Case Fatality Rates for Patients with COVID-19 Requiring Invasive Mechanical Ventilation. A Meta-analysis.

Rationale: Initial reports of case fatality rates (CFRs) among adults with coronavirus disease (COVID-19) receiving invasive mechanical ventilation (IMV) are highly variable.
Objectives: To examine the CFR of patients with COVID-19 receiving IMV.
Methods: Two authors independently searched PubMed, Embase, medRxiv, bioRxiv, the COVID-19 living systematic review, and national registry databases. The primary outcome was the "reported CFR" for patients with confirmed COVID-19 requiring IMV. "Definitive hospital CFR" for patients with outcomes at hospital discharge was also investigated. Finally, CFR was analyzed by patient age, geographic region, and study quality on the basis of the Newcastle-Ottawa Scale.Measurements and
Results: Sixty-nine studies were included, describing 57,420 adult patients with COVID-19 who received IMV. Overall reported CFR was estimated as 45% (95% confidence interval [CI], 39-52%). Fifty-four of 69 studies stated whether hospital outcomes were available but provided a definitive hospital outcome on only 13,120 (22.8%) of the total IMV patient population. Among studies in which age-stratified CFR was available, pooled CFR estimates ranged from 47.9% (95% CI, 46.4-49.4%) in younger patients (age ≤40 yr) to 84.4% (95% CI, 83.3-85.4%) in older patients (age >80 yr). CFR was also higher in early COVID-19 epicenters. Overall heterogeneity is high (I2 >90%), with nonsignificant Egger's regression test suggesting no publication bias.Conclusions: Almost half of patients with COVID-19 receiving IMV died based on the reported CFR, but variable CFR reporting methods resulted in a wide range of CFRs between studies. The reported CFR was higher in older patients and in early pandemic epicenters, which may be influenced by limited ICU resources. Reporting of definitive outcomes on all patients would facilitate comparisons between studies.Systematic review registered with PROSPERO (CRD42020186997).

At a Glance Commentary

Scientific Knowledge on the Subject

Outcome data for patients with severe coronavirus disease (COVID-19) receiving invasive mechanical ventilation have varied substantially. Globally, the case fatality rate (CFR) for patients with COVID-19 admitted to the ICU and receiving invasive mechanical ventilation is high, but overall estimates informed by available studies are lacking.

What this Study Adds to the Field

Of 57,420 adult patients in 69 studies who met the inclusion criteria for this systematic review and meta-analysis of patients with severe COVID-19, the overall estimate for the reported CFR was 45% (95% confidence interval, 38–52%). Definitive hospital outcomes were only available for 13,120 (36.6%) patients. Significant variability in CFR was also present by age of patients and geographic location of the study.

The novel coronavirus disease (COVID-19) pandemic, which is caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), has severely burdened healthcare system capacities in many parts of the world (1). The World Health Organization reports the global crude mortality rate to be 3.9% (2).

The care of critically ill patients with COVID-19 has been rapidly evolving (3). Although there have been promising therapies such as remdesivir (4) and dexamethasone (5), mechanical ventilation continues to be the mainstay of management of severe COVID-19 (6). Hypoxemia (PaO <60 mm Hg) has been commonly reported in hospitalized patients with COVID-19 (7). Early invasive mechanical ventilation (IMV) was promoted early in the pandemic because of concerns of aerosol generation from noninvasive oxygenation therapies facilitating nosocomial viral transmission (8–10).

The case fatality rate (CFR) is defined as the proportion of a population with a disease that dies during a specific period (11). The reported CFRs of critically ill patients with COVID-19 receiving IMV have been observed to be highly variable (12). Causes of this inconsistency likely include the heterogeneity in the management of these patients and in the presentation of outcome data (12, 13). Addressing this knowledge gap will assist in intensive care resource planning and public health strategies.

The aim of this systematic review and meta-analysis was to report the CFR of critically ill adult patients with COVID-19 who received IMV based on the available evidence. The variability in CFR by patient age, geographic region, and study quality was also analyzed in this study.

Methods

This systematic review and meta-analysis was reported using the preferred reporting items for systematic reviews and meta-analyses framework (14) and has been registered on PROSPERO (CRD42020186997). The majority of patients receiving IMV are admitted to the ICU; however, not all ICU patients receive IMV. We therefore included studies explicitly reporting on patients receiving IMV to limit heterogeneity in illness severity. The review process is illustrated in a flow diagram (Figure 1).

Figure 1.

Preferred Reporting Items for Systematic Reviews and Meta-Analyses flowchart of study inclusions and exclusions. Adapted from Reference 14.

Eligibility Criteria

Only studies reporting on consecutive adult patients (≥18 yr of age) with laboratory-confirmed COVID-19 receiving IMV were included. Studies were excluded if 1) the sample size of the cohort was less than 10, 2) they did not report the results of original research, or 3) the cohort consisted only of deceased patients. Studies were also excluded if a significant overlap in patient cohorts was identified.

Search Strategy, Information Sources, and Study Selection

Two authors (Z.J.L. and A.S.) independently searched on the publicly available COVID-19 living systematic review. This dynamic systematic review contains a daily updated list of preprint and published articles relating to COVID-19 obtained from PubMed, EMBASE, medRxiv, and bioRxiv (15). The workflow for obtaining these articles is freely available and has been used previously during the Zika virus epidemic (16). This living platform has been recently validated against an Ovid search relating to COVID-19 (17). Two authors (Z.J.L. and M.P.R.) independently extracted the content of this living systematic review and national registry databases between January 1, 2020, and July 8, 2020. Conflicts in data extraction were resolved by discussion between the reviewers or adjudication by a third author (A.S.). Corresponding authors for all the selected papers were contacted by e-mail for outcome data for patients who were still in the hospital at the time the manuscript was published. The search terms “mortality,” “fatality,” “ICU,” “characteristic,” “invasive,” “mechanical,” “ventilation,” “death,” and “died” were used within the title and abstract columns of the systematic review list. The searching criteria were combined with the Boolean operator “OR.” All studies, including preprint and non-English language articles, were considered. A separate search for COVID-19 national registries was also conducted. Study period and location were analyzed as part of the data collection process.

Definitions

Reported CFR

“Reported CFR” was defined as the CFR among all patients who received IMV, before accounting for patients who were still receiving care in hospital.

Range of estimates for CFR

We also provided a sensitivity analysis of the best possible and worse possible CFRs, assuming all remaining hospitalized patients either lived (lowest possible) or died (highest possible) in the subset of studies that reported the number of patients who received IMV who were still hospitalized at the time of study conclusion.

Definitive CFR

We examined the number of patients receiving IMV who died divided by the number of patients with a known hospital outcome (died or discharged alive) to calculate the definitive CFR.

Quality Assessment and Risk of Bias in Individual Studies

The Newcastle-Ottawa Scale (NOS) is a quality assessment tool used to evaluate nonrandomized studies on the basis of an eight-item score divided into three domains. The NOS has been selected for the purpose of this study because these domains assess selection, comparability, and ascertainment of the outcome of interest. The NOS is the most suitable for the purpose of comparing both reported and definitive CFR values. The NOS was used by the two reviewers (Z.J.L. and U.K.) to independently evaluate the quality of included studies and assess for risk of bias (18). The same set of decision rules was used by each reviewer to score the studies. Any discrepancies from the NOS were reviewed and resolved by two additional authors (A.S. and M.P.R.).

Study Outcomes

The primary outcome was the reported CFR for patients with COVID-19 receiving IMV based on the published studies. However, multiple methods of reporting CFR existed across different studies. Studies have reported the CFR of patients receiving IMV out of all patients receiving IMV, including those still hospitalized, whereas other studies have reported the CFR among patients who have completed their hospital course. This variance in reporting methods therefore resulted in variance in the CFRs reported by authors. As a secondary outcome, we examined the “definitive hospital CFR” for the subgroup of studies for whom we were able to ascertain hospital discharge outcomes. For all studies, we also present a sensitivity analysis that includes all patients showing “lowest possible” CFR for each study (assuming all patients still hospitalized lived) and a “highest possible” CFR (assuming all patients still hospitalized died). Within the appendix, the definitive hospital CFR is calculated by excluding patients who were still hospitalized to report the CFR only among patients with a known hospital outcome. Studies were also stratified based on geographical location (continent), economy (based on United Nations classification 2020), mean age, and study quality.

Data Analysis and Data Collection Process

Statistical analyzes were performed using the statistical software package Stata, version 16.1 (StataCorp). Mean and SD were used for numerical data and proportion was used for categorical data. The random-effects model and the Hartung-Knapp-Sidik-Jonkman method for meta-analysis (19) were used for the pooled prevalence of CFR because these demonstrate better properties in the presence of heterogeneity, accounting for both within-study and between-study variances (20). Results were presented in forest plots. Heterogeneity was tested by using the χ2 test on Cochran’s Q statistic, which was calculated by means of H and I2 indices. The I2 index estimates the percentage of total variation across studies on the basis of true between-study differences rather than on chance. Conventionally, I2 values of 0–25% indicate low heterogeneity, values of 26–75% indicate moderate heterogeneity, and values of 76–100% indicate substantial heterogeneity. Authors conducted subgroup analyzes to identify the possible causes of substantial heterogeneity (21). Univariable metaregression was used, symmetry of the funnel plots was evaluated, and the Egger’s regression test was used to examine for publication bias (22). Confidence interval (CI) was used to evaluate whether differences in CFRs were statistically significant. The 95% CI of prevalence including 0.0% and 100% were calculated using the standard equation (23). As prevalence cannot fall below 0% or above 100%, the CI is trimmed at 0% and 100% (20).

Additional Analyses

We also examined the reported CFR based on age stratification for the subset of studies that reported outcomes by patient age. In addition, we compared the CFRs in studies from different geographic regions and examined difference between reports from cities with an early and dramatic pandemic outbreak, such as Hubei, China, and New York, United States, compared with studies from other cities in the same country.

Results

A total of 5,322 studies were obtained from the living systematic review with 662 unique studies assessed for eligibility via full-text screening (Figure 1). Sixty-nine studies across 23 countries with reported CFRs were included in the final analysis (13, 24–91), including publicly available national registry data from seven countries (29, 56, 59, 65, 66, 80, 90). A summary of the reported CFRs for adult patients receiving IMV is outlined in Table 1. A total of 121,009 patients with confirmed COVID-19 were reported across 69 studies, with 89,405 patients (73.9%) from national registry data. Across 69 studies, 66,900 patients were male (55.3%). The patients’ mean age, as derived by the estimation formula to convert median to mean values (92), was 59.9 years. IMV was administered to 57,420 patients. Fifty-four of the 69 studies reported on the number of patients receiving IMV still hospitalized at the time of study conclusion.

Table 1.

The 69 Studies Selected for the Systematic Review and Meta-analysis

Study	Location of Study		Sample Size (N)	Mean Age (yr)	Sex, M (n)	Received IMV (n)	IMV Patients Still Receiving Care (n)	Died after IMV (n)	IMV Patients with Definitive Hospital Outcome [n (%)]	Primary Outcome: CFR of Patients Requiring IMV by Reported Outcome [% (95% CI)]
Chen et al., May 2020 (24)	Hubei, China		135	NR	78	9	0	6	9 (100)	67 (40–93)
Hu et al., May 2020 (25)	Hubei, China		323	59.0	166	34	NR	31	NR	91 (80–100)
Hu and Li, May 2020 (26)	Hubei, China		105	58.2	66	67	NR	39	NR	58 (47–70)
Hua et al., June 2020 (27)	Hubei, China		469	68.0	76	113	0	104	113 (100)	92 (87–97)
Huang et al., June 2020 (28)	Changsha, China		238	45.0	117	4	NR	2	NR	50 (15–85)
Japan registry, July 2020 (29)	Japan		575	NR	NR	575	67	133	508 (88)	23 (20–27)
Jung et al., May 2020 (30)	South Korea		5,179	44.6	2,295	36	NR	21	NR	58 (43–74)
Liao et al., April 2020 (31)	Sichuan, China		81	51.3	66	10	7	3	3 (30)	30 (5–55)
Nasir et al., June 2020 (32)	Karachi, Pakistan		30	62.5	25	10	0	5	10 (100)	50 (24–76)
Ratanarat et al., July 2020 (33)	Bangkok, Thailand		13	58.0	8	5	0	0	5 (100)	0 (0–27)
Ruan et al., March 2020 (34)	Hubei, China		150	NR	102	25	0	25	25 (100)	100 (91–100)
Shi et al., June 2020 (35)	Hubei, China		671	61.7	322	36	NR	29	NR	81 (68–93)
Sirivongrangson et al., June 2020 (36)	Bangkok, Thailand		19	52.0	15	10	2	0	8 (80)	0 (0–18)
Wang et al., April 2020 (37)	Anhui, China		125	38.8	71	4	3	0	1 (25)	0 (0–30)
Wang, June 2020 (38)	Nationwide, China		141	63.0	99	50	25	25	25 (50)	50 (37–63)
Wang, March 2020 (39)	Hubei, China		18	70.4	10	18	12	5	6 (33)	28 (8–47)
Yang et al., May 2020 (40)	Hubei, China		59	66.1	40	59	NR	36	NR	61 (49–73)
Yang et al., May 2020 (41)	Hubei, China		52	59.7	35	22	3	19	19 (86)	86 (71–100)
Ye et al., June 2020 (42)	Zhejiang, China		856	46.0	439	29	NR	1	NR	3 (0–13)
Young et al., March 2020 (43)	Singapore		18	49.5	9	1	0	0	1 (100)	0 (0–44)
Yu et al., May 2020 (44)	Hubei, China		226	63.0	139	121	6	79	115 (95)	65 (57–74)
Zhao et al., June 2020 (45)	Henan, China		29	51.2	14	5	0	1	5 (100)	20 (0–51)
Zheng et al., May 2020 (46)	Hangzhou, China		34	66.7	23	15	13	0	2 (13)	0 (0–14)
Zhu et al., June 2020 (47)	Hubei, China		102	65.2	59	29	0	25	29 (100)	86 (73–99)
Almazeedi et al., May 2020 (48)	South Surra, Kuwait		1,096	41	888	31	16	13	15 (48)	42 (26–58)
Goshayeshi et al., May 2020 (49)	Mashhad, Iran		1,067	56.9	663	231	32	81	199 (86)	35 (29–41)
Khamis et al., July 2020 (50)	Oman		63	48.0	53	16	NR	5	NR	31 (10–52)
Rinott et al., June 2020 (51)	Israel		403	44.0	220	17	NR	12	NR	71 (50–91)
Shahriarirad et al., June 2020 (52)	South Iran, Iran		113	53.8	71	2	0	2	2 (100)	100 (62–100)
Alfano et al., June 2020 (53)		Modena, Italy	307	65.2	219	53	14	17	39 (74)	32 (20–44)
Busetto et al., May 2020 (54)		Veneto, Italy	92	70.5	57	9	0	0	9 (100)	0 (0–20)
Ceruti et al., May 2020 (55)		Lugano, Switzerland	41	64.0	35	34	4	7	30 (88)	21 (7–34)
France registry, June 2020 (56)		France	4,007	65.0	2,925	2,357	NR	480	NR	20 (19–22)
Giacomelli et al., May 2020 (57)		Lombardy, Italy	233	61.0	72	8	0	7	8 (100)	88 (63–100)
Grasselli et al., April 2020 (58)		Lombardy, Italy	1,591	63.0	1,304	1,150	NR	329	NR	29 (26–31)
ICNARC, 10 July 2020 (59)		United Kingdom	10,421	58.8	320	7,185	426	3,479	6,759 (94)	48 (47–50)
Israelsen et al., May 2020 (60)		Hvidovre, Denmark	175	69.0	85	27	8	17	19 (70)	63 (46–80)
Pavoni et al., May, 2020 (61)		Tuscany, Italy	40	61.0	24	4	1	3	3 (75)	75 (41–100)
Pedersen et al., April 2020 (62)		Zealand, Denmark	17	69.8	12	17	6	7	11 (65)	41 (20–62)
Piano et al., June 2020 (63)		Northern Italy, Italy	584	66.0	357	62	10	18	52 (84)	29 (18–40)
Regina et al., May 2020 (64)		Lausanne, Switzerland	200	66.0	120	38	NR	11	NR	29 (15–43)
Spain registry, July 2020 (65)		Spain	7,695	60.3	5,344	3,867	NR	1,943	NR	50 (49–52)
Sweden registry July, 2020 (66)		Sweden	3,437	59.2	2,530	2,412	58	455	2,354 (98)	19 (17–20)
Zangrillo et al., April 2020 (67)		Lombardy, Italy	73	61.3	61	73	33	17	40 (55)	23 (14–33)
Aggarwal et al., May 2020 (68)		Iowa, United States	16	65.5	12	5	0	0	5 (100)	0 (0–27)
Arentz et al., March 2020 (69)		Washington/Seattle, United States	21	70	11	15	3	10	12 (80)	67 (45–88)
Argenziano et al., May 2020 (70)		New York, United States	1,000	62.7	596	233	86	111	147 (63)	48 (41–54)
Auld et al., May 2,020 (71)		Georgia, United States	217	63.7	119	165	11	56	154 (93)	34 (27–41)
Bhatraju et al., March 2020 (72)		Washington/Seattle, United States	24	64.0	15	18	3	9	15 (83)	50 (29–71)
Buckner et al., May 2020 (73)		Washington/Seattle, United States	105	64.5	53	19	0	10	19 (100)	53 (32–73)
Ferguson et al., July 2020 (74)		California, United States	72	58.1	38	13	4	3	9 (69)	23 (1–45)
Garibaldi et al., May 2020 (75)		Maryland, United States	832	60.4	443	70	24	24	46 (66)	34 (23–45)
Gold et al., May 2020 (76)		Georgia, United States	305	58.8	151	92	6	38	86 (93)	41 (31–51)
Goyal et al., April 2020 (77)		New York, United States	393	61.5	238	130	88	19	42 (32)	15 (8–21)
Klang et al., May 2020 (78)		New York, United States	3,406	NR	1,961	809	0	682	809 (100)	84 (82–87)
Mani et al., June 2020 (79)		New York, United States	184	64.7	111	30	17	13	13 (43)	43 (27–60)
Mexico registry, July 2020 (80)		Mexico	6,898	57.3	4,665	6,898	NR	4,724	NR	68 (67–70)
Mitra et al., June 2020 (81)		Vancouver, Canada	117	68.0	79	74	25	15	49 (66)	20 (11–29)
Palaiodimos et al., July 2020 (82)		New York, United States	200	62.5	98	42	0	32	42 (100)	76 (63–89)
Petrilli et al., May 2020 (83)		New York, United States	2,721	62.7	1,678	647	86	391	561 (87)	60 (57–64)
Reyes Gil et al., May 2020 (84)		New York, United States	217	NR	126	55	0	45	55 (100)	82 (72–92)
Richardson et al., April 2020 (85)		New York, United States	5,700	60.9	3,437	1,151	831	282	320 (28)	25 (22–27)
Salacup et al., July 2020 (86)		Philadelphia, United States	242	66.0	123	54	0	38	54 (100)	70 (58–82)
Shekhar et al., May 2020 (87)		New Mexico, United States	50	54.0	23	22	6	12	16 (73)	55 (35–74)
Shi et al., July 2020 (88)		Michigan, United States	172	61.5	97	61	2	16	59 (97)	26 (15–37)
Suleyman et al., June 2020 (89)		Michigan, United States	355	57.5	204	114	6	91	108 (95)	80 (45–100)
Ziehr et al., June 2020 (13)		Massachusetts, United States	66	56.5	43	66	0	11	66 (100)	17 (8–26)
Brazil registry, July 2020 (90)		Brazil	56,372	68.2	32,940	27,748	NR	19,935	NR	72 (71–72)
Olivares et al., June 2020 (91)		Valdivia, Chile	21	58.9	5	9	0	2	9 (100)	22 (0–47)

Definition of abbreviations: CFR = case fatality rate; CI = confidence interval; ICNARC = Intensive Care National Audit and Research Centre; IMV = invasive mechanical ventilation; NR = not reported.

Primary Outcome: Reported CFR of Patients with Severe COVID-19 Receiving IMV

The reported CFR across these studies was calculated at 45% (95% CI, 39–52%). Although a high heterogeneity was observed across all studies (I2 = 99.52%), our Egger’s regression test for publication bias was 0.43 (nonsignificant). High heterogeneity was observed when studies were analyzed by continent (I2 >90%). The reported CFRs varied between 36% (95% CI, 24–48%) and 52% (95% CI, 19–85%) among different continents, with no significant difference in CFRs. The forest plot is illustrated in Figure 2. Individual study NOS score is illustrated in Table E1 in the online supplement. There was no significant difference in CFR when studies were analyzed based on NOS score (Figure E1).

Figure 2.

Forest plot of the reported case fatality rates (N = 69 studies) for patients receiving invasive mechanical ventilation stratified by continent. CFR = case fatality rate; CI = confidence interval; ICNARC = Intensive Care National Audit and Research Centre; REML = restricted maximum likelihood.

Range of Estimates for CFR

Fifty-four studies reported on the number of patients who were still hospitalized at the time of publication. Across these 54 studies, 15,064 of 35,880 patients (42.0%) received IMV. The sensitivity analysis comparing the “lowest possible” CFR (assuming all patients still hospitalized lived) with the “highest possible” CFR (assuming all patients still hospitalized died) ranged from 43% (95% CI, 36–51%) to 64% (95% CI, 56–72%) (Table E2).

Definitive CFR

A total of 13,120 of 15,064 (87.1%) patients (22.8% of the total IMV cohort) completed their hospital stay. Among these patients, 6,463 of 13,120 patients died (49.5%). The adjusted CFR among these patients was 56% (95% CI, 47–65%) (Figure E2). Within this subset of patients, no statistically significant differences in definitive hospital CFRs were observed when analyzing studies by geographical location (continent), economy, mean age (studies with main age >70 yr had a statistically lower CFR; however, the number of patients who received IMV was small [N = 10]), or study quality (Figures E2–E5). Heterogeneity continued to remain high (I2 >90%) across all analyses.

Analysis of CFR Based on Patient Age and Studies from Early COVID-19 Epicenters

Three studies and three national registries (39, 44, 58, 59, 80, 90) reported on 42,618 IMV patients, of whom 28,547 (67.0%) died, and stratified CFR by age. CFR was >70% among patients aged more than 60 years of age. CFR increased exponentially (y = 0.429e0.1162) with increasing age (Figure 3).

Figure 3.

Reported case fatality rates for patients receiving invasive mechanical ventilation stratified by age, reported in six studies. *Age stratification for ICNARC was 16–39, 40–49, 50–59, 60–69, 70–79, and ≥80. CFR = case fatality rate; CI = confidence interval; Expon. = exponential; ICNARC = Intensive Care National Audit and Research Centre; IMV = invasive mechanical ventilation.

The analysis comparing CFR in Wuhan with that of studies from other regions of China, as well as New York versus other regions in the United States, is illustrated in Figures E6 and E7. The reported CFR across 17 studies (encompassing 640 patients receiving IMV) from China reported an overall CFR of 56% (95% CI, 39–74%). Studies from Wuhan reported a significantly higher CFR of 75% (95% CI, 63–87%) compared with studies from other regions of China (20%; 95% CI 0–45%). Among patients with a known hospital outcome (N = 11 studies), the CFR reported from Wuhan (87%; 95% CI 77–97%) was lower than the CFR reported from other regions in China (33%; 95% CI, 0–82%).

An overall reported CFR of 47% (95% CI, 36–57%) was reported across 21 studies encompassing 3,811 patients with COVID-19 receiving IMV in the United States. Studies from New York reported a CFR of 54% (95% CI, 36–72%) whereas other regions in the United States reported a CFR of 41% (95% CI, 30–53%). When considering definitive outcomes, the overall CFR across 21 studies from the United States was 61% (95% CI 50–72%), with eight studies from New York reporting a significantly higher CFR of 78% (95% CI, 68–88%) compared with other regions in United States (49%; 95% CI, 35–63%).

Univariate and Multivariate Analysis

A simple regression (univariate) analysis and multivariate regression analysis were conducted across the 46 studies with definitive hospital outcome (Table E3). Studies were analyzed by common variables, including geographical location (continent), study quality (NOS score), mean age, and economic status. Poor-quality studies reported significantly lower CFRs compared with good-quality studies (P = 0.035). Multivariate regression did not yield any further statistical significance in study quality. A univariate analysis of studies from earlier epicenters (Wuhan and New York) showed significantly higher CFRs within these epicenters compared with nonepicenter studies in the same country (P = 0.010 for Wuhan vs. other studies in China and P = 0.002 for New York vs. other studies in the United States).

Discussion

This is a large international systematic review and meta-analysis to examine global reports of CFRs for adult patients with COVID-19 receiving IMV. The reported CFR was 45% across all 69 studies, but this included patients still in the hospital. Among all 54 studies, lowest possible to best possible hospital CFR ranged from 43% to 67%. Among patients with a known hospital outcome, the definitive hospital CFR was 56%. We observed no statistical difference between continents. Older patients had a higher CFR, and the CFR was higher in the early COVID-19 epicenters of Wuhan and New York compared with that of other studies from the same country.

The CFR observed in this review of patients with COVID-19 is similar to that of previous outbreaks of severe respiratory infections. Studies from SARS-CoV in 2003 reported a CFR of 45–48% in patients receiving IMV (93, 94), and more recent studies from the Middle East respiratory syndrome reported a 60–74% CFR in critically ill patients (95, 96). In contrast, the CFR is lower in critically ill patients suffering from H1N1 influenza A, in which the CFR of patients receiving IMV was 24.2–26.5% (97). The reported CFR from severe acute respiratory distress syndrome before COVID-19 was lower at 45% (98, 99) when compared with the definitive CFR from COVID-19.

The CFR of patients receiving IMV among studies from Wuhan and New York was significantly higher than that of studies from other regions in China and the United States, respectively. This finding may reflect of the significant challenges faced in the initial stages of the COVID-19 outbreak (100, 101). Reports suggest that prone positioning was infrequent in the initial phase (41), with one Wuhan study reporting only 12% of patients receiving IMV were managed with prone positioning. Variable provider:patient ratios may also have contributed to higher CFR (102–104).

Several factors may account for the large variance in CFRs between studies. ICUs outside of outbreak epicenters may have had the opportunity of time to obtain equipment and consolidate resources before the pandemic (71). This has enabled ICUs to continue at standard patient:provider ratios (71). Closer monitoring and early intensive care for critically ill patients potentially improved patient prognosis (31). Differences in hospital facilities, patient preferences (for which limitations of care may have been in place), and indications for IMV may have also influenced the CFR (12). Finally, the change in triage process considering comorbidities, age, and frailty status in allocating ICU beds and ventilators during the pandemic may have contributed to a lower CFR among patients receiving IMV, in which younger and less frail patients were prioritized for IMV and ICU care (105–108), whereas older and frailer patients were less likely to receive ventilatory support. These older patients potentially died without IMV support, which is not captured in our findings. If significant numbers of older patients died without receiving IMV support that was indicated and desired, this would suggest our CFR estimates for older patients may be low.

Despite stratifying studies on the basis of location and NOS score, high heterogeneity continued to exist across our meta-analysis. This has been reported in other meta-analyzes studying COVID-19 mortality (109–113). Heterogeneity was the lowest at 83.4% among definitive outcomes from Wuhan (Figure E6). Although the reasons for this are not clear, we believe that studies originating from the same geographical location may provide a less heterogeneous cohort, and hence, the I2 value was lower. Other potential factors influencing heterogeneity could be differences in illness severity, thresholds for IMV, admission criteria to the ICU, and regional differences in ICU care.

As demonstrated in a recent editorial, the CFR is substantially higher among older patients, with more than 70% of patients over 60 years of age receiving IMV dying (12). It has also been reported that the CFR for patients in their 80s and 90s receiving IMV with comorbidities has been higher (114). Our findings also suggest that older patients receiving IMV had significantly higher mortality.

There are several limitations to this systematic review. First, most of the included studies had very small numbers of patients; only 17 of 69 studies reported on more than 100 patients receiving IMV. Given the available evidence, we conducted a meta-analysis to account for this variability in sample size. Second, multiple studies may have covered similar patient cohorts. However, each study’s time period, hospital, and location were considered in the final inclusion of studies to minimize overlap in patient cohorts. Third, 14 studies were not peer reviewed, as they were prepublication articles. However, these studies still provided meaningful data on the CFR of the subgroup of patients with COVID-19 who receive IMV. Fourth, the overall heterogeneity was very high (I2 >90%), which may preclude a valid conclusion from pooled results. Although we performed various sensitivity and metaregression analyses, the heterogeneity could not be minimized. This is most likely due to the case mix and the structure of age within included populations. Finally, we were unable to examine the influence of timing in the pandemic because timing and region were highly correlated.

Conclusions

The reported CFR for existing studies of adult patients with COVID-19 receiving IMV was 45%, but many of these reports included patients still in the hospital at the time of publication. Accounting for patients still in the hospital, we found a best possible CFR of 43% and a worst possible CFR of 64%. The CFR increased exponentially in the elderly. Although CFRs did not vary between continents, higher CFRs were noted in early COVID-19 epicenters such as Wuhan and New York compared with other regions in the same country. Additional studies examining long-term CFRs beyond hospital discharge are needed.