Study selection, characteristics and quality assessment
We identified 1592 citations through the literature search, excluded 1276 after initial title and abstract screening, and assessed the full text of 64 studies for eligibility. Another 14 studies were further removed for failing to report seroconversion (n = 6), cross-sectional studies (n = 5), reviews (n = 2), and failing to provide HIV-infected patients (n = 1) (Table S3). Finally, 50 studies with a total of 7160 patients living with HIV were included for our meta-analysis11,12,17-64, 30 studies were included for qualitative analysis of serological antibody titers (Table S4 to 6); 46 studies11,12,17-25,27-32,34-39,41-56,58-64 were included for quantitative analysis of pooled seroconversion rates; 34 studies11,12,17-20,23-25,27,31,35,36,38,39,41-46,48-51,53,55,56,59-64 were used for quantitative analysis of pooled risk ratios for seroconversion following the uncomplete, complete, or booster dose of COVID-19 vaccines between patients living with HIV and HIV-negative vaccine counterparts (Figure 1).
The main characteristics and clinical outcomes of the studies for quantitative analysis were summarized in Table 1 and Table S2. The included studies were published between 2021 and 2022. Of these studies, 21 were from Europe, 11 from Asia, 9 from North America, 3 from South Africa and 2 from South America. The studies comprised 31 prospective studies and 15 retrospective studies. 19 studies were multicenter and 27 were single-center. The number of patients living with HIV in 18 studies was above 100; the follow-up duration in 15 studies were more than 2 months; only 11 studies had adjusted for potential confounders; the patients living with HIV in 40 studies received antiretroviral therapy; the patients living with HIV in 35 studies were not infected with COVID-19 prior to vaccination. In terms of vaccination type, mRNA vaccines were used in 26 studies; adenovirus vaccines were used in 3 studies; inactivated vaccines were used in 10 studies; and another 7 studies involved two or more vaccines or other types of vaccines. Supplementary table S2 presents demographic characteristic, immunoassay and threshold for positive response. Supplementary table S7 shows the detailed risk of bias for each study, and most of studies were regarded as critical or high risk of bias.
Seroconversion rate after uncomplete, complete, and booster vaccination
Sixteen studies, 42 studies, and six studies evaluated the seroconversion rate of patients living with HIV after incomplete, complete, and booster vaccination, respectively. As shown in Figure 2A, the seroconversion rate was 75.0% (95% CI 56.4% to 89.9%) after incomplete vaccination, 89.3% (95%CI 84.2% to 93.5%) after complete vaccination, and 98.4% (95%CI 94.8% to 100%) after booster vaccination. Significant heterogeneity was seen for the pooled seroconversion rate after uncomplete vaccination (I2 > 50%, P < 0.10) (Figure S1A). The funnel plot and Egger’s test (P = 0.47) did not detect the existence of publication bias in these studies (Figure S1B). The sensitivity analysis performed by using the “leave-one-out” did not markedly change our results (Figure S1C). Also, there is significant heterogeneity for the pooled seroconversion rate after complete vaccination (I2 > 50%, P < 0.10) (Figure S2). The funnel plot and Egger’s test (P < 0.01) suggested the existence of publication bias in these studies (Figure S3A). After 10 studies were filled, the funnel plot showed the relative symmetry (Figure S3B), and Egger’s test showed no evidence of significant publication bias (P = 0.49). The pooled seroconversion rate turned to be 96.6% (95% CI, 92.6% - 99.2%) after complete vaccination. The sensitivity analysis did not significantly change our results (Figure S3C). As for the pooled seroconversion rate after booster vaccination, moderate heterogeneity was observed (I2 = 44%, P = 0.11) (Figure S4A), and the funnel plot showed the relative symmetry (Figure S4B), and Egger’s test showed no evidence of significant publication bias (P = 0.63). The results were stable after sensitivity analysis (Figure S4C).
Seroconversion compared with controls after uncomplete, complete, and booster vaccination
Ten studies, 31 studies, and three studies compared the seroconversion with HIV-negative vaccine counterparts after uncomplete, complete, and booster vaccination. As suggested in Figure 2B, the risk ratios were 0.87 (95% CI 0.77 to 0.99) after incomplete vaccination, 0.95 (95%CI 0.92 to 0.98) after complete vaccination, and 0.97 (95%CI 0.94 to 0.99) after booster vaccination. Significant heterogeneity was seen for the pooled risk ratios for seroconversion after uncomplete vaccination (I2 > 50%, P < 0.10) (Figure S5A). The funnel plot and Egger’s test (P < 0.01) suggested the existence of publication bias in these studies (Figure S5B). After 5 studies were filled, the funnel plot showed the relative symmetry (Figure S5C), and Egger’s test showed no evidence of significant publication bias (P = 0.89). The pooled risk ratios for seroconversion changed to be 1.01 (95% CI, 0.95- 1.09) after uncomplete vaccination. The sensitivity analysis performed by using the “leave-one-out” did not markedly change our results except omitting Feng’s, Netto’s or Wong’s study (Figure S5D). Moreover, there is significant heterogeneity for the pooled seroconversion rate after complete vaccination (I2 > 50%, P < 0.10) (Figure S6). The funnel plot and Egger’s test (P < 0.01) suggested the existence of publication bias in these studies (Figure S7A). After 13 studies were filled, the funnel plot showed the relative symmetry (Figure S7B), and Egger’s test showed no evidence of significant publication bias (P = 0.78). The pooled seroconversion rate turned to be 1.00 (95% CI, 0.98 – 1.03) after complete vaccination. The sensitivity analysis did not significantly change our results (Figure S7C). Besides, there was minimal heterogeneity for seroconversion after booster vaccination (I2 = 7%, P = 0.34) (Figure S8A), and the funnel plot showed the relative symmetry (Figure S8B), and Egger’s test showed no evidence of significant publication bias (P = 0.37). The results were stable after sensitivity analysis except omitting Vergori’s study (Figure S8C).
Meta-regression and subgroup analysis for seroconversion rate after complete vaccination
To examine whether the observed heterogeneity could be contributed by possible moderators for the pooled seroconversion rate after complete vaccination, univariate meta-regression was performed and suggested that study location and vaccine type were possible significant moderators (Table S8). Subgroup analyses were further performed to evaluate the potential mediators for the pooled seroconversion rate after complete vaccination (Figure 3, Figure S9-S18). Subgroup analysis according to year of publication demonstrated that the rate was lower in studies published in 2022, compared with studies published in 2021 (87.7% vs. 97.6%, P < 0.01). Subgroup analysis on basis of study location suggested that the rate was lowest in South America (59.1%), compared with Asia (73.1%), South Africa (74.7%), North America (93.9%), Europe (96.0%) (P < 0.01). Subgroup analysis stratified by vaccine type showed that the rate was lowest with inactivated vaccine (59%), compared with adenovirus vaccine (92.8%), mRNA vaccine (96.1%) or other vaccines (88.4%) (P < 0.01). There was no significant heterogeneity among all subgroup comparisons (all P > 0.05), when subgroup analyses were based on study design, source of data, sample size, follow-up duration, adjustment, antiretroviral therapy, or COVID-19 history.
Meta-regression and subgroup analysis for seroconversion compared with controls after complete vaccination
Univariate meta-regression was further performed to explore the origin of heterogeneity for seroconversion compared with controls after complete vaccination, and results showed that study location and vaccine type were also possible significant moderators (Table S9). Subgroup analyses were further performed to evaluate the potential mediators for the pooled seroconversion compared with controls after complete vaccination (Figure 4, Figure S19-S28). Subgroup analysis according to year of publication demonstrated that the risk ratio was lower in studies published in 2022, compared with studies published in 2021 (0.92 vs. 0.99, P < 0.01). Subgroup analysis on basis of source data suggested that the risk ratio was lower in single-center studies (0.93), compared with multi-center studies (0.99) (P = 0.03). Subgroup analysis stratified by vaccine type showed that the risk ratio was lowest with inactivated vaccine (0.73), compared with mRNA vaccine (0.98), adenovirus vaccine (1.03), or other vaccines (0.92) (P < 0.01). There was no significant heterogeneity among all subgroup comparisons (all P > 0.05), when subgroup analyses were based on study location, study design, sample size, follow-up duration, adjustment, antiretroviral therapy, or COVID-19 history.
Grading the quality of evidence
According to the GRADE approach, the quality of evidence was very low for seroconversion rate after uncomplete or complete vaccination, and the quality of evidence was low for overall seroconversion rate after booster vaccination (Table S10A). The quality of evidence was low for seroconversion compared with controls after uncomplete or complete vaccination, and the quality of evidence was moderate for seroconversion compared with controls after booster vaccination (Table S10B). Table S10 provided the detailed criteria to down- or up- grade the level certainty.