Abstract

OBJECTIVE

Large provider caseloads are associated with better patient outcomes after many complex surgical procedures. Mortality rates for pediatric brain tumor surgery in various practice settings have not been described. We used a national hospital discharge database to study the volume-outcome relationship for craniotomy performed for pediatric brain tumor resection, as well as trends toward centralization and specialization.

METHODS

We conducted a cross sectional and longitudinal cohort study using Nationwide Inpatient Sample data for 1988 to 2000 (Agency for Healthcare Research and Quality, Rockville, MD). Multivariate analyses adjusted for age, sex, geographic region, admission type (emergency, urgent, or elective), tumor location, and malignancy.

RESULTS

We analyzed 4712 admissions (329 hospitals, 480 identified surgeons) for pediatric brain tumor craniotomy. The in-hospital mortality rate was 1.6% and decreased from 2.7% (in 1988–1990) to 1.2% (in 1997–2000) during the study period. On a per-patient basis, median annual caseloads were 11 for hospitals (range, 1–59 cases) and 6 for surgeons (range, 1–32 cases). In multivariate analyses, the mortality rate was significantly lower at high-volume hospitals than at low-volume hospitals (odds ratio, 0.52 for 10-fold larger caseload; 95% confidence interval, 0.28–0.94; P = 0.03). The mortality rate was 2.3% at the lowest-volume-quartile hospitals (4 or fewer admissions annually), compared with 1.4% at the highest-volume-quartile hospitals (more than 20 admissions annually). There was a trend toward lower mortality rates after surgery performed by high-volume surgeons (P = 0.16). Adverse hospital discharge disposition was less likely to be associated with high-volume hospitals (P < 0.001) and high-volume surgeons (P = 0.004). Length of stay and hospital charges were minimally related to hospital caseloads. Approximately 5% of United States hospitals performed pediatric brain tumor craniotomy during this period. The burden of care shifted toward large-caseload hospitals, teaching hospitals, and surgeons whose practices included predominantly pediatric patients, indicating progressive centralization and specialization.

CONCLUSION

Mortality and adverse discharge disposition rates for pediatric brain tumor craniotomy were lower when the procedure was performed at high-volume hospitals and by high-volume surgeons in the United States, from 1988 to 2000. There were trends toward lower mortality rates, greater centralization of surgery, and more specialization among surgeons during this period.

Increasing evidence suggests that patient mortality and morbidity rates are lower when care is provided in high-volume settings. For example, the in-hospital mortality rate is lower when complex cancer operations (6,29), pediatric cardiovascular operations (23), and craniotomies for adult intracranial tumors (14,41) are performed at high-volume hospitals or by high-volume surgeons. Pediatric subspecialization has also been correlated with an increased number of complete intracranial tumor resections among pediatric patients (3). However, in-hospital mortality rates have not been previously studied in relation to provider caseloads with respect to pediatric brain tumor resection. The extent to which pediatric brain tumor resection is presently performed in nonspecialized settings in the United States, as well as the results obtained in these settings, remains largely unknown. Although the American Academy of Pediatrics guidelines recommend that such patients be referred to pediatric neurosurgical specialists at the outset of care (4), approximately one-half of children enrolled in Children's Cancer Group protocols between 1986 and 1992 underwent resections performed by general neurosurgeons who performed such operations infrequently (3).

Using a hospital discharge database, we analyzed the results of craniotomy performed between 1988 and 2000 for the resection of pediatric brain tumors in a representative sample of United States nonfederal hospitals. The mortality rates after pediatric brain tumor resections were related to provider caseloads and were examined for changes with time. We also sought to discover whether there was progressive centralization of pediatric brain tumor resections at higher-volume hospitals and concentration of care in the hands of pediatric specialist surgeons during the study period.

PATIENTS AND METHODS

Database Used

The data source for this study was the Nationwide Inpatient Sample (NIS) hospital discharge database for the years 1988 to 2000, which we obtained from the Healthcare Cost and Utilization Project, Agency for Healthcare Research and Quality (Rockville, MD) (56). The NIS database is a hospital discharge database that represents approximately 20% of all inpatient admissions to nonfederal hospitals in the United States. For these years, the NIS database contains discharge data on 100% of discharges from a stratified random sample of nonfederal hospitals in 8 to 28 states, to approximate a representative 20% subsample of all United States nonfederal hospital discharges. Because the NIS database contains data for all patients discharged from sampled hospitals during the year, regardless of age or payer, it can be used to obtain the annual total volume of specified procedures at individual hospitals. Individual patient identifiers are deleted from the NIS database distributed to users, and users are required to report only aggregate results, to prevent inadvertent identification of individuals. For many states, the surgeon who performed the principal procedure during the hospital admission is identified with a unique masked code. An overview of the NIS database is available at http://www.ahcpr.gov/data/hcup/nisintro.htm.

Inclusion and Exclusion Criteria and Definition of End Points

A hospital admission for craniotomy for the resection of a pediatric brain tumor was defined as follows: patient age of 18 years or less; a primary International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) diagnosis code of 191.0 to 191.9 (malignant neoplasm of brain), 225.0 (benign brain neoplasm), 237.5 (brain neoplasm of uncertain behavior), or 239.6 (brain neoplasm not otherwise specified); and a primary procedure ICD-9-CM code of 01.53 (lobectomy) or 01.59 (excision or destruction of tissue or lesion of brain). (The more descriptive Current Procedural Terminology [CPT] codes commonly used in the United States are not available in the NIS database.)

In-hospital death and discharge to institutions other than to home were the primary end points of the study. In-hospital death was coded directly in the NIS database and analyzed by performing logistic regression. Discharge disposition was coded on a four-level scale as death, discharge to a long-term facility, discharge to other facilities, or discharge to home and was analyzed by performing ordinal logistic regression. Secondary end points of length of stay (LOS) and total hospital charges were coded in the NIS database and were analyzed only for patients discharged from the hospital alive. LOS and hospital charge data were highly positively skewed and were analyzed as logarithmic transforms. LOS data for 10 patients (0.2%) with 0 days LOS were recoded as missing. For the period from 1988 to 1992, admissions to zero-weight NIS hospitals were omitted from the study.

Patient Characteristics

Patient age, sex, race, median household income within the postal code of the patient's residence, primary payer (Medicaid, private insurance, self-pay, no charge, or other), and type of admission (emergency, urgent, or elective) were coded in the NIS database. Because age was related to risk in a nonlinear manner, it was either dichotomized (greater than or less than 2 yr) or modeled with the use of a nonparametric smoother (26) in all analyses. Four cases (0.1%) with an admission type of “other” were recoded as routine admissions. More than 5% of discharges had missing values for two variables used principally as stratification factors for other analyses, i.e., race (33% missing) and admission type (12% missing). When these variables were used as stratification factors, missing values were imputed as follows. Missing race was set to Caucasian. Missing admission type was set to emergency for admissions whose source was the emergency room, to urgent for admissions that were transfers from other hospitals, and to routine for admissions from other sources. When race or admission type was the focus of the analysis, imputed values were not used. We tested the validity of the imputations by repeating the main analysis after reclassifying race and admission type, by including a separate “missing value” category for each variable. There was no significant difference between the results of this analysis and the results of the analysis that used imputed variables. Malignant histological features indicate a primary ICD-9-CM diagnosis code of 191.0 to 191.9, and posterior fossa location indicates a primary ICD-9-CM diagnosis code of 191.6 or 191.7. ICD-9-CM classifies all astrocytomas, including both juvenile pilocytic astrocytomas and World Health Organization Grade II astrocytomas, as malignant (59); therefore, the frequency of malignant tumors in our analysis exceeds the value that would be expected with the use of World Health Organization or related grading systems.

Provider and Hospital Characteristics

Hospital region (Northeast, Midwest, South, or West), location (rural or urban), teaching status, and bed size (small, medium, or large) were coded directly in the NIS database. The bed size numbers used to define hospital bed size categories in the NIS database varied according to year, hospital type, and geographic region. For example, among urban Northeast teaching hospitals in 2000, bed size values were 1 to 249, 250 to 424, and ≥425 beds. We defined pediatric hospitals as those at which at least 75% of patients admitted in a given year were 18 years of age or younger. To obtain hospital and surgeon volumes of pediatric brain tumor resections, we counted the cases for each identified surgeon and hospital in the database. Because hospital and physician volumes were positively skewed, the logarithmic transforms were used when volume measurements were entered into the regression models. The percentage of a surgeon's practice that consisted of pediatric surgery was calculated by dividing the number of the surgeon's admissions of patients 18 years of age or younger by the total number of the surgeon's admissions. Only surgical admissions were included in the calculation.

Statistical Methods

Statistical methods included Fisher's exact test, Wilcoxon rank-sum and signed-rank tests, Spearman's rank correlation, and log-linear and linear least-squares, ordinary logistic, and proportional-odds ordinal logistic regression analyses (25,30,45). To correct for possible clustering of similar outcomes within hospitals, which could cause falsely inflated estimates of the statistical significance of regression coefficients, standard errors of the mean were calculated by using a Huber-White sandwich variance-covariance matrix estimated on the basis of the data with adjustment for clustering by hospital (25). LOS and hospital charges were analyzed as logarithmic transforms by performing least-squares regression corrected for clustering as described above. Hospital caseload was modeled in relation to time and hospital-level characteristics with the use of a zero-inflated binomial model, with robust standard errors of the mean adjusted for clustering by hospital to account for within-hospital correlations in caseloads for different years (8,35,39,40).

Projections to the entire United States population were adjusted for the NIS stratified survey method by using SAS PROC SURVEYMEANS (2). Linear regression analyses performed for extrapolated population values (i.e., to test trends in the annual number of admissions or annual mortality rates) were weighted by the inverse of the variance. Other regression analyses treated the sample as a simple random draw from an infinite possible population (i.e., without weighting or correction for a finite sampling fraction).

Calculations were performed with SAS (Version 8.2; SAS Institute, Cary, NC), Stata (Version 7.0; Stata Corp., College Station, TX), and S-Plus (Versions 3.3 and 4.0 for Windows; Insightful, Inc., Seattle, WA) software, with the Hmisc and Design modeling function software libraries written by Harrell (24,25) and the Locfit local likelihood regression library written by Loader (37,38). All P values reported are two-tailed.

RESULTS

We identified 4712 admissions for the resection of pediatric brain tumors between 1988 and 2000 in the NIS database. The clinical characteristics of the patients are presented in Table 1. Most patients were Caucasian and between the ages of 4 and 13 years (Fig. 1), and 55% of the patients were male. At the time of admission, 18% of the patients were 2 years of age or younger and 3% of the patients were 90 days of age or younger. One-half of the admissions were routine, one-fourth were urgent, and one-fourth were emergencies. Eighty-two percent of the patients had malignant tumors (including both pilocytic astrocytomas and World Health Organization Grade II astrocytomas, according to the ICD-9 coding system) and 9% had benign tumors; in the rest of the cases, the histological type was not specified. Locations, coded only for malignant tumors, were frontal (8%), temporal (9%), parietal (7%), occipital (3%), cerebellar (31%), brainstem (7%), and other or unspecified (36%). Most resections were performed at large urban teaching hospitals that were not pediatric hospitals. The in-hospital mortality rate for the cohort was 1.6% (95% confidence interval [CI], 1.3–2.0%), and 0.2% of patients were discharged to long-term facilities, 5.5% were discharged to other facilities such as rehabilitation hospitals, and 92.6% were discharged directly to home.

TABLE 1.

Clinical characteristics of 4712 pediatric patients who underwent brain tumor resection

FIGURE 1.

Histogram showing the age distribution of 4712 pediatric patients admitted for brain tumor resection.

FIGURE 1.

Histogram showing the age distribution of 4712 pediatric patients admitted for brain tumor resection.

Patient Characteristics and Outcomes

Age, sex, race, primary payer for care, median income in the postal code of residence, admission type (emergency, urgent, or routine), and tumor histological type (malignant versus other) and location (posterior fossa versus other) were tested as predictors of death and discharge disposition (Table 2). Age was an important predictor of death and discharge not directly to home (P < 0.001) (Fig. 2). The mortality rate for patients younger than 2 years of age was 5.8%, compared with 1.1% for patients 2 years of age or older, and 11.4% of patients younger than 2 years of age were not discharged directly to home, compared with 6.9% of patients 2 years of age or older. Patient sex was not significantly associated with death or discharge disposition. Caucasian race, private insurance, and residence in postal codes with a high median income were all associated with lower mortality rates and better discharge dispositions. Patients with emergency or urgent admissions demonstrated higher mortality rates and worse discharge dispositions than did those with routine admissions. Malignant tumor histological type predicted significantly higher mortality rates and worse discharge dispositions. Tumor location in the posterior fossa did not affect mortality rates but was associated with worse discharge dispositions. All of these variables, as well as stratification by hospital geographic region, were included in the multivariate analyses described below.

TABLE 2.

Effects of patient characteristics on mortality rates and hospital discharge dispositiona

FIGURE 2.

Probability of death and discharge not directly to home after craniotomy for pediatric brain tumor resection, plotted against age. Patients younger than 2 years of age demonstrated higher rates of death and discharge not directly to home (P < 0.001 for both).

FIGURE 2.

Probability of death and discharge not directly to home after craniotomy for pediatric brain tumor resection, plotted against age. Patients younger than 2 years of age demonstrated higher rates of death and discharge not directly to home (P < 0.001 for both).

Hospital and Surgeon Characteristics and Patient Outcomes

Patients were treated at a total of 329 hospitals. Of this total, 205 (62%) were teaching hospitals, 22 (7%) were pediatric hospitals, and 319 (97%) were urban hospitals. In comparing hospitals in the NIS database at which pediatric brain tumor resections were or were not performed, we observed that the hospitals that were more likely to perform pediatric brain tumor resections were pediatric hospitals, urban hospitals, teaching hospitals, and large-bed size hospitals (P < 0.001 for all). For 45% of the admissions, 480 treating surgeons were identified in the database.

Hospitals and surgeons varied widely with respect to the number of pediatric brain tumor resections performed. Analysis of the data on a per-patient basis revealed that the median annual number of craniotomies for pediatric brain tumor resections was 11 cases/hospital (range, 1–59 admissions; 25th percentile, 5 admissions; 75th percentile, 21 admissions) and 6 cases/surgeon (range, 1–32 admissions; 25th percentile, 2 admissions; 75th percentile, 15 admissions). For 372 patients (8%), the craniotomy for pediatric brain tumor resection was the only one reported during that year at the hospital at which they were treated; for 468 patients (22% of those with identified surgeons), the craniotomy for pediatric brain tumor resection was the only one reported in that year by the surgeon who performed it.

Large hospital and surgeon caseloads were associated with lower in-hospital mortality rates after pediatric brain tumor resection (Fig. 3A). The analysis was adjusted for case mixture by using data on patient age, sex, race, primary payer for care, income in postal code of residence, geographic region, admission type, tumor histological type (malignant versus other), tumor location (posterior fossa versus other), and year of treatment. Odds ratios (ORs) for the importance of hospital and surgeon caseloads are reported for a 10-fold difference in caseload, which approximates the difference between the 25th and 75th percentiles for caseload. In-hospital mortality rates were lower at high-volume hospitals than at low-volume hospitals (OR, 0.52; 95% CI, 0.28–0.94; P = 0.03). The mortality rate at the lowest-volume-quartile hospitals (1–4 admissions/yr) was 2.3%, compared with 1.4% at the highest-volume-quartile hospitals (21 or more admissions/yr). After adjustment for hospital caseload, no other hospital characteristic was a significant predictor of mortality rates. The difference in mortality rates between high- and low-volume hospitals was largest for the youngest patients (Fig. 4A). For patients younger than 2 years of age, the OR for death for a 10-fold increase in hospital caseload was 0.33 (95% CI, 0.15–0.75), compared with an OR of 0.57 (95% CI, 0.43–0.76) for older children. ORs for death with respect to hospital caseload for patients with posterior fossa tumors were not significantly different from those for patients with other tumor types.

FIGURE 3.

In-hospital mortality rates after resection of pediatric brain tumors, plotted against hospital (A) and surgeon (B) caseloads (grouped by quartile). The relationship between larger caseloads and lower mortality rates was significant in the multivariate analysis for hospitals (P = 0.03) but not for surgeons (P = 0.16). Error bars, 95% CI.

FIGURE 3.

In-hospital mortality rates after resection of pediatric brain tumors, plotted against hospital (A) and surgeon (B) caseloads (grouped by quartile). The relationship between larger caseloads and lower mortality rates was significant in the multivariate analysis for hospitals (P = 0.03) but not for surgeons (P = 0.16). Error bars, 95% CI.

FIGURE 4.

Outcomes as a function of age for highest- and lowest-volume quartiles of hospitals, plotted using local likelihood fitting. A, mortality rates; B, rates of discharge other than directly to home. Solid line, lowest-volume-quartile hospitals; dashed line, highest-volume-quartile hospitals. The difference in outcomes between high- and low-volume hospitals was greatest for young patients.

FIGURE 4.

Outcomes as a function of age for highest- and lowest-volume quartiles of hospitals, plotted using local likelihood fitting. A, mortality rates; B, rates of discharge other than directly to home. Solid line, lowest-volume-quartile hospitals; dashed line, highest-volume-quartile hospitals. The difference in outcomes between high- and low-volume hospitals was greatest for young patients.

Mortality rates were lower when high-caseload surgeons performed the resection, but not significantly so (OR, 0.60; 95% CI, 0.29–1.24; P = 0.16) (Fig. 3B). The mortality rate after operations performed by the lowest-volume-quartile surgeons (1 admission/yr) was 1.9%, compared with 1.4% for the highest-volume-quartile surgeons (16 or more admissions/yr).

Adverse discharge disposition was also less likely after pediatric brain tumor resections performed at high-volume hospitals (OR, 0.52; 95% CI, 0.39–0.71; P < 0.001) (Fig. 5A). Case mixture adjustment was performed with multivariate regression, as described above. The rate of discharge not directly to home was 10.7% at the lowest-volume-quartile hospitals (1–4 admissions/yr), compared with 6.3% at the highest-volume-quartile hospitals (21 or more admissions/yr). The difference between high- and low-volume hospitals in rates of discharge not directly to home was slightly greater for younger patients than for older ones (Fig. 4B). After adjustment for hospital volume, pediatric hospitals demonstrated lower rates of discharge not directly to home (OR, 0.58; 95% CI, 0.40–0.85; P = 0.005). No other hospital characteristic was a significant predictor of discharge disposition, after adjustment for hospital volume and pediatric hospital status.

FIGURE 5.

Rates of discharge not directly to home after resection of pediatric brain tumors, plotted against hospital (A) and surgeon (B) caseloads (grouped by quartile). The relationship between larger caseloads and lower mortality rates was significant in multivariate analyses for both hospitals (P < 0.001) and surgeons (P = 0.04). Error bars, 95% CI.

FIGURE 5.

Rates of discharge not directly to home after resection of pediatric brain tumors, plotted against hospital (A) and surgeon (B) caseloads (grouped by quartile). The relationship between larger caseloads and lower mortality rates was significant in multivariate analyses for both hospitals (P < 0.001) and surgeons (P = 0.04). Error bars, 95% CI.

With respect to surgeon caseload, higher volume also predicted better patient outcomes at hospital discharge (OR, 0.70; 95% CI, 0.50–0.98; P = 0.04) (Fig. 5B). The rate of discharges not directly to home was 11.4% after operations performed by the lowest-volume-quartile surgeons (1 admission/yr), compared with 7.7% for the highest-volume-quartile surgeons (16 or more admissions/yr).

Access to High-volume Providers

Patient characteristics were tested as potential predictors that surgery would be performed at a high-volume hospital or by a high-volume surgeon. Age younger than 2 years, race, and primary payer for care were not significantly associated with hospital or surgeon caseload. Patients from higher-income areas of residence were associated with high-volume hospitals (P < 0.001) but not with high-volume surgeons (P = 0.15). Emergency or urgent admissions were more common at low-volume hospitals (P < 0.001) and for low-volume surgeons (P = 0.01). Tumor histological type and location were not associated with hospital or surgeon caseload.

Trends with Time

Increasing Total Annual Caseloads, Centralization, and Specialization of Care

Projected to the United States population, the total annual number of admissions increased from 1260 in 1988 to 1910 in 2000, a 52% relative increase (Fig. 6A). The percentage of patients younger than 2 years of age decreased from 14.6% in 1988 to 1990 to 10.7% in 1997 to 2000 (P = 0.02). In later years in this series, more patients lived in postal codes with higher median incomes (P < 0.001), and there was a trend toward increased private insurance coverage (P = 0.08). Patients with malignant tumors (P = 0.03) or tumors located in the posterior fossa (P < 0.001) constituted less of the study cohort in later years; both of these trends remained significant after adjustment for age. There was no change in the distributions of patient sex, race, or admission type during the study period.

FIGURE 6.

Trends with time for the resection of pediatric brain tumors. A, total number of admissions annually, projected for all United States nonfederal hospitals. B, in-hospital mortality rates for cases in the NIS database. Error bars, 95% CI.

FIGURE 6.

Trends with time for the resection of pediatric brain tumors. A, total number of admissions annually, projected for all United States nonfederal hospitals. B, in-hospital mortality rates for cases in the NIS database. Error bars, 95% CI.

Between 180 and 280 United States hospitals, or 3.2 to 5.6% of the approximately 5000 United States nonfederal hospitals, performed craniotomy for pediatric brain tumor resection in the individual years of the series. There was no increase in the number of hospitals performing pediatric brain tumor resections during the study period (P = 0.8).

The median number of operations per hospital at hospitals performing at least one operation increased from 1.7 (in 1988–1990) to 3 (in 1997–2000), and the 90th percentile hospital caseload value (approximately the 20 highest-volume hospitals in the United States) increased from 10.7 (in 1988–1990) to 16.5 (in 1997–2000) (Fig. 7). In 1988 to 1990, 13.3% of operations were the only pediatric brain tumor resection performed in an individual year at that hospital, compared with 5.5% in 1997 to 2000. Of the 275 hospitals that were in the database for 2 or more years, the caseload for the last sampled year was larger than the caseload for the first sampled year for 35%, the same for 39%, and smaller for 25%, indicating a significant shift toward a larger caseload in the last sampled year than in the first sampled year (P = 0.02). On a per-patient basis, the median hospital caseload increased from 8 admissions (in 1988–1990) to 14 admissions (in 1997–2000).

FIGURE 7.

Selected caseload percentiles for hospitals that performed at least one craniotomy for pediatric brain tumor resection in a given year, plotted using local likelihood fitting. For example, the 90th percentile caseload for hospitals in 1988 was 10 admissions, increasing to 16 in 2000. Caseloads at the largest centers increased disproportionately during the study period.

FIGURE 7.

Selected caseload percentiles for hospitals that performed at least one craniotomy for pediatric brain tumor resection in a given year, plotted using local likelihood fitting. For example, the 90th percentile caseload for hospitals in 1988 was 10 admissions, increasing to 16 in 2000. Caseloads at the largest centers increased disproportionately during the study period.

A greater proportion of patients were treated at teaching hospitals in the later years of the series than in the earlier years. From 1988 to 1990, 58% of admissions were to teaching hospitals, compared with 85% from 1997 to 2000 (P < 0.001) (Fig. 8A). Analyzed on a per-hospital basis, the median, 75th percentile, and 90th percentile hospital caseload values decreased at nonteaching hospitals and increased at teaching hospitals during the study period (Fig. 8, B and C).

FIGURE 8.

Centralization and specialization of pediatric brain tumor resection, 1988 to 2000. A and D, local likelihood fitted curves; B and C, least-squares linear regressions. A, probability of admission to a teaching hospital. It should be noted that the y axis begins at 0.5. B and C, selected caseload percentiles for nonteaching (B) and teaching (C) hospitals that performed at least one craniotomy for pediatric brain tumor resection in a given year. Caseloads increased at teaching hospitals and decreased at nonteaching hospitals. D, per-surgeon proportion of surgical practice consisting of patients 18 years of age or younger. The median increased from 10% pediatric patients to 30%, and the 75th percentile increased from 50% pediatric patients to 90%.

FIGURE 8.

Centralization and specialization of pediatric brain tumor resection, 1988 to 2000. A and D, local likelihood fitted curves; B and C, least-squares linear regressions. A, probability of admission to a teaching hospital. It should be noted that the y axis begins at 0.5. B and C, selected caseload percentiles for nonteaching (B) and teaching (C) hospitals that performed at least one craniotomy for pediatric brain tumor resection in a given year. Caseloads increased at teaching hospitals and decreased at nonteaching hospitals. D, per-surgeon proportion of surgical practice consisting of patients 18 years of age or younger. The median increased from 10% pediatric patients to 30%, and the 75th percentile increased from 50% pediatric patients to 90%.

Zero-inflated negative binomial regression (8,11,35,39,40) was performed to model the volume of craniotomies performed at individual United States hospitals in the NIS database. This model assumes that hospitals perform zero craniotomies during a given year for two classes of reasons. Some hospitals do not perform craniotomies for fixed absolute reasons, such as no affiliated neurosurgeon or no operating rooms. Others might have performed craniotomies but no suitable case presented during that year. The model predicted that clinicians at pediatric hospitals, teaching hospitals, urban hospitals, and large-bed size hospitals were more likely to be capable of offering to perform craniotomies for pediatric brain tumors in a given year (P < 0.001 for each). The pool of hospitals at which craniotomies could be performed decreased with time (P = 0.001). Among the hospitals at which the model predicted craniotomies could be performed for pediatric tumor resection, pediatric hospitals (P = 0.001), teaching hospitals (P = 0.02), and larger-bed size hospitals (P = 0.03) had larger caseloads, and there was a trend toward larger caseloads at individual hospitals with time (P = 0.14).

The median surgeon caseload increased from three admissions (in 1988–1990) to seven admissions (in 1997–2000). High-caseload surgeons tended to be pediatric specialists. The median percent pediatric practice for surgeons who performed 10 or more tumor resections in a year was 88%. Surgeons who performed pediatric brain tumor resections had practices with a greater proportion of pediatric surgery in later years in the series (P = 0.05) (Fig. 8D). For all surgeons in the database who performed pediatric tumor resections, the per-surgeon median proportion of practice that consisted of pediatric surgery increased from 11.5% (in 1988–1990) to 27% (in 1997–2000). In 1988 to 1990, 22% of surgeons had practices that included one-half or more pediatric patients, compared with 41% in 1997 to 2000. The most specialized 25% of surgeons who performed the procedure in 1988 to 1990 had practices that were at least 40% pediatric patients; by 1997 to 2000, however, 25% of surgeons had practices that were 91% or more pediatric patients. Conversely, the proportion of admissions that represented the only pediatric brain tumor resection in the database for that surgeon in that year decreased from 28% (in 1988–1990) to 20% (in 1997–2000).

Decreasing In-Hospital Mortality Rates

For cases in the database, in-hospital death became significantly less frequent with time (P = 0.01). Projected to the United States population, there was a decrease in mortality rates during the study period, from 2.7% in 1988 to 1990 to 1.2% in 2000, a 56% relative decrease (P = 0.08) (Fig. 6B). There was no difference between high- and low-volume hospitals in the relative magnitude of the mortality rate decrease with time, which was between 46% and 74% in all four volume quartiles. After adjustment for age, sex, race, primary payer, median income in the postal code of the patient's residence, admission type, geographic region, tumor histological type and location, and hospital caseload, the decrease in mortality rates with time was not statistically significant (P = 0.4). Expressed as the logarithm of the logistic slope coefficient (β), hospital caseload increases accounted for 32% of the estimated change in mortality rates with time, after adjustment for all other prognostic factors.

LOS and Hospital Charges

LOS decreased significantly during the study period (by 4.3%/yr, P < 0.001); the median was 10 days in 1988 to 1990 and 7 days in 1997 to 2000. After multivariate adjustment for the variables described above and stratification by treatment year, LOS was shorter at larger-volume hospitals, but not significantly so (P = 0.6). In a similar multivariate model, there was a trend toward shorter LOS with larger surgeon caseload (P = 0.2).

Total hospital charges increased significantly during the study period, from a median of $20,000 in 1988 to 1990 to $33,000 in 1997 to 2000 (by 6.7%/yr, P < 0.001). After multivariate adjustment for the variables described above and stratification by treatment year, charges were higher at higher-volume hospitals (charge differential for a 10-fold larger caseload, +6.3%; 95% CI, +1.6 to + 11.2%, P = 0.008). With this model, the predicted difference in charges between the smallest- and largest-caseload hospitals was 11%. If pediatric hospital status was added to the model, then hospital volume was no longer a significant predictor of higher charges. There was a trend toward lower hospital charges for patients of higher-caseload surgeons (charge differential for a 10-fold larger caseload, −4.6%; 95% CI, −10.2 to +1.4%, P = 0.1).

DISCUSSION

This analysis included 4712 admissions for pediatric brain tumor resections performed at United States nonfederal hospitals between 1988 and 2000. The in-hospital mortality rate was 1.6%, and an additional 5.7% of patients were not discharged directly to home. After multivariate adjustment for differences in case mixture, care provided by higher-volume hospitals and surgeons was followed by lower mortality rates and better hospital discharge dispositions. The differences in outcomes between high- and low-volume hospitals were largest for the youngest patients. High-volume care was more readily available to patients who resided in wealthier areas and was associated with minimal changes in hospital charges and a trend toward shorter LOS. The annual number of admissions for resection of pediatric brain tumors increased 52% during the study period, and in-hospital mortality rates decreased 56%. Both centralization and specialization of care took place during the study period.

Outcomes and Provider Volume of Care

The tendency toward better outcomes after care provided by physicians or hospitals with large current caseloads is called the volume-outcome effect. The volume-outcome effect was previously demonstrated for other intracranial procedures, such as clipping of intracranial aneurysms (5,32,55) and microvascular decompression (33), as well as in two population-based studies that addressed the surgical treatment of brain tumors (14,41). Those studies, which demonstrated lower mortality rates after procedures at high-volume hospitals or by high-volume surgeons, excluded pediatric patients and included intracranial tumors of many types, such as primary brain tumors, meningiomas, and metastatic lesions.

The volume-outcome relationship for pediatric brain tumor surgery has been examined for other end points. Albright et al. (3) correlated data on the extent of resection and neurological complications for Children's Cancer Group medulloblastoma and malignant glioma protocols with surgeons’ membership in the American Society of Pediatric Neurosurgeons or self-reported pediatric specialization. Designated specialists performed more complete resections, but data on neurological complications were mixed, with designated specialists who were not Society members having the highest complication rates and Society members the lowest. Hospital caseload effects were not studied. A survey study of 11 institutions reported by Sanford (52) concluded that more experience was correlated with more frequent good outcomes after radical resection of craniopharyngiomas. The volume-outcome effect has also been described for ventriculoperitoneal shunt surgery (7,12,17,31,42,46,54).

We now report lower in-hospital mortality rates and better hospital discharge dispositions after pediatric brain tumor resections by higher-volume hospitals and surgeons. Most of these findings remained statistically significant after case mixture adjustment using both patient- and provider-level variables, including age, sex, race, primary payer for care, median income in postal code of residence, geographic region, emergency or urgent admission, tumor histological type and location, and year of treatment. Better adjustment for baseline risk clearly could be made if detailed patient-level clinical data were available, but administrative databases such as the NIS database typically lack such data. Important risk factors, such as tumor size, details of location, and preoperative performance status, could not be included in the analysis, and biased distribution of these factors between patients treated by high- and low-volume providers could have affected the results of the study. However, most risk factors we identified, such as age younger than 2 years, malignant tumor histological type (as defined by ICD-9 coding, which considers both pilocytic astrocytomas and Grade II astrocytomas to be malignant tumors), and posterior fossa location were evenly distributed between high- and low-volume providers in our data. Emergency and urgent admissions were more common with low-volume providers. Admission type is a “soft” covariate, in that a given clinical presentation might be judged to be an emergency at one hospital and not at another; in this series, 42% of admissions to Northeastern United States hospitals were classified as emergencies, compared with 19% in the Midwest. It is likely that the differences in baseline risk represented by this covariate are imperfectly accounted for in our analysis.

Our analysis of surgeon caseload effects was potentially affected by missing surgeon identifier codes for a large proportion of the cases (55%). Surgeon identifier codes are not reported by several states, including some states with very high-volume hospitals, and our conclusions regarding surgeon volume might have been affected in unpredictable ways by the missing data. However, the cases with identified surgeons still represented approximately 10% of the annual United States caseload, and our results probably can be generalized to the United States population as a whole.

Another limitation of our study is the absence of end points reflecting the efficacy of tumor resection, such as extent of resection and long-term survival rates, and longer-term end points, such as 30-day mortality rates and functional status 1 year after surgery. These data are not contained in the NIS database. Because individual patient identifiers are removed from NIS data, adjuvant postoperative treatments such as radiotherapy and/or chemotherapy also could not be studied. In other cancer surgery settings, the receipt of generally recommended adjuvant treatments has been found to vary with various patient and provider characteristics, including volume of care (22,28,29,49).

Practice Patterns and Trends with Time

To our knowledge, population-based data have not previously been used to study the extent to which pediatric brain tumor resections are performed in relatively low-volume settings. We observed that a substantial number of children undergo craniotomies for tumor resection in practice settings where this is an uncommon event. Albright et al. (3) reported similar findings for children enrolled in Children's Cancer Group protocols.

We observed changing United States practice patterns for pediatric brain tumor craniotomies during the study period. There was a nationwide trend toward increasing total caseloads with time, and malignant tumor histological type and posterior fossa location became relatively less frequent. These changes are consistent with more frequent presentation of pediatric patients with low-grade supratentorial tumors for surgery, perhaps because of more frequent detection or an increasing tendency to resect such tumors. We cannot distinguish between these and other possible explanations with our data source.

In-hospital mortality rates decreased during the study period. This was statistically significant for the cases in the database in univariate analysis, although not after multivariate adjustment. Some of the decrease in mortality rates seemed to be attributable to a shift toward patients with fewer adverse risk factors, as described above. Increasing provider caseloads, with per-patient median hospital and surgeon caseloads approximately doubling during the study period, also explained a portion of the decrease in mortality rates.

We investigated whether the increase in provider caseloads reflected progressive centralization of pediatric brain tumor surgery. Centralization of complex surgery is often advocated when high-volume care is linked to better outcomes, and mandated centralization has sometimes been followed by improved outcomes, such as for pediatric heart surgery in Sweden in 1988 to 1997 (43) or coronary artery bypass surgery in Calgary, Canada, in 1994 to 1998 (27). Centralization of care that occurs spontaneously in relatively unregulated health care systems, such as the United States system, has been less well studied and is poorly defined. We assumed that hallmarks of progressive centralization would include a stable or decreasing total number of hospitals that perform the procedure, a decreasing proportion of hospitals that perform the procedure only rarely, and an increase in caseloads with time at hospitals that continue to perform the procedure.

We observed evidence for each of these trends. Nationwide projections from the NIS database suggested that the number of United States hospitals that offered the procedure remained stable during the study period and the proportion of hospitals at which only one procedure was performed per year decreased from 13% to 5.5%. A comparison of caseloads between the first and last years sampled for 275 hospitals demonstrated a significant increase in caseloads at individual hospitals with time (P = 0.02). A graphical exploratory subgroup analysis (Fig. 8) demonstrated that teaching hospitals had increasing caseloads during the study period, whereas caseloads at nonteaching hospitals decreased.

We used a count regression model, the zero-inflated negative binomial model, to further investigate centralization of pediatric brain tumor operations. This model was previously used to model event counts in populations in which some individuals have a count of zero for fixed reasons, whereas others may generate either a positive count or a count of zero, depending on circumstances (8,11,36,39,40). The large number of hospitals and surgeons in the NIS database each year that performed a single craniotomy for pediatric tumor resection suggests that there was an additional cohort of hospitals and surgeons that would have performed such operations if suitable cases had presented, making the zero-inflated negative binomial model a reasonable choice for modeling. The model demonstrated that the number of United States hospitals prepared to perform a craniotomy for pediatric brain tumor treatment decreased during the study period, with a trend toward larger per-hospital caseloads with time at hospitals that did offer the operation, which is additional evidence of centralization.

We also studied specialization at the individual surgeon level. Surgeons performing pediatric brain tumor resections in the early years of the series, on average (median), had practices that included approximately 10% pediatric patients, a figure that increased to almost 30% by the end of the period (Fig. 8D). The most specialized 25% of surgeons performed at least 40% pediatric surgery in 1988 to 1990, compared with at least 90% pediatric surgery in 1996 to 2000.

Similar changes in hospital-level practice patterns have been demonstrated for other operations in individual states, such as complex cancer surgery in Maryland (10,19,20), including craniotomy for adult intracranial tumors (41), and coronary artery bypass surgery in states that require certification for the operation (58), although not for carotid endarterectomy in New York in 1990 to 1995 (44). In a survey study, pediatric otolaryngologists were more likely than general otolaryngologists to report increasing numbers of referrals, in conjunction with increased complexity (57). Examples of increasing centralization or specialization on a nationwide level have not, to our knowledge, previously been described for neurosurgery. A study of neurosurgical job advertisements (1985–1998) indicated a progressive increase in postings for specialist jobs, including pediatric positions (16). The NIS database is not an ideal data source for a longitudinal study of hospital or surgeon caseloads: individual hospitals are not consistently sampled each year, all hospitals within a geographic service area are not sampled, and surgeons who operate at more than one hospital may have only part of their practice captured by the NIS database. Studies using comprehensive population-based databases for individual areas, such as those maintained by many individual states in the United States, would be valuable in confirming our findings.

Although our study offers support for the general concept of better outcomes after treatment by more experienced surgeons and hospitals, it is not directly relevant to the ongoing dialogue concerning certification of pediatric neurosurgeons (13,34,50,53). Surgical experience can be correlated with outcomes using three measures, i.e., current volume of practice, cumulative lifetime experience (the “learning curve”) (51), and subspecialization, as indicated by membership in specialized societies, by special certification, or by the percentage of actual practice in a specialized area (3). Because the NIS database identifies surgeons with a masked identifier code, classification of surgeons according to certification status was not possible in this study. Direct assessment of the effects of certification on practice patterns will require another data source.

Whether further centralization or specialization of pediatric brain tumor surgery would improve patient outcomes is an important question that merits further research. Although these operations are already restricted to a small subset of United States hospitals, per-hospital and per-surgeon median annual caseloads are presently low. Drawbacks to regionalization of care include longer travel times for patients and families, with corresponding loss of support systems and consequent stress (9,15,47). In addition, deterioration of services in remote locations (“distance decay”) (18,21) is exacerbated by regionalization schemes, partly because of the lack of experienced on-site providers and necessary services when immediate care is required in remote hospitals (1,48). Only 3% of the patients in our study had admissions classified as emergencies and underwent surgery on the day of hospital admission; of those, only three patients (0.06% of the cohort) underwent surgery in rural hospitals. More than 98% of all patients underwent surgery in urban hospitals. It is likely that many pediatric patients who currently undergo craniotomies for tumor treatment in low-volume settings within the United States are in sufficiently stable condition for transfer to nearby high-volume centers without undue risk.

CONCLUSIONS

This study included a large representative sample of pediatric patients who underwent craniotomies for resection of brain tumors in the United States between 1988 and 2000. After multivariate adjustment for other risk factors, adverse outcomes (death or adverse hospital discharge disposition) were less frequent after surgery at high-volume centers or with high-volume surgeons. Increases in the overall numbers of cases treated annually and decreases in mortality rates for centers of all sizes were also observed. The data presented here suggest that centralization of pediatric brain tumor surgery at larger-volume hospitals and progressive specialization in pediatric surgery by surgeons who perform such resections are both currently taking place in the United States. Further research is warranted to assess whether continued centralization and specialization of care might be associated with better patient outcomes on a nationwide scale.

Acknowledgment

We gratefully acknowledge helpful discussions with Paul H. Chapman, M.D.

REFERENCES

1.
Adam
DJ
,
Mohan
IV
,
Stuart
WP
,
Bain
M
,
Bradbury
AW
Community and hospital outcome from ruptured abdominal aortic aneurysm within the catchment area of a regional vascular surgical service
.
J Vasc Surg
 
30
:
922
928
,
1999
.
2.
Agency for Healthcare Research and Quality
:
Healthcare Cost and Utilization Project Nationwide Inpatient Sample: Calculating Nationwide Inpatient Sample Variances, 2000
 .
Rockville
,
Agency for Healthcare Research and Quality
,
2002
.
3.
Albright
AL
,
Sposto
R
,
Holmes
E
,
Zeltzer
PM
,
Finlay
JL
,
Wisoff
JH
,
Berger
MS
,
Packer
RJ
,
Pollack
IF
Correlation of neurosurgical subspecialization with outcomes in children with malignant brain tumors
.
Neurosurgery
 
47
:
879
887
,
2000
.
4.
American Academy of Pediatrics Surgical Advisory Panel
:
Guidelines for referral to pediatric surgical specialists
.
Pediatrics
 
110
:
187
191
,
2002
.
5.
Barker
FG
II
,
Amin-Hanjani
S
,
Butler
WE
,
Ogilvy
CS
,
Carter
BS
In-hospital mortality and morbidity after surgical treatment of unruptured intracranial aneurysms in the United States, 1996–2000: The effect of hospital and surgeon volume
.
Neurosurgery
 
52
:
995
1009
,
2003
.
6.
Begg
CB
,
Cramer
LD
,
Hoskins
WJ
,
Brennan
MF
Impact of hospital volume on operative mortality for major cancer surgery
.
JAMA
 
280
:
1747
1751
,
1998
.
7.
Borgbjerg
BM
,
Gjerris
F
,
Albeck
MJ
,
Borgesen
SE
Risk of infection after cerebrospinal fluid shunt: An analysis of 884 first-time shunts
.
Acta Neurochir (Wien)
 
136
:
1
7
,
1995
.
8.
Cameron
AC
,
Trivedi
PK
Regression Analysis of Count Data
 .
Cambridge
,
Cambridge University Press
,
1998
.
9.
Chang
RK
,
Klitzner
TS
Can regionalization decrease the number of deaths for children who undergo cardiac surgery? A theoretical analysis
.
Pediatrics
 
109
:
173
181
,
2002
.
10.
Chen
H
,
Zeiger
MA
,
Gordon
TA
,
Udelsman
R
Parathyroidectomy in Maryland: Effects of an endocrine center
.
Surgery
 
120
:
948
953
,
1996
.
11.
Cheung
YB
Zero-inflated models for regression analysis of count data: A study of growth and development
.
Stat Med
 
21
:
1461
1469
,
2002
.
12.
Cochrane
DD
,
Kestle
J
Ventricular shunting for hydrocephalus in children: Patients, procedures, surgeons and institutions in English Canada, 1989–2001
.
Eur J Pediatr Surg
 
12
[Suppl
1
]:
S6
S11
,
2002
.
13.
Couldwell
WT
,
Rovit
RL
Rethinking neurosurgical subspecialization
.
Surg Neurol
 
58
:
359
370
,
2002
.
14.
Cowan
JA
Jr
,
Dimick
JB
,
Leveque
JC
,
Thompson
BG
,
Upchurch
GR
Jr
,
Hoff
JT
The impact of provider volume on mortality after intracranial tumor resection
.
Neurosurgery
 
52
:
48
54
,
2003
.
15.
Finlayson
SR
,
Birkmeyer
JD
,
Tosteson
AN
,
Nease
RF
Jr
Patient preferences for location of care: Implications for regionalization
.
Med Care
 
37
:
204
209
,
1999
.
16.
Friedlich
DL
,
Feustel
PJ
,
Popp
AJ
Workforce demand for neurosurgeons in the United States of America: A 13-year retrospective study
.
J Neurosurg
 
90
:
993
997
,
1999
.
17.
George
R
,
Leibrock
L
,
Epstein
M
Long-term analysis of cerebrospinal fluid shunt infections: A 25-year experience
.
J Neurosurg
 
51
:
804
811
,
1979
.
18.
Goodman
DC
,
Fisher
ES
,
Gittelsohn
A
,
Chang
CH
,
Fleming
C
Why are children hospitalized? The role of non-clinical factors in pediatric hospitalizations
.
Pediatrics
 
93
:
896
902
,
1994
.
19.
Goodney
PP
,
Siewers
AE
,
Stukel
TA
,
Lucas
FL
,
Wennberg
DE
,
Birkmeyer
JD
Is surgery getting safer? National trends in operative mortality
.
J Am Coll Surg
 
195
:
219
227
,
2002
.
20.
Gordon
TA
,
Bowman
HM
,
Tielsch
JM
,
Bass
EB
,
Burleyson
GP
,
Cameron
JL
Statewide regionalization of pancreaticoduodenectomy and its effect on in-hospital mortality
.
Ann Surg
 
228
:
71
78
,
1998
.
21.
Gregory
PM
,
Malka
ES
,
Kostis
JB
,
Wilson
AC
,
Arora
JK
,
Rhoads
GG
Impact of geographic proximity to cardiac revascularization services on service utilization
.
Med Care
 
38
:
45
57
,
2000
.
22.
Grilli
R
,
Minozzi
S
,
Tinazzi
A
,
Labianca
R
,
Sheldon
TA
,
Liberati
A
Do specialists do it better? The impact of specialization on the processes and outcomes of care for cancer patients
.
Ann Oncol
 
9
:
365
374
,
1998
.
23.
Hannan
EL
,
Racz
M
,
Kavey
RE
,
Quaegebeur
JM
,
Williams
R
Pediatric cardiac surgery: The effect of hospital and surgeon volume on in-hospital mortality
.
Pediatrics
 
101
:
963
969
,
1998
.
24.
Harrell
FE
Jr
Hmisc and Design libraries for S-Plus for Windows
 . Available at: http://hesweb1.med.virginia.edu/biostat/s/splus.html. Accessed 1/16/2004.
25.
Harrell
FE
Jr
Regression Modeling Strategies: With Applications to Linear Models, Logistic Regression, and Survival Analysis
 .
New York
,
Springer
,
2001
.
26.
Hastie
TJ
,
Tibshirani
RJ
Generalized Additive Models
 .
London
,
Chapman & Hall
,
1990
.
27.
Hemmelgarn
BR
,
Ghali
WA
,
Quan
H
A case study of hospital closure and centralization of coronary revascularization procedures
.
CMAJ
 
164
:
1431
1435
,
2001
.
28.
Hewitt
M
,
Simone
JV
(eds):
Ensuring Quality Cancer Care
 .
Washington, DC
,
National Academy Press
,
1999
.
29.
Hillner
BE
,
Smith
TJ
,
Desch
CE
Hospital and physician volume or specialization and outcomes in cancer treatment: Importance in quality of cancer care
.
J Clin Oncol
 
18
:
2327
2340
,
2000
.
30.
Hosmer
DW
Jr
,
Lemeshow
S
Applied Logistic Regression
 .
New York
,
John Wiley & Sons
,
2000
,
ed 2
.
31.
Jamjoom
AB
,
Jamjoom
ZAB
,
Ur-Rahman
N
,
Mohammed
AA
,
Neel
KF
Factors affecting cerebrospinal fluid shunt revision rate in children at King Khalid University Hospital
.
Saudi Med J
 
19
:
15
18
,
1998
.
32.
Johnston
SC
,
Zhao
S
,
Dudley
RA
,
Berman
MF
,
Gress
DR
Treatment of unruptured cerebral aneurysms in California
.
Stroke
 
32
:
597
605
,
2001
.
33.
Kalkanis
SN
,
Eskandar
EN
,
Carter
BS
,
Barker
FG
II
Microvascular decompression surgery in the United States, 1996 to 2000: Mortality rates, morbidity rates, and the effects of hospital and surgeon volumes
.
Neurosurgery
 
52
:
1251
1262
,
2003
.
34.
Kline
DG
,
Mahaley
MS
Jr
Recognition of special qualifications in neurological surgery
.
J Neurosurg
 
64
:
531
536
,
1986
.
35.
Lambert
D
Zero-inflated Poisson regression with an application to defects in manufacturing
.
Technometrics
 
34
:
1
14
,
1992
.
36.
Lee
AH
,
Stevenson
MR
,
Wang
K
,
Yau
KK
Modeling young driver motor vehicle crashes: Data with extra zeros
.
Accid Anal Prev
 
34
:
515
521
,
2002
.
37.
Loader
C
Local Regression and Likelihood
 .
New York
,
Springer-Verlag
,
1999
.
38.
Loader
CR
LOCFIT for S-Plus (software and electronic documentation)
 .
Murray Hill
,
Lucent Technologies
,
1998
.
39.
Long
JS
Regression Models for Categorical and Limited Dependent Variables
 .
Thousand Oaks
,
Sage Publications
,
1997
.
40.
Long
JS
,
Freese
J
Regression Models for Categorical Dependent Variables Using Stata
 .
College Station
,
Stata Press
,
2001
.
41.
Long
DM
,
Gordon
T
,
Bowman
H
,
Etzel
A
,
Burleyson
G
,
Betchen
S
,
Garonzik
IM
,
Brem
H
Outcome and cost of craniotomy performed to treat tumors in regional academic referral centers
.
Neurosurgery
 
52
:
1056
1065
,
2003
.
42.
Lund-Johansen
M
,
Svendsen
F
,
Wester
K
Shunt failures and complications in adults as related to shunt type, diagnosis, and the experience of the surgeon
.
Neurosurgery
 
35
:
839
844
,
1994
.
43.
Lundstrom
NR
,
Berggren
H
,
Bjorkhem
G
,
Jogi
P
,
Sunnegardh
J
Centralization of pediatric heart surgery in Sweden
.
Pediatr Cardiol
 
21
:
353
357
,
2000
.
44.
Matsen
SL
,
Perler
BA
,
Brown
PM
Jr
,
Roseborough
GS
,
Williams
GM
The distribution of carotid endarterectomy procedures among surgeons and hospitals in New York State: Is regionalization of specialized vascular care occurring?
J Vasc Surg
 
36
:
1146
1153
,
2002
.
45.
McCullagh
P
Regression models for ordinal data
.
J R Stat Soc [B]
 
42
:
109
142
,
1980
.
46.
Odio
C
,
McCracken
GH
Jr
,
Nelson
JD
CSF shunt infections in pediatrics: A seven-year experience
.
Am J Dis Child
 
138
:
1103
1108
,
1984
.
47.
Payne
S
,
Jarrett
N
,
Jeffs
D
The impact of travel on cancer patients’ experiences of treatment: A literature review
.
Eur J Cancer Care (Engl)
 
9
:
197
203
,
2000
.
48.
Petersen
LA
,
Normand
SL
,
Leape
LL
,
McNeil
BJ
Regionalization and the underuse of angiography in the Veterans Affairs Health Care System as compared with a fee-for-service system
.
N Engl J Med
 
348
:
2209
2217
,
2003
.
49.
Pitchforth
E
,
Russell
E
,
Van der Pol
M
Access to specialist cancer care: Is it equitable?
Br J Cancer
 
87
:
1221
1226
,
2002
.
50.
Raimondi
AJ
The education and identification of a pediatric neurosurgeon
.
Childs Nerv Syst
 
8
:
4
7
,
1992
.
51.
Ramsay
CR
,
Grant
AM
,
Wallace
SA
,
Garthwaite
PH
,
Monk
AF
,
Russell
IT
Statistical assessment of the learning curves of health technologies
.
Health Technol Assess
 
5
:
1
79
,
2001
.
52.
Sanford
RA
Craniopharyngioma: Results of survey of the American Society of Pediatric Neurosurgery
.
Pediatr Neurosurg
 
21
[Suppl
1
]:
39
43
,
1994
.
53.
Scott
RM
The fellowship in pediatric neurosurgery
.
Pediatr Neurosurg
 
17
:
281
283
,
1991
–1992.
54.
Smith
ER
,
Butler
WE
,
Barker
FG
II
In-hospital mortality rates after ventriculoperitoneal shunt procedures in the United States, 1998 to 2000: Relation to hospital and surgeon volume of care
.
J Neurosurg
 
100
:
90
97
,
2004
.
55.
Solomon
RA
,
Mayer
SA
,
Tarmey
JJ
Relationship between the volume of craniotomies for cerebral aneurysm performed at New York State hospitals and in-hospital mortality
.
Stroke
 
27
:
13
17
,
1996
.
56.
Steiner
C
,
Elixhauser
A
,
Schnaier
J
The healthcare cost and utilization project: An overview
.
Eff Clin Pract
 
5
:
143
151
,
2002
.
57.
Tunkel
DE
,
Cull
WL
,
Jewett
EA
,
Brotherton
SE
,
Britton
CV
,
Mulvey
HJ
Practice of pediatric otolaryngology: Results of the Future of Pediatric Education II project
.
Arch Otolaryngol Head Neck Surg
 
128
:
759
764
,
2002
.
58.
Vaughan-Sarrazin
MS
,
Hannan
EL
,
Gormley
CJ
,
Rosenthal
GE
Mortality in Medicare beneficiaries following coronary artery bypass graft surgery in states with and without certificate of need regulation
.
JAMA
 
288
:
1859
1866
,
2002
.
59.
World Health Organization
:
International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM): Volume 2, Diseases
 .
Ann Arbor
,
Commission on Professional and Hospital Activities
,
1978
.

COMMENTS

This is an excellent population-based study looking at the association between provider and institutional caseload and outcome for children undergoing resection for brain tumors. Although there are inherent limitations in the coding system used by the database, particularly in terms of histological categorization of the tumors, the statistical analyses of the available data seem to be outstanding and support the logical conclusion that outcomes improve with experience. Other interesting points are that there was a trend over time toward centralization of care at high-volume centers and a small but significant reduction in mortality over time, something that might have been missed in a smaller data set.

Ian F. Pollack

Pittsburgh, Pennsylvania

This is study of a national database of pediatric brain tumors. The authors identified the subjects from International Classification of Diseases, 9th Revision, codes and then looked at mortality and type of discharge as a function of surgeon volume and hospital volume of pediatric craniotomies. The major conclusions were that 1) surgeons with higher caseloads of pediatric brain tumors had a marginally better postoperative mortality and morbidity rate, 2) this was most apparent in the youngest children, and 3) over time there seems to be an increasing concentration of these cases in specialty centers.

This is potentially significant and supports the concept of specialization of pediatric neurosurgeons. Obviously, the better outcomes described are not entirely attributable to the surgeon but are also related to neurointensive care, pediatric anesthesia, and other factors that would differentiate a pediatric neurosurgical center from a community adult-oriented facility.

I am surprised that, in the overall population here, 82% of the tumors were identified as malignant. In a nonselected series of childhood tumors, one would expect something on the order of 50%. I suspect that this was because of miscoding. The Current Procedural Terminology codes may have been entered at admission rather than discharge, and histology was uncertain. In our office, for instance, all new patients are precertified under the 225.0 (benign) code, so that “malignant” does not appear on the billing statement. Also, the procedure codes as they appear here are confusing; normally, one would use Current Procedural Terminology codes (61510 supratentorial, 61518 infratentorial, 61545 craniopharyngioma).

Nonetheless, the same methodology was applied across the board, so that even if it is flawed, the conclusions are probably still valid. If anything, the effects that have been observed are probably even more striking, because the more difficult cases (craniopharyngiomas, pineal tumors, etc.) are probably referred to the pediatric neurosurgical centers.

Leslie N. Sutton

Philadelphia, Pennsylvania

Dr. Barker and colleagues at the Massachusetts General Hospital have contributed yet another well-written analysis of neurosurgical practices. The use of databases such as that of the Agency for Healthcare Research and Quality is becoming more widespread among epidemiologists and researchers around the country, and database analysis has been applied successfully in the examination of traumatic injuries and subsequent implementation of injury-prevention paradigms (1). This is an important article, for two reasons. First, the article simply sought to analyze volume-outcome relationships for pediatric tumors and trends toward centralization and specialization. It is a relatively clever method of noting trends in large numbers of patients on the basis of uniformly collected data. Some incredulous surgeons may simply argue that the actual data are flawed, but with the large number of pediatric tumor cases captured in this study and the specific questions posed, that criticism is unfounded. As predicted, the centralization of pediatric brain tumor operations and the specialization among neurosurgeons have already occurred. This basic observation is worth noting even if most of us would have predicted that by reviewing our local/regional practices. Realistically, this is most likely a reflection of economics, as opposed to outcome-directed medical practices by neurosurgeons. The authors’ second contribution is to correctly infer that with continued centralization and specialization, additional specific studies will be necessary to establish whether that specialized care resulted in better patient outcomes. That constitutes an important but thornier question to answer; however, the answer has the potential to change practice patterns in perpetuity.

Richard G. Ellenbogen

Seattle, Washington

1
Rivara FP: Introduction: The scientific basis for injury control. Epidemiol Rev 25: 20–23, 2003.

Using the Nationwide Inpatient Sample (NIS), the authors performed a multivariant analysis adjusted for a host of factors correlated with mortality and adverse discharge disposition rates for pediatric craniotomies for brain tumor during the years 1988–2000. During this time, the authors noted a progressive centralization, with care being shifted toward large-caseload hospitals and to neurosurgeons whose practice was predominantly pediatric. Mortality and adverse discharge disposition rates were lower with high-volume hospitals with neurosurgeons.

As the authors state in their introduction, there has been ample evidence to show that with a number of complex surgical procedures, increasing caseloads are associated with lower morbidity and mortality, the so-called volume-outcome effect. Interestingly, only 5% of United States hospitals performed one or more craniotomies for pediatric brain tumors during the period studied. In their Discussion, the authors indicate that although the number of United States hospitals able to perform a craniotomy for a patient with a pediatric brain tumor diminished to some degree during the time period studied, the fact that a large number of hospitals and neurosurgeons in the database performed only a single craniotomy for a pediatric brain tumor indicates that there were hospitals and neurosurgeons who would have performed such operations if suitable patients had been available to them. This suggests that referral patterns are a definite factor in the centralization of pediatric brain tumor cases.

The authors obtained their data from the NIS hospital discharge base, which represents 20% of inpatient admissions to nonfederal hospitals in the United States. The authors note that this database is “not an ideal data source for a longitudinal study of hospital or surgeon caseloads; individual hospitals are not consistently sampled each year, all hospitals within a geographic service area are not sampled, and surgeons who operate at more than one hospital may have only part of their practice captured by the NIS database.” Neurosurgeon identifier codes were not available for 55% of the cases, including those in several states with very-high-volume hospitals. A masked identifier code is used to denote each neurosurgeon and does not allow for further classification as to training, experience, or certification. However, knowing the percentage of patients under 18 years of age admitted by a given neurosurgeon allowed the authors to note what percentage of a neurosurgeon's patient population was considered to be pediatric. The authors could not study and evaluate the extent of tumor resection, long-term survival, or neurological function. Also, they did not know whether the pediatric patients received radiation or chemotherapy.

In their Table 1, with regard to clinical characteristics of the patients, the authors note that only 2.2% were Asian or from the Pacific Islands. Being from California probably gives me a different view of breakdown as to race, but 2.2% even for the country as a whole seems a little low. The primary diagnosis for a benign brain tumor was found to be only 9%. As the authors explain, this percentage is artificially low because the International Classification of Diseases, 9th Revision—Clinical Modification, classifies all astrocytomas, including pilocytic and low-grade, as being malignant. The authors also found that increasing caseload had even a more significant effect on mortality if the patients were less than 2 years old. Even though the data are less than perfect, the authors make their case that volume influences outcome.

J. Gordon McComb

Los Angeles, California

This image, taken on January 16, 2004, by the front hazard identification camera on the Mars Exploration Rover Spirit, shows the rover's robotic arm, or instrument deployment device. The arm was deployed from its stowed position beneath the “front porch” of the rover body early that morning. This is the first use of the arm to deploy the microscopic imager, one of four geological instruments located on the arm. The instrument will help scientists analyze and understand Martian rocks and soils through very high-resolution, close-up images. (Courtesy, NASA/JPL/US Geological Survey.)

This image, taken on January 16, 2004, by the front hazard identification camera on the Mars Exploration Rover Spirit, shows the rover's robotic arm, or instrument deployment device. The arm was deployed from its stowed position beneath the “front porch” of the rover body early that morning. This is the first use of the arm to deploy the microscopic imager, one of four geological instruments located on the arm. The instrument will help scientists analyze and understand Martian rocks and soils through very high-resolution, close-up images. (Courtesy, NASA/JPL/US Geological Survey.)