The coronavirus disease 2019 (COVID-19) pandemic exposed critical weaknesses in healthcare systems worldwide, particularly in low-resource settings, where limited access to diagnostic and therapeutic tools posed significant challenges to effective patient management, resulting in widespread infections and mortality. By the end of December 2021, more than 282 million confirmed COVID-19 cases and over 5.4 million deaths had been officially reported globally.¹ However, these numbers likely represent a substantial undercount since reporting systems were not optimal. Estimates of the global excess mortality suggest that the true death toll may have exceeded 14 million in the first 2 years of the pandemic, reflecting limitations in testing, reporting and vital registration systems, particularly in low- and middle-income countries.² In Kenya, approximately 295 000 confirmed cases and 5378 deaths were reported by December 2021.³ However, national seroprevalence surveys and modelling studies suggest that actual infections and deaths were significantly higher than reported, because of limited testing capacity, healthcare access barriers and underreporting of community deaths.^2,4,5,6 These figures underscore the urgent need for more effective triage and case management strategies in resource-constrained healthcare systems to reduce preventable mortality during such health crises.

Coronavirus disease 2019 mortality is driven by age, male sex, comorbidities (e.g., cardiovascular disease, diabetes) and markers of severe illness such as inflammation, respiratory failure and hypoxia. Among these, hypoxia – particularly ‘silent hypoxia’, where patients present with low saturation of peripheral oxygen (SpO₂) without subjective symptoms like shortness of breath – has emerged as a powerful yet under-recognised predictor of mortality. Hypoxia, often overlooked during routine clinical evaluations, has been strongly linked to delayed intervention and increased mortality.^7,8,9 Prior studies have shown that patients with ‘silent hypoxia’ exhibit alarmingly low SpO₂ levels despite appearing clinically stable, leading to missed opportunities for timely intervention.^10,11 For instance, while some studies have found no significant difference in outcome between silent and dyspnoeic hypoxia,¹² others have reported higher risks of intensive care unit (ICU) admission or mechanical ventilation associated with silent hypoxia,¹³ underscoring how the deceptive presentation of silent hypoxia complicates triage – especially in overwhelmed or low-resource settings. Hypoxia is a hallmark of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection, reflecting impaired gas exchange in the lungs and is associated with inflammatory responses, including the cytokine storm observed in critical cases.^14,15 Multiple studies have confirmed that hypoxia correlates with COVID-19 severity, reinforcing its role as a key predictor of poor outcomes.^16,17

Non-invasive pulse oximetry has emerged as an essential and cost-effective tool for early identification of hypoxia. By measuring SpO₂, it provides vital information for guiding clinical decisions, such as hospitalisation and oxygen therapy.¹⁸ Normal SpO₂ levels range from 95% to 100%, while values ≤ 94% are considered hypoxic according to clinical guidelines, including those used in Kenya.^19,20 The World Health Organization (WHO) recommended the use of routine pulse oximetry in COVID-19 case management because of its simplicity, speed, low cost and non-invasiveness.^21,22 Saturation of peripheral oxygen alone or in combination with the fraction of inspired oxygen (SpO₂/FiO₂) has proven valuable and economical for monitoring respiratory function over time in patients with COVID-19.²³ However, in resource-constrained settings, the underutilisation of pulse oximetry – combined with limited oxygen supplies and staffing shortages – exacerbated challenges in recognising and managing hypoxia.^24,25 Moreover, traditional diagnostic tools, such as chest X-rays, often failed to correlate with hypoxia, further complicating management.²⁶ Patients with pre-existing chronic conditions – such as chronic obstructive pulmonary disease, cardiovascular disease or obesity – were particularly vulnerable to poor outcomes, highlighting the need for prompt SpO₂ monitoring.^{27,28,29,30,31,32} These findings are consistent with a recent multicentre study that evaluated predictors and frequency of silent hypoxia in COVID-19 patients. The authors reported a silent hypoxia prevalence of approximately 30%, with silent hypoxia independently associated with increased ICU admissions and mechanical ventilation.¹³ Notably, this form of hypoxia often remained undetected without the aid of pulse oximetry, reinforcing the need for systematic SpO₂ monitoring at admission, particularly in resource-limited settings where clinical signs alone may not reveal hypoxia.

We hypothesised that hypoxia at admission is an independent predictor of in-hospital mortality among COVID-19 patients. This study investigates the prevalence and clinical impact of silent hypoxia among COVID-19 patients admitted to three referral hospitals in Kenya. Analysing patient outcomes, our study underscores the critical role of SpO₂ monitoring in reducing mortality and improving hypoxia management in resource-limited healthcare systems. In addition, we explored the availability of pulse oximetry, the prevalence of hypoxia and its association with mortality during the pandemic, offering actionable insights to address these challenges in similar settings.

This study highlights the critical role of hypoxia in determining the outcomes of patients hospitalised with COVID-19. We found that approximately 8 out of 10 patients hospitalised with COVID-19 had hypoxia (SpO₂ ≤ 94%) at admission, a condition that was associated with significantly higher mortality rates compared to those with normal SpO₂. The prevalence of hypoxia was particularly higher among older patients and those with clinical symptoms consistent with COVID-19, such as cough, chest tightness, low pulse rate, dyspnoea, and underlying comorbidities like hypertension and HIV infection. This relationship suggests that symptomatic COVID-19 disease may have a consistent prognosis even in the absence of diagnostic testing, a finding supported by similar studies.³⁶ Notably, about 40% of our patients experienced silent hypoxia, a proportion that aligns with estimates from other studies, which range between 4.8% and 65%.³⁶ This high burden of hypoxia upon admission emphasises the need to evaluate its prognostic significance, particularly its relationship with mortality and survival outcomes among hospitalised patients.

SpO₂ has emerged as a crucial predictor of survival in COVID-19 patients.²⁷ Mortality among hypoxic patients was 38.0%, nearly threefold that of non-hypoxic patients (13.6%). This is consistent with literature indicating that hypoxia is a key predictor of poor prognosis in respiratory infections.^27,28,37 Moreover, older hypoxic patients had twice the risk of death compared with patients under 40, which is consistent with global findings.³⁸ While hypertension was not independently associated with mortality in our study, it is generally recognised as a risk factor for severe COVID-19 and poor outcomes.³⁹ Our findings suggest that normal SpO₂ levels are associated with better survival probabilities, while hypoxia, along with age and dyspnoea, independently predicts a higher risk of death. Therefore, oxygen supplementation is essential for hypoxic patients, as studies have shown that higher SpO₂ levels after supplementation are linked to reduced mortality,³⁶ and improved survival probabilities.²⁸ The observation that patients with hypoxia had over three times the mortality rate underscores the importance of assessing SpO₂ in all COVID-19 patients, regardless of their symptomatic presentation.

Oxygen supplementation is recommended for hypoxic patients. Although oxygen supplementation is a cornerstone of COVID-19 management, our study’s observational design limits causal inference regarding its impact on survival. While hypoxic patients who received oxygen had lower survival probabilities – likely reflecting underlying disease severity rather than the effect of oxygen itself – this should not be interpreted as evidence that oxygen provision is ineffective. Instead, our findings highlight the urgent need for timely and equitable access to oxygen in line with global recommendations. Studies elsewhere have shown that when administered early, oxygen therapy improves survival outcomes in hypoxic patients.^27,35 Thus, our results reinforce the global consensus on the importance of oxygen access, while recognising that real-world outcomes in resource-constrained settings may be influenced by multiple co-occurring factors such as delayed presentation, limited monitoring or inadequate dosing strategies. It is important to note that only 3 out of 5 (68.6%) of hypoxic patients in our study received oxygen supplementation, reflecting gaps in care delivery. The national guidelines recommend oxygen supplementation for patients with SpO₂ ≤ 94%. While this threshold was the standard at the time, emerging evidence has since suggested that stricter thresholds (e.g., < 92%) might improve early detection of deterioration and might have been beneficial.⁴⁰ We also lacked data on whether follow-up SpO₂ readings were taken after initial drops by 3% – 5%, which is a recommended practice, as has been suggested by Galwankar et al.⁴¹ Community-level SpO₂ monitoring, such as providing oximeters to households,⁴² could be beneficial for ongoing patient management post-discharge, though we did not follow patients after discharge.

Our study had some limitations. We lacked COVID-19 diagnostic information for some cases, relying instead on clinician documentation based on symptomatic presentation. Additionally, SpO₂ data were missing for 5.2% of records, but this low proportion is unlikely to bias our findings. The relatively stringent definition of hypoxia used in this study may have led to the misclassification of some patients as having silent hypoxia, though we adhered to national guidelines. The prevalence of hypoxia in this study was higher than in many global settings. However, our purposive selection of high-volume referral hospitals likely captured a sicker population and may have inflated these estimates. This selection bias, along with the retrospective nature of the study and variability in documentation, limits generalisability to the broader Kenyan context, including patients with milder disease managed in lower-level facilities. Furthermore, retrospective data reviews are prone to missing information, inconsistent clinical documentation and recall bias, all of which may affect data accuracy. To mitigate these issues, we made concerted efforts to verify patient chart contents and collaborated closely with facility health records officers to enhance data quality. A key limitation is that the association between hypoxia and mortality may be confounded by treatment gaps, as not all hypoxic patients received oxygen, making the hazard ratio reflect both disease severity and inadequate management in a low-resource setting. We also did not track patient outcomes of COVID-19 patients post-discharge. Finally, we could not determine whether hypertensive patients had a history of hypertension or elevated blood pressure before admission. Despite these limitations, our study provides valuable insights into the utility of SpO₂ monitoring for the clinical management of hospitalised COVID-19 patients.

Our findings suggest that SpO₂ monitoring is essential in managing hospitalised COVID-19 patients, especially in low-resource settings. While comorbidities may play a role in disease progression, hypoxia and age were the strongest independent predictors of mortality in our cohort. Importantly, the frequent occurrence of silent hypoxia supports its inclusion in triage protocols. Our findings highlight critical patterns in clinical presentation and outcomes that are likely relevant to similarly resourced public facilities across Kenya. Nonetheless, we caution against extrapolating these findings to rural or private settings without further investigation and recommend future studies with broader facility representation. Future research should directly model silent hypoxia as a primary exposure variable to better quantify its independent association with mortality. In addition, studies that incorporate broader severity indicators and follow-up outcomes will be vital in refining clinical care strategies for similar low-resource settings.