This study demonstrates the value of admission vital signs in predicting mortality in patients hospitalised with COVID-19. Lower oxygen saturation, elevated respiratory rate, lower diastolic blood pressure and elevated glucose were found to be significantly associated with in-hospital mortality and comprised components of a promising predictive model. Numerous studies describe the relationship between oxygen saturation, respiratory rate and glucose on outcome [3,4,5,6,7,8,9]; however, the significance of admission diastolic blood pressure is a relatively novel finding. Although this study aimed to assess the influence of admission vital signs as a predictor of COVID-19 outcome, age is another variable that is easily available on initial consultation and was included into the prediction model, yielding an improvement in predictive ability.
COVID-19 remains a major healthcare concern globally and in South Africa. Vital signs are routinely measured for all patients presenting to healthcare facilities across all levels of care, and may serve as an early marker of poor prognosis. Global studies reveal important associations with severity and outcome, which are, however, not necessarily generalisable to the local context given the heterogeneity of the South African population. This study aimed to address the paucity of such data, and secondarily to derive important practical applications for other developing countries as well.
The vital signs evaluated in this study are routinely measured on initial consultation and may provide an early indication of patients with COVID-19 requiring more intensive in-hospital monitoring and treatment. Moreover, the heavy burden of this disease in the public healthcare sector has resulted in patients being managed across all levels of care – district, regional and tertiary – despite substantial difference in resources and skills. Thus, the early identification of patients with vital signs predictive of a poor prognosis may allow for prompt referral to an appropriate center.
Lower oxygen saturation on admission was found to be an independent risk factor for mortality. For every 1 % increase in admission oxygen saturation, the odds of mortality decreased by 7.8%. Given the phenomenon of silent hypoxaemia described in patients with COVID-19, this may represent delayed presentation to a healthcare facility . Although this study focused on hospitalised patients, home oxygen saturation monitoring in non-hospitalised patients with mild infection may also serve as a valuable tool by providing an early warning sign to patients. A large retrospective South African study by Nematswerani et al. yielded significantly lower mortality rates in patents who utilised a pulse oximeter to monitor oxygen saturation at home vs. those that did not . Given the potential benefits, consideration ought to be given to provision of pulse oximeters to high-risk patients with COVID-19 being managed at home, with a low threshold for presentation to hospital.
Measurement of blood oxygen saturation using pulse oximetry is a useful non-invasive tool, however reports of overestimation of true arterial oxygen saturation in darkly pigmented patients raise some concern over its utility in African populations . Even though the non-survivors in this study demonstrated an overall higher mean oxygen saturation when measured by pulse oximetry versus arterial blood gas analysis (86% vs 84.79% respectively), the difference was small, especially at higher oxygen saturations, suggesting that pulse oximeters may be of value in home monitoring of stable patients even in the local population. Further research in this regard is necessary, in predominantly African populations, to identify the ideal method of oxygen saturation measurement for risk stratification - especially at lower oxygen saturation levels.
Several studies concur with our findings - revealing a significant relationship between a low oxygen saturation, elevated glucose and elevated respiratory rate on admission and adverse COVID-19 outcomes [3,4,5,6,7,8,9]; however, the significance of diastolic blood pressure is a novel finding. A study by Fei-Ka Li et al. revealed no significant difference in the mean blood pressure (systolic and diastolic) between patients with critical vs. severe disease with COVID-19. However greater systolic and diastolic blood pressure variation was observed in those with critical disease, and both systolic and diastolic blood pressure variation indices showed positive association with worse outcomes (p = 0.02–0.03 and p = 0.06–0.08 respectively) . However, despite revealing a significant association with mortality on univariate analysis in our study, diastolic blood pressure lost its statistical significance on multivariate analysis – possibly suggesting confounding.
This study failed to demonstrate any significant association between systolic blood pressure, heart rate and temperature on admission and in-hospital mortality. In keeping with this, most other studies found no significant association between admission fever and mortality [11, 12]. Tharakan et al., however, analysed body temperature in 7614 patients with COVID-19 and identified hypothermia as in important marker of poor prognosis in patients with an admission temperature < 36 °C, and even more so < 35.5 °C .
Hypotension and tachycardia are key features of advanced disease and have been shown to indicate a poor prognosis in various generic sepsis-related risk scores . However, a study by Caillon et al. demonstrated high systolic blood pressure measurement on admission as an important component of mortality predication models . Advanced age (with comorbid hypertension) is a major risk factor for mortality in COVID-19 patients, thus it is uncertain whether this represents the burden of uncontrolled hypertension in the deceased population or occurred as a consequence of systemic inflammation and/or interference with angiotensin-converting enzyme 2 (ACE-2) enzymatic activity by SARS-CoV-2 .
Interestingly, although cardiac involvement usually occurs in association with systemic disease, there exists reports of isolated pericardial involvement with SARS-CoV-2 infection – demonstrating the expanding spectrum of cardiac affectation in COVID-19 .
The utilisation of vital signs in disease and outcome prediction has proven to be of immense value. With regards to sepsis prediction in critically ill adults, Mohammed et al. developed a prediction model utilising minute-to-minute physiological data only (heart rate, blood pressure and respiratory rate) and was able to accurately predict sepsis a mean of 17.4 h before sepsis onset with an average test accuracy of 83% . Van Wyk et al. similarly developed a prediction model comprising a minimal set of continuous routinely measured vital signs only – heart rate, respiratory rate, systolic and diastolic blood pressure, temperature and oxygen saturation – and was able to predict sepsis a mean of 5 h prior to onset . Furthermore, the addition of white cell count did not improve model sensitivity. This emphasises that effective prediction models comprising vital signs only exist, and may serve as a critical predictive modelling tool in the realm of COVID-19 as well.
Numerous COVID-19 risk prediction models have been developed; however, they are not necessarily applicable to the context of a developing country given the resource constraints. Further studies with larger cohorts need to be conducted to assess the performance of a prediction model comprising admission vital signs, together with appropriate external validation. Ideally, the models should be sensitive to the resource limitations of the public healthcare sector – comprising variables that are consistently measured and readily available, allowing for utilisation across various levels of care. Despite nearly 2 years since inception, COVID-19 remains a prime healthcare concern globally and in South Africa. Resource limitations coupled with the emergence of novel strains of virus and vaccine hesitancy emphasise the need to remain vigilant. The development and implementation of COVID-19 risk prediction models, comprising easily available parameters, may serve as a vital tool in upcoming waves of infection in resource-constrained settings. Further research in this regard would be forthcoming.
Our study comprised a relatively small sample size spanning June–September 2020. Ever since, newer strains of virus have emerged with postulated differences in behaviour. Thus, the study may not represent characteristic of the current and/or future strains of SARS-CoV-2. Again, highlighting the need for continued research of such nature. Furthermore, the study was limited to the public healthcare sector and did not equally represent race, age groups and socioeconomic status of the population. In addition to admission vital signs, other parameters (comorbidities etc.) are also readily available on initial consultation and may have provided value when integrated into the risk prediction model. The aim of this study is to demonstrate the importance of admission vitals in early risk stratification and its association with COVID-19-related mortality. More extensive machine learning algorithms are beyond the scope of objectives. Hopefully this article prompts further research and development of machine learning models (using larger sample sizes) specifically comprising easily available parameters and targeting developing populations with resource limitations.