We used the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines for reporting this review.¹² We searched PubMed, Cochrane Library, and ClinicalTrials.gov for all primary research studies of any study design, in any language, published from database inception through May 2018 that examined postpartum placement of IUDs. We searched PubMed using the following search strategy: ((((“Intrauterine Devices”[Mesh] OR “Intrauterine Devices, Copper”[Mesh] OR “Intrauterine Devices, Medicated”[Mesh] OR ((intrauterine OR intrauterine) AND (device OR system OR contracept*)) OR IUD OR IUC OR IUCD OR IUS OR mirena OR Skyla OR liletta OR paragard OR “Copper T380” OR CuT380 OR “Copper T380a” OR “Cu T380a”) AND (postpartum OR Puerperium*) NOT (“Animals”[Mesh] NOT “Humans”[Mesh])))). We searched Cochrane Library and ClinicalTrials.gov for any published reviews or additional studies including “Postpartum AND IUD.” We hand-searched relevant articles and reviews for additional references.

We included studies that examined immediate post-partum placement (10 minutes or less after placental delivery) or early postpartum placement (greater than 10 minutes to less than 4 weeks) and reported counts of expulsion and counts of women with follow-up. We included studies with any length of follow-up and any rate of follow-up. We excluded studies that did not report counts of women with IUD placement and occurrence of expulsion, which were required for calculation of pooled estimates of expulsions across studies. We included studies that examined copper-bearing IUDs that are currently available (CuT380A) or were previously available in the United States (Copper 7, TCu200) and the levonorgestrel intrauterine system (LNG-IUS). We excluded studies that evaluated IUDs that were modified from their standard structure and were not commercially available (eg, sutures added to the IUD to anchor the IUD to the endometrium). We included studies that examined IUD placement after vaginal delivery, cesarean delivery, or both.

Two coauthors (T.C.J. and M.K.W.) independently screened all titles and abstracts identified from the initial search to determine whether the studies met inclusion criteria. Full-text articles were reviewed as needed. Pertinent non-English articles were professionally translated by U.S. Centers for Disease Control and Prevention translation services. The lead author (T.C.J.) and one additional coauthor (M.K.W., E.B.-B., N.K.T.) reviewed each included study and abstracted the following information: study author, year of publication, country, funding source, study design, IUD type, timing of IUD placement, delivery method, length of study follow-up, number of women enrolled or randomized, number of IUDs initially placed, number of women with any follow-up, and counts of expulsion (overall, complete, and partial). To assess risk of bias, two coauthors independently reviewed each study according to the U.S. Preventive Services Task Force system to evaluate study design and methodologic features such as potential for selection bias (eg, groups not comparable at baseline for randomized controlled trials), mis-classification (eg, outcome of expulsion diagnosed inconsistently by nonblinded health care provider), and confounding (eg, parity and breastfeeding status not collected or adjusted for).¹³ Two coauthors (E.B.-B. and N.K.T.) assessed one study that was authored by the lead author of this review.¹⁴ Any discrepancies between authors for selection, abstraction, or risk of bias assessment were resolved through discussion.

We calculated expulsion rates by pooling the number of women with IUD placements and the number of IUD expulsions reported for each individual study for women with any follow-up. This calculation of the proportion of a pooled number of expulsions over a pooled number of IUD placements is equivalent to a sum of individual expulsion rates weighed by their corresponding study size. We assumed expulsions were complete expulsions if not otherwise defined, and we excluded partial expulsions from our primary analysis, because not all studies reported them, definitions varied or were absent from studies, or malpositioned IUDs were reported as partial expulsions. Because many of the included randomized controlled trials reported follow-up by intent-to-treat analyses, we were unable to determine how many women received IUDs at the assigned time period per protocol and returned for follow-up; therefore, for those studies that reported only intent-to-treat results, we used the number of women randomized who had follow-up as our denominator.

We calculated pooled expulsion rates within strata defined by factors including: timing of post-partum placement (immediate, early, either immediate or early [mixed], or interval placement at 4 weeks postpartum or greater), delivery method (cesarean, vaginal, either cesarean or vaginal [mixed], or unknown), IUD type (CuT380A, LNG-IUS, either CuT380A or LNG-IUS [mixed], Copper 7, or TCu200), length of study follow-up, study region (defined by the World Health Organization, www.who.int/about/regions/en/), and study quality (defined by U.S. Preventive Services Task Force¹³). The calculations were weighted by number of women in each individual study with respect to different factors listed. We included interval placements only when describing expulsions by placement timing because we wanted to compare the various postpartum periods with interval placements; for other comparisons, we focused only on immediate or early postpartum placements.

Using a log-binomial regression model, we estimated adjusted relative risks (RRs) of IUD expulsion and their associated 95% CIs with inclusion of potential risk factors (timing of IUD placement, delivery method, and IUD type) while adjusting for other covariates including World Health Organization study region, study quality, and length of study follow-up. Finally, we performed a number of sensitivity analyses related to our exclusions based on losses to follow-up and partial expulsion reporting. Women who had an IUD placement and were lost to follow-up were excluded from the primary analysis. To assess the possible effect of loss to follow-up, we included studies with a follow-up rate of at least 70% for additional analyses by assuming that all women lost to follow-up either continued their IUD or experienced an expulsion.¹⁵ In separate analyses, we also estimated adjusted RRs of overall expulsions (complete and partial expulsions together) for timing of placement, delivery method, and IUD type. All analyses were conducted using SAS 9.4.

We identified 1,112 articles in PubMed, 19 studies in ClinicalTrials.gov, and three Cochrane reviews, for a total of 1,134 records (Fig. 1). After removing 69 duplicates and excluding 117 articles published before 1974, the first year when IUDs of interest were initially studied and available, we screened the title and abstract of the 948 articles. We excluded 816 not relevant to our search and then reviewed the full text of 132 articles. Based on full-text review, we excluded 84 articles that did not include IUDs that met inclusion criteria, did not include postpartum IUD placements, or did not provide individual counts for expulsions. We did not identify any additional completed studies with published data from ClinicalTrials.gov. From our Cochrane search, we identified one updated Cochrane review that included 15 trials, of which five met inclusion criteria.³ A total of 48 studies met our inclusion criteria. Two of the articles that met our inclusion criteria were translated from Spanish or French^16,17 and the remaining were published in English.

We describe study characteristics in Table 1. The body of evidence included studies largely published in the past 10 years, from all World Health Organization regions, with levels of evidence of I, II-2, and II-3 and with quality ratings assessed most frequently as fair or poor (compared with good for three studies).^7,18,19 Twenty-two studies included only CuT380A IUDs,^{8,16–18,20–37} 13 included only LNG-IUS,^19,38–49 and nine included a mix of the two^{7,9,10,14,50–54}; the remaining four studies included CuT200^55–57 or Cu7 IUDs.⁵⁸ Twenty-eight studied women with IUDs placed only in the immediate postpartum period,^{9,10,14,17,18,21,23,24,26–30,32–37,40–43,52–54,57,58} four studied only early postpartum placements,^44,47,55,56 and the remaining 16 studied more than one placement period.^{7,8,16,19,20,22,25,31,38,39,45,46,48–50,51} Similar numbers of studies evaluated vaginal and cesarean deliveries, with 14 including only vaginal deliveries,^{14,16,19,20,22,31,34,39,42,44,45,48,53,5816} including only cesarean deliveries,^{7,8,17,18,27–30,35,37,38,40,41,43,46,54} 16 including both delivery methods,^{9,10,21,23–26,32,33,36,47,49,50,51,52,57} and two not reporting delivery method.^55,56 Sample sizes ranged from 7–2,733 women; follow-up duration ranged from 4 weeks–5 years, and follow-up rates ranged from 29–100%.

We describe the crude pooled rates of expulsion by placement timing, delivery method, and IUD type among women with any follow-up in Table 2. Expulsion rates from individual studies ranged from 0% to 46.7% and generally increased with increasing length of study follow-up, ranging from 7.3% (range 0.0–21.4) for studies with follow-up less than 3 months to 18.4% (range 0.0–39.4) for studies with follow-up greater than 6 months. Although the most frequent study follow-up time period was 3–6 months, those with follow-up less than 3 months contributed the largest total number of women to our primary analysis.

By IUD placement timing (n58,569 women with immediate, early, or interval placements), immediate placement had a pooled expulsion rate of 10.0% (range 0.0–26.7%; n54,754), whereas early placement had a pooled expulsion rate of 29.7% (range 0.0–46.7%; n51,372) compared with 1.9% (range 0.0–3.9%; n5633) for interval placement when including all lengths of follow-up (Table 2). By delivery method (n57,936 women with immediate or early placements), the pooled expulsion rate for vaginal deliveries was 14.9% (range 3.3–46.7%; n51,543) and for cesarean deliveries was 3.6% (range 0.0–21.1%; n51,512) with higher rates for vaginal deliveries compared with cesarean delivery for all follow-up intervals. By IUD type including immediate and early placements (n57,936 women with immediate or early placements), pooled expulsion rates were highest for CuT200 IUDs (22.8%, range 3.6–39.4%; n52,179), and rates were higher for LNG-IUS (15.5%, range 0.0–46.7%; n5718) compared with CuT380A (6.7%, range 0.0–19.2%; n54,567). Although CuT380A pooled expulsion rates varied from 5.1%, 8.9%, and 9.4% among studies with less than 3 months, 3–6 months, and longer than 6 months follow-up, respectively, pooled rates of LNG-IUS expulsion decreased among studies with longer follow-up (22.9% during less than 3 months follow-up, 17.4% during 3–6 months follow-up, and 2.5% for follow-up longer than 6 months).

Studies from Africa (n54), Europe (n53), eastern Mediterranean (n57), and Southeast Asia (n59) had pooled postpartum expulsion rates of 2.2%, 3.8%,6.3%, and 6.4%, respectively, whereas three studies pooled from South America had rates of 32.7%. The expulsion rate from 20 studies in North America was 11.3% in comparison with 15.8% from two studies in the Western Pacific. Expulsion rates did not differ by study quality (data not shown).

In multivariable analysis, examining complete expulsions among the sample of women with any follow-up, placement timing, delivery method, and IUD type were associated with risk of expulsion (Table 3). Compared with interval placement, both immediate and early placement had a more than sixfold higher risk of expulsion (adjusted RR 7.63, 95% CI 4.31–13.51; adjusted RR 6.17, 95% CI 3.19–11.93, respectively). Compared with cesarean delivery, IUD placement after vaginal delivery had a fivefold higher risk of expulsion (adjusted RR 5.19, 95% CI 3.85–6.99). Compared with CuT380A, risk of expulsion was higher for the LNG-IUS (adjusted RR 1.91, 95% CI 1.50–2.43) and CuT200 (adjusted RR 1.42, 95% CI 1.06–1.90); however, risk was lower for Cu7 (adjusted RR 0.21, 95% CI 0.08–0.56).

For our sensitivity analyses that included women lost to follow-up (n510,948) (Table 3), we only included 43 studies with a follow-up rate of at least 70%. We first assumed all women lost to follow-up (n52,379) experienced an expulsion and then assumed all women lost to follow-up retained their IUD. Although some variations of estimated risks of complete expulsions existed for both assumptions in comparison with the primary analysis, in general, results were similar with regard to placement timing, delivery methods, and IUD type.

For our sensitivity analysis that examined all expulsions (complete and partial expulsions), the adjusted RR of overall expulsion for early compared with interval placement was nearly twofold higher compared with our primary analysis (adjusted RR 11.17, 95% CI 6.74–18.52) (data not shown). The significantly higher risk of expulsion for immediate placement compared with interval placement remained when examining risk for overall expulsions as did the higher risk of expulsion for vaginal delivery compared with cesarean delivery (data not shown). When examining overall expulsions by IUD type, the higher risk of expulsion for LNG-IUS and CuT200 compared with CuT380A and lower risk for Cu7 remained (data not shown).

This analysis describes pooled postpartum expulsion rates of Cu-IUDs and LNG-IUS and adjusted RRs of expulsion by timing of placement, delivery method, and IUD type. In adjusted multivariable analyses, RRs of expulsion were higher for immediate placement and early placement compared with interval placement, for vaginal compared with cesarean delivery, and for LNG-IUS and CuT200 compared with CuT380A.

These results generally confirm previous findings from narrative systematic reviews^2–4 and provide new information regarding absolute expulsion rates for IUDs placed in the postpartum period. Although several studies have been published recently, most have been pilots or small studies designed to look at continuation rates; sample sizes were too small to detect differences in expulsion rates between groups by timing of placement, delivery method, or IUD type—key clinical factors that may influence expulsion rates. High attrition rates also have been common, and most studies have follow-up periods of 3–6 months. By pooling these data, we had a sample of approximately 8,000 women with follow-up data who had an IUD placed in the postpartum period from study locations across the world, with more than 90% receiving a copper IUD. In addition to our factors of interest, we adjusted for study quality, which varied from poor to good, and length of follow-up. We also included study region because certain areas (eg, Southeast Asia) have had more experience with postpartum IUD placements and thus may have lower expulsion rates.

This analysis is subject to several limitations. First, expulsions were not always clearly defined in the studies, diagnostic criteria were rarely reported, and expulsions may have been ascertained by varied methods, including clinic visit, patient report, or chart review. We assumed when studies reported expulsions not further characterized as complete or partial that these were complete expulsions, which may overestimate our primary outcome. Even for those reporting complete and partial expulsions, definitions distinguishing the two types were rarely reported. The ranges of pooled expulsion rates highlight the lack of data precision even when stratified by placement timing, delivery method, IUD type, and length of follow-up. We were unable to examine potential factors for IUD expulsion such as health care provider experience, insertion technique, or ultrasound use because studies did not consistently report these variables. Because our analysis focused on factors at the study level, we were not able to examine potential confounders at the patient level such as age, parity, and breastfeeding status, which may also influence expulsion. Additionally, we were not able to pinpoint when expulsions most commonly occur because studies did not consistently report timing of expulsions after placement.

To improve this body of evidence and inform a future pooled expulsion analysis, we recommend studies include clear diagnostic criteria for complete and partial expulsions, timing of expulsion, IUD type, insertion technique (eg, manual, ring forceps, or inserter), level of health care provider experience, and breastfeeding status.

Although complete expulsions may be less prone to misclassification, partial expulsions may or may not include malpositioned IUDs, whose clinical significance remains unknown. Our sensitivity analysis examining complete and partial expulsions confirms findings from our primary analysis of complete expulsions and suggests women with early postpartum IUD placements may be at much higher risk for complete and partial expulsions compared with women with interval placements. One recent case–control study found more pregnancies among women with malpositioned IUDs than those with IUDs in normal position as a result of high rates of IUD removal and lack of subsequent highly effective contraception.⁵⁹ None of 28 women with malpositioned IUDs who kept the IUD experienced a pregnancy within 2 years.

Our results, along with safety data and other recommendations, can be used to support essential postpartum contraception initiatives^1–5,60 and aid in counseling women to make an informed choice as to when to initiate their IUD. Although expulsion risk is increased with IUD placement in the immediate and early postpartum periods compared with interval placement, providing contraceptive access to women may outweigh expulsion risk for women who prefer the convenience and other beneficial factors of postpartum IUD placement or who may face additional barriers to placement at a later time. When feasible, provision of IUDs immediately postpartum (within 10 minutes after placental delivery) may result in a lower expulsion risk than early postpartum placement.