Notes for projects with the Vanderbilt Center for Surgical Quality Outcomes Research
Karen Hoffmann radiation paper
on PCOS data (she will also be doing a similar study in the Ceasar cohort, separately.)
could also do the same thing w/ surgery
Purpose is to report 15 year outcomes on rad patients. Anticipate a small sample size because few patients will survive that long.
Also to find predictors of outcomes like functional status
Outcomes: sex, urine, and bowel function scores and bother scores. One is 0-100. There are several more domains within each of sexual function, urine, and bowel.
Want to look at the outcomes separately by tx: radiation and surgery
3 observations (including baseline?) Some will have only 1 or 2.
Need to refine the outcome definition. Percent change? NO! Difference from baseline? Return to bl (Y/N)?
Recommended modeling the post rad assessments as a function of the bl assessment (not a change measurement.)
I proposed using curveRep()? in rms/Hmisc to show representative curves. Sent email with details to TK.
Project 2: looking at radiation patients only, model QOL (functional status) over time as a function of bl function and other bl variables.
CHOICE CEASAR extension grant, Dan Barocas
Dan is applying for a new grant. The part we'll be interested in is developing a quality index for research use.
Meeting 2013 June 3
Attending: JoAnn Tatsuki, Dan
Ceasar study is an observational comparative effectiveness study of treatments for prostate cancer. It has five sites. Patients are followed over time.
Structure: qualifications of MD, resources
Process: imaging, lymphadenectomy, communication with PCP
Stakeholder involvement:
Include patients, payers, providers, hospital administrators
Perhaps develop a separate index for each group?
They need to help us prioritize the quality measures, and perhaps rank the outcomes in terms of importance –it may be that a composite outcome is best, comprised of urinary/sexual/bowel function, satisfaction, complications and cancer control = QUEST. Patients won't care much about efficiency, but will care about safety. Payers will care about efficiency, and will care about quality and safety as they pertain to cost. etc.
Quality Measures: Need to compose a list of candidate measures
Independent variables (other than the quality measures): age, race, SES, disease characteristics, psychosocial measures, treatment in some models (some models will apply to only one treatment)
Outcome Measures: May need to make a composite of function/complication/satisfaction/cancer control. Alternatively, we could select the most important outcome for each stakeholder group and the others could be secondary - e.g., since virtually all patients survive, we could look at QOL as the main outcome and look at others as secondary
Cancer control is one outcome measure. It could be measured by PSA recurrence, need for further therapy, radiographical recurrence, biopsy-proven recurrence.
Modeling: Variable selection (beyond the base model) is the key. How to select – stakeholder opinion, a-priori selection, forward or backward stepwise (frowned upon), LASSO (Least Absolute Shrinkage…) technique.
Consider one model per stakeholder (with composite outcome) rather than multiple models per stakeholder (with multiple outcomes). But different stakeholders may value different outcomes differently???
Order of Operations
Link datasets (Aim 1 of current extension)
Build base model (satisfying Aim 1 and 2 of parent CEASAR grant)
Test individual candidate quality of care measures. These were originally proposed by a Delphi group (perhaps limit to those with satisfactory capture, or those that stakeholders believe have face validity) (Aim 2 of current extension)
Variable selection for inclusion in final model (perhaps with stakeholder involvement)
Build final model, including systematically selected quality measures/ Develop a quality index (Aim 3 of current extension)
Could use LASSO!
Measure the contribution to the outcome of the quality measures collectively, and individually.
Open questions
Optimal method for selecting variables
Optimal modeling technique (based in part on whether or not we use a composite outcome; we could select the most important outcome for each stakeholder group and the others could be secondary - e.g., since virtually all patients survive, we could look at QOL as the main
outcome and look at others as secondary)
Patient-level model??? Wouldn't it make more sense to develop a quality score for each provider and for each facility??
Think about relationship between hospital type (teaching, research, community), travel distance, SES, and outcome. (Travel distance is relatively low priority)
U51 O'Brien grant proposal, Jay Fowke's project
This is one of 3 projects in this proposal
Looks at associations between obesity and immunologic gene expression in prostate and between gene expression and lower urinary tract symptoms LUTS.
How does body fat impact the prostate. (The impact has already been established.)
Meeting 2012 November 12, JoAnn and Tatsuki
Read to understand pathway analysis better. They will use a signature t-score method.
Meeting 2012 November 9
They will measure gene expression in 511 candidate genes that regulate immune response in prostate tissue.
They can probably get about 500 patients, and we need to explain what they can detect.
Recruiting mainly at Urology Associates
Sample population is patients who get a prostate biopsy that are not found to have cancer or precancerous lesions.
Of those biopsied, about one third have cancer.
Data collected: interview, chart, blood, urine, bmi, body fat percentage, waist circumference, waist-hip ratio
LUTS is measured by a 7-item scale ranging from 0-35. It is right skewed with a mean of 7.
Aim 3 involves reassessing LUTS after 5 years. Expect 80% retention for sample size of 400.
Gene expression is a count of the affected ___ in each gene.
We can use lasso as a variable selection technique to choose the important genes.