their individual drug-related risk-benefit profiles.

Polypharmacy and the risk of drug-drug interactions and potentially inappropriate medications in hospital psychiatry.

PURPOSE: The aim of this study was to analyze the epidemiology of polypharmacy in hospital psychiatry. Another aim was to investigate predictors of the number of drugs taken and the associated risks of drug-drug interactions and potentially inappropriate medications in the elderly. METHODS: Daily prescription data were obtained from a pharmacovigilance project sponsored by the Innovations Funds of the German Federal Joint Committee. RESULTS: The study included 47 071 inpatient hospital cases from eight different study centers. The mean number of different drugs during the entire stay was 6.1 (psychotropic drugs = 2.7; others = 3.4). The mean number of drugs per day was 3.8 (psychotropic drugs = 1.6; others = 2.2). One third of cases received at least five different drugs per day on average during their hospital stay (polypharmacy). Fifty-one percent of patients received more than one psychotropic drug simultaneously. Hospital cases with polypharmacy were 18 years older (p < 0.001), more likely to be female (52% vs. 40%, p < 0.001) and had more comorbidities (5 vs. 2, p < 0.001) than hospital cases without polypharmacy. The risks of drug-drug interactions (OR = 3.7; 95% CI = 3.5-3.9) and potentially inappropriate medication use in the elderly (OR = 2.2; CI = 1.9-2.5) substantially increased in patients that received polypharmacy. CONCLUSION: Polypharmacy is frequent in clinical care. The number of used drugs is a proven risk factor of adverse drug reactions due to drug-drug interactions and potentially inappropriate medication use in the elderly. The potential interactions and the specific pharmacokinetics and -dynamics of older patients should always be considered when multiple drugs are used.

Predicting the risk of drug-drug interactions in psychiatric hospitals: a retrospective longitudinal pharmacovigilance study.

OBJECTIVES: The aim was to use routine data available at a patient's admission to the hospital to predict polypharmacy and drug-drug interactions (DDI) and to evaluate the prediction performance with regard to its usefulness to support the efficient management of benefits and risks of drug prescriptions. DESIGN: Retrospective, longitudinal study. SETTING: We used data from a large multicentred pharmacovigilance project carried out in eight psychiatric hospitals in Hesse, Germany. PARTICIPANTS: Inpatient episodes consecutively discharged between 1 October 2017 and 30 September 2018 (year 1) or 1 January 2019 and 31 December 2019 (year 2). OUTCOME MEASURES: The proportion of rightly classified hospital episodes. METHODS: We used gradient boosting to predict respective outcomes. We tested the performance of our final models in unseen patients from another calendar year and separated the study sites used for training from the study sites used for performance testing. RESULTS: A total of 53 909 episodes were included in the study. The models' performance, as measured by the area under the receiver operating characteristic, was 'excellent' (0.83) and 'acceptable' (0.72) compared with common benchmarks for the prediction of polypharmacy and DDI, respectively. Both models were substantially better than a naive prediction based solely on basic diagnostic grouping. CONCLUSION: This study has shown that polypharmacy and DDI can be predicted from routine data at patient admission. These predictions could support an efficient management of benefits and risks of hospital prescriptions, for instance by including pharmaceutical supervision early after admission for patients at risk before pharmacological treatment is established.

Patient Survival After Kidney Transplantation: Important Role of Graft-sustaining Factors as Determined by Predictive Modeling Using Random Survival Forest Analysis.

BACKGROUND: Identification of the relevant factors for death can improve patient's individual risk assessment and decision making. A well-documented patient cohort (n = 892) in a renal transplant program with protocol biopsies was used to establish multivariable models for risk assessment at 3 and 12 months posttransplantation by random survival forest analysis. METHODS: Patients transplanted between 2000 and 2007 were observed for up to 11 years. Loss to follow-up was negligible (n = 15). A total of 2251 protocol biopsies and 1214 biopsies for cause were performed. All rejections and clinical borderline rejections in protocol biopsies were treated. RESULTS: Ten-year patient survival was 78%, with inferior survival of patients with graft loss. Using all pre- and posttransplant variables until 3 and 12 months (n = 65), the obtained models showed good performance to predict death (concordance index: 0.77-0.78). Validation with a separate cohort of patients (n = 349) showed a concordance index of 0.76 and good discrimination of risks by the models, despite substantial differences in clinical variables. Random survival forest analysis produced robust models over a wide range of parameter settings. Besides well-established risk factors like age, cardiovascular disease, type 2 diabetes, and graft function, posttransplant urinary tract infection and rejection treatment were important factors. Urinary tract infection and rejection treatment were not specifically associated with death due to infection or malignancy but correlated strongly with inferior graft function and graft loss. CONCLUSIONS: The established models indicate the important areas that need special attention in the care of renal transplant patients, particularly modifiable factors like graft rejection and urinary tract infection.

HiGHmed - An Open Platform Approach to Enhance Care and Research across Institutional Boundaries.

INTRODUCTION: This article is part of the Focus Theme of Methods of Information in Medicine on the German Medical Informatics Initiative. HiGHmed brings together 24 partners from academia and industry, aiming at improvements in care provision, biomedical research and epidemiology. By establishing a shared information governance framework, data integration centers and an open platform architecture in cooperation with independent healthcare providers, the meaningful reuse of data will be facilitated. Complementary, HiGHmed integrates a total of seven Medical Informatics curricula to develop collaborative structures and processes to train medical informatics professionals, physicians and researchers in new forms of data analytics. GOVERNANCE AND POLICIES: We describe governance structures and policies that have proven effective during the conceptual phase. These were further adapted to take into account the specific needs of the development and networking phase, such as roll-out, carerelated aspects and our focus on curricula development in Medical Inform atics. ARCHITECTURAL FRAMEWORK AND METHODOLOGY: To address the challenges of organizational, technical and semantic interoperability, a concept for a scalable platform architecture, the HiGHmed Platform, was developed. We outline the basic principles and design goals of the open platform approach as well as the roles of standards and specifications such as IHE XDS, openEHR, SNOMED CT and HL7 FHIR. A shared governance framework provides the semantic artifacts which are needed to establish semantic interoperability. USE CASES: Three use cases in the fields of oncology, cardiology and infection control will demonstrate the capabilities of the HiGHmed approach. Each of the use cases entails diverse challenges in terms of data protection, privacy and security, including clinical use of genome sequencing data (oncology), continuous longitudinal monitoring of physical activity (cardiology) and cross-site analysis of patient movement data (infection control). DISCUSSION: Besides the need for a shared governance framework and a technical infrastructure, backing from clinical leaders is a crucial factor. Moreover, firm and sustainable commitment by participating organizations to collaborate in further development of their information system architectures is needed. Other challenges including topics such as data quality, privacy regulations, and patient consent will be addressed throughout the project.

Machine Learning and Data Analytics in Pervasive Health.

INTRODUCTION: This accompanying editorial provides a brief introduction to this focus theme, focused on "Machine Learning and Data Analytics in Pervasive Health". OBJECTIVE: The innovative use of machine learning technologies combining small and big data analytics will support a better provisioning of healthcare to citizens. This focus theme aims to present contributions at the crossroads of pervasive health technologies and data analytics as key enablers for achieving personalised medicine for diagnosis and treatment purposes. METHODS: A call for paper was announced to all participants of the "11th International Conference on Pervasive Computing Technologies for Healthcare", to different working groups of the International Medical Informatics Association (IMIA) and European Federation of Medical Informatics (EFMI) and was published in June 2017 on the website of Methods of Information in Medicine. A peer review process was conducted to select the papers for this focus theme. RESULTS: Four papers were selected to be included in this focus theme. The paper topics cover a broad range of machine learning and data analytics applications in healthcare including detection of injurious subtypes of patient-ventilator asynchrony, early detection of cognitive impairment, effective use of small data sets for estimating the performance of radiotherapy in bladder cancer treatment, and the use negation detection in and information extraction from unstructured medical texts. CONCLUSIONS: The use of machine learning and data analytics technologies in healthcare is facing a renewed impulse due to the availability of large amounts and new sources of human behavioral and physiological data, such as that captured by mobile and pervasive devices traditionally considered as nonmainstream for healthcare provision and management.

mHealth Application Areas and Technology Combinations*. A Comparison of Literature from High and Low/Middle Income Countries.

BACKGROUND: With the continuous and enormous spread of mobile technologies, mHealth has evolved as a new subfield of eHealth. While eHealth is broadly focused on information and communication technologies, mHealth seeks to explore more into mobile devices and wireless communication. Since mobile phone penetration has exceeded other infrastructure in low and middle-income countries (LMICs), mHealth is seen as a promising component to provide pervasive and patient-centered care. OBJECTIVES: The aim of our research work for this paper is to examine the mHealth literature to identify application areas, target diseases, and mHealth service and technology types that are most appropriate for LMICs. METHODS: Based on the 2011 WHO mHealth report, a combination of search terms, all including the word "mHealth", was identified. A literature review was conducted by searching the PubMed and IEEE Xplore databases. Articles were included if they were published in English, covered an mHealth solution/ intervention, involved the use of a mobile communication device, and included a pilot evaluation study. Articles were excluded if they did not provide sufficient detail on the solution covered or did not focus on clinical efficacy/effectiveness. Cross-referencing was also performed on included articles. RESULTS: 842 articles were retrieved and analyzed, 255 of which met the inclusion criteria. North America had the highest number of applications (n=74) followed by Europe (n=50), Asia (n=44), Africa (n=25), and Australia (n=9). The Middle East (n=5) and South America (n=3) had the least number of studies. The majority of solutions addressed diabetes (n=51), obesity (n=25), CVDs (n=24), HIV (n=18), mental health (n=16), health behaviors (n=16), and maternal and child's health (MCH) (n=11). Fewer solutions addressed asthma (n=7), cancer (n=5), family health planning (n=5), TB (n=3), malaria (n=2), chronic obtrusive pulmonary disease (COPD) (n=2), vision care (n=2), and dermatology (n=2). Other solutions targeted stroke, dental health, hepatitis vaccination, cold and flu, ED prescribed antibiotics, iodine deficiency, and liver transplantation (n=1 each). The remainder of solutions (n=14) did not focus on a certain disease. Most applications fell in the areas of health monitoring and surveillance (n=93) and health promotion and raising awareness (n=88). Fewer solutions addressed the areas of communication and reporting (n=11), data collection (n=6), telemedicine (n=5), emergency medical care (n=3), point of care support (n=2), and decision support (n=2). The majority of solutions used SMS messaging (n=94) or mobile apps (n=71). Fewer used IVR/phone calls (n=8), mobile website/email (n=5), videoconferencing (n=2), MMS (n=2), or video (n=1) or voice messages (n=1). Studies were mostly RCTs, with the majority suffering from small sample sizes and short study durations. Problems addressed by solutions included travel distance for reporting, self-management and disease monitoring, and treatment/medication adherence. CONCLUSIONS: SMS and app solutions are the most common forms of mHealth applications. SMS solutions are prevalent in both high and LMICs while app solutions are mostly used in high income countries. Common application areas include health promotion and raising awareness using SMS and health monitoring and surveillance using mobile apps. Remaining application areas are rarely addressed. Diabetes is the most commonly targeted medical condition, yet remains deficient in LMICs.

Physical Activity in Bariatric Surgery Patients: Does Temperament Matter?

OBJECTIVE: Our aim was to investigate if physical activity (PA) in bariatric surgery patients is related to temperament. METHODS: Preoperative (n = 70) and post-operative (n = 73) patients were categorized as being physically 'active' versus 'inactive' on the basis of objective PA monitoring. Assessment included the behavioural inhibition system (BIS)/behavioural activation system (BAS) scales, the effortful control (EC) subscale of the Adult Temperament Questionnaire-Short Form, a numeric pain rating scale and measures for depressive and eating disorder symptoms. RESULTS: 'Active' did not differ from 'inactive' patients with regard to temperament (BIS, BAS, and EC). Regressions with PA grouping as dependent variable (adjusted for age, gender, body mass index (BMI), depressive or eating disorder symptoms, or pain intensity) indicated an association between lower BMI and more PA in the preoperative and the post-operative group. In the post-operative group, in addition to lower BMI, also lower age and higher BIS reactivity contributed to more PA. Furthermore, there was a significant interaction between BMI and BIS suggesting that low BMI was only associated with more PA in post-operative patients with high BIS. DISCUSSION: The results indicate that temperament per se does not contribute to the level of PA in bariatric surgery patients. However, in post-operative patients, lower BMI was associated with a higher likelihood of being physically active particularly in patients with anxious temperament. These preliminary findings need further investigation within longitudinal studies. Copyright © 2017 John Wiley & Sons, Ltd and Eating Disorders Association.

Modelling of Operative Report Documents for Data Integration into an openEHR-Based Enterprise Data Warehouse.

In order to integrate operative report documents from two operating room management systems into a data warehouse, we investigated the application of the two-level modelling approach of openEHR to create a shared data model. Based on the systems' analyses, a template consisting of 13 archetypes has been developed. Of these 13 archetypes, 3 have been obtained from the international archetype repository of the openEHR foundation. The remaining 10 archetypes have been newly created. The template was evaluated by an application system expert and through conducting a first test mapping of real-world data from one of the systems. The evaluation showed that by using the two-level modelling approach of openEHR, we succeeded to represent an integrated and shared information model for operative report documents. More research is needed to learn about the limitations of this approach in other data integration scenarios.

Physical Activity, Decision-Making Abilities, and Eating Disturbances in Pre- and Postbariatric Surgery Patients.

BACKGROUND: Physical activity (PA) is considered to have a beneficial influence on executive functioning, including decision-making. Enhanced decision-making after bariatric surgery may strengthen patients' ability to delay gratification, helping to establish appropriate eating behavior. The objectives of this study were to (1) compare a preoperative group with a postoperative group with regard to daily PA, decision-making, and eating disturbances; and (2) investigate the relationship between these variables. METHODS: The study included 71 bariatric surgery candidates (78 % women, BMI [kg/m2] M = 46.9, SD = 6.0) and 73 postoperative patients (78 % women, BMI M = 32.0, SD = 4.1; 89 % Roux-en-Y gastric bypass, 11 % sleeve gastrectomy; months postoperative M = 8.2, SD = 3.5; total weight loss [%] M = 33.2, SD = 8.9) who completed SenseWear Pro2 activity monitoring. Decision-making was assessed using a computerized version of the Iowa Gambling Task and eating disorder psychopathology using the Eating Disorder Examination-Questionnaire. RESULTS: The number of patients who were classified as physically inactive was similarly high in the pre- and postoperative groups. No group differences emerged with regard to decision-making, but the postoperative group exhibited less eating disturbances than the preoperative group. No significant associations were found between PA, decision-making, and eating behavior. CONCLUSIONS: Patients after bariatric surgery were not more physically active than bariatric surgery candidates, which should be considered in care programs. Additionally, future research is needed to explore the possible link between PA, patients' decision-making abilities, and eating disturbances concerning dose-response questions.

Associations between sensor-based physical activity behaviour features and health-related parameters.

OBJECTIVES: Wearable actimetry devices are used increasingly in cohort and cross-sectional studies to assess physical activity (PA) behaviour objectively. Thus far, the medical relevance of distinct PA groups, as identified by using new methods of sensor data analysis, remains unclear. The objective of this research paper is to evaluate whether such PA groups differ in commonly accepted health risk parameters. METHODS: PA sensor data and corresponding outcome data of the NHANES 2005-06 study were obtained. Data pre-processing included elimination of potential outliers, data splitting and the computation of PA parameters, including a novel regularity measure. PA groups were identified using the x-Means clustering algorithm, and groups were evaluated for differences in CRP, BMI and HDL. RESULTS: Data sets of 7334 NHANES participants were analysed, and four distinct PA groups were identified. Statistically significant group differences were found for CRP and BMI (p<0.001), but not for HDL (p=0.67). CONCLUSIONS: PA groups derived from objective accelerometer mass data differ in exemplary health-related outcome parameters. The novel PA regularity measure is of particular interest and may become part of future PA assessments, especially when regarding low-intensity, short-lived PA events. Further research in pattern recognition methods and analytic algorithms for PA data from current multi-sensing devices is necessary.

A Method to Find Generic Thresholds for Identifying Relevant Physical Activity Events in Sensor Data.

The increasing use of wearable actimetry devices in cohort studies can provide a deep and objective insight in physical activity (PA) patterns. For reliable and reproducible pattern recognition, and to minimize the influence of specific device characteristics, there is a need for a generic method to identify relevant PA events in sensor data sets on the basis of comprehensive features such as PA duration and intensity. The objectives of this paper are to present a method to identify universal event detection thresholds for such parameters, and to attempt to find stable meta-clusters of PA behaviour. PA events of 5, 10, 20 and 30 min with low, medium and high intensity thresholds found in literature and intensity deciles were computed for a random sample (N = 100) of the NHANES 2005-06 accelerometer data set (N = 7457). On the basis of all combinations of the above, activity events were detected, and parameters mean duration, mean intensity and event regularity were computed. Results were clustered using x-Means clustering and visualized for 5-, 10-, 20-, and 30-min events. Stable clustering results are obtained with intensity thresholds up to the 8th decile and for event durations up to 10 min. Two stable meta-clusters were detected: 'irregularly active' (intensity at 52nd percentile) and 'regularly active' (intensity at 42nd percentile). Distinct generic thresholds could be identified and are proposed. They may prove useful for further investigations of similar actimetry data sets, minimising the influence of specific device characteristics. The results also confirm that distinct PA event patterns - including event regularity - can be identified using wearable sensor devices, especially when regarding low-intensity, short-term activities which do not correspond to current PA recommendations. Further research is necessary to evaluate actual associations between sensor-based PA parameters and health outcome. The author identified generic intensity and duration thresholds for analysing objective PA data from wearable devices. This may contribute to further analyses of PA patterns along with their relations with health outcome parameters.

The Night Eating Syndrome (NES) in Bariatric Surgery Patients.

The night eating syndrome (NES) has been included into the Diagnostic and Statistical Manual of Mental Disorders 5 as an example of an 'other-specified feeding or eating disorder'. The prevalence of NES has found to be higher in obese populations than in the general population and seems to rise with increasing body mass index. Recent studies suggest a prevalence of 2%-20% in bariatric surgery samples. Given that the core feature of this eating disorder may involve a shift in the circadian pattern of eating that disrupts sleep, and not the ingestion of objectively large amounts of food, it is a pattern that can continue after bariatric surgery. Nonetheless, symptoms of NES appear to decrease after weight loss surgery, and there is no evidence that pre-surgery NES negatively impacts weight loss following surgery. Prospective and longitudinal studies of the course of night eating symptoms are warranted using clear criteria and standardized assessment instruments.

Physical activity is unrelated to cognitive performance in pre-bariatric surgery patients.

OBJECTIVE: To investigate the relationship between physical activity (PA) and cognitive performance in extreme obesity. METHODS: Seventy-one bariatric surgery candidates (77.5% women) with a mean body mass index (BMI) of 46.9 kg/m2 (SD=6.0) and a mean age of 41.4 (SD=11.9) years completed SenseWear Pro2 activity monitoring for seven days. Cognitive functioning was assessed by a computerized test battery including tasks of executive function (Iowa Gambling Task), visuospatial short-term memory (Corsi Block Tapping Test) and verbal short-term memory (Auditory-Verbal Learning Test). Questionnaires assessing eating disturbances and depressive symptoms were administered. Somatic comorbidities were assessed by medical chart review. RESULTS: The level of PA was low with mean steps per day within wear time being 7140 (SD=3422). Most patients were categorized as sedentary (31.0%) or low active (26.8%). No significant association between PA estimates and cognitive performance was found. Lower PA was modestly correlated with higher BMI but not with age, somatic comorbidity or depressive symptoms. Moderated regression analyses suggested a significant interaction effect between depression and PA in predicting performance on the Corsi Block Tapping Test. Patients with (29.6%) and without (70.4%) regular binge eating did not differ with respect to PA or cognitive function. CONCLUSION: The findings indicate no association between daily PA and cognitive performance in morbidly obese patients. Future studies should explore the relationship between the variables with regard to dose-response-questions, a broader BMI range and with respect to potential changes after substantial weight loss due to bariatric surgery.

Clinical evaluation of a mobile sensor-based gait analysis method for outcome measurement after knee arthroplasty.

Clinical scores and motion-capturing gait analysis are today's gold standard for outcome measurement after knee arthroplasty, although they are criticized for bias and their ability to reflect patients' actual quality of life has been questioned. In this context, mobile gait analysis systems have been introduced to overcome some of these limitations. This study used a previously developed mobile gait analysis system comprising three inertial sensor units to evaluate daily activities and sports. The sensors were taped to the lumbosacral junction and the thigh and shank of the affected limb. The annotated raw data was evaluated using our validated proprietary software. Six patients undergoing knee arthroplasty were examined the day before and 12 months after surgery. All patients reported a satisfactory outcome, although four patients still had limitations in their desired activities. In this context, feasible running speed demonstrated a good correlation with reported impairments in sports-related activities. Notably, knee flexion angle while descending stairs and the ability to stop abruptly when running exhibited good correlation with the clinical stability and proprioception of the knee. Moreover, fatigue effects were displayed in some patients. The introduced system appears to be suitable for outcome measurement after knee arthroplasty and has the potential to overcome some of the limitations of stationary gait labs while gathering additional meaningful parameters regarding the force limits of the knee.

Mining geriatric assessment data for in-patient fall prediction models and high-risk subgroups.

BACKGROUND: Hospital in-patient falls constitute a prominent problem in terms of costs and consequences. Geriatric institutions are most often affected, and common screening tools cannot predict in-patient falls consistently. Our objectives are to derive comprehensible fall risk classification models from a large data set of geriatric in-patients' assessment data and to evaluate their predictive performance (aim#1), and to identify high-risk subgroups from the data (aim#2). METHODS: A data set of n = 5,176 single in-patient episodes covering 1.5 years of admissions to a geriatric hospital were extracted from the hospital's data base and matched with fall incident reports (n = 493). A classification tree model was induced using the C4.5 algorithm as well as a logistic regression model, and their predictive performance was evaluated. Furthermore, high-risk subgroups were identified from extracted classification rules with a support of more than 100 instances. RESULTS: The classification tree model showed an overall classification accuracy of 66%, with a sensitivity of 55.4%, a specificity of 67.1%, positive and negative predictive values of 15% resp. 93.5%. Five high-risk groups were identified, defined by high age, low Barthel index, cognitive impairment, multi-medication and co-morbidity. CONCLUSIONS: Our results show that a little more than half of the fallers may be identified correctly by our model, but the positive predictive value is too low to be applicable. Non-fallers, on the other hand, may be sorted out with the model quite well. The high-risk subgroups and the risk factors identified (age, low ADL score, cognitive impairment, institutionalization, polypharmacy and co-morbidity) reflect domain knowledge and may be used to screen certain subgroups of patients with a high risk of falling. Classification models derived from a large data set using data mining methods can compete with current dedicated fall risk screening tools, yet lack diagnostic precision. High-risk subgroups may be identified automatically from existing geriatric assessment data, especially when combined with domain knowledge in a hybrid classification model. Further work is necessary to validate our approach in a controlled prospective setting.