Time series analysis as input for clinical predictive modeling: modeling cardiac arrest in a pediatric ICU

Development of machine learning models predicting mortality using routinely collected observational health data from 0-59 months old children admitted to an intensive care unit in Bangladesh: critical role of biochemistry and haematology data

INTRODUCTION

Treatment in the intensive care unit (ICU) generates complex data where machine learning (ML) modelling could be beneficial. Using routine hospital data, we evaluated the ability of multiple ML models to predict inpatient mortality in a paediatric population in a low/middle-income country.

METHOD

We retrospectively analysed hospital record data from 0-59 months old children admitted to the ICU of Dhaka hospital of International Centre for Diarrhoeal Disease Research, Bangladesh. Five commonly used ML models- logistic regression, least absolute shrinkage and selection operator, elastic net, gradient boosting trees (GBT) and random forest (RF), were evaluated using the area under the receiver operating characteristic curve (AUROC). Top predictors were selected using RF mean decrease Gini scores as the feature importance values.

RESULTS

Data from 5669 children was used and was reduced to 3505 patients (10% death, 90% survived) following missing data removal. The mean patient age was 10.8 months (SD=10.5). The top performing models based on the validation performance measured by mean 10-fold cross-validation AUROC on the training data set were RF and GBT. Hyperparameters were selected using cross-validation and then tested in an unseen test set. The models developed used demographic, anthropometric, clinical, biochemistry and haematological data for mortality prediction. We found RF consistently outperformed GBT and predicted the mortality with AUROC of ≥0.87 in the test set when three or more laboratory measurements were included. However, after the inclusion of a fourth laboratory measurement, very minor predictive gains (AUROC 0.87 vs 0.88) resulted. The best predictors were the biochemistry and haematological measurements, with the top predictors being total CO2, potassium, creatinine and total calcium.

CONCLUSIONS

Mortality in children admitted to ICU can be predicted with high accuracy using RF ML models in a real-life data set using multiple laboratory measurements with the most important features primarily coming from patient biochemistry and haematology.

Forecasting ICU Census by Combining Time Series and Survival Models

Capacity planning of ICUs is essential for effective management of health safety, quality of patient care, and the allocation of ICU resources. Whereas ICU length of stay (LOS) may be estimated using patient information such as severity of illness scoring systems, ICU census is impacted by both patient LOS and arrival patterns. We set out to develop and evaluate an ICU census forecasting algorithm using the Multiple Organ Dysfunction Score (MODS) and the Nine Equivalents of Nursing Manpower Use Score (NEMS) for capacity planning purposes.

DESIGN

Retrospective observational study.

SETTING

We developed the algorithm using data from the Medical-Surgical ICU (MSICU) at University Hospital, London, Canada and validated using data from the Critical Care Trauma Centre (CCTC) at Victoria Hospital, London, Canada.

PATIENTS

Adult patient admissions (7,434) to the MSICU and (9,075) to the CCTC from 2015 to 2021.

INTERVENTIONS

None.

MEASUREMENTS AND MAIN RESULTS

We developed an Autoregressive integrated moving average time series model that forecasts patients arriving in the ICU and a survival model using MODS, NEMS, and other factors to estimate patient LOS. The models were combined to create an algorithm that forecasts ICU census for planning horizons ranging from 1 to 7 days. We evaluated the algorithm quality using several fit metrics. The root mean squared error ranged from 2.055 to 2.890 beds/d and the mean absolute percentage error from 9.4% to 13.2%. We show that this forecasting algorithm provides a better fit when compared with a moving average or a time series model that directly forecasts ICU census. Additionally, we evaluated the performance of the algorithm using data during the global COVID-19 pandemic and found that the error of the forecasts increased proportionally with the number of COVID-19 patients in the ICU.

CONCLUSIONS

It is possible to develop accurate tools to forecast ICU census. This type of algorithm may be important to clinicians and managers when planning ICU capacity as well as staffing and surgical demand planning over a short time horizon.

Binned Data Provide Better Imputation of Missing Time Series Data from Wearables

The presence of missing values in a time-series dataset is a very common and well-known problem. Various statistical and machine learning methods have been developed to overcome this problem, with the aim of filling in the missing values in the data. However, the performances of these methods vary widely, showing a high dependence on the type of data and correlations within the data. In our study, we performed some of the well-known imputation methods, such as expectation maximization, k-nearest neighbor, iterative imputer, random forest, and simple imputer, to impute missing data obtained from smart, wearable health trackers. In this manuscript, we proposed the use of data binning for imputation. We showed that the use of data binned around the missing time interval provides a better imputation than the use of a whole dataset. Imputation was performed for 15 min and 1 h of continuous missing data. We used a dataset with different bin sizes, such as 15 min, 30 min, 45 min, and 1 h, and we carried out evaluations using root mean square error (RMSE) values. We observed that the expectation maximization algorithm worked best for the use of binned data. This was followed by the simple imputer, iterative imputer, and k-nearest neighbor, whereas the random forest method had no effect on data binning during imputation. Moreover, the smallest bin sizes of 15 min and 1 h were observed to provide the lowest RMSE values for the majority of the time frames during the imputation of 15 min and 1 h of missing data, respectively. Although applicable to digital health data, we think that this method will also find applicability in other domains.

Cardiac Arrest in Children Following Cardiac Surgery: A Scoping Review of Contributing Factors

Nearly half of children experiencing cardiac arrest following cardiac surgery do not survive hospital discharge and patients who survive often experience significant neurological impairment. Additionally, increased resource utilization following cardiac arrest translates into adverse logistical and financial consequences. Although some studies have identified patient characteristics that increase the risk of cardiac arrest after pediatric cardiac surgery, modifiable risk factors, which could provide a foundation for effective prevention strategies, have been elusive. This scoping review explores the current knowledge surrounding risk factors associated with cardiac arrest in children following cardiac surgery and provides recommendations for future research.

A Review of Efficient Real-Time Decision Making in the Internet of Things

Emerging applications of IoT (the Internet of Things), such as smart transportation, health, and energy, are envisioned to greatly enhance the societal infrastructure and quality of life of individuals. In such innovative IoT applications, cost-efficient real-time decision-making is critical to facilitate, for example, effective transportation management and healthcare. In this paper, we formally define real-time decision tasks in IoT, review cutting-edge approaches that aim to efficiently schedule real-time decision tasks to meet their timing and data freshness constraints, review state-of-the-art approaches for efficient sensor data analytics in IoT, and discuss future research directions.

Artificial Intelligence to Improve Health Outcomes in the NICU and PICU: A Systematic Review

CONTEXT

Artificial intelligence (AI) technologies are increasingly used in pediatrics and have the potential to help inpatient physicians provide high-quality care for critically ill children.

OBJECTIVE

We aimed to describe the use of AI to improve any health outcome(s) in neonatal and pediatric intensive care.

DATA SOURCE

PubMed, IEEE Xplore, Cochrane, and Web of Science databases.

STUDY SELECTION

We used peer-reviewed studies published between June 1, 2010, and May 31, 2020, in which researchers described (1) AI, (2) pediatrics, and (3) intensive care. Studies were included if researchers assessed AI use to improve at least 1 health outcome (eg, mortality).

DATA EXTRACTION

Data extraction was conducted independently by 2 researchers. Articles were categorized by direct or indirect impact of AI, defined by the European Institute of Innovation and Technology Health joint report.

RESULTS

Of the 287 publications screened, 32 met inclusion criteria. Approximately 22% (n = 7) of studies revealed a direct impact and improvement in health outcomes after AI implementation. Majority were in prototype testing, and few were deployed into an ICU setting. Among the remaining 78% (n = 25) AI models outperformed standard clinical modalities and may have indirectly influenced patient outcomes. Quantitative assessment of health outcomes using statistical measures, such as area under the receiver operating curve (56%; n = 18) and specificity (38%; n = 12), revealed marked heterogeneity in metrics and standardization.

CONCLUSIONS

Few studies have revealed that AI has directly improved health outcomes for pediatric critical care patients. Further prospective, experimental studies are needed to assess AI's impact by using established implementation frameworks, standardized metrics, and validated outcome measures.

A predictive framework in healthcare: Case study on cardiac arrest prediction

Data-driven healthcare uses predictive analytics to enhance decision-making and personalized healthcare. Developing prognostic models is one of the applications of predictive analytics in medical environments. Various studies have used machine learning techniques for this purpose. However, there is no specific standard for choosing prediction models for different medical purposes. In this paper, the ISAF framework was proposed for choosing appropriate prediction models with regard to the properties of the classification methods. As one of the case study applications, a prognostic model for predicting cardiac arrests in sepsis patients was developed step by step through the ISAF framework. Finally, a new modified stacking model produced the best results. We predict 85 % of heart arrest cases one hour before the incidence (sensitivity> = 0.85) and 73 % of arrest cases 25 h before the occurrence (sensitivity> = 0.73). The results indicated that the proposed prognostic model has significantly improved the prediction results compared to the two standard systems of APACHE II and MEWS. Furthermore, compared to previous research, the proposed model has extended the prediction interval and improved the performance criteria.

Wireless monitoring and real-time adaptive predictive indicator of deterioration

To assist in the early warning of deterioration in hospitalised children we studied the feasibility of collecting continuous wireless physiological data using Lifetouch (ECG-derived heart and respiratory rate) and WristOx2 (pulse-oximetry and derived pulse rate) sensors. We compared our bedside paediatric early warning (PEW) score and a machine learning automated approach: a Real-time Adaptive Predictive Indicator of Deterioration (RAPID) to identify children experiencing significant clinical deterioration. 982 patients contributed 7,073,486 min during 1,263 monitoring sessions. The proportion of intended monitoring time was 93% for Lifetouch and 55% for WristOx2. Valid clinical data was 63% of intended monitoring time for Lifetouch and 50% WristOx2. 29 patients experienced 36 clinically significant deteriorations. The RAPID Index detected significant deterioration more frequently (77% to 97%) and earlier than the PEW score ≥ 9/26. High sensitivity and negative predictive value for the RAPID Index was associated with low specificity and low positive predictive value. We conclude that it is feasible to collect clinically valid physiological data wirelessly for 50% of intended monitoring time. The RAPID Index identified more deterioration, before the PEW score, but has a low specificity. By using the RAPID Index with a PEW system some life-threatening events may be averted.

Harnessing repeated measurements of predictor variables for clinical risk prediction: a review of existing methods

BACKGROUND

Clinical prediction models (CPMs) predict the risk of health outcomes for individual patients. The majority of existing CPMs only harness cross-sectional patient information. Incorporating repeated measurements, such as those stored in electronic health records, into CPMs may provide an opportunity to enhance their performance. However, the number and complexity of methodological approaches available could make it difficult for researchers to explore this opportunity. Our objective was to review the literature and summarise existing approaches for harnessing repeated measurements of predictor variables in CPMs, primarily to make this field more accessible for applied researchers.

METHODS

MEDLINE, Embase and Web of Science were searched for articles reporting the development of a multivariable CPM for individual-level prediction of future binary or time-to-event outcomes and modelling repeated measurements of at least one predictor. Information was extracted on the following: the methodology used, its specific aim, reported advantages and limitations, and software available to apply the method.

RESULTS

The search revealed 217 relevant articles. Seven methodological frameworks were identified: time-dependent covariate modelling, generalised estimating equations, landmark analysis, two-stage modelling, joint-modelling, trajectory classification and machine learning. Each of these frameworks satisfies at least one of three aims: to better represent the predictor-outcome relationship over time, to infer a covariate value at a pre-specified time and to account for the effect of covariate change.

CONCLUSIONS

The applicability of identified methods depends on the motivation for including longitudinal information and the method's compatibility with the clinical context and available patient data, for both model development and risk estimation in practice.

Early identification of impending cardiac arrest in neonates and infants in the cardiovascular ICU: a statistical modelling approach using physiologic monitoring data

OBJECTIVE

To develop a physiological data-driven model for early identification of impending cardiac arrest in neonates and infants with cardiac disease hospitalised in the cardiovascular ICU.

METHODS

We performed a single-institution retrospective cohort study (11 January 2013-16 September 2015) of patients ≤1 year old with cardiac disease who were hospitalised in the cardiovascular ICU at a tertiary care children's hospital. Demographics and diagnostic codes of cardiac arrest were obtained via the electronic health record. Diagnosis of cardiac arrest was validated by expert clinician review. Minute-to-minute physiological monitoring data were recorded via bedside monitors. A generalized linear model was used to compute a minute by minute risk score. Training and test data sets both included data from patients who did and did not develop cardiac arrest. An optimal risk-score threshold was derived based on the model's discriminatory capacity for impending arrest versus non-arrest. Model performance measures included sensitivity, specificity, accuracy, likelihood ratios, and post-test probability of arrest.

RESULTS

The final model consisting of multiple clinical parameters was able to identify impending cardiac arrest at least 2 hours prior to the event with an overall accuracy of 75% (sensitivity = 61%, specificity = 80%) and observed an increase in probability of detection of cardiac arrest from a pre-test probability of 9.6% to a post-test probability of 21.2%.

CONCLUSIONS

Our findings demonstrate that a predictive model using physiologic monitoring data in neonates and infants with cardiac disease hospitalised in the paediatric cardiovascular ICU can identify impending cardiac arrest on average 17 hours prior to arrest.

Electroencephalographic patterns preceding cardiac arrest in neonates following cardiac surgery

AIM

To identify EEG changes that could predict impending cardiac arrest (CA) in neonates with congenital heart disease undergoing postoperative continuous EEG monitoring.

METHODS

Single-center observational study of neonates who underwent cardiac surgery and had CA postoperatively while undergoing EEG monitoring from 2012-2018. Clinical data were extracted from the medical record. EEG backgrounds were evaluated at defined time-points using standardized terminology.

RESULTS

We assessed 22 neonates. The median gestational age was 38.7 weeks (IQR 37.6, 39), the median age at surgery was 5 days (IQR 2, 8), 12 patients (55%) underwent repair for hypoplastic left heart syndrome, and the median time from cardiac intensive care unit arrival postoperatively to CA was 9.5 h (IQR 7, 23). The initial EEG background was abnormal in 15 (68%). All 22 neonates (100%) had worsening of the EEG background prior to initiation of chest compressions for CA at a median of 3 min (IQR 1.5, 3). Eighteen neonates (82%) had an EEG change more than 1 min prior to chest compressions. The EEG backgrounds immediately prior to CA were continuous low voltage in 1 (5%), excessive discontinuity in 8 (36%), burst-suppression in 2 (9%), and low voltage suppression in 11 (50%).

CONCLUSION

EEG background was abnormal in 68% of neonates at EEG monitoring onset and worsened in all minutes before CA. EEG background changes may be an early sign of impending CA and indicative of developing cerebral dysfunction. Further study is needed to determine whether rapid identification of EEG changes could drive implementation of interventions to prevent CA.

A deep learning model for real-time mortality prediction in critically ill children

Background

The rapid development in big data analytics and the data-rich environment of intensive care units together provide unprecedented opportunities for medical breakthroughs in the field of critical care. We developed and validated a machine learning-based model, the Pediatric Risk of Mortality Prediction Tool (PROMPT), for real-time prediction of all-cause mortality in pediatric intensive care units.

Methods

Utilizing two separate retrospective observational cohorts, we conducted model development and validation using a machine learning algorithm with a convolutional neural network. The development cohort comprised 1445 pediatric patients with 1977 medical encounters admitted to intensive care units from January 2011 to December 2017 at Severance Hospital (Seoul, Korea). The validation cohort included 278 patients with 364 medical encounters admitted to the pediatric intensive care unit from January 2016 to November 2017 at Samsung Medical Center.

Results

Using seven vital signs, along with patient age and body weight on intensive care unit admission, PROMPT achieved an area under the receiver operating characteristic curve in the range of 0.89–0.97 for mortality prediction 6 to 60 h prior to death. Our results demonstrated that PROMPT provided high sensitivity with specificity and outperformed the conventional severity scoring system, the Pediatric Index of Mortality, in predictive ability. Model performance was indistinguishable between the development and validation cohorts.

Conclusions

PROMPT is a deep model-based, data-driven early warning score tool that can predict mortality in critically ill children and may be useful for the timely identification of deteriorating patients.

Electronic supplementary material

The online version of this article (10.1186/s13054-019-2561-z) contains supplementary material, which is available to authorized users.

Analysis and prediction of unplanned intensive care unit readmission using recurrent neural networks with long short-term memory

BACKGROUND

Unplanned readmission of a hospitalized patient is an indicator of patients' exposure to risk and an avoidable waste of medical resources. In addition to hospital readmission, intensive care unit (ICU) readmission brings further financial risk, along with morbidity and mortality risks. Identification of high-risk patients who are likely to be readmitted can provide significant benefits for both patients and medical providers. The emergence of machine learning solutions to detect hidden patterns in complex, multi-dimensional datasets provides unparalleled opportunities for developing an efficient discharge decision-making support system for physicians and ICU specialists.

METHODS AND FINDINGS

We used supervised machine learning approaches for ICU readmission prediction. We used machine learning methods on comprehensive, longitudinal clinical data from the MIMIC-III to predict the ICU readmission of patients within 30 days of their discharge. We incorporate multiple types of features including chart events, demographic, and ICD-9 embeddings. We have utilized recent machine learning techniques such as Recurrent Neural Networks (RNN) with Long Short-Term Memory (LSTM), by this we have been able to incorporate the multivariate features of EHRs and capture sudden fluctuations in chart event features (e.g. glucose and heart rate). We show that our LSTM-based solution can better capture high volatility and unstable status in ICU patients, an important factor in ICU readmission. Our machine learning models identify ICU readmissions at a higher sensitivity rate of 0.742 (95% CI, 0.718-0.766) and an improved Area Under the Curve of 0.791 (95% CI, 0.782-0.800) compared with traditional methods. We perform in-depth deep learning performance analysis, as well as the analysis of each feature contribution to the predictive model.

CONCLUSION

Our manuscript highlights the ability of machine learning models to improve our ICU decision-making accuracy and is a real-world example of precision medicine in hospitals. These data-driven solutions hold the potential for substantial clinical impact by augmenting clinical decision-making for physicians and ICU specialists. We anticipate that machine learning models will improve patient counseling, hospital administration, allocation of healthcare resources and ultimately individualized clinical care.

Learning from different perspectives: Robust cardiac arrest prediction via temporal transfer learning

Predicting and preventing cardiac arrest is one of the biggest challenges of contemporary cardiology, as a patients survival depends on the effectiveness of the emergency response teams. While black-box models have shown to have better predictive accuracies for cardiac risk stratification, early warning scoring systems are more prominent in the hospital setting due to their ease of implementation and interpretability. We propose a temporal transfer learning approach to utilize information from adjacent time points to yield an early cardiac arrest prediction model that is robust in predictive accuracies as well as maintains the interpretability of the model coefficients. Our model estimates the logistic regression coefficients simultaneously for various time points to share knowledge from different observation windows. This framework can overcome small sample size issues, and result in robust estimation of the model coefficients. We find that our model consistently outperforms a logistic regression model fit only on vital sign data from a single time slice for 763 intensive care unit patients. Moreover, we find that the estimated coefficients from our model captures temporal trends in the data.

Establishing and Sustaining an ECPR Program

The use of extracorporeal support after failed return of a spontaneous ciruculation during cardiopulmonary resuscitation (ECPR) is well described. There are 4 distinct phases for resuscitation with ECPR and the time spent in each phase is critical for successful outcome. Recommendations for ECPR previously published by the American Heart Association provide the context for implementing a consistent and well-rehearsed system for ECPR, by people with the knowledge, experience and resources to deploy ECPR in the most optimal time frame possible in selected patient populations. In this manuscript we review the current status of ECPR for acute cardiac failure and the components we believe are necessary to develop and sustain a reliable and resilient program.

A window-based time series feature extraction method

This study proposes a robust similarity score-based time series feature extraction method that is termed as Window-based Time series Feature ExtraCtion (WTC). Specifically, WTC generates domain-interpretable results and involves significantly low computational complexity thereby rendering itself useful for densely sampled and populated time series datasets. In this study, WTC is applied to a proprietary action potential (AP) time series dataset on human cardiomyocytes and three precordial leads from a publicly available electrocardiogram (ECG) dataset. This is followed by comparing WTC in terms of predictive accuracy and computational complexity with shapelet transform and fast shapelet transform (which constitutes an accelerated variant of the shapelet transform). The results indicate that WTC achieves a slightly higher classification performance with significantly lower execution time when compared to its shapelet-based alternatives. With respect to its interpretable features, WTC has a potential to enable medical experts to explore definitive common trends in novel datasets.

Technical challenges related to implementation of a formula one real time data acquisition and analysis system in a paediatric intensive care unit

Most existing, expert monitoring systems do not provide the real time continuous analysis of the monitored physiological data that is necessary to detect transient or combined vital sign indicators nor do they provide long term storage of the data for retrospective analyses. In this paper we examine the feasibility of implementing a long term data storage system which has the ability to incorporate real-time data analytics, the system design, report the main technical issues encountered, the solutions implemented and the statistics of the data recorded. McLaren Electronic Systems expertise used to continually monitor and analyse the data from F1 racing cars in real time was utilised to implement a similar real-time data recording platform system adapted with real time analytics to suit the requirements of the intensive care environment. We encountered many technical (hardware and software) implementation challenges. However there were many advantages of the system once it was operational. They include: (1) The ability to store the data for long periods of time enabling access to historical physiological data. (2) The ability to alter the time axis to contract or expand periods of interest. (3) The ability to store and review ECG morphology retrospectively. (4) Detailed post event (cardiac/respiratory arrest or other clinically significant deteriorations in patients) data can be reviewed clinically as opposed to trend data providing valuable clinical insight. Informed mortality and morbidity reviews can be conducted. (5) Storage of waveform data capture to use for algorithm development for adaptive early warning systems. Recording data from bed-side monitors in intensive care/wards is feasible. It is possible to set up real time data recording and long term storage systems. These systems in future can be improved with additional patient specific metrics which predict the status of a patient thus paving the way for real time predictive monitoring.

The initiation of a multidisciplinary bladder cancer clinic and the uptake of neoadjuvant chemotherapy: A time-series analysis

INTRODUCTION

While level 1 evidence supports the use of neoadjuvant chemotherapy (NAC) for patients with muscle-invasive bladder cancer (MIBC), its uptake has been underwhelming, even in academic centres. Our aim was to determine if the initiation of a multidisciplinary bladder cancer clinic (MDBCC) in 2008 at our institution, where patients are assessed simultaneously by bladder cancer-focused urologists and radiation oncologists with easy access to a medical oncologist, was associated with an increased use of NAC.

METHODS

Patients with MIBC initiating treatment between July 2000 and June 2013 were identified and classified by academic year (July 1 to June 30). Time-series analyses using interventional autoregressive integrated moving average (ARIMA) models with ramp intervention functions were then conducted. A sensitivity analysis was performed on clinical N0 patients.

RESULTS

The cohort included 278 patients: 168 from 2000-2007 and 110 from 2008-2012 (academic years). Forty-two (15.1%) patients received NAC and 74 (26.6%) received adjuvant chemotherapy (AC). Overall the proportion of patients receiving NAC increased from 7.7% before the MDBCC to 47.6% in 2012 (Interventional ARIMA p=0.036). The results were similar when restricting to cN0 patients (p<0.001). NAC use gradually increased over time regardless of MDBCC attendance, although the proportion of patients receiving NAC appears to have risen more sharply among MDBCC attendees.

CONCLUSIONS

At our institution, the initiation of the MDBCC was temporally associated with increased use of NAC. In addition to multidisciplinary collaboration, having a critical mass of NAC physician advocates and support from institutional leaders are essential to the uptake of NAC.

Using Time Series Analysis to Predict Cardiac Arrest in a PICU

OBJECTIVES

To build and test cardiac arrest prediction models in a PICU, using time series analysis as input, and to measure changes in prediction accuracy attributable to different classes of time series data.

DESIGN

Retrospective cohort study.

SETTING

Thirty-one bed academic PICU that provides care for medical and general surgical (not congenital heart surgery) patients.

SUBJECTS

Patients experiencing a cardiac arrest in the PICU and requiring external cardiac massage for at least 2 minutes.

INTERVENTIONS

None.

MEASUREMENTS AND MAIN RESULTS

One hundred three cases of cardiac arrest and 109 control cases were used to prepare a baseline dataset that consisted of 1,025 variables in four data classes: multivariate, raw time series, clinical calculations, and time series trend analysis. We trained 20 arrest prediction models using a matrix of five feature sets (combinations of data classes) with four modeling algorithms: linear regression, decision tree, neural network, and support vector machine. The reference model (multivariate data with regression algorithm) had an accuracy of 78% and 87% area under the receiver operating characteristic curve. The best model (multivariate + trend analysis data with support vector machine algorithm) had an accuracy of 94% and 98% area under the receiver operating characteristic curve.

CONCLUSIONS

Cardiac arrest predictions based on a traditional model built with multivariate data and a regression algorithm misclassified cases 3.7 times more frequently than predictions that included time series trend analysis and built with a support vector machine algorithm. Although the final model lacks the specificity necessary for clinical application, we have demonstrated how information from time series data can be used to increase the accuracy of clinical prediction models.

Time-series analysis in the medical domain: a study of Tacrolimus administration and influence on kidney graft function

There exists a major concern regarding toxic effects of immunosuppressive medication on the kidney graft during post-transplant care, with observed variation in individual susceptibility to adverse drug effects amongst patients. To date, there has been no possibility to identify susceptible patients prospectively. This study analyzes medical data which includes time series of measures of renal function and trough levels of immunosuppressive drug Tacrolimus, with the main aim of identifying patients susceptible to drug toxicity. We evaluate a plethora of time-series distance measures, determining their appropriateness to the domain based on two criteria: (1) preserving the expected correlations between distances, and (2) ability to detect the expected patterns of interaction between immunosuppressive drug levels and renal function. Besides identifying the most suitable time-series distance measures, we observed that the majority of patients do not exhibit an association between impaired graft function and higher Tacrolimus dosing. On the other hand, the minority of patients determined most sensitive to varying Tacrolimus levels showed a strong tendency to prefer low Tacrolimus dosing.

Septic Shock Prediction for Patients with Missing Data

Sepsis and septic shock are common and potentially fatal conditions that often occur in intensive care unit (ICU) patients. Early prediction of patients at risk for septic shock is therefore crucial to minimizing the effects of these complications. Potential indications for septic shock risk span a wide range of measurements, including physiological data gathered at different temporal resolutions and gene expression levels, leading to a nontrivial prediction problem. Previous works on septic shock prediction have used small, carefully curated datasets or clinical measurements that may not be available for many ICU patients. The recent availability of a large, rich ICU dataset called MIMIC-II has provided the opportunity for more extensive modeling of this problem. However, such a large clinical dataset inevitably contains a substantial amount of missing data. We investigate how different imputation selection criteria and methods can overcome the missing data problem. Our results show that imputation methods in conjunction with predictive modeling can lead to accurate septic shock prediction, even if the features are restricted primarily to noninvasive measurements. Our models provide a generalized approach for predicting septic shock in any ICU patient.

Application of seemingly unrelated regression in medical data with intermittently observed time-dependent covariates

Background. In many studies with longitudinal data, time-dependent covariates can only be measured intermittently (not at all observation times), and this presents difficulties for standard statistical analyses. This situation is common in medical studies, and methods that deal with this challenge would be useful. Methods. In this study, we performed the seemingly unrelated regression (SUR) based models, with respect to each observation time in longitudinal data with intermittently observed time-dependent covariates and further compared these models with mixed-effect regression models (MRMs) under three classic imputation procedures. Simulation studies were performed to compare the sample size properties of the estimated coefficients for different modeling choices. Results. In general, the proposed models in the presence of intermittently observed time-dependent covariates showed a good performance. However, when we considered only the observed values of the covariate without any imputations, the resulted biases were greater. The performances of the proposed SUR-based models in comparison with MRM using classic imputation methods were nearly similar with approximately equal amounts of bias and MSE. Conclusion. The simulation study suggests that the SUR-based models work as efficiently as MRM in the case of intermittently observed time-dependent covariates. Thus, it can be used as an alternative to MRM.

Multivariate temporal symptomatic characterization of cardiac arrest

We model the temporal symptomatic characteristics of 171 cardiac arrest patients in Intensive Care Units. The temporal and feature dependencies in the data are illustrated using a mixture of matrix normal distributions. We found that the cardiac arrest temporal signature is best summarized with six hours data prior to cardiac arrest events, and its statistical descriptions are significantly different from the measurements taken in the past two days. This matrix normal model can classify these patterns better than logistic regressions with lagged features.

A molecular diffusion based utility model for Drosophila larval phototaxis

Background

Generally, utility based decision making models focus on experimental outcomes. In this paper we propose a utility model based on molecular diffusion to simulate the choice behavior of Drosophila larvae exposed to different light conditions.

Methods

In this paper, light/dark choice-based Drosophila larval phototaxis is analyzed with our molecular diffusion based model. An ISCEM algorithm is developed to estimate the model parameters.

Results

By applying this behavioral utility model to light intensity and phototaxis data, we show that this model fits the experimental data very well.

Conclusions

Our model provides new insights into decision making mechanisms in general. From an engineering viewpoint, we propose that the model could be applied to a wider range of decision making practices.