Computed tomography severity score as a predictor of disease severity and mortality in COVID-19 patients: A systematic review and meta-analysis

Abstract

Background

Prediction of outcomes in severe COVID-19 patients using chest computed tomography severity score (CTSS) may enable more effective clinical management and early, timely ICU admission. We conducted a systematic review and meta-analysis to determine the predictive accuracy of the CTSS for disease severity and mortality in severe COVID-19 subjects.

Methods

The electronic databases PubMed, Google Scholar, Web of Science, and the Cochrane Library were searched to find eligible studies that investigated the impact of CTSS on disease severity and mortality in COVID-19 patients between 7 January 2020 and 15 June 2021. Two independent authors looked into the risk of bias using the Quality in Prognosis Studies (QUIPS) tool.

Results

Seventeen studies involving 2788 patients reported the predictive value of CTSS for disease severity. The pooled sensitivity, specificity, and summary area under the curve (sAUC) of CTSS were 0.85 (95% CI 0.78–0.90, I² =83), 0.86 (95% CI 0.76–0.92, I² =96) and 0.91 (95% CI 0.89–0.94), respectively. Six studies involving 1403 patients reported the predictive values of CTSS for COVID-19 mortality. The pooled sensitivity, specificity, and sAUC of CTSS were 0.77 (95% CI 0.69–0.83, I² = 41), 0.79 (95% CI 0.72–0.85, I² = 88), and 0.84 (95% CI 0.81–0.87), respectively.

Discussion

Early prediction of prognosis is needed to deliver the better care to patients and stratify them as soon as possible. Because different CTSS thresholds have been reported in various studies, clinicians are still determining whether CTSS thresholds should be used to define disease severity and predict prognosis.

Conclusion

Early prediction of prognosis is needed to deliver optimal care and timely stratification of patients.  CTSS has strong discriminating power for the prediction of disease severity and mortality in patients with COVID-19.

Introduction

Coronavirus disease (COVID-19) outbreak caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) arose in Wuhan, China in December 2019. The World Health Organization (WHO) declared the outbreak a global pandemic in March 2020 [1].

Health care systems around the world continue to encounter substantial obstacles years after the first SARS-CoV-2-infection and it is crucial to know the factors that predict poor COVID-19 outcomes. COVID-19 patients experience fever, fever, dry cough, dyspnea, headaches, dizziness, weakness, diarrhea, and pneumonia. Many individuals have minor symptoms. Patients with acute respiratory distress syndrome (ARDS) who have COVID-19 experience a wide range of respiratory symptoms, from mild to severe hypoxia. However, those who have been diagnosed with a serious illness such as ARDS, septic shock, or even multiple organ dysfunction syndrome have poor prognosis [2]. It was found that the many patients with ARDS or severe pneumonia had pulmonary fibrosis [3].

The timeframe between the onset of symptoms and the onset of ARDS suggests that respiratory symptoms may appear suddenly. Therefore, early detection of critical cases is vital. Although a chest computed tomography (CT) scan can aid in the diagnosis of COVID-19, the reverse transcription-polymerase chain reaction (RT-PCR) test remains the gold standard.

A high-sensitivity diagnostic tool for interstitial pneumonia, chest CT may play an essential role in assessing the extent of lung involvement [4]. Non-contrast high-resolution computed tomography (HRCT) chest imaging is useful for detecting early disease, particularly in patients who have had false-negative real-time RT-PCR findings, as well as monitoring and detecting infection progression [5]. Several clinical trials have revealed and addressed the advantages of HRCT in the diagnosis of COVID-19 infection [6,7]. Although RT-PCR test remains the gold standard, chest CT scan can aid in the diagnosis of COVID-19. Additionally, the diagnostic accuracy of CT scans in COVID-19 patients have been analyzed by several systematic reviews [8], [9], [10].

The severity of the condition can also be determined using imaging findings, aiding clinicians in making an informed clinical judgment and facilitating prompt and effective treatment. In critically ill patients, the severity of the disease can also influence the prognosis, allowing for optimal selection of early intensive care unit (ICU) involvement. Hence, we can halt the disease from progressing, save healthcare resources, and decrease mortality. CT can detect COVID-19 lung disorders earlier [11] than RT-PCR (98% vs. 71% [12]) and is far more sensitive than RT-PCR. Patients with negative RT-PCR but significant CT findings should be segregated to prevent infection.

The chest CT severity score (CTSS) is a scoring method for evaluating COVID-19 lung variations and involvement based on an estimate of pulmonary affected areas. Based on the extent of lung lobe involvement, chest CTSS were calculated as follows: 0%, (0 points, no involvement), 1–25% (1 point, minimal involvement), 26–50% (2 points, mild involvement), 51–75% (3 points, moderate involvement), and 76–100% (4 points, severe involvement) [13]. The five lobe indices (range 0–20) were added together to calculate the CTSS. The chest CTSS was found to significantly correlate with clinical scores of pneumonia severity (CURB-65) and Pneumonia Severity Index/Pneumonia Outcome Research Trial (PSI/PORT), inflammatory markers (C-reactive protein, ferritin, neutrophil/lymphocyte ratio, lymphopenia). It has been suggested that the size of pulmonary lesions may indicate the intensity of the inflammatory response throughout the body [14,15].

Several studies evaluating the CTSS's prediction accuracy for severity and mortality during the COVID-19 pandemic have been published [16,17]. The severity and mortality following COVID-19 infection and CTSS were found to be significantly correlated. As a result, the area of the lung lesions in early CT scans following the onset of symptoms may be used as a potential indicator of patient mortality. However, the findings were inconclusive due to differences in a heterogeneous population, clinical settings, and cut-off values. Because different CTSS thresholds have been reported in various studies, clinicians are unsure whether CTSS thresholds should be used to define illness severity and predict prognosis. The purpose of this systematic review and meta-analysis was to assess CTSS's predictive accuracy on disease severity and mortality in COVID-19 patients.

Methods

The protocol was prospectively registered via PROSPERO (CRD42021244069). The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement [18]. and the Cochrane Handbook for Systematic Reviews of Interventions [19] guidelines were followed.

Selection criteria for studies

Following databases were searched for published articles: PubMed, Web of Science, Google Scholar, and the Cochrane Library for articles published between 7 Jan 2020 and 15 June 2021. We also checked the references of related journals to make sure we didn't skip any studies. The key words used in search strategy were COVID-19, non-contrast High resolution computed tomography, HRCT, Computed tomography severity score, Chest CT score, Prediction, severity and mortality. The search was limited to only articles published in English and no filter was applied for age, and type of study.

All citations were reviewed twice, and any inconsistencies were resolved through conversation and, if applicable, third-party adjudication. Two authors (JP, AKY) independently assessed all potentially significant references in two stages, first reading titles and abstracts and then articles for those that met the criteria. A third author (KS) resolved disagreements. We kept track of the exclusion criteria throughout the entire article review process.

Search strategies

We used electronic search engines and databases (Google Scholar, PubMed, Web of Science and the Cochrane Library) for obtaining the relevant articles from the literature.  The following keywords were used “COVID-19″, “CORONA”, "sars cov 2″, “Computed tomography”, “Chest CT”, “Chest computed tomography”, “X-Ray CT Scan”, “CT Severity”, “Mortality”, “Death”, “Decease”, “Severity”.

Boolean operators (OR, and AND) were used in succession. The filters applied were the English language and Human subjects. The references included in the studies were also screened for other potentially relevant publications. [See Supplementary Table 1]

Types of studies and eligibility criteria

To evaluate the CTSS as a tool for predicting disease severity and mortality in COVID-19 patients, cohort studies (retrospective or prospective) were selected for the review as there were no randomized control trial study. The studies were chosen using the inclusion criteria: (a) The predictive value of CTSS in COVID-19 patients affecting disease severity or mortality was assessed; (b) a 2×2 table of findings [adequate data to assess true positive (TP), false positive (FP), false negative (FN), and true negative (TN)] could be built. However, case reports, case series (explaining only phenomenology and with a sample size of less than ten), letters to editors, review articles, abstract publications, comments, and conference papers and inability to generate a 2×2 table were excluded.

Participants/population

Participants of COVID-19, diagnosed with RT-PCR testing and in whom a CTSS measure was determined.

Exposure

CTSS using any cutoff value that reported in studies for disease severity and mortality in COVID-19 patients

Comparison

Comparisons of the CTSS among:

Mild-moderate vs severe-critical and survival vs non-survival cases

Types of outcome assessment

To assess the predictive accuracy of CTSS for disease severity and mortality

Data extraction

Two reviewers (JP and AKY) extracted data from selected studies independently and in duplicate, including the first author, articles published year, country, group of patients (male/female), mean age, cut-off value, area under curve (AUC), TP, TN, FP, FN, sensitivity, and specificity using predefined data abstraction forms. Differences of opinion were addressed through discussion, agreement, or consultation with a third reviewer (KS).

Assessment of quality of studies

Two authors (JP and AKY) conducted the risk of bias assessments. We employed the Quality In Prognosis Studies (QUIPS) tool [20] for prognostic factor to evaluate the risk of bias (RoB). This tool highlights the six bias categories, as well as overall rating appraisals, prompted items and relevant factors. The QUIPS tool classifies RoB with each of the following aspects: study participation, study attrition, prognostic factor measurement, outcome measurement, study confounding, and statistical analysis and reporting as "low," "moderate," or "high."

Statistical analysis

In order to predict CTSS, this meta-analysis comprises studies that fulfilled all of the criteria for inclusion with none of the exclusion criteria. Direct extraction or indirect calculations were used to get requisite data. DerSimonian and Laird random-effects models were utilized in our meta-analyses. The study weights were determined using the inverse variance approach. The sensitivity, specificity, positive likelihood ratio, negative likelihood ratio, diagnostic odds ratio (DOR), and related 95% confidence intervals (CI) were calculated using a bivariate random-effects model. To assess overall predictive accuracy, summary receivers operating characteristic (sROC) curves were plotted. If AUC value will be higher, the diagnostic power will be better [21]. The Cochrane Q test and the I² for heterogeneity have been used to assess the study heterogeneity [22]. Deek funnel plot was used to see the publication bias. Cochrane collaboration guidelines [23] were used to convert interquartile ranges (IQR) to standard deviations (SD) for continuous variables. The relation between likehood ratio, pre and post-test probability was determined using a Fagan nomograph.

To investigate potential variables within and between studies, bivariate meta-regression analyses was performed. We looked at possible factors of heterogeneity using the different continuous variables as covariate/moderator variables such as mean age, mean D-dimer level, mean C-reactive protein (CRP), mean lymphocyte count, percentage of male gender, percentage of cardiac disease, percentage of respiratory disease, percentage of hypertensive subjects, and percentage with diabetes.

Taking into account the clinical relevance and findings of the meta-regression analysis, we performed a sub-group analysis to formative sources of heterogeneity amongst included studies. We performed subgroup analyses for disease severity using cut-off values (continuous variable). Because there were fewer studies, a subgroup analysis for mortality was not possible. STATA version 13.0 was used for all statistical analysis (StataCorp. 2013. Stata Statistical Software: Release 13. College Station, TX: StataCorp LP).

Results

Search results and description of studies

After the literature search, 118 potentially relevant studies have been identified. 34 duplicate studies were excluded after screening. After screening titles and abstracts, 47 studies were excluded on the basis of inclusion and exclusion criteria. 37 studies provided additional insight after reading the full text. Finally, the present meta-analysis included 22 studies, sixteen of which reported predictive value for severity [15,[24], [25], [26], [27], [28], [29], [30], [31], [32], [33], [34], [35], [36], [37], [38], five of which reported predictive value for mortality [39], [40], [41], [42], [43], and one of which reported predictive value for both severity and mortality [44] (Fig. 1 ).

Fig. 1.

PRISMA flow diagram.

Fig. 2 depicts the risk of bias in the different studies used for the present meta-analysis. The risk of bias was low in fourteen studies [15], [24], [25], [26],28,29,32,33,35,38,39,[42], [43], [44], a moderate in three studies [31,37,41], and high in five studies [27,30,34,36,40].

Fig. 2.

Risk of bias summary.

Table 1 summarizes the demographic parameters and features of included studies. Notably, the sensitivity, specificity, and AUC of CTSS to predict disease severity and mortality differed considerably across studies. Severity was determined by the degree of involvement (50–75%) as lung opacification.

Table 1.

Demographic parameters and characteristics of the included studies in prediction of CTSS for disease severity and mortality.

	Study	Country	Duration of data collection	No. of patients	Male/female	Mean age(±SD)	Cut-off	AUC	TP	TN	FN	FP	SEN (%)	SPE (%)
Severity	Li [24]	China	1 month	83	44/39	45.5  ±  12.3	7	0.87	20	48	5	10	80	82.8
	Tan [25]	China	24 days	27	11/16	48.89 ± 18.47	7	0.71	3	19	3	2	50	91
	Sun [26]	China	27 days	84	37/47	46 ± 11.85	8.2	0.94	21	56	2	5	91.3	91.8
	Li K [27]	China	27 days	78	38/40	44.6  ±  17.9	7.5	0.92	7	70	1	0	82.6	100
	Yang [28]	China	15 days	102	53/49	52.83 ± 12.96	19.5	0.89	15	79	3	5	83.3	94
	Xiao [29]	China	1 month 21 days	243	105/138	47 ± 51.11	11	0.86	34	159	6	44	85	78.3
	Li S [30]	China	1 month	53	27/26	64.8  ±  17.78	5.25	0.75	28	12	7	6	81	69.2
	Nair [31]	India	4.5 month	67	64/3	45 ± 37.78	21.5	0.87	22	28	1	16	95.7	63.6
	Bellos [15]	Greece	N/A	42	29/13	56.64 ± 14.12	10.5	0.81	8	22	3	10	75	70
	Palwa [32]	Pakistan	2 months	500	407/93	54.76 ± 13.995	18.5	0.96	97	356	18	29	84.3	92.5
	Devie [33]	France	2 month 4 days	158	84/74	68 ± 14	15	0.67	31	81	27	19	53.4	81
	Li Y [34]	China	2 months	123	62/61	64.43 ± 14.02	6	0.66	23	54	16	30	59	64.3
	Salahshour [44]	Iran	1 month	439	272/167	53.7  ±  17	8	0.77	59	207	9	164	86.8	55.8
	Ye Y [35]	China	5 months	196	107/89	51.7  ±  12.98	8.87	0.91	41	109	3	43	93.2	71.5
	Saad [36]	Egypt	6 months	192	127/65	37.86 ± 11.8	7	N/A	28	157	1	7	95.9	96
	Rana [37]	India	5 month 15 days	250	175/75	56.89 ± 13.67	13	0.996	94	150	6	0	94.4	100
	Dogan [38]	Turkey	N/A	151	73/78	51 ± 53.33	15	0.83	17.8	89.0	2.2	41.9	89	68
Mortality	Zhou [39]	China	N/A	134	85/49	68 ± 12.59	16.5	0.817	50	51	22	11	69.4	82.3
	Li Y [40]	China	25 days	98	65/33	71.1  ±  8.5	14.5	0.875	38	40	8	12	83.3	77.3
	Tabatabaei [41]	Iran	2 months	90	54/36	44.2  ±  55.9	7.5	0.89	25	52	5	8	83	87
	Raoufi [42]	Iran	1 month	380	251/129	53.57 ± 16.66	12	0.80	22	266	7	85	75.8	75.8
	Salahshour [44]	Iran	1 month	439	272/167	53.7  ±  17	15	0.80	18	357	10	54	64.3	86.9
	Abbasi [43]	Iran	19 days	262	172/90	58 ± 17.78	10	0.839	47	136	9	70	84	66

CTSS- computed tomography severity score, AUC- area under curve, TP- true positive, TN- true negative, FN- false negative, FP- false negative, SEN- sensitivity, SPE- specificity, SD- standard deviation.

Outcomes

Predictive value of CTSS on disease severity

It was reported in seventeen studies involving 2788 patients. The pooled estimates of sensitivity and specificity were 0.85 (95% CI 0.78–0.90, I2 =83) and 0.86 (95% CI 0.76–0.92, I2 =96), respectively (Fig. 3a ). CTSS had a summary AUC (sAUC) of 0.91 (95% CI 0.89–0.94) for predicting disease severity, suggesting a strong diagnostic value (Fig. 4a). The pooled estimates of positive likelihood ratio and negative likelihood ratio were 6.2 (95% CI 3.4–11.3) and 0.18. (95% CI 0.12–0.27), respectively. DOR had been 35 (95% CI 14–88). Fagan nomogram showed that when pre-test probability was 50%, the post-test probability of CTSS for identification of severe cases was 86%, however, when CTSS was lower than cut-off value, it was diminished to 15% (Fig. 5a).

Fig. 3a.

Forest plot demonstrating CTSS's sensitivity and specificity for predicting disease severity in COVID-19 patients.

Fig. 3b.

Forest plot demonstrating CTSS's sensitivity and specificity for predicting mortality in COVID-19 patients.

Fig. 4.

sROC for the included studies. (a) CTSS's AUC for predicting disease severity was 0.91 (95% CI 0.89–0.94). (b) CTSS's AUC for predicting mortality was 0.84 (95% CI 0.81–0.87).

Predictive value of CTSS on mortality

A total of 1403 patients were enrolled in six studies that used CTSS to predict mortality. The pooled estimates of sensitivity and specificity were 0.77 (95% CI 0.69–0.83, I² = 41) and 0.79 (95% CI 0.72–0.85, I² = 88), respectively (Fig. 3b ). The sAUC of CTSS for predicting mortality was 0.84 (95% CI 0.81–0.87) (Fig. 4b). The pooled estimates of positive likelihood ratio and negative likelihood ratio were 3.7 (95% CI: 2.8–4.8) and 0.29 (95% CI: 0.23–0.38), respectively. The DOR was 13 (95% CI: 9–18). The Fagan nomogram revealed that when the pre-test probability was adjusted to 50%, the post-test probability of CTSS for mortality was 79%; when the CTSS was less than the cut-off value, the post-test probability was 23% (Fig. 5b ).

Fig. 5.

(a) Fagan nomogram of CTSS for predicting disease severity. (b) Fagan nomogram of CTSS for predicting mortality.

There was no significant publication bias of the funnel plot for disease severity (P = 0.45) or mortality (P = 0.08), respectively implying that the study findings were valid (See Supplementary Figure 1 and 2).

We investigated the source of heterogeneity on the effect size using major clinical parameters (age, D-dimer level, CRP, lymphocyte count, gender, cardiac disease, respiratory disease, hypertension, and diabetes), but we found that these variables won't significantly illustrate the factors that contribute on pooled sensitivity and pooled specificity disease severity and mortality, respectively (See Supplementary Figure 3 and 4).

Analyses of subgroups and sensitivity

The CTSS cut-off value was used to conduct subgroup analysis. In eight studies for predicting severity, the cut-off value was greater than 10, resulting in “high cut-off value” (cut-off > 10) subgroup. In the “equal or low cut-off value” (cut-off <10) subgroup, nine other studies were included. For cut-off > 10 and cut-off <10, the AUC were 0.91 (95% CI 0.89–0.94) and 0.92 (95% CI 0.89–0.94), respectively [Table 2 ].

Table 2.

Subgroup analysis based on cut-off values.

Categories	Number of studies	Sensitivity (95% CI)	Specificity (95% CI)	sAUC (95% CI)	DOR (95% CI)
Cut-off  < 10	9	0.86 (0.75–0.92)	0.85 (0.71–0.93)	0.92 (0.89–0.94)	35 (10–123)
Cut-off > 10	8	0.84 (0.74–0.91)	0.87 (0.71–0.95)	0.91 (0.89–0.94)	37 (9–144)

sAUC- summary area under curve, DOR- diagnostic odds ratio, CI- confidence interval.

Discussion

The present meta-analysis, which had a comprehensive literature search, a pre-registered protocol, an emphasis on only COVID-19 patients, the QUIPS tool to assess bias, and the integration of recent studies, suggested that CTSS may be a reliable predictor of severity and mortality in COVID-19 patients. As far as we know, there has not been another meta-analysis that used CTSS to forecast severity and mortality in COVID-19 patients.

Clinicians are unsure of whether CTSS thresholds should be used to define illness severity and predict prognosis because different CTSS thresholds have been reported in various studies. The current meta-analysis reveals that CTSS in COVID-19 patients can be used to not only predict disease severity (AUC = 0.91, sensitivity = 0.85, and specificity = 0.86) but also mortality (AUC = 0.84, sensitivity = 0.77, and specificity = 0.79).

Management for mild and severe patients is vastly varied. Early prediction of prognosis is needed to deliver the best care to patients and stratify them as soon as possible. In severe cases, however, mortality remains high despite the use of a variety of interventions such as mechanical ventilation, extracorporeal membrane oxygenation (ECMO), and continuous renal replacement therapy (CRRT) [45,46]. A thoracic CT scan can also be used to evaluate COVID-19 disease progression. Non-ICU patients had bilateral ground-glass opacities and subsegmental consolidations, whereas ICU patients had bilateral multiple lobular and subsegmental consolidations [47]. The respiratory rate, oxygen saturation, and ratio of arterial oxygen partial pressure to fractional inspired oxygen (PaO2/FiO2) are used to categorize mild and severe instances. Two meta-analyses [9,48,49]. have confirmed the diagnostic value of chest CT in identifying COVID-19. However, none of them examined CTSS's predictive values for severity and mortality.

COVID-19 can be detected with a sensitivity of > 90% on chest CT, with false-negative results most common in patients who have had symptoms for less than 3 days [3,50]. Furthermore, the study population, COVID-19 prevalence, COVID-19 stage, disease severity at the time of imaging, and the presence of concomitant lung disease all have an impact on chest CT diagnostic accuracy [51,52].

The cut-off values of individual studies varied widely from study to study, which could be attributed to integrating the degree of pulmonary involvement with different attenuation patterns (i.e., normal, ground-glass, and consolidation). A consensus has yet to emerge for the standard cut-off value of CT scans. The major parameters for identifying mild and severe cases are respiratory rate, oxygen saturation, and PaO2/FiO2. In our study, we did not find differences in sensitivity and specificity for studies having a lower and higher cut-off for the severity of the disease. A similar observation was made for mortality due to different population settings, comorbidities, the timing of the measurement of CTSS, and excessive inflammation and immune suppression. However, even with individual data, this would have been possible when the same CT machine and standard definition for the severity of COVID-19 cases and death due to COVID-19 cases were used. The absence of publication bias further validated the findings of the current meta-analysis.

The heterogeneity among the study should be taken into consideration. The reason for heterogeneity may be that the studies have been conducted in different health care systems, in different time frame and countries capacity to conduct study using HRCT might have affected the conduct of the study. Also, in case of mortality, we find study only from limited countries and the availability of effective treatment approaches may have possible effect on the mortality at the time of conduct of study.

Limitations

The data was subject to confounding impacts because most studies were retrospective. We also see many variations among the studies, as indicated by I², implying that well-designed prospective trials are required. Many factors influenced disease development and prognosis, including age, sex, comorbidities, and other inflammatory markers, which were not explored in the present study due to a lack of data. Despite doing meta-regression and subgroup analyses, the heterogeneity was not explored, resulting in low-quality evidence. The CTSS cut-off value was inconsistent among the meta-analysis studies, which could be attributed to differences in clinical situations or patient comorbidities and settings, which could have caused heterogeneity.

Conclusion

Our meta-analysis shows that CTSS provides strong discriminating power for predicting disease severity and mortality in COVID-19 patients. If CTSS is validated in more prospective studies, it may aid clinicians in recognizing possibly severe cases earlier, instituting early triage, and initiating comprehensive COVID-19 control on time, thus lowering COVID-19 mortality.

CRediT authorship contribution statement

Jay Prakash: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Validation, Visualization, Writing – original draft, Writing – review & editing. Naveen Kumar: Conceptualization, Data curation, Funding acquisition, Software, Investigation, Resources, Validation, Visualization, Writing – review & editing. Khushboo Saran: Investigation, Validation, Writing – review & editing. Arun Kumar Yadav: Data curation, Formal analysis, Methodology, Writing – review & editing. Amit Kumar: Data curation, Formal analysis, Methodology, Writing – original draft, Writing – review & editing. Pradip Kumar Bhattacharya: Conceptualization, Investigation, Supervision, Validation, Visualization, Writing – review & editing. Anupa Prasad: Methodology, Visualization.