Hospital Variation in Responses to Safety Warnings about Power Morcellation in Hysterectomy

Abstract

BACKGROUND:

Safety warnings about power morcellation in 2014 considerably changed hysterectomy practice, especially for laparoscopic supracervical hysterectomy that typically requires morcellation to remove the corpus uteri while preserving the cervix. Hospitals might vary in how they responded to the safety warnings and altered hysterectomy procedures to avoid use of power morcellation. Yet there has been little data on how hospitals differ in their practice changes.

OBJECTIVES:

To examine whether hospitals varied in their use of laparoscopic supracervical hysterectomy after safety warnings about power morcellation and compare the risk of surgical complications at hospitals that had different response trajectories in use of laparoscopic supracervical hysterectomy.

STUDY DESIGN:

This was a retrospective analysis of data from the New York Statewide Planning and Research Cooperative System, as well as the State Inpatient Databases and State Ambulatory Surgery and Services Databases from 14 other states. We identified women age ≥18 years undergoing hysterectomy for benign indications in hospital inpatient and outpatient settings from 10/1/2013–9/30/2015. We calculated a risk-adjusted utilization rate of laparoscopic supracervical hysterectomy for each hospital in each calendar quarter after accounting for patient clinical risk factors. Applying a growth mixture modeling approach, we identified distinct groups of hospitals that exhibited different trajectories of using laparoscopic supracervical hysterectomy over time. Within each trajectory group, we compared patients’ risk of surgical complications in the pre-warning (2013Q4–2014Q1), transition (2014Q2–2014Q4), and post-warning (2015Q1–2015Q3) period using multivariable regressions.

RESULTS:

Among 212,146 women undergoing benign hysterectomy at 511 hospitals, use of laparoscopic supracervical hysterectomy decreased from 15.1% in 2013Q4 to 6.2% in 2015Q3. Use of laparoscopic supracervical hysterectomy at these 511 hospitals exhibited four distinct trajectory patterns: persistent low use (mean risk-adjusted utilization rate of laparoscopic supracervical hysterectomy changed from 2.8% in 2013Q4 to 0.6% in 2015Q3), decreased medium use (17.0% to 6.9%), decreased high use (51.4% to 24.2%), and rapid abandonment (30.5% to 0.8%). Meanwhile, use of open abdominal hysterectomy increased by 2.1, 4.1, 7.8, and 11.8 percentage points between the pre-warning and post-warning period in these four trajectory groups, respectively. Compared to the pre-warning period, risk of major complications in the post-warning period decreased among patients at “persistent low use” hospitals (adjusted odds ratio [OR] = 0.88, 95% confidence interval [CI]: 0.81–0.94). In contrast, the risk of major complications increased among patients at “rapid abandonment” hospitals (adjusted OR = 1.48, 95% CI: 1.11–1.98) and the risk of minor complications increased among patients at “decreased high use” hospitals (adjusted OR = 1.31, 95% CI: 1.01–1.72).

CONCLUSIONS:

Hospitals varied in their use of laparoscopic supracervical hysterectomy after safety warnings about power morcellation. Complication risk increased at hospitals that shifted considerably towards open abdominal hysterectomy.

INTRODUCTION

The safety of uterine power morcellation, a technique once commonly used in gynecologic surgery, has been scrutinized in recent years. The U.S. Food and Drug Administration (FDA) issued a warning in April 2014 cautioning that uterine power morcellation may inadvertently disseminate cancer cells if patients have undiagnosed uterine malignancies, leading to upstaging and compromised survival.¹ The FDA further requested in November 2014 that laparoscopic power morcellators should include a boxed warning about this safety concern.² These actions led to important changes in the practice of hysterectomies.

Several studies reported a significant shift away from laparoscopic hysterectomies, especially laparoscopic supracervical hysterectomy (LSH) – which typically requires morcellation to remove the corpus uteri laparoscopically while preserving the cervix.^3–8 However, data on how hospitals vary in their practice changes are sparse. Hospital variation in management of hysterectomy (e.g., surgical route, concomitant adnexal procedures, and patient outcomes) is well-documented.^9–14 Since uncontained power morcellation can be avoided by utilizing contained power morcellation,¹⁵ alternative types of minimally invasive hysterectomy, or open abdominal hysterectomy, hospitals may differ considerably in their responses.

The shift away from laparoscopic hysterectomy has raised concerns for a potential increase in surgical complications since abdominal hysterectomy has greater morbidity risk than laparoscopic hysterectomy.¹⁶ However, empirical evidence is mixed. Many studies found no significant change in complications among hysterectomies overall, whereas some reported higher rates of complications in selected patient groups.^6,8,17 This is likely due to heterogeneity in how practice changed across hospitals with varying degrees of reversion to abdominal hysterectomy, which could mask its impact on patient outcomes.

To better elucidate practice changes and safety implications, this study examined hospital variation in the trajectory of LSH use after safety warnings about power morcellation. For hospitals that had different response trajectories, we compared their institutional characteristics, changes in reversion to abdominal hysterectomy, and risk of surgical complications.

MATERIALS AND METHODS

Data and Study Population

This study used the State Inpatient Databases (SID) and State Ambulatory Surgery and Services Databases (SASD) from the Agency for Healthcare Research and Quality Healthcare Cost and Utilization Project,^18,19 as well as data from the New York Statewide Planning and Research Cooperative System (SPARCS).²⁰ The SID, SASD and SPARCS databases contained complete enumeration of all hospital inpatient and outpatient encounters in a given state-year regardless of payer. Patient sociodemographic characteristics and diagnosis/procedure-related information were available for each encounter. We acquired SID and SASD databases from the following 14 states that documented both inpatient and outpatient procedures and hospital identifiers: Colorado, Florida, Iowa, Kentucky, Maryland, Minnesota, Nebraska, Nevada, New Jersey, North Carolina, Oregon, Utah, Vermont, and Wisconsin. Along with New York, these states span all four Census regions and represent 31.8% of the U.S. population.²¹

We limited our sample to adult women who underwent a hysterectomy from 10/1/2013–9/30/2015. The fourth quarter of 2013 (2013Q4) served as the baseline because the potential role of power morcellation in spreading undiagnosed uterine cancer was first publicized in a Wall Street Journal article on December 18, 2013²² and data in our sample suggested that use of LSH continued to increase until 2013Q4. We used data up to the third quarter of 2015 (2015Q3) to capture subsequent changes in practice. More recent quarters/years were excluded because transition of the International Classification of Diseases (ICD) coding system from the 9^th to the 10^th Revision in October 2015 could confound measurement of changes in procedures and morbidities. Hysterectomies were identified using ICD-9 procedure codes and current procedural terminology (CPT) codes.

To focus on routine gynecologic patients with benign indications, we excluded women who underwent hysterectomy for obstetric conditions, were admitted from the emergency department or were transferred in, had any cancer diagnosis, underwent a radical hysterectomy or pelvic exenteration, or had elevated risk for gynecologic cancer (endometrial hyperplasia, cervical dysplasia or other cervical abnormality, elevated CA125, genetic susceptibility to breast, ovarian or endometrial cancer). These eligibility criteria were determined based on admission source and diagnosis/procedure codes. Hysterectomies performed at freestanding ambulatory surgery centers were excluded as we could not measure characteristics of these facilities. To ensure an adequate sample size for assessing each hospital’s practice and non-linear trend in practice changes, we limited our analysis to hospitals that had data from the baseline quarter and at least two additional calendar quarters and had at least 10 eligible hysterectomies per quarter.

We linked hospitals in the SID, SASD and SPARCS data to the 2014 American Hospital Association (AHA) annual survey database to measure institutional characteristics.²³ The AHA annual survey data provided comprehensive measures of each hospital’s infrastructure, capacity, organization, and financial characteristics. This study was approved by the Yale Human Investigation Committee.

Measures

For each patient, we defined a binary indicator (yes/no) for whether she underwent an LSH, as opposed to other routes of hysterectomy. This categorization was based on ICD-9 procedure code 68.31 and CPT procedure codes 58541–58544.

We measured occurrence of major and minor complications for each patient during the hospital stay. Major complications included in-hospital mortality, transfer-out, and severe morbidities (e.g., sepsis, hemorrhage/blood transfusion, stroke, pulmonary or deep vein embolism). Minor complications included less severe outcomes such as operative wound disruption, urinary tract infection, and nausea/vomiting. These conditions were determined based on patients’ disposition status and diagnosis/procedure codes.^24,25

Patient clinical risk factors included age, surgical indication, smoking status, comorbidities, and concomitant procedures. Using diagnosis codes, we classified surgical indication into categories such as uterine leiomyoma, menstrual disorders, genital prolapse, endometriosis, female pelvic inflammatory diseases, and postmenopausal bleeding. We measured the Elixhauser comorbidities^26,27 based on diagnosis codes. Common comorbidities (obesity, anemia, hypertension, diabetes, chronic pulmonary disease, hypothyroidism, and depression) were included as individual binary indicators (yes/no), whereas other comorbidities were accounted for using a count variable because of their low frequency. Using procedure codes, we constructed two additional binary indicators for whether the patient underwent any concomitant abdominopelvic surgery (yes/no) or concomitant other surgery (yes/no).

Hospital characteristics included teaching status, urban/rural location, type of ownership, affiliation with a multi-hospital system, annual volume, and patient composition. Annual volume was measured based on a hospital’s average annual number of benign hysterectomies that met sample eligibility criteria in our study period. Patient composition measured the proportion of a hospital’s eligible hysterectomies where the patients were racial/ethnic minorities (i.e., nonwhite); where Medicaid, self-pay or no charge was the primary source of payment (i.e., safety-net proportion); and where the patients’ income (based on zip code medium household income) was in the highest quartile.

Statistical Analysis

We estimated a patient’s likelihood of undergoing LSH (yes/no) using logistic regression via the generalized estimating equation (GEE) approach to account for clustering of data by hospital,²⁸ while adjusting for patient age, surgical indication, smoking status, comorbidities, concomitant procedures, and indicators of calendar quarter. From this model, we calculated a risk-adjusted rate of LSH use for each hospital in each calendar quarter by dividing its observed number of LSHs by expected number of LSHs (conditional on its patient characteristics) and then multiplying by the sample overall observed rate of LSH use in the corresponding calendar quarter (see Appendix A).²⁸

Using these risk-adjusted rates of LSH use, we applied a growth mixture modeling technique to identify distinct hospital groups that exhibited different trajectories of LSH use (see Appendix A).²⁹ Trajectories were modeled as a function of time, i.e., number of calendar quarters since baseline. The highest order polynomial term of time (e.g., quadratic or cubic function of time) that was statistically significant determined the shape of each trajectory. We evaluated models with an increasing number of trajectories and selected the optimal number of trajectories based on the Bayesian information criterion³⁰ and posterior probability of trajectory group membership (i.e., the probability that a hospital belonged to a give trajectory based on its observed data).²⁹ Each hospital was assigned to the trajectory group for which it had the maximum posterior probability of membership. We then compared hospital characteristics across the different trajectory groups using chi-square test for categorical variables and Wilcoxon rank sum test for continuous variables. Further pairwise comparisons were performed by adjusting for multiple comparison.^31–35

We also examined changes in patients’ risk of complications over time. Since the FDA warnings about power morcellation were released in April and November 2014, we defined 2013Q4–2014Q1 as the pre-warning period, 2014Q2–2014Q4 as the transition period, and 2015Q1–2015Q3 as the post-warning period. Within each trajectory group, we compared the risk of major complications and minor complications between these time periods, first using chi square tests and then using logistic regressions via the GEE approach accounting for patients’ clinical risk factors and clustering of hysterectomies by hospital. Using similar analysis, we also examined changes in complication risk among a subset of women who underwent hysterectomy for uterine leiomyoma - the patient population that was the focus of the FDA warnings.

P values below 0.05 were considered statistically significant. Analyses were performed using SAS version 9.4 (SAS Institute Inc., Cary, NC).

RESULTS

Sample Characteristics

A total of 212,146 women from 511 hospitals met sample eligibility criteria (mean age=46.9 years, standard deviation=10.5). Uterine leiomyoma (58.4%), menstrual disorder (50.9%), and endometriosis (31.1%) were the most common indications (categories not mutually exclusive) (Table 1). Most hysterectomies (62.7%) were outpatient procedures. In the overall sample, use of LSH decreased from 15.1% of benign hysterectomies in 2013Q4 to 6.2% in 2015Q3.

Table 1.

Characteristics of patients in the sample (N=212,146)

Patient Characteristics	N (%)
Age, years
18–34	18,466 (8.7)
35–44	76,649 (36.1)
45–54	78,799 (37.1)
55–64	21,115 (10.0)
≥65	17,117 (8.1)
Race/ethnicitya
Non-Hispanic white	117,213 (62.1)
Non-Hispanic black	35,391 (18.7)
Hispanic	18,675 (9.9)
Other	9,240 (4.9)
Unknown	8,297 (4.4)
Primary payer
Private insurance	153,631 (72.4)
Medicare	21,201 (10.0)
Medicaid	27,308 (12.9)
Self-pay	3,410 (1.6)
Other payer	5,063 (2.4)
No charge	1,277 (0.6)
Unknown	256 (0.1)
Household income (based on zip code)
Lowest quartile	47,311 (22.3)
Lower middle quartile	56,123 (26.5)
Upper middle quartile	53,344 (25.1)
Highest quartile	52,711 (24.8)
Unknown	2,657 (1.3)
Surgical indicationb
Uterine leiomyoma	123,809 (58.4)
Other disorders of corpus uteri	29,559 (13.9)
Menstrual disorder	107,997 (50.9)
Endometriosis	66,040 (31.1)
Genital prolapse	39,957 (18.8)
Inflammatory diseases of female pelvic organs	61,547 (29.0)
Disorders of ovary or fallopian tube	61,092 (28.8)
Noninflammatory disorders of cervix	5,029 (2.4)
Postmenopausal bleeding	7,629 (3.6)
Other menopausal disorder	2,817 (1.3)
Urinary incontinence	21,782 (10.3)
Abdominal pain	2,318 (1.1)
Abdominal/pelvic mass	1,401 (0.7)
Other female genital disorders	42,498 (20.0)
Setting
Inpatient	79,057 (37.3)
Outpatient	133,089 (62.7)
Concomitant abdominopelvic procedure	21,769 (10.3)
Concomitant other procedure	1,370 (0.6)
Smoking	35,189 (16.6)
Comorbidities
Hypertension	48,907 (23.1)
Diabetes	14,635 (6.9)
Anemia	31,799 (15.0)
Obesity	27,515 (13.0)
Chronic pulmonary disease	20,583 (9.7)
Hypothyroidism	17,188 (8.1)
Depression	16,780 (7.9)

Percentages may not add to 100 due to rounding.

^a.N=23,330 had missing data. Race/ethnicity was not available in some state-years.

^b.Surgical indications were not mutually exclusive. A patient could have more than one indication.

The 511 hospitals all had at least 10 eligible patients in every calendar quarter from 2013Q4–2015Q3. These hospitals were diverse in characteristics: 55.4% were teaching institutions, 11.9% were in rural areas, and 75.3% had private non-profit ownership (Table 2). Their median annual volume of eligible benign hysterectomies was 155 (interquartile range: 100–266).

Table 2.

Characteristics of hospitals in the sample (N=511) and association with the different response trajectories

Hospital Characteristics	Overall Sample (N=511)	Persistent Low Use (N=246)	Decreased Medium Use (N=199)	Decreased High Use (N=48)	Rapid Abandonment (N=18)	P Value
Teaching status						0.001a
Yes	283 (55.4)	118 (48.0)	131 (65.8)	27 (56.3)	7 (38.9)
No	228 (44.6)	128 (52.0)	68 (34.2)	21 (43.8)	11 (61.1)
Location						0.004b
Urban	450 (88.1)	205 (83.3)	188 (94.5)	42 (87.5)	15 (83.3)
Rural	61 (11.9)	41 (16.7)	11 (5.5)	6 (12.5)	3 (16.7)
Type of ownership						0.19
Government (non-federal)	53 (10.4)	31 (12.6)	19 (9.5)	3 (6.3)	0 (0.0)
Private non-profit	385 (75.3)	173 (70.3)	158 (79.4)	38 (79.2)	16 (88.9)
Private for-profit	73 (14.3)	42 (17.1)	22 (11.1)	7 (14.6)	2 (11.1)
Affiliation with a multi-hospital system						0.13
Yes	388 (75.9)	180 (73.2)	157 (78.9)	40 (83.3)	11 (61.1)
No	123 (24.1)	66 (26.8)	42 (21.1)	8 (16.7)	7 (38.9)
Annual volume (number of eligible benign hysterectomies)	155 (100–266)	136 (90–202)	229 (116–363)	151 (100–261)	114 (83–151)	<0.001c
% of patients that were racial/ethnic minoritiesd	26.6 (11.7–49.0)	23.4 (9.3–45.1)	27.6 (14.2–49.0)	26.2 (17.2–52.5)	26.5 (12.8–61.7)	0.16
% of patients with Medicaid or were uninsured or no charge	15.0 (7.4–24.5)	18.1 (10.0–27.4)	12.0 (5.6–20.6)	11.9 (4.7–22.1)	13.0 (8.3–19.5)	<0.001e
% of patients in highest income quartile	12.9 (2.6–36.7)	6.7 (0.7–19.2)	21.1 (7.4–43.9)	26.1 (2.9–61.3)	17.3 (1.9–68.2)	<0.001f

Data were reported as n (%) or median (interquartile range). Percentages may not add to 100 due to rounding.

^a.In pairwise comparison (with p values adjusted to account for multiple comparison), “Persistent low use” differed significantly from “Decreased medium use” (p=0.001).

^b.In pairwise comparison (with p values adjusted to account for multiple comparison), “Persistent low use” differed significantly from “Decreased medium use” (p=0.002).

^c.In pairwise comparison (with p values adjusted to account for multiple comparison), “Decreased Medium Use” differed significantly from “Persistent low use” (p<0.001), “Decreased High Use” (p=0.03), and “Rapid Abandonment” (p<0.001).

^d.Based on N=451 hospitals (data in some state-years did not have race/ethnicity measures).

^e.In pairwise comparison (with p values adjusted to account for multiple comparison), “Persistent low use” differed significantly from “Decreased medium use” (p<0.001) and “Decreased high use” (p=0.03).

^f.In pairwise comparison (with p values adjusted to account for multiple comparison), “Persistent low use” differed significantly from “Decreased medium use” (p<0.001) and “Decreased high use” (p<0.001).

Distinct Response Trajectories

The growth mixture model classified the 511 hospitals into four distinct trajectories of LSH use: 1) persistent low use (mean risk-adjusted rate of LSH use changed from 2.8% in 2013Q4 to 0.6% in 2015Q3); 2) decreased medium use (17.0% to 6.9%); 3) decreased high use (51.4% to 24.2%), and 4) rapid abandonment (30.5% to 0.8%) (Figure 1). These trajectory patterns accounted for 48.1% (n=246), 38.9% (n=199), 9.4% (n=48), and 3.5% (n=18) of the hospitals, respectively, and 37.9% (n=80,303), 51.5% (n=109,182), 8.6% (n=18,228), and 2.1% (n=4,433) of the patients, respectively.

Figure 1.

Distinct response trajectories among hospitals regarding use of laparoscopic supracervical hysterectomy over time LSH = laparoscopic supracervical hysterectomy; Q = quarter. Rate of LSH use reflects risk-adjusted rate for each hospital in each calendar quarter after accounting for differences in patient clinical risk factors. Solid lines reflect mean risk-adjusted rate among hospitals in each trajectory. Dashed lines reflect trajectories predicted by the growth mixture model, with error bars reflecting 95% confidence intervals.

Hospitals in the four trajectory groups differed significantly in characteristics (Table 2). Compared to hospitals in the “decreased medium use” group, those in the “persistent low use” or “rapid abandonment” groups had smaller volume (median=229 versus 136 and 114, respectively, p<0.001 for both). Compared to hospitals in the “decreased medium use” group, those in the “persistent low use” group were also more likely to be in rural locations (5.5% versus 16.7%, p=0.002) and less likely to be teaching hospitals (65.8% versus 48.0%, p=0.001). Additionally, compared to hospitals in the “decreased medium use” and “decreased high use” groups, those in the “persistent low use” group had a higher safety-net proportion (median=12.0% and 11.9% versus 18.1%, p<0.001 and p=0.03, respectively) and a lower proportion of patients with high income (median=21.1% and 26.1% versus 6.7%, p<0.001 for both).

Surgical Outcomes

The proportion of hysterectomies with a major complication decreased from 8.6% in the pre-warning period to 8.0% in the post-warning period in the “persistent low use” group (p=0.048), but increased from 6.3% to 7.1% in the “decreased high use” group (p=0.006) and from 4.8% to 7.0% in the “rapid abandonment” group (p=0.02) (Table 3). The rate of major complications remained stable in the “decreased medium use” group (pre-warning: 8.3%, post-warning: 8.2%, p=0.50). After adjusting for clinical risk factors, patients in the “rapid abandonment” group had a significantly higher risk of major complications in the transition and post-warning period than in the pre-warning period (adjusted odds ratio [OR]=1.44, 95% confidence interval [CI]: 1.04–2.01; and adjusted OR=1.48, 95% CI: 1.11–1.98; respectively). In contrast, among women in the “persistent low use” group, their risk of major complications was lower in the post-warning than pre-warning period (adjusted OR=0.88, 95% CI: 0.81–0.94).

Table 3.

Changes in risk of surgical complications among patients undergoing hysterectomy, stratified by trajectory group

Complicationsa	Persistent Low Use (N=80,303)	Decreased Medium Use (N=109,182)	Decreased High Use (N=18,228)	Rapid Abandonment (N=4,433)
Major complication
Unadjusted, n/N (%)
Pre-warning	1,745/20,272 (8.6)	2,309/27,842 (8.3)	302/4,775 (6.3)	58/1,198 (4.8)
Transition	2,588/30,910 (8.4)	3,516/41,718 (8.4)	559/7,113 (7.9)	125/1,730 (7.2)
Post-warning	2,331/29,121 (8.0)	3,249/39,622 (8.2)	449/6,340 (7.1)	105/1,505 (7.0)
P value	0.048	0.50	0.006	0.02
Adjusted odds ratio (95% CI)b
Pre-warning	Reference	Reference	Reference	Reference
Transition	0.95 (0.89–1.01)	1.00 (0.91–1.09)	1.21 (1.05–1.40)	1.44 (1.04–2.01)
Post-warning	0.88 (0.81–0.94)	0.95 (0.87–1.03)	1.07 (0.92–1.24)	1.48 (1.11–1.98)
Minor complication
Unadjusted, n/N (%)
Pre-warning	890/20,272 (4.4)	1,189/27,842 (4.3)	156/4,775 (3.3)	35/1,198 (2.9)
Transition	1,363/30,910 (4.4)	1,817/41,718 (4.4)	279/7,113 (3.9)	62/1,730 (3.6)
Post-warning	1,274/29,121 (4.4)	1,680/39,622 (4.2)	269/6,340 (4.2)	61/1,505 (4.1)
P value	0.98	0.70	0.03	0.29
Adjusted odds ratio (95% CI)b
Pre-warning	Reference	Reference	Reference	Reference
Transition	1.00 (0.90–1.10)	1.00 (0.92–1.09)	1.21 (0.93–1.57)	1.19 (0.63–2.25)
Post-warning	0.97 (0.87–1.08)	0.95 (0.88–1.03)	1.31 (1.01–1.72)	1.60 (0.97–2.64)

CI = confidence interval.

^a.Major complications and minor complications were not mutually exclusive. A patient could have both major and minor complications.

^b.Based on multivariable regression models that also adjusted for patient age, surgical indication, smoking status, comorbidities, and concomitant procedures, while accounting for clustering of hysterectomies by hospital.

When evaluated by type of major complications (Table 4), hemorrhage/blood transfusion was the category that increased the most in the “rapid abandonment” group (pre-warning: 2.3%, post-warning: 4.3%, p=0.02). In contrast, rate of hemorrhage/blood transfusion decreased in the “persistent low use” group (pre-warning: 5.1%, post-warning: 4.5%, p=0.01). Other types of major complications had lower frequencies and their changes between the pre- and post-warning period were generally small.

Table 4.

Changes in different categories of major complications, stratified by trajectory group

Categories of Major Complicationsa	Persistent Low Use (N=80,303)	Decreased Medium Use (N=109,182)	Decreased High Use (N=18,228)	Rapid Abandonment (N=4,433)
Hemorrhage/blood transfusion, n/N (%)
Pre-warning	1,033/20,272 (5.1)	1,292/27,842 (4.6)	187/4,775 (3.9)	28/1,198 (2.3)
Transition	1,483/30,910 (4.8)	2,072/41,718 (5.0)	342/7,113 (4.8)	68/1,730 (3.9)
Post-warning	1,312/29,121 (4.5)	1,942/39,622 (4.9)	268/6,340 (4.2)	65/1,505 (4.3)
P value	0.01	0.13	0.052	0.02
Operative injury, n/N (%)
Pre-warning	453/20,272 (2.2)	534/27,842 (1.9)	69/4,775 (1.4)	14/1,198 (1.2)
Transition	692/30,910 (2.2)	821/41,718 (2.0)	132/7,113 (1.9)	39/1,730 (2.3)
Post-warning	610/29,121 (2.1)	712/39,622 (1.8)	101/6,340 (1.6)	23/1,505 (1.5)
P value	0.41	0.19	0.20	0.07
Infection/sepsis, n/N (%)
Pre-warning	91/20,272 (0.4)	119/27,842 (0.4)	21/4,775 (0.4)	-b
Transition	108/30,910 (0.3)	160/41,718 (0.4)	19/7,113 (0.3)	-b
Post-warning	115/29,121 (0.4)	152/39,622 (0.4)	21/6,340 (0.3)	-b
P value	0.21	0.60	0.28
Cardiac/pulmonary complications, n/N (%)
Pre-warning	321/20,272 (1.6)	466/27,842 (1.7)	69/4,775 (1.4)	14/1,198 (1.2)
Transition	474/30,910 (1.5)	778/41,718 (1.9)	113/7,113 (1.6)	19/1,730 (1.1)
Post-warning	437/29,121 (1.5)	719/39,622 (1.8)	95/6,340 (1.5)	22/1,505 (1.5)
P value	0.76	0.17	0.81	0.63
Other, n/N (%)c
Pre-warning	181/20,272 (0.9)	314/27,842 (1.1)	29/4,775 (0.6)	14/1,198 (1.2)b
Transition	291/30,910 (0.9)	391/41,718 (0.9)	56/7,113 (0.8)	16/1,730 (0.9)b
Post-warning	267/29,121 (0.9)	396/39,622 (1.0)	53/6,340 (0.8)	12/1,505 (0.8)b
P value	0.85	0.047	0.36	0.61

^a.Categories were not mutually exclusive. A patient could have more than one category of major complications.

^b.To protect confidentiality and avoid reporting counts ≤10, patients with “Infection/sepsis” were combined with those who had “Other” complications for the “rapid abandonment” group.

^c.“Other” complications included acute kidney failure, stroke, generalized ischemic cerebrovascular disease, postoperative shock, anaphylaxis, ostomy complications, foreign body left behind during procedure, reoperation, death, and transfer-out.

Regarding minor complications, the risk increased significantly in the “decreased high use” group (pre-warning: 3.3%, post-warning: 4.2%, adjusted OR=1.31, 95% CI: 1.01–1.72) (Table 3). The risk of minor complications also increased in the “rapid abandonment” group but it did not reach statistical significance (adjusted OR=1.60, 95% CI: 0.97–2.64, p=0.07), likely due to the smaller sample size in this group.

These patterns of change in complication risk are consistent with the different extent of reversion to open abdominal hysterectomies across the trajectory groups. Between the pre-warning and post-warning period, the proportion of benign hysterectomies performed via an open abdominal approach increased the most in the “decreased high use” and “rapid abandonment” groups (by 7.8 and 11.8 percentage points, respectively) but the least in the “persistent low use” group (by 2.1 percentage points) (Appendix B). Further analysis focusing on women undergoing hysterectomy for uterine leiomyoma generated similar results (Appendix C).

COMMENT

Principal Findings

We identified important variation among hospitals in how they responded to the FDA warnings about power morcellation. Some hospitals virtually eliminated LSH, whereas other hospitals retained a substantial level of LSH use. Hospitals also varied considerably in how much they shifted hysterectomies to an open abdominal approach. Risk of complications increased in two trajectory groups where there were considerable shifts to abdominal hysterectomy.

Results in the Context of What is Known

Prior research evaluating the impact of power morcellation warnings has focused on practice changes in the overall population. We showed that hysterectomies in the “decreased medium use” group accounted for the largest proportion of patients (51.5%) where there was no significant change in complication risk. The experience of these patients might dominate evaluation of the overall population, which can mask important safety impact of practice changes and might explain why prior research often did not find changes in complication risk. By identifying distinct groups of hospitals and analyzing patients’ experience within each group separately, we found evidence partially supporting the concern that reversion to abdominal procedures could adversely affect patients’ outcomes. However, only a small proportion of hospitals shifted their practice substantially to an abdominal approach and incurred worsening outcomes. Hospitals that had minimal increases in use of abdominal hysterectomy actually had a reduction in complication risk, especially in the risk of hemorrhage/blood transfusion, likely reflecting quality improvement efforts over time.

Research and Clinical Implications

Hospitals in the “rapid abandonment” and “decreased high use” groups acted quickly to avoid inadvertent morcellation of cancerous tissues by abandoning or considerably reducing LSH use. However, this was accompanied by a substantial shift to abdominal hysterectomy and patients at these hospitals experienced a significant increase in risk of surgical complications. Efforts to promote appropriate use of alternative approaches of minimally invasive hysterectomy may be beneficial. For instance, uterine tissues can often be removed without morcellation via total laparoscopic hysterectomy or laparoscopically-assisted vaginal hysterectomy, allowing patients to still benefit from the lower complication risk of minimally invasive surgery.³⁶

Despite reduced use of LSH, hospitals in the “decreased high use” group still performed 24.2% of its hysterectomies via LSH in the post-warning period. As our data lacked clinical detail regarding preoperative evaluations (e.g., assessment for potential occult uterine cancer) and method of tissue extraction (e.g., contained power morcellation or other methods), we were unable to investigate the appropriateness of these LSH procedures. The risk of occult uterine cancer needs to be closely monitored at these hospitals.

Several hospital characteristics were associated with the distinct response trajectories. Hospitals in rural locations, with smaller volumes, and serving more socioeconomically disadvantaged patients tended to follow trajectories with low use of LSH or rapid abandonment of LSH. Conversely, urban hospitals, teaching institutions, hospitals with larger volumes, and hospitals with lower safety-net proportions tended to retain higher use of LSH. It is likely that smaller, rural, and less resource-rich hospitals lack the capacity to adapt practices (e.g., lack of expertise or technology). This is consistent with prior research suggesting that smaller, nonteaching, and rural hospital are slower in adopting medical innovations.^17,37,38 Efforts to better understand the needs at these hospitals will help them enhance care.

Strengths and Limitations

Unique strengths of this study include our examination of hospital heterogeneity in practice and the inclusion of both inpatient and outpatient hysterectomies from a large number of states across the country. However, we recognize several study limitations. First, we measured complications during in-hospital stay and could not assess complications after discharge. As many complications following abdominal hysterectomy occur after discharge, we might have underestimated the adverse consequences of shifting from laparoscopic to abdominal hysterectomies.³⁹ Second, we relied on administrative data to measure complications and risk factors, which may lack sufficient clinical detail and accuracy. Likewise, we limited our analysis to the utilization pattern of LSH, instead of power morcellation, since ICD/CPT procedure codes in administrative data could not identify use of power morcellation. Third, our sample included hospitals with at least 10 eligible hysterectomies per quarter which were relatively large hospitals. To inform generalizability of our findings to smaller hospitals, we conducted a sensitivity analysis including hospitals with at least 5 eligible hysterectomies per quarter and found very similar trajectory patterns. Fourth, despite our large overall sample size, the number of hospitals or patients in some analyses were small. This limited the statistical power in detecting differences in hospital characteristics across trajectory groups or patient complication risk over time. Finally, our data lacked information on surgeon characteristics, which can also influence the selection of surgical route and patient outcomes. Potential variation in surgeons’ practices warrants further research.

Conclusions

Hospitals varied in how they responded to safety warnings about power morcellation. Some hospitals had a substantial shift from laparoscopic to abdominal hysterectomies with a concomitant increase in complication risk. Efforts to promote safe use of alternative minimally invasive techniques, as well as targeted support for lower capacity hospitals, may help improve patient outcomes.

AJOG at a Glance:

1. Why was the study conducted?

   • Safety warnings about power morcellation have changed hysterectomy practice; yet there were little data on whether hospitals differed in their responses and what impact this had on patient outcomes.
2. What are the key findings?

   • There were distinct groups of hospitals that differed considerably in how they reduced use of laparoscopic supracervical hysterectomy over time.
   • Hospitals in these distinct Journalgroupsalsovariedin howmuchtheyshiftedhysterectomy to an open abdominal approach.
   • Complication risk decreased at hospitals that had minimal shifts to open abdominal hysterectomy but increased at hospitals that had large shifts to open abdominal hysterectomy.
3. What does this study add to what is already known?

   • It is important to understand heterogeneity in hospital practice and the implications for patient safety.

Appendix A. Additional Technical Detail

Risk-Adjustment Model

Using data from all eligible patients in the entire study period (2013Q4–2015Q3), we estimated a patient’s likelihood of undergoing laparoscopic supracervical hysterectomy (LSH) (yes/no) using logistic regression via the generalized estimating equation approach to account for clustering of data by hospital. Following the National Quality Forum Measure Evaluation Criteria for the design of risk-adjustment models, our risk-adjustment model included patients’ pre-existing clinical risk factors but not factors “related to disparities in care or the quality of care”.¹ The model adjusted for patient age, surgical indication, smoking status, comorbidities, concomitant procedures, and indicators of calendar quarter. Performance of the risk-adjustment model was evaluated by the c-statistic, which was 0.69 in our analysis. Surgical indication, smoking status, comorbidities, concomitant procedures, and indicators of calendar quarter were all significantly associated with LSH use.

Similar to the methodology used in prior research,² we calculated a risk-adjusted rate of LSH use for each hospital in each calendar quarter based on results from the risk-adjustment model. Specifically, for each patient at a given hospital in a particular calendar quarter, we calculated her expected probability of undergoing LSH based on her clinical risk factors and coefficient estimates from the risk-adjustment model. The sum of these probabilities across all patients at this hospital in this calendar quarter reflected the expected number of LSHs for the hospital-quarter. The risk-adjusted rate of LSH for this hospital-quarter was calculated by dividing the observed number of LSHs among these patients by the expected number of LSHs, and then multiplying by the sample overall observed rate of LSH use in the corresponding calendar quarter. Multiplication by the sample overall observed rate of LSH use from the corresponding calendar quarter helped convert the risk-adjusted measure to the original scale of LSH utilization rate and facilitate assessment of temporal trends in LSH use over time.

Identification of Distinct Hospital Trajectory Groups

To identify distinct groups of hospitals that exhibited different trajectories of LSH use, we analyzed hospital-quarter specific data on risk-adjusted rate of LSH use using a group-based growth mixture model via a SAS procedure PROC TRAJ.³ The group-based approach assumed that there were different groups of hospitals that followed qualitatively distinct trajectories of LSH use. Hospitals within the same group shared a same pattern of trajectories, whereas hospitals in different groups had distinct trajectories. The trajectories of LSH use were modeled as a function of time, i.e., number of calendar quarters since baseline. The highest order polynomial term of time (e.g., quadratic or cubic function of time) that was statistically significant determined the shape of the trajectory in each group. The model allowed for different parameter estimates for different trajectory groups and used a maximum likelihood function to estimate the relevant parameters (e.g., coefficients of the polynomial terms of time and proportion of hospitals in each group). We compared alternative models with an increasing number of trajectory groups and determined the optimal number of groups based on the Bayesian Information Criteria (BIC) and the average posterior probability of trajectory group membership among hospitals. Posterior probability of trajectory group membership reflects the probability that a hospital belonged to a given trajectory based on its observed data. Higher values of average posterior probability of group membership among hospitals indicate better model performance. Preference was given to a model with a smaller number of groups if alternative models only had marginal improvement in BIC but substantively similar trajectory patterns with low prevalence of certain groups or large decline in average posterior probability of group membership. Our final analysis used a four-group model because BIC score plateaued in subsequent models and the four-group model achieved better average posterior probability of group membership than subsequent models.

Appendix B. Proportion of benign hysterectomies performed via an open abdominal approach over time, stratified by trajectory group

Appendix C. Changes in risk of surgical complications among the subset of patients who underwent hysterectomy for uterine leiomyoma, stratified by trajectory group

Complicationsa	Persistent Low Use (N=44,116)	Decreased Medium Use (N=65,481)	Decreased High Use (N=11,375)	Rapid Abandonment (N=2,837)
Major complication
Unadjusted, n/N (%)
Pre-warning	1,072/10,942 (9.8)	1,546/16,432 (9.4)	202/2,954 (6.8)	40/731 (5.5)
Transition	1,638/16,960 (9.7)	2,361/24,923 (9.5)	388/4,394 (8.8)	92/1,123 (8.2)
Post-warning	1,470/16,214 (9.1)	2,233/24,126 (9.3)	307/4,027 (7.6)	72/983 (7.3)
P value	0.08	0.70	0.006	0.08
Adjusted odds ratio (95% CI)b
Pre-warning	Reference	Reference	Reference	Reference
Transition	0.95 (0.87–1.04)	0.98 (0.90–1.08)	1.21 (0.9999–1.47)	1.53 (1.03–2.27)
Post-warning	0.88 (0.80–0.97)	0.95 (0.87–1.04)	1.06 (0.89–1.28)	1.52 (1.07–2.15)
Minor complication
Unadjusted, n/N (%)
Pre-warning	502/10,942 (4.6)	676/16,432 (4.1)	83/2,954 (2.8)	23/731 (3.1)
Transition	800/16,960 (4.7)	1,122/24,923 (4.5)	171/4,394 (3.9)	49/1,123 (4.4)
Post-warning	726/16,214 (4.5)	1,032/24,126 (4.3)	179/4,027 (4.4)	34/983 (3.5)
P value	0.58	0.15	0.002	0.34
Adjusted odds ratio (95% CI)b
Pre-warning	Reference	Reference	Reference	Reference
Transition	1.01 (0.88–1.16)	1.08 (0.97–1.20)	1.28 (0.90–1.83)	1.36 (0.62–2.97)
Post-warning	0.95 (0.81–1.10)	1.01 (0.91–1.12)	1.52 (1.12–2.05)	1.37 (0.75–2.53)

CI = confidence interval.

^a.Major complications and minor complications were not mutually exclusive. A patient could have both major and minor complications.

^b.Based on multivariable regression models that also adjusted for patient age, surgical indication (other than uterine leiomyoma), smoking status, comorbidities, and concomitant procedures, while accounting for clustering of hysterectomies by hospital.