Abstract
Background:
Recent literature suggests that volume-outcome associations for pancreatectomy have attenuated over time, leading some researchers to question the continued relevance of volume thresholds. However, this perceived attenuation may reflect the methodological limitations of single, binary cutoffs rather than a true weakening of the underlying relationship. Stratum-specific likelihood ratios (SSLRs) generate multiple empirically derived volume strata that may detect persistent gradients that are obscured by binary stratification. The objectives of this study were to (i) define volume strata for pancreatectomy using SSLR, (ii) assess the robustness of these strata across multiple outcomes, and (iii) examine whether the association persists in modern cohorts.
Methods:
Patients who underwent pancreatectomy between 2004 and 2021 were identified using the National Cancer Database. The volume strata were defined by SSLR based on the 90-day postoperative mortality. The temporal threshold stability was assessed by stratified outcome analysis (chi-squared test).
Results:
Overall, 61,920 patients underwent pancreatectomy at 982 facilities with a 90-day mortality rate of 5.4%. SSLR analysis yielded 6 volume strata: ≤3, 4 to 9, 10 to 20, 21 to 47, 48 to 120, and ≥121 procedures/year with decreasing 90-day mortality (≤3: 11%; 4–9: 7.3%; 10–20: 6.1%; 21–47: 4.2%; 48–120: 3.3%; ≥121: 2.3%; P < .001). The optimized threshold of 21 procedures/year was identified based on SSLR. Temporal stratification into 5-year periods (2006–2010, 2011–2015, and 2016–2020) demonstrated persistent volume-outcome associations across volume strata (P < .001).
Conclusion:
SSLR reveals persistent volume-outcome associations for pancreatectomy across multiple empirically derived strata, even in contemporary data. These findings suggest that reports of an attenuated volume-outcome relationship may reflect limitations of single, static cutoffs rather than true weakening of this association.
Keywords: Hospital volume, Pancreatectomy, Volume-outcome relationship
Introduction
Higher operative volumes are associated with better patient outcomes for complex surgical procedures, including resection for malignant disease of the pancreas [1-8]. This volume-outcome relationship has informed quality reporting efforts and centralization of care discussions over the past decade [2,6]. The Leapfrog Group, a healthcare quality organization, currently recommends a minimum annual volume threshold of 20 pancreatic resections annually, a singular volume cutoff that has remained largely unchanged since its initial adoption despite substantial evolution in perioperative care [9,10]. However, a growing body of literature has questioned whether this volume-outcome relationship is still relevant in contemporary practice. Several recent studies have suggested that improvements in perioperative care have attenuated the volume-outcome relationship over time, leading some researchers to conclude that current volume thresholds may no longer be meaningful or that the relationship itself has weakened [11-14]. If true, singular cutoffs derived from historical data may inadequately reflect the current relationship between hospital volume and patient outcomes.
Reports of an attenuated volume-outcome relationship may alternatively reflect the methodological limitations that are inherent to a single, static volume cutoff rather than a true weakening of the volume-outcome association. Previous studies on volume-outcome relationship have largely used binary thresholds, such as the Leapfrog Group’s recommended 20-case threshold, or arbitrary groups, such as tertiles and quartiles, that do not reflect natural breaks in outcome data [6,9,10,15,16]. These approaches may obscure trends in the volume-outcome relationship, particularly as overall outcomes improve and mortality differences narrow.
Stratum-specific likelihood ratios (SSLRs) offer an alternative approach for generating multiple empirically derived volume strata based on points in the data in which the risk of adverse outcomes changes most precipitously [17,18]. Unlike methods that use predetermined cutoffs, SSLR identifies natural groups in outcome data without complex analytical assumptions [19,20]. SSLR has been used to derive reproducible volume strata in other fields, such as orthopedic surgery. However, data on complex abdominal surgery are lacking [17-22].
The current study hypothesizes that the volume-outcome relationship for pancreatectomy persists and that the apparent attenuation reported in recent literature reflects the limitations of a single, static volume threshold. Therefore, the objectives of this study were to (i) define hospital-level volume thresholds for pancreatectomy using SSLR based on 90-day operative mortality, (ii) explore the robustness of SSLR strata across multiple outcome measures, and (iii) examine whether the volume-outcome relationship persists across temporal cohorts, directly testing whether the relationship has attenuated in more recent years.
Methods
Patient selection
Patients with malignant disease of the pancreas who underwent a pancreatic resection from 2004 to 2021 were retrospectively identified in the National Cancer Database (NCDB). The NCDB is a joint project of the American Cancer Society and the Commission on Cancer (CoC) of the American College of Surgeons and captures approximately 70% of all newly diagnosed malignancies in the United States annually [23]. Patients were excluded from this study if they had their index surgery performed at a different facility or if they had metastatic disease on pathology or clinical staging variables. A pancreatic resection was defined as any partial or total pancreatic resection corresponding to Surgery of the Primary Site codes 30 to 80. This study was deemed nonhuman subjects research by the Indiana University Institutional Review Board.
Volume strata
For each facility, the surgical volume was measured as the total number of pancreatic resections captured in the NCDB divided by the number of years in which the facility had at least 1 pancreatic resection captured in the NCDB. This was rounded to the nearest whole number. In addition, the facilities were grouped based on facility volume thresholds, with whole-number groups used as individual strata for the SSLR computation. SSLR was performed using 90-day mortality as the primary outcome. The SSLR methodology has been described elsewhere [24]. In brief, the likelihood ratios and CIs for each outcome were calculated in each volume stratum. Adjacent strata with overlapping CIs were iteratively collapsed until there remained a set of strata and corresponding likelihood ratios with nonoverlapping CIs [17,18,22].
Outcomes
Of note, 90- and 30-day mortality were analyzed across identified volume thresholds. A textbook outcome for pancreatic resection was considered if a patient met all 4 of the following criteria: length of stay of < 14 days (approximately 75th percentile in the current cohort), negative (R0) margin, no 90-day mortality, and no unplanned readmission within 30 days. This definition was chosen as a more robust outcome than mortality alone based on a representative outcome of past literature, considering the availability of data within the NCDB [25-28]. Each outcome was assessed for the overall cohort and reported. Patients were divided into volume groups based on the volume thresholds identified at the facility in which they underwent their surgical resection. Patient demographics, crowfly (linear travel distance in miles), and hospital factors were analyzed and compared across the groups.
To evaluate the effect of the implementation of volume-based thresholds on the number of patients displaced, specific patients at low-volume facilities were “displaced” from their care. Patients displaced were calculated as the number of patients that would not receive care at a specific low-volume center if it closed based on a volume threshold established. Therefore, these patients would be “displaced” and either received care at an alternative location or received no care [29]. The number needed to refer was calculated by the inverse of the absolute risk reduction in 90-day mortality.
Change over time
The cohort was divided into 3 groups based on years of diagnosis (2006–2010, 2011–2015, and 2016–2020). The volume for each hospital was recalculated for each 5-year time frame. Textbook outcome, 30-day mortality, and 90-day mortality were assessed across volume thresholds established from SSLR analysis to assess the consistency of relationships over time. The chi-squared test was used to assess the relationship of outcomes, and the Cochran-Armitage test was used to assess the trend in outcomes across the strata [30,31].
Identification of singular threshold
Using SSLR-derived volume groups, a singular threshold was deemed optimal where the likelihood ratios for 90-day mortality crossed 1 [18]. The cohort was divided at this singular threshold into low and high volumes. Comparisons of the number of patients, number of hospitals, and patient outcomes were described for the 2 groups.
Sensitivity analysis using pancreatoduodenectomies
To further validate the SSLR model, a sensitivity analysis was performed using only pancreatic resections that would be considered pancreatoduodenectomies (PDs; NCDB Surgery of Primary Site codes 35, 36, 37, and 70). The volume strata were developed using the same SSLR methodology. Outcomes and variables were similarly analyzed.
Statistical analysis
The Kruskal-Wallis rank-sum test was used to compare continuous variables, and the chi-squared test was used to compare categorical variables. All statistical analyses were performed using R statistical software (R Foundation for Statistical Computing) and Stata software (StataCorp LLC). A P value of < .05 was considered statistically significant.
Results
There were 61,920 patients identified at 982 unique facilities who underwent a pancreatic resection for malignant disease of the pancreas. Most patients were male (31,512 [51%]), non-Hispanic White (47,974 [77%]), from a metropolitan area (49,701 [85%]), and on Medicare insurance (32,237 [52%]) (Table 1). In addition, the average patient traveled 61 miles (SD, 174) for their care. Most tumors were located in the head of the pancreas (37,032 [60%]), followed by the tail of the pancreas (11,599 [19%]).
Table 1.
Sample patient characteristics across volume thresholds.
| Characteristic | Volume strata | P value | ||||||
|---|---|---|---|---|---|---|---|---|
| Overall | Strata: 1–3 | Strata: 4–9 | Strata: 10–20 | Strata: 21–47 | Strata: 48–120 | Strata: ≥121 | ||
| N | 61,920 | 3940 | 11,187 | 14,242 | 19,741 | 9562 | 3248 | |
| Sex, male | 31,512 (51.0) | 2004 (51.0) | 5569 (50.0) | 7208 (51.0) | 10,096 (51.0) | 4958 (52.0) | 1677 (52.0) | .058 |
| Age group | < .001 | |||||||
| < 50 y | 5956 (9.6) | 385 (9.8) | 1026 (9.2) | 1284 (9.0) | 2033 (10.0) | 890 (9.3) | 338 (10.0) | |
| 50–60 y | 11,853 (19.0) | 798 (20.0) | 2170 (19.0) | 2601 (18.0) | 3858 (20.0) | 1813 (19.0) | 613 (19.0) | |
| 60–70 y | 20,274 (33.0) | 1226 (31.0) | 3629 (32.0) | 4643 (33.0) | 6577 (33.0) | 3187 (33.0) | 1012 (31.0) | |
| 70–80 y | 18,146 (29.0) | 1168 (30.0) | 3282 (29.0) | 4390 (31.0) | 5576 (28.0) | 2769 (29.0) | 961 (30.0) | |
| ≥80 y | 5691 (9.2) | 363 (9.2) | 1080 (9.7) | 1324 (9.3) | 1697 (8.6) | 903 (9.4) | 324 (10.0) | |
| Race ethnicity | < .001 | |||||||
| Asian | 1834 (3.0) | 138 (3.5) | 393 (3.5) | 434 (3.0) | 573 (2.9) | 166 (1.7) | 130 (4.0) | |
| Hispanic | 3502 (5.7) | 311 (7.9) | 869 (7.8) | 972 (6.8) | 986 (5.0) | 255 (2.7) | 109 (3.4) | |
| Non-Hispanic Black | 6278 (10.0) | 450 (11.0) | 1475 (13.0) | 1556 (11.0) | 1831 (9.3) | 713 (7.5) | 253 (7.8) | |
| Non-Hispanic White | 47,974 (77.0) | 2851 (72.0) | 7954 (71.0) | 10,785 (76.0) | 15,650 (79.0) | 8065 (84.0) | 2669 (82.0) | |
| Other/unknown | 2332 (3.8) | 190 (4.8) | 496 (4.4) | 495 (3.5) | 701 (3.6) | 363 (3.8) | 87 (2.7) | |
| Metropolitan status | < .001 | |||||||
| Metro | 49,701 (85.0) | 3290 (86.0) | 9311 (86.0) | 11,856 (86.0) | 15,657 (85.0) | 6907 (82.0) | 2680 (93.0) | |
| Rural | 950 (1.6) | 76 (2.0) | 234 (2.2) | 192 (1.4) | 300 (1.6) | 130 (1.5) | 18 (0.6) | |
| Urban | 7660 (13.0) | 481 (13.0) | 1286 (12.0) | 1780 (13.0) | 2525 (14.0) | 1411 (17.0) | 177 (6.2) | |
| Income quartile | < .001 | |||||||
| < $46,277 | 8220 (15.0) | 584 (17.0) | 1650 (18.0) | 2133 (17.0) | 2490 (14.0) | 1194 (16.0) | 169 (5.3) | |
| $46,277–$57856 | 11,057 (21.0) | 862 (25.0) | 2011 (21.0) | 2576 (20.0) | 3570 (21.0) | 1693 (23.0) | 345 (11.0) | |
| $57,857–$74,062 | 12,853 (24.0) | 874 (25.0) | 2389 (25.0) | 2979 (23.0) | 4323 (25.0) | 1780 (24.0) | 508 (16.0) | |
| ≥$74,063 | 21,427 (40.0) | 1126 (33.0) | 3362 (36.0) | 5023 (40.0) | 6997 (40.0) | 2746 (37.0) | 2173 (68.0) | |
| Insurance status | < .001 | |||||||
| Private | 23,304 (38.0) | 1308 (33.0) | 3804 (34.0) | 5182 (36.0) | 7716 (39.0) | 3780 (40.0) | 1514 (47.0) | |
| Uninsured + Medicaid | 4531 (7.3) | 465 (12.0) | 1171 (10.0) | 1063 (7.5) | 1304 (6.6) | 480 (5.0) | 48 (1.5) | |
| Medicare | 32,237 (52.0) | 2064 (52.0) | 5915 (53.0) | 7580 (53.0) | 10,274 (52.0) | 4814 (50.0) | 1590 (49.0) | |
| Other/unknown | 1848 (3.0) | 103 (2.6) | 297 (2.7) | 417 (2.9) | 447 (2.3) | 488 (5.1) | 96 (3.0) | |
| Great circle distance | < .001 | |||||||
| Median (IQR) | 19 (8–50) | 7 (3–17) | 10 (5–24) | 15 (7–34) | 25 (11–62) | 47 (17–110) | 42 (19–129) | |
| Charlson-Deyo score | < .001 | |||||||
| 0 | 39,336 (64.0) | 2406 (61.0) | 7019 (63.0) | 8953 (63.0) | 12,585 (64.0) | 6170 (65.0) | 2203 (68.0) | |
| 1 | 15,963 (26.0) | 1071 (27.0) | 2848 (25.0) | 3699 (26.0) | 5081 (26.0) | 2435 (25.0) | 829 (26.0) | |
| 2 | 4262 (6.9) | 301 (7.6) | 831 (7.4) | 1010 (7.1) | 1336 (6.8) | 624 (6.5) | 160 (4.9) | |
| ≥3 | 2359 (3.8) | 162 (4.1) | 489 (4.4) | 580 (4.1) | 739 (3.7) | 333 (3.5) | 56 (1.7) | |
| Tumor location | < .001 | |||||||
| Head | 37,032 (60.0) | 2057 (52.0) | 6449 (58.0) | 8585 (60.0) | 11,945 (61.0) | 5965 (62.0) | 2031 (63.0) | |
| Body | 6276 (10.0) | 377 (9.6) | 1143 (10.0) | 1442 (10.0) | 1967 (10.0) | 993 (10.0) | 354 (11.0) | |
| Tail | 11,599 (19.0) | 1034 (26.0) | 2229 (20.0) | 2590 (18.0) | 3638 (18.0) | 1573 (16.0) | 535 (16.0) | |
| Other/overlapping | 7013 (11.0) | 472 (12.0) | 1366 (12.0) | 1625 (11.0) | 2191 (11.0) | 1031 (11.0) | 328 (10.0) | |
Data are presented as number (percentage), unless otherwise specified.
Statistically significant findings are noted in bold.
Volume strata
The overall 90-day mortality rate for this cohort was 5.4%. SSLR analysis yielded 6 volume strata: ≤3, 4 to 9, 10 to 20, 21 to 47, 48 to 120, and ≥121 cases per year (Table 2). Of note, the 90-day mortality decreased across the volume strata (≤3: 11.0%; 4–9: 7.3%; 10–20: 6.1%; 21–47: 4.2%; 48–120: 3.3%; ≥121: 2.3%; P < .001).
Table 2.
SSLR volume cutoffs and subsequent volume strata.
| Volume cutoff | Volume strata | SSLR | Low CI | High CI |
|---|---|---|---|---|
| 1 | 1–3 | 2.18 | 1.98 | 2.40 |
| 4 | 4–9 | 1.39 | 1.30 | 1.48 |
| 10 | 10–20 | 1.14 | 1.07 | 1.21 |
| 21 | 21–47 | 0.78 | 0.73 | 0.82 |
| 48 | 48–120 | 0.61 | 0.55 | 0.68 |
| 121 | 121+ | 0.41 | 0.32 | 0.51 |
SSLR, stratum-specific likelihood ratio.
A singular threshold was identified at 21 cases per year (in which the likelihood ratio becomes < 1).
Patient analysis
Across volume groups, NCDB-reported patient sex did not differ significantly (P > .05) (Table 1). Although age groups differed statistically across volume groups, these differences were not clinically significant. Non-Hispanic White patients were more likely to be seen at high-volume facilities than at low-volume facilities (≥121: 82%; 48–120: 84%; 21–47: 79%; 10–20: 76%; 4–9: 71%; 1–3: 72%; P < .001 for all thresholds). In addition, patients seen at a high-volume facility were more likely to be in a metropolitan area (≥121: 93%; 48–120: 82%; 21–47: 85%; 10–20: 86%; 4–9: 86%; 1–3: 86%; P < .001 for all thresholds). Patients with private insurance were more likely to be treated at high-volume facilities than at low-volume facilities (≥121: 47%; 48–120: 40%; 21–47: 39%; 10–20: 36%; 4–9: 34%; 1–3: 33%; P < .001 for all thresholds). High-volume facilities had greater distances traveled than low-volume facilities (Table 1).
Most facilities were in the low-volume strata (1–3) (Fig. 1A). However, most patients were treated at the 21- to 47-volume strata (Fig. 1B). This trend was consistent in the overall cohort and when the cohort was divided into three 5-year groups. A decreasing number of facilities and patients were noted in the lowest-volume group over time, with corresponding increases in the patient cohorts in the greater-volume thresholds (Fig. 1). If volume thresholds were enforced in this cohort, implementing a 4-case-per-year minimum would displace 3940 patients (6.4%).
Figure 1.
Hospital and patients within each volume strata over time. A, Hospitals within each volume thresholds (percentage of total). B, Patients within each volume threshold (percentage of total).
Outcomes
The outcomes improved across all strata for 30-day mortality (≤3: 6.2%; 4–9: 3.8%; 10–20: 3.0%; 21–47: 2.1%; 48–120: 1.3%; ≥121: 0.7%; P < .001) (Table 3). In addition, textbook outcome increased across each threshold (≤3: 50%; 4–9: 56%; 10–20: 59%; 21–47: 64%; 48–120: 66%; ≥121: 69%; P < .001).
Table 3.
Outcomes in overall cohort.
| Volume strata | 30-d mortality (n = 58,340) |
90-d mortality (n = 58,177) |
Textbook outcome (n = 57,239) |
|---|---|---|---|
| All patients | 1555 (2.7) | 3124 (5.4) | 35,028 (61.0) |
| 1–3 | 231 (6.2) | 408 (11.0) | 1828 (50.0) |
| 4–9 | 396 (3.8) | 751 (7.3) | 5720 (56.0) |
| 10–20 | 399 (3.0) | 802 (6.1) | 7755 (59.0) |
| 21–47 | 389 (2.1) | 787 (4.2) | 11,692 (64.0) |
| 48–120 | 118 (1.3) | 305 (3.3) | 5868 (66.0) |
| ≥121 | 22 (0.7) | 71 (2.3) | 2165 (69.0) |
Data are presented as number (percentage), unless otherwise specified.
Temporal relationships
Temporal stratification into 5-year periods (2006–2010, 2011–2015, and 2016–2020) demonstrated persistent volume-outcome associations across volume strata for 30-day mortality, 90-day mortality, and textbook outcome (all P < .001) (Fig. 2). Each volume group showed improvement over time for all outcomes (eg, 30-day mortality trended down from 2006–2010 to 2015–2020 for all volume groups).
Figure 2.
Volume-outcome associations across volume strata.
Singular threshold identification
A singular volume threshold was identified where the likelihood ratio became < 1 within the SSLR analysis at 21 cases per year (Table 2). There were more hospitals (91% vs 9%) and fewer patients overall (47% vs 53%) in the low-volume group (< 21 cases per year) than in the high-volume group (≥21 cases per year). All outcomes were significantly better in the high-volume group than in the low- volume group (30-day mortality: 1.7% vs 3.8%, respectively; P < .001; 90-day mortality: 3.8% vs 7.2%, respectively; P < .001; textbook outcome: 65.3% vs 56.6%, respectively; P < .001). Implementing a minimum case threshold of 21 cases per year would displace 29,369 patients (47%).
Sensitivity analysis using PDs
A total of 38,480 patients (62% of the total cohort) underwent PD at 868 facilities, with an overall 90-day mortality rate of 6.0% (n = 2172). SSLR analysis yielded the following volume strata: 1, 2, 3 to 6, 7 to 15, 16 to 71, and ≥72. Significant improvements in 30-day mortality were observed across all strata (1: 11.0%; 2: 6.3%; 3–6: 4.3%; 7–15: 3.1%; 16–71: 2.0%; ≥72: 0.6%; P < .001). Similarly, 90-day mortality (1: 18.0%; 2: 11.0%; 3–6: 8.3%; 7–15: 6.3%; 16–71: 4.3%; ≥72: 2.3%; P < .001) and textbook outcome (1: 38.0%; 2: 43.0%; 3–6: 50.0%; 7–15: 56.0%; 16–71: 61.0%; ≥72: 67.0%; P < .001) improved across all strata (Table 4). SSLR yielded a singular volume threshold at 16 PDs/year. When stratifying into three 5-year periods (2006–2010, 2011–2015, and 2016–2020), volume-outcome associations persisted across volume strata for all 3 outcomes (P < .001 across all outcomes) (Supplementary Figure Panels A-C).
Table 4.
Outcomes in the cohort with pancreatoduodenectomy.
| Volume strata | 30-d mortality (n = 36,588) |
90-d mortality (n = 36,500) |
Textbook outcome (n = 36,063) |
|---|---|---|---|
| All patients | 1080 (3.0) | 2172 (6.0) | 20,570 (57.0) |
| 1 | 59 (11.0) | 94 (18.0) | 192 (38.0) |
| 2 | 104 (6.3) | 184 (11.0) | 694 (43.0) |
| 3–6 | 285 (4.3) | 551 (8.3) | 3321 (50.0) |
| 7–15 | 287 (3.1) | 585 (6.3) | 5082 (56.0) |
| 16–71 | 333 (2.0) | 711 (4.3) | 9893 (61.0) |
| ≥72 | 12 (0.6) | 47 (2.3) | 1388 (67.0) |
Data are presented as number (percentage), unless otherwise specified.
Discussion
The relevance of the historical volume-outcome association for procedures, such as pancreatectomy, has been increasingly debated, with recent literature suggesting that the static thresholds, such as the Leapfrog Group’s 20-case standard, may no longer apply to contemporary practice. This study demonstrated that robust volume-outcome associations persist for pancreatectomy when examined using empirically derived SSLR strata. SSLR identified 6 distinct volume thresholds with progressively decreasing 90-day mortality. As volume strata increased, the secondary outcome measures also improved, with a decrease in 30-day mortality and improved textbook outcomes. In addition, the volume-outcome association remained significant across all three 5-year periods examined, challenging recent assertions that the volume-outcome relationship has attenuated in modern clinical practice.
Previous approaches used to define volume thresholds have relied on tertiles or expert consensus and imposed arbitrary groupings rather than identifying natural breaks in outcome data [6,15,16]. Cubic splines and other modeling approaches require analyst inputs that limit reproducibility and contribute to the wide variation in current thresholds, ranging from 16 to 105 pancreatectomy cases annually [2,32-36]. SSLR yielded 6 distinct volume strata, each representing a statistically and clinically meaningful change in the mortality risk. The singular optimal threshold identified at 21 cases/year closely reflects the Leapfrog Group’s standard, providing empirical support for this benchmark while also demonstrating that outcomes continue to improve at higher-volume hospitals beyond this single binary cutoff. In addition, the SSLR strata derived in this study demonstrated consistent associations across multiple outcome measures, including 30-day mortality, 90-day mortality, and textbook outcome, with progressive improvement observed at each volume strata for all 3 measures. This consistency across outcomes suggests that the strata reflect meaningful differences in the overall quality of care rather than associations driven by a single outcome metric.
Several recent studies have concluded that the volume-outcome relationship has attenuated over time. Jogerst et al. [12] observed a narrowing mortality difference between low- and high-volume hospitals from 2004 to 2013, suggesting that improvements in perioperative care have weakened the volume-outcome association. In addition, a recent study similarly questioned the continued applicability of minimum case volume recommendations for complex oncologic surgery. The analysis examined multiple procedures, including pancreatectomy, and concluded that the diminishing mortality differences between volume strata challenge the relevance of fixed thresholds, such as those used by the Leapfrog Group [14]. This work builds on previous studies from the same group demonstrating that the Leapfrog Group’s volume standards have limited ability to discriminate hospital performance with proposed thresholds [14]. The current analysis directly tested this by examining the volume-outcome associations across 3-time cohorts. Significant outcome gradients persisted across all time periods, including the most recent years. Although mortality rates improved across all strata over time, which is consistent with previous literature, the relative differences between volume strata remained. These findings suggest that the attenuation observed in previous studies may reflect the limitations of binary or tertile-based volume stratification rather than the weakening of the underlying volume-outcome relationship. When multiple empirically derived volume strata are used rather than singular cutoffs, the volume-outcome association remains even as overall outcomes improve.
The persistent volume-outcome association across empirically derived strata has several clinical implications. For quality reporting, these findings suggest that binary classifications based on singular thresholds may obscure meaningful variation across different volume hospitals. Reporting frameworks that incorporate multiple volume strata could provide more granular and informative quality benchmarks. For research, SSLR-derived strata offer an empirically derived alternative to arbitrary tertile or quartile approaches for volume-stratified analysis and are more reproducible than expert consensus. In addition, the consistency of the SSLR-derived strata over time suggests that previously established volume thresholds remain relevant in modern clinical practice. However, policy metrics that use multiple volume strata and conceptualize volume on a continuum rather than a binary threshold may more accurately reflect the volume-outcome relationship.
This study has several limitations. First, this was a retrospective review of the NCDB data. NCDB lacks certain specific variable characteristics, such as hospital-specific information (bed count, nurse-to-patient ratios, intensive care unit availability, etc.), that likely mediate the volume-outcome relationship through mechanisms, such as failure-to- rescue capabilities. The NCDB does not have surgeon-level volume and lacks historical data on minimally invasive approaches before 2010, limiting these additional analyses. In addition, the NCDB is limited to CoC facilities, which may exclude facilities performing pancreatic resections for nonmalignant disease of the pancreas and nonmalignant indications. However, a recent study demonstrated that > 85% of pancreaticoduodenectomies are performed for malignant indications [37]. The NCDB may lack representation from additional low-volume institutions, biasing the results to higher-volume centers. If non-CoC low-volume institutions had consistent outcomes as the low-volume facilities in this study, the results demonstrating high-volume centers’ superior outcomes would be even more profound in favor of higher-volume facilities having better outcomes. Second, we initially derived SSLR from an unadjusted model. In post hoc analyses, we also derived SSLRs from risk-adjusted outcomes models using multivariate regression. In addition, risk-adjusted SSLRs demonstrated decreasing odds of mortality as surgical volume increased. Third, volume thresholds are limited to yearly analysis. This may skew data to some facilities being labeled as low volume if the facility performed more in the first or last quarter of a year. Finally, the textbook outcome is a generalized term, with variable definitions [25-28]. The results in this study may have been altered with the use of a different definition of textbook outcome.
Conclusion
SSLR analysis identified persistent, robust, volume-outcome associations for pancreatectomy across 6 empirically derived strata, which were consistent across multiple outcome measures and across temporal cohorts. These findings suggest that reports of an attenuated volume-outcome relationship may reflect methodological limitations of a single, static volume cutoff rather than a true weakening of the volume-outcome association.
Supplementary Material
Funding
Kristen N. Kaiser and Alexa J. Hughes are supported on a training grant from the National Cancer Institute (grant number: NCI T32CA282070). The funding agency has no relationship to the current work. None of the authors have financial disclosures.
Appendix A. Supporting information
Supplementary data associated with this article can be found in the online version at doi:10.1016/j.gassur.2026.102349.
Footnotes
Declaration of competing interest
The authors declare no competing interests.
References
- [1].Birkmeyer JD, Stukel TA, Siewers AE, Goodney PP, Wennberg DE, Lucas FL. Surgeon volume and operative mortality in the United States. N Engl J Med 2003;349(22):2117–27. 10.1056/NEJMsa035205 [DOI] [PubMed] [Google Scholar]
- [2].Birkmeyer JD, Siewers AE, Finlayson EVA, Stukel TA, Lucas FL, Batista I, et al. Hospital volume and surgical mortality in the United States. N Engl J Med 2002;346(15):1128–37. 10.1056/NEJMsa012337 [DOI] [PubMed] [Google Scholar]
- [3].Tay E, Gambhir S, Stopenski S, Hohmann S, Smith BR, Daly S, et al. Outcomes of complex gastrointestinal cancer resection at US news & world report top-ranked vs non-ranked hospitals. J Am Coll Surg 2021;233(1):21–7. 10.1016/j.jamcollsurg.2021.02.012.e1. [DOI] [PubMed] [Google Scholar]
- [4].Franchi E, Donadon M, Torzilli G. Effects of volume on outcome in hepatobiliary surgery: a review with guidelines proposal. Glob Health Med 2020;2(5):292–7. 10.35772/ghm.2020.01013 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [5].Fischer C, Alvarico SJ, Wildner B, Schindl M, Simon J. The relationship of hospital and surgeon volume indicators and post-operative outcomes in pancreatic surgery: a systematic literature review, meta-analysis and guidance for valid outcome assessment. HPB 2023;25(4):387–99. 10.1016/j.hpb.2023.01.008 [DOI] [PubMed] [Google Scholar]
- [6].Lidsky ME, Sun Z, Nussbaum DP, Adam MA, Speicher PJ, Blazer DG 3rd. Going the extra mile: improved survival for pancreatic cancer patients traveling to high-volume centers. Ann Surg 2017;266(2):333–8. 10.1097/sla.0000000000001924 [DOI] [PubMed] [Google Scholar]
- [7].Sweigert PJ, Eguia E, Baker MS, Paredes AZ, Tsilimigras DI, Dillhoff M, et al. Assessment of textbook oncologic outcomes following pancreaticoduodenectomy for pancreatic adenocarcinoma. J Surg Oncol 2020;121(6):936–44. 10.1002/jso.25861 [DOI] [PubMed] [Google Scholar]
- [8].Sato Y, Kaneko R, Yano Y, Kamada K, Kishimoto Y, Ikehara T, et al. Volumeoutcome relationship in cancer survival rates: analysis of a regional populationbased cancer registry in Japan. Healthcare 2022;11(1):16. 10.3390/healthcare11010016 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [9].Popovich DL, Vogus TJ, Iacobucci D, Austin JM. Are hospital ratings systems transparent? An examination of Consumer Reports and the Leapfrog Hospital Safety Grade. Health Mark Q 2020;37(1):41–57. 10.1080/07359683.2020.1713578 [DOI] [PubMed] [Google Scholar]
- [10].Smith SN, Reichert HA, Ameling JM, Meddings J. Dissecting leapfrog: how well do leapfrog safe practices scores correlate with hospital compare ratings and penalties, and how much do they matter? Med Care 2017;55(6):606–14. 10.1097/mlr.0000000000000716 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [11].Wasif N, Etzioni D, Habermann EB, Mathur A, Chang YH. Contemporary improvements in postoperative mortality after major cancer surgery are associated with weakening of the volume-outcome association. Ann Surg Oncol 2019;26(8):2348–56. 10.1245/s10434-019-07413-9 [DOI] [PubMed] [Google Scholar]
- [12].Jogerst KM, Chang Y-HH, Etzioni DA, Mathur AK, Habermann EB, Wasif N. Identifying the optimal case-volume threshold for pancreatectomy in contemporary practice. Am J Surg 2022;223(2):318–24. 10.1016/j.amjsurg.2021.03.030 [DOI] [PubMed] [Google Scholar]
- [13].Finks JF, Osborne NH, Birkmeyer JD. Trends in hospital volume and operative mortality for high-risk surgery. N Engl J Med 2011;364(22):2128–37. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [14].Kartik A, Ven Fong Z, Stucky CC, Cronin PA, Etzioni D, Chang YH, et al. Revisiting minimum case volume recommendations for complex surgery in contemporary practice. JAMA 2025;334(21):1945–8. 10.1001/jama.2025.17274 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [15].Amini N, Spolverato G, Kim Y, Pawlik TM. Trends in hospital volume and failure to rescue for pancreatic surgery. J Gastrointest Surg 2015;19(9):1581–92. 10.1007/s11605-015-2800-9 [DOI] [PubMed] [Google Scholar]
- [16].Colavita PD, Tsirline VB, Belyansky I, Swan RZ, Walters AL, Lincourt AE, et al. Regionalization and outcomes of hepato-pancreato-biliary cancer surgery in USA. J Gastrointest Surg 2014;18(3):532–41. 10.1007/s11605-014-2454-z [DOI] [PubMed] [Google Scholar]
- [17].Navarro SM, Haeberle HS, Mont MA, Krebs V, Ramkumar PN. Stratum-specific likelihood ratio analysis: an evidence-based and pragmatic approach to meaningful thresholds in lower extremity arthroplasty. Surg Technol Int 2019;34:415–20. [PubMed] [Google Scholar]
- [18].Peirce JC, Cornell RG. Integrating stratum-specific likelihood ratios with the analysis of ROC curves. Med Decis Mak 1993;13(2):141–51. 10.1177/0272989×9301300208 [DOI] [PubMed] [Google Scholar]
- [19].Schwartz AM, Farley KX, Nazzal EM, Manz WJ, Bradbury TL Jr, Guild GN 3rd. . Evidence-based hospital procedural volumes as predictors of outcomes after revision hip arthroplasty. J Arthroplast 2020;35(10):2952–9. 10.1016/j.arth.2020.05.008 [DOI] [PubMed] [Google Scholar]
- [20].Wilson S, Marx RG, Pan TJ, Lyman S. Meaningful thresholds for the volume-outcome relationship in total knee arthroplasty. J Bone Jt Surg Am 2016;98(20):1683–90. 10.2106/jbjs.15.01365 [DOI] [PubMed] [Google Scholar]
- [21].Ghomrawi HMK, Huang LW, Hiredesai AN, French DD. Using the stratum-specific likelihood ratio method to derive outcome-based hospital volume categories for total knee replacement. Med Care 2024;62(4):250–5. 10.1097/mlr.0000000000001985 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [22].Wade D, Varker T, Forbes D, O’Donnell M. The Alcohol Use Disorders Identification Test-Consumption (AUDIT-C) in the assessment of alcohol use disorders among acute injury patients. Alcohol Clin Exp Res 2014;38(1):294–9. 10.1111/acer.12247 [DOI] [PubMed] [Google Scholar]
- [23].Mallin K, Browner A, Palis B, Gay G, McCabe R, Nogueira L, et al. Incident cases captured in the national cancer database compared with those in U.S. population based central cancer registries in 2012-2014. Ann Surg Oncol 2019;26(6):1604–12. 10.1245/s10434-019-07213-1 [DOI] [PubMed] [Google Scholar]
- [24].Kaiser KN, Hughes AJ, Chung JW, Wilk AS, Ross-Driscoll K, Patzer RE, et al. Reproducibility and consistency of methods to define hospital-level procedural volume thresholds for pancreatectomy. J Surg Oncol 2025;133(1):92–100. 10.1002/jso.70134 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [25].Katayama ES, Thammachack R, Woldesenbet S, Khalil M, Munir MM, Tsilimigras D, et al. The association of established primary care with postoperative outcomes among Medicare patients with digestive tract cancer. Ann Surg Oncol 2024;31(12):8170–8. 10.1245/s10434-024-16042-w [DOI] [PMC free article] [PubMed] [Google Scholar]
- [26].Partelli S, Fermi F, Fusai GK, Tamburrino D, Lykoudis P, Beghdadi N, et al. The value of textbook outcome in benchmarking pancreatoduodenectomy for non-functioning pancreatic neuroendocrine tumors. Ann Surg Oncol 2024;31(6):4096–104. 10.1245/s10434-024-15114-1 [DOI] [PubMed] [Google Scholar]
- [27].Mohamed A, Nicolais L, Fitzgerald TL. Textbook outcome as a composite measure of quality in hepaticopancreatic surgery. J Hepato-bil Pancreat Sci 2023;30(10):1172–9. 10.1002/jhbp.1351 [DOI] [PubMed] [Google Scholar]
- [28].Tsilimigras DI, Pawlik TM, Moris D. Textbook outcomes in hepatobiliary and pancreatic surgery. World J Gastroenterol 2021;27(15):1524–30. 10.3748/wjg.v27.i15.1524 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [29].Dimick JB, Cowan JA Jr, Upchurch GR Jr, Colletti LM. Hospital volume and surgical outcomes for elderly patients with colorectal cancer in the United States. J Surg Res 2003;114(1):50–6. 10.1016/s0022-4804(03)00207-5 [DOI] [PubMed] [Google Scholar]
- [30].Armitage P. Tests for linear trends in proportions and frequencies. Biometrics 1955;11(3):375–86. 10.2307/3001775 [DOI] [Google Scholar]
- [31].Cochran WG. Some methods for strengthening the common χ2 tests. Biometrics 1954;10(4):417–51. 10.2307/3001616 [DOI] [Google Scholar]
- [32].Ju MR, Karalis JD, Leonard G, Mansour JC, Karagkounis G, Wang SC, et al. Defining volume targets for regionalization of pancreaticoduodenectomy, esophagectomy, and major lung resection centers to improve oncologic and surgical outcomes. Ann Surg Oncol 2024;31(1):499–513. 10.1245/s10434-023-14339-w [DOI] [PubMed] [Google Scholar]
- [33].Ju MR, Blackwell JM, Zeh HJ, Yopp AC, Wang SC, Porembka MR. Redefining high-volume gastric cancer centers: the impact of operative volume on surgical outcomes. Ann Surg Oncol 2021;28(9):4839–47. 10.1245/s10434-021-09655-y [DOI] [PubMed] [Google Scholar]
- [34].Conroy PC, Calthorpe L, Lin JA, Mohamedaly S, Kim A, Hirose K, et al. Determining hospital volume threshold for safety of minimally invasive pancreaticoduodenectomy: a contemporary cutpoint analysis. Ann Surg Oncol 2022;29(3):1566–74. 10.1245/s10434-021-10984-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [35].Krautz C, Nimptsch U, Weber GF, Mansky T, Grützmann R. Effect of hospital volume on in-hospital morbidity and mortality following pancreatic surgery in Germany. Ann Surg 2018;267(3):411–7. 10.1097/sla.0000000000002248 [DOI] [PubMed] [Google Scholar]
- [36].Perri G, van Hilst J, Li S, Besselink MG, Hogg ME, Marchegiani G. Teaching modern pancreatic surgery: close relationship between centralization, innovation, and dissemination of care. BJS Open 2023;7(5):zrad081. 10.1093/bjsopen/zrad081 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [37].Ellis RJ, Brajcich BC, Bertens KA, Chan CHF, Castillo CFD, Karanicolas PJ, et al. Association between biliary pathogens, surgical site infection, and pancreatic fistula: results of a randomized trial of perioperative antibiotic prophylaxis in patients undergoing pancreatoduodenectomy. Ann Surg 2023;278(3):310–9. 10.1097/sla.0000000000005955 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.