Inclusion, clinical characteristics and demographic data

A total of 913 medical records of patients with SLE were reviewed, 841 (92.1%) were assessed for eligibility, and 109 (11.9%) were included, with a total of 155 pregnancies Fig. 1. The median disease duration was 72 months [36; 108], and the mean age at the time of pregnancy was 28.2 years (\(\pm\)5.5); 94 (60.6%) patients had at least one comorbidity. The median number of consultations during pregnancy was 4 [2; 5]. The distribution of the ACR/97 criteria, autoantibody profile, and comorbidities are described in Table 1.

Fig. 1

Flowchart describing the number of records obtained in the eligibility assessment, inclusion, and those available for analysis

Table 1 Demographic and clinical characteristics of the 155 pregnancies in 109 patients with SLEPrepregnancy period

In the period prior to pregnancy, SLE was in remission in 91 pregnancies (61.9%), while it was active in 56 pregnancies (38.1%) with a median SLEDAI of 6 [4; 8]. Among the active cases, 39 (69.6%) had nephritis, 30 (53.5%) had cutaneous activity, 26 (46.4%) had hematological disorders, and 18 (32.1%) had arthritis. Immunosuppressants were used in 131 pregnancies (88.5%), antimalarials in 103 (69.6%), antihypertensives in 39 (26.4%), and anticoagulants in 6 (4.1%). Disease activity, medication and comparisons are shown in Table 2.

Table 2 Medications and SLE status before and during 155 pregnancies, along with comparisons between these periodsPregnancy period

In 77 (50.0%; N=154) pregnancies, patients had SLE manifestations; nephritis occurred in 54 (70.1%), cutaneous activity in 44 (57.1%), arthritis in 28 (36.6%), and hematological disorders in 23 (29.8%). The incidence of SLE activity was 29.7%; the proportion of patients with active disease increased by 12.3% compared to that in the prepregnancy period, there was also a worsening of SLEDAI (Z=-4.0; p\( < \)0.001); renal activity increased by 13.8%, cutaneous activity by 9.9%, and the others showed no significant change. There was an increase of 27.1% in the use of methyldopa, 15.8% in the use of ASA, and 6% in the use of anticoagulants. The other medications showed no significant variation. Table 2.

Postpartum period

After pregnancy, SLE was in remission in 61 patients (40.9%; N=149). Among the 88 patients (59.1%) with active disease, 58 (65.9%) had nephritis, 57 (64.7%) had cutaneous activity, 33 (37.5%) had arthritis, and 14 (15.9%) had hematological activity. There was a 20.4% increase (p\( < \)0.001; N=142) in the proportion of SLE activity cases compared to the prepregnancy period and a worsening of the SLEDAI (4 [1; 8] vs. 2 [0; 6]; Z=-4.8; N=146; p\( < \)0.001). Of the patients who were in remission before pregnancy, 25 (28.7%) showed SLE flare during pregnancy, and after delivery, this proportion increased to 47.1% (p=0.005; N=87), with an incidence of 18.4%.

Analysis of maternal and fetal/perinatal outcomes

One hundred and sixteen pregnancies (80.0%; N=145) had some maternal complication. Fifty-one pregnancies (41.5%; N=123) experienced hypertension during pregnancy. There was pregnancy loss in 35 (22.9%; N=153), preeclampsia in 26 (16.8%; N=155), SLE flare during pregnancy in 27 (18.5%; N=146), and flare after delivery in 29 patients (19.5%; N=149). In 10 pregnancies, there were no data to determine if a complication had occurred. Blood pressure data were unavailable for 6 cases (3.8%), and disease activity data were unavailable for 4 cases (2.5%).

Fetal and perinatal complications occurred in 92 pregnancies (63.9%; N=144). There were 118 live births (77.1%; N=153), of which 74 (62.7%) were term and 44 (37.3%) were premature. The median duration of pregnancy was 37 weeks [35; 39], the mean birth weight was 2772.7 grams (\(\pm\) 692.2), the median birth length was 47 cm [44.5; 49], and the median head circumference was 33 cm [32; 35]. Low birth weight for gestational age was observed in 16 cases (16.2%; N=99), and IUGR was observed in 15 cases (15.6%; N=96). There were 35 cases of fetal death (22.9%; N=153), 24 spontaneous abortions (15.6%; N=153) and 11 stillbirths (7.2%; N=153). In 11 pregnancies (7.1%), sufficient data were not available to determine if at least one fetal and/or perinatal complications had ocurred. Of these, 9 (81.8%) were related to birth weight, and in 2 cases (18.1%), no data were available at all.

Table 3 shows the comparison of the number of consultations during pregnancy, maternal age at the time of pregnancy, disease duration, and SLEDAI before and/or during pregnancy between pregnancies with and without complications. Table 4 presents the adjusted odds ratio of associations with maternal and fetal/perinatal complications for associations with \(\alpha \leq 0.25\); variables showing multicollinearity were excluded from the adjustment, as shown in Tables A1 and A2 of the appendix.

Table 3 Comparison of clinical and demographic characteristics between SLE pregnancies with or without adverse outcomesTable 4 Adjusted odds ratio of characteristics associated with adverse maternal and fetal outcomesPredictors of maternal, fetal, and perinatal complications

The logistic regression model for maternal complications included 141 (90.9%) pregnancies, had a significant reduction in the -2 log-likelihoods \((^2=31.88, p < 0.001)\), and had a satisfactory fit of the data according to the Hosmer and Lemeshow test \((^2=1.94, p=0.983)\); the accuracy was 80.9%. The ACR/97 criteria for immunological disorders 3.02 [1.11; 8.21], renal disorder 3.74 [1.25; 11.21], and SLEDAI 1.30 [1.04; 1.64] were risk factors, while the use of hydroxychloroquine before pregnancy was protective 0.23 [0.08; 0.67].

In the model analyzing fetal and perinatal complications, 102 pregnancies (65.8%) were included in the model. This model showed a significant reduction in the -2 Log-likelihoods \((^2=40.26, p < 0.001)\), and had a satisfactory fit of the data according to the Hosmer and Lemeshow test \((^210.38, p=0.168)\); the accuracy was 77.5%. The use of hydroxychloroquine before and during pregnancy was a protective factor 0.14 [0.04; 0.44], while the use of ASA 5.22 [1.33; 20.54] and SLEDAI 1.31 [1.14; 1.52] were risk factors for these events. Table 5 describes the regression parameters.

Table 5 Logistic regression models to determine risk factors for adverse events in SLE pregnanciesPost-hoc analysis of ASA association and adverse fetal and perinatal events

Considering that ASA use is a well-established protective factor for fetal complications in pregnancy Askie et al. [20], Rolnik et al. [21], and our results unexpectedly identified it as a risk factor, we conducted a post-hoc analysis. Stratification of the association between ASA use and adverse fetal events by aPL, APS, and previous thromboembolic events (PTE) revealed a higher proportion of ASA use and fetal complications among individuals without these conditions: 13 (92.9%), 14 (87.5%) and 17 (85.0%), respectively.

The difference in proportions showed high variability, with a relative error of 26% and a wide confidence interval. Logistic regression analysis further demonstrated variability, with ASA showing a standard error of 69.8%. To address this, we applied a bootstrap analysis with 10000 resamplings, resulting in coefficient convergence and corrected standard errors. Despite these adjustments, ASA remained strongly associated with adverse fetal/perinatal outcomes. Table 6.

Table 6 Logistic regression model for fetal adverse events after bootstrap adjustment

We then assessed the probability of sign error (Type-S) and exaggeration factor (Type-M) using the methodology of Gelman and Carlin [22] the effect size from Askie et al. [20]. The analysis revealed a 100% probability of a Type-S error and a Type-M error of -0.9996, yielding a corrected– of -1.091 and an OR of 0.33 [0.33; 0.34] for ASA association.

These findings should be interpreted with caution, as the 100% probability of a Type S error indeed supports the interpretation that aspirin acted as a protective factor, as previously discussed. The observed variability in the data suggests the potential influence of unmeasured confounding factors, such as treatment adherence, drug exposure, diagnostic access bias, or confounding by indication. However, such variables were not available in the our patient registry data.