Socioeconomic Disparities in Mortality after Cancer Surgery: The Role of Failure to Rescue

Abstract

Importance

Disparities in operative mortality due to socioeconomic status have been consistently demonstrated, but the mechanisms underlying this disparity are not well understood.

Objective

To determine whether variations in failure to rescue (FTR) contribute to socioeconomic disparities in mortality following major cancer surgery.

Design, Setting and Exposure

A retrospective cohort study using the Medicare Medical Provider Analysis and Review (MedPAR) and Denominator files. A summary measure of socioeconomic status (SES) was created for each US ZIP code using Census data linked to residence. Multivariable logistic regression was used to examine the influence of SES on rates of FTR, and fixed-effects hierarchical regression was used to evaluate the extent to which disparities could be attributed to differences between hospitals.

Participants

All patients undergoing esophagectomy, pancreatectomy, partial or total gastrectomy, colectomy, lobectomy or pneumonectomy, and cystectomy for cancer during the years 2003 to 2007 (N=596,222)

Main Outcome Measures

Operative mortality, post-operative complications, and failure to rescue (case-fatality following one or more major complications).

Results

Patients in the lowest quintile of SES had mildly increased rates of complications (25.6% in the lowest quintile vs. 23.8% in the highest quintile, p<0.01), a larger increase in mortality (10.2% vs. 7.7%, p<0.001), and the greatest increase in rates of FTR (26.7% vs. 23.2%,p<0.01). Analysis of hospitals revealed a higher FTR rate for all patients (regardless of SES) at hospitals treating the largest proportion of low SES patients. Adjusted odds of FTR according to SES ranged from 1.04 [0.95 – 1.14] for gastrectomy, to 1.45 [1.21 – 1.73] for pancreatectomy. Additional adjustment for hospital effect nearly eliminated the disparity observed in FTR across levels of SES.

Conclusions

Patients in the lowest quintile of SES have significantly increased rates of FTR. This appears to be, at least in part, a function of the hospital where low SES patients are treated. Future efforts to ameliorate socioeconomic disparities should concentrate on hospital processes and characteristics that contribute to successful rescue.

INTRODUCTION

Disparities in post-operative mortality based on socioeconomic status (SES) have been consistently demonstrated following major cancer surgery. Low SES patients undergoing gastrectomy are 55% more likely to die following surgery compared to those with higher SES, and operative mortality following lung resection is 37% higher in low-income patients.^{1, 2} While some authors have posited that patient characteristics account for a portion of these differences,³ other evidence suggests that hospital quality plays an important role in the socioeconomic variations observed in mortality.¹

The hospital mechanisms that contribute to increased mortality rates at centers that disproportionately treat patients of low SES remain poorly understood. While it has long been assumed that increased rates of mortality are a consequence of higher rates of complications, more recent studies of mortality variations following major surgery have challenged this notion. Instead, they assert that the timely recognition and treatment of complications once they occur may be a larger concern. This idea, first described by Silber and colleagues,⁴ is termed "failure to rescue," as it signifies the inability to rescue a patient from death following a major complication. This notion has become an increasingly important concept in the current understanding of mortality variation, as it explains a large portion of the variation in mortality rates between hospitals.⁵

The objective of this study is to examine whether failure to rescue helps explain socioeconomic disparities in mortality rates following cancer surgery in Medicare patients who underwent one of six major cancer operations. For this analysis, the exposure variable SES was defined by a summary measure which links US census data (income, education, and employment) to ZIP code of residence, and multivariable logistic regression was used to examine its influence on rates of failure to rescue. Insight into hospital level mechanisms, such as failure to rescue, that contribute to the inequalities seen in this subset of surgery patients, could have significant implications for future health policy aimed at reducing variations in mortality following major cancer surgery.

METHODS

Patients and Databases

We used data from the Medicare Provider Analysis and Review (MEDPAR) file, which includes inpatient claim file data from the national Medicare database for the years 2003–2007. These files contain hospital discharge records for fee-for-service, acute care hospitalizations of all Medicare recipients. We used the Medicare denominator file to determine the vital status of patients 30 days after surgery. Medicare patients enrolled in managed care plans were not included in this analysis, as they do not appear in the MEDPAR files. We excluded Medicare patients under the age of 65 and over the age of 99. The Institutional Review Board of the University of Michigan and the CMS approved this protocol and waived the requirement for informed consent.

Using appropriate International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnosis and procedure codes, we identified all patients with a corresponding cancer diagnosis undergoing one of six operations during the study period: esophagectomy (procedures 43.99, 42.40, 42.41, 42.42; diagnoses 150–150.9; N=14,562), pancreatectomy (52.51, 52.53, 52.6, 52.7; 152–152.9, 156–157.9; N=15,239), partial or total gastrectomy (43.5–43.99; 151–151.9; N=39,584), colectomy (45.73 – 45.76; 153–153.9, 154; N=423,474), lobectomy or pneumonectomy (32.4, 32.5; 162–165.9; N=80,395), and cystectomy (57.7–57.79; 188–189.9; N=22,968). These particular procedures were selected because they are complex high-risk operations, and they represent a broad range of cancer diagnoses and specialties. In addition, each operation has a risk of mortality high enough to ensure the necessary power to evaluate potential causative factors influencing variations in surgical mortality by SES. Patient demographic data, as well as admission acuity (emergent, urgent, elective), were determined from the MEDPAR record. Other types of admissions (trauma, newborn, etc.) were excluded. Comorbid conditions were defined by the Elixhauser method, which uses ICD-9-CM codes to classify secondary diagnoses on the MEDPAR record into 30 different comorbid conditions.⁶

Socioeconomic Status

We constructed a summary measure of SES for each US ZIP code, using data on income, education, and occupation from the 2000 US Census linked to the patient’s ZIP code of residence in the Medicare files. Individual variables chosen (Table 1) and methods for calculating the summary measure were based on previously developed methods.⁷ In brief, for each ZIP code, a z-score was estimated for each variable by subtracting the overall mean and dividing by the standard deviation. The median SES summary score was created by summing the z-scores of all 6 variables. These scores ranged from −20 to 20; larger scores indicate greater socioeconomic disparity. Patients were sorted according to SES summary score and grouped into quintiles. Quintiles were used to evaluate SES as they allow for accurate modeling of a dose-response relationship -- regardless of the shape of that relationship.⁸ Hospitals were also sorted according to the average summary SES score of their patient populations to form quintiles that would identify hospitals that predominantly treat patients of either low or high SES. To confirm the significance of the relationship between SES and FTR, similar models incorporating SES as a continuous variable were evaluated separately.

Table 1.

Patient characteristics by SES quintile

Patient Characteristics	Quintiles of SES					p value
	Lowest	Low	Middle	High	Highest
Age (median)	75.0	76.0	76.0	76.0	76.0	p =0.09
Gender (% female)	54.6	54.5	54.6	54.6	54.9	p =0.52
Race (% non-white)	24.6	10.9	7.5	7.2	7.2	p <.001
≥3 comorbidities	20.1	18.4	17.5	16.5	14.6	p <.001
Median SES summary score (range)	−4 (−10 – −2)	−1 (-2 – 0)	1 (0 – 2)	4 (2 – 6)	8 (6 – 17)
Wealth/Income
Median household income ($)	28603	35082	40024	47928	64622	p <.001
Median value of housing units ($)	64500	83800	98700	132000	205100	p <.001
Households with interest, dividend, or rental income (%)	21.5	31.4	37.6	42.5	53.2	p <.001
Education
Adult residents who completed ≥ high school (%)	67.8	77.9	83.0	87.5	92.7	p <.001
Adult residents who completed ≥ college (%)	14.5	20.2	25.3	34.2	51.2	p <.001
Employment
Employed residents with management, professional, or related occupations	21.4	25.5	29.0	35.9	48.7	p <.001

(SES: socioeconomic status).

Primary Outcomes and Statistical Analysis

Primary outcomes for this study included operative mortality, post-operative complications, and failure to rescue. Operative mortality was defined as death within 30 days of the index procedure or before hospital discharge. Post-operative complications were classified as medical or surgical, and were identified by ICD-9-CM codes using previously validated methods.^{9, 10} Medical post-operative complications included: myocardial infarction (410.00–410.91), pulmonary failure (518.81. 518.4, 518.5, 518.8), pneumonia (481, 482.0–482.9, 483, 484, 485, 507.0), acute renal failure (584), and venous thromboembolism (415.1, 451.11, 451.19, 451.2, 451.81, 453.8). Surgical complications included: post-operative hemorrhage (998.1), surgical site infection (958.3, 998.3, 998.5, 998.59, 998.51), and gastrointestinal hemorrhage (530.82, 531.00–531.21, 531.40, 531.41, 531.60, 531.61, 532.00–532.21, 532.40, 532.41, 532.60, 532.61, 533.00–533.21, 533.40, 533.41, 533.60, 533.61, 534.00–534.21, 534.40, 534.41, 534.60, 534.61, 535.01, 535.11, 535.21, 535.31, 535.41, 535.51, 535.61, 578.9).

Failure to rescue was classified as a case-fatality among patients with one or more of the defined major complications. Crude rates of the primary outcome measures were determined for each SES quintile. We first used multivariable logistic regression to determine risk-adjusted rates of our three primary outcome measures. We then used unique hospital identifiers in fixed-effects hierarchical regression models to assess the relationship between SES and failure to rescue following cancer surgery, adjusting for patient characteristics and the hospital at which patients received their care. Given that the datasets used include all Medicare patients treated during the study period, a fixed-effects model was chosen to determine population estimates. This modeling adjusts for the hospital-level effects that influence failure to rescue of all patients, in order to isolate the socioeconomic differences in failure to rescue within hospitals.¹¹ If the socioeconomic disparity in failure to rescue is no longer significant after accounting for hospital-level variation, the differences in mortality can be assumed to arise from differences between hospitals. Patients with missing data were excluded, as the total amount of missing data was very small (less than 0.019 percent), and it was assumed to be missing at random. Final risk-adjustment models had C-statistics ranging from 0.75 (pancreatectomy) to 0.83 (colectomy). All analyses were performed using STATA version 11 (StataCorp, College Station, TX), with two-sided tests and alpha set at 0.05.

RESULTS

A total of 596,222 patients underwent one of the six cancer operations during the five-year study period. The demographic and socioeconomic characteristics of included patients are shown in Table 1. The median age of all patients in the study population was 76 years old, and there were slightly more females than males. The percentage of minority patients in each quintile decreased as SES score increased. The lowest quintile of SES included 24.6% non-white patients, while the highest quintile included only 14.6% non-white patients.

Rates of major complications, mortality, and failure to rescue for all operations combined across patient quintiles of SES are shown in Figure 1. While major complications from all operations were mildly different across socioeconomic levels (25.6% in the lowest quintile vs. 23.8% in the highest quintile, p<0.01), greater disparities were seen in operative mortality across SES quintiles. Patients in the lowest SES quintile had 1.3-fold greater odds of experiencing perioperative mortality than patients in the highest (10.2% vs. 7.7%, p<0.001). Similar disparities in SES were observed for rates of failure to rescue (26.7% in the lowest quintile vs. 23.2% in the highest quintile, p<0.01).

Figure 1.

Rates of mortality, failure to rescue, and major complications for all operations across quintiles of SES (SES: socioeconomic status)

Figure 2 illustrates the extent to which disparities in failure to rescue are attributable to the effect of the hospital at which patients were treated. For each operation, there is a higher rate of failure to rescue at hospitals treating more patients with a low SES. However, all patients treated at those hospitals, regardless of the patient’s SES, experience this effect. For each operation, patients in the highest quintile of SES treated at the lowest SES hospital experienced a higher rate of failure to rescue than patients from the lowest quintile of SES being treated at the highest SES hospital. For some operations, such as cystectomy and pancreatectomy, this disparity was marked.

Figure 2.

Adjusted failure to rescue among the lowest and highest patient quintiles of SES within the lowest and highest hospital quintiles of SES (SES: socioeconomic status; FTR: failure to rescue)

The crude and adjusted odds of failure to rescue for each operation are shown in Table 2. These were determined first after adjusting for patient characteristics, and then for patient characteristics and fixed-hospital effects, in the lowest quintile of SES as compared to the highest. The crude odds ratio for failure to rescue for all operations was 1.20 [95% CI: 1.16–1.25], and these disparities were significant for all individual operations except cystectomy. The highest crude odds of death following a complication were seen following pancreatectomy, with an odds ratio of 1.43 [95% CI: 1.21–1.70]. When adjusted for patient characteristics alone, the odds ratios for failure to rescue decreased slightly but remained significant for all operations combined [1.16, 95% CI:1.12–1.19], colectomy [1.17, 95% CI: 1.13–1.21], lung resection [1.27, 95% CI:1.16–1.40], and pancreatectomy [1.45, 95% CI:1.21–1.73].

Table 2.

Influence of patient characteristics and hospital effects on variation in failure to rescue rates between the lowest and highest SES quintiles

Lowest vs. Highest SES Quintiles	Crude OR (95% CI)	OR Adjusted for Patient Characteristics (95% CI)	OR Adjusted for Patient Characteristics and Hospital Effect (95% CI)
All Operations	1.20 [1.16 – 1.25]	1.16 [1.12 – 1.19]	1.05 [1.01 – 1.09]
Cystectomy	1.15 [0.94 – 1.39]	1.18 [0.97 – 1.44]	0.96 [0.73 – 1.27]
Colectomy	1.10 [1.06 – 1.14]	1.17 [1.13 – 1.21]	1.07 [1.02 – 1.11]
Lung Resection	1.23 [1.12 – 1.35]	1.27 [1.16 – 1.40]	1.02 [0.91 – 1.15]
Pancreatectomy	1.43 [1.21 – 1.70]	1.45 [1.21 – 1.73]	1.22 [0.95 – 1.57]
Gastrectomy	1.17 [1.05 – 1.31]	1.04 [0.95 – 1.14]	0.95 [0.85 – 1.07]
Esophagectomy	1.28 [1.04 – 1.56]	1.12 [0.95 – 1.32]	0.95 [0.76 – 1.20]

(SES: socioeconomic status; OR: odds ratio; CI: confidence interval).

However, once adjusted for patient characteristics and hospital effects, the odds of failure to rescue decreased substantially for all operations. As shown in Table 2, the adjusted odds ratio for all operations decreased to 1.05 [95% CI: 1.01–1.09]. The odds of failure to rescue also decreased considerably for each individual operation. For example, the odds ratio for failure to rescue following pancreatectomy decreased to 1.22 [95% CI: 0.95 – 1.57], and lost statistical significance. Only colectomy retained statistical significance following adjustment for patient characteristics and hospital effects [1.07, 95% CI: 1.02–1.11]. For both multivariable logistic regression and hierarchical regression models, results were similar when SES was evaluated as a continuous variable.

DISCUSSION

In our analysis, we found that patients in the lowest quintiles of SES undergoing major oncologic operations have somewhat increased rates of major complications, but even higher rates of failure to rescue. A greater percentage of patients from the lowest quintile of SES are treated at hospitals with higher failure to rescue rates. Adjustment for patient characteristics and hospital effects using fixed-effect hierarchical modeling nearly eliminated disparities in failure to rescue across SES. Thus, this disparity appears to be driven, at least in part, by the hospital at which the patient was treated, as patients of all SES levels experience higher failure to rescue rates at low SES hospitals.

These findings correlate with previous work evaluating the influence of SES on surgical outcomes. Birkmeyer and colleagues, among others, have shown that SES is a significant predictor of operative mortality in both cardiovascular and oncologic operations.^{1, 2, 12–15} Numerous theories have been offered to explain these findings. Osborne and colleagues assert that patient characteristics, such as the number and severity of comorbidities, play the largest role in observed disparities.³ However, they studied patients undergoing cardiovascular operations, where peri-operative processes may be different as a result of longstanding quality improvement efforts. Multiple studies have documented the importance of access to care, showing that those without insurance coverage have worse mortality rates than those with coverage.^16–18 And finally, Daley and colleagues showed that patients of low SES most often receive care at lower quality hospitals, as determined by lack of available technology and decreased quality-of-care ratings (assessing the technical competence of staff, monitoring of care quality, coordination of work, and surgical leadership, among other factors).¹⁹

The identification of hospital quality as an important contributor to the disparities in surgical outcomes observed in disadvantaged populations is also supported by recent studies evaluating racial disparities, which report similar findings.^{20, 21} A recent systematic review of disparities in trauma care by Haider and colleagues identified numerous potential mechanisms for these variations, including hospital quality, within the domains of “host factors,” “pre-hospital factors,” “hospital / provider factors,” and “post-hospital care / rehabilitation factors.”²² Dimick and colleagues investigated mechanisms of racial disparities and found that residential segregation and race-related differences in referral patterns likely contribute to use of low-quality hospitals.²¹ This study adds to the literature by highlighting specific mechanisms within low-quality hospitals, such as failure to rescue, that likely play a significant role in the socioeconomic disparities of operative mortality seen among patients undergoing major inpatient surgery within the United States.^{5, 23} Failure to rescue has been previously shown to play an important role in the outcomes of many different types of oncology patients,^24–27 and it is likely one of the principal mechanisms through which available technology, staff technical competence, and work coordination influence the mortality variations that result in socioeconomic disparities.

This study has several important limitations. First, the use of Medicare data limits our analyses to patients 65 and older and could threaten the generalizability of our results. However, this subset of patients accounts for a large percentage of all patients diagnosed with cancer.²⁸ The use of administrative data could be considered a second limitation of this study. This type of data lacks clinical details such as cancer stage and disease location, which influence the outcome and prognosis of oncology patients. While this is less of a concern when evaluating rates of 30-day mortality, it does limit adjustment for operative complexity. Though administrative data has been criticized for inaccuracy in coding comorbidities and complications, we utilized the best-performing comorbidity index currently available.²⁹

A third potential limitation of this study is reliance on SES measured at the level of ZIP code, as opposed to Census tracts. While studies that have compared different geographic units of measurement have generally found that census-based units produce the greatest gradients in health according to SES,^{30, 31} those and other studies have found that ZIP code-level measures provide similar, but more conservative, estimates of socioeconomic disparity.^30–32 And finally, although we did not have access to individual-level data, the potential misclassification of SES according to ZIP code has been shown to be random, and as a result, the bias that results is a conservative estimate of the true influence of SES on health.³³ Consequently, the significant differences identified in this study are likely just a portion of the true disparities that exist between patients of varying SES.

This work has important implications for quality improvement efforts directed toward cancer surgery in disadvantaged populations. Many organizations, such as the Center for Medicare and Medicaid Services (CMS), focus their quality improvement efforts on the reduction of complications. The Surgical Care Improvement Project (SCIP), for example, targets process compliance for surgical site infection, venous thromboembolism, and myocardial infarction. Through the Hospital Value-Based Purchasing Program, CMS offers incentive payments to hospitals based on these measures.³⁴ This work adds to expanding literature that asserts complications and mortality are not causally-related at the hospital-level,^{24, 25, 35, 36} and as such, advocates a different approach. Moreover, these findings highlight the possibility that value-based purchasing could actually have a paradoxical effect on socioeconomically disadvantaged populations. By financially rewarding only high-performing hospitals, which care for fewer patients in the lowest SES quintile, this program could instead have the unintended consequence of widening the disparities observed in these populations. With this in mind, organizations wishing to reduce socioeconomic disparities and improve mortality rates following major cancer surgery could instead direct their efforts and resources toward processes and systems aimed at the timely recognition and management of complications once they occur.

Multiple hospital characteristics, such as nursing environments, ICU staffing, and teaching status, have been proposed to explain differences in failure to rescue rates between hospitals. Aiken and colleagues showed that, after adjusting for other hospital factors, each additional patient per nurse beyond a 4:1 ratio resulted in a 7% increase in the likelihood of dying, and a similar increase in the rate of failure to rescue.³⁷ Similarly, Friese and colleagues found a 37% increased odds of death for patients undergoing oncologic surgery in hospitals with poor nursing environments.³⁸ Multiple studies have illustrated decreased operative mortality in hospitals with daily ICU rounds.^{39, 40} Finally, Silber and colleagues established that patients undergoing surgery at hospitals with high teaching intensity had a 15% lower odds of death.⁴¹

Though various hospital characteristics have been identified as potential contributors to successful rescue from complications, it is likely that some combination of hospital resources, attitudes, and behaviors is what yields an environment most conducive to the timely recognition and effective management of complications.⁴² Future work to rectify the origins of socioeconomic disparities in perioperative care should concentrate on these factors also, not just compliance measures directed towards complications and their prevention. Future national hospital quality improvement initiatives can use these findings as evidence to support efforts to improve rescue rates in poorly performing hospitals, as part of a broad strategy directed toward effectively reducing socioeconomic disparities in cancer surgery mortality.