Charlson comorbidity index and a composite of poor outcomes in COVID-19 patients: A systematic review and meta-analysis.

BACKGROUND AND AIMS: The ongoing COVID-19 pandemic is disproportionately affecting patients with comorbidities. Therefore, thorough comorbidities assessment can help establish risk stratification of patients with COVID-19, upon hospital admission. Charlson Comorbidity Index (CCI) is a validated, simple, and readily applicable method of estimating the risk of death from comorbid disease and has been widely used as a predictor of long-term prognosis and survival.
METHODS: We performed a systematic review and meta-analysis of CCI score and a composite of poor outcomes through several databases.
RESULTS: Compared to a CCI score of 0, a CCI score of 1-2 and CCI score of ≥3 was prognostically associated with mortality and associated with a composite of poor outcomes. Per point increase of CCI score also increased mortality risk by 16%. Moreover, a higher mean CCI score also significantly associated with mortality and disease severity.
CONCLUSION: CCI score should be utilized for risk stratifications of hospitalized COVID-19 patients.

Coronavirus disease of 2019

Charlson comorbidity index

Odd Ratios

Hazard Ratios

Preferred Reporting Items for Systematic Reviews and Meta-Analyses

Mean difference

System for Research In Primary Care

angiotensin-converting enzyme-2

Severe acute respiratory syndrome coronavirus-2

Transmembrane protease serine 2

Renin-angiotensin system

Nonsteroidal anti-inflammatory drugs

C-reactive protein

Intensive care unit

Introduction

Since the emergence of Coronavirus Disease 2019 (COVID-19) in Wuhan in late December 2019, the total of confirmed cases and deaths of this contagious respiratory disease keeps increasing worldwide. As of July 17, 2020, the World Health Organisation (WHO) has declared more than 13 million people as a positive confirmed COVID-19 case that results in more than 580.000 deaths [1]. Through descriptive observational studies, it is well established that patients with comorbidities are disproportionately affected by COVID-19 and associated with worse clinical outcomes [[2], [3], [4], [5]].

Therefore, it is crucial to have a thorough assessment of comorbidities to establish risk stratification of patients with COVID-19 upon hospital admission. Charlson Comorbidity Index (CCI) is a validated, simple, and readily applicable method of estimating the risk of death from comorbid disease and has been widely used as a predictor of long-term prognosis and survival [[6], [7], [8]]. Thus, to delineate better the advantage of using CCI for risk stratifications in COVID-19 patients, we performed a systematic review and meta-analysis aimed to assess the association between CCI and a composite of poor outcomes in COVID-19 patients.

Methods

Search and selection criteria

A systematic literature search was performed through several databases, including Pubmed, EuropePMC, EBSCOhost, Proquest, Cochrane library and two preprint servers (preprint.org and Medrxiv). The keywords used were (“Charlson Comorbidity Index” OR “CCI” OR “Charlson Index”) AND (“COVID-19" OR “SARS-CoV-2" OR “Novel Coronavirus” OR “2019-nCov”). The inclusion criteria of this study were studies of COVID-19 patients that reported any of the following: (1) odds ratios (ORs) and hazard ratios (HRs) of CCI score with a composite of poor outcomes (2) Mean CCI score for a composite of poor outcomes vs. no outcome, (3) per point HRs or ORs of CCI score and mortality. A composite of poor outcomes consists of mortality, need for critical care, severe disease presentation, mechanical ventilation. If two or more studies are consisting of the same population, we select the study that reported the most complete data regarding the inclusion criteria. We excluded: review articles, non-research letters, communications, and commentaries; studies with samples <20; case reports and small case series; non-English language articles; research in pediatric populations (17 years of age and younger). We finalized our systematic search on July 15, 2020. The search was performed by two independent researchers (JH and SL), and discrepancies were resolved by discussion with a third person (RP). This systematic search is reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA).

Data extraction

Data extraction was carried out by JH, RP, and SL using a standardized form containing the following details: author name, country, study design, number of subjects, sex, age, outcome, and CCI score types. Data of CCI score that was reported other than mean ± SD was transformed accordingly using a calculator available online, derived from Wan et al. and Luo et al. studies [[9], [10], [11]]. The risk of bias of the included studies was assessed using the Newcastle-Ottawa Score by 2 independent authors and discrepancies were resolved via discussion [12].

Statistical analysis

Review Manager 5.4 was used for the meta-analysis [13]. To characterize the association between CCI score (1–2 and ≥3) and a composite of poor outcomes, and per point CCI score and mortality, we calculated the pooled estimates and its 95% confidence interval in the form of odds ratios (ORs) and hazard ratios (HRs), respectively, using the generic inverse variance method. The CCI 0 was used as the reference of comparison. Whereas, to characterize the association between a composite of poor outcomes and mean CCI score, we calculated the pooled estimates in the form of a mean difference (MD) and its standard deviation. To account for interstudy variability regardless of the heterogeneity, a random-effects model was assigned. We used two-tailed p values with a significance set at ∼0.05. To assess heterogeneity across studies, we used the inconsistency index (I2) with a value above 50% or p < 0.10 indicates significant heterogeneity, whereas I <25% is considered low heterogeneity. Each individual component of the composite of poor outcomes was then sub-analyzed. A sensitivity analysis using the leave-one-out method was set to assess statistical robustness and detect the source of heterogeneity. Finally, an inverted funnel-plot analysis was used to detect any publication bias qualitatively.

Results

Study selection and characteristics

Figure one shows the study profile. A total of 20 studies were included in the qualitative and quantitative synthesis (Fig. 1 , Table 1 ) [2,[14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [26], [27], [28], [29], [30], [31], [32]]. One study which we included described an OR value. Nonetheless, we imputed it as an HR, because the other studies included defining the prognostic of per point CCI score and mortality used ORs.

Fig. 1

Study profile.

Table 1

Characteristic of the included studies.

No.	Author	Country	Study Design	Preprint	Subjects	Male	Overall ageMean± SD/Median (IQR)	Outcome	Newcastle-Ottawa Scale
1	Christensen et al.	Denmark	Retrospective cohort	No	798 vs 3265417 vs 3265	2144/4480	55 (31)	Severe outcomeMortality	9
2	Garibaldi et al.	USA	Prospective cohort	Yes	113 vs 719	443/832	63 (26)	Severe disease and mortality	9
3	Giacomelli et al.	Italy	Prospective cohort	Yes	48 vs 185	39/48 vs 122/185	NA	Mortality	8
4	Iaccarino et al.	Italy	Cross-Sectional	No	188 vs1403	125/188 vs892/1403	79.6 ± 0.8 vs64.7 ± 0.4	Mortality	9
5	Imam et al.	USA	Retrospective cohort	No	20 vs 1285	702/1305	61.0 ± 16.3	Mortality	8
6	Rossi et al.	Italy	Retrospective cohort	Yes	11,205 vs31,721	26873/42926	69 (22)	Mortality	9
7	Burn et al.	Spain	Retrospective cohort	Yes	2791 vs 15228	1669/2791Vs8223/15228	81 (14) vs61 (25)	Mortality	9
8	Narain et al.	USA	Retrospective Cohort	Yes	3098	2006/3098	NA	Mortality	9
9	Price-Haywood	USA	Retrospective cohort	No	326 vs 1056	NA	NA	Mortality	9
10	Sánchez-Montalvá et al.	Spain	Prospective cohort	Yes	22 vs 60	16/22 vs36/60	75.2 ± 6.2 vs53.3 ± 19.9	Mortality	8
11	Regina et al.	Switzerland	Retrospective cohort	Yes	37 vs163	29/37 vs91/163	66 (22) vs71 (27)	Mechanical ventilation	9
12	Castro et al.	USA	In silico cohort	Yes	707	213 vs 494	62.7 ± 15.1 vs62.2 ± 19	Mechanical ventilation	9
13	Shashikumar	USA	Prospective cohort	Yes	10 vs 1659 vs 343	7/10 vs 9/16;40/59 vs176/343	52.8 (23.6) vs57.6 (36.4)61.5 (23) vs65 (31)	Mechanical ventilation	7
14	Chroboczek et al.	France	Retrospective cohort	Yes	35 vs 35	27/35 vs 26/35	64 ± 1058 ± 14	Intubation	9
15	Ji et al.	Korea	Retrospective case-control	Yes				Severe disease	8
16	Bhargava et al.	USA	Retrospective cohort	Yes	197	45/74 vs58/123	63.1 (13.9) vs59.1 (17.3)	Severe disease	9
17	Marcos et al.	Spain	Retrospective cohort (?)	Yes	918	363 vs 555	79.2 (11.5) vs68.6 (14.7)	Severe disease	7
19	Mejía-Vilet et al.	Mexico	Prospective cohort	Yes	329	115 vs 214	53 (19) vs49 (20)	Critical care	7
20	Balnis et al.	USA	Prospective cohort	Yes	41	13/19 vs15/22	61.2 (27) vs58.7 (14)	Worse outcomes	7

The prognosis of CCI score 1–2 on mortality

A total of three studies showed that CCI 1–2 was significantly associated with mortality compared to CCI 0. The pooled HR was 1.41 [1.27, 1.57; p < 0.001] with high heterogeneity between studies [I 2, 64%; p = 0.04] (Fig. 2 ). Upon sensitivity analysis by removing Burn (2) et al. study, heterogeneity can be reduced while maintaining the significant association with mortality (HR 1.33 (1.28, 1.39), p < 0.001; I 2 0%, p = 0.46) are still maintained.

Fig. 2

The prognosis of CCI score 1–2 on mortality.

The prognosis of CCI score ≥3 on mortality

A total of three studies showed that CCI ≥3 was significantly associated with mortality (HR 1.77 (1.68, 1.86), p < 0.001; I 2 0%, p = 0.62) (Fig. 3 ).

Fig. 3

The prognosis of CCI score ≥ 3 on mortality.

Per point CCI score and mortality

Pooled HRs across four studies showed a non-significant association between increased per point CCI score and mortality (HR 1.09 (0.97, 1.23), p = 0.13; I 2 77%, p = 0.005). Moreover, upon removal of Price-Haywood study, heterogeneity can be reduced, indicating statistical robustness while maintaining significant associations (HR 1.16 (1.07, 1.25), p < 0.001; I 2 0%, p = 0.50) (Fig. 4 ).

Fig. 4

Per point CCI score and mortality.

The association between CCI score 1–2 and a composite of poor outcomes

A total of two studies showed that CCI 1–2 was significantly associated with a composite of poor outcomes (mortality and disease severity) (OR 1.90 (1.61, 2.24), p < 0.001; I 2 0%, p = 0.47) (Fig. 5 ).

Fig. 5

The association between CCI score 1–2 and a composite of poor outcomes.

The association between CCI score ≥3 and a composite of poor outcomes

A total of two studies showed that CCI ≥3 was significantly associated with a composite of poor outcomes (mortality and disease severity) (OR 2.95 (2.39, 3.65), p < 0.001; I 2 28%, p = 0.23) with considerable subgroup differences (I 2 72.8%, p = 0.06). Furthermore, subgroup analysis showed that CCI score ≥3 was significantly associated with mortality (OR 3.51 (2.69, 4.57), p < 0.001; I 2 0%, p = 0.44) and disease severity (OR 2.49 (1.97, 3.13), p < 0.001; I 2 0%, p = 0.61) (Supplementary Fig. 1). Upon sensitivity analysis by removing Christensen et al. study (severity, 3–4), heterogeneity can be reduced while maintaining the significant association with a composite of poor outcomes (OR 3.19 (2.57, 3.96), p < 0.001; I 2 3%, p = 0.38).

Mean CCI score and a composite of poor outcomes

Meta-analysis showed that pooled mean CCI score was higher in the group with poor outcomes (MD 0.69 (0.20, 1.18), p = 0.006; I 2 94%, p < 0.001) (Supplementary Fig. 2). Furthermore, subgroup analysis showed that mean CCI score was significantly higher in the mortality (MD 2.03 (1.20, 2.85), p= <0.001; I 2 64%, p = 0.01) and the severe group (MD 1.05 (0.68, 1.41), p < 0.001; I 2 65%, p = 0.02). Interestingly, lower mean CCI score was associated with mechanical ventilation, albeit non-significant (MD -0.46 (−1.25, 0.35), p = 0.27; I 2 68%, p = 0.01). Additionally, the subgroup differences were significantly high [I 2, 89.4%; p < 0.001]. Upon sensitivity analysis by removing Iaccarino et al. study, heterogeneity can be reduced; (MD 0.56 (0.06, 1.06), p = 0.03; I 2 85%, p < 0.001).

Publication bias & small-study effects

Funnel plot analysis showed an asymmetrical shape for mean CCI score and composite of poor outcomes (Supplementary Fig. 3).

Egger’s test showed no indication of small-study effects for the CCI score 1–2 (p = 0.734), CCI >3 (p = 0.544), and a composite of poor outcomes. However, there was a statistically significant small-study effect for the mean CCI and a composite of poor outcomes analysis.

Discussion

This systematic review and meta-analysis showed that higher CCI was associated with increased mortality and disease severity in patients with COVID-19. The risk for mortality increases by 16% for each increase in CCI.

The maximum score for CCI is 24 (updated version) or 29 (older version). However, the studies did not provide mean/median for CCI >3, which can be anywhere between 3 and 24/29, this imprecision is a potential cause of heterogeneity as studies with higher mean CCI for the category CCI >3 may show worse prognosis.

The source of the heterogeneity in Burns et al. study, in part, is caused by employing a primary care database from System for Research In Primary Care (SIDIAP), which did not provide detailed descriptions during hospitalization. Thus, other than age, no multivariable adjustments can be made for the HR, which might inaccurately show high HR [26].

Moreover, the high heterogeneity in per point CCI score and mortality was attributed to Price-Haywood study. This study employed adjustments with different sets of confounding variables compared to other studies, which might reveal other covariates that render the HR of per point CCI score and mortality insignificant [20].

Regarding the mean Charlson score and a composite of poor outcomes, after excluding the Iaccarino study, the heterogeneity can be reduced, albeit still high. One major difference between this study and others is that the population Charlson score was clustered around the mean, reflected by the low standard deviation [31].

The Charlson Comorbidity Index (CCI) originally was developed to predict the risk of mortality within 1 year of hospitalization. Scores are based on a number of comorbidities, each given a weighted integer from one to six depending on the severity of the morbidity [33]. It is a well-validated, simple, easy-to-apply index to evaluate patients’ prognosis and survival. During the current pandemic, the severity and mortality of COVID-19 are often predicted by age, gender, and the presence of comorbidities, such as diabetes, cardiovascular, cerebrovascular, and respiratory diseases [[34], [35], [36], [37], [38], [39], [40]]. Advanced age and multiple comorbidities are independent risk factors of mortality for patients with COVID-19 [32]. The CCI score, which accumulates ages and summarizes comorbidity measures, predicts death among COVID-19 patients by an exponential increase in the odds ratio at each point of score [6,31].

Among various conditions, hypertension and diabetes mellitus are the most prevalent conditions associated with increased severity and death of COVID-19 cases [41,42]. Individuals with chronic diseases are frequently found to have overexpression of angiotensin-converting enzyme (ACE)-2 receptor. Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) may invade the respiratory tract or other organs by binding to the ACE2 receptor at human cells following spike protein activation by transmembrane protease serine 2 (TMPRSS2). In patients with multiple comorbidities, renin-angiotensin system (RAS) inhibitors are commonly used and it is thought that these drugs upregulate ACE2 expression which consequently facilitates the entry of SARS-CoV-2 into the targeted cells. Nonetheless, regular administration of ACE inhibitors or angiotensin receptor blockers are not associated with severity and mortality in COVID-19 and are still recommended to control blood pressure and ultimately prevent cardiovascular complications [43]. Besides, the use of nonsteroidal anti-inflammatory drugs (NSAID) and corticosteroid is quite prevalent in people with long-term, chronic illnesses, but it is important to remember that these drugs must be used with caution considering its side effects [44,45]. However, it is found that the use of NSAID and RAS inhibitors had no significant effect on AKI in the first 48 h or increased death, while relative immunosuppression due to steroid consumption and high prevalence of comorbidities raise concerns about the development of poor outcomes [32,46].

Various biomarkers such as C-reactive protein (CRP), D-dimer, procalcitonin, and ferritin, are often elevated in severe COVID-19 cases and evaluation of these parameters can be useful in predicting severe outcomes and complications during such pandemic [47]. Lymphopenia was also shown to be associated with higher mortality [48]. Following SARS-CoV-2 invasion, the pathogen induces hyperinflammation or cytokine release syndrome which is thought as the plausible mechanism for multiple organ dysfunction, especially acute kidney injury, acute liver injury, and coagulopathy, and the development of other serious complications in COVID-19 [49,50]. The application of CCI scoring in the context of the COVID-19 outbreak can be very useful to forecast the need for intensive care unit (ICU) admission, respiratory support, or the probability for hospital readmission. Patients with comorbidities are often at higher risk for developing acute cardiovascular diseases, although COVID-19 in patients with comorbidity are concerning, it should not prevent or delay adequate treatment [51,52]. With the pandemic still growing worldwide, understanding the patients’ clinical characteristics and risk factors that anticipate the poor outcomes in COVID-19 transmission is crucial for planning comprehensive treatment and allocating valuable resources [31].

Limitations

The included studies did not report the mean/median for CCI >3 which potentially leads to imprecision and heterogeneity. Although a pooled HR showed a 16% increased risk for every one-point increase, we cannot assess the non-linearity of the association because the studies did not fulfill the prerequisites for a non-linear dose-response analysis.

Conclusion

A CCI score above 0 was prognostically associated with mortality, with per point CCI score increment associated with a 16% increase of mortality risk. A CCI score above 0 also was associated with a composite of poor outcomes. Finally, a higher mean CCI score was associated with mortality and disease severity, but not mechanical ventilation. However, there was a publication bias and significant small study effect of Mean CCI score and a composite of poor outcomes, indicated by the asymmetrical shape of the inverted funnel plot analysis and by the Egger’s test, respectively.