Linking Big Data and Prediction Strategies: Tools, Pitfalls, and Lessons Learned

Critical Data for Critical Care: A Primer on Leveraging Electronic Health Record Data for Research From Society of Critical Care Medicine's Panel on Data Sharing and Harmonization

A growing body of critical care research draws on real-world data from electronic health records (EHRs). The bedside clinician has myriad data sources to aid in clinical decision-making, but the lack of data sharing and harmonization standards leaves much of this data out of reach for multi-institution critical care research. The Society of Critical Care Medicine (SCCM) Discovery Data Science Campaign convened a panel of critical care and data science experts to explore and document unique advantages and opportunities for leveraging EHR data in critical care research. This article reviews and illustrates six organizing topics (data domains and common data elements; data harmonization; data quality; data interoperability and digital infrastructure; data access, sharing, and governance; and ethics and equity) as a data science primer for critical care researchers, laying a foundation for future publications from the SCCM Discovery Data Harmonization and Sharing Guiding Principles Panel.

Chinese experts' consensus on the application of intensive care big data

The development of intensive care medicine is inseparable from the diversified monitoring data. Intensive care medicine has been closely integrated with data since its birth. Critical care research requires an integrative approach that embraces the complexity of critical illness and the computational technology and algorithms that can make it possible. Considering the need of standardization of application of big data in intensive care, Intensive Care Medicine Branch of China Health Information and Health Care Big Data Society, Standard Committee has convened expert group, secretary group and the external audit expert group to formulate Chinese Experts' Consensus on the Application of Intensive Care Big Data (2022). This consensus makes 29 recommendations on the following five parts: Concept of intensive care big data, Important scientific issues, Standards and principles of database, Methodology in solving big data problems, Clinical application and safety consideration of intensive care big data. The consensus group believes this consensus is the starting step of application big data in the field of intensive care. More explorations and big data based retrospective research should be carried out in order to enhance safety and reliability of big data based models of critical care field.

Rapid prediction of secondary neurologic decline after traumatic brain injury: a data analytic approach

Secondary neurologic decline (ND) after traumatic brain injury (TBI) is independently associated with outcome, but robust predictors of ND are lacking. In this retrospective analysis of consecutive isolated TBI admissions to the R. Adams Cowley Shock Trauma Center between November 2015 and June 2018, we aimed to develop a triage decision support tool to quantify risk for early ND. Three machine learning models based on clinical, physiologic, or combined characteristics from the first hour of hospital resuscitation were created. Among 905 TBI cases, 165 (18%) experienced one or more ND events (130 clinical, 51 neurosurgical, and 54 radiographic) within 48 h of presentation. In the prediction of ND, the clinical plus physiologic data model performed similarly to the physiologic only model, with concordance indices of 0.85 (0.824-0.877) and 0.84 (0.812-0.868), respectively. Both outperformed the clinical only model, which had a concordance index of 0.72 (0.688-0.759). This preliminary work suggests that a data-driven approach utilizing physiologic and basic clinical data from the first hour of resuscitation after TBI has the potential to serve as a decision support tool for clinicians seeking to identify patients at high or low risk for ND.

High resolution data modifies intensive care unit dialysis outcome predictions as compared with low resolution administrative data set

High resolution clinical databases from electronic health records are increasingly being used in the field of health data science. Compared to traditional administrative databases and disease registries, these newer highly granular clinical datasets offer several advantages, including availability of detailed clinical information for machine learning and the ability to adjust for potential confounders in statistical models. The purpose of this study is to compare the analysis of the same clinical research question using an administrative database and an electronic health record database. The Nationwide Inpatient Sample (NIS) was used for the low-resolution model, and the eICU Collaborative Research Database (eICU) was used for the high-resolution model. A parallel cohort of patients admitted to the intensive care unit (ICU) with sepsis and requiring mechanical ventilation was extracted from each database. The primary outcome was mortality and the exposure of interest was the use of dialysis. In the low resolution model, after controlling for the covariates that are available, dialysis use was associated with an increased mortality (eICU: OR 2.07, 95% CI 1.75-2.44, p<0.01; NIS: OR 1.40, 95% CI 1.36-1.45, p<0.01). In the high-resolution model, after the addition of the clinical covariates, the harmful effect of dialysis on mortality was no longer significant (OR 1.04, 95% 0.85-1.28, p = 0.64). The results of this experiment show that the addition of high resolution clinical variables to statistical models significantly improves the ability to control for important confounders that are not available in administrative datasets. This suggests that the results from prior studies using low resolution data may be inaccurate and may need to be repeated using detailed clinical data.

Pressure Injuries in Critical Care Patients in US Hospitals: Results of the International Pressure Ulcer Prevalence Survey

PURPOSE

The purpose of this secondary analysis was to examine pressure injury (PI) prevalence, PI risk factors, and prevention practices among adult critically ill patients in critical care units in the United States using the International Pressure Ulcer Prevalence™ (IPUP) Survey database from 2018 to 2019.

DESIGN

Observational, cohort study with cross-sectional data collection and retrospective data analysis.

SUBJECTS AND SETTING

The sample comprised 41,866 critical care patients drawn from a sample of 296,014 patients in US acute care facilities who participated in the 2018 and/or 2019 IPUP surveys. The mean age among critical care patients was 63.5 years (16.3) and 55% were male. All geographic regions of the United States were represented in this sample, with the greatest percentages from the Southeast (47.5%) and Midwest (17.5%) regions.

METHODS

Overall critical care PI prevalence and hospital-acquired PI (HAPI) rates were obtained and analyzed using the 2018/2019 IPUP survey database. Critical care PI risk factors included in the database were analyzed using frequency distributions. Prevention practices among critically ill patients were analyzed to evaluate differences in practices between patients with no PIs, superficial PIs (stage 1, stage 2), and severe PIs (stage 3, stage 4, unstageable, deep tissue pressure injury).

RESULTS

The overall PI prevalence for critical care patients was 14.3% (n = 5995) and the overall HAPI prevalence was 5.85% (n = 2451). In patients with severe HAPIs, the most common risk factors were diabetes mellitus (29.5%), mechanical ventilation (27.6%), and vasopressor agents (18.9%). Significant differences between patients with no PIs as compared to those with superficial or severe HAPIs (P = .000) for all prevention practices were found.

CONCLUSIONS

Study findings support the gaps elucidated in previous critical care studies on PI development in this population. The 2 most persistent gaps currently challenging critical care practitioners are (1) accurate risk quantification in this population and (2) the potential for unavoidability in PI development among critically ill patients.

Risk Factors for Pressure Injury Development Among Critical Care Patients

Identification of the appropriate pressure injury (PI) risk factors is the first step in successful PI prevention. Measuring PI risk through formalized PI risk assessment is an essential component of any PI prevention program. Major PI risk factors identified in the empirical literature in the critical care population include age, diabetes, hypotension, mobility, prolonged intensive care unit admission, mechanical ventilation and vasopressor administration. Future risk assessment using sophisticated data analytics available in the electronic medical record may result in earlier, targeted PI prevention and will improve our understanding of risk factors that may contribute to unavoidable PIs.

Lessons Learned From the First Pilot Study of the Compensatory Reserve Index After Congenital Heart Surgery Requiring Cardiopulmonary Bypass

BACKGROUND

Early warning systems that utilize dense physiologic data and machine learning may aid prediction of decompensation after congenital heart surgery (CHS). The Compensatory Reserve Index (CRI) analyzes changing features of the pulse waveform to predict hemodynamic decompensation in adults, but it has never been studied after CHS. This study sought to understand the feasibility, safety, and potential utility of CRI monitoring after CHS with cardiopulmonary bypass (CPB).

METHODS

A single-center prospective pilot cohort of patients undergoing pulmonary valve replacement was studied. Compensatory Reserve Index was continuously measured from preoperative baseline through the first 24 postoperative hours. Average CRI values during selected procedural phases were compared between patients with an intensive care unit (ICU) length of stay (LOS) <3 days versus LOS ≥3 days.

RESULTS

Twenty-three patients were enrolled. On average, 17,445 (±3,152) CRI data points were collected and 0.33% (±0.40) of data were missing per patient. There were no adverse events related to monitoring. Five (21.7%) patients had an ICU LOS ≥3 days. Compared to the ICU LOS <3 days group, the ICU LOS ≥3 days group had a greater decrease in CRI from baseline to immediately after CPB (-0.3 ± 0.1 vs -0.1 ± 0.2, P = .003) and were less likely to recover to baseline CRI during the monitoring period (20% vs 83%, P = .017).

CONCLUSIONS

Compensatory Reserve Index monitoring after CHS with CPB seems feasible and safe. Early changes in CRI may precede meaningful clinical outcomes, but this requires further study.

Evaluation of standard and semantically-augmented distance metrics for neurology patients

Background

Patient distances can be calculated based on signs and symptoms derived from an ontological hierarchy. There is controversy as to whether patient distance metrics that consider the semantic similarity between concepts can outperform standard patient distance metrics that are agnostic to concept similarity. The choice of distance metric can dominate the performance of classification or clustering algorithms. Our objective was to determine if semantically augmented distance metrics would outperform standard metrics on machine learning tasks.

Methods

We converted the neurological findings from 382 published neurology cases into sets of concepts with corresponding machine-readable codes. We calculated patient distances by four different metrics (cosine distance, a semantically augmented cosine distance, Jaccard distance, and a semantically augmented bipartite distance). Semantic augmentation for two of the metrics depended on concept similarities from a hierarchical neuro-ontology. For machine learning algorithms, we used the patient diagnosis as the ground truth label and patient findings as machine learning features. We assessed classification accuracy for four classifiers and cluster quality for two clustering algorithms for each of the distance metrics.

Results

Inter-patient distances were smaller when the distance metric was semantically augmented. Classification accuracy and cluster quality were not significantly different by distance metric.

Conclusion

Although semantic augmentation reduced inter-patient distances, we did not find improved classification accuracy or improved cluster quality with semantically augmented patient distance metrics when applied to a dataset of neurology patients. Further work is needed to assess the utility of semantically augmented patient distances.

Identifying Risk Factors for Pressure Injury in Adult Critical Care Patients

BACKGROUND

Critically ill patients have a variety of unique risk factors for pressure injury. Identification of these risk factors is essential to prevent pressure injury in this population.

OBJECTIVE

To identify factors predicting the development of pressure injury in critical care patients using a large data set from the PhysioNet MIMIC-III (Medical Information Mart for Intensive Care) clinical database.

METHODS

Data for 1460 patients were extracted from the database. Variables that were significant in bivariate analyses were used in a final logistic regression model. A final set of significant variables from the logistic regression was used to develop a decision tree model.

RESULTS

In regression analysis, cardiovascular disease, peripheral vascular disease, pneumonia or influenza, cardiovascular surgery, hemodialysis, norepinephrine administration, hypotension, septic shock, moderate to severe malnutrition, sex, age, and Braden Scale score on admission to the intensive care unit were all predictive of pressure injury. Decision tree analysis revealed that patients who received norepinephrine, were older than 65 years, had a length of stay of 10 days or less, and had a Braden Scale score of 15 or less had a 63.6% risk of pressure injury.

CONCLUSION

Determining pressure injury risk in critically ill patients is complex and challenging. One common pathophysiological factor is impaired tissue oxygenation and perfusion, which may be nonmodifiable. Improved risk quantification is needed and may be realized in the near future by leveraging the clinical information available in the electronic medical record through the power of predictive analytics.

A practical approach to storage and retrieval of high-frequency physiological signals

OBJECTIVE

Storage of physiological waveform data for retrospective analysis presents significant challenges. Resultant data can be very large, and therefore becomes expensive to store and complicated to manage. Traditional database approaches are not appropriate for large scale storage of physiological waveforms. Our goal was to apply modern time series compression and indexing techniques to the problem of physiological waveform storage and retrieval.

APPROACH

We deployed a vendor-agnostic data collection system and developed domain-specific compression approaches that allowed long term storage of physiological waveform data and other associated clinical and medical device data. The database (called AtriumDB) also facilitates rapid retrieval of retrospective data for high-performance computing and machine learning applications.

MAIN RESULTS

A prototype system has been recording data in a 42-bed pediatric critical care unit at The Hospital for Sick Children in Toronto, Ontario since February 2016. As of December 2019, the database contains over 720,000 patient-hours of data collected from over 5300 patients, all with complete waveform capture. One year of full resolution physiological waveform storage from this 42-bed unit can be losslessly compressed and stored in less than 300 GB of disk space. Retrospective data can be delivered to analytical applications at a rate of up to 50 million time-value pairs per second.

SIGNIFICANCE

Stored data are not pre-processed or filtered. Having access to a large retrospective dataset with realistic artefacts lends itself to the process of anomaly discovery and understanding. Retrospective data can be replayed to simulate a realistic streaming data environment where analytical tools can be rapidly tested at scale.