Predictive factors for failure of the levonorgestrel releasing intrauterine system in women with heavy menstrual bleeding

The general rule of thumb studying one predictor variable for every 10 events was taken into account for the analysis. Eight potential prognostic factors were considered, based on previous research and on clinical knowledge: age, BMI, history of caesarean section, vaginal delivery, previous treatment for HMB, use of anticoagulants, dysmenorrhea, and baseline PBAC score [6, 9‐13]. These characteristics were first tested in a univariable model. When fitting the model, the relation of these potential prognostic factors with failure of treatment was tested using various forms, including categorization, dichotomization or continuous. In the final model, all factors were included as dichotomized variables. Age was dichotomized as ≤ 45 or > 45 years, BMI as ≥ 25 or < 25 kg/m², caesarean section, vaginal delivery, previous HMB treatment, and anticoagulant use as yes or no, and baseline PBAC score as ≥ 425 or < 425 (above and below baseline median PBAC score). Dysmenorrhea (measured with a 4-point Likert scale) was dichotomized as present or non-present, and to estimate the risk more specifically for women who experienced severe dysmenorrhea, we also dichotomized it as ‘severe’ or ‘none to moderate’. In addition, the model was adjusted for any effect of randomization by including a variable to code the RCT and observational data. All variables which were found to be risk factors in previous research together with statistically significant variables in the univariable analyses were entered in a multivariable logistic regression model using the forced entry method. Stochastic regression imputation was used to impute missing baseline data, with the assumption that values were missing at random. A sensitivity analysis was performed in which women who were included in general practice and did not receive an ultrasound were excluded, in order to rule out bias due to unknown intracavitary pathology.

Baseline continuous variables were presented either as means and standard deviations (SD), or median and interquartile range (IQR). Categorical data were presented as numbers and percentages. Relative risks (RR) with 95% confidence intervals (CI) were estimated using log-binomial regression. A significance level of 5% was used for all statistical tests. Statistical analyses were performed using the software Statistical Package for the Social Sciences (SPSS), version 25 (IBM Corp, New York, USA).

In this study we aimed to identify prognostic factors associated with failure of the LNG-IUS in women with HMB. Age below 45 years and the presence of severe dysmenorrhea were found to be independently associated with a higher risk of failure.

Strengths of this study are that the data was prospectively collected and the study had a relatively long follow-up period (24 months). Information about reinterventions was not only obtained from participants’ medical records, but also from patient questionnaires, with which we minimized the risk of missing data in women who did not receive a reintervention in the hospital where they received the LNG-IUS. The MIRA-trial was a pragmatic trial in which women could be treated at both the general practice as the gynecology department, reflecting daily practice. As a consequence, the underlying cause of HMB was not extensively examined and women who were included at the general practice did not routinely receive an ultrasound before insertion of the LNG-IUS. This is in line with the Dutch general practice guideline, in which this is described as an optional investigation [15]. One could argue that intracavitary pathology or a distorted uterine cavity by intramural fibroids might have been present in some of these women which could have led to a higher reintervention rate. Therefore, we performed a sensitivity analysis in which we excluded women where no ultrasound was performed (N = 10). The results of this sensitivity analysis were comparable with the results of the primary analysis. Another limitation of this study was the relatively small number of women included in the study, by which the number of potential prognostic factors included in the model was limited and some factors were possibly not found to be significant predictors due to a lack of power alone. To conduct a sample size estimation a priori, or power calculation post-hoc however was not feasible. Power in a prognostic model is driven by the number of prognostic factors in the model, the incidence of the outcome, and the number of observations [16]. A larger sample size can therefore always be considered preferable and a minimum required sample size is not straight forward. Simulation of different scenarios for sample size estimation as proposed by van Smeden et al. was not conducted [17].

For some baseline characteristics data was missing, most importantly BMI and PBAC-score. Multiple regression imputation was used to prevent a complete case analysis. Regression imputation is a commonly used and reliable method to impute missing data and preferred over complete case analysis. We combined the results of the RCT and the prospective cohort arm of the MIRA trial which caused methodological heterogeneity. Given, however, that both the RCT and observational data collection were conducted under the same protocol, we consider the effect of this heterogeneity small. Moreover, discontinuation rates between women randomized and women followed in the observational cohort were comparable and to rule out a small influence which we were not able to objectify, we added study arm to the multivariable analysis.

Outcome predictors of LNG-IUS treatment in women with HMB are not extensively researched. Six studies were found which investigated potential prognostic factors [6, 9‐13]. These studies used different outcome measures including quality of life, menstrual bleeding pattern, discontinuation of care and presence of HMB at follow-up or receiving a hysterectomy. Age was investigated as a prognostic factor for failure in three of these studies [6, 9, 12]. It was measured as a continuous factor or as dichotomous factor (≤ 35 or > 35 years) and was not found to be a risk factor for failure. A cut-off value of 45 was chosen in our study because this value was found to be associated with failure in endometrial ablation, another frequently applied HMB treatment, in previous research [18]. From a clinical perspective, the lower risk of failure of women who are at least 45 years old might be explained because these women approach (or already reached) menopause. Therefore, they might be more inclined to wait until their menstrual bleeding decreases spontaneously. Younger women have to bridge a longer period which could be the reason why they faster consider to remove the LNG-IUS and opt for another intervention.

Presence of dysmenorrhea in general was not shown to be associated with failure, though women with severe dysmenorrhea were found to have a higher failure risk. For clinical practice this implicates physicians should question women about the severity of their dysmenorrhea symptoms as this influences the chance of successful treatment. Two previous studies investigated if dysmenorrhea was associated with a higher failure risk. One study found no association [6] and the other study found abdominal and back pain to be related to the risk of receiving a hysterectomy [12]. In the present study the cause of dysmenorrhea was not investigated. Adenomyosis and endometriosis have been found to be important causes of dysmenorrhea. In women in which one of these disorders are present, the LNG-IUS has been shown to reduce dysmenorrhea complaints [19, 20]. Therefore, the LNG-IUS still seems to be a good treatment option in women with both HMB and dysmenorrhea. Future research with a special focus on women with combined HMB and dysmenorrhea is necessary to determine which treatment has the highest chance of success.

The presence of menorrhagia as predictor for LNG-IUS failure was investigated in two studies [9, 13]. In both studies it was found that women without severe bleeding had a lower risk of failure. In our study only women with HMB (PBAC-score above 150 points) were included and a higher PBAC-score (≥ 425 points) was not found to be a risk factor for failure. One study investigated if BMI affected the success chance of LNG-IUS treatment and showed the LNG-IUS to be less beneficial in women with a BMI of 25 kg/m² or less [6]. However, this was explained by the greater efficacy of usual medical treatment in leaner women in the control group. Consistent with the results in our study, no significant influence of BMI was found in the LNG-IUS group itself [6]. Two studies investigated the presence of myomas as risk factors for failure, of which one study found women with a myoma of 2.5 cm or larger had a higher risk of failure [10], and the other study found no influence [11]. In our study women with intracavitary or large intramural myomas were excluded because we expected the LNG-IUS and the compared treatment (endometrial ablation) to be less effective in these women. Therefore, we could not investigate the influence of myomas in the present study. In one study an association was found between depression symptoms and receiving a hysterectomy during LNG-IUS treatment (adjusted OR of 3.30) [12]. We did not measure depression symptoms in our study and no other studies were found which showed this association.

Overall, the results of this study can be used to inform women who are below 45 years old and or have severe dysmenorrhea about the higher risk of needing an additional intervention. More research, in a larger study population with specific attention to underlying pathology may identify additional prognostic factors for LNG-IUS failure. In other frequently applied HMB treatments as endometrial ablation, younger age and dysmenorrhea were also found to be risk factors for receiving a reintervention [18]. Therefore, these factors will not be a distinguishing factor to either advise or discourage the use of LNG-IUS treatment. Based upon the available evidence counselling should not be specified to differences in patient characteristics but should instead rather focus on the expectation of women and the specific features of the available treatment options in order for women to make a well-informed decision. Future research comparing differences between prognostic factors for the different treatment options for HMB might be helpful to personalize counselling of women with HMB.