Very high SARS-CoV-2 load at the emergency department presentation strongly predicts the risk of admission to the intensive care unit and death

To the Editor,

During the first wave of COVID-19 pandemic in Italy, the high number of deaths was mostly due to an uncontrolled viral spread and to a limited knowledge of the clinical and laboratory factors leading to the worst complications [1], [2], [3]. A better risk stratification may be of help in facing the increase in patients requiring hospitalization during the ongoing second wave of the pandemic in Europe. The possible importance of viral load as predictive marker for the clinical outcome was recently supported by its evaluation on different clinical samples, even if the evidence is still conflicting [4], [5], [6], [7], [8]. Several clinical scores were described to date in order to be used in the stratification of COVID-19 patients. However, none of them includes the possible role played by the amount of virus detected at clinical presentation [9], [10]. Therefore, further studies are certainly needed the evaluate the possible prognostic role of the viral load, together with other clinical parameters collected at the emergency department (ED) presentation, in order to provide possible additional patients’ stratification tools to physicians.

We evaluated complete clinical and laboratory data of 121 patients presenting to the ED of San Raffaele Hospital, Milan, Italy from March 23rd to May 23rd, 2020. Median age was 60.3 years (IQR = 25), with 89 patients under 70 years and 32 over 70 years. Sixty-three patients were male (52.1%): 46 (73%) under 70 years and 17 (27%) over 70 years. Viral load was inferred on nasopharyngeal swabs through cycle threshold (Ct) determination with Cobas^® SARS-CoV-2 Test (Roche), which detects ORF-1a/b and E gene regions on SARS-CoV-2 genome. Non-infectious plasmid DNA containing a specific SARS-CoV-2 sequence and a pan-Sarbecovirus sequence were used in the test as positive control. A non-Sarbecovirus related RNA construct was used as internal control. The test was designed to be used on the automated Cobas^® 6800 Systems under Emergency Use Authorization (EUA). To avoid possible bias in the study related to different amount of material collected, we examined 30 representative nasopharyngeal swabs for cellular target gene (β-globin): we evaluated the coefficient of variation (CV) of the obtained Ct values and observed that β-globin had a CV of 5% (Ct mean: 38.1; standard deviation: 1.9), ORF-1a/b had a CV of 17.2% (Ct mean: 27.9; standard deviation: 4.8) and E gene had a CV of 20% (Ct mean: 30; standard deviation: 6.0). For the sake of brevity, only Ct values for ORF-1a/b (the SARS-CoV-2 specific target) will be discussed, but no substantial differences were observed in the final analyses when using E gene results. The overall median Ct value was 29.5 (IQR = 8.3): significantly lower in >70 [27.5 (IQR = 8.6)] than in ≤70 patients [30.2 (IQR = 6.8)]. All other clinical and laboratory parameters considered in our study are reported in Table 1. All parameters were used in univariate Cox models and were correlated to the risk of admission to the intensive care unit (ICU) and death. Subsequently, we identified the best multivariate Cox model according to Akaike Information Criterion by the Best Subset Selection method, using all the variables tested in univariate analyses. The adequacy of the chosen model was confirmed by testing the proportional hazards assumption for the Cox regression model. The study was reviewed and approved by San Raffaele Hospital IRB in the COVID‐19 Biobanking project “COVID‐BioB” N° CE: 34/int/2020 19/March/2020 ClinicalTrials.gov Identifier: NCT04318366 and the patients gave their informed consent at hospital admission.

The 121 patients were categorized in three groups according to the viral load: 13 (10.7%) patients were included into the very high (Ct ≤ 21), 56 (46.3%) into the medium-high (Ct 21–31) and 52 (43%) into the low (Ct >31) viral load group. Twenty-seven (22.3%) patients were admitted to the ICU or died, with 6/13 (46.1%) in the very high, 15/56 (26.8%) in the medium-high and 6/52 (11.5%) in the low viral load group. The number of deaths in the same groups were 6/13 (46.1%), 12/56 (21.4%) and 5/52 (9.2%), respectively.

The univariate Cox models analyses revealed a clinical and statistical significantly increased risk of ICU admission or death for age >70 years-old (HR = 2.93; p=0.004), coronary artery disease (HR = 3.02; p=0.008) and chronic kidney disease (HR = 3.14; p=0.0133). Although not correlated with an increased risk, other statistically significant variables were oxygen saturation at ED presentation (HR = 0.91; p<0.0001), PaO₂/FiO₂ ratio (HR = 1.0; p=0.0059), blood glucose (HR = 1.01; p=0.0047), aspartate aminotransferase (HR = 1.01; p=0.02), lactate dehydrogenase (HR = 1.01; p<0.0001) and C-reactive protein levels (HR = 1.01; p=0.0002). Importantly, viral load was a statistically significant variable correlated with an increased risk of ICU admission or death: in particular, a significantly higher risk was observed in the case of medium-high (HR = 2.78; p=0.0325) and, even most significantly, in the case of very high viral load (HR = 12.5; p< 0.0001) (Figure 1A).

Figure 1:

Kaplan-Meier survival curves. (A) Kaplan–Meier curves for ICU admission or death as function of viral cycle threshold (Ct) detected for SARS-CoV-2-specific ORF-1a/b at ED admission for all subjects (log-rank test p<0.0001). Plots of the fitted Cox model using median clinical parameters for (B) 70 years and (C) over 70 years patients.

In the multivariate analysis, only age >70 (HR = 1.86; p=0.18), coronary artery disease (HR = 1.84; p=0.23), PaO₂ (HR = 1.02; p=0.06), lactate dehydrogenase (HR = 1.01; p<0.0001), serum creatinine (HR = 1.27; p=0.04) and viral load (p<0.0001) were retained in the chosen model. Regarding the viral load, a higher risk of ICU admission or death was observed in the medium-high (HR = 3.9; p=0.0176) and, also in this case most significantly, in the very high group (HR = 26.54, p<0.0001). The importance of viral load in influencing the risk of ICU admission and death in our cohort was made even clearer when the fitted Cox model was applied to patients with different risk profiles. In other words, the effect of different viral loads was evaluated in simulated patients featuring different risk profiles according to the other parameters included in the model. As shown in Figure 1B, C, the effect of higher viral loads is evident in plots of the fitted model performed using the median values of all other clinical parameters in our cohort. This is appreciable both in younger (≤70 years) patients (no coronary artery disease; PaO₂ 9.9 kPa; lactate dehydrogenase 5.7 μkat/L; serum creatinine 75.2 μmol/L) and in older patients (no coronary artery disease; PaO₂ 8.5 kPa; lactate dehydrogenase 6.1 μkat/L; serum creatinine 100.8 μmol/L). The effect of viral load is also evident in the two age groups when using the worst and the best values for each of the parameters included in our model, further confirming its role as independent prognostic factor (Supplementary Figure 1; descriptive statistics for younger and older patients in Supplementary Tables 1 and 2).

The lack of virological markers associated to the clinical outcomes of SARS-CoV-2 infected subjects contribute to hinder the prompt stratification of COVID-19 patients presenting at the ED. Under this perspective, our results suggest that nasopharyngeal viral load is a strong prognostic predictor for ICU admission or death, when determined in symptomatic subjects at ED presentation. This is evident especially when its effect is considered in combination with other established risk factors, such as older age, presence of comorbidities, impaired respiratory function and high values of systemic inflammatory markers. In other words, as further suggested by fitting our model to patients with different risk profiles, SARS-CoV-2 viral load has not an absolute prognostic value per se in all infected individuals. It should be evaluated in the context of the general clinical presentation and of the clinical history of each patient. We are aware that larger studies are certainly needed to confirm its role in helping identify patients at higher risks for serious COVID-19 complications, but our data further support the opportunity of including viral load in clinical scores to be used for a more rational stratification of patients during the ongoing second phase of the pandemic.