Abstract
Background:
Thyroid surgery at high-quality hospitals is associated with fewer complications. We evaluated the impact of referring older adults with thyroid cancer to higher-performing local hospitals.
Methods:
We performed a simulation study of Surveillance, Epidemiology, and End Results–Medicare patients, aged ≥66 years, who underwent a thyroidectomy for well-differentiated thyroid cancer (2013–2017). An 80% sample was used to calculate each hospital’s risk-standardized 30-day serious adverse event rate, dividing hospitals into quartiles by performance. Hospitals located ≤30 miles of the remaining 20% of patients were compared, and 30-day serious adverse event rates and costs were simulated as if patients were treated at higher-quality hospitals using logistic regression with each alternative hospital’s fixed-effect.
Results:
We identified 8,946 patients who underwent thyroid resection at 843 hospitals. Average risk-adjusted serious adverse event rates ranged from 13.9% to 52.9% between quartile 1 and 4 hospitals (P < .001). We identified higher-quality hospitals for 43.7% of patients. Simulating care at the best local hospital reduced predicted serious adverse event rates from 25.6% (95% confidence interval, 24.7–26.4) to 16.2% (95% confidence interval, 15.5–16.8; P < .001), while modestly lowering average costs from $12,883 (95% confidence interval, 12,500–13,267) to $12,679 (95% confidence interval, 12,304–13,056; P =.029).
Conclusion:
Simulated care at higher-performing hospitals decreased serious adverse event rates after thyroid resection. Optimizing hospital selection may reduce postoperative morbidity without compromising preferences for local treatment.
Introduction
Older adults undergoing thyroid resection have an increased risk of complications and readmissions compared with younger adults,1-3 accounting for substantial additional costs.4 Despite their increased risk and complexity, older adults with well-differentiated thyroid cancer often receive lower-quality surgical care. For example, older patients and those with Medicare insurance are less likely to undergo surgical resection with a high-volume surgeon.4
Because thyroidectomy remains the best treatment for well-differentiated thyroid cancer at all ages, strategies to enhance the quality of surgical care provided to older patients may improve both short- and long-term outcomes. Many have suggested referring older patients to high-volume centers through regionalization,5,6 which directs patients to designated centers of excellence regardless of convenience. However, patients demonstrate a preference for health care near their homes. Older age, high illness burden, and financial concerns are prohibitive for many patients recommended to travel to regional centers, even when it may improve outcomes.7,8
An alternative approach to regionalization involves directing patients to the best hospital within an acceptable distance of their homes, which may improve outcomes without placing undue travel burdens on older adults. In this study, we took advantage of existing methods established by the Centers for Medicare & Medicaid Services9 to assess risk-adjusted hospital performance in thyroid resection. The intent of this research was to estimate the potential clinical and financial impacts of referring older patients with thyroid cancer to higher-performing local hospitals. We performed a simulation study of older adults who underwent thyroid resection for well-differentiated thyroid cancer to examine 30-day serious adverse events and costs as if patients had received care at alternative, higher-performing local hospitals.
Methods
Setting and participants
Using the linked Surveillance, Epidemiology, and End Results (SEER)-Medicare registry, we identified patients ≥66 years with a diagnosis of well-differentiated thyroid cancer from 2013 to 2017.10 Using the International Classification of Diseases, Ninth Revision (ICD-9) and Tenth Revision (ICD-10) procedure codes and Current Procedural Terminology codes (Supplemental Table SI), we then identified patients who underwent a partial thyroidectomy, total thyroidectomy without a neck dissection, or total thyroidectomy with any neck dissection using the Medicare Provider Analysis and Review and Part B files. Patients lacking continuous Medicare Fee-For-Service coverage or having health maintenance organization coverage from 12 months before 9 months after their surgery date were excluded owing to the potential for unreported claims data. Patients with missing data for any covariate were excluded from the analysis (Figure 1). We excluded patients who underwent a partial thyroidectomy followed by completion thyroidectomy within the 30-day follow-up period.
Figure 1.
Consort diagram of included patients. HMO, health maintenance organization.
Variables
The primary outcome was the risk-adjusted rate of any serious adverse event ≤30 days after surgery. Serious adverse events included general postoperative complications (eg, atrial fibrillation, cardiac or cerebrovascular events, transient ischemic attack, respiratory complications, renal infection, renal dysfunction, sepsis, and venous thromboembolism), thyroid surgery–specific complications (eg, vocal cord or voice dysfunction, hypocalcemia, tracheostomy, hematoma, and surgical site infection), and death. To avoid misclassification of pre-existing conditions as complications, patients were considered to have experienced an adverse event if a relevant ICD or Current Procedural Terminology code was identified in the 30 days after surgery but not present on admission or within the year before admission (Supplemental Table SII).
The secondary outcome was 30-day costs, defined as Medicare payments for the index operative encounter, subsequent outpatient visits, professional fees, and durable medical equipment claims ≤30 days after surgery. All dollar amounts were inflated to 2017 dollars using the annual “medical care” component of the consumer price index.11
Patient demographic variables were obtained from the Master Beneficiary Summary File and SEER Cancer File. We included age, sex at birth, race, and dual eligibility. To identify comorbidities, we searched the Medicare Provider Analysis and Review (inpatient), outpatient, Part B, and durable medical equipment files for preexisting conditions within the year before the index operative encounter using the Elixhauser Comorbidity Index software (Agency for Healthcare Research and Quality, Rockville, MD).12 We also searched the diagnoses fields flagged as “present on admission” at the time of the index operative encounter. Using a similar method, we assessed physical frailty with a validated, claims-based frailty index.13 A binary variable indicating whether the patient was treated during the ICD-9 or ICD-10 era was generated to mitigate potential bias due to differences in procedure and complication coding. Inpatient or outpatient setting for the index operative encounter was classified based on the source of the claim.
Variables describing cancer-specific disease characteristics were obtained from the SEER Cancer File. Papillary and follicular histology was categorized using the ICD-O-3 (Supplemental Table SI).14 The SEER summary stage was used to the classify cancer spread as in situ, localized, regional, distant, or unstaged.
Descriptive hospital information including bed size, region, and classification as an American College of Surgeons (ACS) cancer center, cancer center affiliate, or academic hospital was attained from the American Hospital Association Annual Survey Database15 and SEER-Medicare hospital files. Travel distance between the patient’s home and hospital was calculated as the geodetic distance in miles between the 2 zip code centroids.16
Model overview and simulation strategy
The hospitals were ranked and divided into quartiles according to performance based on their risk-standardized rate of serious adverse events in an 80% training cohort using the Centers for Medicare & Medicaid Services methodology.9 Additional details can be found in the Supplemental material. To evaluate the impact of referral based on hospital performance, postoperative outcomes were simulated in the 20% testing cohort as if those patients had been treated at local hospitals in a better quartile when available at ≤30 miles.17 To avoid the instability of estimating complication rates at low case volumes, we restricted the pool of eligible alternative hospitals to those performing ≤10 thyroidectomies during the study period (n = 238 hospitals, which treated 79.5% of the total patient population).
The simulation models assumed that patients were willing to seek care at alternative, higher-performing hospitals if available. We also assumed that the choice of an inpatient setting was deliberate for the patients who underwent inpatient thyroidectomy (eg, due to patient comorbidities or preference). We therefore limited the set of potential alternative hospitals to those that offer inpatient care for only the patients who received inpatient thyroidectomy.
Two counterfactual hospital selection scenarios were evaluated. First, we examined outcomes as if all patients had received care at their closest hospital located ≤30 miles in a higher-performance quartile (closest better hospital). Second, we simulated care at the hospital with the best performance ranking located ≤30 miles (best hospital). In some circumstances, the same hospital might represent both the closest better and best hospital, depending on the quality quartile distribution of hospitals located ≤30 miles. The study was designed according to recommendations by the Consolidated Health Economic Evaluation Reporting Standards.18
Statistical analysis
Serious adverse event risk was predicted using simulation models with hospital fixed effects for each patient undergoing care at their original hospital, closest better hospital, or best hospital. Thus, the predicted outcomes at each hospital differed only in the relative shift based on the original and alternative hospitals’ fixed effect values as observed in the training model. Each patient’s risk estimate for the care received at the original hospital was compared with the estimate for each alternative hospital selection scenario using paired t tests. We evaluated the impact of care at alternative hospitals on 30-day costs with the same methods using linear regression. The population median travel distance was calculated between each patient’s original and alternative hospitals and compared using Wilcoxon signed-rank tests.
Descriptive statistics included 1-way analysis of variance for continuous variables and χ2 analysis for the categorical variables. Statistical analysis was performed using Stata, version 17.0 (Stata- Corp, LLC, College Station, TX) and SAS, version 9.4 (SAS Institute, Cary, NC). This study was approved by the Institutional Review Board at the University of Pennsylvania (Protocol #848800).
Results
Patient characteristics
We identified 8,946 patients with well-differentiated thyroid cancer who underwent thyroid resection. Histology was papillary in 8,203 patients (91.7%) and follicular in 743 patients (8.3%). The cohort had predominantly localized disease (n = 6,269; 70.1%), with some regional (n = 2,223; 24.8%) and distant (n = 317; 3.5%) disease. Patients underwent partial thyroidectomy (n = 2,184; 24.4%), total thyroidectomy without a neck dissection (n = 4,631; 51.8%), and total thyroidectomy with any neck dissection (n = 2,131; 23.8%) at 843 different hospitals. Medical comorbidities were common in our population (Supplemental Table SIII). The training and testing cohorts had similar demographic characteristics, comorbidities, and disease characteristics (Table I).
Table I.
Characteristics of study population
| Training cohort (n = 7,173) | Testing cohort (n = 1,773) | Total (n = 8,946) | P value | |
|---|---|---|---|---|
| Age (y) | 69.2 (10.8) | 69.1 (10.9) | 69.2 (10.8) | .584 |
| Female sex | 4,946 (69.0) | 1,222 (68.9) | 6,168 (68.9) | .980 |
| Race | .621 | |||
| Black | 538 (7.5) | 126 (7.1) | 664 (7.4) | |
| White | 5,905 (82.3) | 1,477 (83.3) | 7,382 (82.5) | |
| Other | 730 (10.2) | 170 (9.6) | 900 (10.1) | |
| Dual eligible | 1,645 (22.9) | 418 (23.6) | 2,063 (23.1) | .565 |
| Claims-based frailty index | 0.163 (0.059) | 0.164 (0.058) | 0.163 (0.058) | .493 |
| Cancer histology | .828 | |||
| Papillary | 6,575 (91.7) | 1,628 (91.8) | 8,203 (91.7) | |
| Follicular | 598 (8.3) | 145 (8.2) | 743 (8.3) | |
| SEER summary stage | .852 | |||
| in situ | 28 (0.4) | <11 (<0.6) | <39 (<0.4) | |
| Local | 5,026 (70.1) | >1,238 (>69.8) | >6,264 (>70) | |
| Regional | 1,781 (24.8) | 442 (24.9) | 2,223 (24.8) | |
| Distant | 259 (3.6) | 58 (3.3) | 317 (3.5) | |
| Unstaged | 79 (1.1) | 24 (1.4) | 103 (1.2) | |
| Selected preoperative comorbidities | ||||
| Vocal cord dysfunction | 266 (3.7) | 63 (3.6) | 329 (3.7) | .756 |
| Hypocalcemia | 161 (2.2) | 45 (2.5) | 206 (2.3) | .461 |
| Tracheostomy | 16 (0.2) | <11 (<0.6) | <27 (<0.3) | .354 |
| Hypothyroidism | 3,581 (49.9) | 828 (46.7) | 4,409 (49.3) | .015 |
| Procedure type | .257 | |||
| Partial thyroidectomy | 1,727 (24.1) | 457 (25.8) | 2,184 (24.4) | |
| Total thyroidectomy | 3,719 (51.8) | 912 (51.4) | 4,631 (51.8) | |
| Total thyroidectomy with neck dissection | 1,727 (24.1) | 404 (22.8) | 2,131 (23.8) | |
| Visit type | .828 | |||
| inpatient | 2,270 (31.6) | 562 (31.7) | 2,832 (31.7) | |
| Outpatient | 4,903 (68.4) | 1,211 (68.3) | 6,114 (68.3) | |
| ICD era | .424 | |||
| ICD-9 | 3,816 (53.2) | 962 (54.3) | 4,778 (53.4) | |
| ICD-10 | 3,357 (46.8) | 811 (45.7) | 4,168 (46.6) |
A full list of medical covariates can be found in Supplemental Table SIII. The continuous variables are presented as mean (SD) and the categorical variables are presented as n (%) unless otherwise specified. Due to restrictions of the data use agreement, cell counts >0 and <11 are depicted as <11. ICD, international Classification of Diseases; ICD-9, ICD, Ninth Revision; ICD-10, ICD, Tenth Revision; SEER, Surveillance, Epidemiology, and End Results.
Assessment of hospital performance
Examining postoperative outcomes, 2,072 patients (23.2%) suffered a serious adverse event at ≤30 days of surgery (Table II). The most common serious adverse event was hypocalcemia (n = 1,243; 13.9%), followed by voice disturbances (n = 251; 2.8%), vocal cord dysfunction (n = 169; 1.9%) and cardiac events (n = 270; 3%). Patients who underwent a total thyroidectomy with neck dissection had the highest risk of serious adverse event (n = 653; 30.6%; P < .001) compared with patients who underwent a partial thyroidectomy (n = 352; 16.1%) or total thyroidectomy without a neck dissection (n = 1,067; 23%).
Table II.
Serious adverse events by procedure type
| Partial thyroidectomy (n = 2,184) |
Total thyroidectomy without neck dissection (n = 4,631) |
Total thyroidectomy with neck dissection (n = 2,131) |
|
|---|---|---|---|
| Death | 16 (0.7) | 14 (0.3) | <11 (<0.5) |
| Serious adverse event | 352 (16.1) | 1,067 (23) | 653 (30.6) |
| Procedure-specific complication | |||
| Vocal cord dysfunction | 37 (1.7) | 66 (1.4) | 66 (3.1) |
| Voice disturbance | 65 (3.0) | 109 (2.4) | 77 (3.6) |
| Hypocalcemia | 98 (4.5) | 707 (15.3) | 438 (20.6) |
| Tracheostomy | 22 (1.0) | 23 (0.5) | 28 (1.3) |
| Bleeding/hematoma | 20 (0.9) | 42 (0.9) | 15 (0.7) |
| Surgical site infection | 15 (0.7) | <11 (<0.2) | <11 (<0.5) |
| Lateral neck complication | <11 (<0.5) | <11 (<0.2) | <11 (<0.5) |
| General complication | |||
| Atrial fibrillation | 26 (1.2) | 52 (1.1) | 24 (1.1) |
| Cardiac events | 71 (3.3) | 131 (2.8) | 68 (3.2) |
| Cerebrovascular complication | 20 (0.9) | 37 (0.8) | 18 (0.8) |
| Renal infection | <11 (<0.5) | <11 (<0.2) | <11 (<0.5) |
| Renal dysfunction | 18 (0.8) | 61 (1.3) | 32 (1.5) |
| Respiratory complication | <11 (<0.5) | <11 (<0.2) | <11 (<0.5) |
| Sepsis | 11 (0.5) | 15 (0.3) | <11 (<0.5) |
| Transient ischemic attack | <11 (<0.5) | <11 (<0.2) | <11 (<0.5) |
| Venous thromboembolism | 14 (0.6) | 23 (0.5) | 17 (0.8) |
Results are presented as n (%). Due to restrictions of the data use agreement, cell counts >0 and <11 are depicted as <11.
Risk-standardized serious adverse event rates ranged from 13.9% at quartile 1 hospitals to 52.9% at quartile 4 hospitals (P < .001) (Table III). Higher-performing hospitals were larger and more often had cancer programs accredited by the ACS. Lower-performing hospitals were more likely to be smaller and nonteaching. Higher-performing hospitals treated more Medicare patients with thyroid cancer over the 5-year study period, with a median of 16 cases (interquartile interval [IQI] = 11 –31.25) in quartile 1 and 4 cases (IQI = 3–9.75) in quartile 4 (P < .001).
Table III.
Hospital characteristics and performance by quality quartile
| Quartile 1 (n = 104) | Quartile 2 (n = 104) | Quartile 3 (n = 104) | Quartile 4 (n = 104) | P value | |
|---|---|---|---|---|---|
| Region | .343 | ||||
| Northeast | >37 (>35.6) | 39 (37.5) | 27 (26.0) | 33 (32.4) | |
| South | 22 (21.2) | 17 (16.3) | 29 (27.9) | 20 (19.6) | |
| Midwest | <11 (<10.6) | 11 (10.6) | <11 (<10.6) | 13 (12.7) | |
| West | 34 (32.7) | 37 (35.6) | >37 (>35.6) | 36 (35.3) | |
| Urban hospital | 104 (100.0) | >93 (>89.4) | 104 (100.0) | >93 (>89.4) | .257 |
| Rural hospital | 0 (0.0) | <11 (<10.6) | 0 (0.0) | <11 (<10.6) | |
| ACS cancer center | 81 (77.9) | 78 (75.0) | 70 (67.3) | 60 (57.7) | .007 |
| Cancer center affiliate | 18 (17.3) | 20 (19.2) | 25 (24.0) | 31 (29.8) | .136 |
| Medical school affiliation | .010 | ||||
| Major | 48 (46.2) | 36 (34.6) | 30 (28.8) | 20 (19.2) | |
| Graduate | 17 (16.3) | 19 (18.3) | 21 (20.2) | 20 (19.2) | |
| Limited or No Affiliation | 39 (37.5) | 49 (47.1) | 53 (50.9) | 64 (61.5) | |
| Bed size | .005 | ||||
| Small (<100) | <11 (<10.6) | <11 (<10.6) | <11 (<10.6) | 12 (11.5) | |
| Medium (100–299) | 29 (27.9) | 36 (34.6) | 37 (35.6) | 48 (46.2) | |
| Large (300+) | >64 (>61.5) | >57 (>54.8) | >56 (>53.8) | 44 (42.3) | |
| Procedure volume over study period, median (IQI) | 16.00 (11.00–31.25) | 14.50 (7.00–29.50) | 9.00 (4.00–21.75) | 4.00 (3.00–9.75) | < .001 |
| Procedure mix, mean % of total volume (SD) | |||||
| Partial thyroidectomy | 23.8% (16.8) | 24.2% (18.1) | 22.6% (20.1) | 28.5% (25) | .163 |
| Total thyroidectomy without neck dissection | 52.1% (21.2) | 52.9% (23.3) | 56.3% (26.8) | 51.0% (29) | .469 |
| Total thyroidectomy with neck dissection | 24.0% (19.1) | 22.9% (19.1) | 21.2% (23.7) | 20.5% (23.6) | .617 |
| P:E ratio | 0.582 (0.154) | 1.058 (0.121) | 1.480 (0.132) | 2.373 (0.521) | < .001 |
| Risk-standardized adverse event risk, mean (SD) | 13.9% (4.3) | 25.1% (5.8) | 35.5% (7.4) | 52.9% (11.1) | < .001 |
Results are presented as n (%). Of the 843 hospitals assessed, 427 hospitals accounting for 11.8% (n = 850) of the patients in the training cohort were ineligible for inclusion in the simulation analysis due to perfectly predicting serious adverse event rates, either through having a 100% or 0 complication rate. Due to restrictions of the data use agreement, cell counts greater than 0 and less than 11 are depicted as <11.
ACS, American College of Surgeons; IQI, interquartile interval; P:E, predicted-to-expected.
Simulation of alternative hospital selection
Using the testing cohort, we were able to simulate results for 1,514 patients (85.4%) (Supplemental Table SIV). Among these patients, we identified a better-performing hospital located ≤30 miles for 662 (43.7%) patients. Patients with a better-performing hospital in simulation were more often treated in the inpatient setting and had demographic and disease characteristics similar to those who were originally treated at their optimal local hospital (Table IV).
Table IV.
Characteristics of testing cohort in simulation
| Remained at original hospital (n = 852) | Moved in simulation to higher-performing hospital (n = 662) |
P value | |
|---|---|---|---|
| Age (y) | 69.2 (10.8) | 69.1 (11.1) | .863 |
| Female sex | 589 (69.1) | 446 (67.4) | .465 |
| Race | .004 | ||
| Black | 48 (5.6) | 56 (8.5) | |
| White | 738 (86.6) | 531 (80.2) | |
| Other | 66 (7.7) | 75 (11.3) | |
| Dual eligible | 202 (23.7) | 138 (20.8) | .185 |
| Claims-based frailty index | 0.160 (0.055) | 0.166 (0.061) | .066 |
| Cancer histology | .294 | ||
| Papillary | 779 (91.4) | 615 (92.9) | |
| Follicular | 73 (8.6) | 47 (7.1) | .645 |
| SEER summary stage | |||
| In situ | <11 (<1.3) | <11 (<1.7) | |
| Local | >601 (>70.5) | >454 (>68.5) | |
| Regional | 212 (24.9) | 167 (25.2) | |
| Distant | 27 (3.2) | 19 (2.9) | |
| Unstaged | <11 (<1.3) | 11 (1.7) | |
| Selected comorbidities before surgery | |||
| Vocal cord dysfunction | 23 (2.7) | 29 (4.4) | .075 |
| Hypocalcemia | 21 (2.5) | 20 (3.0) | .508 |
| Tracheostomy | <11 (<1.3) | <11 (<1.7) | .858 |
| Hypothyroidism | 395 (46.4) | 314 (47.4) | .679 |
| Procedure type | .197 | ||
| Partial thyroidectomy | 214 (25.1) | 149 (22.5) | |
| Total thyroidectomy | 425 (49.9) | 361 (54.5) | |
| Total thyroidectomy with neck dissection | 213 (25.0) | 152 (23.0) | |
| Visit type | .162 | ||
| Inpatient | 252 (29.6) | 218 (32.9) | |
| Outpatient | 600 (70.4) | 444 (67.1) | |
| ICD era | .261 | ||
| ICD-9 | 445 (52.2) | 365 (55.1) | |
| ICD-10 | 407 (47.8) | 297 (44.9) | |
| Distance from home to original hospital, median (IQI) | 15.00 (6.62–36.65) | 8.60 (4.30–16.05) | < .001 |
Results are presented as n (%). Due to restrictions of the data use agreement, cell counts >0 and <11 are depicted as <11. Patients in the testing cohort were included in the simulation if they received care from a hospital that was ranked using the training cohort. Patients with an unranked original hospital were thus excluded (n = 259). Patients were eligible for simulated care at a higher-quality hospital if a hospital in a higher quartile with >10 cases was identified at ≤30 miles.
ICD, International Classification of Diseases; ICD-9, ICD, Ninth Revision; ICD-10, ICD, Tenth Revision; IQI, interquartile interval; SEER, Surveillance, Epidemiology, and End Results.
In the testing cohort, we predicted a serious adverse event rate of 25.6% (95% CI, 24.7–26.4) with treatment at the set of original hospitals. Simulating care at the closest better hospital reduced the population’s predicted serious adverse event rate to 18.1% (95% CI, 17.5–18.8; P < .001). Similarly, simulating care at the best hospital further reduced the population’s serious adverse event rate to 16.2% (95% CI, 15.5–16.8; P < .001) (Figure 2A). A sensitivity analysis using prediction models that included both stage and procedure type in addition to the other covariates had similar results (Supplemental Table SV).
Figure 2.
Predictions of serious adverse events (A) and costs (B) at alternative hospitals. Squares represent point estimates with 95% CIs. Tests of model performance had a C statistic of 0.722 and a Brier score of 0.165 when evaluated on the training cohort and a C statistic of 0.617 and a Brier score of 0.173 when evaluated on the testing cohort.
In simulation, the predicted mean 30-day costs were $12,883 (95% CI, 12,500–13,267) for care at the original hospital. Simulating alternative hospital selection reduced costs to $12,397 (95% CI, 12,027–12,767; P < .001) with care at the closest better hospital. Costs were also decreased slightly to $12,679 (95% CI, 12,304–13,056; P = .029) with care at the best hospital located ≤30 miles (Figure 2B).
Patients in the testing cohort traveled a median of 11.5 miles (IQI = 5.3–26.3) to their original hospital for treatment. If all eligible patients chose the closest better hospital located ≤30 miles, the population’s median travel distance would not increase (12.3 miles; IQI = 6.4–24.5; P = .075). If all patients traveled to the best hospital located ≤30 miles, the median travel distance would increase to 16.6 miles (IQI = 8.0–27.2; P < .001).
Discussion
In this study, the simulated referral of older adults with well-differentiated thyroid cancer to higher-performing local hospitals decreased predicted serious adverse events. We found that 23.2% of older adults experienced a serious adverse event, with average risk-adjusted rates ranging from 13.9% to 52.9% among hospitals in the highest- and lowest-performing quartiles. Directing patients to the best local hospital led to a predicted absolute risk reduction of 9.4% and relative risk reduction of 36.7%. The improvements in serious adverse event rates were not associated with clinically meaningful differences in cost. Referral to higher-performing hospitals also appeared convenient for patients, given the median travel distance was only increased by 5.1 miles to reach the best local hospitals.
Like prior studies, we found hospital-level variation in surgical outcomes for patients with thyroid cancer. It has been established that hospitals vary widely in surgical markers of quality for thyroid cancer treatment, including guideline adherence,19 margin positivity,6 and use of intraoperative nerve monitoring.20 Several prior studies have also found that high-volume, specialized hospitals offer lower rates of postoperative morbidity, recurrence, and disease-specific mortality.6,19,20 For example, the risk-adjusted odds of recurrent laryngeal nerve injury varied from 0.16 to 18.2 among hospitals in the ACS–National Surgical Quality Improvement Program, with corresponding rates of nerve monitoring ranging from 55.7% to 37.7% between high- and low-performing hospitals.20 Although there are many indicators of hospital quality, we elected to focus on postoperative outcomes (rather than processes of care) due to the direct impact on patients. We highlighted the potential of hospital choice to serve as a mechanism to improve outcomes.
Our findings were consistent with published literature reporting a high burden of complications among older adults with thyroid cancer. Papaleontiou et al21 found that 19.1% of older adults in SEER experienced a thyroid-specific complication, including 13.6% with hypoparathyroidism and 7.1% with recurrent laryngeal nerve injury. Similarly, older adults treated at ACS–National Surgical Quality Improvement Program hospitals had an 8% risk of recurrent laryngeal nerve injury.22 In a survey of older adults after thyroidectomy for malignancy, 12% reported persistent voice dysfunction at 3 months after surgery and 5% reported vocal cord paralysis diagnosed with laryngoscopy.23 Given our findings that 43.7% of older adults in our cohort did not receive treatment at their optimal local hospital, the high rates of postoperative complications may be reduced through informed hospital selection.
We found that the 30-day costs of thyroid resection were similar in magnitude across alternative hospital selection scenarios. Costs for thyroidectomy have been estimated around $10,000,24 with higher costs for inpatient surgery and neck dissections.24 Prior studies have found that complications, longer length of stay, and readmissions account for substantial additional costs among Medicare beneficiaries with thyroid cancer.4 Although reimbursements for the index procedure are standardized by Medicare, the additional costs of complications and associated treatments continue to accrue over time. Although we found a small but statistically significant decrease in costs of care at higher-performing hospitals, our estimates do not include the long-term costs of persistent complications that require further treatment after 30 days.
Prior literature has proposed regionalization as a mechanism to standardize and improve outcomes for patients with thyroid cancer.5,6 The travel burden of regionalized care is often impractical for older adults and inconsistent with patient preferences. Our approach to use data to inform hospital selection demonstrates the ability to improve clinical outcomes while allowing patients to receive care close to home at similar cost. Our simulated findings suggested that strict regionalization may not be required to improve outcomes for older adults.
Our simulation study had several limitations. First, the study was designed to estimate the potential benefit of optimized hospital choice for older patients with Medicare coverage, an ideal population for this intervention due to universal insurance coverage among the included patients and hospitals. Thus, the results may not generalize to younger patients or those with marketplace insurance who may have fewer available options for in-network hospitals. Second, we compared the hospitals surrounding each patient using geographic distance. Future studies should consider the impact of travel time or driving distance on available hospitals. Third, we were unable to estimate risk-adjusted serious adverse event rates for 14.6% of the population, due to care at hospitals with either very low procedure volumes or a lack of variation in the outcome measure in the training cohort. Fourth, hospitals are motivated to adequately capture patient complexity and procedures for reimbursement, but differences in coding practices may occur across institutions. Finally, we were unable to capture the permanence or severity of each complication due to the limitations of administrative claims data. Although this study focuses on the initial 30 days after surgery to capture hospital performance during the index operation, future studies are planned to examine the impact of hospital choice longer-term outcomes, such as survival, recurrence, and persistent complications.
Finally, our simulation study does not address issues related to implementation. Incorporating outcomes into the hospital selection process requires an understanding of the factors important to patient and referring providers when selecting a hospital for surgery. Further qualitative work is required to understand the most effective ways to present quality information, because outcomes data presently are rarely used in the referral process.25
Our simulation study confirmed that the current delivery of surgical care for older adults with well-differentiated thyroid cancer is suboptimal. By choosing the local hospital with the best disease- and procedure-specific outcomes, patients can reduce their risk of adverse events without increasing costs or compromising preference to be treated close to home. These results highlighted a tremendous opportunity to transform the way in which patients and referring providers consider their choice of hospital for surgery.
Supplementary Material
Funding/Support
Caitlin B. Finn receives salary support from the National Institutes of Health T32 Training Program, 5T32CA251063. Heather Wachtel receives salary support from the National Institutes of Health NCATS KL2 TR001879. Rachel R. Kelz receives salary support from the National Institute on Aging R01 AG060612 and the National Cancer Institute R01 CA228399.
Biographies
Dr Sareh Parangi (Boston, MA): Did you look at the websites of the better-performing quartile hospitals to see if they had advertised their better performance?
Dr Caitlin Finn: Due to the limitations of our data set, we did not look at individual hospitals. Our performance measure was specific to thyroid surgery. Hospitals generally advertise in a more global fashion, such as US News & World Report rankings or Top Doctors rankings. They are less likely to advertise specific performance measures for specific patient populations or types of procedure. It would be interesting to look at and hopefully incorporate more procedure-specific performance and quality information into these global performance metrics for the patients.
Dr Carolyn Gardner (Denton, TX): As a high-volume community surgeon, I would submit it is not the hospital that has the outcome, it is the surgeon. When patients are referred to me as the surgeon, I determine the best place for this patient to have an operation. I would say, in addition to the idea of the high-volume hospital, we should ask the question, how many dedicated endocrine surgeons use that hospital and are the cases at that hospital done by endocrine, general otolaryngology, or general surgeons?
Dr Caitlin Finn: We agree. We are hoping to study this on a surgeon level as well. You bring up a great point that patient preferences are critical. It is important to study why patients choose the hospital that they choose and why some patients may choose to go to a lower-performing hospital. We need to ask what are these hospitals offering that patients cannot find at high-performing, high-volume centers?
Dr William B. Inabnet, III (Lexington, KY): I have a comment and a very short question. I can think of no better study for the American College of Surgeons—patient outreach programs that are designed to identify hospitals and in the future hospital systems that provide high-quality care in 1 component or 1 standard. The question is about the data set in your model. Did you differentiate the administrative data from the clinical data, because there is a difference, and scoring systems often use administrative data, which has its pitfalls; so, any comments on that?
Dr Caitlin Finn: Our data were from the Surveillance, Epidemiology, and End Results (SEER) Medicare registry. The SEER registry is a cancer registry that has variables, such as tumor stage and histology. The Medicare claims data report administrative billing claims from the hospitals. So, the SEER data gave us the patient information and tumor characteristics and the Medicare data gave us the health care use across all hospitals. The advantage of this is that it is a geographically based data source, so it allows comparison of all hospitals within a particular region, including high-performing and low-performing ones. You also mentioned hospital systems and optimizing care within systems, and I think this study has the potential to optimize the location of care within a hospital system; knowing whether patients get their treatment at the main academic center or a referral/satellite site is valuable for optimizing care.
Footnotes
Presented at the 42nd Annual Meeting of the American Association of Endocrine Surgeons, May 22–24, 2022, Cleveland, OH.
Conflict of interest/Disclosure
The authors have no conflicts of interests or disclosures to report.
Supplementary materials
Supplementary material associated with this article can be found, in the online version, at https://doi.org/10.1016/j.surg.2022.05.047.
References
- 1.Sahli ZT, Ansari G, Gurakar M, et al. Thyroidectomy in older adults: an ACS-NSQIP study of outcomes. J Surg Res. 2018;229:20. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Grogan RH, Mitmaker EJ, Hwang J, et al. A population-based prospective cohort study of complications after thyroidectomy in the elderly. J Clin Endocrinol Metab. 2012;97:1645–1653. [DOI] [PubMed] [Google Scholar]
- 3.Papaleontiou M, Hughes DT, Guo C, Banerjee M, Haymart MR. Population-based assessment of complications following surgery for thyroid cancer. J Clin Endocrinol Metab. 2017;102:2543–2551. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Sosa JA, Mehta PJ, Wang TS, Boudourakis L, Roman SA. A population-based study of outcomes from thyroidectomy in aging Americans: at what cost? J Am Coll Surg. 2008;206:1097–1105. [DOI] [PubMed] [Google Scholar]
- 5.Sharma RK, Lee J, Liou R, McManus C, Lee JA, Kuo JH. Optimal surgeon-volume threshold for neck dissections in the setting of primary thyroid malignancies. Surgery. 2022;171:172–176. [DOI] [PubMed] [Google Scholar]
- 6.Youngwirth LM, Adam MA, Scheri RP, Roman SA, Sosa JA. Patients treated at low-volume centers have higher rates of incomplete resection and compromised outcomes: analysis of 31,129 patients with papillary thyroid cancer. Ann Surg Oncol. 2016;23:403–409. [DOI] [PubMed] [Google Scholar]
- 7.Resio BJ, Chiu AS, Hoag JR, et al. Motivators, barriers, and facilitators to traveling to the safest hospitals in the United States for complex cancer surgery. JAMA Network Open. 2018;1:e184595–e184595. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Symer MM, Abelson JS, Yeo HL. Barriers to regionalized surgical care: public perspective survey and geospatial analysis. Ann Surg. 2019;269:73–78. [DOI] [PubMed] [Google Scholar]
- 9.Drye E, Normand SLT, Wang Y, et al. Comparison of hospital risk-standardized mortality rates using in-hospital and 30-day models: implications for hospital profiling. Ann Intern Med. 2012;156(1 Pt 1):19. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.SEER-Medicare Linked Data Resource. https://healthcaredelivery.cancer.gov/seermedicare/medicare/table.html. Accessed February 1, 2022. [Google Scholar]
- 11.Bureau of Labor Statistics. Databases, tables & calculators by subject. https://data.bls.gov/timeseries/CUUR0000SAM?output_view=data. Accessed February 6, 2022. [Google Scholar]
- 12.Agency for Healthcare Research and Quality. Elixhauser Comorbidity Software Refined for ICD-10-CM. https://www.hcup-us.ahrq.gov/toolssoftware/comorbidityicd10/comorbidity_icd10.jsp. Accessed February 9, 2022. [Google Scholar]
- 13.Kim DH, Schneeweiss S, Glynn RJ, Lipsitz LA, Rockwood K, Avorn J. Measuring frailty in Medicare data: development and validation of a claims-based frailty index. J Gerontol A Biol Sci Med Sci. 2018;73:980–987. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Fritz A, Percy C, Jack A, et al. ICD-O International Classification of Diseases for Oncology. Third Edition. Geneva (Switzerland): World Health Organization; 2000. [Google Scholar]
- 15.American Hospital Association. AHA Annual Survey Database. https://www.ahadata.com/aha-annual-survey-database. Accessed October 13, 2021. [Google Scholar]
- 16.SAS Help Center: ZIPCITYDISTANCE Function. https://documentation.sas.com/doc/en/pgmsascdc/9.4_3.2/lefunctionsref/n1r333fdkrofhxn10vmhu9bq5m85.htm. Accessed February 13, 2022. [Google Scholar]
- 17.Ho V, Short MN, Ku-Goto MH. Can centralization of cancer surgery improve social welfare? Forum Econ Policy. 2012;15:1–25. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Husereau D, Drummond M, Petrou S, et al. Consolidated Health Economic Evaluation Reporting Standards (CHEERS) statement. BMJ. 2013;346:f1049. [DOI] [PubMed] [Google Scholar]
- 19.Megwalu UC, Ma Y, Hernandez-Boussard T, Divi V, Lin Gomez S. The impact of hospital quality on thyroid cancer survival. Otolaryngology Head Neck Surg. 2020:269–276. [DOI] [PubMed] [Google Scholar]
- 20.Liu JB, Sosa JA, Grogan RH, et al. Variation of thyroidectomy-specific outcomes among hospitals and their association with risk adjustment and hospital performance. JAMA Surg. 2018;153:e174593–e174593. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Papaleontiou M, Hughes DT, Guo C, Banerjee M, Haymart MR. Population-based assessment of complications following surgery for thyroid cancer. J Clin Endocrinol Metab. 2017;102:2543. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Gunn A, Oyekunle T, Stang M, Kazaure H, Scheri R. Recurrent laryngeal nerve injury after thyroid surgery: an analysis of 11,370 patients. J Surg Res. 2020;255:42–49. [DOI] [PubMed] [Google Scholar]
- 23.Kovatch KJ, Reyes-Gastelum D, Hughes DT, Hamilton AS, Ward KC, Haymart MR. Assessment of voice outcomes following surgery for thyroid cancer. JAMA Otolaryngol Head Neck Surg. 2019;145:823–829. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Lee L, Mitmaker EJ, Chabot JA, Lee JA, Kuo JH. Cost-effectiveness of diagnostic lobectomy versus observation for thyroid nodules >4 cm. Thyroid. 2016;26: 271–279. [DOI] [PubMed] [Google Scholar]
- 25.Finn CB, Tong JK, Alexander HE, et al. How referring providers choose specialists for their patients: a systematic review. J Gen Intern Med. 2022;37: 3444–3452. [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.