Download presentation
Presentation is loading. Please wait.
Published byHilary Walters Modified over 8 years ago
1
A Preliminary Feature Extraction Based Regression Analysis for Predicting Patient Satisfaction on Physician-Patient Cancer Prognosis Communication Nan Kong Weldon School of Biomedical Engineering Purdue University Shuai Fang, Wenting Shi, and Cleveland G. Shields July 10, 2014 @ ICSH, Beijing, China
2
Outline Background and Introduction Patient-centeredness State-of-the-art research Analyzed Data on Cancer Prognosis Communication Human subject studies Measurement instrument design Principal Component Analysis (PCA) Preliminary Case Study PCA vs. alternatives Conclusions and Future Work Incorporation of conversation data 2
3
Patient-Physician Communication Good communication Treat patients as unique individuals Support patient-centered care Many barriers exist Language barriers Racial and ethnic concordance Effects of disabilities on patient’s experience Physicians differ in communication proficiency o Cultural competency of physicians’ 3
4
Note: * Estimate. Expenditures shown in $US PPP (purchasing power parity). Source: Calculated by The Commonwealth Fund based on 2007 International Health Policy Survey; 2008 International Health Policy Survey of Sicker Adults; 2009 International Health Policy Survey of Primary Care Physicians; Commonwealth Fund Commission on a High Performance Health System National Scorecard; and Organization for Economic Cooperation and Development, OECD Health Data, 2009 (Paris: OECD, Nov. 2009). AUSCANGERNETHNZUKUS OVERALL RANKING (2010)3641527 Quality Care4752136 Effective Care2763514 Safe Care6531427 Coordinated Care4572136 Patient-Centered Care2536174 Access6.553142 Cost-Related Problem63.5 2517 Timeliness of Care6721345 Efficiency2653417 Equity4531627 Long, Healthy, Productive Lives1234567 Health Expenditures/Capita, 2007$3,357$3,895$3,588$3,837*$2,454$2,992$7,290 Country Rankings 1.00–2.33 2.34–4.66 4.67–7.00 4 Healthcare Comparative Assessment
5
Patient-Centered Care Take into account patient’s needs, values, and perspectives Provide guidance to patients in the context of full and unbiased information One of six interrelated factors constituting high-quality care In current practice, unclear how to effectively implement patient-centered care in various settings How to carry out patient-centered communication? Factors that result in communication effectiveness 5
6
State-of-the-Art Research Hypothesis-driven research conducted by health psychologists and communication researchers Measurement instrument design Human subject experiment design Identify important markers (features) on communication effectiveness Markers tend to be redundant and irrelevant Limited help in development of promising communication behavior recommendation/training strategies 6
7
Making the Communication Smarter 7 As more measure instruments become available, an increasing number of prediction models are being developed that include more features than any hypothesis testing research that has dealt with. Choose/synthesize from highly correlated features from survey data to make better predictions with relatively few subjects Analytic framework development
8
Prognosis Discussion Prognosis discussion tends to be emotionally difficult; hampered by the different focuses More so for end-stage cancer prognosis discussion In the literature, many factors shown to be effective Data collected initially for Shields et al. (2009) Main hypothesis in Shields et al. (2009): eliciting and validating patient concerns is a marker of physician willingness to discuss prognosis. Controlled study with standard patients Outcome: patient satisfaction Feature categories: (1) eliciting and validating; (2) attentive voice tone; (3) prognosis communication assessment; (4) miscellaneous 8 Shields CG, Coker CJ, Poulsen SS, Doyle JM, Fiscella K, Epstein RM, Griggs JJ. (2009). Patient-centered communication and prognosis discussions with cancer patients. Patient Education and Counseling, 77(3), 437-442.
9
Human Subject Study Design 20 Family physicians and 19 oncologists recruited for a pilot study Patient characteristics controlled by using standard patients (SPs) SP methodology used extensively in primary care research, but not in examination of oncologist visit Complete medical record sent to the recruited physicians to make the visits believable SP carried two recorders to the visit 9
10
Outcome Measure (I) A post-visit questionnaire given to the SPs 5 sections SP’s perception on the physician’s communication (5 questions, 1 – 5 scale, sum calculated) SP’s belief on how well the physician knows him (4 questions, 1 – 5 scale, sum calculated) How satisfied the SP is with the physician (1 question, 1 – 6 scale) SP’s trust on the physician (7 questions, 1 – 5 scale, mean calculated) SP’s overall trust (1 question, 0 – 10 scale) 10
11
Outcome Measure (II) 11
12
Predictive Variables (I) Measurement Instruments -- Eliciting and validating items (Shields et al. 2009) 12
13
Predictive Variables (II) Response categories for coding eliciting and validating concerns during prognosis discussion (Shields et al. 2009) Satisfactory level of coding reliability 13
14
Predictive Variables (III) Voice tone Attentive, anxious, hostile E.g., attentive voice tone, rating four separate factors: warmth, concern, worry, and openness 1-7 scale Prognosis communication assessment Ten items based on the components of the SPIKES protocol for delivering bad news (Baile et al. 2000) 1- 5 scale Physician use of certainty in the language Tally the amount of certainty words said by physicians E.g., absolute, certain, clear, complete, confident, definite, sure Miscellaneous 14 Baile WR, Buckman R, Lenzi R, Glober G, Beale EA, Kudelka AP. (2000). SPIKES – a six step protocol for delivering bad news: Application to the patients with cancer. Oncologist, 5(4), 302-11.
15
Predictive Variables & Descriptive Stats 15 In total, 13 input variables
16
Correlation of Selected Variables 16 Some variables are highly correlated Explore feature extraction techniques, e.g., PCA
17
Principal Component Analysis (PCA) A commonly used multivariate statistical analysis technique for finding patterns and reducing correlations in high-dimensional data Uses orthogonal transformation to convert a set of observations of possibly correlated variables into a set of linearly uncorrelated variables called principal components Applied in feature extraction 17
18
Our Comparative Study Standardization input variables Outcome measures and then aggregate them With PCA Varying the number of PCs, making selections with reasonably high variance interpreted Making the linear transform onto the space constructed by the selected PCs Linear regression Alternative methods Generalized linear regression with model selection 18
19
Feature Extraction 19
20
Comparative Results 20
21
Conclusions and Future Work Applied PCA in prediction of patient satisfaction in end-stage cancer prognosis communication Limitation: SP approach Provide guideline for improving patient-physician communication and shared decision making in the era of smart health The greatest untapped resource Challenge: ill-compressed sequence data Continue to explore dimensionality reduction techniques Coded conversation sequence data (see next slide) Audio recordings and scripts for recommendation system development 21
22
Conversation Sequence Data 22 Portion of the original coded data from one sequence Topics include Appointment, Depression, Family, Medical information, Other information, Medication, Pain management, Treatment Data formatting and presentation Feature extraction and regression: Lasso, regression shrinkage and selection
23
Acknowledgements Dr. Cleveland G. Shields, PhD Associate Professor, Department of Human Development and Family Studies, Purdue University Discovery Park Undergraduate Research Internship Program Dr. Xuegong Zhang, PhD Professor, Department of Automation, Tsinghua University Tsinghua-Purdue Two-Way Summer Undergraduate Research Internship Program Anonymous reviewers Conference organizers 23
24
THANK YOU! Questions and Comments? nkong@purdue.edu
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.