Acute Respiratory Mortality Surveillance (ARMS) for Coronavirus Infection (COVID-19): A globally relevant technology to strengthen mortality surveillance for acute respiratory deaths in many countries lacking complete medical certification of death
To develop an enhanced verbal autopsy module to identify deaths from COVID-19. This will serve as a model for the next novel pathogen-as near as possible to real time in settings without routine medical certification of death.
- To distinguish COVID-19 from other causes of respiratory deaths using an "Acute Respiratory Mortality Surveillance" (ARMS) module
- Establish baseline distributions of usual acute respiratory deaths, as a comparator for COVID-19 deaths, and to inform modelling.
- Improve mortality assessments for any subsequent COVID-19 waves.
The current global infectious threat, COVID-19, has not yet been widely detected in sub-Saharan Africa or other low income countries in Asia. It is almost inevitable that it will reach those places. While unusual spikes in infection-related deaths can register quickly in higher income countries and in China, they can go unrecognized for weeks or months in low-income settings where even very ill people do not go to a hospital, infecting others. Detecting a mortality signal is important and may be the first step in recognizing a serious outbreak. We propose to build on our extensive experience using verbal autopsy (VA) in the long-running Indian Million Death Study, and ongoing studies in China, Hong Kong, Ethiopia and Sierra Leone to develop an enhanced verbal autopsy module to identify deaths from COVID-19. This will serve as a model for the next novel pathogen-as near as possible to real time in settings without routine medical certification of death. We will test three hypotheses: #1 An "Acute Respiratory Mortality Surveillance" (ARMS) module can be added quickly to the WHO VA instrument and validated against hospitalized cases and deaths (paired with epidemiological information and machine learning) to distinguish COVID-19 from other causes of respiratory deaths. #2 Early deployment of ARMS in China, Hong Kong, India, Sierra Leone, and Ethiopia will help establish baseline distributions of usual acute respiratory deaths, as a comparator for COVID-19 deaths, and to inform modelling. #3 Effective knowledge translation of an open-source, widely-available ARMS module will improve the global response to COVID-19, particularly in the lowest income countries and help to improve mortality assessments for any subsequent COVID-19 waves. A successful ARMS will contribute to stopping the current outbreak and add novel surveillance tools. All materials and results will be made available globally to ensure the broadest use.
Design: Retrospective cause of death study using electronic verbal autopsy, dual physician coding, and machine coding
Setting: China: hospital records from Zhongda Hospital, Nanjing; Hong Kong: hospital records; India: hospital records or a community disease register in Amravati (a tribal district) in rural Maharashtra and Anand district in Gujarat; Sierra Leone: urban/peri-urban areas of Bo City and within enumeration areas from nationally representative surveys of deaths; Ethiopia: death regisration in East Gojjam and North Showa rural districts.
Methodology: #1 Create and validate the ARMS module: rapid review of diagnostic guidelines for COVID-19 and the published literature on symptoms, signs, and other key features, including contact history among severe COVID-19 cases; clinician expert consultation on symptom patterns and chronologies to be the basis of the COVID-19 differential diagnosis; mapping the list of symptoms to our existing e-VA tool to determine gaps, and then develop a checklist and sequence of symptoms; ODK conversion of questions into Android-based phone/tablet applications. #2 Deploy ARMS in five countries: implementation of ARMS among both confirmed COVID-19 and suspected acute respiratory deaths; collaborator organization of teams to interview households of the deceased; automated quality checks of each e-VA collected and storage in a secure cloud server; assignment of ICD-10 codes and three COVID-19 categories (none, possible, and certain) to each death by local, trained physicians using ARMS data with location, seasonality, contact history, influenza patterns, etc.; anonymous reconciliation of differences between physician coders and adjudication of persistent differences by a senir doctor; adaptation of the InSilico VA algorithm and natural language processing to the e-VAs, including symptoms suggestive of COVID-19 diagnosis versus other causes. #3 Knowledge translation: online training for disease surveillance program officers on how to use ARMS as an investigative tool for outbreaks and unusual clusters.
Outcomes: Sensitivity/specificity and ROC curves for COVID-19 versus other etiologies; causes of death in five countries using physician and automated/NLP assignment; cause-specific mortality fractions for acute respiratory deaths; cluster detection in each setting