The search strategy generated 7798 records. After the exclusion of duplicates, 6820 articles were screened according to the inclusion and exclusion criteria. Then 106 articles were retrieved for full-text review. Finally, 34 articles were included in this study. (Figure 1)
Characteristics of included studies
Of the 34 included studies (Table 1),2-4, 17, 20-23, 31-56 8 were on SARS,2, 20, 21, 31-35 1 was on MERS,3 and 25 were on COVID-19.4, 17, 22, 23, 36-56 A total of 3230 coronavirus-infected patients underwent LFTs, including 905 SARS patients, 73 MERS patients and 2252 COVID-19 patients. Three articles published in Chinese,21, 34, 35 one article published in Spanish,41 and all others were in English. The majority of studies were from Asia (n=18) and Europe (n=14), including 16 from China,2, 4, 21, 23, 32-38, 40, 46, 48, 52, 56 one each from Singapore20 and South Kore,3 and 14 articles from 10 European countries. The other two studies were from Canada.31, 43 The prospective cohort design2-4, 17, 23, 31, 32, 36, 38, 42-45, 51-53 was most commonly used (n=16), followed by 9 prospective follow-up studies20, 22, 33, 37, 39, 41, 49, 55, 56, 4 retrospective cohort studies40, 46-48, 3 retrospective follow-up studies21, 34, 35, and 2 cross-sectional study studies50, 54. Twenty-three studies were conducted on the entire hospital setting, 6 were on the severe/critical setting,22, 37, 44, 45, 51, 55 and 5 were on the non-severe/critical setting.17, 38, 41, 49, 50 20 studies were evaluated as "good" in the quality assessment, and the risk of bias was judged as low.2, 3, 20-23, 31-34, 40, 41, 43, 44, 46, 48, 49, 51, 53, 55 Another 14 studies had a moderate risk of bias.4, 17, 35-39, 42, 45, 47, 50, 52, 54, 56
Predicted values and damage rates of lung function on average
In total, 23 of the included studies reported the predicted value of lung function, involving seven indices.2, 3, 17, 20-22, 31, 34, 36-39, 43-47, 49-53, 56 (Table 2) The overall pooled DLCO %-predicted (79.2 (95% CI, 76.2-82.2)) of 1428 patients in 19 studies was the lowest and the only one below the normal range. (Figure 2A) The pooled predicted values of the other 6 indices (FVC, FEV1, FEV1/FVC, TLC, RV, Kco) was 93.8 (95% CI, 91.2-96.4), 94.5 (95% CI, 91.7-97.2), 83.3 (95% CI, 81.3-85.2), 94.8 (95% CI, 91.7-97.9), 97.2 (95% CI, 90.7-103.7), and 90.2 (95% CI, 85.4-95.0), respectively. The heterogeneities of the seven indices were high, with I2 ranging from 91.1% to 94.3%. To explore the source of heterogeneity, we performed meta-regression on the virus, country, disease setting, and measurement time for the seven indices, finding that there were no statistical differences across all the studies. And, publication bias was only found on the FEV1/FVC index. (P = 0.046) (Table 2) Twenty-four studies were included in the meta-analyses to pool the damage rate of lung function in seven indices,2-4, 17, 20, 22, 23, 32-34, 36, 38-41, 43, 46, 48, 51-56 and the DLCO got the highest rate with 35.2% (95% CI, 28.7-41.8%). (Figure 2B) The rate of the other six indices varied from 6.2% to 21.5%. (Table 2) Similarly, the heterogeneities of the seven indices were high, (I2 = 79.6-96.4%) and no positive findings were found in the meta-regression. However, publication bias was the opposite that it existed in all but DLCO. (P = 0.052) (Table 2) Several studies also identified the extent of lung function damage, and the mild impairment accounted for most, with the pooled 83%-100% in FVC, FEV1, TLC, DLCO, and Kco.17, 20, 22, 23, 33, 39, 46, 56 (Table 2)
Severe/critical vs. Non-Severe/critical
A total of 10 cohort studies detailed the predicted values of lung function in two groups of severe/critical and non-severe/critical. (283 vs. 531 patients)2, 3, 17, 36, 38, 46, 47, 51-53 Differences between the two groups of the seven indices were summarized by meta-analyses. (Table 3) Predicted values in the severe/critical group were found to be worse than that in the non-severe/critical group (all P < 0.013), except for FEV1/FVC (no difference between two groups, P = 0.585). And, DLCO got the largest gap, whose lung function in severe/critical decreased -11.6 (WMD, 95% CI -14.2--9.0) compared with non-severe/critical.2, 3, 17, 38, 46, 47, 51-53 (Figure 3A) All the heterogeneities were acceptable. (I2: 0-58.5%) (Table 3) Nine cohort studies4, 17, 36, 38, 46, 47, 51-53 compared the damage rates of the seven indices between the two groups, two of which were retrospective cohort studies46, 47 and the others were prospective studies. (483 vs. 492 patients) Meta-analyses found that the risk of lung function damage of DLCO and TLC in the severe/critical group was 1.74 (RR, 95% CI 1.46-2.07) (Figure 3B) and 2.00 (RR, 95% CI 1.38-2.90) times higher than that in the non-severe/critical group, respectively. (P = 0.000) However, there were no significant differences in the other five indices. (P > 0.088) Heterogeneities were low, with I2 varied from 0 to 13.9%. (Table 3)
Follow-up Changes
Five articles reported the changes of lung function in predicted value during a one-year follow-up.2, 22, 31, 37, 50 Meta-analyses showed improvements in FVC, FEV1/FVC, and Kco. (P < 0.049) While there were no statistically significant changes in the other four indices. (P > 0.074) (Table 4) Given the high heterogeneity of most indices, we performed subgroup analyses. It showed that all indices got statistically improved in the severe/critical group, (P < 0.049) while they had no changes in the non-severe/critical group, except for Kco. (P > 0.208) (Supplementary Figure 1) Only three articles covered the damage rate changes of lung function.2, 22, 55 Analyses showed that there was no statistically significant change in any of the seven indices. (P > 0.182) (Table 4) And except that DLCO had a high heterogeneity, (I2 = 84.6%) no heterogeneity was found in other indices. (I2 = 0%) (Table 4)