Advertisement

Design of the Association of Uterine Perforation and Expulsion of Intrauterine Device study: a multisite retrospective cohort study

Open AccessPublished:January 15, 2021DOI:https://doi.org/10.1016/j.ajog.2021.01.003

      Background

      Intrauterine devices are effective and safe, long-acting reversible contraceptives, but the risk of uterine perforation occurs with an estimated incidence of 1 to 2 per 1000 insertions. The European Active Surveillance Study for Intrauterine Devices, a European prospective observational study that enrolled 61,448 participants (2006–2012), found that women breastfeeding at the time of device insertion or with the device inserted at ≤36 weeks after delivery had a higher risk of uterine perforation. The Association of Uterine Perforation and Expulsion of Intrauterine Device (APEX-IUD) study was a Food and Drug Administration–mandated study designed to reflect current United States clinical practice. The aims of the APEX-IUD study were to evaluate the risk of intrauterine device–related uterine perforation and device expulsion among women who were breastfeeding or within 12 months after delivery at insertion.

      Objective

      We aimed to describe the APEX-IUD study design, methodology, and analytical plan and present population characteristics, size of risk factor groups, and duration of follow-up.

      Study Design

      APEX-IUD study was a retrospective cohort study conducted in 4 organizations with access to electronic health records: Kaiser Permanente Northern California, Kaiser Permanente Southern California, Kaiser Permanente Washington, and Regenstrief Institute in Indiana. Variables were identified through structured data (eg, diagnostic, procedural, medication codes) and unstructured data (eg, clinical notes) via natural language processing. Outcomes include uterine perforation and device expulsion; potential risk factors were breastfeeding at insertion, postpartum timing of insertion, device type, and menorrhagia diagnosis in the year before insertion. Covariates include demographic characteristics, clinical characteristics, and procedure-related variables, such as difficult insertion. The first potential date of inclusion for eligible women varies by research site (from January 1, 2001 to January 1, 2010). Follow-up begins at insertion and ends at first occurrence of an outcome of interest, a censoring event (device removal or reinsertion, pregnancy, hysterectomy, sterilization, device expiration, death, disenrollment, last clinical encounter), or end of the study period (June 30, 2018). Comparisons of levels of exposure variables were made using Cox regression models with confounding adjusted by propensity score weighting using overlap weights.

      Results

      The study population includes 326,658 women with at least 1 device insertion during the study period (Kaiser Permanente Northern California, 161,442; Kaiser Permanente Southern California, 123,214; Kaiser Permanente Washington, 20,526; Regenstrief Institute, 21,476). The median duration of continuous enrollment was 90 (site medians 74–177) months. The mean age was 32 years, and the population was racially and ethnically diverse across the 4 sites. The mean body mass index was 28.5 kg/m2, and of the women included in the study, 10.0% had menorrhagia ≤12 months before insertion, 5.3% had uterine fibroids, and 10% were recent smokers; furthermore, among these women, 79.4% had levonorgestrel-releasing devices, and 19.5% had copper devices. Across sites, 97,824 women had an intrauterine device insertion at ≤52 weeks after delivery, of which 94,817 women (97%) had breastfeeding status at insertion determined; in addition, 228,834 women had intrauterine device insertion at >52 weeks after delivery or no evidence of a delivery in their health record.

      Conclusion

      Combining retrospective data from multiple sites allowed for a large and diverse study population. Collaboration with clinicians in the study design and validation of outcomes ensured that the APEX-IUD study results reflect current United States clinical practice. Results from this study will provide valuable information based on real-world evidence about risk factors for intrauterine devices perforation and expulsion for clinicians.

      Key words

      Introduction

      Although intrauterine devices (IUDs) are highly effective long-acting reversible contraceptives,
      Practice Bulletin no. 186 summary: long-acting reversible contraception: implants and intrauterine devices.
      their use carries a risk of uterine perforation, with estimated incidences of 1 to 2 perforations per 1000 insertions.
      • Barnett C.
      • Moehner S.
      • Do Minh T.
      • Heinemann K.
      Perforation risk and intra-uterine devices: results of the EURAS-IUD 5-year extension study.
      ,
      • O’Brien P.A.
      • Pillai S.
      Uterine perforation by intrauterine devices: a 16-year review.
      These rates were reported in the prospective observational European Active Surveillance Study for Intrauterine Devices (EURAS-IUD), designed to evaluate the risk of uterine perforation among users of levonorgestrel (LNG)-releasing and copper IUDs used in a routine clinical setting, conducted in 6 European countries, with recruitment from 2006 to 2012.
      • Heinemann K.
      • Reed S.
      • Moehner S.
      • Minh T.D.
      Risk of uterine perforation with levonorgestrel-releasing and copper intrauterine devices in the European Active Surveillance Study on Intrauterine Devices.
      Here, 2 cohorts were included: new users of LNG-releasing IUDs (n=43,078) or copper IUDs (n=18,370), with follow-up after 12 months. The study did not identify a relevant difference in perforation rate by IUD type; however, this study suggested that breastfeeding at the time of IUD insertion was associated with a 6-fold increase in the relative risk of uterine perforation. Furthermore, there was an increased risk of uterine perforation among those with IUD insertions within 36 weeks after the most recent delivery.

      Why was this study conducted?

      The European Active Surveillance Study for Intrauterine Devices showed an increased risk of uterine perforation among women who were breastfeeding or ≤36 weeks after delivery at intrauterine device (IUD) insertion. The US Food and Drug Administration suspected differences in breastfeeding practices and postpartum timing of IUD placement and mandated a study to assess risks among women in the United States.

      Key findings

      Retrospective, real-world data can be used to estimate risks with more timely results than are achieved with prospective studies. Pooling data across multiple sites can provide demographic variation and confirm results across sites.

      What does this add to what is known?

      APEX-IUD study design and patient characteristics are presented. Final study results will provide valuable information for clinical practice based on real-world evidence about the risk of perforation and expulsion.
      Consistent with clinical practice in the European countries participating in the study, EURAS-IUD provided little data on IUD placement immediately after delivery, a more accepted practice in the United States.
      Practice Bulletin no. 186 summary: long-acting reversible contraception: implants and intrauterine devices.
      Given the results of EURAS-IUD,
      • Heinemann K.
      • Reed S.
      • Moehner S.
      • Minh T.D.
      Risk of uterine perforation with levonorgestrel-releasing and copper intrauterine devices in the European Active Surveillance Study on Intrauterine Devices.
      the US Food and Drug Administration (FDA) mandated a study to evaluate uterine perforation risks associated with US clinical practices for IUD insertion. Although the FDA initially recommended a prospective observational study, the research team instead suggested that a retrospective study drawing data from electronic health records (EHRs) would provide greater efficiency and more timely results while reflecting current US clinical practice. Following a validation study demonstrating that algorithms could be used to identify uterine perforation and IUD expulsion in proposed data sources and that breastfeeding could be identified in EHRs of women who had given birth,
      • Anthony M.S.
      • Armstrong M.A.
      • Getahun D.
      • et al.
      Identification and validation of uterine perforation, intrauterine device expulsion, and breastfeeding in four health care systems with electronic health records.
      the Association of Uterine Perforation and Expulsion of IUD (APEX-IUD) study was planned in coordination with Bayer and the FDA.
      The APEX-IUD study was a multisite, retrospective cohort study that used data from EHRs and a health information exchange to assess the outcomes of uterine perforation and IUD expulsion in association with potential risk factors, including breastfeeding, postpartum timing of IUD insertion, recent history of menorrhagia, and IUD type in the setting of usual healthcare.
      European Network of Centres for Pharmacoepidemiology and Pharmacovigilance (ENCePP)
      Study on the association of uterine perforation and intrauterine device (IUD) expulsion with breastfeeding status at the time of IUD insertion and postpartum timing of IUD insertion in electronic medical record databases. A postmarketing requirement for Mirena.
      This manuscript describes the study design, methodology, and characteristics of the study population. In addition, comparisons with the methods of the prospective cohort study, EURAS-IUD, are discussed. The study results of the APEX-IUD study will be described separately.

      Materials and Methods

      Study population

      The study population included all women aged ≤50 years with evidence of an IUD insertion and with ≥12 months of enrollment history preceding IUD insertion (for Kaiser Permanente [KP] sites) or a clinical visit ≥12 months before insertion (for Regenstrief Institute [RI]). Here, the first potential date for a woman’s inclusion varied by research site based in part on when EHR data became available at the beginning of the year at each site, 2001 at the RI, 2007 at KP Washington (KPWA), 2009 at KP Southern California (KPSC), and 2010 at KP Northern California (KPNC); and the last date for inclusion at all sites was on April 30, 2018. All IUD insertions from eligible women have been included.

      Study design

      The APEX-IUD study was a retrospective cohort study. Among women with an IUD insertion identified within EHR data, this study has evaluated the association of 2 primary potential risk factors, breastfeeding status at the time of IUD insertion and timing of IUD insertion during the postpartum period, and the outcomes of IUD-related uterine perforation and IUD expulsion in the usual healthcare setting (ie, the data were abstracted from patient care records rather than data collected for a clinical trial). In addition, the association of 2 secondary potential risk factors, recent history of menorrhagia and IUD type, and said outcomes were evaluated. Women were observed from IUD insertion date (index date) to the earliest date of IUD-related uterine perforation, IUD expulsion, IUD removal, IUD reinsertion, pregnancy, hysterectomy or other sterilization procedures, IUD expiration, disenrollment from the healthcare system (KP sites), last clinical encounter (RI), end of the study period (June 30, 2018), or death. All person-time at risk data were included with no requirement for minimum or maximum follow-up time.
      All participating research sites received an approval or a waiver to conduct this study from their respective institutional review boards. In addition, KPSC received approval from the California Health and Human Services Agency and the California Department of Public Health Center for Health Statistics and Informatics (state birth and death files). Each site submitted data for analysis in a deidentified, standard format in accordance with the data structure template for variables (Table 1).
      Table 1The Association of Uterine Perforation and Expulsion of IUD study variables
      VariableDefinitionCategoriesFormat
      Potential risk factor variables
      Postpartum time of IUD insertionTime since most recent delivery date ≤index date
      Data source: calculated
      • 0–≤3 d
      • >3 d–≤6 wk
      • >6–≤14 wk
      • >14–≤52 wk
      • >52 wk or with no evidence of delivery
      4 categories; 5 categories; binary
      Breastfeeding statusBreastfeeding status at index date
      Data source: structured questionnaires
      ,
      Data source: codes
      ,
      Data source: natural language programming
      • No
      • Yes
      • Undetermined
        Those with no evidence of a delivery in the 52 weeks before index date were excluded from the breastfeeding analysis
      3 categories
      IUD typeIUD type at the time of the index date
      Data source: codes
      ,
      Data source: natural language programming
      • LNG
      • Copper
      • Unknown
      3 categories
      MenorrhagiaMenorrhagia before or on index date
      Data source: codes
      • Yes, if diagnosis of menorrhagia within 12 mo before the index date
      • No, if no menorrhagia diagnosis within the 12 mo before index date
      Binary
      Outcome variables
      Uterine perforationUterine perforation event during person-time at risk
      Data source: codes
      ,
      Data source: natural language programming
      • Yes (partial, complete)
      • No
      Binary
      IUD expulsionIUD expulsion event during person-time at risk
      Data source: codes
      ,
      Data source: natural language programming
      • Yes (partial, complete)
      • No
      Binary
      Covariates
      Demographic
      AgeAge in years at IUD insertion
      Data source: other sources.
      • ≤28 y
      • >28–≤36 y
      • >36–≤50 y
      Integer; 3 categories
      Race and ethnicityRace and ethnicity
      Data source: other sources.
      • Asian or Pacific Islander
      • Hispanic Black
      • Hispanic other
      • Hispanic White
      • Non-Hispanic Black
      • Non-Hispanic White
      • Other or multiple
      • Unknown
      8 categories
      SmokingSmoked within 365 d of index date
      Data source: structured questionnaires
      ,
      Data source: codes
      ,
      Data source: natural language programming
      • No
      • Yes
      • Missing or unknown
      3 categories
      Calendar year of IUD insertionCalendar year of IUD insertion
      Data source: codes
      ,
      Data source: natural language programming
      • 2001–2009
      • 2010
      • 2011
      • 2012
      • 2013
      • 2014
      • 2015
      • 2016
      • 2017
      • 2018
      Categorical
      Duration of lookback periodDuration of lookback period
      Data source: calculated
      • ≥1–≤2 y
      • >2–≤4 y
      • >4–≤6.5 y
      • >6.5 y
      • Dichotomized:
      • ≥1–≤4 y
      • >4 y
      Continuous; 4 categories; 2 categories
      Clinical
      BMIBMI (kg/m2) at index date or closest date
      Data source: codes
      ,
      Data source: natural language programming
      • Underweight (<18.5 kg/m2)
      • Normal weight (18.5–24.9 kg/m2)
      • Overweight (25.0–29.9 kg/m2)
      • Obesity (≥30.0 kg/m2)
      • Missing
      Continuous; 5 categories
      DysmenorrheaDysmenorrhea before or on index date
      Data source: codes
      • Diagnosis of dysmenorrhea within the 12 mo before the index date but not diagnosed before that time
      • No diagnosis of dysmenorrhea within the 12 mo before the index date but a diagnosis of dysmenorrhea before that time
      • Diagnosis of dysmenorrhea recorded both within the 12 mo before the index date and before that time
      • No diagnosis of dysmenorrhea found before or on the index date
      4 categories
      MenorrhagiaMenorrhagia before or on index date
      Data source: codes
      • Diagnosis of menorrhagia within the 12 mo before the index date but not diagnosed before that time
      • No diagnosis of menorrhagia within the 12 mo before the index date but a diagnosis of menorrhagia before that time
      • Diagnosis of menorrhagia recorded both within the 12 mo before the index date and before that time
      • No diagnosis of menorrhagia found before or on the index date
      4 categories
      Uterine fibroidsFibroids before or on index date
      Data source: codes
      • No
      • Yes
      Binary
      ParityCumulative number of viable pregnancies (ie, carried to at least 20 wk gestation) before the index date
      Data source: codes
      ,
      Data source: natural language programming
      ,
      Data source: other sources.
      • Dichotomized for analyses of IUD type and menorrhagia:
      • 0
      • >0
      • Parity dichotomized for analyses of breastfeeding and postpartum period:
      • ≤1
      • >1
      • Categorical:
      • 0
      • 1
      • 2
      • ≥3
      • Missing
      Integer; 2 categories; 2 categories; 5 categories
      Cesarean deliveryAny cesarean delivery during the lookback period (summarized only if a woman had one or more parity)
      Data source: codes
      • No
      • Yes
      Binary
      Cesarean delivery for most recent deliveryCesarean delivery for most recent delivery (summarized only if there was a recorded delivery within 52 wk before or on the index date)
      Data source: codes
      • No
      • Yes
      Binary
      Concomitant gynecologic procedureAny of the following procedures at insertion
      Data source: calculated
      • No
      • Yes
      Binary
       Abortion procedureAbortion
      Data source: codes
      • No
      • Yes
      Binary
       Aspiration and curettageAspiration and curettage
      Data source: codes
      • No
      • Yes
      Binary
       Dilation and curettageDilation and curettage
      Data source: codes
      • No
      • Yes
      Binary
       Biopsy of the cervix or uterusBiopsy
      Data source: codes
      • No
      • Yes
      Binary
       AblationAblation
      Data source: codes
      • No
      • Yes
      Binary
       Colposcopy and other cervical proceduresColposcopy and cervical procedures
      Data source: codes
      • No
      • Yes
      Binary
       Hysteroscopy procedureHysteroscopy at insertion
      Data source: codes
      • No
      • Yes
      Binary
       Laminaria procedureLaminaria
      Data source: codes
      • No
      • Yes
      Binary
       LaparoscopyLaparoscopy
      Data source: codes
      • No
      • Yes
      Binary
       Lysis adhesionsLysis adhesions
      Data source: codes
      • No
      • Yes
      Binary
       MyomectomyMyomectomy
      Data source: codes
      • No
      • Yes
      Binary
       Nerve procedureNerve procedure
      Data source: codes
      • No
      • Yes
      Binary
       Salpingectomy or oophorectomySalpingectomy or oophorectomy
      Data source: codes
      • No
      • Yes
      Binary
      Indicator of a difficult IUD insertionAny of the following indicators at insertion
      Data source: calculated
      • No
      • Yes
      Binary
       DilationCervical dilation
      Data source: codes
      ,
      Data source: natural language programming
      • No
      • Yes
      Binary
       Ultrasound guidanceUltrasound guidance
      Data source: codes
      ,
      Data source: natural language programming
      • No
      • Yes
      Binary
       Paracervical blockParacervical block
      Data source: codes
      ,
      Data source: natural language programming
      • No
      • Yes
      Binary
       Provider noted difficult insertionProvider noted difficult insertion
      Data source: natural language programming
      • No
      • Yes
      Binary
       MisoprostolMisoprostol before insertion
      Data source: codes
      ,
      Data source: natural language programming
      • No
      • Yes
      Binary
      Clinician-related characteristics (available in KPNC, KPSC, and KPWA but not in RI)
      Number of IUD insertions performedNumber of IUD insertions performed
      Data source: other sources.
      • <50 insertions
      • ≥50 insertions
      • Missing
      Continuous; 3 categories
      Annualized number of insertions in previous yearAnnualized number of insertions in previous year
      Data source: calculated
      • 4 categories based on quartiles
      • 2 categories based on median
      • Missing
      Continuous; 5 categories; 3 categories
      Length of employment in previous yearLength of employment in previous year, in days
      Data source: other sources.
      Continuous and missing
      Other
      Live birthLive birth for the most recent delivery within the past 52 wk of the index date (summarized only among women who had a delivery within 1 y before the index date)
      Data source: codes
      ,
      Data source: other sources.
      • No
      • Yes
      Binary
      SiteStudy site
      Data source: other sources.
      • KPNC
      • KPSC
      • KPWA
      • RI
      4 categories
      BMI, body mass index; IUD, intrauterine device; KPNC, Kaiser Permanente Northern California; KPSC, Kaiser Permanente Southern California; KPWA, Kaiser Permanente Washington; LNG, levonorgestrel; RI, Regenstrief Institute.
      Anthony et al. Association of uterine perforation and expulsion of intrauterine device study methods. Am J Obstet Gynecol 2021.
      a Data source: calculated
      b Data source: structured questionnaires
      c Data source: codes
      d Data source: natural language programming
      e Those with no evidence of a delivery in the 52 weeks before index date were excluded from the breastfeeding analysis
      f Data source: other sources.

      Data sources

      Data were obtained from 3 integrated healthcare systems with EHRs (KPNC, KPSC, and KPWA) and the RI, an organization with research access to a health information exchange with access to EHRs. These study sites were included on the basis of their ability to access population-based EHR data, data quality, and variation in demographics. Appendix A provides additional detail.
      Potential risk factors, covariates, and outcomes were identified from the EHRs and included both structured data (eg, International Classification of Diseases, Ninth Revision, or International Classification of Diseases, Tenth Revision, Clinical Modification [ICD-9 or ICD-10-CM] diagnosis and procedure codes, medication codes, Current Procedural Terminology codes, and Healthcare Common Procedural Coding System codes) and unstructured data (eg, clinical notes). Algorithms were developed to identify potential risk factors and outcomes using operational definitions, natural language processing (NLP), and medical record review at all sites.
      • Anthony M.S.
      • Armstrong M.A.
      • Getahun D.
      • et al.
      Identification and validation of uterine perforation, intrauterine device expulsion, and breastfeeding in four health care systems with electronic health records.
      These algorithms were developed collaboratively to capture the same concepts but implemented separately at each site and differed where appropriate; for instance, some ICD codes performed better at some sites than others owing to the variation in local coding practices. Algorithms for the outcome variables, uterine perforation and IUD expulsion, were validated by obstetrician-gynecologist clinicians from each site via chart review while ICD-9-CM codes were in use, prior to use of ICD-10-CM coding. The proportions of women at risk who had 1 of these outcomes were calculated and compared before and after the implementation of ICD-10-CM coding to assess temporal consistency over coding terms (Appendix B).

      Potential risk factors

      Two primary risk factors were evaluated: time of IUD insertion after delivery and breastfeeding status at time of IUD insertion. Postpartum time of IUD insertion, originally categorized into 4 discrete periods in agreement with FDA, was later expanded to 5 periods, such that the ≤6-week period (which was strongly bimodal) was further divided: (1) 0 to ≤3 days, (2) >3 days to ≤6 weeks, (3) >6 to ≤14 weeks, (4) >14 to ≤52 weeks, and (5) >52 weeks or with no evidence of delivery. Breastfeeding status at the time of IUD insertion was classified as “yes” (last breastfeeding date within 30 days before IUD insertion or any time after IUD insertion to 52 weeks after delivery), “no” (last breastfeeding date >30 days before IUD insertion or first nonbreastfeeding date before IUD insertion), or “undetermined.” Women with undetermined breastfeeding status were excluded from the breastfeeding cohort analyses. Other key potential risk factors included recent menorrhagia, defined as a menorrhagia code in the 12 months before IUD insertion, and IUD type (LNG-IUDs, including 52 mg, 19.5 mg, and 13.5 mg reservoir devices, and copper IUDs).

      Outcomes

      Two key outcomes were assessed: (1) uterine perforation, defined as complete perforation (with clinical evidence of IUD in the pelvis or abdominal cavity) or partial perforation (ie, IUD removed after being visualized as partially embedded in the myometrium on imaging or hysteroscopy or partial perforation noted by clinician at time of removal), and (2) IUD expulsion, defined as complete expulsion (IUD located in the vagina, not present in the uterus or abdomen on imaging, or patient reported IUD fell out) or partial expulsion (any portion of IUD in the cervix on imaging, documented visualization by a clinician, or IUD considered malpositioned on imaging and removed by the clinician). At the RI, KPNC, and KPWA, potential uterine perforations identified by structured and unstructured data were confirmed using medical record review.

      Covariates

      Demographic and clinical characteristics were assessed before the index date for each eligible IUD insertion. All available data, but a minimum of 12 months before the index date, were used to evaluate women’s characteristics and potential confounders, thus reducing the misclassification of demographic and clinical characteristics.
      • Brunelli S.M.
      • Gagne J.J.
      • Huybrechts K.F.
      • et al.
      Estimation using all available covariate information versus a fixed look-back window for dichotomous covariates.
      Demographic covariates included age, self-reported race and ethnicity and smoking status, month and year of the index date, and duration of the lookback period; clinical covariates included body mass index (BMI; kg/m2), dysmenorrhea, uterine fibroids, parity, cesarean delivery (for women with a delivery before the index date), and indicators of a difficult insertion (Table 1). Other procedure-related characteristics (eg, concomitant gynecologic procedure) and clinician-related covariates (eg, number of IUD insertions in the previous year) were also collected (Table 1).

      Study variables

      For the outcome evaluations, person-time at risk was calculated from the IUD insertion date to the first occurrence of an outcome or censoring date, at which point follow-up ends for that outcome. The censoring date was the earliest of the following dates: IUD removal or reinsertion, start of new pregnancy, hysterectomy, bilateral oophorectomy and other types of sterilization, expiration of IUD, disenrollment from the healthcare system (KP sites), last clinical encounter in the healthcare system (RI), death, and end of the study period (June 30, 2018).

      Study size

      Power calculations for uterine perforation using the expected number of IUD insertions and risk factor allocation based on the validation study
      • Anthony M.S.
      • Armstrong M.A.
      • Getahun D.
      • et al.
      Identification and validation of uterine perforation, intrauterine device expulsion, and breastfeeding in four health care systems with electronic health records.
      and EURAS-IUD
      • Heinemann K.
      • Reed S.
      • Moehner S.
      • Minh T.D.
      Risk of uterine perforation with levonorgestrel-releasing and copper intrauterine devices in the European Active Surveillance Study on Intrauterine Devices.
      were performed using PASS 14 software (NCSS Statistical Software, Kaysville, UT) for a 2-sided test of the hazard ratio (HR).
      • Schoenfeld D.A.
      Sample-size formula for the proportional-hazards regression model.
      Because IUD expulsion rates are higher than uterine perforation rates, the power for IUD expulsion comparisons is greater than for perforation.

      Statistical analysis

      Descriptive analyses for all variables of interest were conducted overall and stratified by research site. For categorical variables, frequencies and percentages were calculated. For continuous variables, the mean, standard deviation (SD), minimum, maximum, median, and quartiles were computed. Missing data were treated as missing, and no imputation was performed. All analyses were performed using Statistical Analysis System (SAS) software, (version 9.3 or higher; SAS Institute, Inc, Cary, NC).
      Crude incidence rates and cumulative incidence of the outcomes will be estimated for each risk factor group (eg, breastfeeding status and postpartum IUD insertion timing). Crude HRs will be calculated for each outcome for each site. Confounding in the multivariable models will be controlled through propensity scores based on the values of covariates at the time of IUD insertion. Separate propensity score models will be developed using logistic regression for each pairing of a potential risk factor with an outcome. Covariates will be assessed for inclusion in propensity score models based on association with the study outcome if the crude HR is greater than 1.11 or less than 0.90 and not outcome blinded. Additional confounders will be selected for inclusion if at least a 10% change in the HR of the risk factor - outcome relationship occurs when adjusting for that variable. From the fitted logistic regression models, propensity scores will be estimated for each IUD insertion.
      Propensity scores will be used to calculate weights for each IUD insertion within each risk factor group. The overlap weighting method
      • Li F.
      • Morgan K.L.
      • Zaslavsky A.M.
      Balancing covariates via propensity score weighting.
      is used which has an advantage of not requiring trimming of observations; furthermore, observations with significant overlap between groups will be up-weighted, and observations with very little overlap will be down-weighted, compared with regular inverse probability treatment weighting. To assess whether covariates are balanced across risk factor groups after weighting, the distribution of each variable will be compared between categories of the risk factor variable, and balance parameters (ie, standardized differences) will be calculated.
      • Austin P.C.
      • Stuart E.A.
      Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies.
      Appendix C provides additional detail.

      Results

      The results presented include a description of the study design, methodology, and characteristics of the study population. The final study results will be published separately.

      Participants

      The study included 326,658 women with at least 1 IUD insertion identified during the study period. The number of women included from each study site differed (KPNC, 161,442; KPSC, 123,214; KPWA, 20,526; and RI, 21,476). Approximately 19% of women had multiple insertions; the information reported here only reflects the first IUD insertion.

      Size of risk factor groups

      Nearly 95,000 women across all sites had an IUD insertion within 52 weeks after delivery and had information in their EHR sufficient to determine breastfeeding status at IUD insertion. Twice as many of these women were determined to be breastfeeding at the time of IUD insertion (n=64,186) compared with women not breastfeeding (n=30,631) (Table 2; Figure). Most women (228,834 of 326,658) had an IUD placement more than 52 weeks after delivery or had no evidence of a delivery in their EHR. At the time of the first identified IUD insertion, 10% of women (range across sites, 5.6%–12.8%) had a menorrhagia diagnosis within 12 months before IUD insertion. Approximately 79% of the first identified IUD insertions were LNG-releasing IUDs; 20% were copper IUDs; and IUD type was not available (approximately 1%).
      Table 2Size of the risk factor groups, pooled and by research site
      VariablePooled (N=326,658)KPNC (n=161,442)KPSC (n=123,214)KPWA (n=20,526)RI (n=21,476)
      Person-years at risk641,427325,552241,92337,49636,456
      Characteristic
      Breastfeeding status (patients at ≤52 wk postpartum)
       Yes64,186 (65.6)34,357 (74.8)23,679 (57.7)3964 (68.1)2186 (43.8)
       No30,631 (31.3)10,996 (23.9)17,027 (41.5)875 (15.0)1733 (34.7)
       Undetermined (excluded)3007 (3.1)578 (1.3)363 (0.9)986 (16.9)1080 (21.6)
      Postpartum time of IUD insertion
       0–≤3 d2788 (0.9)2001 (1.2)106 (0.1)27 (0.1)654 (3.0)
       >3 d–≤6 wk17,272 (5.3)10,615 (6.6)4818 (3.9)747 (3.6)1092 (5.1)
       >6–≤14 wk56,047 (17.2)24,259 (15.0)25,880 (21.0)3682 (17.9)2226 (10.4)
       >14–≤52 wk21,717 (6.6)9056 (5.6)10,265 (8.3)1369 (6.7)1027 (4.8)
       >52 wk or no delivery228,834 (70.1)115,511 (71.5)82,145 (66.7)14,701 (71.6)16,477 (76.7)
      Menorrhagia
       ≤12 mo before insertion32,552 (10.0)13,593 (8.4)15,727 (12.8)2027 (9.9)1205 (5.6)
       >12 mo or no diagnosis294,106 (90.0)147,849 (91.6)107,487 (87.2)18,499 (90.1)20,271 (94.4)
      Menorrhagia (patients at >52 wk after delivery)
       ≤12 mo before insertion31,600 (13.8)13,204 (11.4)15,297 (18.6)1961 (13.3)1138 (6.9)
       >12 mo or no diagnosis197,234 (86.2)102,307 (88.6)66,848 (81.4)12,740 (86.7)15,339 (93.1)
      IUD type
      Site-specific results are not presented in keeping with data use agreements with Kaiser Permanente research sites.
       LNG-IUD259,234 (79.4)
       Copper IUD63,664 (19.5)
       Unknown3,760 (1.2)
      Data are presented as number (percentage).
      IUD, intrauterine device; KPNC, Kaiser Permanente Northern California; KPSC, Kaiser Permanente Southern California; KPWA, Kaiser Permanente Washington; LNG, levonorgestrel; RI, Regenstrief Institute.
      Anthony et al. Association of uterine perforation and expulsion of intrauterine device study methods. Am J Obstet Gynecol 2021.
      a Site-specific results are not presented in keeping with data use agreements with Kaiser Permanente research sites.
      Figure thumbnail gr1
      FigureSchematic of study population and risk factor groups
      IUD, intrauterine device; KPNC, Kaiser Permanente Northern California; KPSC, Kaiser Permanente Southern California; KPWA, Kaiser Permanente Washington; RI, Regenstrief Institute.
      Anthony et al. Association of uterine perforation and expulsion of intrauterine device study methods. Am J Obstet Gynecol 2021.

      Censoring events and duration of continuous enrollment

      Reasons for censoring were similar across sites and overall (Table 3). Removal and/or replacement of IUD (32.0%), end of the study period (32.0%), and end of enrollment or follow-up (25.6%) accounted for most censoring events (89.6%). The proportion of censoring because of IUD expiration, the approach used if no other censoring event or outcome occurred before this time, was 4.5%. Across all sites, the median duration of continuous enrollment was 90.0 (site range, 74.0–176.8) months (Table 4). Median continuous enrollment after the index date was 28.7 (site range, 21.8–30.7) months overall.
      Table 3Percentages of censoring events, pooled and by research site
      Censoring eventPooled, %KPNC, %KPSC, %KPWA, %RI, %
      Removal of IUD (single reason)24.926.024.423.121.2
      Subsequent IUD insertion (single reason)1.40.81.92.02.2
      Both removal and subsequent insertion5.76.15.95.11.5
      Pregnancy (single reason)1.20.91.61.61.6
      Hysterectomy (single reason)0.40.10.70.50.8
      Bilateral oophorectomy or other sterilization (single reason)0.30.20.40.20.3
      IUD expiration (single reason)
      Added 3 months to the IUD expiration in the product label to allow for delayed medical appointments
      4.54.84.14.43.7
      Death (single reason)0.10.10.1<0.10.1
      End of enrollment or follow-up (single reason)25.623.726.143.620.2
      End of the study period (single reason)32.032.831.216.645.2
      Other multiple reasons recorded on the censoring date0.91.00.70.41.4
      Outcome event
      May have had a censoring events in addition to uterine perforation and/or IUD expulsion recorded on the same date.
      3.03.42.92.41.8
      Total100.0100.0100.0100.0100.0
      IUD, intrauterine device; KPNC, Kaiser Permanente Northern California; KPSC, Kaiser Permanente Southern California; KPWA, Kaiser Permanente Washington; RI, Regenstrief Institute.
      Anthony et al. Association of uterine perforation and expulsion of intrauterine device study methods. Am J Obstet Gynecol 2021.
      a Added 3 months to the IUD expiration in the product label to allow for delayed medical appointments
      b May have had a censoring events in addition to uterine perforation and/or IUD expulsion recorded on the same date.
      Table 4Average length of continuous enrollment for the study population, pooled and by research site
      CharacteristicPooled (N=326,658)KPNC (n=161,442)KPSC (n=123,214)KPWA (n=20,526)RI (n=21,476)
      The data set used from RI does not contain enrollment date; first and last clinical encounters were used.
      Continuous enrollment (mo)
       Mean (SD)88.6 (46.3)81.3 (32.5)83.5 (37.1)81.0 (43.0)180.0 (78.1)
       Median (Q1–Q3)90.0 (52.7–114.0)89.0 (52.0–114.0)85.0 (49.3–126.0)74.0 (44.0–119.0)176.8 (123.2–230.4)
       Min–max12.0–438.212.0–114.012.0–126.013.0–150.012.0–438.2
      Continuous enrollment on or before index date (mo)
       Mean (SD)51.9 (42.0)44.1 (25.8)46.1 (28.7)45.7 (31.1)149.7 (77.2)
       Median (Q1–Q3)39.9 (23.1–67.4)37.5 (22.4–61.0)38.2 (22.4–63.3)36.0 (21.5–60.9)145.9 (89.4–201.6)
       Min–max12.0–435.212.0–112.012.0–124.012.1–148.012.0–435.2
      Continuous enrollment on or after index date (mo)
       Mean (SD)36.8 (29.5)37.3 (28.2)37.4 (30.6)35.4 (32.9)30.3 (29.1)
       Median (Q1–Q3)28.7 (11.9–57.1)30.7 (13.0–58.2)28.5 (11.6–58.6)24.3 (9.5–52.8)21.8 (9.6–39.0)
       Min–max0.0–209.60.0–101.90.0–114.00.1–137.90.0–209.6
      KPNC, Kaiser Permanente Northern California; KPSC, Kaiser Permanente Southern California; KPWA, Kaiser Permanente Washington; min, minimum; max, maximum; Q1, lower quartile (ie, 25th percentile); Q3, upper quartile (ie, 75th percentile); RI, Regenstrief Institute; SD, standard deviation.
      Anthony et al. Association of uterine perforation and expulsion of intrauterine device study methods. Am J Obstet Gynecol 2021.
      a The data set used from RI does not contain enrollment date; first and last clinical encounters were used.

      Baseline characteristics

      Table 5 shows the demographics and characteristics of the study population stratified by research site. The mean age in this population was 32 years, and the population was racially and ethnically diverse across the 4 sites. Mean BMI was 28.5 kg/m2 (SD, 7.0 kg/m2); 10% of women were recent smokers. Population characteristics were generally well balanced across sites, with some variations by site in clinical characteristics, including BMI, smoking history, menorrhagia, and presence of uterine fibroids. The demographic characteristics of the study population were representative of the regions covered by the participating healthcare systems as verified by comparison with census data.
      Table 5Characteristics of the study population at the time of the first observed intrauterine device insertion, pooled and by research site
      VariablesPooled (N=326,658)KPNC (n=161,442)KPSC (n=123,214)KPWA (n=20,526)RI (n=21,476)
      Person-years at risk641,427325,552241,92337,49636,456
      Characteristic
      Age (y)
       Mean (SD)32.0 (8.3)32.2 (8.3)32.2 (8.3)31.3 (8.2)30.1 (8.0)
       Categories
      ≤28 y119,469 (36.6)56,832 (35.2)44,859 (36.4)8007 (39.0)9771 (45.5)
      >28–≤36 y107,871 (33.0)54,047 (33.5)39,915 (32.4)7042 (34.3)6867 (32.0)
      >36–≤50 y99,318 (30.4)50,563 (31.3)38,440 (31.2)5477 (26.7)4838 (22.5)
      Race and ethnicity
       Asian or Pacific Islander38,911 (11.9)26,216 (16.2)9998 (8.1)2122 (10.3)575 (2.7)
       Hispanic Black696 (0.2)96 (0.1)524 (0.4)54 (0.3)22 (0.1)
       Hispanic other56,180 (17.2)33,967 (21.0)21,284 (17.3)716 (3.5)213 (1.0)
       Hispanic White42,501 (13.0)2000 (1.2)38,649 (31.4)584 (2.8)1268 (5.9)
       Non-Hispanic Black28,323 (8.7)12,678 (7.9)11,397 (9.2)1234 (6.0)3014 (14.0)
       Non-Hispanic White137,102 (42.0)72,745 (45.1)36,439 (29.6)13,097 (63.8)14,821 (69.0)
       Other or multiple16,357 (5.0)12,249 (7.6)2913 (2.4)492 (2.4)703 (3.3)
       Unknown6588 (2.0)1491 (0.9)2010 (1.6)2227 (10.8)860 (4.0)
      Recent smoker
       Yes32,623 (10.0)14,929 (9.3)11,288 (9.2)1680 (8.2)4726 (22.0)
       No288,539 (88.3)144,366 (89.4)110,831 (90.0)16,592 (80.8)16,750 (78.0)
       Unknown or missing5496 (1.7)2147 (1.3)1095 (0.9)2254 (11.0)0 (0.0)
      BMI (kg/m2)
       Mean (SD)28.5 (7.0)28.0 (6.8)28.9 (7.0)28.0 (7.1)30.0 (8.2)
      BMI categories
       Underweight3689 (1.1)1956 (1.2)1306 (1.1)217 (1.1)210 (1.0)
       Normal weight113,675 (34.8)61,437 (38.1)39,041 (31.7)8010 (39.0)5187 (24.2)
       Overweight96,181 (29.4)47,887 (29.7)37,631 (30.5)5638 (27.5)5025 (23.4)
       Obese107,674 (33.0)49,371 (30.6)44,925 (36.5)6011 (29.3)7367 (34.3)
       Missing5439 (1.7)791 (0.5)311 (0.3)650 (3.2)3687 (17.2)
      Dysmenorrhea diagnosis
       Recent (≤12 mo before index only)10,893 (3.3)3861 (2.4)5651 (4.6)863 (4.2)518 (2.4)
       Past (>1 y before index only)18,080 (5.5)6473 (4.0)7473 (6.1)1904 (9.3)2230 (10.4)
       Diagnosis in recent and past periods4373 (1.3)1437 (0.9)2257 (1.8)477 (2.3)202 (0.9)
       No diagnosis293,312 (89.8)149,671 (92.7)107,833 (87.5)17,282 (84.2)18,526 (86.3)
      Menorrhagia diagnosis
       Recent (≤12 mo before index only)23,398 (7.2)10,373 (6.4)10,826 (8.8)1362 (6.6)837 (3.9)
       Past (>1 y before index only)13,288 (4.1)4698 (2.9)5501 (4.5)1343 (6.5)1746 (8.1)
       Diagnosis in recent and past periods9154 (2.8)3220 (2.0)4901 (4.0)665 (3.2)368 (1.7)
       No diagnosis280,818 (86.0)143,151 (88.7)101,986 (82.8)17,156 (83.6)18,525 (86.3)
      Fibroid diagnosis17,416 (5.3)7742 (4.8)8096 (6.6)1271 (6.2)307 (1.4)
      Parity
       061,920 (19.0)36,814 (22.8)18,980 (15.4)3973 (19.4)2153 (10.0)
       >0225,925 (69.2)112,478 (69.7)95,495 (77.5)8161 (39.8)9791 (45.6)
       Missing38,813 (11.9)12,150 (7.5)8739 (7.1)8392 (40.9)9532 (44.4)
      Cesarean delivery any time before the index date
       Yes54,295 (16.6)25,233 (15.6)22,939 (18.6)2295 (11.2)3828 (17.8)
       No171,630 (52.5)87,245 (54.0)72,556 (58.9)5866 (28.6)5963 (27.8)
       Nullipara or missing100,733 (30.8)48,964 (30.3)27,719 (22.5)12,365 (60.2)11,685 (54.4)
      Cesarean delivery at most recent delivery before the index date
       Yes23,245 (7.1)10,081 (6.2)10,638 (8.6)1402 (6.8)1124 (5.2)
       No74,579 (22.8)35,850 (22.2)30,431 (24.7)4423 (21.6)3875 (18.0)
       No delivery in past year228,834 (70.1)115,511 (71.6)82,145 (66.7)14,701 (71.6)16,477 (76.7)
      Concomitant gynecologic procedure
       Yes
      At least 1 of the following: abortion, aspiration and curettage, dilation and curettage, excision or biopsy of the cervix or uterus, ablation, colposcopy and other cervical procedures, hysteroscopy procedure, laminaria procedure, laparoscopy, lysis adhesions, myomectomy, nerve procedure, or salpingectomy or oophorectomy.
      26,234 (8.0)13,494 (8.4)10,770 (8.7)1275 (6.2)695 (3.2)
      Difficult insertion indicator
       Any difficult insertion29,777 (9.1)19,685 (12.2)4273 (3.5)2324 (11.3)3495 (16.3)
       Cervical dilation10,209 (3.1)8501 (5.3)33 (0.0)102 (0.5)1573 (7.3)
       Ultrasound guidance4628 (1.4)3620 (2.2)252 (0.2)194 (0.9)562 (2.6)
       Paracervical block14,731 (4.5)12,788 (7.9)1051 (0.9)654 (3.2)238 (1.1)
       Difficult insertion noted2987 (0.9)1701 (1.1)767 (0.6)230 (1.1)289 (1.3)
       Use of misoprostol8689 (2.7)3827 (2.4)2329 (1.9)1295 (6.3)1238 (5.8)
      Calendar year of IUD insertion
       2001–200916,524 (5.1)010,840 (8.8)4585 (22.3)1099 (5.1)
       201031,563 (9.7)18,206 (11.3)11,032 (9.0)1847 (9.0)478 (2.2)
       201132,747 (10.0)17,974 (11.1)12,311 (10.0)1986 (9.7)476 (2.2)
       201236,584 (11.2)19,911 (12.3)13,728 (11.1)2111 (10.3)834 (3.9)
       201334,303 (10.5)18,694 (11.6)12,377 (10.0)2012 (9.8)1220 (5.7)
       201433,946 (10.4)18,769 (11.6)11,699 (9.5)1963 (9.6)1515 (7.1)
       201537,621 (11.5)19,144 (11.9)13,072 (10.6)2006 (9.8)3399 (15.8)
       201641,302 (12.6)20,242 (12.5)14,894 (12.1)1773 (8.6)4393 (20.5)
       201746,518 (14.2)21,688 (13.4)17,284 (14.0)1681 (8.2)5865 (27.3)
       201815,550 (4.8)6814 (4.2)5977 (4.9)562 (2.7)2197 (10.2)
      Duration of lookback period (mo)
       Mean (SD)56.8 (42.3)53.9 (28.7)46.1 (28.7)45.7 (31.1)149.7 (77.2)
       Min–max12–43512–11212–12412–14812–435
      Data are presented as number (percentage), unless otherwise specified.
      BMI, body mass index; IUD, intrauterine device; KPNC, Kaiser Permanente Northern California; KPSC, Kaiser Permanente Southern California; KPWA, Kaiser Permanente Washington; min, minimum; max, maximum; RI, Regenstrief Institute; SD, standard deviation.
      Anthony et al. Association of uterine perforation and expulsion of intrauterine device study methods. Am J Obstet Gynecol 2021.
      a At least 1 of the following: abortion, aspiration and curettage, dilation and curettage, excision or biopsy of the cervix or uterus, ablation, colposcopy and other cervical procedures, hysteroscopy procedure, laminaria procedure, laparoscopy, lysis adhesions, myomectomy, nerve procedure, or salpingectomy or oophorectomy.

      Comment

      The EURAS-IUD and APEX-IUD studies employed distinct and complementary designs to explore different research questions. The EURAS-IUD recruited women over a 7-year period (2006–2012) to evaluate the risk of uterine perforation among IUD users in routine practice.
      • Heinemann K.
      • Reed S.
      • Moehner S.
      • Minh T.D.
      Risk of uterine perforation with levonorgestrel-releasing and copper intrauterine devices in the European Active Surveillance Study on Intrauterine Devices.
      During 1 year of follow-up among 61,448 women, 81 uterine perforations were identified.
      • Heinemann K.
      • Reed S.
      • Moehner S.
      • Minh T.D.
      Risk of uterine perforation with levonorgestrel-releasing and copper intrauterine devices in the European Active Surveillance Study on Intrauterine Devices.
      In a 5-year extension among 39,009 women, an additional 23 perforations were identified.
      • Barnett C.
      • Moehner S.
      • Do Minh T.
      • Heinemann K.
      Perforation risk and intra-uterine devices: results of the EURAS-IUD 5-year extension study.
      In comparison, protocol submission and initiation of data collection for the APEX-IUD study occurred in 2018; final results will be available after 20 months. Also the APEX-IUD study provided up to 9.5 years of follow-up data, with more than 5 times more women enrolled (a total of 326,658 women in the United States), and the study evaluated multiple additional potential risk factors for uterine perforation and IUD expulsion.
      The APEX-IUD population was younger than the EURAS-IUD population
      • Heinemann K.
      • Reed S.
      • Moehner S.
      • Minh T.D.
      Risk of uterine perforation with levonorgestrel-releasing and copper intrauterine devices in the European Active Surveillance Study on Intrauterine Devices.
      : mean age was 32.2 years vs 37.4 years among LNG-IUD users and 31.2 years vs 33.3 years among copper IUD users. Across the 4 research sites, the APEX-IUD population reflected the general populations in the geographic regions from which they were drawn and was racially and ethnically diverse. The EURAS-IUD population included proportionally more smokers (23.2% current smokers) than the APEX-IUD population (10.2% who were smokers in ≤12 months before IUD insertion) and proportionally more women with normal weight as evidence by a BMI of <25 kg/m2 (47.3% vs 36.5% in the APEX-IUD population).

      Clinical implications

      The APEX-IUD study will provide additional information on IUD-related uterine perforation reported by EURAS-IUD investigators from a European population. The APEX-IUD study provides information reflective of demographic characteristics and clinical care in the United States. In addition, the APEX-IUD study will include IUD expulsion as a study outcome and provide additional details regarding the impact of menorrhagia on outcomes. The study will contribute significantly to the knowledge regarding IUD risks that might lead to unintended pregnancy, providing a comprehensive picture that aids in clinical decision-making.

      Research implications

      There are advantages to prospective cohort studies, including the EURAS-IUD, such as the ability to specify the data to be collected, to query participants in real time, and to include those who might not be represented in healthcare data sources. However, prospective real-world studies rely on volunteers who might not completely reflect the populations of interest and require intensive follow-up procedures to ensure participant retention. As a US retrospective cohort study involving EHR data within insured healthcare systems, the APEX-IUD study represents a satisfactory approach for an FDA-mandated study.

      Strengths and limitations

      Key advantages of the APEX-IUD study are the retrospective design that enabled a larger cohort, a shorter time to the availability of results, a longer duration of follow-up, and greater efficiency relative to a prospective design. Retrospective data collection from a healthcare system data source is limited to information about care received while enrolled in the healthcare system and assumes enrollees do not seek care outside the system during their enrollment. Thus, risk estimates could be underestimated. The results of this study depend on the accurate capture of health information and definitions of variables selected. Misclassification is possible because variables were determined from codes and clinical notes (via NLP). There is also the potential for misclassification of outcomes and risk factors within the data sources if women were not aware, women did not seek treatment, or if treatments were not documented in the EHR.

      Conclusion

      This study has a large, sociodemographically diverse population with rich data providing the opportunity to evaluate the risk of uterine perforation and IUD expulsion in the setting of usual clinical practice in the United States. Relative to a prospective study, the retrospective design of the APEX-IUD study enabled a larger study population with a longer follow-up time and was a satisfactory alternative to an FDA request for prospective data. When reliable risk factor, outcome, and covariates can be ascertained from the EHR, real-world data can provide informative risk estimates that can be used to inform clinical practice. Future results from this study can provide valuable information based on real-world evidence about risk factors for IUD perforation and expulsion for clinicians.

      Highlights

      • Study methods and cohort data of the APEX-IUD study are presented in this report.
      • The Food and Drug Administration mandated the APEX-IUD study to evaluate IUD-related uterine perforation and IUD expulsion.
      • Key risk factors are as follows: breastfeeding and postpartum timing at IUD insertion.
      • Variables were identified through structured data and natural language processing of 326,658 women with ≥1 IUD insertion (median continuous enrollment, 7.5 years).

      Acknowledgments

      The team members of the APEX-IUD study would like to thank Kaiser Permanente Northern California, Southern California, and Washington, and Regenstrief Institute members who contributed electronic health information to this study. In addition, the team members would like to thank Kate Lothman (RTI Health Solutions) for her medical writing contributions.

      Supplemental Material

      Appendix A: data sources

      Four data sources with electronic health records (EHRs) were used for this study: Kaiser Permanente Northern California (KPNC), Kaiser Permanente Southern California (KPSC), Kaiser Permanente Washington (KPWA), and Regenstrief Institute (RI). Data in different files within each data source were linked by the patient’s identification number. Descriptions of the healthcare system for each source of data follow.

      Kaiser Permanente Northern California

      The KPNC region in California extends from Santa Rosa and Sacramento in the north, to Modesto in the east, to San Jose and Fresno in the south and includes the entire San Francisco Bay Area. It covers 21 hospitals and 238 medical offices. KPNC covers approximately 4 million patients, representing half of the commercially insured patients and one-quarter of patients with Medicare in the area. The patient population represents the diversity of age, sex, and race and ethnicity in the regions served.
      Data for KPNC are housed within a comprehensive EHR system that captures every patient encounter in every department, including hospital, emergency, ambulatory surgical, specialist, and generalist care encounters; clinic visits and telephone encounters; physiological measures; procedures; laboratory and radiology testing; and diagnoses. The comprehensive EHR system was fully implemented in 2009. Standardized research data sets—including enrollment, sociodemographics, pharmacy, encounters, diagnoses, procedures, vital signs, census, and laboratory results—are maintained for the purposes of research. Data are linked across all data sets via a unique medical record number. Infant records are maintained and can be linked to the mother’s delivery record data.
      In the validation study, continuous enrollment in KPNC was measured via enrollment files. Of all intrauterine device (IUD) insertions in this healthcare system, 67% (more than 100,000) were in women with at least 12 months of continuous enrollment before the date of insertion.

      Kaiser Permanente Southern California

      Kaiser Permanente Southern California is Kaiser Permanente’s largest region, with 4.6 million members who broadly represent the diversity of age, sex, and race and ethnicity in the Southern California population. KPSC covers 15 hospitals and over 234 medical offices.
      The KPSC EHR system was fully implemented in 2008 and integrates all aspects of care, including pharmacy and laboratory services, appointments, registration, and billing. Standardized research data sets are maintained similar to those in KPNC, including date and site of care, diagnosis codes, procedure codes, vaccinations, prescription medications and dispensing activities, vital signs, radiology, clinical reports, telephone encounters, laboratory and pathology results, and member demographics and enrollment information.
      Each KPSC member is assigned a unique medical record number on joining the health plan. This number is retained for life, irrespective of leaving and rejoining the health plan. This unique number allows for the linkage across all data sets (both clinical and administrative). The prenatal data set includes data on live births, and infant records can be linked to the mother’s data.
      In the validation study, continuous enrollment in KPSC was measured via enrollment files. Of all IUD insertions in this healthcare system, 67% (more than 80,000) were in women with at least 12 months of continuous enrollment before the date of insertion.

      Kaiser Permanente Washington

      Based in state of Washington, KPWA (formerly the Group Health Cooperative, a nonprofit health system) currently serves approximately 700,000 members and provides primary, specialty, hospital, home health, and inpatient skilled nursing care on a prepaid (capitation) basis. Members in the healthcare system represent the diversity of age, sex, and race and ethnicity of the geographic region. About two-thirds of members receive comprehensive care at Kaiser Permanente medical facilities (Group Practice). KPWA has 36 care locations across the state plus 4 additional facilities with specialties, such as eye care and mental health, in addition to 24/7 urgent care centers. The remaining members receive care from contracted provider networks in geographic areas not served by KPWA medical facilities but reimbursed by KPWA.
      The EHR system was fully implemented in 2006 and includes data sets on enrollment, encounters, diagnoses, procedures, vital signs, radiology, pathology, laboratory tests, and pharmacy dispensing. Data are linked across all data sets via a unique member identifier. The mother-infant data set includes data on women with live births and linked infant records.
      In the validation study, continuous enrollment in KPWA was measured via enrollment files. Of all IUD insertions in this healthcare system, 64% (more than 15,000) were in women with at least 12 months of continuous enrollment before the date of insertion.

      Regenstrief Institute

      The RI has research access to the Indiana Health Information Exchange, which represents over 16 million patients and includes clinical data from 117 Indiana hospitals, 38 core hospital systems, and the state and local public health departments of Indiana. Data from healthcare encounters were available for this study since 2001 and were captured in a standardized fashion. Data are linked via a unique identifier across institutions.
      In the validation study, continuous enrollment was measured via healthcare encounters. Of all IUD insertions in this healthcare system, 74% (approximately 5700) were in women with at least 1 clinical encounter 12 months or more before the date of insertion.

      Appendix B: evaluation of change from International Classification of Diseases, Ninth Revision, to International Classification of Diseases, Tenth Revision, Clinical Modification codes on outcomes

      Algorithms for the outcome variables of uterine perforation and IUD expulsion were validated in these 4 data sources before use of International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) coding. No formal validation of the algorithms with ICD-10-CM codes to identify uterine perforation or IUD expulsion was done in this study. However, the proportion of patients at risk who had 1 of these outcomes was evaluated before and after the implementation of ICD-10-CM coding to evaluate consistency over time.
      In the pooled data (Supplemental Table), the proportion of women with a uterine perforation was 0.11% in the 12 months before ICD-10-CM code implementation and 0.12% in the 12 months after ICD-10-CM code implementation (which occurred on October 1, 2015). The proportion of women with an IUD expulsion in the 12 months before and 12 months after ICD-10-CM code implementation was 0.72% and 0.77%, respectively. At each research site, the proportions before and after ICD-10-CM code implementation were relatively consistent, with the exception of RI, where the estimates were based on a small number of events. The RI identified most cases from natural language processing (so there would be little impact of the ICD-9 or ICD-10-CM code changes), and they reviewed the records of all potential cases to verify case status.

      Appendix C: propensity score modeling

      For the final analyses, propensity scores are used to calculate weights for each IUD insertion within each risk factor group. The weights are “overlap weights.”
      • Li F.
      • Morgan K.L.
      • Zaslavsky A.M.
      Balancing covariates via propensity score weighting.
      This method has an advantage of not requiring trimming of observations; therefore, observations where there is significant overlap among groups are up-weighted, and observations where there is very little overlap are down-weighted, compared with regular inverse probability treatment weighting. To assess whether covariates are balanced across risk factor groups after weighting, the distribution of each variable is compared among categories of the risk factor variable, and balance parameters (ie, standardized differences) are calculated.
      • Austin P.C.
      • Stuart E.A.
      Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies.
      Pairwise balance parameters (ie, pairwise standardized differences) are used for the categorical risk factor variable (postpartum timing), in which each category is compared with the referent group. The balance among risk factor groups is assessed overall and within each site. If the groups are unbalanced on key covariates after application of overlap weighting, then the logistic regression model is revised by including interaction terms with the data source, and the covariate balance among the groups overall and within each data source is reevaluated on the basis of the revised model.
      • Rosenbaum P.R.
      • Rubin D.B.
      Reducing bias in observational studies using subclassification on the propensity score.
      ,
      • Austin P.C.
      An introduction to propensity score methods for reducing the effects of confounding in observational studies.
      When satisfactory balance between the exposed and unexposed groups is achieved (in general, absolute standardized difference of <0.2), the weighting is incorporated in modeling for confounder-adjusted outcome assessments.
      Supplemental TableProportion of patients with an IUD who had a uterine perforation or IUD expulsion within the 12 months before and 12 months after the implementation of ICD-10-CM coding
      Site12 mo before ICD-10 implementation12 mo after ICD-10 implementation
      Patients at risk, NUterine perforation, n (%)IUD expulsion, n (%)Patients, NUterine perforation, n (%)IUD expulsion, n (%)
      Pooled84,92993 (0.11)614 (0.72)91,851108 (0.12)709 (0.77)
      KPNC46,29757 (0.12)392 (0.85)49,86665 (0.13)455 (0.91)
      KPSC31,11620 (0.06)168 (0.54)32,66829 (0.09)191 (0.58)
      KPWA44426 (0.14)35 (0.79)43135 (0.12)38 (0.88)
      RI307410 (0.33)19 (0.62)50049 (0.18)25 (0.50)
      ICD-10-CM, International Classification of Diseases, Tenth Revision, Clinical Modification; IUD, intrauterine device; KPNC, Kaiser Permanente Northern California; KPSC, Kaiser Permanente Southern California; KPWA, Kaiser Permanente Washington; RI, Regenstrief Institute.
      Anthony et al. Association of uterine perforation and expulsion of intrauterine device study methods. Am J Obstet Gynecol 2021.

      References

      1. Practice Bulletin no. 186 summary: long-acting reversible contraception: implants and intrauterine devices.
        Obstet Gynecol. 2017; 130: 1173-1175
        • Barnett C.
        • Moehner S.
        • Do Minh T.
        • Heinemann K.
        Perforation risk and intra-uterine devices: results of the EURAS-IUD 5-year extension study.
        Eur J Contracept Reprod Health Care. 2017; 22: 424-428
        • O’Brien P.A.
        • Pillai S.
        Uterine perforation by intrauterine devices: a 16-year review.
        J Fam Plann Reprod Health Care. 2017; 43: 289-295
        • Heinemann K.
        • Reed S.
        • Moehner S.
        • Minh T.D.
        Risk of uterine perforation with levonorgestrel-releasing and copper intrauterine devices in the European Active Surveillance Study on Intrauterine Devices.
        Contraception. 2015; 91: 274-279
        • Anthony M.S.
        • Armstrong M.A.
        • Getahun D.
        • et al.
        Identification and validation of uterine perforation, intrauterine device expulsion, and breastfeeding in four health care systems with electronic health records.
        Clin Epidemiol. 2019; 11: 635-643
        • European Network of Centres for Pharmacoepidemiology and Pharmacovigilance (ENCePP)
        Study on the association of uterine perforation and intrauterine device (IUD) expulsion with breastfeeding status at the time of IUD insertion and postpartum timing of IUD insertion in electronic medical record databases. A postmarketing requirement for Mirena.
        (Available at:)
        http://www.encepp.eu/encepp/viewResource.htm?id=33469
        Date: 2020
        Date accessed: May 1, 2020
        • Brunelli S.M.
        • Gagne J.J.
        • Huybrechts K.F.
        • et al.
        Estimation using all available covariate information versus a fixed look-back window for dichotomous covariates.
        Pharmacoepidemiol Drug Saf. 2013; 22: 542-550
        • Schoenfeld D.A.
        Sample-size formula for the proportional-hazards regression model.
        Biometrics. 1983; 39: 499-503
        • Li F.
        • Morgan K.L.
        • Zaslavsky A.M.
        Balancing covariates via propensity score weighting.
        J Am Stat Assoc. 2018; 113: 390-400
        • Austin P.C.
        • Stuart E.A.
        Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies.
        Stat Med. 2015; 34: 3661-3679