Process-oriented Iterative Multiple Alignment for Medical Process Mining

Adapted from biological sequence alignment, trace alignment is a process mining technique used to visualize and analyze workflow data. Any analysis done with this method, however, is affected by the alignment quality. The best existing trace alignment techniques use progressive guide-trees to heuristically approximate the optimal alignment in O(N2L2) time. These algorithms are heavily dependent on the selected guide-tree metric, often return sum-of-pairs-score-reducing errors that interfere with interpretation, and are computationally intensive for large datasets. To alleviate these issues, we propose process-oriented iterative multiple alignment (PIMA), which contains specialized optimizations to better handle workflow data. We demonstrate that PIMA is a flexible framework capable of achieving better sum-of-pairs score than existing trace alignment algorithms in only O(NL2) time. We applied PIMA to analyzing medical workflow data, showing how iterative alignment can better represent the data and facilitate the extraction of insights from data visualization.

Management Issues in Critically Ill Pediatric Patients with Trauma

The management of critically ill pediatric patients with trauma poses many challenges because of the infrequency and diversity of severe injuries and a paucity of high-level evidence to guide care for these uncommon events. This article discusses recent recommendations for early resuscitation and blood component therapy for hypovolemic pediatric patients with trauma. It also highlights the specific types of injuries that lead to severe injury in children and presents challenges related to their management.

Region-based Activity Recognition Using Conditional GAN

We present a method for activity recognition that first estimates the activity performer's location and uses it with input data for activity recognition. Existing approaches directly take video frames or entire video for feature extraction and recognition, and treat the classifier as a black box. Our method first locates the activities in each input video frame by generating an activity mask using a conditional generative adversarial network (cGAN). The generated mask is appended to color channels of input images and fed into a VGG-LSTM network for activity recognition. To test our system, we produced two datasets with manually created masks, one containing Olympic sports activities and the other containing trauma resuscitation activities. Our system makes activity prediction for each video frame and achieves performance comparable to the state-of-the-art systems while simultaneously outlining the location of the activity. We show how the generated masks facilitate the learning of features that are representative of the activity rather than accidental surrounding information.

Progress Estimation and Phase Detection for Sequential Processes

Process modeling and understanding are fundamental for advanced human-computer interfaces and automation systems. Most recent research has focused on activity recognition, but little has been done on sensor-based detection of process progress. We introduce a real-time, sensor-based system for modeling, recognizing and estimating the progress of a work process. We implemented a multimodal deep learning structure to extract the relevant spatio-temporal features from multiple sensory inputs and used a novel deep regression structure for overall completeness estimation. Using process completeness estimation with a Gaussian mixture model, our system can predict the phase for sequential processes. The performance speed, calculated using completeness estimation, allows online estimation of the remaining time. To train our system, we introduced a novel rectified hyperbolic tangent (rtanh) activation function and conditional loss. Our system was tested on data obtained from the medical process (trauma resuscitation) and sports events (Olympic swimming competition). Our system outperformed the existing trauma-resuscitation phase detectors with a phase detection accuracy of over 86%, an F1-score of 0.67, a completeness estimation error of under 12.6%, and a remaining-time estimation error of less than 7.5 minutes. For the Olympic swimming dataset, our system achieved an accuracy of 88%, an F1-score of 0.58, a completeness estimation error of 6.3% and a remaining-time estimation error of 2.9 minutes.

A Data-driven Process Recommender Framework

We present an approach for improving the performance of complex knowledge-based processes by providing data-driven step-by-step recommendations. Our framework uses the associations between similar historic process performances and contextual information to determine the prototypical way of enacting the process. We introduce a novel similarity metric for grouping traces into clusters that incorporates temporal information about activity performance and handles concurrent activities. Our data-driven recommender system selects the appropriate prototype performance of the process based on user-provided context attributes. Our approach for determining the prototypes discovers the commonly performed activities and their temporal relationships. We tested our system on data from three real-world medical processes and achieved recommendation accuracy up to an F1 score of 0.77 (compared to an F1 score of 0.37 using ZeroR) with 63.2% of recommended enactments being within the first five neighbors of the actual historic enactments in a set of 87 cases. Our framework works as an interactive visual analytic tool for process mining. This work shows the feasibility of data-driven decision support system for complex knowledge-based processes.

Medical Workflow Modeling Using Alignment-Guided State-Splitting HMM

Process mining techniques have been used to discover and analyze workflows in various fields, ranging from business management to healthcare. Much of this research, however, has overlooked the potential of hidden Markov models (HMMs) for workflow discovery. We present a novel alignment-guided state-splitting HMM inference algorithm (AGSS) for discovering workflow models based on observed traces of process executions. We compared the AGSS to existing methods using four real-world medical workflow datasets and a more detailed case study on one of them. Our numerical results show that AGSS not only generates more accurate workflow models, but also better represents the underlying process. In addition, with trace alignment to guide state splitting, AGSS is significantly more efficient (by a factor of O(n)) than previous HMM inference algorithms. Our case study results show that our approach produces a more readable and accurate workflow model that existing algorithms. Comparing the discovered model to the hand-made expert model of the same process, we found three discrepancies. These three discrepancies were reconsidered by medical experts and used for enhancing the expert model.

Process Mining for Trauma Resuscitation

We present our process mining system for analyzing the trauma resuscitation process to improve medical team performance and patient outcomes. Our system has four main parts: trauma resuscitation process model discovery, process model enhancement (or repair), process deviation analysis, and process recommendation. We developed novel algorithms to address the technical challenges for each problem. We validated our system on real-world trauma resuscitation data from the Children's National Medical Center (CNMC), a level 1 trauma center. Our results show our system's capability of supporting complex medical processes. Our approaches were also implemented in an interactive visual analytic tool.

Language-Based Process Phase Detection in the Trauma Resuscitation

Process phase detection has been widely used in surgical process modeling (SPM) to track process progression. These studies mostly used video and embedded sensor data, but spoken language also provides rich semantic information directly related to process progression. We present a long-short term memory (LSTM) deep learning model to predict trauma resuscitation phases using verbal communication logs. We first use an LSTM to extract the sentence meaning representations, and then sequentially feed them into another LSTM to extract the meaning of a sentence group within a time window. This information is ultimately used for phase prediction. We used 24 manually-transcribed trauma resuscitation cases to train, and the remaining 6 cases to test our model. We achieved 79.12% accuracy, and showed performance advantages over existing visual-audio systems for critical phases of the process. In addition to language information, we evaluated a multimodal phase prediction structure that also uses audio input. We finally identified the challenges of substituting manual transcription with automatic speech recognition in trauma resuscitation.

Exploring Design Opportunities for a Context-Adaptive Medical Checklist Through Technology Probe Approach

This paper explores the workflow and use of an interactive medical checklist for trauma resuscitation-an emerging technology developed for trauma team leaders to support decision making and task coordination among team members. We used a technology probe approach and ethnographic methods, including video review, interviews, and content analysis of checklist logs, to examine how team leaders use the checklist probe during live resuscitations. We found that team leaders of various experience levels use the technology differently. Some leaders frequently glance at the checklist and take notes during task performance, while others place the checklist on a stand and only interact with the checklist when checking items. We compared checklist timestamps to task activities and found that most items are checked off after tasks are performed. We conclude by discussing design implications and new design opportunities for a future dynamic, adaptive checklist.

Poster Abstract: 3D Activity Localization With Multiple Sensors

We present a deep learning framework for fast 3D activity localization and tracking in a dynamic and crowded real world setting. Our training approach reverses the traditional activity localization approach, which first estimates the possible location of activities and then predicts their occurrence. Instead, we first trained a deep convolutional neural network for activity recognition using depth video and RFID data as input, and then used the activation maps of the network to locate the recognized activity in the 3D space. Our system achieved around 20cm average localization error (in a 4m × 5m room) which is comparable to Kinect's body skeleton tracking error (10-20cm), but our system tracks activities instead of Kinect's location of people.

Accuracy of Postresuscitation Team Debriefings in a Pediatric Emergency Department

STUDY OBJECTIVE

Guideline committees recommend postresuscitation debriefings to improve performance. "Hot" postresuscitation debriefings occur immediately after the event and rely on team recall. We assessed the ability of resuscitation teams to recall their performance in team-based, hot debriefings in a pediatric emergency department (ED), using video review as the criterion standard. We hypothesized that debriefing accuracy will improve during the course of the study.

METHODS

Resuscitation physician and nurse leaders cofacilitated debriefings after ED resuscitations involving cardiopulmonary resuscitation (CPR) or intubation. Debriefing teams recorded their self-assessments of clinical performance measures with standardized debriefing forms. The debriefing form data were compared with actual performance measured by video review at 2 pediatric EDs over 22 months. CPR performance measures included time to automated external defibrillator pad placement, epinephrine administration timing, and compression pause timing. Intubation measures included occurrences of oxygen desaturation, number of intubation attempts, and use of end-tidal carbon dioxide monitoring.

RESULTS

We analyzed 100 resuscitations (14 cardiac arrests, 22 cardiac arrests with intubation, and 64 intubations). The accuracy of debriefing answers was 87%, increasing from 83% to 91% between the first and second halves of the study period (7.7% difference; 95% confidence interval 0.2% to 15%). Debriefings that acknowledged an error in certain performance measures (ie, automated external defibrillator pad placement delay, multiple intubation attempts, and occurrence of oxygen desaturation) had significantly worse performance in those specific measures on video review.

CONCLUSION

Teams in postresuscitation debriefings had a higher degree of debriefing answer accuracy in the final 50 debriefings than in the first 50. Teams also distinguished various degrees of resuscitation performance.

Checklist as a Memory Externalization Tool during a Critical Care Process

We analyzed user interactions with a paper-based checklist in a regional trauma center to inform the design of digital cognitive aids for safety-critical medical teamwork. An initial review of paper checklists from actual trauma resuscitations revealed that trauma team leaders frequently wrote notes on the checklist. To understand this notetaking practice, we performed content analysis of 163 checklists collected over the period of four months. We found nine major categories of information that leaders recorded during resuscitations, including patient values, physical assessment findings, and pre-hospital information. An analysis of types and amount of notes written by leaders of different experience levels showed that more experienced leaders recorded more patient values and physical findings, while less experienced leaders recorded more notes about their activities and task completion status. These findings suggested that a checklist designed for a high-risk, fast-paced medical event has evolved into a dual function tool, serving both as a compliance and memory aid. Based on these findings, we derived requirements for designing digital cognitive aids to support safety-critical medical teamwork.

Online Process Phase Detection Using Multimodal Deep Learning

We present a multimodal deep-learning structure that automatically predicts phases of the trauma resuscitation process in real-time. The system first pre-processes the audio and video streams captured by a Kinect's built-in microphone array and depth sensor. A multimodal deep learning structure then extracts video and audio features, which are later combined through a "slow fusion" model. The final decision is then made from the combined features through a modified softmax classification layer. The model was trained on 20 trauma resuscitation cases (>13 hours), and was tested on 5 other cases. Our results showed over 80% online detection accuracy with 0.7 F-Score, outperforming previous systems.

Deep Learning for RFID-Based Activity Recognition

We present a system for activity recognition from passive RFID data using a deep convolutional neural network. We directly feed the RFID data into a deep convolutional neural network for activity recognition instead of selecting features and using a cascade structure that first detects object use from RFID data followed by predicting the activity. Because our system treats activity recognition as a multi-class classification problem, it is scalable for applications with large number of activity classes. We tested our system using RFID data collected in a trauma room, including 14 hours of RFID data from 16 actual trauma resuscitations. Our system outperformed existing systems developed for activity recognition and achieved similar performance with process-phase detection as systems that require wearable sensors or manually-generated input. We also analyzed the strengths and limitations of our current deep learning architecture for activity recognition from RFID data.

Deep Neural Network for RFID-Based Activity Recognition

We propose a Deep Neural Network (DNN) structure for RFID-based activity recognition. RFID data collected from several reader antennas with overlapping coverage have potential spatiotemporal relationships that can be used for object tracking. We augmented the standard fully-connected DNN structure with additional pooling layers to extract the most representative features. For model training and testing, we used RFID data from 12 tagged objects collected during 25 actual trauma resuscitations. Our results showed 76% recognition micro-accuracy for 7 resuscitation activities and 85% average micro-accuracy for 5 resuscitation phases, which is similar to existing system that, however, require the user to wear an RFID antenna.

Association Between Trauma Center Type and Mortality Among Injured Adolescent Patients

IMPORTANCE

Although data obtained from regional trauma systems demonstrate improved outcomes for children treated at pediatric trauma centers (PTCs) compared with those treated at adult trauma centers (ATCs), differences in mortality have not been consistently observed for adolescents. Because trauma is the leading cause of death and acquired disability among adolescents, it is important to better define differences in outcomes among injured adolescents by using national data.

OBJECTIVES

To use a national data set to compare mortality of injured adolescents treated at ATCs, PTCs, or mixed trauma centers (MTCs) that treat both pediatric and adult trauma patients and to determine the final discharge disposition of survivors at different center types.

DESIGN, SETTING, AND PARTICIPANTS

Data from level I and II trauma centers participating in the 2010 National Trauma Data Bank (January 1 to December 31, 2010) were used to create multilevel models accounting for center-specific effects to evaluate the association of center characteristics (PTC, ATC, or MTC) on mortality among patients aged 15 to 19 years who were treated for a blunt or penetrating injury. The models controlled for sex; mechanism of injury (blunt vs penetrating); injuries sustained, based on the Abbreviated Injury Scale scores (post-dot values <3 or ≥3 by body region); initial systolic blood pressure; and Glasgow Coma Scale scores. Missing data were managed using multiple imputation, accounting for multilevel data structure. Data analysis was conducted from January 15, 2013, to March 15, 2016.

EXPOSURES

Type of trauma center.

MAIN OUTCOMES AND MEASURES

Mortality at each center type.

RESULTS

Among 29 613 injured adolescents (mean [SD] age, 17.3 [1.4] years; 72.7% male), most were treated at ATCs (20 402 [68.9%]), with the remainder at MTCs (7572 [25.6%]) or PTCs (1639 [5.5%]). Adolescents treated at PTCs were more likely to be injured by a blunt than penetrating injury mechanism (91.4%) compared with those treated at ATCs (80.4%) or MTCs (84.6%). Mortality was higher among adolescents treated at ATCs and MTCs than those treated at PTCs (3.2% and 3.5% vs 0.4%; P < .001). The adjusted odds of mortality were higher at ATCs (odds ratio, 4.19; 95% CI, 1.30-13.51) and MTCs (odds ratio, 6.68; 95% CI, 2.03-21.99) compared with PTCs but was not different between level I and II centers (odds ratio, 0.76; 95% CI, 0.59-0.99).

CONCLUSION AND RELEVANCE

Mortality among injured adolescents was lower among those treated at PTCs, compared with those treated at ATCs and MTCs. Defining resource and patient features that account for these observed differences is needed to optimize adolescent outcomes after injury.

Activity Recognition for Medical Teamwork Based on Passive RFID

We describe a novel and practical activity recognition system for dynamic and complex medical settings using only passive RFID technology. Our activity recognition approach is based on the use of objects that are specific for a given activity. The object-use status is detected from RFID data and the activities are predicted from the statuses of use of different objects. We tagged 10 objects in a trauma room of an emergency department and recorded RFID data for 10 actual trauma resuscitations. More than 20,000 seconds of data were collected and used for analysis. The system achieved a 96% overall accuracy with a 0.74 F-score for detecting use of 10 common resuscitation objects and 95% accuracy with a 0.30 F Score for activity recognition of 10 medical activities.

Design Challenges in Converting a Paper Checklist to Digital Format for Dynamic Medical Settings

We describe a mobile digital checklist that we designed and developed for trauma resuscitation-a dynamic, fast-paced medical process of treating severely injured patients. The checklist design was informed by our analysis of user interactions with a paper checklist that was introduced to improve team performance during resuscitations. The design process followed an iterative approach and involved several medical experts. We discuss design challenges in converting a paper checklist to its digital counterpart, as well as our approaches for addressing those challenges. While we show that using a digital checklist during a fast-paced medical event is feasible, we also recognize several design constraints, including limited display size, difficulties in entering notes about the medical process and patient, and difficulties in replicating user experience with paper checklists.

Use of Payer as a Proxy for Health Insurance Status on Admission Results in Misclassification of Insurance Status among Pediatric Trauma Patients

The purpose of this study was to quantify health insurance misclassification among children treated at a pediatric trauma center and to determine factors associated with misclassification. Demographic, medical, and financial information were collected for patients at our institution between 2008 and 2010. Two health insurance variables were created: true (insurance on hospital admission) and payer (source of payment). Multivariable logistic regression was used to determine which factors were independently associated with health insurance misclassification. The two values of health insurance status were abstracted from the hospital financial database, the trauma registry, and the patient medical record. Among 3630 patients, 123 (3.4%) had incorrect health insurance designation. Misclassification was highest in patients who died: 13.9 per cent among all deaths and 30.8 per cent among emergency department deaths. The adjusted odds of misclassification were 6.7 (95% confidence interval: 1.7, 26.6) among patients who died and 16.1 (95% confidence interval: 3.2, 80.77) among patients who died in the emergency department. Using payer as a proxy for health insurance results in misclassification. Approaches are needed to accurately ascertain true health insurance status when studying the impact of insurance on treatment outcomes.

Use of Payer as a Proxy for Health Insurance Status on Admission Results in Misclassification of Insurance Status among Pediatric Trauma Patients

The purpose of this study was to quantify health insurance misclassification among children treated at a pediatric trauma center and to determine factors associated with misclassification. Demographic, medical, and financial information were collected for patients at our institution between 2008 and 2010. Two health insurance variables were created: true (insurance on hospital admission) and payer (source of payment). Multivariable logistic regression was used to determine which factors were independently associated with health insurance misclassification. The two values of health insurance status were abstracted from the hospital financial database, the trauma registry, and the patient medical record. Among 3630 patients, 123 (3.4%) had incorrect health insurance designation. Misclassification was highest in patients who died: 13.9 per cent among all deaths and 30.8 per cent among emergency department deaths. The adjusted odds of misclassification were 6.7 (95% confidence interval: 1.7, 26.6) among patients who died and 16.1 (95% confidence interval: 3.2, 80.77) among patients who died in the emergency department. Using payer as a proxy for health insurance results in misclassification. Approaches are needed to accurately ascertain true health insurance status when studying the impact of insurance on treatment outcomes.

Mortality Among Injured Children Treated at Different Trauma Center Types

IMPORTANCE

Trauma is the leading cause of death among US children. Whether pediatric trauma centers (PTCs), mixed trauma centers (MTCs), or adult trauma centers (ATCs) offer a survival benefit compared with one another when treating injured children is controversial. Ascertaining the optimal care environment will better inform quality improvement initiatives and accreditation standards.

OBJECTIVE

To evaluate the association between type of trauma center (PTC, MTC, or ATC) and in-hospital mortality among young children (5 years and younger), older children (aged 6-11 years), and adolescents (aged 12-18 years).

DESIGN, SETTING, AND PARTICIPANTS

In this retrospective cohort study, injured children aged 18 years or younger who were hospitalized in the United States from January 1, 2010, to December 31, 2013, were observed for the duration of their admission until discharge or death. We included patients with an Abbreviated Injury Score of 2 or greater in at least 1 body region. Random-intercept multilevel regression was used to evaluate the association between center type and in-hospital mortality after adjusting for confounders. Stratified analyses in young children, older children, and adolescents were performed. We conducted secondary analyses limited to patients with severe injuries (Injury Severity Score ≥25). Both analyses were performed between January 1 and August 31, 2014. Data were derived from 252 US level I and II trauma centers voluntarily participating in the American College of Surgeons adult or pediatric Trauma Quality Improvement Program.

MAIN OUTCOME AND MEASURE

In-hospital mortality.

RESULTS

We identified 175 585 injured children. Crude mortality rates were 2.3% for children treated at ATCs, 1.8% for children treated at MTCs, and 0.6% for children treated at PTCs. After adjustment, children had higher odds of dying when treated at ATCs (odds ratio [OR], 1.57; 95% CI, 1.15-2.14) and MTCs (OR, 1.45; 95% CI, 1.05-2.01) compared with those treated at PTCs. In stratified analyses, young children had higher odds of death when treated at ATCs vs PTCs (OR, 1.78; 95% CI, 1.05-3.40), but there was no association between center type and mortality among older children (OR, 1.17; 95% CI, 0.65-2.11) and adolescents (OR, 1.23; 95% CI, 0.82-1.85). Results were similar in analyses of severely injured children: those treated at ATCs (OR, 1.75; 95% CI, 1.25-2.44) and MTCs (OR, 1.62; 95% CI, 1.15-2.29) had higher odds of death when compared with those treated at PTCs.

CONCLUSIONS AND RELEVANCE

Injured children treated at ATCs and MTCs had higher in-hospital mortality compared with those treated at PTCs. This association was most evident in younger children and remained significant in severely injured children. Quality improvement initiatives geared toward ATCs and MTCs are required to provide optimal care to injured children.

Reassessing mechanism as a predictor of pediatric injury mortality

BACKGROUND

The use of mechanism of injury as a predictor of injury outcome presents practical challenges because this variable may be missing or inaccurate in many databases. The purpose of this study was to determine the importance of mechanism of injury as a predictor of mortality among injured children.

METHODS

The records of children (<15-y-old) sustaining a blunt injury were obtained from the National Trauma Data Bank. Models predicting injury mortality were developed using mechanism of injury and injury coding using either abbreviated injury scale post-dot values (low-dimensional injury coding) or injury International Classification of Diseases, Ninth Revision codes and their two-way interactions (high-dimensional injury coding). Model performance with and without inclusion of mechanism of injury was compared for both coding schemes, and the relative importance of mechanism of injury as a variable in each model type was evaluated.

RESULTS

Among 62,569 records, a mortality rate of 0.9% was observed. Inclusion of mechanism of injury improved model performance when using low-dimensional injury coding but was associated with no improvement when using high-dimensional injury coding. Mechanism of injury contributed to 28% of model variance when using low-dimensional injury coding and <1% when high-dimensional injury coding was used.

CONCLUSIONS

Although mechanism of injury may be an important predictor of injury mortality among children sustaining blunt trauma, its importance as a predictor of mortality depends on the approach used for injury coding. Mechanism of injury is not an essential predictor of outcome after injury when coding schemes are used that better characterize injuries sustained after blunt pediatric trauma.