EURURO Vol. 72 No. 3

where S stands for the standard error of the observed outcome

[35_TD$DIFF]

O, taking

the model-based expected outcome E to be a fixed quantity, and

calculated

-values from the standard normal distribution.

For each QI, we classified outlier hospitals performing worse than

expected as poor outliers and those performing better than expected as

superior outliers. We experimented with adjusting for multiple testing

through Bonferroni correction and control of false discovery rate

[15]

but chose to use the fixed

-value threshold of 0.05 for the purpose of

classifying outlier hospitals for further analyses as previously reported

[10,12]

. Additionally, for data representation, we multiplied the

observed to expected outcomes ratio for each hospital by the national

average to determine the case-mix adjusted QI

[13_TD$DIFF]

[14]

. Internal consistency

of the resulting outlier classifications was investigated via Venn

diagrams, pairwise Kendall’s tau correlations, and Cronbach

statistic.

Patients treated in outlier versus nonoutlier hospitals were compared by

30-d mortality, 90-d mortality (logistic regression), and overall mortality

(Cox regression). Mortality models were fitted using generalized

estimating equations allowing for within-hospital correlation, both

with and without case-mix adjustment. We calculated a composite

quality measure encompassing hospital performance across the five QIs,

hereafter referred to as the Renal Cancer Quality Score (RC-QS). Briefly,

for each QI a hospital

[14_TD$DIFF]

identified as a superior outlier

[15_TD$DIFF]

received one point,

for being a poor outlier one point was deducted, and for being a

nonoutlier zero points were awarded. The final RC-QS was a summation

of the points received across each indicator for an individual hospital. We

investigated the association of the RC-QS to outcomes using similar

regression models as described above, as well as to hospital location,

volume, and type of institution. Statistical

-values for the hospital

characteristics associations were calculated from chi-square (location)

and two-sample Wilcoxon tests (volume and type), comparing hospitals

with positive sum scores to hospitals with negative sum scores, omitting

zero scores. The statistical analyses were performed in SAS software

version 9.3 (SAS Institute, NC, USA) and R statistical environment version

3.2.2 (R Foundation for Statistical Computing, Vienna, Austria).

Results

Variations in the quality of RCC surgical care were captured

across five QIs: MIS, PN, PM, LOS, and RP. The number of

patients analyzed for each QI and their corresponding

comorbidities, tumor-, and treatment-related characteris-

tics are summarized in

Table 1 .

Statistically significant

interhospital variation was observed for all QIs (

0.001)

with random effects models indicating between hospital

variance proportions of 31%, 17%, 12%, 15%, and 20% for the

MIS, PN, PM, LOS, and RP indicators, respectively. All of the

random effects remained significant when adjusting for

case-mix (

0.001). For each QI, more than 1100 hospitals

were benchmarked for quality performance against the

national average utilizing our case-mix adjusted QIs. A total

Table 1 – Study cohort characteristics

MIS

LOS

[31_TD$DIFF]

Total (%)

34 150

87 408

71 422

126 289

131 319

Median age, yr (IQR)

62 (53–70)

60 (51–69)

59 (50–68)

62 (53–71)

Sex

Male

20 740 (61)

51 833 (59)

43 639 (61)

78 579 (62)

81 804 (62)

Charlson/Deyo Score

23 357 (68)

61 512 (70)

51 153 (72)

87 665 (69)

91 613 (70)

[32_TD$DIFF]

7834 (23)

19 487 (22)

15 748 (22)

28 657 (23)

29 520 (22)

[33_TD$DIFF]

2959 (9)

6409 (8)

4521 (6)

9967 (8)

10 186 (8)

T-Stage

26 050 (76)

87 408 (T1a: 100)

71 422 (100)

69 134 (55)

71 517 (54)

8100 (24)

20 994 (17)

21 803 (17)

34 253 (27)

35 962 (27)

1908 (1)

2037 (2)

Lymph node status

29951 (88)

83341 (95)

68976 (96)

102245 (81)

106061 (81)

3878 (11)

3935 (4)

2389 (3)

19149 (15)

20036 (15)

321 (1)

132 (1)

57 (1)

4895 (4)

5222 (4)

Metastases

117 121 (93)

122 240 (92)

9168 (7)

9990 (8)

Histology

Clear cell

[5_TD$DIFF]

27 010 (79)

68 277 (78)

54 022 (76)

103 320 (82)

107 555 (82)

Papillary

[6_TD$DIFF]

4548 (13)

13 419 (15)

12 326 (17)

13 104 (10)

13 549 (10)

Chromophobe

[7_TD$DIFF]

2195 (6)

4618 (5)

4214 (6)

6467 (5)

6643 (5)

Other

[8_TD$DIFF]

397 (2)

1094 (1)

860 (1)

3398 (3)

3572 (3)

Median tumor size, cm (IQR)

5.0 (3.5–7.0)

2.6 (2.0–3.4)

2.6 (2.0–3.6)

5.5 (3.7–8.0)

5.5 (3.8–8.1)

Tumor grade

G1/2

18 562 (54)

59 073 (67)

47 090 (66)

65 746 (52)

67 846 (52)

G3/4

8807 (26)

14 604 (17)

12 367 (17)

43 125 (34)

44 855 (34)

6855 (20)

13 731 (16)

11 965 (17)

17 418 (14)

18 618 (14)

Median yr of diagnosis (IQR)

2011 (2010–2012)

2009 (2007–2012)

2010 (2007–2012)

2008 (2006–2011)

No. hospitals assessed

1155

1207

1131

1245

1254

[4_TD$DIFF]

Denotes number of patients included in the analysis for each quality indicator.

IQR = interquartile range; LOS = length of stay; MIS = T1-2 tumors receiving a minimally invasive (laparoscopic or robotic) approach for radical nephrectomy;

NA = not applicable; PM = positive surgical margin following PN for T1 tumors; PN = T1a tumors undergoing partial nephrectomy.

E U R O P E A N U R O L O G Y 7 2 ( 2 0 1 7 ) 3 7 9 – 3 8 6

381