Study selection process:
We showed the search results and the selection process in Fig. 1.
Description of included trials:
After considering duplicates (182), 1044 articles were screened between 4 April and 1 November 2021 from the database searches. The full text of 81 articles was recovered, while fifteen did not have inclusion criteria, eighteen did not have data for PE or GH, and in fifteen, data was insufficient. Twenty-seven articles were eligible, but six papers were not acceptable after a detailed survey by statistician (Fig. 1). The present meta-analysis was performed on fifteen studies that their participants were pregnant women with SCH, including eighteen hypotheses (comparisons) (5117 vs. 49457 participants) on GH (14–20, 22–23, 25–28) and eleven hypotheses (2162 vs. 26401) on PE (14–17, 21–24, 26) (Table 1). It should be noted that some involved studies included two or more hypotheses with different sample sizes or other conditions of the survey. Therefore, we first referred to the number of these hypotheses (comparisons) and then mentioned preliminary studies in parentheses. Seventeen hypotheses were cohorts (14–15, 17–18, 20, 22–28), and three were RCTs (16, 19, 21). Eight hypotheses compared treated with euthyroid (15, 18, 24, 26), and two with control (19, 28), one of them compared SCH with participants who their TSH levels were equal or lesser than 2.5 without treatment, but may be TPOAb+ (28). Two hypotheses compared SCH with placebo in RCT (16) and cohort (25) studies. Eight hypotheses (14, 17, 20–23, 27) compared treated SCH with untreated SCH. The trimester of study entry in ten was first, five was second, one was third, three were first or second, and in one was first, second, or third trimesters. For GH outcome, the time to start treatment was during gestation except for one that was unknown (25) and another (27) that was before or during pregnancy; and for PE outcome, in all was during pregnancy.
Results of five studies were summarized for OH, in which ten hypotheses (1165 vs. 27829) evaluated the effect of levothyroxine on GH (16, 26, 29–30) and eight (866 vs. 24994) on PE (16, 24, 26, 30). In seven hypotheses, the trimester of study entry was first, in three was second, and in Casey et al.’ study was first or second. The time to start treatment in all was during gestation. One hypothesis was RCT (16), and others were cohorts (24, 26, 29–30). Seven compared treated women with euthyroid (24, 26, 30), two with controls (29), and one with placebo (16). One compared treated and untreated women with OH in the first trimester (29).
Participants in five studies had OH or SCH in which four hypotheses reported GH rates (17068 vs. 554508) (31, 32, 35), while five stated PE rates (17074 vs. 553716) (31, 33–35). The trimester of study entry was different among involved studies, while treatment in all except one (33) started before or during pregnancy. Four hypotheses were cohorts (31–32, 35) and two case controls (33–34) which compared treated participants with controls. Two hypotheses compared participants whose TSH levels were controlled and uncontrolled (35).
Three studies, including three hypotheses (1127 vs. 306), estimated the effect of levothyroxine on OR of GH in treated TPOAb + women versus untreated (20, 36–37). The trimester of study entry was first except one (36), which was first or second, and treatment in all started during pregnancy. Two studies were cohorts (20, 37), and one was RCT (37).
Seven studies were met-analyzed in which eight hypotheses (1633 vs. 5341) compared the ORs of GH (20, 28, 36–39), and five (404 vs. 2693) the ORs of PE (36–37, 40) in TPOAb + compared to TPOAb-. The trimester of study entry was first, second, or third in one study (39), first in four studies, first or second in other four, and finally, second trimester in one. The time of starting trimester in all except one (39) was during pregnancy. Three hypotheses compared treated TPOAb + with control TPOAb- (20, 36–37). Also, two treated TPOAb + with treated TPOAb- (38–39) and one treated and untreated TPOAb + with treated and untreated TPOAb- (28), which we performed a subgroup meta-analysis on all three (Fig. 11). Two hypotheses compared treated and placebo TPOAb + with treated and placebo TPOAb- (40). One compared untreated TPOAb + with untreated TPOAb- (28), and another untreated with control (36).
Two hypotheses (430 vs.2065) compared the OR of GH in TPOAb- with TPOAb- (20).
Quality of the included studies:
The quality of the included RCTs, cohorts, and case controls have been shown in Tables 2a, 2b, and 2c.
Bias Assessment:
We provided the risk of bias assessment in Table 2a, and deficiency cases were classified as not written (NR). In all studies, RCTs have been developed by a numerical list created by a computer system. The sequence generation was observed in all except one study (36). Allocation concealment from researchers and participants and also implementations were not observed in most studies. The blinding of participants, personnel, outcome assessor and analysis approach were not written in all studies. Baseline variables were statistically similar in all studies; also, incomplete outcomes and reporting bias were not observed in all. Quality in cohort studies was rated as good in eleven studies, and as fair in ten studies (Table 2b). Quality in case-control studies was good in two studies considering that one item (non-response rate) has not been reported (Table 2c).
Meta-analysis Results For Studies Included Pregnant Women With
SCH
Random-effects model in SCH showed that ORs of GH (18 comparisons) and PE (11 comparisons) after LT4 therapy were not significant between compared groups [OR = 1.03, 95% CI: (0.85, 1.25), I2 = 35.25%, P = 0.78, OR = 1.02, 95% CI: (0.66,1.58), I2 = 46.86%, P = 0.94, respectively] (Figs. 2, 3). For both analyses, the heterogeneity was not significant. Also, Egger’s test did not indicate any evidence of publication bias for GH outcome (P = 0.69), whereas, for PE, the publication bias was seen (P = 0.04). According to the fill and trim method, the value of the adjusted OR of PE was 1.37 with 95% CI (0.86, 2.18). Therefore, based on 11 observed hypotheses, the overall effect size may be smaller than that in the absence of publication bias. The sensitivity analysis results indicated that after excluding each study, the total effect size did not change.
Subgroup meta-analysis based on the design study to analyze its effect on the association between levothyroxine treatment and two outcomes did not demonstrate significant effects (Figs. 1–2 supplementary file). Also, the type of two compared groups (treated versus control, placebo, euthyroid, or untreated) were not significantly effective on obtained results for the two outcomes (Figs. 4–5). Furthermore, the subgroup meta-analysis based on the trimester of study entry did not show significant effects between subgroups for both outcomes (Figs. 3–4 supplementary file). For both outcomes, age, BMI, TSH levels, and TPOAb status were insignificant by meta-regression (P > 0.05).
OH
Meta-analysis indicated that the ORs of GH (10 comparisons) and PE (8 comparisons) after levothyroxine therapy were not significantly different between compared groups [OR = 1.23, 95% CI: (0.86,1.78), I2 = 34.29%, P = 0.26, OR = 1.32, 95% CI: (0.83, 2.09), I2 = 0.00%, P = 0.24, respectively] (Figs. 6–7). The heterogeneity was not significant for both models. Egger’s test demonstrated no publication bias for both models (for GH: P = 0.44 and PE: P = 0.62). The sensitivity analysis results indicated that after excluding each study, the total effect size did not change. For both outcomes, all studies were cohort except one (16) was RCT. For GH outcome, the comparison was conducted between treated versus euthyroid or control, except one (16) that compared treated versus placebo and also another (29) that compared treated versus untreated in the first trimester. For PE outcome, all comparisons were made between treated versus euthyroid except in one (16) that compared treated with placebo. Therefore, subgroup meta-analysis based on study design and compared groups were not needed. We performed a subgroup meta-analysis for both outcomes based on the trimester of study entry, considering that one study (16) was deleted due to its trimester that was first or second. For GH outcome, subgroup meta-analysis showed significant heterogeneity between two subgroups of first, also second trimesters (Fig. 5 supplementary file). The overall effect size of each group showed that in the second-trimester subgroup, the levothyroxine treatment was associated with an increased risk of GH [OR = 1.97, 95% CI: (1.34, 2.90)], whereas there was no significant effect in the first-trimester subgroup [OR = 0.95, 95% CI: (0.62, 1.47)]. The test of group differences demonstrated that the group-specific overall effect sizes were statistically different (Qb=6.04, P = 0.01). For PE outcome, the subgroup meta-analysis did not show any significant heterogeneity between subgroups (Fig. 6 supplementary file). Meta-regression did not show any significant effect of age and BMI.
Sch Or Oh
Random-effects model on four hypotheses for GH and five for PE in women with SCH or OH revealed that the ORs of GH and PE were not significantly different [OR = 1.22, 95% CI: (0.56, 2.66), I2 = 94.71%, P = 0.62, OR = 0.56, 95% CI: (0.13, 2.31), I2 = 98.13%, P = 0.42, respectively] (Figs. 8–9). For both analyses, the heterogeneity was significant. Also, Egger’s test did not indicate any publication bias (for GH: P = 0.42 and for PE: P = 0.19). The sensitivity analysis results for GH showed that after excluding each study, the total effect size did not change. However, for PE outcome, after omitting the Das, et al.’ study, the overall effect size will be significant without heterogeneity (OR = 1.36, 95% CI: (1.24, 1.49), P = 0. 00) due to the Turunen et al.’ study that has the highest effect. For GH outcome, all hypotheses were cohort. For PE, three hypotheses were cohort, and two hypotheses were case-control. Therefore, we conducted a subgroup meta-analysis based on the design study. However, there was no significant difference between the two subgroups (Fig. 7 supplementary file). The overall effect size of each group determined by subgroup meta-analysis based on the type of compared groups for GH outcome indicated that participants whose TSH levels were controlled (less than 2.5) had lesser OR of GH compared to uncontrolled [OR = 0.63, 95% CI: (0.42, 0.95)], whereas there was no significant effect for the subgroup that compared treated with controlled participants [OR = 2.17, 95% CI: (0.94, 5.04)]. The test of group differences demonstrated that the group-specific overall effect sizes were statistically different (Qb=6.68, P = 0.01) (Fig. 8 supplementary file). However, for PE outcome, there was no evidence of significant heterogeneity between the two subgroups (Fig. 9 supplementary file). For both outcomes, the trimester of study entry was different among studies, so the information was not enough for subgroup meta-analysis.
Antibody +&+
Meta-analysis on three hypotheses to estimate the OR of GH in TPOAb + women showed that levothyroxine treatment decreased the risk of GH compared with untreated (OR = 0.43, 95% CI: (0.30, 0.62), I2 = 0.00%, P = 0.00) (Fig. 10). The heterogeneity was insignificant, and publication bias was not seen (P = 0.56). The sensitivity meta-analysis indicated that after excluding Yang et al.’ study, the OR of GH will not be significant (OR = 0.65, 95% CI: (0.22, 1.91) ), I2 = 0.00%, P = 0.43), whereas after excluding two other studies (36, 37), the overall total effect will stay significant.
Two studies were cohort (20, 36), and one was RCT (37). The trimester of study entry was the first, except one that was the first or second (37), and the time to start treatment was during pregnancy for all studies. For PE, the studies were not enough to conduct a mate-analysis.
Antibody + versus -
Using eight hypotheses that compared TPOAb + women with TPOAb- women, the random-effects model revealed that the ORs of GH was not significantly different from compared groups (OR = 0.99, 95% CI: (0.8, 1.24), I2 = 0.00%, P = 0.96) (Fig. 11). Also, using five hypotheses that compared the ORs of PE, there was not seen any significant association (OR = 1.13, 95% CI: (0.67, 1.92), I2 = 0.00%, P = 0.65) (Fig. 12). For both analyses, the heterogeneity was not significant. Also, there was no publication bias for both outcomes (GH: P = 0.73 and PE: P = 0.78). The sensitivity analysis results indicated that after excluding each study, the total effect size did not change. For GH, all studies were cohort except one, which was RCT (27). For PE, two studies were cohort, and three studies were RCT. However, the subgroup meta-analysis based on the study design did not show any significant differences between the two subgroups (Fig. 10 supplementary file). The effect of the type of compared groups for both outcomes was surveyed by subgroup meta-analysis, which did not demonstrate effects between subgroups (Figs. 11–12 supplementary file).
For GH, the trimester of study entry in four hypotheses was the first, one was the second (38), two hypotheses were first or second (36), and one was the first, second, or third trimester (39). However, a considerable difference didn't show between subgroups by subgroup meta-analysis (Fig. 13 supplementary file). For PE outcome, the trimester of study entry in four hypotheses was the first or second trimester, except one which was the first. For both outcomes, the time to start treatment in all studies was during pregnancy.