The World Health Organisation (WHO) report on excess mortality due to Covid-19, puts India’s excess mortality -- people who might not have died if there had been no pandemic -- at a staggering 4.75 million as against the official government count of about 0.5 million cumulative deaths linked to Covid-19, as of May 2022. The WHO estimate is nearly 10 times the government count. The politicisation of the discourse and the raging debate that it has sparked between the supporters and opponents of the government on who is right and who is wrong is quite beside the point. The wide discrepancy in numbers is not trivial and has long-term implications that must, from a public governance perspective, concern all citizens.
The real question is: What must we do to improve our health information system and Civil Registration (CR) data system such that we generate more scientific and accurate data that can provide decision support to improve health outcomes.
Deaths are life events recorded as ‘vital statistics’ in government parlance. Routine statistics are maintained by the CR Systems to facilitate policymaking and target-setting in public health interventions. Mortality statistics support epidemiological and policy research, the findings from which may have positive implications for arresting premature mortality. Excess deaths are a proxy for the mortality effects of SARS-CoV-2 infection. The key assumption is that increases in all-cause mortality during peak weeks of Covid-19 compared to pre-pandemic periods are nearly all due to the infection, even if SARS-CoV-2 infection was not confirmed. The excess death estimates are subject to substantial variation and also include indirect effects of the pandemic and the effects of measures taken.
Much debate about Covid-19 death attribution has centred around the use of testing to establish the diagnosis. Under-counting of deaths may occur when no testing is done or testing is false-negative, and over-counting may ensue from false-positive testing. Clinically-based attribution of death causes may correct some testing misclassifications or may add further misclassifications.
Eventually, the question to ask is not whether Covid-19 deaths are over-counted or under-counted. Improving processes for registration of deaths, mortality data reporting, and clinical attribution of both hospital and domiciliary deaths in the future should constitute a matter of concern.
The tragic loss of life during the Covid-19 pandemic must be carefully measured, to illuminate the dynamics of the pandemic and the best use of interventions. Attribution of death typically uses the WHO guidance and various national guidelines. Yet, there are huge variations across countries and even between different health systems and physicians in the same country in how deaths are attributed. Even before the major reclassification of death causes due to Covid-19, registration of deaths and death certificates were known to be notoriously error-prone. Three million of India's annual 10 million deaths are not registered, with the largest gaps in poorer states and among women. Eight million deaths lack medical certification of the cause.
Covid-19 is a pandemic in which most deaths occur in people with several underlying diseases or co-morbidity conditions. Dissecting the relative contribution of each disease/condition to death can be difficult. Careful collection of information on patient characteristics, comorbidities and their severity is essential to get reliable estimates for death counts. Notably, the infection fatality rate is markedly higher in hospital or nursing home patients than in the community-dwelling persons of the same age; and the difference can be rather wide, if limited to people institutionalised in palliative care for terminal disease. The extent to which deaths of patients in palliative care with minimum life expectancy are attributed to Covid-19, therefore, varies across countries and locations. There is large variation across countries on the percentages of people who die at home, at the hospital, or in institutionalised care.
These settings may differ in how they pursue diagnosis (or over-diagnosis) of Covid-19 as cause of death. One may question whether Covid-19 should ever be listed as primary cause of death in patients with known terminal disease. Regardless, person-year calculations would be less biased, if this background information becomes available. Mistrusting reported Covid-19 death counts, several analysts focus on excess death assessments. Excess deaths should be scrutinised for death causes accentuated by the pandemic versus by measures taken against the pandemic: for instance, deaths due to disruption of essential healthcare services, deferring much needed life-saving surgery, non-availability of hospital beds for non-covid emergencies, suicides, diseases of despair, starvation, tuberculosis, and more. Meticulous audit of medical records may offer insights, but even these records are likely unreliable at present.
Overall, given these difficulties, equating excess mortality to Covid-19 itself is probably naïve and flawed. Excess mortality in 2020–2021 has been substantially higher than reported Covid-19 deaths in several countries across the globe and not just in India. Regardless of the methodology adopted by various agencies, each of these estimates has shortcomings; and therefore, estimating Covid-deaths with statistical confidence may prove elusive. But all estimates suggest that the death toll in India from the pandemic is likely to be an order of magnitude greater than the official count of about 0.5 million.
To improve the quality of CR-generated vital statistics of deaths it would be useful to: (a) reclassify ‘mandatory’ fields in the CR database such that Medically Certified Cause of Deaths aligned with International Classification of Diseases (ICD) are mandatorily entered if a death is medically certified; and (b) reform ‘cause of death’ certification by increasing the coverage of hospitals under the umbrella of the Medically Certified Cause of Death (MCCD) scheme and by mandating and universalising medical certification of deaths.
As a possible next step, a Machine Learning model should be developed to predict mortality risk: for different demographic groups; for different medical care pathways; and for each of the main killers modelled after the (i) Reynolds Risk Score and (ii) a Poisson regression model to estimate variable parameters. Understanding and engaging with the data-based estimates is necessary because in this morbid process of counting deaths, attendant accountability, will matter now but even more in the future.
(The writer is Director, Public Affairs Centre)