tetano
Editor, Senior Moderator
J Glob Health
. 2024 May 31:14:05013.
doi: 10.7189/jogh.14.05013. Construction and validation of a covariate-based model for district-level estimation of excess deaths due to COVID-19 in India
Anand Krishnan[SUP] 1 [/SUP], Mahasweta Dubey[SUP] 1 [/SUP], Rakesh Kumar[SUP] 1 [/SUP], Harshal R Salve[SUP] 1 [/SUP], Ashish Datt Upadhyay[SUP] 2 [/SUP], Vivek Gupta[SUP] 3 [/SUP], Sumit Malhotra, Ravneet Kaur[SUP] 1 [/SUP], Baridalyne Nongkynrih[SUP] 1 [/SUP], Mohan Bairwa[SUP] 1 [/SUP]
Affiliations
Background: Different statistical approaches for estimating excess deaths due to coronavirus disease 2019 (COVID-19) pandemic have led to varying estimates. In this study, we developed and validated a covariate-based model (CBM) with imputation for prediction of district-level excess deaths in India.
Methods: We used data extracted from deaths registered under the Civil Registration System for 2015-19 for 684 of 713 districts in India to estimate expected deaths for 2020 through a negative binomial regression model (NBRM) and to calculate excess observed deaths. Specifically, we used 15 covariates across four domains (state, health system, population, COVID-19) in a zero inflated NBRM to identify covariates significantly (P < 0.05) associated with excess deaths estimate in 460 districts. We then validated this CBM in 140 districts by comparing predicted and estimated excess. For 84 districts with missing covariates, we validated the imputation with CBM by comparing estimated with predicted excess deaths. We imputed covariate data to predict excess deaths for 29 districts which did not have data on deaths.
Results: The share of elderly and urban population, the under-five mortality rate, prevalence of diabetes, and bed availability were significantly associated with estimated excess deaths and were used for CBM. The mean of the CBM-predicted excess deaths per district (x̄ = 989, standard deviation (SD) = 1588) was not significantly different from the estimated one (x̄ = 1448, SD = 3062) (P = 0.25). The estimated excess deaths (n = 67 540; 95% confidence interval (CI) = 35 431, 99 648) were similar to the predicted excess death (n = 64 570; 95% CI = 54 140, 75 000) by CBM with imputation. The total national estimate of excess deaths for all 713 districts was 794 989 (95% CI = 664 895, 925 082).
Conclusions: A CBM with imputation can be used to predict excess deaths in an appropriate context.
. 2024 May 31:14:05013.
doi: 10.7189/jogh.14.05013. Construction and validation of a covariate-based model for district-level estimation of excess deaths due to COVID-19 in India
Anand Krishnan[SUP] 1 [/SUP], Mahasweta Dubey[SUP] 1 [/SUP], Rakesh Kumar[SUP] 1 [/SUP], Harshal R Salve[SUP] 1 [/SUP], Ashish Datt Upadhyay[SUP] 2 [/SUP], Vivek Gupta[SUP] 3 [/SUP], Sumit Malhotra, Ravneet Kaur[SUP] 1 [/SUP], Baridalyne Nongkynrih[SUP] 1 [/SUP], Mohan Bairwa[SUP] 1 [/SUP]
Affiliations
- PMID: 38813676
- DOI: 10.7189/jogh.14.05013
Background: Different statistical approaches for estimating excess deaths due to coronavirus disease 2019 (COVID-19) pandemic have led to varying estimates. In this study, we developed and validated a covariate-based model (CBM) with imputation for prediction of district-level excess deaths in India.
Methods: We used data extracted from deaths registered under the Civil Registration System for 2015-19 for 684 of 713 districts in India to estimate expected deaths for 2020 through a negative binomial regression model (NBRM) and to calculate excess observed deaths. Specifically, we used 15 covariates across four domains (state, health system, population, COVID-19) in a zero inflated NBRM to identify covariates significantly (P < 0.05) associated with excess deaths estimate in 460 districts. We then validated this CBM in 140 districts by comparing predicted and estimated excess. For 84 districts with missing covariates, we validated the imputation with CBM by comparing estimated with predicted excess deaths. We imputed covariate data to predict excess deaths for 29 districts which did not have data on deaths.
Results: The share of elderly and urban population, the under-five mortality rate, prevalence of diabetes, and bed availability were significantly associated with estimated excess deaths and were used for CBM. The mean of the CBM-predicted excess deaths per district (x̄ = 989, standard deviation (SD) = 1588) was not significantly different from the estimated one (x̄ = 1448, SD = 3062) (P = 0.25). The estimated excess deaths (n = 67 540; 95% confidence interval (CI) = 35 431, 99 648) were similar to the predicted excess death (n = 64 570; 95% CI = 54 140, 75 000) by CBM with imputation. The total national estimate of excess deaths for all 713 districts was 794 989 (95% CI = 664 895, 925 082).
Conclusions: A CBM with imputation can be used to predict excess deaths in an appropriate context.