2.1 Search strategy

To identify studies meeting the inclusion criteria, a literature search was conducted across PubMed, Web of Science, Excerpt Medica Database (EMBASE) and Cochrane Library. The search period lasted from inception to December 16, 2022, and no language restrictions were applied. The keywords combination we adopted was as follows: (‘COVID-19′ OR ‘SARS-CoV-2′) AND (‘reproductive techniques, assisted’) OR (‘sperm injections, intracytoplasmic’) OR (‘fertilization in vitro’) OR (‘embryo transfer’) OR (‘embryo implantation’). Furthermore, we thoroughly examined the bibliographies of the selected paper along with the conference proceedings of ESHRE and ASRM to ensure that potentially relevant records were not omitted.

2.2 Study selection

The eligibility of the searched literature was evaluated according to the inclusion criteria as follows: 1) the study was observationally designed, such as cohort; 2) the participants were infertile women who underwent ART treatment cycles; 3) the exposed group had been infected with COVID-19 in the past and now recovered, while previously uninfected individuals served as the control group; 4) the outcomes encompassed any predefined information concerning cycle characteristics, laboratory indicators, and pregnancy results. The cycle characteristics contained data on the duration of stimulation, the total dose of gonadotropin, the peak level of estradiol, and the thickness of the endometrium. The laboratory parameters comprised the number of retrieved oocytes, the number and rate of mature oocytes, the number and rate of fertilized oocytes, as well as the number and rate of good-quality embryos. The pregnancy indicators contained the rate of embryo implantation, the rate of early miscarriage, as well as biochemical, clinical and ongoing pregnancy rates.

Studies were excluded if they: 1) were cell or animal experiments, case reports or series, reviews, editorials, communications, or comments; 2) lacked original outcome data or a control group; 3) were only abstracted without full data; 4) were duplication publications in conference; 5) investigated the impact of COVID-19 infection in males; or 6) studied the effect of COVID-19 pandemics or current COVID-19 infection. Studies with larger sample sizes were given priority when containing overlapping data of the same cohort.

A two-step approach was employed for the identification of studies. Firstly, the titles and/or abstracts of the paper were examined by two independent reviewers (Yiqi Liu and Shen Chen) based on the selection criteria described above. Then, the literature was reviewed in its entirety to ensure its eligibility for inclusion. Any disagreement that arose during the selection process was discussed with the third author (Mengyi Chen) until an agreement was reached.

2.3 Data extraction and study quality assessment

Data extraction was completed by two reviewers (Yiqi Liu and Shen Chen) independently and recorded into the pre-designed Excel form in a standardized manner. To avoid potential errors, we also conducted a reciprocal check with a third reviewer (Mengyi Chen), and the inconsistency was resolved by consulting with the corresponding author (Jialyu Huang). The information gathered was as follows: name of the first author, publication year, country of study, design and period of study, sample size, participants’ age, diagnosis of SARS-CoV-2 infection, disease severity, the time interval from infection to treatment, embryo transfer (ET) strategy, and targeted outcome measures.

Given the final inclusion of cohort studies, the methodological quality was evaluated via the Newcastle-Ottawa Scale (NOS).18 Three main dimensions include cohort selection, comparability, and outcome evaluation, which could be further specified into eight aspects including the representativeness of the exposed group, selection of the control group, determination of exposure, lack of outcome at the outset, comparability of the design or analysis, evaluation of outcome, and the duration and sufficiency of follow-up. Based on the total score, studies were classified as low (0–3), medium^4-6 and high^7-9 quality. Again, any disputes arising from this process were settled through consultation with the corresponding author (Qiongfang Wu).

2.4 Statistical analysis

Review Manager (version 5.4; Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2020) was employed to perform data synthesis. Post-matching data had priority for analysis in all included studies. For continuous outcomes, data were pooled as mean differences (MDs), while odds ratios (ORs) were adopted to depict pooled effects for dichotomous data. According to Luo et al.¹⁹ and Wan et al.,²⁰ when studies were reported with the complete or partial five-number summary, the mean and standard deviation of the sample were computed from medians, ranges, or interquartile ranges (https://www.math.hkbu.edu.hk/∼tongt/papers/median2mean.html). Since the validity of the heterogeneity test might be affected by a small sample size, the random-effects model was adopted for synthesis even if low heterogeneity was detected in included studies.²¹

We estimated the heterogeneity across studies via Cochran's Q test, and the inconsistency index (I²) < 50% was considered to indicate low heterogeneity.²² Subgroup analyses were conducted for the two primary outcomes of the retrieved oocyte number and clinical pregnancy rate according to the sample size (≥ or < 100) as well as the study design (retrospective or prospective). Additionally, sensitivity analysis was used to estimate the leverage of individual studies and robustness of the overall results, by sequentially excluding one study at a time. Egger's test was completed through Stata (version 16.0; StataCorp LLC) to evaluate the publication bias. In Egger's test, p < .10 was considered to be of statistical significance,²³ while p < .05 was adopted in any other analyses.

3.1 Study selection and characteristics

Initially, a total of 1482 studies were identified, with 1480 records from database search and an additional two records from other sources by manual reviewing. After removing duplicates, 846 records remained. Subsequently, 804 records were excluded by title and/or abstract filtering, followed by 29 records through full-text assessment. Finally, a total of 15 studies were included in this systemic review and meta-analysis (Figure 1).^{14, 15, 24-36}

The main characteristics of the included studies are enumerated in Table 1. These studies were distributed in eight countries including China, Israel, the USA, Argentina, Italy, Spain, Russia and Jordan, and all were published between 2021 and 2022. Most of these studies (n = 10) employed a retrospective cohort design^{14, 25-27, 29, 30, 32-35} and five were prospective cohorts.^{15, 24, 28, 31, 36} All studies were single-centre except for the studies by Youngster et al.³⁴ and Albeitawi et al.,²⁶ whose participants were enrolled from two centres. The sample size of each study ranged between 7 and 452. The infection of SARS-CoV-2 was diagnosed in four studies by polymerase chain reaction test,^{14, 24, 33, 35} five by serum antibodies,^{27, 28, 30, 31, 36} and three using both.^{15, 29, 32} The remaining studies did not mention the details of diagnosis. The severity of COVID-19 was investigated in seven studies, among which three were asymptomatic or mild,^{29, 30, 32} two were mild,^{27, 35} one was mild or severe²⁸ and one was mild or moderate.³⁶ Most studies (n = 11) considered the interval between COVID-19 and the initiation of the post-infection cycle, except those conducted by Morris et al.,³⁰ Odeh-Natour et al.,³¹ Chico-Sordo et al.,²⁸ and Albeitawi et al.²⁶ In terms of ET strategy reported in 10 studies, five used fresh ET,^{26, 31, 32, 34, 36} three used frozen-thawed ET^{25, 30, 33} and two used both.^{24, 35} Overall, the literature included in this review had high methodological quality, as indicated by the mean NOS score of 8 (all ≥ 7) (Table S1).

3.2 Meta-analysis of cycle characteristics

The quantity of included studies and participants varied in reporting cycle characteristics, with five (n = 572) providing data on stimulation duration,^{14, 24, 26, 32, 35} five (n = 663) on gonadotropin dose,^{14, 24, 32, 34, 35} seven (n = 467) on peak estradiol level^{14, 15, 24, 27, 31, 34, 35} and six (n = 957) on endometrial thickness.^{24, 25, 31, 33-35} As shown in Figure 2, the results revealed that none of the characteristics were significantly affected. The pooled MDs for the comparison of previously infected versus uninfected cycles were 0.10 days (95% confidence interval [CI]: −0.82–0.62; p = .79; I² = 57%) for duration of stimulation, 58.83 IU (95% CI: −124.65–242.31; p = .53; I² = 0) for total dose of gonadotropin, −57.28 pg/mL (95% CI: −366.03–251.47; p = .72; I² = 21%) for peak level of estradiol, and 0.12 mm (95% CI: −0.20–0.45; p = .47; I² = 23%) for endometrial thickness.

3.3 Meta-analysis of laboratory parameters

Among the included studies, 12 reported the number of oocytes retrieved, comprising 536 infected versus 677 uninfected cycles.^{14, 15, 24, 26-29, 31, 32, 34-36} The mean oocyte number was estimated to be 10.67 versus 10.61 (MD = 0.06; 95% CI: −0.75–0.88; p = .88) after meta-analysis (Figure 3A). Six studies (n = 517) provided information on the number of fertilized oocytes,^{14, 26, 27, 31, 35, 36} and a significant increase was noticed in the infected group (MD = 0.79; 95% CI: 0.20–1.38; p = .008) (Figure 3D). The pooled results of the other laboratory parameters were comparable in the number of mature oocytes (8 studies; MD = −0.29; 95% CI: −1.65–1.06; p = .67),^{14, 15, 26-29, 31, 36} oocyte maturation rate (five studies; MD = −3.20; 95% CI: −7.43–1.04; p = .14),^{14, 15, 28, 34, 36} fertilization rate (five studies; MD = 1.97; 95% CI: −1.54–5.48; p = .27),^{14, 26, 28, 34, 36} number of good-quality embryos (four studies; MD = −0.60; 95% CI: −2.32–1.12; p = .50),^{14, 24, 35, 36} and good-quality embryo rate (three studies; MD = −14.76; 95% CI: −33.99–4.47; p = .13)^{14, 15, 36} (Figure 3B,C,E–G). Among the above seven comparisons, the heterogeneity was identified as low in most outcomes, except for the number and rate of good-quality embryos, whose I² were 70% and 77%, respectively.

3.4 Meta-analysis of pregnancy outcomes

Data on clinical pregnancy rate was provided by nine studies, including a total of 435 previously infected and 969 uninfected cycles.^{24-26, 30-34, 36} After pooled analysis, the mean OR was calculated to be 0.96 with a 95% CI of 0.74–1.24 (p = .74; I² = 0) (Figure 4B). There was also no significant impact identified on other pregnancy outcomes. In detail, the ORs were 1.10 for biochemical pregnancy (95% CI: 0.76–1.60; p = .61; I² = 0),^{24-26, 30, 32} 0.86 for embryo implantation (95% CI: 0.46–1.61; p = .64; I² = 50%),^{25, 32, 33} 2.69 for early miscarriage (95% CI: 0.85–8.52; p = .09; I² = 0),^{32, 33, 36} and 0.77 for ongoing pregnancy (95% CI: 0.42–1.43; p = .41; I² = 36%)^{25, 30, 33} (Figure 4A,C–E).

3.5 Subgroup analysis

As shown in Figure S1, the pooled result of oocyte number remained consistently comparable in retrospective (MD = −0.15; 95% CI: −1.42–1.11; p = .81) and prospective (MD = 0.20; 95% CI: −1.06–1.47; p = .75) cohort studies. Moreover, no significant change was observed in subgroups according to sample size ≥ 100 (MD = 0.22; 95% CI: −0.70–1.13; p = .64) and < 100 (MD = −0.51; 95% CI: −2.27–1.25; p = .57). Similarly, the clinical pregnancy rate remained stable in retrospective and prospective design subgroups, with pooled ORs of 0.94 (95% CI: 0.70–1.28; p = .70; I² = 0) and 1.00 (95% CI: 0.60–1.68; p = .99; I² = 0), respectively (Figure S2). Also, there was no significant alteration by different sample sizes (≥ 100: OR = 1.03; 95% CI: 0.78–1.36; p = .82; < 100: MD = 0.59; 95% CI: 0.26–1.32; p = 0.20). There was little to no heterogeneity across the meta-analyses (I² = 0%–21%).

3.6 Sensitivity analysis

The two primary outcomes of retrieved oocyte number and clinical pregnancy rate remained stable when one study was omitted at a time (Figure S3A,B). However, when we removed the paper by Odeh-Natour et al.,³¹ the correlation between previous COVID-19 infection and the number of fertilized oocytes was rendered insignificant (MD = 0.47; 95% CI: −0.28–1.22) (Figure S3C).

3.7 Publication bias

According to Egger's test, no significant publication bias was found in any of the 15 outcomes analyzed (p = .197–.976), except for the embryo implantation rate (p = .030), as shown in Table S2.