11

Critique-An Integrated Data Mining Approach to Real-time Clinical Monitoring and Deterioration WarningABDELFATTAH AL ZAQQA

22An Integrated Data Mining Approach to Real-time Clinical Monitoring and Deterioration Warning

Authors: Yi Mao

Wenlin Chen

Wenlin Chen

Chenyang Lu

Marin Kollef

Thomas C. Bailey

Publisher: ACM New York, NY, USA ©2012, ACM SIGKDD international conference on Knowledge discovery and data mining.

Date: August 12-16, 2012 Beijing, China.

Washington University in St.Louis, Saint Louis, USA

33What is the specific problem that the paper is intended to address?

• In general hospital units, patients' vital signs are typically collected manually by a nurse, at a granularity of only a handful of readings per day, which confronts us with the challenged of sparseness and irregularity of data.

• In most hospitals, only intensive care units (ICUs) are equipped with real-time electronic medical sensors.

• Mining multiple time series is a more challenging problem.

• Granularity :  is the extent to which a system is broken down into small parts, either the system itself or its description or observation..

44What is the motivation for this paper?

Improve the quality of patient care in hospitals.

Every year, 4-17% of patients undergo cardiopulmonary or respiratory arrest while in hospitals.

Lots of these patients could have been saved if warning of serious clinical events could be provided early, before its occurrence rather than when it is happening.

55What are the previous solutions and why are they inadequate?

Most prior studies focus on some specific disease prediction.

There have been little study on general prediction and warning for serious clinical deterioration and death.

Time series data has rich information that previous work did not make full use of.

66What is the proposed solution (hypothesis, idea, design)?

Monitor real-time clinical data. Real-time Data Sensing(RDS)

Performs feature selection to select the most relevant and discriminative features.

integrated data mining approach to identify the signs of clinical deterioration and provide early warning for possible mortality.

RDS enables patients' vital signs data in general hospital units to be collected via wireless sensors.

77logic, argument, evidence, artifacts, or experiments

multiple time series, features such as correlation and coherence can be used.

Through pilot study at the Barnes Jewish Hospital. RDS system has been installed in its step-down unit.

Based on the MIMIC II (Multiparameter Intelligent Monitoring in Intensive Care).(because no enough data)

Barnes Jewish affiliated to Washington University

88

Cont.

Data were collected between 2001 and 2008 from a variety of ICUs

Comparison of linear and nonlinear separation using DFA(Detrended Fluctuation Analysis)

final logistic regression model is approved by medical experts for a clinical trial at a major hospital

99

contributions

1. Develop an integrated data mining approach to early deterioration warning based on real-time monitoring data.(ICU and non ICU patients)

2. Bridge the gap between biomedical community and data mining community by incorporating popular features and methods from both areas.

3. Strengthen the early warning system by applying important data mining techniques such as feature selection to address the challenging multi dimensional time-series classification problem.

4. Integrated approach to a large collection of real patient data recorded from ICUs and show significant improvement over previous methods.

1010Does the work appear to be original and significant?

Original: the way of collecting data is different( here by electronic devices at ICU and wireless sensors).

Zhou proposed a Multi-Task Learning Formulation for Predicting Alzheimer's Disease.By integration of heterogeneous data (neuro images, demographic, and genetic measures).(designed for some specific diseases ).

Significant: not only because it’s Preventing clinical deterioration and death in hospital patients they try to bridge the gap in reading data in minutes instead of hours).

1111What are future directions for this research (author’s and yours, perhaps driven by shortcomings or other critiques)?

“Since our RDS system is still in a smaller scale clinical trial and does not yield enough data.”

From this I think authors will test their approach based on RDS system data, after it be considered in acceptable scale for making experiments based on it.

1212What are the most important citations in the paper to follow up on if you were to read further on this topic? Why?

[21] J. Zhou, Y. Lei, J. Liu, and J. Ye. A multi-task learning formulation for predicting disease progression. KDD '11 Proceedings of the 17th ACM SIGKDD international conference on Knowledge discovery and data mining.

Why? Same context.

1313

Thanks