CoChat
Sign Up
Cited by 170Machine Learning in HealthcareArtificial Intelligence in HealthcareChronic Obstructive Pulmonary Disease (COPD) Research

The DOI matches its Crossref record; 41 of the 41 references it cites were found in a registry or an index.

Predictive Modeling of the Hospital Readmission Risk from Patients’ Claims Data Using Machine Learning: A Case Study on COPD

Open access
Min, Xu, Yu, Bin, Wang, Fei·2019·Scientific Reports
DOI 10.1038/s41598-019-39071-yPMC PMC6382784
Open this paper in CoChat — free accountOpen PDF

Abstract

Chronic Obstructive Pulmonary Disease (COPD) is a prevalent chronic pulmonary condition that affects hundreds of millions of people all over the world. Many COPD patients got readmitted to hospital within 30 days after discharge due to various reasons. Such readmission can usually be avoided if additional attention is paid to patients with high readmission risk and appropriate actions are taken. This makes early prediction of the hospital readmission risk an important problem. The goal of this paper is to conduct a systematic study on developing different types of machine learning models, including both deep and non-deep ones, for predicting the readmission risk of COPD patients. We evaluate those different approaches on a real world database containing the medical claims of 111,992 patients from the Geisinger Health System from January 2004 to September 2015. The patient features we build the machine learning models upon include both knowledge-driven ones, which are the features extracted according to clinical knowledge potentially related to COPD readmission, and data-driven features, which are extracted from the patient data themselves. Our analysis showed that the prediction performance in terms of Area Under the receiver operating characteristic (ROC) Curve (AUC) can be improved from around 0.60 using knowledge-driven features, to 0.653 by combining both knowledge-driven and data-driven features, based on the one-year claims history before discharge. Moreover, we also demonstrate that the complex deep learning models in this case cannot really improve the prediction performance, with the best AUC around 0.65.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s41598-019-39071-y

Cites

41 of 41 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 Three types of patient representations for incorporating the temporal information. (a) Sequence Representation; (b) Matrix Representation with Regular Time Intervals (MR-RTI); (c) Matrix Representation with Irregular Time Intervals (MR-ITI).
Figure 1 Three types of patient representations for incorporating the temporal information. (a) Sequence Representation; (b) Matrix Representation with Regular Time Intervals (MR-RTI); (c) Matrix Representation with Irregular Time Intervals (MR-ITI).CC BY · PubMed Central
Figure 2 AUC performance achieved by predictive models with different types of features and machine learning models. In (a), ‘hos’, ‘lace’, ‘hand’, ‘knowledge’ represent HOSPITAL score, LACE index, handcrafted feature, and the combination of all these three kinds of features. For the legend, ‘lr’, ‘lrl1’, ‘lrl1’, ‘rf’, ‘svm’, ‘gbdt’, ‘mlp’ represent logistic regression, logistic regression with L1 penalty, logistic regression with L2 penalty, random forest, support vector machine, gradient boosting decision tree, and multi-layer perceptron. The same naming convention is also applied in the legends of the follow-up figures. In (b), COPD readmission prediction performance with combined data -driven features. For the x-axis, ‘data_bow’, ‘data_tfidf’ and ‘data_bbow’ represent BoW, TFIDF and BBoW features.
Figure 2 AUC performance achieved by predictive models with different types of features and machine learning models. In (a), ‘hos’, ‘lace’, ‘hand’, ‘knowledge’ represent HOSPITAL score, LACE index, handcrafted feature, and the combination of all these three kinds of features. For the legend, ‘lr’, ‘lrl1’, ‘lrl1’, ‘rf’, ‘svm’, ‘gbdt’, ‘mlp’ represent logistic regression, logistic regression with L1 penalty, logistic regression with L2 penalty, random forest, support vector machine, gradient boosting decision tree, and multi-layer perceptron. The same naming convention is also applied in the legends of the follow-up figures. In (b), COPD readmission prediction performance with combined data -driven features. For the x-axis, ‘data_bow’, ‘data_tfidf’ and ‘data_bbow’ represent BoW, TFIDF and BBoW features.CC BY · PubMed Central
Figure 3 Comparisons of the predictive performance of different types of data-driven features on COPD readmission. In (a), ‘dx_bow’, ‘dx_tfidf’, ‘dx_bbow’ represent BoW, TFIDF and BBoW feature for diagnosis records. ‘dx_ccs_bow’, ‘dx_ccs_tfidf’, ‘dx_ccs_bbow’ represent Bow, TFIDF and BBoW feature for grouped diagnosis codes using CCS hierarchy. The same naming convention also applies to (b–d).
Figure 3 Comparisons of the predictive performance of different types of data-driven features on COPD readmission. In (a), ‘dx_bow’, ‘dx_tfidf’, ‘dx_bbow’ represent BoW, TFIDF and BBoW feature for diagnosis records. ‘dx_ccs_bow’, ‘dx_ccs_tfidf’, ‘dx_ccs_bbow’ represent Bow, TFIDF and BBoW feature for grouped diagnosis codes using CCS hierarchy. The same naming convention also applies to (b–d).CC BY · PubMed Central
Figure 4 Performance comparison among different time fusion and different embedding strategies. In (a), ‘basic’ indicates the most basic model where we use the sequence input without any time weighting or attention mechanisms. ‘basic_day’, ‘basic_week’ and ‘basic_month’ indicate the model using matrix input with regular time interval, whose time granularity is day, week and month. ‘time_day’, ‘att_day’, ‘time_att_day’ indicate the model using matrix input of irregular time interval, plus the time weighting layer, attention weighting layer, and both layers. For all models, we adopt Word2Vec embedding, and let the embedding layer be trainable when training the deep models. In (b),’skipgrams’,’skipgrams_w’ and’med2vec’ indicate the models using embedding matrix learned by Skip-grams model, the time weighted Skip-grams model, and the Med2vec model. The suffix’_fixed’ means that we keep the parameters in the embedding layer fixed during training.’one-hot’ indicates the model simply uses the one-hot embedding layer.
Figure 4 Performance comparison among different time fusion and different embedding strategies. In (a), ‘basic’ indicates the most basic model where we use the sequence input without any time weighting or attention mechanisms. ‘basic_day’, ‘basic_week’ and ‘basic_month’ indicate the model using matrix input with regular time interval, whose time granularity is day, week and month. ‘time_day’, ‘att_day’, ‘time_att_day’ indicate the model using matrix input of irregular time interval, plus the time weighting layer, attention weighting layer, and both layers. For all models, we adopt Word2Vec embedding, and let the embedding layer be trainable when training the deep models. In (b),’skipgrams’,’skipgrams_w’ and’med2vec’ indicate the models using embedding matrix learned by Skip-grams model, the time weighted Skip-grams model, and the Med2vec model. The suffix’_fixed’ means that we keep the parameters in the embedding layer fixed during training.’one-hot’ indicates the model simply uses the one-hot embedding layer.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 37 references

Cited by

Sorted by citations

Items 101–170 of 170

Related

See all

Metadata

Authors
Min, X., Yu, B., & Wang, F.
Affiliations
Weill Cornell Medicine (US)
Corresponding author
Wang, Fei
ORCID
Venue
Scientific Reports
Publisher
Springer Science and Business Media LLC
Published
2019-02-20
Volume
9
Issue
1
Pages
2362-2362
ISSN
2045-2322 (ISSN-L)
DOI
10.1038/s41598-019-39071-y
PMC
PMC6382784
OpenAlex
W2923546398
Cited by
170
References
37
Citations by year
2026: 8 · 2025: 12 · 2024: 23 · 2023: 35 · 2022: 21 · 2021: 32 · 2020: 27 · 2019: 11
FWCI
10.38
Type
journal article
Language
en
Open access
Licence
Locations
  • Scientific Reports (publishedVersion, cc-by)
  • Europe PMC (PubMed Central) (publishedVersion, other-oa)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
Funders
National Science Foundation
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning in Healthcare
  • Health Sciences › Health Professions › Health Information Management › Artificial Intelligence in Healthcare
  • Health Sciences › Medicine › Pulmonary and Respiratory Medicine › Chronic Obstructive Pulmonary Disease (COPD) Research
Keywords
hospital readmission prediction, chronic obstructive pulmonary disease, health insurance claims data, feature extraction, deep learning, 30-day readmission risk, Geisinger Health System
MeSH terms
Machine Learning, China, Female, Forecasting, Humans, Male, Patient Discharge, Patient Readmission, Risk Factors, ROC Curve, Models, Statistical, Pulmonary Disease, Chronic Obstructive

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review