Present: Peter Callas, Emily Houston, Jennifer Oshita,
Liliane Savard, Connie van Eeghen (5)
1.
Warm Up: PhD degrees… always taking longer than expected, sort of like finalizing
manuscripts with many co-authors!
2.
Jen’s methods
and survival analysis: Jen will explain
the project and methods, then ask for feedback on the covariates used to build
her model, on a secondary analysis, and how best to present the results
a. NHATS: National
health trends survey of older adults set up as a cohort study from 2011, still
ongoing
i.
This study has nine years of data: 2011-2020
ii.
What is the hazard of dying of someone with communication
disabilities (CD) vs someone without?
iii.
Including only community dwelling older adults: those living
independently with no assistive services imbedded in an institutional setting
iv.
Excluding persons who do not speak English or have a cognitive
impairment (dx of dementia)
1. Used an
orientation question to assess impairment, plus word recall exercises
v.
5K+ beneficiaries
1. No CD (about 3K)
2. Any CD: could
have single subtype of CD
a. Hearing only
(self-reported)
b. Expressive only
(self-reported)
c. Cognitive only
d. Multiple of
above CDs
b. Analysis
i.
If you have a CD at baseline, what is your HR over a nine year
period, adjusting for covariates that might change over time (excluding gender,
race, ethnicity), such as self rating of health, comorbidities, other
disabilities (number of ADLs), mobility disabilities,…
1. Not BMI or
other biometric measures
2. Not employment
status (all are 65+ years of age) – Jen will recheck this for an association
3. Not falls – Jen
will recheck this too; check with Nancy Gell
4. Consider lag
time of covariates over the 9 years for a future study (not this time)
ii.
Secondary analysis
1. Planning to break
down CDs by subtypes, but Expressive group is relatively small compared to
Hearing and Cognitive groups; the outcomes vary among groups
a. Good to help
assess which subtypes might be drivers of the outcomes
b. Sample sizes
are not too small
iii.
Assumptions: because Jen is using time as a covariate, and death
removes data for that and subsequent years, Jen had to impute missing data to prevent
Stata from removing those records
1. Checked against
the UCSD check list of assumptions
2. Reviewed Kaplan
Meier plots for logical outcomes
3. Goodness of fit
analysis (Nelson-Aalen cumulative hazard compared to Cox-Snell residual): Jen
will find out more about how to interpret these results
4. How does sudden
death affect the outcome? Likely shows
up similarly in both groups, so not a confounder.
5. Include the
number (median months?) with the presentation of the graph, along with IQR
6. Table 2:
represents the final model at a high level (none, any CDs) – looks good
c. Next steps
after following up on additional mediators and more investigation on interpretation
i.
Check on falls and employment
ii.
Follow up with Peter on Cox-Snell interpretation
iii.
Finish writing up
3.
Next week: TBD
Recorded by: CvE
Posts
