- To quantify in absolute terms the potential benefit of regionalisation of care from low- to high-volume hospitals.
Charlson comorbidity Index
International Classification of Diseases
Nationwide Inpatient Sample
(prolonged) length of stay
Many studies have shown that patients have more favourable outcomes postoperatively at high-volume hospitals than at low-volume hospitals [1-4]. Such reality has prompted several organisations to advocate a volume-based referral to improve quality of care and number of lives saved [5, 6]. In the context of radical cystectomy (RC) for patients with urothelial carcinoma of the urinary bladder, a volume–outcome relationship has also been recently recorded .
Despite the confirmation of the volume–outcome relationship in patients treated with RC, the quantification of the real impact of volume-based referral in absolute terms has never been performed. Given the shift, or at least the encouragement towards higher use of high-volume centres for the treatment of invasive bladder cancer in recent years [4, 6-8], a real assessment of the benefits of regionalisation is necessary. In this sense, we sought to simulate the effect of regionalisation by calculating the number of events that can be potentially avoided if patients were redirected from low- to high-volume hospitals. To accomplish these analyses, we focused on commonly assessed metrics (intraoperative and postoperative complications, blood transfusions, prolonged length of stay [pLOS], and in-hospital mortality) within a nationally representative sample of RC patients in the USA.
Data from the most contemporary years (1998–2009) of the Nationwide Inpatient Sample (NIS) were abstracted. The NIS includes inpatient discharge data collected via federal-state partnerships, as part of the Agency for Healthcare Research and Quality's Healthcare Cost and Utilization Project. As of the year 2009, the NIS contained administrative data on 8 043 415 discharges from 1044 hospitals within 40 states, approximating to 20% of community hospitals within the USA, including public hospitals and academic medical centres. The NIS is the sole hospital database in the USA with charge information on all patients regardless of payer, including persons covered by Medicare, Medicaid, private insurance, and the uninsured.
Relying on discharge records, all patients with a primary diagnosis of bladder cancer (International Classification of Diseases ICD-9-CM code 188) were identified. The cystectomy procedure code (ICD-9-CM 57.7) resulted in the identification of 16 180 patients who underwent RC between 1998 and 2009. Secondary diagnostic codes (ICD-9-CM 197.0, 197.7, 198.x) were used to identify patients with metastases, which were excluded from any further analysis. Relying on the ICD-9 procedure codes, patients who received a continent (orthotopic neobladder or continent cutaneous reservoir, ICD-9 code 57.87) or incontinent (ileal conduit, ICD-9 code 56.51) urinary diversion were identified.
Hospital volume was quantified using the annual hospital caseload, which represents the number of RCs performed annually, as was done previously . Patients were divided into three groups according to the annual caseload of the institution where their RC was performed: low, below the lower 20-percentile (1–3 RCs per year), high, above the upper 20-percentile (≥24 RCs per year), and intermediate (4–23 RCs per year)].
Patient age at RC was categorised into four subgroups: ≤59, 60–69, 70–79 and ≥80 years. Those aged <18 years were not included. Gender and race (White, Black, Hispanic, other races including Asian, Pacific Islander, Native American, or unspecified, and unknown), were also examined. Baseline Charlson comorbidity Index (CCI, 0 vs ≥1) was calculated according to Charlson et al. , and adapted according to Deyo et al. . To ensure uniformity of coding across data sources, detailed insurance categories are combined in the more general groups, namely private insurance, Medicare, Medicaid, and other. Income at the patient-level was not available within the NIS. Consequently, we relied on the median household income of the patient's ZIP code of residence, which were derived from the USA Census. Five categories (American dollars) were available within the database: (1) <25 000$, (2) 25 000–34 999$, (3) 35 000–44 999$, (4) ≥45 000$, and (5) unknown.
The NIS records up to 15 diagnoses and procedures for each admission. The presence of any complications was defined using ICD-9 diagnoses numbers 2–15. The specific ICD-9 codes used for postoperative complications were previously described , and recently updated . Intraoperative complication was defined as accidental puncture or laceration during a procedure (ICD-9: 998.2). Blood transfusion recipients were identified using the ICD-9 procedure codes: 99.02, 99.04. Finally, vital status was used to define in-hospital mortality. In-hospital mortality information is coded from disposition of patient. LOS was first continuously coded. Subsequently, for the purpose of analyses, admissions with a LOS above the upper 20th percentile (≥14 days) were defined as pLOS.
To make national estimates of the number of discharges more accurate, survey weights were applied in all models. All subsequent analyses were performed on the weighted population. In the first step, we attempted to confirm that a volume–outcome relationship existed in the present cohort, by fitting multivariable logistic regression models for prediction of intraoperative and postoperative complications, pLOS, and in-hospital mortality, where the primary variable of interest was hospital volume categories. All models were adjusted for patient age, baseline comorbidities (CCI), race, gender, insurance status, urinary diversion, and median annual income.
The second step consisted of our primary analyses, which aimed to quantify the real impact of a volume-based referral in the context of RC. As such, multivariable models were fitted only amongst high hospital volume-treated individuals (≥24 RCs/year). Adjustment was made for patient age, baseline CCI, race, gender, insurance status, urinary diversion, and median annual income. Subsequently, we applied each of the covariates' coefficients obtained from these models onto those exclusively treated at low-volume hospitals (≤3 RCs/year), with the intent of simulating the assumption that care was delivered at a high hospital volume setting. This calculation resulted in the definition of ‘predicted rates’ (e.g. predicted postoperative complication rates). Specifically, for each patient, based on his/her baseline characteristics (e.g. age, CCI), a predicted rate was obtained. The predicted rates were then compared with the actual observed complications rates among patients treated at low-volume hospitals. The difference between the observed rates and the predicted rates was defined as the excess of adverse events, corresponding to potentially avoidable outcomes.
Finally, the potential number needed to treat at a high-volume hospital in order to avoid one detrimental outcome at a low-volume hospital was calculated for each endpoint of interest, defined as the reciprocal of the absolute risk reduction (i.e. observed minus predicted rates) . This step allowed the expression of the impact of redirecting low-volume treated individuals to high-volume hospitals in absolute terms, with regard to adverse outcomes. Sub-analyses were repeated amongst patients aged ≥80 years and those with a baseline CCI of ≥1. Sensitivity analyses were performed by repeating the aforementioned steps to simulate the redirection of patients treated at low-volume hospitals to intermediate-volume hospitals. All tests were two-sided with a statistical significance set a P < 0.05. Analyses were conducted using the statistical package for R (the R foundation for Statistical Computing, version 2.12.2).
A weighted estimate of 79 859 patients who underwent a RC for non-metastatic urothelial carcinoma of the urinary bladder between 1998 and 2009 were identified. The mean (median, interquartile range) age was 68 (70, 62–76) years. Patient and sociodemographic characteristics differed according to hospital-volume groups, as described in Table 1. Treatment at various hospital-volume groups according to year of hospitalisation is shown in Figure 1.
|Characteristics||Overall (%)||Low HV (1–3), %||Intermediate HV(4–23), %||High HV (≥24), %||P|
|No. of patients (%)||79 859 (100)||16 739 (21.0)||46 675 (58.5)||16 445 (20.6)|
|Median income, $:|
In multivariable analyses, treatment at low-volume hospitals was associated with 26, 39, 17, and 75% higher odds of intraoperative complications, postoperative complications, blood transfusions, and pLOS compared with high-volume hospitals (all P < 0.03, Table 2). Additionally, patients treated at low-volume hospitals were at a 2.2-fold higher odds of succumbing to in-hospital death than patients treated at high-volume hospitals (P < 0.001).
|Hospital volume||Odds ratio (95% CI)|
|Intraoperative complication||Postoperative complication||Blood transfusion||pLOS (≥14 days)||In-hospital mortality|
|High||1.0 (ref.)||1.0 (ref.)||1.0 (ref.)||1.0 (ref.)||1.0 (ref.)|
|Intermediate||1.14 (1.01–1.28)||1.17 (1.12–1.27)‡||0.92 (0.88–0.96)‡||1.42 (1.35–1.48)‡||1.45 (1.26–1.67) ‡|
|Low||1.26 (1.10–1.45)||1.39 (1.32–1.45)‡||1.17 (1.18–1.23)‡||1.75 (1.66–1.86)‡||2.22 (1.90–2.58) ‡|
In the second part, we attempted to elucidate the impact of regionalisation of care from low- to high-volume hospitals. First, we applied the coefficients obtained for each endpoint of interest from multivariable models developed amongst high-volume hospitals-treated onto patients treated at low-volume hospitals. Under the assumption that care is delivered at high-volume hospitals instead of low-volume hospitals, the predicted rates of intraoperative complications, postoperative complications, blood transfusions, pLOS, and in-hospital mortality were 2.5, 30.2, 30.4, 16.1, and 1.8% respectively. The actual observed rates for respectively the same endpoints were 3.1, 37.6, 33.2, 25.5, and 3.8% (data not shown).
Excess rates were quantified by subtracting the predicted rates from the observed rates, which resulted in respectively 0.6, 7.4, 2.8, 9.4, and 2.0% for intraoperative complications, postoperative complications, blood transfusions, pLOS, and in-hospital mortality (Fig. 2A). In sub-analyses analyses, our results showed that excess rates for persons aged ≥80 years old were 0.9, 3.1, 5.9, 8.0, and 3.1%, respectively (Fig. 2B). Similarly, excess rates for persons with a baseline CCI of ≥1 were 0.4, 7.4, 2.0, 12.2, and 3.2%, respectively (Fig. 2C). In sensitivity analyses, excess rates between the predicted rates obtained from intermediate-volume hospitals and actual observed rates for the same endpoints were 0.3, 3.9, 5.0, 3.8, and 1.3%, respectively (Fig. 3A). The excess rates for persons aged ≥80 years old were 1.3, 0.4, 9.9, 1.8, and 0.8% (Fig. 3B), respectively. Similarly, the excess rates for persons with a baseline CCI of ≥1 were 0.3, 4.3, 5.2, 5.8, and 2.6%, respectively (Fig. 3C). The derived number of patients needed to redirect to intermediate- and high-volume hospitals in order to avoid one detrimental outcome at low-volume hospital are given in Table 3 .
|Endpoints||Overall||Aged ≥80 years||CCI ≥1|
|(A) Low- to high-volume|
|(B) Low- to intermediate-volume|
A growing body of evidence supports the presence of a volume–outcome relationship in several types of major surgery, including urological and gastrointestinal cancer surgery [2, 3, 14]. Given its technically demanding nature and the requirement for specialised postoperative care, outcomes after RC are very strongly linked to hospital volume , and, to a lesser extent, surgeon volume [11, 15]. This has led to the vast majority of RCs being performed at high-volume and primarily urban, academic centres [7, 16]. However, despite robust evidence that improved outcomes are achieved in higher volume centres, there is a paucity of data on the benefits that such regionalisation of bladder cancer surgery may achieve.
We therefore sought to address this void by using a large, contemporary, population-based analysis to attempt to quantify the absolute benefit that regionalisation may bring for patients undergoing RC. Specifically, we simulated the effect of regionalised care delivery at high-volume hospitals instead of low-volume hospitals, with respect to several specific endpoints: intraoperative complications, postoperative complications (overall and according to 10 specific postoperative complications), blood transfusions, pLOS, and in-hospital mortality.
The first part of the present analysis adds to the volume–outcome literature on RC. After controlling for patient age, race, gender, insurance status, baseline CCI, urinary diversion and median annual income, we found that outcomes at high-volume institutions were significantly better than those at lower-volume centres in all five of our principal outcome measures, in accordance with prior data [17, 18]. Patients treated at low-volume institutions were 26, 39, 17, and 75% more likely to experience intraoperative complications, postoperative complications, blood transfusions, and have a pLOS, respectively, compared with those operated at high-volume hospitals. Most strikingly, patients with bladder cancer having surgery at low-volume centres had a 2.2-fold increased risk of an in-hospital death compared with those at high-volume centres. These data serve to confirm the need to centralise or regionalise complex cancer surgeries, such as RC. However, despite the unequivocal evidence that superior outcomes are achieved at higher volume hospitals, they are unable to quantify the benefit that may accrue as a result of such regionalisation. In the absence of a randomised comparison of outcomes after RC at high- and low-volume centres, we addressed this question by analysing observed and expected complication rates at low-volume hospitals and thereby quantifying the benefits of regionalisation in relative terms.
These latter analyses showed that low-volume centres had excess rates of 0.6, 7.4, 2.8, 9.4 and 2.0% for intraoperative complications, postoperative complications, blood transfusion, pLOS and in-hospital mortality, respectively. Using these risk reductions, we were able to calculate a putative ‘number needed to treat’ to quantify the number of RCs that need to be performed at high-volume rather than low-volume hospitals to achieve a reduction of one complication (i.e. the benefit of regionalisation). These translate to 166, 14, 36, 11 and 50 for intraoperative complications, postoperative complications, blood transfusions, pLOS, and in-hospital mortality, respectively. Similar patterns were reported when analyses were restricted to patients aged ≥ 80 years or those with a baseline CCI of ≥1. Moreover, the present results showed that the benefit of regionalisation was not unimportant even by redirecting patients treated at low-volume hospitals to intermediate-volume hospitals, which comprised 60% of our patient cohort.
Given the current incidence of bladder cancer in the USA (an estimated 73 510 cases will be diagnosed in 2012 ), and that ≈25% of these will be muscle-invasive , with 45% of such patients undergoing RC , we can therefore surmise that 50 patients undergoing RC for muscle-invasive bladder cancer this year would be spared an intraoperative complication should they be operated at a high- rather than a low-volume hospital. Similarly, 230 patients would be spared a blood transfusion, 591 patients would avoid a postoperative complication and 752 patients would not have a pLOS. Most importantly, regionalisation could theoretically prevent the deaths of 267 patients with bladder cancer, which is ≈2% of the current death rate from the disease. Moreover, for older and more highly co-morbid patients, a patient group recognised as less likely to receive care at a high-volume institution , regionalisation may achieve an even greater benefit. The excess rates of in-hospital mortality were higher in this sub-group, translating into a figure of 32 and 31 RCs that would need to be performed to prevent one in-hospital death in patients aged ≥ 80 years, or those with a CCI of ≥1, respectively. Whilst these figures are clearly theoretical, they serve to provide, for the first time, a quantitative estimate of the potential benefit that regionalisation of RC may achieve in the USA.
Healthcare regionalisation has gained increasing acceptance over the past decade, with good evidence that lives may be saved by adopting such a policy [6, 23, 24], and initiatives undertaken to centralise complex surgery to hospitals meeting particular volume thresholds . However, there is opposition to this initiative, with some evidence, particularly from Canada, that it may not achieve as significant a benefit as purported . Furthermore, it is argued that competition lowers hospital prices paid by health insurers, and that it may lead to improved outcomes as well as lower patient costs , which is particularly significant when considering that nearly 20% of the USA Gross Domestic Product is projected to be spent on healthcare by 2012 . However, economic analysis has suggested that whilst the benefits of regionalisation may not be as large as is widely assumed, there are still gains to the healthcare consumer arising from such a policy . Moreover, there is little evidence that it would substantially impact on smaller, rural hospitals , one of the groups that stand most to lose from centralisation of services. Therefore, the present data add quantitative weight behind the move to regionalise RC services that has been occurring in the last few years [7, 16].
Nevertheless, there are important limitations to the present study. First, despite adjusting for confounding variables, several assumptions were made in the present analyses. For example, we assumed that the effect of race on postoperative complications in those treated at higher-volume hospitals is similar to those treated at lower-volume hospitals. An ideal method to delineate the benefit that may be achieved with transferring care from low- to high-volume hospitals would be a randomised controlled trial. However, this would be unethical and hence, the present study design represents an acceptable alternative. Secondly, the definition of low-, medium- and high-volume hospitals was made according to percentiles, whereas it is possible that a specific volume threshold may better distinguish between volume thresholds. Furthermore, the present study cannot confirm a causal relationship between volume and outcomes, some espouse the ‘practice-makes-perfect’ hypothesis, whereby hospitals and surgeons get better at procedures after doing more of them, whereas others may argue that other certain hospitals are intrinsically better equipped to provide a higher quality of care in general, which may result in more referrals. Or it could be possible that the observed associations are a result of additional variables that are beyond measure. Finally, from a practical perspective, if RC services were transferred from low- to high-volume institutions, the latter may not have the capacity to take on the extra caseload, thereby obviating any outcome benefits that may arise. In this case, the present analyses do show that a non-insignificant benefit can also be achieved by redirecting treatment from low- to intermediate-volume institutions. As such, a more realistic approach to improving outcomes may be to selectively avoid the lower-volume hospitals.
In conclusion, while accepting the limitations of the data, the present study, for the first time, quantifies the potential benefit of regionalisation of RC for muscle-invasive bladder cancer to higher-volume hospitals. Many complications may be averted and a substantial improvement in mortality may occur should such a move take place, with older patients and those with more co-morbidities standing most to benefit.
P.I.K. is partially supported by the University of Montreal Health Centre Urology Specialists, Fonds de la Recherche en Santé du Québec, the University of Montreal Department of Surgery and the University of Montreal Health Centre (CHUM) Foundation.