EURURO Vol. 72 No. 2

309 320

included time since treatment, treatment type, race/ethnicity, and all
their interactions as independent variables. In these unadjusted models,
responses included all of each patient’s scores over time within a
particular domain, including the baseline score. Time since beginning
treatment (time since baseline survey for AS patients) was modeled as a
continuous variable. The relationship between time and mean function
was modeled as a restricted cubic spline, permitting nonlinearity. We
used generalized estimating equations (GEEs) with an independence
working covariance matrix to calculate standard errors for the regression
coefficients.
2.4. Adjusted comparisons
To identify racial/ethnic differences in the effect of treatment on
functional outcomes at 1 yr after treatment, we fitted a second set of
longitudinal models with interactions between race/ethnicity and
treatment type and between time since treatment and race, adjusting
for the following baseline factors: time since beginning treatment,
pretreatment function, patient age, comorbidity tumor characteristics
(PSA corrected for 5-a reductase inhibitor use, Gleason score [ 6, 3 + 4,
4 + 3, or 8], and T stage [T1 or T2]), psychosocial measures (educational
attainment, insurance type, employment type, marital status, Short-
Form 36 physical function score, social support, CES-D score, and
participatory decision-making index), receipt of hormone therapy, and
study site. Adjusted models were fitted using the same approach as for
the unadjusted models (GEE), but here the baseline score was used as a
covariate rather than one of the responses. Treatment effects are
characterized by differences in function score at 1 yr after treatment
between treatment groups, and racial differences in treatment effects
were characterized by the difference between races in these treatment
effects. Thus, our estimate of interest is a difference in differences (DID)
accompanied by a 95% confidence interval (CI). In a sensitivity analysis,
we used propensity score regression adjustment as an alternative means
of accounting for systematic pretreatment differences between patients
receiving different treatments. We used a multinomial logistic regres-
sion model to estimate the log odds of receiving each of the three
treatments. The fitted values from this model were then included in a
second version of our main analysis model.
Some regression coefficients had missing values; the most often
missing[6_TD$DIFF] variable contained 5% missing. These values were first imputed
using multiple imputation via predictive mean matching to avoid
casewise deletion of patient records missing any model covariates
. Fifteen cycles of updating imputations were performed to create
one final data set used to fit the analysis models. Because AA men were
more likely to undergo open RP than robotic RP (and similarly were less
likely to receive a nerve-sparing operation and intensity-modulated
radiation therapy [IMRT]), we performed a second sensitivity analysis
to assess the impact of treatment technique on the results by excluding
men who did not have a robotic nerve-sparing operation and those
who did not receive IMRT. All analyses were conducted using R version
3.2.2
3. Results
Among the 2338 CEASAR participants in the analytic cohort,
1835 (79%) were white, 324 (14%) were AA, and 179 (8%)
were Hispanic. presents the distributions of selected
demographic, socioeconomic, and clinical characteristics by
race/ethnicity. In general, white men had a higher level of
educational attainment and were more likely to be married
when compared to AA and Hispanic men. Hispanic and AA
men were more likely to be uninsured or insured by
Medicaid and were more likely to have income of less than
$30 000 per year. AA men were more likely to harbor high-
risk disease according to the D’Amico criteria, and were
more likely to undergo open rather than robotic RP.
Baseline function also varied significantly by race and
ethnicity . AA and Hispanic men reported lower
EPIC domain scores for sexual function at baseline in
comparison to white men, and Hispanic patients reported
lower scores for urinary irritative symptoms and the
urinary incontinence domain. No clinically significant
differences were noted in the baseline domain scores for
bowel function.
3.1. Urinary incontinence
Overall, RP was associated with lower adjusted mean scores
for urinary incontinence when compared to AS and EBRT at
1 yr after therapy. The adjusted mean score for urinary
incontinence at 1 yr was 19.9 points (95% CI 17.2–22.7;
p < 0.001) lower for RP compared to AS and 21.9 points (95%
CI 19.2–24.6; p < 0.001) points lower compared to EBRT.
While this association between RP and incontinence was
observed across all race/ethnic groups, the decline was
greater for AA than for white men (adjusted DID 8.4 points,
95% CI 2.0–14.8; p = 0.01; ). Despite this result,
baseline function and primary treatment appeared to be far
more important in predicting post-treatment urinary
incontinence than race/ethnicity .
Because AA men reported a greater decline in urinary
incontinence function after RP compared to white men, we
tested whether AA men had greater odds of reporting
moderate or severe bother secondary to overall urinary
function compared to white men after RP (Supplementary
Table 2). Notably, there were no apparent between-race
differences in the odds of moderate or severe bother by
overall urinary function, despite lower scores for the
continuous domain (p = 0.15).
3.2. Sexual, bowel, urinary irritative, and hormone function
There was no evidence of any clinically significant
differences by race/ethnicity in the effects of treatment
on EPIC scores for sexual, bowel, or hormone function
) or for bother scores in these domains (Supple-
mentary Table 2). The average difference in effect of RP on
urinary irritative symptoms between white and AA men
was 4.4 (95% CI 0.8–8.0; p = 0.02; ).
3.3. Sensitivity analyses
Because there was evidence of differential adoption of
modern treatment modalities among minority populations,
we performed a sensitivity analysis excluding patients who
did not undergo a robotic nerve-sparing operation and
those who did not receive IMRT. The results of this analysis
were similar to our main analysis with respect to racial
differences in treatment effects. There was, however, an
even greater decline in the post-RP incontinence domain
among AA compared to white men (DID 14.1 points, 95% CI
5.4–22.9; p = 0.002).
E U R O P E A N U R O L O G Y 7 2 ( 2 0 1 7 ) 3 0 7 – 3 1 4 309