Introduction

The COVID-19 pandemic, declared in March 2020, has remained a global health threat. Between 2020 and 2021, COVID-19 became the third leading cause of death in the United States, and to date, it has been associated with more than 6.9 million deaths worldwide [1,2]. Although the World Health Organization declared in May 2023 that COVID-19 no longer constituted a Public Health Emergency of International Concern, the disease continues to pose a threat that requires ongoing surveillance. In Colombia, according to the latest 2025 report from the National Institute of Health, there have been 6,415,764 confirmed cases since the onset of the pandemic, with a case fatality rate of 2.24%. Notably, the burden has been disproportionately high among individuals over the age of 60 [3], highlighting its persistent impact on public health.

A wide range of predictive models have been used during the pandemic to estimate the risk of deterioration, including severity scores originally developed for critical illness, such as APACHE II and SOFA, which, although not designed specifically for COVID-19, demonstrated reasonable prognostic value in this population. A recent systematic review by Appel et al. reported that 242 scores for COVID-19 outcome prognosis have been developed, though most exhibited considerable risk of bias or insufficient validation. Among these, the 4C Mortality Score emerged as the most consistently validated tool, integrating clinical, physiological, and laboratory variables and demonstrating robust predictive performance across multiple populations [4].

In addition, Marcolino et al. developed and externally validated the ABC₂-SPH risk score, which also incorporates clinical parameters, including heart rate and blood pressure, alongside physiological and laboratory markers, and demonstrated strong discriminative ability for in-hospital mortality in cohorts from Brazil and Spain [5].

Despite this, the shock index (SI) has been rarely incorporated into these models, despite its long-standing role as a rapid and reliable bedside marker of hemodynamic compromise in trauma, sepsis, and obstetric emergencies. Its simplicity and immediate availability make it particularly valuable in scenarios requiring swift clinical assessment, including patients presenting with COVID-19. Although most studies suggest favorable performance of the SI in patients with COVID-19 [6–8], the evidence remains inconsistent (9), with reports ranging from poor predictive capacity to excellent discrimination of outcomes such as intensive care unit (ICU) admission and mortality [9,10]

Several variants of the SI have been previously developed to improve its performance across different clinical settings by incorporating additional physiological parameters. The modified SI (mSI) replaces systolic blood pressure with mean arterial pressure, the diastolic SI (dSI) incorporates diastolic pressure, the estimated SI (eSI) adjusts for vascular resistance via pulse pressure, and the hypoxia-adjusted SI (HaSI) integrates oxygen saturation to account for respiratory compromise. Evidence from clinical studies indicates that mSI and dSI have demonstrated potential improvement in predictive performance over SI in certain populations, such as patients with trauma, sepsis, or cardiogenic shock [11–15]. Current evidence regarding the prognostic value of SI and its variants in COVID-19 remains controversial, as studies have reported inconsistent predictive performance. Moreover, evidence from Latin American cohorts is limited, making it difficult to generalize existing findings to settings with distinct demographic and comorbidity profiles. To address these gaps, we comprehensively evaluated the prognostic performance of SI and its variants in predicting hospital mortality and ICU admission, as well as the need for invasive mechanical ventilation, among patients with moderate COVID-19 treated at a tertiary care center in a middle-income country.

Materials and methods

Results

A total of 283 patients were included (Fig 1), with a median age of 61 years (IQR: 48.0–75.0). The majority were male (58.7%). Regarding cardiovascular comorbidities, 51.9% had cardiac disease, 49.1% had hypertension, and 7.1% had coronary artery disease. Only 5.7% had a history of heart failure. Other conditions, such as atrial fibrillation (5.3%), atrioventricular block (0.4%), or ventricular arrhythmias (0.4%), were uncommon. Concerning pulmonary diseases, 9.5% had chronic lung disease, including COPD (4.9%), asthma (1.4%), and pulmonary fibrosis (0.7%). No cases of interstitial lung disease, cirrhosis, or hepatitis were reported. Chronic kidney disease was present in 21% of patients, primarily in stages 3–5, with stage 5 being the most frequent (11.0%). Diabetes mellitus was identified in 21.2% (7.4% non-insulin-dependent and 13.8% insulin-dependent), although data were unavailable for 78.8% of the cohort. A total of 14.5% had a cancer diagnosis, with hematologic malignancies (29.3%) and prostate cancer (19.5%) being the most common types (Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Sociodemographic and clinical characteristics.

https://doi.org/10.1371/journal.pone.0341634.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Patient selection flowchart.

Inclusion and exclusion criteria of the patients.

https://doi.org/10.1371/journal.pone.0341634.g001

At admission, vital signs showed a mean systolic blood pressure of 123.0 mmHg, a heart rate of 89 bpm, and oxygen saturation of 96%. All patients had a Glasgow Coma Scale score of 15, and the NEWS2 score had a median of 3.0 (IQR: 1.0–5.0). The shock index had a median value of 0.7 (IQR: 0.6–0.9). A qSOFA score ≥2 was observed in 6.8% of patients. Inflammatory markers included leukocytes at 6,690 (IQR: 5,070–9,925), neutrophils at 4,800 (3,365–7,965), lymphocytes at 1,000 (695–1,470), and a neutrophil-to-lymphocyte ratio (NLR) of 5.1 (2.7–9.5) (Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Admission physiological variables and clinical outcomes.

https://doi.org/10.1371/journal.pone.0341634.t002

Regarding complications, 23.3% developed acute respiratory distress syndrome, 9% had a high-risk NEWS2 score, and 29.3% required ICU admission, with a median stay of 8 days (IQR: 3–18.5). Invasive mechanical ventilation was administered to 12% of patients, with a median duration of 10.5 days (IQR: 7.0–17.75) (Table 2).

The median length of hospital stay was 7 days (IQR: 4–14). At discharge, 88% of patients were discharged alive, while 12% died. The most common discharge destination was other (73.1%), followed by homecare (11.7%) (Table 2).

ICU admission

None of the indices demonstrated acceptable discriminative ability, with AUC values ranging from 0.495 to 0.514 (Table 3). The highest sensitivity (56.6%) was observed with mSI, while eSI showed the highest specificity (60.0%) and negative predictive value (NPV, 73.2%). Positive likelihood ratios (LR+) were low across all indices (range: 1.13–1.21), and the Youden index was below 0.10 in all cases (Table 3) (Fig 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Diagnostic performance of shock index variants.

https://doi.org/10.1371/journal.pone.0341634.t003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. ROC analysis for each shock index variant predicting ICU admission.

Predictive performance of mSI, dSI, eSI, and HaSI for ICU admission.

https://doi.org/10.1371/journal.pone.0341634.g002

Hospital length of stay

The eSI showed the best overall performance (AUC = 0.574) (Fig 2), with a specificity of 69.2%, NPV of 79.1%, and LR+ of 1.60. The highest sensitivity was observed with HaSI (84.0%), at the cost of low specificity (33.7%). The highest Youden indices were recorded for dSI (0.197) and eSI (0.185) (Table 3) (Fig 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. ROC analysis each shock index variant predicting hospital length stay.

Predictive performance of mSI, dSI, eSI, and HaSI for hospital length of stay.

https://doi.org/10.1371/journal.pone.0341634.g003

Discharge status (mortality)

The eSI demonstrated the best performance in predicting mortality, with an AUC of 0.699 (Fig 3), sensitivity of 59.0%, specificity of 73.5%, positive predictive value (PPV) of 94.2%, and NPV of 19.7%, with an LR+ of 2.23. HaSI followed closely, with an AUC of 0.681 and comparable diagnostic performance. The mSI had the highest specificity (91.2%) but very low sensitivity (21.3%) (Table 3) (Fig 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. ROC analysis each shock index variant predicting in-hospital mortality.

Predictive performance of mSI, dSI, eSI, and HaSI for in-hospital mortality.

https://doi.org/10.1371/journal.pone.0341634.g004

Discussion

This study evaluated the utility of the shock index and its variants in predicting clinical outcomes in patients with moderate COVID-19, including the need for inpatient management (ICU admission, hospital stay), and in-hospital mortality. The findings show that the traditional shock index demonstrated limited discriminative capacity, with areas under the curve (AUCs) of 0.51 for ICU admission, 0.56 for hospital length of stay, and 0.52 for mortality.

When exploring variations of the index, including the use of mean arterial pressure, age, and oxygen saturation, no substantial improvement in performance was observed, except for the HaSI index, which achieved an AUC of 0.70 for mortality prediction, indicating acceptable performance in that context.

Although some predictive models, such as NEWS2 and qSOFA, show favorable performance in specific parameters, their overall usefulness in early risk stratification of patients with COVID-19 remains limited. NEWS2 demonstrates relatively higher sensitivity compared with other scores; however, its values generally remain below 70%, meaning that a substantial proportion of at-risk patients may still be missed. In contrast, qSOFA exhibits very high specificity, approaching 97% for outcomes such as in-hospital mortality, ICU admission, need for mechanical ventilation, and vasopressor use, indicating a strong ability to correctly identify patients unlikely to deteriorate. Nevertheless, the limited sensitivity and modest ability to rule out adverse outcomes observed across these models reduce their reliability for confidently excluding risk. Taken together, these trade-offs underscore the challenges of depending on these tools to identify patients who are truly low risk at emergency department presentation [17].

Our findings are consistent with studies such as that by Van Rensen et al., who reported very low sensitivity (12.3%) for SI in predicting mortality among hospitalized COVID-19 patients, despite high specificity (93.6%) and limited PPV (42.4%). In their analysis, using a low SI cutoff (>0.57)—within the normal range—did not improve its performance in predicting ICU admission. In this context, SI was not useful for early identification of clinical deterioration in hospitalized patients [18]. Other studies have explored modified versions of the shock index. In the study by Trichur et al., the use of SI adjusted for mean arterial pressure was significantly associated with mortality (p = 0.04), whereas the traditional SI was not (p = 0.205), suggesting that certain modifications may improve predictive performance [19].

These variable performances across studies can be attributed to differences in disease severity, patient populations, and the choice of cutoff values. Van Rensen et al.‘s study included a broader range of hospitalized patients, while our cohort focused specifically on moderate COVID-19 cases, where more subtle hemodynamic alterations may not be captured by traditional SI. Additionally, their use of a threshold within the normal range inherently limits sensitivity. The superiority of modified indices, such as the SI adjusted for mean arterial pressure in Trichur’s study and the eSI in our analysis (AUC 0.699 for mortality), suggests that incorporating additional hemodynamic parameters better reflects the complex pathophysiology of COVID-19-related deterioration. However, even the best-performing index in our study showed only modest discriminative ability for mortality and poor performance for ICU admission and hospital length of stay (AUC 0.574), reinforcing the need for multiparametric approaches rather than reliance on single hemodynamic indices for risk stratification in moderate COVID-19.

Similarly, studies in older adults have shown results in line with ours. One analysis of elderly patients with COVID-19 reported low AUCs for SI (0.590) and modified SI (0.608) in predicting mortality [20]. Furthermore, there was no strong correlation with other outcomes such as ICU admission, mechanical ventilation, or the development of shock, which aligns with our results showing poor performance for ICU admission across all shock index variants. Consistently, Doğanay et al. reported that mortality increased significantly with age (26.9% in patients <56 years vs. 91.4% in those >77 years), and that oxygen saturation was a critical factor, as no patient with SpO₂ > 95% died [7].

In contrast, Hsieh et al. found better performance with the age-adjusted SI and HaSI, both showing AUCs > 0.7 and statistically significant associations with ICU admission, intubation, and mortality (p < 0.001) [6].

Combining SI with other clinical parameters may improve its diagnostic utility. In this regard, Eldaboosy et al. proposed a combination of SI > 0.7 and PaO₂/FiO₂ < 250 (SI-PaFi), which demonstrated notable performance: an AUC of 0.89 for ICU admission and 0.90 for mortality, with sensitivity and specificity above 85%, outperforming APACHE II and CURB-65 scores [21]. This approach may represent a promising modification for evaluating patients with COVID-19 and warrants further investigation.

These mixed findings are reflected in the broader literatura. A recent systematic review and meta-analysis by Alsagaff et al. (n = 4,576) showed that an elevated SI was associated with increased mortality (OR: 7.52; 95% CI: 3.72–15.19; p < 0.00001), with an AUC of up to 0.84, sensitivity of 78%, and specificity of 76.8%. It was also associated with ICU admission (OR: 5.81; 95% CI: 1.18–28.58; p = 0.03). However, the main limitation of this meta-analysis was the high heterogeneity among included studies (>95%), which restricts the generalizability of its conclusions [8]. Five studies were conducted in Turkey, with others in Saudi Arabia, Taiwan, and the Netherlands; none were from Latin America or similar middle-income settings. The studies used varying SI cutoffs (0.6 to 1.0), different outcome definitions (in-hospital, 14-day, 30-day, and 90-day mortality), and one study defined high SI as >0.7 combined with hypoxemia (PaO2/FiO2 < 250) [21], creating a composite index rather than evaluating SI alone. Notably, Van Rensen et al., also included in that meta-analysis, showed results consistent with ours. The substantial heterogeneity and the inclusion of studies with widely varying methodologies suggest that the pooled estimates may not be uniformly applicable across all clinical contexts, particularly in settings with different patient characteristics and healthcare resources like ours.

Several factors specific to our population and setting may explain SI’s limited discriminative capacity in our study. First, SI was originally designed for hemorrhagic and septic shock, where the heart rate-to-blood pressure ratio reflects intravascular volume depletion and compensatory cardiovascular responses. However, in moderate COVID-19 pneumonia, the primary pathophysiology is hypoxemic respiratory failure with initially preserved hemodynamic; patients may maintain relatively normal vital signs while developing progressive hypoxemia, pulmonary microthrombi, and inflammatory responses that do not necessarily manifest as classic shock physiology at emergency department presentation.

Second, our patients received standardized evidence-based COVID-19 treatment protocols including corticosteroids, anticoagulation, and early respiratory support. These measures may have altered the natural hemodynamic trajectory of the disease and diminished the predictive value of initial vital sign-based indices.

Third, as a tertiary care center in a middle-income country, ICU admission decisions may be influenced by resource constraints and bed availability, introducing selection pressures that differ from settings with greater critical care capacity. These contextual factors suggest that SI performance is highly setting-dependent, and our findings underscore the critical importance of prospective, context-specific validation before implementing any triage tool in clinical practice, particularly in Latin American healthcare systems and specific COVID-19 severity subgroups where data remain limited. Rather than contradicting the existing literature, our study contributes to understanding the external validity limitations of SI and highlights that pooled estimates from heterogeneous studies should be interpreted cautiously and validated locally before clinical implementation.

This study has several limitations that should be considered when interpreting its findings. First, it is a retrospective, single-center study, which limits the representativeness of our cohort and hinders the generalizability of the results to other institutions with different population characteristics, available resources, or clinical management practices. Furthermore, our final sample size was smaller than initially estimated, which may have reduced the statistical power to detect significant differences between shock index variants. This limited sample size may also have contributed to the modest AUCs observed and affects the precision of the diagnostic parameters calculated, particularly in subgroup analyses.

It is important to note that the data analyzed refer exclusively to the point of hospital admission. Most patients did not present with severe hypoxemia or require immediate support, which may explain the limited performance of the indices. Clinical instability may develop in later stages of the disease, suggesting the need for serial monitoring of the shock index and its variants during hospitalization, as well as the development of multivariable models that integrate the shock index with other clinical and laboratory markers to compare their performance against conventional prognostic tools.

Conclusion

Early identification of patients at risk for complications in moderate COVID-19 is crucial for optimizing hospital resources, particularly in settings with high healthcare demand. Our findings demonstrate that the shock index and its variants have limited discriminative ability as standalone predictors for mortality, ICU admission, and hospital length of stay in patients with moderate COVID-19. While some evidence suggests potential benefits of combining the shock index with other clinical parameters, significant heterogeneity across studies limits the generalizability of these findings. Our results indicate that shock index variants should not be relied upon in isolation for risk stratification in moderate COVID-19. Future research should focus on developing and prospectively validating multivariable prediction models that incorporate multiple clinical, laboratory, and imaging biomarkers, with careful attention to local contexts and patient populations.