Identifying Differences in Physician Communication Styles with a Log-Linear Transition Component Model
Authors: Byron Wallace, Issa Dahabreh, Thomas Trikalinos, Michael Barton Laws, Ira Wilson, Eugene Charniak
AAAI 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We propose a novel approach toward this end in which we model speech act transitions in conversations via a log-linear model incorporating physician specific components. We train this model over transcripts of outpatient visits annotated with speech act codes and then cluster physicians in (a transformation of) this parameter space. We find significant correlations between the induced groupings and patient survey response data comprising ratings of physician communication. We perform a hierarchical regression analysis to test for correlation between physician cluster assignments and survey data comprising patient assessments of physician communication. Our data comprises transcripts manually segmented and annotated according to the General Medical Interaction Analysis System (GMIAS) (Laws et al. 2011a; Laws 2013). |
| Researcher Affiliation | Academia | Byron C. Wallace and Issa J. Dahabreh and Thomas A. Trikalinos Center for Evidence-Based Medicine, Brown University {byron wallace, issa dahabreh, thomas trikalinos}@brown.edu M. Barton Laws and Ira Wilson Health Services, Policy & Practice, Brown University {michael barton laws, ira wilson}@brown.edu Eugene Charniak Computer Science, Brown University ec@cs.brown.edu |
| Pseudocode | No | The paper describes the model and optimization process (e.g., 'We fit this model via gradient descent (specifically, Newton optimization)...'), but it does not include any pseudocode or a formally labeled algorithm block. |
| Open Source Code | No | The paper does not provide any explicit statements about open-sourcing the code for the described methodology, nor does it provide a link to a code repository. |
| Open Datasets | Yes | Our data comprises transcripts manually segmented and annotated according to the General Medical Interaction Analysis System (GMIAS) (Laws et al. 2011a; Laws 2013). Details on operational labeling criteria are provided elsewhere (Laws 2013). |
| Dataset Splits | No | The paper states, 'For this work, we experiment with 360 physician-patient visits annotated with GMIAS speech act codes.' It describes the dataset size and composition but does not specify how this data was split into training, validation, and test sets for model development and evaluation. It talks about 'fitting the model' and then 'assessing association' with survey data, but no typical ML-style data splits. |
| Hardware Specification | No | The paper does not provide any specific details regarding the hardware (e.g., CPU, GPU, memory, cloud resources) used to perform the experiments or train the models. |
| Software Dependencies | No | The paper mentions methods like 'gradient descent (specifically, Newton optimization)' and references a book for 'hierarchical regression analysis (Rabe-Hesketh and Skrondal 2008)', but it does not specify any software dependencies (e.g., programming languages, libraries, frameworks) with version numbers. |
| Experiment Setup | Yes | We place Gaussian priors on all physician specific terms (πdr, σdr s and λdr ρ ) Normal(0, 0.25) (recall that this is on the log scale). We continued descent until likelihood ceased to increase or a maximum number of iterations was reached (here, 100). Because we have only 41 doctors in total, we set k=2. |