1. RegionAl: an Optimized Regional Classifier to Predict Mortality in
Chronic Obstructive Pulmonary (COPD)Disease Patients Haohan Zhang, Chunlei Tang, PhD, Joseph M. Plasek, Yun Xiong, PhD, Jing Ma, MD, PhD, Li Zhou, MD, PhD, David W. Bates, MD
Importance
Results
Materials and Methods
Discussion and Conclusion
Studies have shown that there exists a mismatch between patients’ wishes and actual care at the end of life. In United States, up to 60% of deaths happen in an acute care hospital, with patients receiving aggressive care in their final days due to physicians’ tendency to over-estimate prognoses or their ability to treat the patients.
Chronic Obstructive Pulmonary Disease (COPD) is a leading cause of mortality in America; affecting an estimated 14.7 million adult Americans diagnosed annually with COPD. Observing similar COPD conditions within a particular COPD stage is very important due to COPD’s slow progression.
Materials: We extracted a cohort containing 15,500 COPD patients who had received care at Partners Healthcare network and died between 2011 and 2017 from Partners research patient data registry.
Methods: We propose an algorithm called Region Atlas (RegionAl), which is an optimized variant of our previously-developed regional classifier based on a spiral timeline. We focus on two aspects of optimization.
Designing an automatic procedure to identify representative sentences by extracting the initial sentences containing a fixed minimum of 30 percent of words that map to a specific topic identified by Latent Dirichlet Allocation.
Using a tolerance range to achieve window (i.e., time interval) enlargement. Tolerance range is defined as the distance of observing similar COPD conditions between the upper and lower specification limit. Given a fixed disjoint n-day window as interval, the tolerance range is
This method is an improvement upon our previously developed regional classifier.
Tang C, Zhang H, et al. Developing a Regional Classifier to Track Patient Needs in Medical Literature Using Spiral Timelines on a Geographical Map. Proceedings of the IEEE BIBM 2017;
RegionAl is able to 1) efficiently give a probability of death for the individual patient by comparing a particular clinical document with the atlas under the permissible tolerance range; and 2) to visualize COPD conditions.
The experimental results demonstrate that RegionAl produces interpretable, accurate, and reliable results in estimating COPD mortality.
The atlas could also address a proxy problem – to predict the probability of death for a given patient within a permissible tolerance range, and to help make recommendations for palliative care referral.
Future work will evaluate our algorithm using diverse data sets with different mortality rates and from different data resources.
Spiral Atlas
Results
Author Affiliations
Shanghai Key Laboratory of Data Science, Fudan University, Shanghai, CHN;
Division of General Medicine and Primary Care, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA, USA;
Clinical and Quality Analysis, Partners HealthCare System, Boston, MA, USA;
Department of Biomedical Informatics, University of Utah School of Medicine, Salt Lake City, UT, USA;
Department of Population Medicine, Harvard Medical School, Boston, MA, USA.
Figure 1. A spiral atlas with representative sentences
Sentence Representations
These data demonstrate a very severe (FEV1 <35% of predicted) obstructive ventilatory deficit.
These data suggest a severe (FVC between 35 and 49% of predicted) restrictive ventilatory deficit.
Nonspecific T wave abnormality now evident in Inferior leads.
LDA results
FVC normal reduced FEV1 data predicted FEV1/FVC bronchodilator severe deficit restrictive suggest TLC abnormality spirometry moderate change diffusion range interpretation flow obstructive critical evident demonstrate wave
180 to 270 days before death
90 to 180 days before death ……
270 to 360 days before death
0 to 90 days before death
630 to 720 days before death 1
360 to 450 days before death
450 to 540 days before death
540 to 630 days before death
0 to 65 days before death
(b) time intervals changed
by the tolerance range
c) pulmonary notes within 0 to 65
days before death.
(a) a 90-day spiral atlas

2. RegionAl: an Optimized Regional Classifier to Predict Mortality in
Chronic Obstructive Pulmonary (COPD)Disease Patients Haohan Zhang, Chunlei Tang, PhD, Joseph M. Plasek, Yun Xiong, PhD, Jing Ma, MD, PhD, Li Zhou, MD, PhD, David W. Bates, MD
Importance
Results
Materials and Methods
Discussion and Conclusion
Studies have shown that there exists a mismatch between patients’ wishes and actual care at the end of life. In United States, up to 60% of deaths happen in an acute care hospital, with patients receiving aggressive care in their final days due to physicians’ tendency to over-estimate prognoses or their ability to treat the patients.
Chronic Obstructive Pulmonary Disease (COPD) is a leading cause of mortality in America; affecting an estimated 14.7 million adult Americans diagnosed annually with COPD. Observing similar COPD conditions within a particular COPD stage is very important due to COPD’s slow progression.
Materials: We extracted a cohort containing 15,500 COPD patients who had received care at Partners Healthcare network and died between 2011 and 2017 from Partners research patient data registry.
Methods: We propose an algorithm called Region Atlas (RegionAl), which is an optimized variant of our previously-developed regional classifier based on a spiral timeline. We focus on two aspects of optimization.
Designing an automatic procedure to identify representative sentences by extracting the initial sentences containing a fixed minimum of 30 percent of words that map to a specific topic identified by Latent Dirichlet Allocation.
Using a tolerance range to achieve window (i.e., time interval) enlargement. Tolerance range is defined as the distance of observing similar COPD conditions between the upper and lower specification limit. Given a fixed disjoint n-day window as interval, the tolerance range is
This method is an improvement upon our previously developed regional classifier.
Tang C, Zhang H, et al. Developing a Regional Classifier to Track Patient Needs in Medical Literature Using Spiral Timelines on a Geographical Map. Proceedings of the IEEE BIBM 2017;
RegionAl is able to 1) efficiently give a probability of death for the individual patient by comparing a particular clinical document with the atlas under the permissible tolerance range; and 2) to visualize COPD conditions.
The experimental results demonstrate that RegionAl produces interpretable, accurate, and reliable results in estimating COPD mortality.
The atlas could also address a proxy problem – to predict the probability of death for a given patient within a permissible tolerance range, and to help make recommendations for palliative care referral.
Future work will evaluate our algorithm using diverse data sets with different mortality rates and from different data resources.
Spiral Atlas
Results
Author Affiliations
Shanghai Key Laboratory of Data Science, Fudan University, Shanghai, CHN;
Division of General Medicine and Primary Care, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA, USA;
Clinical and Quality Analysis, Partners HealthCare System, Boston, MA, USA;
Department of Biomedical Informatics, University of Utah School of Medicine, Salt Lake City, UT, USA;
Department of Population Medicine, Harvard Medical School, Boston, MA, USA.
Figure 1. A spiral atlas with representative sentences
Sentence Representations
These data demonstrate a very severe (FEV1 <35% of predicted) obstructive ventilatory deficit.
These data suggest a severe (FVC between 35 and 49% of predicted) restrictive ventilatory deficit.
Nonspecific T wave abnormality now evident in Inferior leads.
LDA results
FVC normal reduced FEV1 data predicted FEV1/FVC bronchodilator severe deficit restrictive suggest TLC abnormality spirometry moderate change diffusion range interpretation flow obstructive critical evident demonstrate wave
180 to 270 days before death
90 to 180 days before death ……
270 to 360 days before death
0 to 90 days before death
630 to 720 days before death 2
360 to 450 days before death
450 to 540 days before death
540 to 630 days before death
0 to 65 days before death
(b) time intervals changed
by the tolerance range
c) pulmonary notes within 0 to 65
days before death.
(a) a 90-day spiral atlas