aAbortions could be excluded for more than 1 reason.
bThere are 50 different possible settings that include settings such as schools, homeless shelters, inpatient hospital, skilled nursing facilities, and others. Facility type was classified as an ambulatory surgical center (ASC) when the place-of-service code variable (stdplac) equaled 24 (“ambulatory surgery center”) and office-based setting when the place-of-service code variable equaled 11 (“office”).
cEvents identified through individual review of cases with inpatient encounters, emergency department visits, and complication diagnosis codes were mutually exclusive (ie, a patient’s disposition that was found by more than 1 of the 3 methods was counted once).
dEvents identified through programming were not mutually exclusive; ie, an abortion could have had 1 or more of the programmed events. These abortions were not individually reviewed by the clinician coder.
eTable 1. Diagnosis, Treatment, and Prescriptions used to create the dataset and the variables in the dataset
eTable 2. Unadjusted associations between variables controlled for in adjusted analyses and study outcomes
eTable 3. Multivariable logistic regressions of odds of abortion-related morbidity or adverse events and adjusted incidence rates of events (aInc) after abortion in ambulatory surgery centers (ASCs) vs office-based settings among total sample. Full regression results
eTable 4. Multivariable logistic regressions of odds of any event after abortion in ambulatory surgery centers (ASCs) vs. office-based settings among total sample. Checking effect of each covariate on main effect of interest.
eTable 5. Multivariable logistic regressions of odds of abortion-related morbidities and adverse events abortion in ambulatory surgery centers (ASCs) vs. office-based settings among total sample. Alternative version of missed ectopic pregnancies (i.e. no diagnosis or treatment within 14 days). n=50 286 abortions
eTable 6. Multivariable logistic regressions of odds of abortion-related morbidities and adverse events in ambulatory surgery centers (ASCs) vs. office-based settings among total sample. Adding additional antibiotics that were present in the dataset but not commonly used to treat abortion-related infections as indicators of infection. n=50,311 abortions
eTable 7. Sample description of Elixhauser Comorbidity Index by facility type (n=50,311 abortions)
eTable 8. Multivariable logistic regressions of odds of abortion-related morbidities and adverse events in ambulatory surgery centers (ASCs) vs. office-based settings among total sample using >=1 Comorbidities from Elixhauser Comorbidity conditions instead of hypertension, diabetes, number of previous visits (n=50,311 abortions)
eTable 9. Multivariable logistic regressions of odds of abortion-related morbidities and adverse events in ambulatory surgery centers (ASCs) vs. office-based settings among total sample Elixhauser Comorbidity Readmission Score instead of hypertension, diabetes, number of previous visits (n=50,311 abortions)
Customize your JAMA Network experience by selecting one or more topics from the list below.
Roberts SCM, Upadhyay UD, Liu G, et al. Association of Facility Type With Procedural-Related Morbidities and Adverse Events Among Patients Undergoing Induced Abortions. JAMA. 2018;319(24):2497–2506. doi:10.1001/jama.2018.7675
Is there an association between the type of facility in which an abortion is performed and abortion-related morbidities and adverse events?
In this retrospective cohort study of 50 311 induced abortions among 49 287 women with private health insurance, performance of the abortion in an ambulatory surgery center compared with an office-based setting was not associated with a significant difference in abortion-related morbidities and adverse events (adjusted odds ratio, 0.97).
Rates of abortion-related morbidities and adverse events did not significantly differ by whether the abortion was performed in an ambulatory surgery center vs an office-based setting.
Multiple states have laws requiring abortion facilities to meet ambulatory surgery center (ASC) standards. There is limited evidence regarding abortion-related morbidities and adverse events following abortions performed at ASCs vs office-based settings.
To compare abortion-related morbidities and adverse events at ASCs vs office-based settings.
Design, Setting, and Participants
Retrospective cohort study of women with US private health insurance who underwent induced abortions in an ASC or office-based setting (January 1, 2011-December 31, 2014). Outcomes were abstracted from a large national private insurance claims database during the 6 weeks following the abortion (date of final follow-up, February 11, 2015).
Facility type for abortion (ASCs vs office-based settings, including facilities such as abortion clinics, nonspecialized clinics, and physician offices).
Main Outcomes and Measures
The primary outcome was any abortion-related morbidity or adverse event (such as retained products of conception, abortion-related infection, hemorrhage, and uterine perforation) within 6 weeks after an abortion. Two secondary outcomes, both subsets of the primary outcome, were major abortion-related morbidities and adverse events (such as hemorrhages treated with a transfusion, missed ectopic pregnancies treated with surgery, and abortion-related infections that resulted in an overnight hospital admission) and abortion-related infections.
Among 49 287 women (mean age, 28 years [SD, 7.3]) who had 50 311 induced abortions, (23 891 [47%] first-trimester aspiration, 13 480 [27%] first-trimester medication, and 12 940 [26%] second trimester or later), 5660 abortions (11%) were performed in ASCs and 44 651 (89%) in office-based settings. Overall, 3.33% had an abortion-related morbidity or adverse event; 0.32% had a major abortion-related morbidity or adverse event; and 0.74% had an abortion-related infection. In adjusted analyses, there was no statistically significant difference between ASCs vs office-based settings, respectively, in the rates of abortion-related morbidities or adverse events (3.25% vs 3.33%, difference, −0.08%; 95% CI, −0.58% to 0.43%; adjusted OR, 0.97; 95% CI, 0.81-1.17), major morbidities or adverse events (0.26% vs 0.33%; difference, −0.06%; 95% CI, −0.18% to 0.06%; adjusted OR, 0.78; 95% CI, 0.45-1.37), or infections (0.58% vs 0.77%; difference, −0.16%; 95% CI, −0.35% to 0.03%; adjusted OR, 0.75; 95% CI, 0.52-1.09).
Conclusions and Relevance
Among women with private health insurance who had an induced abortion, performance of the abortion in an ambulatory surgical center compared with an office-based setting was not associated with a significant difference in abortion-related morbidities and adverse events. These findings, in addition to individual patient and individual facility factors, may inform decisions about the type of facility in which induced abortions are performed.
In June 2016, the US Supreme Court ruled in Whole Woman’s Health vs Hellerstedt1 that the Texas law requiring abortion facilities to meet ambulatory surgery center (ASC) standards was unconstitutional. Despite this ruling, 13 states currently have laws that require abortions to be provided in ASCs.2 These laws include such requirements as specified hall and door widths or separate procedure and recovery rooms.2 Many of these apply only at a specific gestational week (or gestational duration), typically in the second trimester.2 Supporters of ASC laws have argued that ASC requirements make abortion safer.1
Limited published peer-reviewed research has directly compared abortion-related morbidities and adverse events across facility types. One study found fewer abortion-related events in clinics than hospitals3 and a recent review found similar rates of abortion-related events after first-trimester abortion procedures across study populations in different facility types, including hospitals, ASCs, and office-based settings.4
More than 95% of induced abortions are provided in outpatient, non–hospital-based settings—in abortion clinics, nonspecialized clinics, or physician offices.5 Abortions have been performed in these settings for more than 45 years.6 Ambulatory surgery centers developed in the 1980s to move some surgeries and procedures from hospitals to non–hospital-based outpatient settings.7 In the 2000s, some states began passing laws that required abortions to be provided in ASCs, and some abortion clinics that had been office-based settings became licensed as ASCs to continue to perform abortions.8 There are no published national estimates of the proportion of abortions performed in ASCs vs office-based settings. State-specific data indicate that the minority of outpatient abortion facilities are ASCs.9,10 The present study used a private insurance claims database to compare abortion-related morbidities and adverse events at ASCs vs office-based settings.
This retrospective cohort study used 2011-2014 data from the Truven Health MarketScan Commercial Claims and Encounters database to compare abortion-related morbidities and adverse events across 2 facility types, ASCs and office-based settings. The University of California, San Francisco, institutional review board considered this study exempt because it examined existing data that are publicly available. The Penn State institutional review board considered this study not human subjects research.
The claims database is a commercially available health insurance claims database. It includes claims data for a sample of privately insured people in all 50 US states, including demographic characteristics (ie, age and sex), health care utilization, dates of service, diagnosis codes, procedure codes, and facility type. The data represent claims from clinicians, hospitals, and pharmacies that have been adjudicated for payment and are obtained directly from a convenience sample of large employers and health plans that agree to participate in the database. Although no attempts are made to correct or change information received from the participating employers and health plans, Truven Health has a quality-control process to verify that the data meet criteria for quality and completeness.11 The database only includes data from fully paid and adjudicated claims and the diagnosis and procedure codes are compared with codes in effect when raw data were collected and are edited as necessary.11 This database has been used in other studies examining complications and follow-up care after health care procedures.12,13
The study population included all beneficiaries in this database who were 11 years or older who had an induced abortion between January 1, 2011, and December 31, 2014, in an ASC or office-based setting and who were enrolled in their insurance plan for at least a year prior to the index abortion and at least 6 weeks after the abortion. Current Procedural Terminology (CPT) codes used to identify abortions are listed in eTable 1 in the Supplement. Ectopic pregnancies diagnosed, treated or both within 7 days of the abortion and molar pregnancies were excluded because ectopic pregnancies and molar pregnancies receive different treatments than abortions and have different expected potential morbidities and adverse events.
Facility type was identified based on the standardized place-of-service code variable, which indicates the setting where the service occurred. Settings in the standardized place-of-service code variable are defined by the Centers for Medicare & Medicaid Services and used across health care billing.14 There are 50 different possible settings that include settings such as schools, homeless shelters, inpatient hospital, skilled nursing facilities, and others.15 Facility type was classified as ASC when the standardized place-of-service code variable (stdplac) equaled 24 (“ambulatory surgery center”) and office-based setting when the place-of-service code variable equaled 11 (“office”), which includes most office-based settings.15 These settings included facilities such as abortion clinics, nonspecialized clinics, and physician offices. Abortions performed in other settings were excluded.
The primary outcome was abortion-related morbidities and adverse events occurring within 6 weeks of the abortion. Secondary outcomes, which were both subsets of the primary outcome, were major abortion-related morbidities and adverse events and abortion-related infections.
In accordance with methods published in a recent study of abortion-related events using billing data, abortion-related events were estimated by examining and evaluating all diagnoses and treatments at all health care encounters on the day of and within 6 weeks of the abortion.3 Each index abortion was coded as to whether an abortion-related morbidity or adverse event occurred within the 6 weeks subsequent to the abortion. Events were defined as any abortion-related morbidity or adverse event that received an abortion-related diagnosis or treatment code at any care location, including emergency departments (EDs), the original abortion facility, other health care sites, or pharmacy within 6 weeks of an abortion. Events included those that occurred during, on the day of, or up to 6 weeks after the index abortion. Potential events were identified through an examination of International Classification of Diseases, Ninth Revision (ICD-9) codes in either primary or secondary positions, Health Care Common Procedure Coding System (HCPCS) codes, CPT codes, and medication codes for each health care encounter within 6 weeks of the abortion.
The Procedural Abortion Incident Reporting and Surveillance (PAIRS) Framework,16 which was originally developed for first-trimester aspiration abortions, was used to classify specific events into 1 or more specific diagnoses: retained products of conception, failed abortion, hemorrhage, infection, uterine perforation, anesthesia reaction, symptomatic intrauterine material, postabortal hematometra, cervical injury, disseminated intravascular coagulation, missed ectopic pregnancy, and other or undetermined. Retained placenta was added to the definition of retained products of conception and disseminated intravascular coagulation was added to account for additional types of events that could occur for second-trimester or later abortions. Events were classified as major if they required overnight hospital admission, additional surgery, or blood transfusion. All others were classified as minor.
Identifying abortion-related events involved the following steps. First, each abortion with a subsequent ED visit, a diagnosis code indicating an abortion or miscarriage complication on the day of the index abortion, a subsequent health care encounter with a diagnosis code indicating an abortion or miscarriage complication, or a subsequent inpatient visit was individually coded by a clinically trained reviewer who evaluated all available billing data (ICD-9 and CPT codes, laboratory tests, and medications) for encounters that occurred within 6 weeks after these abortions, including on the day of the abortion. Diagnosis codes for miscarriage complications were included because they seemed unlikely to be separate pregnancies and, instead, were likely billing coding errors because the ICD-9 codes for miscarriage complications and abortion complications only differ in 1 number. The reviewer, blinded to abortion facility type, classified each index abortion with a subsequent ED visit, a complication diagnosis code, or a subsequent inpatient visit as having an abortion-related event or not and then classified each case with an abortion-related event into 1 or more of the 12 possible types of diagnoses. Missed ectopic pregnancies were identified by searching all encounters within 6 weeks and ectopic pregnancies not diagnosed or treated within 7 days after the index abortion were classified as missed ectopic pregnancies.
All encounters within 6 weeks that were not individually coded were searched to identify injection and intravenous (IV) antibiotics commonly used to treat abortion-related infections. All encounters within 6 weeks that were not individually coded were searched to identify repeat procedures (abortion, miscarriage, or dilation and curettage procedures, or additional doses of misoprostol). These repeat procedures were further classified as incomplete, failed, or other or undetermined based on diagnosis codes (eTable 1 in the Supplement) and were coded as events. The injection and IV antibiotics and repeat procedures were added to the individually coded data set.
In addition, the included study cohort as well as the entire raw cohort of abortions (including those that were excluded) were searched for deaths. The entire raw cohort was searched because it was possible that, if someone died within 6 weeks of the abortion, that they might be disenrolled from their insurance and thus would not have met inclusion criteria.
Variables controlled for in the adjusted analyses included: abortion type (first-trimester aspiration abortion performed through 12-14 weeks, first-trimester medication abortion typically provided through 9 weeks at the time abortions in this study were provided,17 and second-trimester and later abortion performed after 12-14 weeks), diabetes, hypertension, age, number of previous-year outpatient health care visits, 1 or more inpatient visits in the previous year (as proxies for underlying health conditions), US census region, and year. To have more complete data on chronic health condition and health care utilization variables, only women insured for at least 1 year before their abortion were included.
When planning the study, we identified 3530 induced abortions in ASCs and 15 444 induced abortions in office-based settings in the 2012 data set. Based on assumptions that there would be a similar number of abortions in each year from 2012 through 2014 and that about half of the abortions would meet eligibility criteria, a sample size calculation for a difference in proportions indicated that there would be sufficient power to detect up to a 0.06 per 100 difference (assuming 0.80 power) between the 2.1 per 100 events expected based on prior published research3 vs a possible 1.5 per 100 in ASCs. Based on the Cohen H effect size descriptions, this would translate to a small effect size.18 Even when controlling for potential confounders in logistic regression, this sample size would have 0.80 power to detect small effects, ie, an odds ratio (OR) of 0.7 to 0.74 for ASCs vs office-based settings.19 Upon extracting the data for 2012 through 2014 and identifying the abortions that met other eligibility criteria, the ratio of abortions in ASCs to office-based settings was closer to 1:7 than the 1:5 originally estimated. Prior to analyzing data, an updated sample size calculation was conducted and indicated that an additional year of data (2011) was needed to have sufficient power to detect a 0.5% difference. The final data set had 0.80 power to detect a 0.5% difference in events in ASCs vs office-based settings, assuming 2.1 per 100 events in office-based settings.
Analysis was conducted in Stata 14.2. In regression models, any abortion-related event was the main outcome and facility type the main exposure variable. The first model specified for each outcome included only facility type and the outcome. Then, adjusted models that simultaneously added all of the potential confounders were estimated. Analysis included generalized estimating equations with exchangeable correlation structure, logit link, binomial distribution, and robust standard errors to account for possible clustering by individuals who had more than one abortion during the study. The QIC command in Stata was used to select the correlation structure.20 In accordance with a priori study plans, these analyses were repeated for major events and for infections (because infections could be an event type for which variation across procedure facility type is possible). Then, according to a priori study plans, subgroup analyses were conducted for any event by abortion type using interaction terms because the rate of abortion-related morbidities and adverse events varies by abortion type3 and some laws regarding ASC requirements apply specifically to second-trimester and later abortions.2 Subgroup analyses were rerun with first-trimester medication abortions as the reference group and then with second-trimester and later abortions as the reference group to get interpretable ORs for these abortion types. Analyses for the secondary outcomes and the subgroup analyses were exploratory.
The postestimation margins command was used to obtain adjusted incidence rates and adjusted differences in incidence rates. As a supplementary analysis, a series of regressions that examined the effect on the main relationship of interest of adding each covariate to the model were also conducted. Additionally, the QIC program20 was used to compare nested models for the primary outcome: those that included abortion type and those that did not. Although it is not possible to determine the extent to which data are missing, the analysis assumes that missing data are rare since these are adjudicated billing claims used to determine payments to clinicians, hospitals, and pharmacies.
Two prespecified sensitivity analyses that used different definitions of abortion-related morbidities and adverse events were conducted. First, there are considerable challenges of measuring whether an ectopic pregnancy was missed based only on billing data. In particular, there is no information available in billing data about whether the clinician suspected an ectopic pregnancy at the time of the abortion visit and the timing of follow-up could be influenced by when test results came back and when the patient was able to present at a facility that provides care for ectopic pregnancies. Because the seven-day cut-off was somewhat arbitrary and could be overly restrictive, the definition of missed ectopic pregnancies was changed to those not diagnosed or treated within 14 days. Second, additional injection or IV antibiotics that are not commonly used to treat abortion-related infections were present in the data set and were added for a sensitivity analysis. A third sensitivity analysis was conducted post hoc. This sensitivity analysis used the Elixhauser Comorbidity Index21 as a control variable instead of the prespecified control variables of diabetes, hypertension, number of previous outpatient visits, and 1 or more previous inpatient visits. This analysis used a binary score of 1 or more of the 30 comorbidities in the Elixhauser Comorbidity Index21,22 and, in a separate analysis, used the Elixhauser Comorbidity Index readmission score.23 Statistical testing was 2-sided and used a P < .05 significance level.
The database included 104 106 induced abortions during the study period. Of those, 53 795 abortions were excluded because the abortions were not provided in the defined venues of an ASC or in an office-based setting, patients had been enrolled in their insurance plan for less than 1 year prior or 6 weeks subsequent to the abortion, or patients were younger than 11 years old. Two hundred sixteen abortions involved a molar pregnancy and 765 an ectopic pregnancy diagnosed or treated within 7 days of the index abortion and were excluded (Figure). Most of the 17 621 abortions in the database that were not performed in an ASC or office-based setting during the study period were provided in inpatient, outpatient, or ED hospital–based settings (n = 16 909). The only other type of facility in which more than 1% of the 17 621 abortions excluded due to facility type were performed was through Federally Qualified Health Centers (n = 297). The study cohort included 49 287 patients who had 50 311 abortions in an ASC or an office-based setting and who were continuously enrolled in their insurance plan for at least 1 year before and at least 6 weeks after the abortion. The date of the final follow-up was February 11, 2015.
Multiple abortions occurred in 798 patients (1822 abortions) during the study period. Of the 1822 multiple abortions, 88% of them occurred in the same facility type for each abortion.
The mean age was 28 years (range, 11-59 years; SD, 7.3 years). Of the 50 311 included abortions, 23 891 (47%) were first-trimester aspiration, 13 480 (27%) first-trimester medication, and 12 940 (26%) second-trimester or later abortions (Table 1). Five thousand six hundred sixty abortions (11%) were performed in ASCs and 44 651 (89%) in office-based settings. The study population differed by facility type: Patients in ASCs were slightly older (28.6 vs 28.1 years; P < .001); more patients in ASCs had had a previous year inpatient encounter (9.8% vs 8.9%; P = .03); fewer medication abortions were in ASCs (2.6% of abortions in ASCs vs 29.9% of abortions in office-based settings; P < .001 for comparison of abortion type by facility type); and abortions in ASCs were more common in the South and Midwest (25.1% of abortions in ASCs and 12.6% of abortions in office-based settings were in the South; 22.6% of abortions in ASCs and 11.5% of abortions in office-based settings were in the Midwest, P < .001 for comparison of region by facility type).
Among the 50 311 abortions, 1674 (3.33%) had an abortion-related event, 163 (0.32%) had a major event, and 374 (0.74%) had an abortion-related infection (Table 2). Of first-trimester abortions, 603 of 23 891 aspiration abortions (2.52%) and 730 of 13 480 medication abortions (5.42%) had an abortion-related event. Of 12 940 abortions performed in the second-trimester or later, 341 (2.64%) had an abortion-related event. At least 0.2% or more of abortions had 1 or more of the following types of events: retained products of conception, infection, other, symptomatic intrauterine material, hemorrhage, and missed ectopic pregnancy. The remaining types of events (failed abortion, disseminated intravascular coagulation, hematometra, uterine perforation, anesthesia-reaction, cervical injury) occurred in fewer than 0.05% of abortions or were not present in the data set (Table 2). In the entire raw cohort (including the excluded abortions), there were 2 deaths. These 2 patients were hospital inpatients and were thus excluded because they did not have their abortions in either an ASC or office-based setting. Additionally, the deaths occurred more than 6 weeks after the abortion (outside of the study period).
In unadjusted analyses examining the primary outcome, abortion-related events were less common in ASCs (2.60%, 95% CI, 2.21-3.05 in ASCs) than in office-based settings (3.42%; 95% CI, 3.26%-3.59% in office-based settings; P = .001). In unadjusted analyses examining secondary outcomes, there were no significant differences in major events with 0.25% (95% CI, 0.15%-0.42%) in ASCs vs 0.33% (95% CI, 0.28%-0.39%) in office-based settings. Similarly, there was not a significant difference in abortion-related infections in ASCs (0.58%, 95% CI, 0.41%-0.82%) vs office-based settings (0.76%, 95% CI, 0.69%-0.85%; Table 2). Unadjusted associations between variables controlled for in adjusted analyses and study outcomes are in eTable 2 in the Supplement.
The QIC test indicated that the exchangeable correlation structure was a better fit for the data than the independent correlation structure (QIC of 307 274.59 for exchangeable vs 307 277.23 for independent). In adjusted analyses, there were not significant differences in abortion-related events (adjusted OR, 0.97; 95% CI, 0.81-1.17; adjusted incidence rate, 3.25% vs 3.33%; adjusted difference, −0.08%; 95% CI, −0.58% to 0.43%) between ASCs and office-based settings (Table 3). Full regression results including 95% CIs for adjusted incidence rates are in eTable 3 in the Supplement. Abortion-type was the only variable controlled for in the adjusted analyses that affected the main association of interest between facility type and events. The model with only facility type had an OR of 0.75 (P=.001); models with all control variables other than abortion type had ORs of 0.74 or 0.75 (P=.001). The model that included abortion type had an OR of 0.97 (P=.77; eTable 4 in the Supplement). The QIC test indicated that the model that included abortion type was a better fit for the data than was the model that did not include abortion type (QIC of 307 274.59 for model with vs 310 114.09 for model without abortion type).
There were also not significant differences in the secondary outcomes of major abortion-related events (adjusted OR, 0.78; 95% CI, 0.45-1.37; adjusted incidence rate, 0.26% vs 0.33%; adjusted difference, −0.06%; 95% CI, −0.18% to 0.06%) or infections by facility type (adjusted OR, 0.75; 95% CI, 0.52-1.09; adjusted incidence rate, 0.58% vs 0.77%; adjusted difference, −0.16%; 95% CI, −0.35% to 0.03%; Table 3).
In unadjusted subgroup analyses by abortion type, there were not significant differences across facility type in abortion-related events among first-trimester aspiration abortions: 78 events (2.15%) of 3630 in ASCs (95% CI, 1.72%–2.68%) vs 525 (2.59%) of 20 261 in office-based settings (95% CI, 2.38%-2.81%). Second-trimester abortions were also similar by setting: 53 events (2.81%) of 1883 in ASCs (95% CI, 2.16%-3.67%) vs 288 (2.60%) of 11 057 in office-based settings (95% CI, 2.32%-2.92%). Among first-trimester medication abortions, the rate of abortion-related adverse events was significantly greater at ASCs, with events occurring in 10.88% (16 of 147) of abortions performed in ASCs (95% CI, 6.74%-17.12%) vs 5.36% (714 of 13 333) of abortions performed in office-based settings (95% CI, 4.99%-5.75%; P = .003; Table 2).
In adjusted subgroup analyses, there were not significant differences between ASCs and office-based settings in abortion-related events for first-trimester aspiration abortion (adjusted OR, 0.84; 95% CI, 0.66-1.07; adjusted incidence rate, 2.19% vs 2.59%; adjusted difference, −0.38%; 95% CI, −0.88% to 0.12%). The interaction term for second-trimester and later abortion × facility type was not significant, indicating that there also was no statistical difference in events in ASCs vs office-based settings for second-trimester and later abortions, (adjusted OR from model with second trimester and later as reference group, 1.01; 95% CI, 0.75-1.37; adjusted incidence rate, 2.62% vs 2.59%; adjusted difference, 0.03%; 95% CI, −0.70% to 0.76%). There were significant differences in odds of abortion-related events in ASCs vs office-based settings for medication abortions, P value for interaction term was .001 (adjusted OR from model with first-trimester medication abortion as reference group; 2.23; 95% CI, 1.30-3.80; adjusted incidence rate, 11.22% vs 5.42%; adjusted difference, 5.54%; 95% CI, 5.12%-10.56%; Table 3).
There were no substantive differences in the sensitivity analyses using different definitions of abortion-related morbidities and adverse events and adjusting for comorbidities using the Elixhauser Comorbidity Index (eTables 5 through 9 in the Supplement).
In this retrospective analysis of more than 50 000 induced abortions in the United States between 2011 and 2014, performance of abortions in ASCs compared with office-based settings was not associated with a significant difference in abortion-related morbidities and adverse events. The lack of a significant association between abortions performed in ASCs vs office-based settings and the rates of abortion-related morbidities and adverse events is consistent with the small body of literature that compares the safety of other outpatient procedures across ASCs and office-based settings.24
This study reinforced that there are low rates of abortion-related morbidities and adverse events after abortion, with major events occurring in only one-third of one percent of cases. This study also confirmed that there are low rates of abortion-related morbidities and adverse events in both ASCs and office-based settings. Although the estimate of the overall abortion-related event rate was higher than a previous estimate using claims data, the estimate of major events was similar.3 The estimates for events after first-trimester aspiration abortions were higher than previous estimates using claims data but were within the range of estimates in other published studies,3,4 as were the estimates for second-trimester and later abortions.25-27 The estimates for events after medication abortions are within the range of previous published estimates of events after medication abortions using claims data.3
The study was powered to detect a 0.5% difference in any abortion-related morbidities and adverse events. This difference is smaller than the predetermined acceptable risk difference of 2% used in a large noninferiority study comparing events after abortions performed by advance practice clinicians and physicians.28 In that study, the acceptable risk difference was determined before the start of the study by a panel of researchers and clinicians.28 The upper bound of the 95% CI around the 0.1% observed difference in this study was 0.6%, which is within the 2% specified as clinically insignificant in that previous study. The upper bound of the 95% CIs for observed differences for first-trimester aspiration and second-trimester and later abortions were less than 1%, which is within the 2% threshold for any events. The upper bound of the 95% CI for the 0.1% observed difference in major events was 0.2%, which is slightly smaller than the 0.3% difference in major events defined as clinically important in a recent study.29 The upper bound of the 95% CI for the 0.2% observed difference in infections was 0.4%, which is also not clinically important.
Abortion-related morbidities and adverse events appear more common among women having first-trimester medication abortions in ASCs than in office-based settings. The upper bound of the difference (11%) may be clinically important. Because medication abortions in ASCs were rare (only 2.6% of abortions in ASCs and 0.3% of all abortions in the sample), the significant association between facility type and abortion-related morbidities and adverse events for first-trimester medication abortions should be interpreted with caution. It is possible, given that medication abortions are rarely performed in ASCs, that women who have medication abortions in ASCs may travel a long distance and may receive follow-up care at sites that may be more likely to provide an additional treatment.
Observational studies include the risk of unbalanced study groups, for which differences are not adequately controlled for in analyses. Although there were some significant differences in the measured covariates across ASCs and office-based settings, the patient-level differences identified in this study were small (ie, half a year of age, <1% difference in proportion with ≥1 inpatient encounters in the previous year). Only 1 of the observed covariates—abortion type—confounded the relationship between facility type and incidents (eTable 4 in the Supplement). This confounding was due to first-trimester medication abortions being much more common in office-based settings and, similar to previous research,3 having more events (>5%) than first-trimester aspiration and second -trimester and later abortions, which had closer to 2.5%.
This study has several strengths. First, the study used a national sample of claims data from a database that is often used to examine safety of health care procedures.12,13 Using an insurance claims database that includes a national sample allows direct comparison of abortion-related morbidities and adverse events across ASCs vs office-based settings and provides a sufficiently large sample to detect differences, avoid biases associated with focusing on a few facilities where practice could be unrepresentative, and control for potential confounders. Second, there was little loss to follow-up because of the use of claims data. Claims databases are useful for examining morbidities and adverse events after health care procedures (including abortion) because they routinely capture health care visits and treatments that occur subsequent to the procedure,3 thereby increasing chances that most postprocedure events will be captured in the data set and limiting biases from loss to follow-up that have been noted in other studies examining abortion-related morbidities and adverse events.28 Third, there was a large sample of second-trimester and later abortions that allowed assessment of whether overall findings held among this subset of abortions.
This study has several limitations. First, although the study differentiated between first- and second-trimester abortions, it was not possible to know the weeks’ gestation at which the abortion was provided, a potential limitation given that the risk of abortion-related events increases with each week of gestation.25,26 If there were differences in timing of abortions within the first and second trimesters across office-based settings and ASCs, this could be an unmeasured confounder. Second, it is not known whether the ectopic pregnancies classified as missed were suspected at the index abortion visit. The approach of including all ectopic pregnancies diagnosed, treated, or both after 7 days as missed was conservative. Third, by virtue of the data included in the database used in this study, the study only included abortions paid for by private insurance. Only about 15% of the almost 1 million abortions provided each year in the United States are paid for by private insurance.5,30 Thus, findings may not be generalizable to all abortions in the United States. Fourth, other potential confounders, including body mass index, race, and previous cesarean section, were not available in the database and thus could not be controlled. However, recent research has not indicated associations between obesity and abortion-related events, so this should partially address this concern.25 Other indicators of health status, such as frequency of health care visits, were controlled for. Because no anesthesia-related reactions were identified, it does not appear that not being able to control for anesthesia has biased the results. Fifth, the database did not include information on the specific type of facility (ie, abortion clinics, nonspecialized clinics, and physician offices) in which the abortions were performed. This study thus was unable to assess whether abortion-related morbidities and adverse events vary by whether the abortion was performed in an abortion clinic, nonspecialized clinic, or a physician office. Sixth, the database used in the study only includes inpatient deaths. If someone died without being admitted to a hospital, this would not be captured. The fact that there were no deaths identified among the study cohort is not surprising, because the mortality rate from abortion in the United States is less than 1 death per 100 000 abortions.31
Among women with private health insurance who had an induced abortion, performance of the abortion in an ambulatory surgery center compared with an office-based setting was not associated with a significant difference in abortion-related morbidities and adverse events. These findings, in addition to individual patient and individual facility factors, may inform decisions about the type of facility in which induced abortions are performed.
Corresponding Author: Sarah C. M. Roberts, DrPH, Advancing New Standards in Reproductive Health, University of California, San Francisco, 1330 Broadway, Ste 1100, Oakland, CA 94612 (firstname.lastname@example.org).
Accepted for Publication: May 17, 2018.
Correction: This Results section of this article was corrected on July 24, 2018, to change the difference in “the rates of abortion-related morbidities or adverse events” to “−0.08.”
Author Contributions: Drs Roberts and Liu had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Concept and design: Roberts, Upadhyay, Liu, Leslie.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Roberts, Liu.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Roberts, Liu, Ba.
Obtained funding: Roberts.
Administrative, technical, or material support: Liu, Kerns, Beam, Leslie.
Supervision: Roberts, Upadhyay, Liu, Kerns, Leslie.
Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest and none were reported.
Funding/Support: This study was supported by a grant from the Society of Family Planning Research Fund, SFPRF10-10.
Role of the Funder/Sponsor: The Society of Family Planning Research Fund had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; or decision to submit the manuscript for publication.
Additional Contributions: We thank Bonnie Scott Jones, JD, for help understanding ASC laws, and Sara Daniel, MPH, and Beckie Kriz, RN, MS, for database preparation. The efforts of Mss Jones, Daniel, and Kriz also were supported by SFPRF10-10. All 3 worked at Advancing New Standards in Reproductive Health in the Department of Obstetrics, Gynecology, and Reproductive Sciences at the University of California, San Francisco, during the study.