3.4.5.1 Sensitivity analysis and exploration of heterogeneity
In the above meta-analysis, the heterogeneity test with meta-analysis results showed that some of the meta-analysis effect results are heterogeneous. Therefore, we performed a sensitivity analysis excluding studies one by one for results with high heterogeneity (I2 > 50%) to explore the source of the heterogeneity.
The RCT group showed heterogeneity in the five outcomes, including duration of hospital stay, any adverse events, serious adverse events, new use of mechanical ventilation or ECMO at baseline, and new use of noninvasive mechanical ventilation or high-flow oxygen at baseline. After performing sensitivity analysis, in the results of new use of noninvasive mechanical ventilation of high-flow oxygen at baseline, it was found that the heterogeneity decreased significantly after excluding the study by spinner et al [28]. About new use of non-invasive mechanical ventilation or high-flow oxygen at baseline [10, 28, 37], spinner et al [28] had less severe disease of patients than other two studies, and therefore disease severity may influence the need for noninvasive mechanical ventilation and high-flow oxygen distribution, leading to heterogeneity. We analyzed the direction of the effect value of the three studies [10, 28, 37], and found that the RR values of the three studies were all less than 1, indicating that the use of remdesivir can reduce the new use of noninvasive mechanical ventilation or high-flow oxygen at baseline. The sensitivity analysis of other results, unfortunately, did not find that excluding a particular study would significantly affect the heterogeneity of the meta-analysis, indicating that the results were relatively stable (Appendix 5 Figure S22-38).
There was heterogeneity in the observational study group in terms of mortality, duration of hospital stay, recovery, new use of invasive mechanical ventilation or ECMO at baseline, clinical improvement, and discharge. After sensitivity analysis, the heterogeneity of recovery result was greatly reduced after excluding the study of Carlos K. H. Wong et al [26] (Appendix 5 Figure S25). The four studies [26, 29, 45, 46] were compared, and found that the patient population in the Carlos K. H. Wong et al [26] study classified as early mild patients, while the patient populations in the other three studies [29, 45, 46] classified as severe patients. Differences in the severity of their patients' disease may lead to differences in recovery. Then, we analyzed the effect size direction of the four studies [26, 29, 45, 46], and found that the RR values of the four studies were all greater than 1, indicating that the use of remdesivir helped to improve the recovery (Fig. 3A). In the sensitivity analysis results of new use of mechanical ventilation or ECMO at baseline, the heterogeneity was found to be derived from the study of Eun-Jeong Joo et al [29] (Appendix 5 Figure S28). However, there was no significant difference in study design and subjects among the four studies included in the analysis by cross-sectional comparison [22, 27, 29, 31]. Moreover, the effect size RR values of its four studies are on both sides of 1, so the impact of the use of remdesivir on the need for mechanical ventilation or ECMO is not yet clear (Appendix 4 Figure S3). About the meta-analysis of clinical improvement, although the heterogeneity was found in the study of JM Jeetendra Kumar et al [34] (Appendix 5 Figure S31), no significant difference in study design and subjects was found among the three included studies [25, 26, 34] by horizontal comparison (Appendix 4 Figure S8). According to the effect size direction and meta-analysis results of the three studies, the use of remdesivir helps to improve the clinical improvement.
The results of sensitivity analysis of kidney injury revealed that the heterogeneity was derived from the study of Quratulain Shaikh et al. [41] (Appendix 5 Figure S33). However, by comparing the three studies [31, 39, 41], no significant difference was found in study design and subjects. Moreover, the effect value RR of the three studies is scattered on both sides of 1, with clinical heterogeneity, so the effect of remdesivir on kidney injury is not yet clear (Appendix 4 Figure S13).
The meta-analysis of the duration of hospital stay and new admission of ICU at baseline of remdesivir combined with steroids showed the heterogeneity. Through sensitivity analysis, the heterogeneity was found to originate from Thomas Benfield [53] and Toshiki Kuno [54] (Appendix 5 Figure S35-36). However, through horizontal comparison, no significant difference in study design and subjects of the five studies was found [49–51, 53, 54]. Moreover, RR values were dispersed on both sides of 1, indicating that the efficacy of Remdesivir combined with steroids is not clear (Appendix 4 Figure S16-17).
The meta-analysis results of the observational study group of remdesivir combined with tocilizumab showed high heterogeneity. It was found that the heterogeneity was derived from Sohini Sengupta research by sensitivity analysis [45] (Appendix 5 Figure S37). Comparing the three studies [56–58], they were found no significant differences in study design and subjects. The effect size RR values of the three studies were all greater than 1, indicating that the combination of remdesivir and tocilizumab would increase mortality (Appendix 4 Figure S19).
3.4.5.2 Meta-regression analysis and subgroup analysis.
Due to the small number of outcomes reported in the included RCTs. Therefore, we only perform meta-regression analysis and subgroup analysis on the mortality meta-analysis results of the observational study group [14]. In the above results, we found that remdesivir use appears to be associated with COVID-19 severity. In addition, the severity of COVID-19 is related to age [63]. Therefore, in the meta-regression, age and severity of COVID-19 patients were used as covariates for the meta-regression. The meta-regression results showed that the severity of disease in patients with COVID-19 was associated with the use of remdesivir (Table 1). Then, we used the severity of the disease as the grouping standard, and divided into mild group, severe group, and all group with both for subgroup analysis. In the severe group, the use of remdesivir significantly reduced mortality [RR = 0.57, 95%CI (0.49,0.67), P = < 0.00001] by subgroup analysis (Fig. 6). In the all group, the use of remdesivir was not shown to have an effect. These results suggest that remdesivir use reduces mortality in severe COVID-19 patients.
Table 1
Result of Meta-regression
Meta-regression
|
Number of study = 16
|
REML estimate of between-study variance
|
tau2 = .05197
|
% residual variation due to heterogeneity
|
I-squared_res = 43.42%
|
Proportion of between-study variance explained
|
Adj R-squared = 48.55%
|
Joint test for all covariates
|
Model F(2,13) = 4.10
|
With Knapp-Hartung modification
|
Prob > F = 0.0415
|
_ES
|
Coef.
|
Std. Err.
|
t
|
P>|t|
|
[95% Conf. Interval]
|
age
|
.1918155
|
.2138775
|
0.90
|
0.386
|
− .2702386 .6538697
|
severity
|
− .3706237
|
.1461122
|
-2.54
|
0.025
|
− .68628 − .0549674
|
_cons
|
.3819634
|
.4856754
|
0.79
|
0.446
|
− .6672745 1.431201
|