key: cord-0957091-7nvxh83t authors: Otiende, M.; Nyaguara, A.; Bottomley, C.; Walumbe, D.; Mochamah, G.; Amadi, D.; Nyundo, C.; Kagucia, E.; Etyang', A.; Adetifa, I.; Maitha, E.; Chondo, E.; Nzomo, E.; Aman, R.; Mwangangi, M.; Amoth, P.; Kasera, K.; Ng'ang'a, W.; Barasa, E.; Tsofa, B.; Mwangangi, J.; Bejon, P.; Agweyu, A.; Williams, T.; Scott, A. title: Impact of the COVID-19 epidemic on mortality in rural coastal Kenya date: 2022-04-07 journal: nan DOI: 10.1101/2022.04.06.22273516 sha: d68f01a45abbe0dca86f9cafec5210a86f0cdb59 doc_id: 957091 cord_uid: 7nvxh83t Background: The impact of COVID-19 on all-cause mortality in sub-Saharan Africa remains unknown. Methods: We monitored mortality among 306,000 residents of Kilifi Health and Demographic Surveillance System, Kenya, through four COVID-19 waves from April 2020-September 2021. We calculated expected deaths using negative binomial regression fitted to baseline mortality data (2010-2019) and calculated excess mortality as observed-minus-expected deaths. We excluded deaths in infancy because of under-ascertainment of births during lockdown. In February 2021, after two waves of wild-type COVID-19, adult seroprevalence of anti-SARS-CoV-2 was 25.1%. We predicted COVID-19-attributable deaths as the product of age-specific seroprevalence, population size and global infection fatality ratios (IFR). We examined changes in cause of death by Verbal Autopsy (VA). Results: Between April 2020 and February 2021, we observed 1,000 deaths against 1,012 expected deaths (excess mortality -1.2%, 95% PI -6.6%, 5.8%). Based on SARS-CoV-2 seroprevalence, we predicted 306 COVID-19-attributable deaths (a predicted excess mortality of 30.6%) within this period. Monthly mortality analyses showed a significant excess among adults aged [≥]45 years in only two months, July-August 2021, coinciding with the fourth (Delta) wave of COVID-19. By September 2021, overall excess mortality was 3.2% (95% PI -0.6%, 8.1%) and cumulative excess mortality risk was 18.7/100,000. By VA, there was a transient reduction in deaths attributable to acute respiratory infections in 2020. Conclusions: Normal mortality rates during extensive transmission of wild-type SARS-CoV-2 through February 2021 suggests that the IFR for this variant is lower in Kenya than elsewhere. We found excess mortality associated with the Delta variant but the cumulative excess mortality risk remains low in coastal Kenya compared to global estimates. By December 2021, the global mortality attributable to COVID-19 was more than 5.4 million 1 . In Europe and South America, there were 204 and 274 COVID-19 deaths 6 usually acquired through decennial censuses, 3-5 yearly cluster sample surveys (Demographic and Health Surveys) or open cohort studies (Health and Demographic Surveillance Systems). Here, we use mortality data from the Kilifi Health and Demographic Surveillance System (KHDSS), a surveillance population of over 300,000 residents followed for 20 years, to examine the mortality impact of COVID- 19 in an area typical of rural Kenya. In a random population survey in the KHDSS between December 2020 and April 2021, 14.5% of children (0-15 years) and 25.1% of adults (16 years) were anti-SARS-Cov-2 IgG positive, suggesting widespread transmission by early 2021 19 . We examine mortality trends up to 23 rd September 2021 and investigate the impact of the pandemic on morbidity using a hospital surveillance system linked to the KHDSS. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) 8 this population. Admissions to KCH are linked in real-time to the KHDSS population register to provide a hospital-based passive disease surveillance system. We restricted morbidity analysis to adults aged 15 years or older and analyzed the patterns of clinical pneumonia admissions and all admissions. Clinical pneumonia, an existing admission diagnosis on our surveillance system, was defined as the presence of any two of the following symptoms within the 14 days before admission: cough, fever, chest pain, crackles, haemoptysis and dyspnoea. Rates of mortality and hospital admissions were calculated as the number of deaths or admissions divided by person years of observation (PYO). PYO were calculated as time from the latest of birth or in-migration or study start date to the earliest of study end date or outmigration or death. Individuals' periods of residence outside the KHDSS area were excluded. Models fitted to rates of mortality, all admissions and pneumonia admissions in 2010-2019 were used to predict these outcomes in 2020-2021. We chose January 2010 to December 2019 as the baseline period because mortality rates were stable during this decade 21 . We censored all analyses on 23 rd September 2021, the date our 50 th KHDSS re-enumeration round was completed, and we locked our demographic database on 14 th January 2022, the date of completion of the 51 st round ( Figure S1 ). We modelled monthly death counts using negative binomial regression. The model included a log-linear trend, sine and cosine terms to account for seasonality, and an offset to account for changes in person years of observation (Equation S1). It was fitted using the glm.nb function in R. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 7, 2022. ; We modelled the rates of monthly admissions as a local-level time series model 24 to account for the stochastic trend of admission rates (Equation S2). The model was fitted, without covariates, to pre-pandemic hospital admissions. The months January-December 2017 were excluded as missing because of a prolonged healthcare workers' strike. Fitting was done using the Arima function in R; the local level model is equivalent to an ARIMA (0,1,1) model. We modelled the log rate of pneumonia admissions using a linear regression model with ARMA errors to account for autocorrelation. Because there were zero admissions in some months, we used quarterly rates of admission. The model included a linear trend term, but no terms for seasonality (Equation S3 ). To fit the model, we used the Arima function in R. Excess mortality was calculated as the difference between the observed and predicted number of deaths in the COVID-19 period expressed as a percentage of predicted number of deaths or as a risk based on the initial population size (per capita excess mortality). We used 1st April 2020 as the start of the analysis period because the interval from infection to death is approximately 2-3 weeks and it is unlikely there would have been any COVID-19 deaths before April 2020. We analysed two periods of cumulative excess mortality; in the first, we censored deaths and PYO at 17 th February 2021, the reference date for the KHDSS serosurvey of SARS-CoV-2 antibodies; in the second, we censored at 23 rd September 2021. In each analysis we included any events captured on or before 14 th January 2022. In the last 10 re-enumeration rounds, 99% of deaths were ascertained by the end of the first complete re-enumeration round after the death (Table S2 ). The remaining . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 7, 2022. ; https://doi.org/10.1101/2022.04.06.22273516 doi: medRxiv preprint 1% were not detected until a second interview was undertaken because the initial respondent, who may have been a member of an adjacent household, may not have been aware of the death. To adjust for this under-ascertainment bias we divided the number of observed deaths occurring between 4 th May and 23 rd September 2021 ( Figure S1 ) by 0.99. The KHDSS population is re-enumerated every 4 months because infant deaths may be missed with longer intervals; a child may be born and die without any enumeration contact. The suspension of the KHDSS field operations on 23 rd March 2020 extended the re-enumeration interval to 11 months and it is likely that the ascertainment of infant deaths was reduced in this period. Therefore, we excluded infants from the estimation of excess mortality. In-migrants may also enter and die before they can be enumerated. To exclude mortality changes attributable to variable detection of in-migrants, we compared the annual mortality risk of the cohort of KHDSS residents on 23 rd March in 2020 to similar snapshot cohorts on 23 rd March in each year of the baseline period. Expected COVID-19 deaths were calculated as the product of the global IFR for SARS-CoV-2 infection 25 , the cumulative incidence of SARS-CoV-2 and the number of KHDSS residents in each age stratum. We calculated confidence intervals for the expected number of COVID-19 deaths using the delta method 26 . The cumulative incidence of SARS-CoV-2 was derived from a serosurvey of a random sample of 850 KHDSS residents between 1 st December 2020 and 28 th April 2021 (median sample date 17 th Feb 2021). The serosurvey sampled 100 individuals in each 5-year age stratum from 0-14 years, 50 individuals in each 5-year age . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 7, 2022. stratum between 15-64 years and 50 individuals aged 65 or greater 19 . We compared deaths predicted by this calculation with the total number of excess deaths observed in all KHDSS residents aged 1 year or more between 1 st April 2020 and 17 th February 2021. We calculated cause-specific mortality fractions in 10 categories; the 9 leading causes of death in children and adults, namely, Acute Respiratory Infections (ARI), unspecified cardiac disease, stroke, pulmonary tuberculosis, malaria, HIV-related deaths, digestive neoplasm, acute abdominal conditions, and Road Traffic Accidents (RTA), and all other causes. To identify COVID-19-specific deaths we used the 6 additional VA COVID-19 questions; for any death with at least one positive response, we applied two discriminating processes. Firstly, we invited two independent reviewers to conduct Physician Certified Verbal Autopsy (PCVA) using clinical information collected during VA interview and classified each death as a probable, possible or unlikely COVID-19 death. Discordant cases were resolved jointly by the reviewers. Secondly, we used the COVID-19 Rapid Mortality Surveillance (CRMS) software 27 , which is a simplified version of the probabilistic modelling methods used in the InterVA-4 models, to derive the probability that the death was COVID-19 related. We used a probability cut-off value of 0.89 based on a validation study conducted in Brazil 28 . We analyzed 15,379 deaths and 12,295 hospital admissions between 1 st Jan 2010 -23 rd September 2021 with a total of 3,157,610 PYO. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. During the pandemic period, the observed rate of all admissions was generally lower than the predicted rates in all age subgroups. Pneumonia admission rates declined in the pre-pandemic period across all age groups ( Figure 1E-H) . For all adults, the incidence declined from approximately 100 cases per 100,000 PYO in 2010 to approximately 50 cases per 100,000 PYO at the end of 2019 ( Figure 1E ). As with all admissions, the observed rate of pneumonia admissions during the pandemic was lower than the predicted rates in all age subgroups. Monthly observed mortality rates did not vary consistently from the predicted rates Between 1 st April 2020 and 17 th February 2021, we predicted 1,012 deaths beyond infancy but observed 1,000 (excess mortality -1.2%, 95% PI -6.6%, 5.8%, Table 1 ). Excess mortality was significantly negative among neonates (-25.8% 95% PI -37.5, -1.5%) and older infants (-38.0%, 95% PI -48.8, -16.5%, Table S3 ) up to 17 th . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 7, 2022. ; February 2021. Between 1 st April 2020 and 23 rd September 2021, we predicted 1,718 deaths beyond infancy but observed 1,768 (Table 1) ; this gives an excess mortality of 2.9% (95% PI -1.0%, 7.8%) and excess mortality risk of 16.8/100,000. After adjusting for under-ascertainment of deaths occurring between 4 th May-23 rd September, excess mortality was 3.2% (95% PI -0.6%, 8.1%) and excess mortality risk was 18.7/100,000 (Table S5 ). The adjusted excess mortality deviated significantly from zero only among adults aged ≥ 65 years (11.0% 95% PI 5.0%, 19 .8%). Based on the SARS-CoV-2 seroprevalence estimates and the global IFR, the cumulative number of expected COVID-19 related deaths up to 17 th February 2021 was 306 (95% CI, 249-376, Table 2 ) which represents a total excess mortality of 30.6%. This model did not predict any deaths in infancy and, within this same period, we observed 12 fewer deaths than predicted, giving a negative excess mortality of - There was no evidence that field interviewers were unable to reach appropriate respondents during the pandemic leading to under-reporting of deaths (Table S6 ). In annual period survival analyses for the KHDSS cohorts that were resident on 23 rd March each year between 2010 and 2020, there was no evidence of decreased survival in 2020 (Table S7, Figure S5 ). proportion attributable to ARI fell in 2020 in all age groups but returned to pre-. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The number of patients diverted from the KCH general wards to these isolation facilities was small, not all of them were residents of the KHDSS, and even after . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 7, 2022. ; https://doi.org/10.1101/2022.04.06.22273516 doi: medRxiv preprint accounting for these additional hospital admissions we did not observe a signal of increased hospital utilization ( Figure S4 ). In several countries, the pandemic has led to an increase in deaths at home because patients have been unable or unwilling to travel to hospital for fear of acquiring SARS-CoV-2 32 . It is possible that COVID-19 patients, resident in the KHDSS area, stayed at home for similar reasons. In Kilifi, the excess mortality analysis showed no evidence of an increase in deaths among adults during the first 11 months of the pandemic. Because of the disruption to fieldwork in 2020, infants who were born and died between household visits may not have been reported to the field interviewers. We found a 30% reduction in infant mortality between 1 st April 2020 and 17 th February 2021 (Table S3 ) which is best explained by this under-ascertainment and we therefore excluded infants from our analyses of overall mortality. This problem is unlikely in resident children aged 1 year or more, whose births had already been captured by the demographic surveillance before the disruption. Similarly, unhealthy in-migrants, who migrated and died in the period when fieldwork was suspended, may not have been reported. Our mortality analysis excluding in-migrant deaths did not reveal a reduction in survival in 2020 compared to previous years (Table S7 , Figure S5 ). In the verbal autopsy data, undetected COVID-19 deaths are likely to be identified as Acute Respiratory Infections (ARI). However, the proportion of all deaths attributable to ARI declined in the first year of the pandemic in all adult age groups and then returned to baseline levels in 2021 for most age groups. Among children aged <5 years, ARI declined as a cause of death from 26% in 2019 to 19% in 2020 and rose again in early 2021 consistent with a transient decline in social contacts and . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The estimate of cumulative incidence of SARS-CoV-2 in the first 11 months of the pandemic is highly dependent on the accuracy of ELISA for SARS-CoV-2 IgG. This assay, validated in Kenyan control populations, has a specificity of 99.0% (95% CI 98.1-99.5%) and sensitivity of 92.7% (95% CI 87.9-96.1%) 34 . The assay also has 97% agreement with the WANTAI total IgG ELISA adopted by WHO across Africa 35 . The prediction of 306 COVID-19 related deaths in KHDSS derived from the seroprevalence results relies heavily on estimates in older persons yet seroprevalence was estimated in only 101 adults in the three strata above 60 years, of whom 24 were seropositive. Collapsing these age strata and applying a seroprevalence of 24% to all adults aged ≥ 60 years would lead to a prediction of 234 deaths. Although lower, this still represents a substantial mortality excess which contrasts sharply with the observation of 12 fewer deaths in the same period, suggesting that the IFR of the wild-type virus in Kilifi was substantially lower than was estimated globally 7 . . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 7, 2022. ; Excess mortality in South Africa and North Africa 13,36 has been high during the pandemic period but mortality data from countries in other parts of Africa are sparse. Given seroprevalence estimates of 50% 37 , 60% and higher 38,39 in many settings in tropical Africa, the lack of a corresponding excess in detectable mortality raises questions about the severity of COVID-19 in this context. Although early waves of the pandemic did not result in an observable risk in mortality in Kilifi, we did observe a discrete peak of mortality coincident with the national Delta wave confined to adults aged ≥ 45 years. Normally, mortality reporting is subject to lags. To report Delta-wave mortality promptly we have adjusted observed estimates using ascertainment levels derived from many rounds of prior re-enumeration. This yields an excess mortality of 3.2% and an excess mortality risk of 18.7/100,000 across the first four waves of COVID-19 which compares with a global excess mortality risk of 120.3/100,000 16 . Taken together with hospital surveillance showing lower utilization during the pandemic and serological data confirming extensive spread of SARS-CoV-2 infection, these data confirm that the public health impact of the COVID-19 pandemic in coastal Kenya in 2020-21 has been substantially lower than elsewhere in the world. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 7, 2022. ; Individual verbal consent to participate in a continuous health and demographic surveillance system was sought at the household level using a specific informed consent form. Written informed consent was obtained by interviewers from all VA respondents. This study was approved by the Ethical Review Committee of the Kenya Medical Research Institute (approval number: KEMRI/SERU/CGMR-C/007/3057). Underlying individual data include geo-located residence and individual hospital records and hence would be high risk for identifiability. Intermediary data will be CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 7, 2022. ; version for submission. All authors declare no competing interests. We gratefully thank the residents of Kilifi who have participated in the surveillance activities of the KHDSS. We acknowledge the tremendous work of the verbal autopsy and census field staff, and data supervisors who collect and process this information, and the Community Liaison Group who run the community engagement programmes. This article is published with the permission of the Director of the Kenya Medical Research Institute. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 7, 2022. Table S5 ) . It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 7, 2022. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 7, 2022. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 7, 2022. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 7, 2022. ; Calculated as (observed deaths -expected deaths)/PYO. Excess mortality rate for all ages above infancy (≥ 1 year) is the weighted average of the age-specific rates. The weights are the proportion of each age-group in the KHDSS population. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 7, 2022. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 7, 2022. ; Max Roser Coronavirus Pandemic (COVID-19) The Conundrum of Low COVID High seroprevalence of SARS-CoV-2 but low infection fatality ratio eight months after introduction in SARS-CoV-2 Seroprevalence in a Rural and Urban Household Cohort during First and Second Waves of Infections Population seroprevalence of SARS-CoV-2 antibodies in Anambra State SeroTracker: a global SARS-CoV-2 seroprevalence dashboard Why is There Low Morbidity and Mortality of COVID-19 in Africa? Estimation of all-cause excess mortality by age-specific mortality patterns for countries with incomplete vital statistics: a population-based study of the case of Peru during the first wave of the COVID-19 pandemic Estimating excess mortality due to the COVID-19 pandemic: a systematic analysis of COVID-19-related mortality Economic Commision for Africa. Report on the Status of civil registration and vital statistics in Africa Births and deaths must be registered in Africa