Adapting Surgical Models to Individual Hospitals Using Transfer Learning

Lee, Gyemin; Rubinfeld, Ilan; Syed, Zeeshan

doi:10.1109/icdmw.2012.93

Cited by 21 publications

(20 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Changes in the coefficients of the risk model were monitored for different temporal updating schemes, but performance measures for discrimination and calibration were not investigated. Recently in the informatics field approaches to transfer learning for adapting risk tools from one hospital to another have developed [32, 40]. These rely on global maximization of an objective function that sums over individual hospitals, allowing individual hospitals to collect different predictors.…”

Section: Discussionmentioning

confidence: 99%

Improving patient prostate cancer risk assessment: Moving from static, globally-applied to dynamic, practice-specific risk calculators

Ströbl

Vickers

Calster

et al. 2015

Journal of Biomedical Informatics

View full text Add to dashboard Cite

Clinical risk calculators are now widely available but have generally been implemented in a static and one-size-fits-all fashion. The objective of this study was to challenge these notions and show via a case study concerning risk-based screening for prostate cancer how calculators can be dynamically and locally tailored to improve on-site patient accuracy. Yearly data from five international prostate biopsy cohorts (3 in the US, 1 in Austria, 1 in England) were used to compare 6 methods for annual risk prediction: static use of the online US-developed Prostate Cancer Prevention Trial Risk Calculator (PCPTRC); recalibration of the PCPTRC; revision of the PCPTRC; building a new model each year using logistic regression, Bayesian prior-to-posterior updating, or random forests. All methods performed similarly with respect to discrimination, except for random forests, which were worse. All methods except for random forests greatly improved calibration over the static PCPTRC in all cohorts except for Austria, where the PCPTRC had the best calibration followed closely by recalibration. The case study shows that a simple annual recalibration of a general online risk tool for prostate cancer can improve its accuracy with respect to the local patient practice at hand.

show abstract

Section: Discussionmentioning

confidence: 99%

Improving patient prostate cancer risk assessment: Moving from static, globally-applied to dynamic, practice-specific risk calculators

Ströbl

Vickers

Calster

et al. 2015

Journal of Biomedical Informatics

View full text Add to dashboard Cite

show abstract

“…Different systems have differing data distributions [ 53 ] and collected attributes, impacting the performance and applicability of a model trained using one system’s data for another system [ 54 - 57 ]. To address these two issues, one can perform transfer learning and use other source systems’ information to improve model accuracy for the target system [ 54 , 58 , 59 ]. Transfer learning typically requires using other source systems’ raw data [ 60 , 61 ].…”

Section: Limitations Of Current Patient Identification Methods For Asmentioning

confidence: 99%

A Roadmap for Optimizing Asthma Care Management via Computational Approaches

Luo

Sward

2017

JMIR Med Inform

View full text Add to dashboard Cite

Asthma affects 9% of Americans and incurs US $56 billion in cost, 439,000 hospitalizations, and 1.8 million emergency room visits annually. A small fraction of asthma patients with high vulnerabilities, severe disease, or great barriers to care consume most health care costs and resources. An effective approach is urgently needed to identify high-risk patients and intervene to improve outcomes and to reduce costs and resource use. Care management is widely used to implement tailored care plans for this purpose, but it is expensive and has limited service capacity. To maximize benefit, we should enroll only patients anticipated to have the highest costs or worst prognosis. Effective care management requires correctly identifying high-risk patients, but current patient identification approaches have major limitations. This paper pinpoints these limitations and outlines multiple machine learning techniques to address them, providing a roadmap for future research.

show abstract

“…Previous work on using transfer learning to improve performance of clinical risk stratification models has focused on feature-representation transfer [26] and parameter-transfer [13]. In [26], the authors show that including hospital-specific features can improve performance on the task of interest.…”

Section: Related Workmentioning

confidence: 99%

“…They also include data from related tasks in training, but they do not consider weighting the auxiliary examples based on the similarity to the target task. In [13], the authors show that adapting a global model to a local institution can lead to improved performance over simply using the global model.…”

Section: Related Workmentioning

confidence: 99%

“…Risk stratification models are used to determine a patient's category of risk, such as high-risk or low-risk. Although this limits the data that can be used, previous work in the field of cardiac surgery, and in the context of other clinical questions, has shown that models that account for institution-and procedure-specific characteristics can have improved performance over those that do not [26,13,19,11]. Even at the largest of medical centers, the amount of data available for a specific surgery is small relative to the number of potentially relevant features.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Instance Weighting for Patient-Specific Risk Stratification Models

Gong

Sundt

Rawn

et al. 2015

Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

View full text Add to dashboard Cite

Accurate risk models for adverse outcomes can provide important input to clinical decision-making. Surprisingly, one of the main challenges when using machine learning to build clinically useful risk models is the small amount of data available. Risk models need to be developed for specific patient populations, specific institutions, specific procedures, and specific outcomes. With each exclusion criterion, the amount of relevant training data decreases, until there is often an insufficient amount to learn an accurate model. This difficulty is compounded by the large class imbalance that is often present in medical applications.In this paper, we present an approach to address the problem of small data using transfer learning methods in the context of developing risk models for cardiac surgeries. We explore ways to build surgery-specific and hospital-specific models (the target task) using information from other kinds of surgeries and other hospitals (source tasks). We propose a novel method to weight examples based on their similarity to the target task training examples to take advantage of the useful examples while discounting less relevant ones.We show that incorporating appropriate source data in training can lead to improved performance over using only target task training data, and that our method of instance weighting can lead to further improvements. Applied to a surgical risk stratification task, our method, which used data from two institutions, performed comparably to the risk model published by the Society for Thoracic Surgeons, which was developed and tested on over one hundred thousand surgeries from hundreds of institutions.

show abstract

Adapting Surgical Models to Individual Hospitals Using Transfer Learning

Cited by 21 publications

References 12 publications

Improving patient prostate cancer risk assessment: Moving from static, globally-applied to dynamic, practice-specific risk calculators

Improving patient prostate cancer risk assessment: Moving from static, globally-applied to dynamic, practice-specific risk calculators

A Roadmap for Optimizing Asthma Care Management via Computational Approaches

Instance Weighting for Patient-Specific Risk Stratification Models

Contact Info

Product

Resources

About