OUP user menu

Patient characteristics and hospital quality for colorectal cancer surgery

Wei Zhang, John Z. Ayanian, Alan M. Zaslavsky
DOI: http://dx.doi.org/10.1093/intqhc/mzl047 11-20 First published online: 25 September 2006

Abstract

Objective. To assess associations of patient characteristics with quality-related characteristics of the hospitals where they were treated for colorectal cancer and the role of these associations in disparities in treatment quality affecting vulnerable patient groups or variations across health plans.

Setting. Population-based cancer registry in California.

Participants. A total of 38 237 patients diagnosed with stage I–III (non-metastatic) colorectal cancer in California between 1994 and 1998.

Methods. Registry data were linked with hospital discharge abstracts, US census data, and Medicare enrollment data. The associations of patients’ sociodemographic, clinical, and geographic covariates with treatment at high-volume institutions were assessed with logistic regression. The associations of patients’ covariates with the risk-adjusted 30-day mortality rates of the hospitals where they received surgery were tested with linear regression.

Results. Patients with more advanced tumor stage or more extensive comorbidity, those of Hispanic or Asian race/ethnicity, and those from less affluent communities were less likely to undergo surgery at high-volume institutions and were treated at hospitals with higher risk-adjusted 30-day postoperative mortality rates than those who were less severely ill, white, or more affluent, respectively (all P < 0.05). Black patients also received surgery at hospitals with above-average mortality. Among patients 65 years and older, Medicare managed-care enrollees underwent surgery in higher-volume hospitals than Medicare fee‐for-service enrollees, and there was substantial variation in hospital volume and adjusted hospital mortality among Medicare managed-care plans.

Conclusion. Improving access of sicker, poorer, and minority patients to high-quality hospitals for cancer surgery may improve their outcomes. Further study of processes affecting hospital referral is warranted.

  • case-mix adjustment
  • colorectal neoplasms
  • colorectal surgery
  • delivery of health care
  • hospitals
  • outcome and process assessment
  • quality of health care

Selection of a hospital for colorectal cancer surgery, unlike emergency care for myocardial infarction or stroke, usually allows some time for planning and consideration. Patients’ and physicians’ perceptions of quality may play important roles in hospital referral [1]. Little is known about the effect of patient characteristics on referral, but some studies suggest that older, minority, and low-income patients are more often treated at lower-quality medical institutions [2–4].

Risk-adjusted 30-day mortality rate is a generally accepted indicator of the quality of surgical treatment, reflecting the technical expertise and experience of hospital staff [5,6]. Owing to the difficulty of developing such outcome-based measures, however, many studies of quality have instead relied on structural characteristics such as patient volume, justified by the numerous studies showing better surgical outcomes in high-volume hospitals [2,3,7–12], particularly for treatment of both colon and rectal cancers [13].

Several patient attributes might affect hospital selection, including tumor stage, comorbidities, and socioeconomic status. Managed care plans might direct their enrollees to institutions based on cost or on quality considerations [14–16]. Patients’ tendency to seek care at nearby hospitals [17–19] might mediate some racial variations in quality of care through patterns of residential racial concentration [20].

In this study, we explored the association between patient attributes and the quality of the hospital where they underwent surgery, comparing the conclusions obtained with structural (volume) and outcome (30-day mortality) measures of quality. To assess racial disparities in outcomes of cancer care [21–23], we considered the role of each of these factors as mediators of racial/ethnic disparities. We also investigated variations by insurance type among Medicare patients.

Methods

Data sources

The study used data from the California Cancer Registry on 46 573 California residents diagnosed with stage I, II, or III non-metastatic invasive carcinoma of the colon or rectum from January 1994 through December 1998 who had a hospital admission during which colorectal surgery was performed within 4 months of diagnosis. The California Cancer Registry is the largest population-based cancer registry for a geographically contiguous area in the US, meeting the data quality standards of the National Cancer Institute Surveillance, Epidemiology, and End Results (SEER) Program [24] and the North American Association of Central Cancer Registries [25].

The Registry provided data on patients’ age, gender, race/ethnicity, date of diagnosis, date of surgery, tumor site and stage, residence, vital status, and surgery hospital. It also collected SEER Extent-of-Disease (EOD) codes, which were converted into tumor stage as classified by the American Joint Committee on Cancer using an algorithm developed by the SEER program [26,27].

Registry data were linked to hospital discharge abstracts maintained by the California Office of Statewide Health Planning and Development, using a probabilistic matching algorithm based on patients’ Social Security number, date of birth, sex, and zip code [28,29]. Information on specific surgical procedures and comorbidities was derived from these abstracts. Using patients’ discharge diagnosis codes for admissions from 18 months before to 6 months after their diagnosis with colorectal cancer [30], we calculated a modified Charlson comorbidity index [31]. We linked our data to Medicare enrollment files to identify beneficiaries’ health insurance type (fee-for-service versus managed care) and specific plan. The study protocol was approved by the Institutional Review Boards of the California Department of Health Services, Public Health Institute, Northern California Cancer Center, and Harvard Medical School in accord with assurances filed with and approved by the US Department of Health and Human Services.

Variables and outcome measures

Our patient characteristics fall into five groups. Geography was represented by the 58 counties of patients’ residence. Clinical characteristics included tumor stage (I, II, and III) and site (colon and rectum) categorized into six combinations, and comorbidity (modified Charlson index). Demographic variables were age (categorized as <45, 45–54, 55–64, 65–74, 75–84, and >85 years), gender, and race (white, Black, Hispanic, Asian-American/Pacific Islanders, and other). Patients older than 65 years were identified as Medicare fee-for-service enrollees or as members of a Medicare managed-care plan; indicator variables were defined for three major Medicare plans.

Socioeconomic status was represented by the median household income of persons of the patient’s race/ethnicity in the patient’s census block, based on linked 1990 census data. The 7% of patients who could not be geocoded to a census block group were matched to colorectal cancer patients of the same race/ethnicity in their zip code to obtain multiple imputations of median household income [32].

Hospital volume was defined as the average annual number of patients undergoing cancer-related colorectal surgery in the hospital from 1994 through 1998. Hospitals were categorized into quartiles that contained approximately equal numbers of patients.

Two outcomes were analyzed: (i) admission to a high-volume hospital (the highest quartile) for surgery and (ii) the risk-adjusted 30-day mortality rate of the hospital at which surgery was performed.

Statistical analysis

Descriptive statistics.

We tested the associations of hospital volume quartiles with patient characteristics, with chi-squared tests for nominal categories (gender, race/ethnicity, and health insurance type) and with the Mantel–Haenszel trend test for ordinal categories (age, tumor stage, socioeconomic status quartiles, and comorbidity). Mean unadjusted 30-day mortality rates (measured from the date of surgery) were calculated for each clinical and sociodemographic subgroup and by hospital volume quartile.

Risk-adjusted hospital mortality rates.

We adjusted hospital mortality rates for differences in hospital case-mix. We first fitted a logistic regression model for 30-day survival including severity-related patient covariates (age, gender, race, tumor stage by site, comorbidity, and income) and hospital dummies. We then calculated the predicted mortality rate for each hospital if every patient in our sample had been treated at that hospital and tabulated these mean adjusted 30-day hospital mortality rates by patient characteristics.

Multivariate analysis of referral to high-volume and low-mortality hospitals.

Logistic regression models assessed the extent to which patient demographic, socioeconomic, clinical, payer, and geographic covariates predicted surgery in hospitals of the highest-volume quartile; we tested the overall significance of county effects by the likelihood ratio test. Similarly, we regressed adjusted hospital-specific mortality rates on sociodemographic, clinical, and payer covariates and county-of-residence indicator variables, thus assessing how patient covariates jointly predicted use of hospitals with different levels of adjusted mortality for surgery.

Differences related to health insurance types.

For Medicare patients 65 years and older, we calculated the rate of admission to a high-volume hospital and the mean adjusted hospital mortality rate, each by insurance type and among the largest health plans.

Mediators of differences by race.

We explored the between-hospital racial effect on patient outcome by using our models to predict mean 30-day mortality rates of hospitals where white, Black, Hispanic, and Asian patients received treatment through the use of (i) county indicator variables of patient residence only; (ii) clinical covariates only; (iii) socioeconomic covariates only; and (iv) race covariates only (the residual racial effect). With each set of predictions, the differences of predicted mortality rates between white and other race groups were then summarized to demonstrate the extent of racial differences in outcome mediated by hospital selection owing to the included covariates. Finally, we compared the contribution of each effect with the total race difference, which is the sum of cross-race differences in hospital mortality rates due to all four groups of covariates.

Results

Cohort description

Of the original cohort of 46 573 colorectal cancer patients, we excluded 8336 individuals who could not be linked to a hospital discharge abstract or Medicare enrollment data, leaving 38 237 patients (82.1% of the original sample), whose registry data were linked with hospital discharge abstracts, US census data, and Medicare enrollment data (for those at least 65 years old). Female patients (83.2 versus 81.0%, P < 0.0001), patients older than 65 (83.3 versus 79.0%, P < 0.0001), and patients with more advanced tumor stage (85.5% for stage III, 82.9% for stage II, and 77.4% for stage I, P < 0.0001) were more likely to appear in the final linked data set.

At the 383 hospitals where colorectal cancer surgery was performed, the number of operations performed ranged from 1 to 204 per year, with a patient-weighted median of 67 cases per year (interquartile range 35–91). The 29 hospitals in the highest-volume quartile had a lower mean crude 30-day mortality rate (2.4%) than hospitals in the lowest-volume quartile (4.6%, P < 0.01).

The use of high-volume hospitals varied greatly across counties of patient residence. In 3 of the 58 counties in California, more than half of colorectal cancer patients were treated in high-volume hospitals, as were between 25 and 50% of patients in 9 other counties; conversely, in 31 counties fewer than 5% of patients went to high-volume hospitals for surgery.

Sociodemographic and clinical characteristics of the 38 237 linked patients by hospital volume quartiles appear in Table 1. Patients were more likely than average to have surgery in high-volume hospitals if they were male, were younger than 65, had no comorbidities, had a stage I tumor, or were from a high-income neighborhood. Compared with white patients, Asian-American and Hispanic patients were less likely and Black patients more likely to go to high-volume hospitals for surgery. Patients older than 65 years in managed care plans were more likely to have surgery in high-volume hospitals than those in Medicare fee-for-service plans and were much less likely to have surgery in hospitals of the lowest-volume quartile.

View this table:
Table 1

Hospital volumes, unadjusted mortality rates, and adjusted hospital mortality rates after colorectal cancer surgery, by patient characteristics

nHospital volume quartile (number of patients per year)P-valueUnadjusted 30-day individual mortality (%)P-valueMean adjusted 30-day hospital mortality (%)P-value
Quartile 1 (<35)Quartile 2 (35–67)Quartile 3 (68–91)Quartile 4 (>91)
n 38 2379593 (25.1)9574 (25.0)9669 (25.3)9401 (24.6)
Gender0.010.100.12
    Male19 19624.525.425.224.93.43.2
    Female19 04125.724.625.424.23.13.2
Age (years)<0.0001<0.0010.01
    <45143625.822.125.826.30.73.1
    45–54302724.323.825.226.70.73.0
    55–64605624.624.223.927.31.13.1
    65–7411 69224.225.725.224.91.93.2
    75–8411 84625.825.225.523.54.63.3
    >84418026.626.026.920.69.43.4
Race<0.0001    0.010.01
    White29 01823.525.026.125.33.53.2
    Black224025.524.121.129.33.13.5
    Hispanic384234.222.824.318.72.73.5
    Asian297828.428.621.521.52.33.2
    Others15925.823.323.927.03.83.2
Stage<0.00011<0.00120.10
    I (colon)701723.925.325.225.72.23.2
    II (colon)12 30426.724.325.323.64.23.2
    III (colon)906625.325.124.724.93.73.3
    I (rectum)353322.125.227.025.72.03.2
    II (rectum)316225.525.225.623.73.43.2
    III (rectum)315523.726.824.824.72.53.3
Comorbidity<0.0001<0.0010.01
    018 54023.024.426.426.21.43.1
    1814926.325.024.324.33.93.3
    ≥211 54827.626.124.222.15.93.3
Payer (among patients of age >65)    0.001<0.0010.05
    Medicare fee-for-service12 17326.825.624.623.04.53.3
    Medicare managed care643717.326.029.926.73.83.2
Census income<0.0001<0.0010.01
    ≤$25 00010 81032.025.223.419.44.03.4
    $25 000 – $35 000887326.724.824.723.83.73.3
    $35 000 – $50 00010 19322.326.225.426.12.93.3
    >$50 000836117.923.728.130.32.33.0
  • 1 P = 0.0002 among colon cancer stages; P = 0.01 among rectal cancer stages.

  • 2 P < 0.0001 among colon cancer stages; P = 0.001 among rectal cancer stages.

Compared with overall 30-day postoperative mortality rate of 3.3%, unadjusted mean mortality rates (Table 1, middle columns) were significantly higher for patients who were older than 65, white, had Charlson score >2, or were from neighborhoods in the lowest income quartile.

Effects of patient characteristics on mortality within hospital

In multivariate logistic regression models with hospital indicator variables to control for between-hospital variation (Table 2), the odds of 30-day mortality increased with age, more advanced tumor stage, and more severe comorbidity. Higher socioeconomic status and female sex were associated with lower 30-day mortality rates. Racial differences in mortality were small and insignificant, after controlling for other characteristics. To insure adequate case-mix adjustment of hospital mortality rates, we retained all of the relevant variables in the model regardless of statistical significance.

View this table:
Table 2

Individual-level predictors of mortality, controlling for hospital effects (adjustment model)

Patients’ characteristicsOR 30-day mortality (95% CI)
GenderP < 0.0001
    Male1.0
    Female0.76 (0.67–0.85)
Age (years)P < 0.0001
    <450.63 (0.33–1.23)
    45–540.60 (0.37–0.97)
    55–640.79 (0.57–1.08)
    65–741.0
    75–842.24 (1.90–2.64)
    >844.69 (3.92–5.62)
RaceP = 0.50
    White1.0
    Black0.86 (0.64–1.16)
    Hispanic0.86 (0.68–1.07)
    Asian0.87 (0.65–1.17)
Stage
    ColonP < 0.0001
        I1.0
        II1.82 (1.51–2.20)
        III1.87 (1.53–2.28)
    RectumP = 0.003
        I1.03 (0.77–1.39)
        II1.78 (1.37–2.31)
        III1.54 (1.15–2.04)
ComorbidityP < 0.0001
    01.0
    12.09 (1.76–2.48)
    ≥22.84 (2.44–3.31)
Census incomeP = 0.01
    ≤$25 0001.0
    $25 000 – $35 0000.94 (0.80–1.11)
    $35 000 – $50 0000.80 (0.67–0.95)
    >$50 0000.75 (0.62–0.92)
  • CI, confidence interval; OR, odds ratio.

Risk-adjusted hospital mortality

After controlling for severity-related covariates, we calculated the mean adjusted postoperative 30-day mortality rates for each hospital and reported mean adjusted mortality for hospitals averaged by patient subgroup (Table 1, final columns). Patients from lower-income neighborhoods and those of Black and Hispanic ethnicity obtained care from hospitals with mean adjusted 30-day mortality greater than the 3.2% overall rate, as did patients 85 years and older and those with more severe comorbidity.

After adjustment, high-volume hospitals had significantly lower mean 30-day adjusted mortality rates than low-volume hospitals (2.6, 2.7, 3.4, 4.2% by quartile from highest-to-lowest volume, P < 0.01).

Multivariate analysis of hospital referral

In the multivariate model that controlled for both county and other patient covariates, the odds of undergoing colorectal surgery in a high-volume hospital increased with higher socioeconomic status or better health status (Table 3). Patients with more advanced tumor stage were significantly less likely than those with stage I tumors to have their surgeries performed in a high-volume institution, as were patients with more severe comorbidity. Compared with white patients, Black patients were significantly more likely to undergo surgery in a high-volume hospital. For patients 65 years and older, the odds of admission to a high-volume hospital decreased as patient age increased and did not differ significantly between enrollees in Medicare fee-for-service plans and those in Medicare managed-care plans. Finally, the county effect was highly significant (P < 0.001), reflecting geographical variation in use of high-volume hospitals. Nonetheless, although not every county contained a high-volume hospital, at least some patients from every county received care in one of those hospitals.

View this table:
Table 3

Multivariable models for having surgery in highest-volume hospitals and for adjusted 30-day hospital mortality rate

Odds ratio for surgery in a high-volume hospital (95% CI)Adjusted differences in adjusted 30-day hospital mortality rate (percentage points) (P-value)
GenderP = 0.16P = 0.003
    Male10
    Female0.96 (0.92–1.02)0.04 (0.08)
Age (years)P < 0.0001P < 0.0001
    <450.75 (0.65–0.88)−0.08 (0.19)
    45–540.79 (0.70–0.89)−0.16 (0.001)
    55–640.83 (0.75–0.91)−0.08 (0.04)
    65–7410
    75–840.90 (0.85–0.96)0.06 (0.03)
    >840.76 (0.69–0.84)0.12 (0.002)
RaceP < 0.0001P < 0.0001
    White10
    Black1.24 (1.11–1.38)0.22 (<0.0001)
    Hispanic0.67 (0.61–0.74)0.20 (<0.0001)
    Asian0.67 (0.61–0.74)0.18 (<0.0001)
StageP = 0.0003 and 0.271P = 0.14 and 0.292
    I (colon)10
    II (colon)0.86 (0.80–0.93)0.04 (0.18)
    III (colon)0.88 (0.81–0.95)0.06 (0.05)
    I (rectum)0.94 (0.85–1.04)0.07 (0.09)
    II (rectum)0.85 (0.77–0.95)0.03 (0.49)
    III (rectum)0.90 (0.81–1.00)0.11 (0.01)
ComorbidityP < 0.0001P < 0.0001
    010
    10.94 (0.88–1.00)0.08 (0.003)
    ≥20.84 (0.79–0.89)0.12 (<0.0001)
PayerP < 0.001P = 0.02
    Medicare FFS10
    Medicare MCO1.01 (0.94–1.10)−0.09 (0.005)
    Others (≥65)0.99 (0.92–1.06)−0.01 (0.80)
    MCO (<65)1.83 (1.66–2.01)−0.09 (0.04)
Census incomeP < 0.0001P < 0.0001
    ≤$25 00010
    $25 000 – $35 0001.21 (1.12–1.31)−0.03 (0.28)
    $35 000 – $50 0001.19 (1.10–1.28)−0.02 (0.52)
    >$50 0001.37 (1.27–1.49)−0.24 (<0.0001)
Overall county effectP < 0.0013P < 0.014
  • 1 P = 0.0003 for colon, 0.27 for rectal.

  • 2 P = 0.14 for colon, 0.29 for rectal.

  • 3 Likelihood ratio test.

  • 4 F-test.

In multivariate analyses (Table 3), variables that predicted having surgery in a hospital with higher adjusted mortality (in descending order of effect magnitude) included residence in a neighborhood with less than $50 000 median income, minority race/ethnic group, more severe comorbidity, more advanced tumor stage (III), and very old age. Enrollment in a Medicare managed-care plan was associated with undergoing surgery in hospitals with lower adjusted 30-day mortality. Finally, the effects of county of residence were significant (P < 0.01), demonstrating the importance of geographical variation in hospital mortality rates.

Medicare health plans and hospital referral

We compared hospital assignments by health plan for 19 879 patients who were at least 65 years old with Medicare enrollment data (Table 4). Managed-care plan enrollees were more likely than fee-for-service enrollees to have surgery in high-volume hospitals, but mean adjusted 30-day hospital mortality rates were similar for hospitals used by the two groups. Patterns of hospital assignment varied significantly by volume and by 30-day mean mortality across the three largest Medicare managed-care plans (accounting for 74% of Medicare managed-care enrollment).

View this table:
Table 4

Hospital selections by health plan types for patients at least 65 years old

Distribution (%) by hospital volume quartileMean adjusted 30-day hospital mortality (%)
Quartile 1 (lowest)Quartile 2Quartile 3Quartile 4 (highest)
Fee-for-service129.426.224.220.23.3
Health plan218.624.127.429.93.3
    A2.816.022.958.42.62
    B31.032.328.28.64.12
    C22.022.525.929.63.22
    Others22.328.633.415.73.52
  • 1 Significant difference between fee-for-service and managed care health plans as a whole for volume (P < 0.0001) but not for mean mortality rates.

  • 2 Significant differences among plans (P < 0.0001) for volume distribution and mean mortality rates.

Direct and mediated effects of race on volume and 30-day mortality of chosen hospital

To summarize effects on racial disparities in outcomes of differences in hospital referral, we present the predicted race-specific mean adjusted 30-day hospital rates in Table 5. Predicting hospital selection only by county of residence, the mean adjusted 30-day mortality rates of hospitals where Black and Hispanic patients were treated would be higher than the corresponding mean for whites by 0.07 and 0.11% (respectively) and would be lower for Asian patients by 0.11%. Clinical or socioeconomic covariates accounted for only small differences in mean mortality rates of hospitals used by patients of different races/ethnicities. With only race variables in the model, Black, Hispanic, and Asian patients were treated at hospitals with higher predicted mean mortality rates than those of hospitals treating white patients by 0.22, 0.20, and 0.19%, respectively. Thus, racial disparities in use of low-mortality hospitals were mostly attributable to the residual race effect itself, not to racial differences in clinical or socioeconomic status or geographical distribution. Similarly, although Blacks and Asians were concentrated in counties with greater rates of admission to high-volume hospitals, most of the racial differences in admission to these institutions were attributable to residual race effects (positive for Blacks and negative for Hispanics and Asians).

View this table:
Table 5

Analysis of race effect on hospital mortality rate and admission to high-volume hospitals

Hospital mortality rateAdmission to high-volume hospital
BlackHispanicAsianBlackHispanicAsian
Geographic covariates only0.070.11−0.111.4−0.31.8
Clinical covariates only−0.03−0.04−0.04−0.2−0.10.1
Socioeconomic covariates only0.030.03−0.02−0.7−0.20.7
Residual race effect only0.220.200.193.5−5.9−6.5
Total difference from whites0.310.310.024.0−6.6−3.9
  • Entries represent differences (in percentage points) in mean mortality rate of hospitals chosen by patients of each race compared with that for hospitals chosen by whites, or in probability of admission to a high-volume hospital, predicted by the mean values for patients of that race on the corresponding set of covariates.

Discussion

This study focused on the relationship of patient characteristics with those of hospitals where they obtained colorectal cancer surgery, finding associations of clinical status, age, race, and socioeconomic status with use of high-volume and low-mortality hospitals. The largest effects of patient characteristics on mean hospital 30-day mortality rates were about 0.24 percentage points, which appears small but is appreciable compared with overall 30-day mortality (3.3%). Effects of this relative magnitude are of concern if projected to the groups with higher overall mortality (such as the oldest patients) or if the effects of hospital quality persist beyond the first month of survival.

Despite their critical needs, patients with worse clinical status (especially more severe comorbidity) were less likely to obtain surgery at hospitals with high-volume or low adjusted mortality rates. This may be due to sicker patients’ more urgent need for surgery, their reluctance to travel long distances, or their long-standing relationships with local health care practitioners.

Among Medicare beneficiaries, managed care members were more likely than fee-for-service beneficiaries to enter hospitals with high patient volume, but adjusted mortality rates were similar for hospitals used by the two groups. These contrasting findings highlight the importance of measuring outcomes to assess quality whenever possible, rather than relying solely on structural measures. The various large Medicare health plans sent their patients to hospitals with very different distributions of patient volume and adjusted mortality, presumably due to their distinctive hospital contracting and referral strategies and possibly differences in geographical distribution.

To better understand racial disparities in treatment, we explored the sources of racial variation in hospital referral. Most of the racial variation in outcomes within a hospital stems from patients’ clinical status. The remaining small and statistically insignificant within-hospital differences in outcome by race/ethnicity might be explained by incomplete adjustment for clinical factors. Conversely, racial differences in hospital selection were substantial and were mostly not mediated by geographic, clinical, or socioeconomic effects. Thus, there are still racial barriers to high-quality surgical care for colorectal cancer patients, independent of other patient characteristics. Furthermore, although Black patients are concentrated in dense urban areas and are more likely than whites to obtain care at high-volume hospitals, the high-volume hospitals at which they obtain care are not necessarily the ones that obtain superior outcomes.

Some patients may prefer treatment in a hospital with a concentration of providers or patients from their race/ethnic group. In the county with the highest percentage of Asian patients (31%), for example, one high-volume hospital served almost exclusively (>99%) Asian patients. Other patients may have limited ability to travel to hospitals further from their homes, or insurance and cost considerations may affect their access to some hospitals.

A major strength of this study is its focus on the role of hospital effects on outcomes through a measure of adjusted hospital mortality rates. In patient-level analyses, such effects are confounded with the effects of clinical characteristics and social supports that cause variations in outcomes within each hospital. For example, Black patients have lower unadjusted mortality than whites but are treated in hospitals with higher adjusted mortality rates. Other strengths of this study include its large, population-based cohort, the availability of clinical, sociodemographic, and geographic information for adjustment, and relatively precise race codes. The California cohort represents a diverse population of 33 million people and is not limited to Medicare beneficiaries. We studied moderate-risk procedures that typically allow some time between diagnosis and surgery (median = 6 days) during which the treating hospital could be selected. Thus, our work extends findings of previous studies that focused on rural patients or patients requiring high-risk procedures or urgent care.

This study had several limitations. First, we were able to link only 82% of the eligible subjects identified by the registry to the auxiliary data sets. Second, some colorectal cancer patients switched hospitals before surgery, had surgeries in multiple hospitals, or were treated in out-of-state hospitals. Third, stage and comorbidity explain only part of the variation in clinical severity, so additional measures of patient’s clinical status might have improved our case-mix adjustment. Although we controlled for fixed county effects, our study did not consider systematic associations of county characteristics with hospital quality or differences in geographic access within counties. Finally, although our focus on 30-day mortality was appropriate to assessing quality of surgery, we did not assess longer-term survival, which is likely to be affected by non-hospital care such as adjuvant therapy, surveillance, and supportive services.

This study demonstrated disparities in hospital selection across racial groups and across clinical and socioeconomic strata. Many patients of severe clinical status were treated in hospitals of lower quality. Although clinical factors remain the most important determinant of outcomes, hospital referral is one of the most important mediators of outcomes that is modifiable by the health care system. Further research is needed to assess the roles of patient access factors and preferences, health care providers, and health plans in referrals for cancer surgery [33].

Acknowledgements

The authors acknowledge the major contributions of Mark Allen of the California Cancer Registry to the development of linked data sets and of Robert Wolf of Harvard Medical School to data management and programming, and insightful comments from William Wright of the California Cancer Registry. This study was supported by grants R01 HS09869 and U01 CA93324 from the Agency for Healthcare Research and Quality and the National Cancer Institute.

The collection of cancer incidence data used in this study was supported by the California Department of Health Services as part of the statewide cancer reporting program mandated by California Health and Safety Code Section 103885; the National Cancer Institute’s Surveillance, Epidemiology and End Results Program under contract N01-PC-35136 awarded to the Northern California Cancer Center, contract N01-PC-35139 awarded to the University of Southern California, and contract N02-PC-15105 awarded to the Public Health Institute; and the Centers for Disease Control and Prevention’s National Program of Cancer Registries, under agreement #U55/CCR921930-02 awarded to the Public Health Institute. The ideas and opinions expressed herein are those of the authors, and endorsement by the State of California, Department of Health Services, the National Cancer Institute, and the Centers for Disease Control and Prevention or their contractors and subcontractors is not intended nor should be inferred.

References

View Abstract