Derivation and validation of a severity scoring tool for COVID-19 illness in low-resource setting

Pigoga, Jennifer L

Derivation and validation of a severity scoring tool for COVID-19 illness in low-resource setting

Doctoral Thesis

2021

Abstract

Background The COVID-19 pandemic has profoundly impacted some of the most vulnerable populations in lowresource settings (LRS) across the globe. These settings tend to have underdeveloped healthcare systems that are exceptionally vulnerable to the strain of an outbreak such as SARS-CoV-2. LRS-based clinicians are in need of effective and contextually appropriate triage and assessment tools that have been purpose-designed to aid in evaluating the severity of potential COVID-19 patients. In the context of the COVID-19 crisis, a low-input severity scoring tool could be a cornerstone of ensuring timely access to appropriate care and justified use of critically limited resources. Aim and objectives The aim of this research was to develop and validate a tool to assist frontline providers in rapidly predicting severe COVID-19 disease in LRS. To achieve this aim, the following objectives were defined: identify existing methods of risk stratification of suspected COVID-19 patients worldwide; establish predictors of severe COVID-19 illness measurable in LRS; derive a risk stratification tool to assist facility-based healthcare providers in LRS in evaluating in-hospital mortality risk; and validate tool SST in the African setting using real-world data. Methods To achieve the aim of this dissertation, quantitative and review methodologies were employed across four studies. First, a scoping review was conducted to identify all studies describing screening, triage, and severity scoring of suspected COVID-19 patients worldwide. These tools were then compared to usability and feasibility standards for LRS emergency units, to determine viable tool options for such settings. Following this, a systematic review and meta-analysis were undertaken to evaluate existing literature for associations between COVID-19 illness severity, and historical characteristics, clinical presentations, and investigations measurable in LRS. Three online databases were searched to identify all studies assessing potential associations between clinical characteristics and investigations, and COVID-19 illness severity. Data for all variables that were statistically analysed in relation to COVID19 disease severity were extracted and a meta-analysis was conducted to generate pooled odds ratios for individual variables' predictive abilities. In the third study, machine learning was used on data from a retrospective cohort of Sudanese COVID-19 patients to derive the AFEM COVID-19 Mortality Score (AFEM-CMS), a contextually appropriate mortality index for COVID-19. Following this, a fourth study was conducted with a more recent Sudanese dataset to validate the tool. Results The scoping review identified COVID-19 risk stratification 23 tools with potential feasibility for use in LRS. Of these, none had been validated in LRS. The systematic review then identified 79 eligible articles, including data from 27713 individual patients with laboratory-confirmed COVID-19. A total of 202 features were studied in relation to COVID-19 severity across these articles, of which 81 were deemed feasible for assessment in LRS. Meta-analysis of two demographic features, 21 comorbidities, and 21 presenting signs and symptoms with appropriate data available identified 19 significant predictors of severe COVID-19, including: past medical history of stroke (pOR: 3.08 (95% CI [1.95, 4.88])), shortness of breath (pOR: 2·78 (95% CI [2·24-3·46])), chronic kidney disease (pOR: 2.55 (95% CI [1.52-4.29])), and presence of any comorbidity (pOR: 2.41 (95% CI [2.01-2.89])). These significant predictors of severe COVID-19 were then considered for inclusion in the AFEM-CMS. Data from 467 COVID-19 patientsin Sudan were used to derive two versions of the tool. Both include age, sex, number of comorbidities, Glasgow Coma Scale, respiratory rate, and systolic blood pressure; in settings with pulse oximetry, oxygen saturation is included and, in settings without access, heart rate is included. The AFEM-CMS showed good discrimination: The model including pulse oximetry had a C-statistic of 0.775 (95% CI: 0.737-0.813) and the model excluding it had a C-statistic of 0.719 (95% CI: 0.678- 0.760). The tool was then validated against a second set of data from Sudan and found to once again have reasonable discriminatory power in identifying those at greatest risk of death from COVID-19: The model including pulse oximetry had a C-statistic of 0.732 (95% CI: 0.687-0.777) and the model excluding pulse oximetry had a C-statistic of 0.696 (0.645-0.747). Conclusions and relevance This dissertation establishes what is, to our knowledge, the first COVID-19 mortality prediction tool intentionally designed for frontline providers in LRS and validated in such a setting. The derivation and validation of the AFEM-CMS highlight the feasibility and potential impact of real-time development of clinical tools to improve patient care, even in times of surge in LRS. This study is just one of hundreds of efforts across all resource levels suggesting that rapid use of machine learning methodologies holds promise in improving responses to pandemics and other emergencies. It is our hope that, in future health crises, LRS-based clinicians and researchers can refer to these techniques to inform contextually and situationally appropriate clinical tools and reduce morbidity and mortality.

Keywords

General Surgery

Reference:

Collections

PhD / Doctoral

Full item page