Crude mortality, complication, and failure-to-rescue (FTR) rates in black and white patients at hospitals with high–teaching-intensity (resident-to-bed ratio [RB] = 0.6) vs nonteaching hospitals (RB = 0). The relative differences between outcomes at hospitals with an RB of 0 vs an RB of 0.6 for black vs white patients were significant at the P = .001 level for death and P < .001 for FTR comparisons; the relevance difference for complications failed to reach statistical significance (P = .498).
Standardized mortality, complication, and failure-to-rescue (FTR) rates in black and white patients at hospitals with high teaching intensity (resident-to-bed ratio [RB] = 0.6) vs those with low teaching intensity (RB = 0). These are directly standardized results derived from model 1 of Table 5. The model was used to predict the outcomes of an artificial population in which the distribution of risk factors was the same for black and white patients and for patients at teaching-intensive and nonteaching hospitals. The relative differences between outcomes at hospitals with an RB of 0.6 vs 0 for black vs white patients were significant at the P < .001 level for death and FTR comparisons; complications failed to reach statistical significance.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Silber JH, Rosenbaum PR, Romano PS, et al. Hospital Teaching Intensity, Patient Race, and Surgical Outcomes. Arch Surg. 2009;144(2):113–120. doi:10.1001/archsurg.2008.569
To determine if the lower mortality often observed in teaching-intensive hospitals is because of lower complication rates or lower death rates after complications (failure to rescue) and whether the benefits at these hospitals accrue equally to white and black patients, since black patients receive a disproportionate share of their care at teaching-intensive hospitals.
A retrospective study of patient outcomes and teaching intensity using logistic regression models, with and without adjusting for hospital fixed and random effects.
Three thousand two hundred seventy acute care hospitals in the United States.
Medicare claims on general, orthopedic, and vascular surgery admissions in the United States for 2000-2005 (N = 4 658 954 unique patients).
Main Outcome Measures
Thirty-day mortality, in-hospital complications, and failure to rescue (the probability of death following complications).
Combining all surgeries, compared with nonteaching hospitals, patients at very major teaching hospitals demonstrated a 15% lower odds of death (P < .001), no difference in complications, and a 15% lower odds of death after complications (failure to rescue) (P < .001). These relative benefits associated with higher resident-to-bed ratio were not experienced by black patients, for whom the odds of mortality and failure to rescue were similar at teaching and nonteaching hospitals, a pattern that is significantly different from that of white patients (P < .001).
Survival after surgery is higher at hospitals with higher teaching intensity. Improved survival is because of lower mortality after complications (better failure to rescue) and generally not because of fewer complications. However, this better survival and failure to rescue at teaching-intensive hospitals is seen for white patients, not for black patients.
Outcomes are generally better in hospitals with higher teaching intensity,1-6 but it is unclear how this benefit is achieved. Lower risk-adjusted mortality at teaching hospitals might result from the prevention of some complications, prevention of death after a complication has occurred, both, or even one effect offsetting the other. While teaching hospitals are generally larger and have more advanced technology, greater volume, and better nurse staffing4,7,8 (attributes that may aid in both preventing complications and successfully treating complications), it is by no means clear whether all patients benefit equally from these attributes.
This study first examines surgical outcomes to determine whether differences in complication and failure-to-rescue rates explain observed differences in mortality rates between more and less teaching-intensive hospitals. As a measure of death after complications, failure to rescue provides an important test of how well hospitals do in treating patients who develop complications.9-11 We then examine how race is associated with outcomes at hospitals with higher or lower teaching intensity, as black patients both compose a disproportionate share of patients at teaching-intensive hospitals and obtain a disproportionate share of their surgical care at more teaching-intensive hospitals.12,13
A description of the data set and the selection/exclusion criteria has been previously reported in the Resident Hours Study,14 which examined all Medicare patients admitted to short-term general nonfederal acute care hos pitals from July 1, 2000, to June 30, 2005, with principal procedure/diagnosis-related group classification of general, orthopedic, or vascular surgery. The initial sample included 6 610 766 surgical patients from 5736 acute care hospitals within 50 states. After exclusions, a total sample of 4 658 594 patients from 3270 hospitals was left.
Outcome measures were death within 30 days of hospital admission, in-hospital complications, and failure to rescue. A patient was considered to have developed a complication if any complication was noted in the index hospitalization based on an algorithm published previously using the 1999-2000 Medicare Provider Analysis and Treatment File and available in the electronic appendix.11 Failure to rescue was defined as a death following an in-hospital complication and has been described in detail in other publications.9-11,15-19
The risk adjustment approach used was developed by Elixhauser and colleagues,20 Glance and colleagues,21 and Quan and colleagues22 at the Agency for Healthcare Research and Quality. We used age, sex, and 27 comorbidities (excluding fluid and electrolyte disorders and coagulopathy)21,22 and added 37 interaction terms derived from previous models using a 180-day lookback for comorbidities.11,23-26 There were a total of 82 diagnosis-related group/principal procedure groups.10,11 Racial assignments were based on self-reports. To simplify the presentation, we sorted patients only by black and white race; other racial groups were coded as “other” and not reported in this analysis.
The number of residents per hospital was obtained from Medicare Cost Reports. The resident-to-bed ratio is defined as the ratio of (interns + residents)/average operating beds4,14,27 and has been used in previous studies2,5,6,14 to differentiate “very major” teaching (resident-to-bed ratio > 0.6) from less teaching-intensive hospitals.
We tested the robustness of our findings with both logit regression models fitted by SAS Logistic and hospital random effects in a hierarchical model using SAS GLIMMIX.28 To help illustrate the impact of race and teaching intensity on outcomes, we also used direct standardization.19,29,30 After fitting our models with race and resident-to-bed ratio, we calculated the estimated probability of each outcome under 4 alternative assumptions: all patients were (1) white and in a hospital with a resident-to-bed ratio of 0.6; (2) white and a resident-to-bed ratio of 0; (3) black and a resident-to-bed ratio of 0.6; and (4) black and a resident-to-bed ratio of 0. In this way, we can better illustrate comparisons of outcome rates by race and teaching intensity, adjusting for differences in the health of patients based on the distribution of risk factors for the entire study population.
Table 1 describes the patient population. Black patients generally were younger but had more comorbidities than white patients. Patients at more teaching-intensive hospitals were younger and had fewer comorbidities. Black patients undergoing procedures at nonteaching hospitals (resident-to-bed ratio = 0) were somewhat older and had more comorbidities than black patients at teaching-intensive hospitals (resident-to-bed ratio > 0.6).
In Table 2, we display the associations between the resident-to-bed ratio and other hospital characteristics often associated with better patient outcomes.1-4,11,17,18 Larger resident-to-bed ratios were associated with higher proportions of hospitals with characteristics traditionally associated with better outcomes. In this report, we use the resident-to-bed ratio as a marker for a type of hospital, that is, as a proxy for the hospital characteristics associated with teaching intensity. Our models do not determine whether residents themselves cause the differences we observe.
For each outcome measure in Table 3, we provide results for each surgical category separately and then for the combined group. There were 4 658 954 patients in the mortality and complication models, but only 2 021 314 in the failure-to-rescue model because the failure-to-rescue analysis included only those who had a complication or died. For patients with similar comorbidities and procedures at hospitals with high teaching intensity (resident-to-bed ratio = 0.6 residents per bed) vs no residents (resident-to-bed ratio = 0), the fitted odds of death were 15% lower (95% confidence interval [CI] = 13%-16%) for the combined surgery group, with similar findings for subgroups. Results fitting a random-effects model using individual hospital indicators were similar. Adding income (as defined by median income in the patient's zip code31) to the combined surgery random-effects model did not change these results, suggesting that the observed differences between teaching-intensive and nonteaching hospitals are not reflecting unequal access by income level.
In contrast, the associations between the resident-to-bed ratio and complication rates indicated no consistent relationship, with odds ratios (ORs) overlapping 1.0. However, similar to mortality, failure-to-rescue rates were consistently lower in hospitals with higher resident-to-bed ratios. Hospitals of high teaching intensity (resident-to-bed ratio = 0.6) compared with nonteaching hospitals (resident-to-bed ratio = 0) were associated with 14% (95% CI, 12%-15%) lower odds of failure to rescue for combined surgery, with again similar findings for subgroup analysis. The random-effects model also produced similar results.
In Table 4, we can see that black patients had higher mortality rates and higher complication rates than white patients. White patients displayed a lower odds of death at teaching-intensive hospitals vs nonteaching hospitals (OR, 0.92; 95% CI, 0.90-0.95), whereas black patients displayed a slightly increased but statistically insignificant odds of dying at the teaching-intensive vs nonteaching hospital (OR, 1.03; 95% CI, 0.97-1.09). For both white and black patients, there was a modest reduction in complications at the teaching-intensive hospitals. Finally, whereas white patients displayed a reduction in failure-to-rescue rates in the teaching-intensive hospitals vs the nonteaching hospitals (OR, 0.94; 95% CI, 0.92-0.97), black patients displayed an increased failure-to-rescue rate (OR, 1.06; 95% CI, 1.00-1.12). Figure 1 displays the unadjusted results. For both black and white patients, rates of death were lower in the higher–teaching-intensity hospitals than in the nonteaching hospitals, although differences for black patients were far smaller than for white patients when comparing nonteaching with teaching-intensive hospitals. We compared the relative advantage of teaching intensity for black patients by calculating the odds of an outcome between the higher– (resident-to-bed ratio > 0.6) and lower– (resident-to-bed ratio = 0) teaching-intensity hospitals for black vs white patients. The relative benefit (if <1) or worsening (if >1) for black vs white patients at teaching-intensive vs nonteaching hospitals was 1.12 (95% CI, 1.05-1.19; P = .001) for death, 0.99 (95% CI, 0.96-1.02; P = .50) for complications, and 1.12 (95% CI, 1.06-1.20; P < .001) for failure to rescue.
Model 1 in Table 5 includes resident-to-bed ratio and race and their interactions. In model 1, in the first row of Table 5, the mortality model tests whether the odds of dying for black patients were higher or lower than for white patients in hospitals without residents (resident-to-bed ratio = 0). The adjusted odds of dying were 0.96 (95% CI, 0.95-0.98) for black patients compared with white patients at nonteaching hospitals (resident-to-bed ratio = 0). We then compared the odds of dying in a highly teaching-intensive hospital (resident-to-bed ratio = 0.6) with a hospital with a resident-to-bed ratio of 0. For combined surgery, for white patients, the odds of dying were 0.83 (95% CI, 0.81-0.84), representing 17% (95% CI, 16%-19%) lower mortality in hospitals with a resident-to-bed ratio of 0.6 vs 0. However, black patients displayed an OR of 1.04 (95% CI, 0.99-1.08) comparing hospitals with a resident-to-bed ratio of 0.6 with hospitals with a resident-to-bed ratio of 0. Hence, the mortality differences associated with teaching hospitals differed substantially for white and black patients (17% lower for white patients and 4% higher for black patients) when compared with nonteaching hospitals. The ratio 1.04/0.83 of these ORs (ie, the race × resident-to-bed ratio interaction) is 1.25 (95% CI, 1.20-1.31) and represents the relative benefit (if <1) or worsening (in >1) in black patients relative to white patients when comparing nonteaching hospitals (resident-to-bed = 0) with teaching-intensive hospitals (resident-to-bed ratio = 0.6) (see model 2 of Table 5). When we adjusted for Medicare geographic region in these models, our results were unchanged (results not shown).
The next analysis compares black and white patients in the same hospital, in contrast to comparing black and white patients at different hospitals with the same resident-to-bed ratio. We fit a model in which the resident-to-bed ratio was replaced by hospital indicators or fixed effects, so resident-to-bed ratio appears in the model only in the interaction with race. As can be seen from the row of Table 5 labeled “Model 3: Hospital Fixed Effects,” in this alternative fixed-effects model, the overall ratio of the black patient to white patient ORs comparing resident-to-bed ratio of 0.6 with resident-to-bed ratio of 0 is still large but somewhat smaller, namely 1.13 (95% CI, 1.07-1.19; P < .001). This suggests that even after controlling for the individual hospital, black patients benefit less than white patients when comparing highly teaching-intensive and nonteaching hospitals. The random-effects model specifying each hospital with an indicator variable (Model 4 in Table 5) gave similar results for the resident-to-bed ratio × black patient interaction, with the OR for death being 1.18 (95% CI, 1.12-1.24).
Complications displayed a very different pattern. Among patients in the combined surgery group, at nonteaching hospitals (resident-to-bed ratio = 0), black patients had a 14% (95% CI, 13%-15%) higher odds of developing a complication than white patients. White patients in highly teaching-intensive hospitals (resident-to-bed ratio = 0.6) had no different odds of complications than white patients in nonteaching hospitals (OR = 1.0; 95% CI, 0.99-1.00). Similarly, for black patients, the odds of developing complications were not different (OR, 0.99; 95% CI, 0.97-1.01) and not significant when comparing hospitals with a resident-to-bed ratio of 0.6 vs 0. The 2 models with fixed or random hospital indicators yielded the same conclusion.
Failure-to-rescue results are reported in the last column of Table 5 and are generally similar in pattern to the mortality results. White patients, but not black patients, had lower failure-to-rescue rates at teaching hospitals compared with nonteaching hospitals.
To better display the associations noted in Table 5, the directly standardized rates of death, complication, and failure to rescue are displayed in Figure 2 for each of 4 hypothetical (or “standardized”) patient groups: a black or white patient in a hospital with a resident-to-bed ratio of 0.6 or 0. In terms of mortality and failure to rescue, white patients do better at teaching hospitals than nonteaching hospitals, but black patients do about the same at teaching and nonteaching hospitals. While complication rates were higher in black patients than white patients, resident-to-bed ratio had little association with whether white or black patients developed complications. However, whereas white patients were more likely to survive after experiencing a complication at a teaching-intensive hospital as compared with a hospital without residents, this was not the case for black patients.
We found, as have others,1,3,11,32,33 that hospitals with higher teaching intensity appear to have lower risk-adjusted mortality after major surgery than less teaching-intensive hospitals. Previous studies have shown similar or higher postoperative complication rates at teaching hospitals than at nonteaching hospitals.34-38 We now demonstrate that the lower mortality rates in surgical cases are mediated by fewer deaths among patients who experienced complications (lower failure to rescue) and not by lower rates of complications. Moreover, this finding does not change when adjustments are made for zip code level income, suggesting that lower failure-to-rescue rates in this population are not generated by unequal access to higher teaching-intensity hospitals by patients of different incomes.
It is therefore of interest to find, when using data from the entire Medicare population in the United States, that black patients, unlike white patients, do not experience lower surgical mortality and failure-to-rescue rates at teaching-intensive hospitals. It appears that black patients fare about equally well in teaching and nonteaching hospitals, whereas white patients have significantly better risk-adjusted mortality and failure to rescue at teaching hospitals than at nonteaching hospitals.
Why does this racial disparity in mortality and failure to rescue exist? This disparity is smaller, though still substantial, in the model with a separate fixed effect for each hospital. This indicates that some, but by no means all, of the disparity stems from black patients going to teaching hospitals with similar resident-to-bed ratios but worse mortality and failure-to-rescue rates than their white counterparts (a similar effect was reported by Lucas et al39 and Barnato et al40). However, our study found that the within-hospital disparities are large, significant, and more substantial than those observed in previous work.13,40-42
In earlier work, we also studied racial differences in the length of surgery for comparable procedures and found lower-income black Medicare patients had surgery that took on average 29 minutes longer than white patients of similar income (P < .001). In part, this was because black patients tended to go to teaching hospitals that had longer procedure times.43,44 However, even when adjusting for the individual hospital, procedure time remained significantly longer in black patients, but now by 7 minutes (P < .001). Inside some very major teaching hospitals, the black-white difference was not apparent, while in others, the difference was more than 16 minutes for comparable surgery. The observed racial disparities in adjusted procedure length raise questions as to whether there are potential differences in who provides care to these populations at teaching-intensive hospitals.
Why racial differences in failure to rescue should occur within hospitals is not well understood, but there are many possibilities. Chan et al45 report that black patients were 22% (P < .009) more likely to experience a delay in initiating defibrillation than their white counterparts, with arrests occurring in unmonitored beds more often than in white patients (P < .001). Are black patients being monitored in the same way as their white counterparts? In search of a more general cause, Balsa and McGuire46 have described a process of “statistical” discrimination in which unintentional actions potentially based on poor communication may lead to disparities in outcomes. This could be exacerbated in time-pressured environments in which relatively inexperienced providers deliver much of the care. Unintentional differences in communication47 might lead to less appropriate or less accurate monitoring of black patients or less involvement in their care by personnel who could make a difference in reducing failure to rescue. In our previous work, we considered the possibility that the differences in surgical procedure length between whites and black patients may be because of different levels of involvement of physicians in training in black vs white patients.43,44
How does the difference in income between black and white patients relate to the disparity in failure to rescue? This is a complex issue because these are Medicare, non–health maintenance organization patients and, in principle, income should not be a factor in care, though gaps between principle and practice might occur. We did adjust for median income within the zip code of residence, and after adjustment, teaching-intensive hospitals still had lower failure-to-rescue rates than nonteaching hospitals in white patients but not black patients, suggesting that the apparent benefit of teaching intensity is not an artifact of unequal income.
It also was interesting that at nonteaching hospitals, black patients actually had slightly lower overall adjusted mortality than white patients, although the crude mortality rates were higher for black patients than white patients in nonteaching hospitals.48-51 We would not want to make too much of our finding since the coefficient on the race difference in nonteaching hospitals was small (an OR of 0.96) and recent work by Volpp et al49 and Polsky et al50 reports that black patients were noted to have lower 30-day mortality than white patients for a number of conditions, but this reversed with longer follow-up.
There are limitations to our study. Although we report on a very large sample size based on Medicare claims data, the trade-off is that these records do not contain medical record–based data. For example, we do not have details on the sequencing or severity of complications and do not know whether subgroups in this study had a different distribution of complications that may partially explain our findings.52,53 Relying on claims data, and not medical record review, does leave open the possibility that racial differences in mortality and failure to rescue may be because of unmeasured severity. However, our study compared white patients at less teaching-intensive hospitals with white patients at more teaching-intensive hospitals and the same for black patients. Hence, for our severity adjustment to be inadequate, even after our extensive risk adjustment, white patients entering teaching hospitals would need to be in better health than white patients entering nonteaching hospitals, but blacks entering teaching hospitals would need to be in the same health as black patients entering nonteaching hospitals. If black patients were sicker than white patients in the same unmeasured ways on admission to all hospitals, this by itself would not produce the pattern of mortality and failure-to-rescue rates that we found.
In conclusion, teaching-intensive hospitals with high resident-to-bed ratios have lower risk-adjusted mortality rates after major surgery than hospitals with lower ratios or without residents. This better survival is mainly because of better failure-to-rescue rates after postoperative complications. However, on average, while white patients have lower mortality and failure-to-rescue rates at teaching-intensive hospitals, black patients do not.
Correspondence: Jeffrey H. Silber, MD, PhD, Center for Outcomes Research, The Children's Hospital of Philadelphia, 3535 Market St, Ste 1029, Philadelphia, PA 19104 (email@example.com).
Accepted for Publication: September 8, 2008.
Author Contributions: Drs Silber and Volpp had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Silber, Rosenbaum, Romano, Rosen, and Volpp. Acquisition of data: Silber, Wang, Halenar, Even-Shoshan, and Volpp. Analysis and interpretation of data: Silber, Rosenbaum, Romano, Teng, and Volpp. Drafting of the manuscript: Silber, Rosenbaum, Rosen, Teng, and Volpp. Critical revision of the manuscript for important intellectual content: Silber, Rosenbaum, Romano, Wang, Halenar, Even-Shoshan, and Volpp. Statistical analysis: Silber, Rosenbaum, Rosen, Wang, and Teng. Obtained funding: Silber and Volpp. Administrative, technical, and material support: Silber, Halenar, Even-Shoshan, and Volpp. Study supervision: Silber.
Financial Disclosure: None reported.
Funding/Support: This work was funded through National Heart, Lung, and Blood Institute grant R01 HL082637, Department of Veterans Affairs grant IIR 04-202, and National Science Foundation grant SES 0646002.
Role of the Sponsors: The sponsors/funders had no role in the design or conduct of the study; collection, management, analysis, or interpretation of the data; and preparation, review, or approval of the manuscript.
Additional Contributions: Laura J. Bressler, BA, and Traci Frank, Center for Outcomes Research, The Children's Hospital of Philadelphia, helped conduct this research.