key: cord-0299122-3791udg0
authors: Surendra, H.; Salama, N.; Lestari, K. D.; Adrian, V.; Widyastuti, W.; Oktavia, D.; Lina, R. N.; Djaafara, B. A.; Fadilah, I.; Sagara, R.; Ekawati, L. L.; Nurhasim, A.; Ahmad, R. A. A.; Kekalih, A.; Syam, A. F.; Shankar, A. H.; Thwaites, G.; Baird, J. K.; Hamers, R. L.; Elyazar, I. R.
title: Pandemic inequity in a megacity: a multilevel analysis of individual, community, and health care vulnerability risks for COVID-19 mortality in Jakarta, Indonesia
date: 2021-11-26
journal: nan
DOI: 10.1101/2021.11.24.21266809
sha: 673231adb4577c260e9efe5d8d0bda99ef640b9a
doc_id: 299122
cord_uid: 3791udg0

Background The 33 recognized megacities comprise approximately 7% of the global population, yet account for 20% COVID-19 deaths. The specific inequities and other factors within megacities that affect vulnerability to COVID-19 mortality remain poorly defined. We assessed individual, community-level and health care factors associated with COVID-19-related mortality in a megacity of Jakarta, Indonesia, during two epidemic waves spanning March 2, 2020, to August 31, 2021. Methods This retrospective cohort included all residents of Jakarta, Indonesia, with PCR-confirmed COVID-19. We extracted demographic, clinical, outcome (recovered or died), vaccine coverage data, and disease prevalence from Jakarta Health Office surveillance records, and collected sub-district level socio-demographics data from various official sources. We used multi-level logistic regression to examine individual, community and sub-district-level health care factors and their associations with COVID-19-mortality. Findings Of 705,503 cases with a definitive outcome by August 31, 2021, 694,706 (98.5%) recovered and 10,797 (1.5%) died. The median age was 36 years (IQR 24-50), 13.2% (93,459) were <18 years, and 51.6% were female. The sub-district level accounted for 1.5% of variance in mortality (p<0.0001). Individual-level factors associated with death were older age, male sex, comorbidities, and, during the first wave, age <5 years (adjusted odds ratio (aOR) 1.56, 95%CI 1.04-2.35; reference: age 20-29 years). Community-level factors associated with death were poverty (aOR for the poorer quarter 1.35, 95%CI 1.17-1.55; reference: wealthiest quarter), high population density (aOR for the highest density 1.34, 95%CI 1.14-2.58; reference: the lowest), low vaccine coverage (aOR for the lowest coverage 1.25, 95%CI 1.13-1.38; reference: the highest). Interpretation In addition to individual risk factors, living in areas with high poverty and density, and low health care performance further increase the vulnerability of communities to COVID-19-associated death in urban low-resource settings.

However, there is a general scarcity of data in LMIC assessing the influence of communitylevel socio-demographics factors on COVID-19-related mortality.

Indonesia, the world's fourth most populous country (population 274 million), is a lower-middle 149 income country (LMIC) featuring great geographic, cultural and socio-economic diversity the highest number of COVID-19 confirmed cases and deaths in Southeast Asia, second only 153 to India in all of Asia 20 , at 4,253,598 cases and 143,744 deaths (3·4% case fatality rate (CFR)) 154 up to November 22, 2021 21 , of which 20% (863,482) of cases and 9·4% (13,574) of deaths 155 occurred in its capital Jakarta, a megacity (7,659 Km 2 , and estimated population 10·6 million) 156 that features stark health inequalities and socio-demographic heterogeneity. The first SARScalled the Special Capital City Area Jakarta (Daerah Khusus Ibukota, DKI Jakarta), the PHDI 165 ranged from 64% in North Jakarta to 68% in East Jakarta districts. However, the five districts is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted November 26, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 community-level risk factors and individual outcomes (deceased vs recovered) of cases living in the corresponding sub-district ( Figure 1A ). In accordance with Indonesia's national COVID-185 19 guidelines 25 , confirmatory SARS-CoV-2 PCR testing was conducted on naso-and/or 186 oropharyngeal swab specimens in COVID-19 reference laboratories. 

Completed forms were submitted to the DKI Jakarta Health Office for cleaning and verification 194 (checking for completeness, inconsistency, error, and duplication) and entered into a 195 surveillance database. We extracted data regarding SARS-CoV-2 PCR testing, hospital 196 admission, and outcomes (recovered or deceased), along with age, sex, and pre-existing 197 comorbidities (based on clinical assessment or cases self-report) 25 .

Sub-district-level data (community-level risk factors) were collected from official government 

. CC-BY-NC-ND 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted November 26, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 

Mann-Whitney U test, χ² test, or Fisher's exact test to compare characteristics between 226 deceased and recovered cases. We set statistical significance at 0·05, and all tests were two 227 sided.

We used bivariable and multivariable multi-level logistic regression models to determine the 230 risk of death, expressed as odds ratio (OR) with 95% confidential intervals (CI). Sub-district 231 was treated as the random effect variable to adjust for clustering of observations within sub-232 districts. We did null model analysis (no predictor was added) and the result justified the use 233 of the multi-level models. We excluded cases from two sub-districts with insufficient sample 234 size (Kepulauan Seribu Selatan and Utara). All independent variables with p-value <0·10 in 235 bivariable analysis were included in the multivariable models. Final model selection was 236 informed by intra class correlation postestimation test. We used interaction terms to examine 237 potential effect modification by age, sex, and time. In the presence of interaction, the stratum-238 specific OR and 95% CI were calculated, adjusting for other variables with p-value <0·10 in 239 bivariable analysis. Additionally, we used a restricted cubic spline mixed effect regression to 240 model the OR of death over time.

There was a substantial proportion of missing data for chronic comorbidities (58%). Missing-243 indicator analysis by risk factor stratification and by regression analysis identified bias of 244 missing data with respect to mortality, thus we additionally conducted analysis to assess 245 sensitivity of risk factor identification due to missing data. We performed multi-level logistic is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted November 26, 2021. ; https://doi.org/10.1101/2021.11.24.21266809 doi: medRxiv preprint secondary analysis of anonymised routine surveillance data 21 .

The funder of the study had no role in study design, data collection, data analysis, data 260 interpretation, or writing of the report. The corresponding author had full access to all of the 261 data and the final responsibility to submit for publication. All authors were not precluded from 262 accessing data in the study, and accepted responsibility to submit for publication. Table 1 is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint Figure 1B ).

Compared to recovered cases, deceased cases were older (median 59 vs 35 years); more 301 likely to be males (1·8% vs 1·3%), to have one or more comorbidities (9·4% vs 1·0%), to be 302 infected in the first wave (1·9% vs 1·2%), and to live in sub-districts with higher population 303 density (highest density: 1·7% vs lowest density: 1·4%), higher poverty (highest: 1·6% vs 304 lowest: 1·4%); higher nurse-population ratio (lowest: 1·6% vs highest: 1·5%); and lower 305 vaccine coverage (lowest: 1·7% vs highest: 1·4%) ( Table 1) . Compared with the first wave,

there was a notable decrease in CFRs across sub-districts during the second wave of the 307 epidemic ( Figure 2E -F, and Figure 3B ). Moreover, the sub-districts with higher population 308 density, poverty, and lower vaccine coverage tended to be the sub-districts with higher CFR 309 ( 

In bivariable analysis (Table 2) , the risk of death was significantly associated with older age, 312 male sex, comorbidities, first wave, and higher sub-district population density, poverty, higher 313 prevalence of tuberculosis, all-cause mortality among under 5 years, and lower COVID-19 314 vaccine coverage.

In the final multivariable multi-level logistic regression model ( Figure 4A ), the risk of death was 317 increased for age groups 30-39 years (aOR 1·94, 95%CI 1·62-2·33), 40-49 years (aOR 4·51, 318 95%CI 3·82-5·33), 50-59 years (aOR 12·65, 95%CI 10·80-14·81), 60-69 years (aOR 18·64, 319 95%CI 15·87-21·89), ≥70 years (aOR 32·91, 95%CI 27·97-38·72) compared to 20-29 years; 320 for males (aOR 1·29, 95%CI 1·24-1·34); for individuals with at least one comorbidity (aOR 321 3·96, 95%CI 3·56-4·41); for residents of sub-districts with highest population density (aOR 322 1·34, 95%CI 1·14-1·58, reference: lowest density), higher poverty (Q3) (aOR 1·35, 95%CI 323 1·17-1·55, reference: lowest poverty), and with lowest vaccine coverage (aOR 1·25, 95%CI 324 1·13-1·38, reference: highest coverage). We found no associations with proportion of poor 325 sanitation areas, doctor-population ratio, nurse-population ratio, prevalence of hypertension, 326 diabetes, and tuberculosis (p>0·05 each). The sensitivity analysis revealed similar findings,

suggesting there was no significant bias introduced by missing data in our dataset; it also 328 suggested that the risk of death was increased for cases who had at least one comorbidity 329 . CC-BY-NC-ND 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted November 26, 2021. ; https://doi.org/10.1101/2021.11.24.21266809 doi: medRxiv preprint (aOR 4·25, 95%CI 3·81-4·75) compared to those who had no comorbidity (Supplementary  Table 2 ).

We found that the effect of age was modified by time (first wave), and poverty was modified 1·04-2·35) compared to adult age 20-29 years in the first pandemic wave, but not in the second 336 wave (Supplementary Table 3 ). We found that mortality risk significantly decreased over time 337 especially for children 0-9 years ( Figure 4B ). In addition, we found that the risk of death was 338 higher for sub-districts with highest level of poverty and density compared to sub-districts with 339 lowest poverty and density (aOR 1·27, 95%CI 1·10-1·47) (Supplementary Table 4 

In line with previous reports from various settings, the strongest independent risk factors of 353 deaths were older age, male sex, and the presence of one or more chronic comorbidities.

Important novel findings were that sub-district-level socio-demographic factors, especially high 355 population density and poverty, and health care factors, especially low COVID-19 vaccine 356 coverage, further increased the risk of COVID-19-related death in metropolitan Jakarta.

A previous US study conducted in the early phase of the pandemic showed a significant 359 association between household crowding and COVID-19 outcomes 18 ; counties with the 360 highest household crowding had a nearly 2-fold higher COVID-19 mortality rate than counties 361 with the lowest crowding. Concordant with that study, we found that residents in sub-districts 362 with the highest population density had a 34% higher risk of death than those residing in sub- is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted November 26, 2021. ; https://doi.org/10.1101/2021.11.24.21266809 doi: medRxiv preprint prone riverbanks. These sub-districts are also known to have relatively higher prevalence of non-communicable diseases such as hypertension and diabetes, and poverty-related infectious disease such as tuberculosis (Supplementary Table 5 

to those studies, we found that the risk of death was 40% higher for resident of sub-districts 380 with higher poverty (quarter 3) relative to those of lowest poverty. The interaction between 381 poverty and population density also revealed that the risk of death was 30% higher for sub-382 districts with highest level of poverty and density compared to sub-districts with lowest poverty 383 and density. Urban crowding and poverty impose very many disadvantages to health, here 384 shown to include elevated risk of death as a consequence of SARS-CoV-2 infection.

The risk of COVID-19-related death in DKI Jakarta was 25% higher for resident of sub-districts is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted November 26, 2021. ; https://doi.org/10.1101/2021.11.24.21266809 doi: medRxiv preprint that older age, male sex and presence of underlying comorbidities were associated with higher the youngest populations have been suffering most from gaps in access to health services, 

This study had some limitations. The retrospective design and reliance on routine surveillance 416 data meant that, for some key baseline variables, data were incomplete or uniformly 417 unavailable (e.g., type of comorbidities and disease severity classification). As in many other 418 settings, the individual-level socio-demographics data were not recorded in the current Indonesia's national database 22 . Comorbidities were often self-reported or could be under-420 diagnosed, potentially resulting in underreporting and hence underestimation of effect sizes.

Details on supportive care and treatment received were also not available for this analysis.

There are several other relevant socio-demographics variables such as human development 423 index and public health development index that may represent population and health system 424 vulnerability 24 but were only available at the district level, and were not included in our analysis.

However, our analysis included all available key variables that compose those indicators 426 (prevalence of infectious and non-communicable diseases, health care workers-population 427 ratio, universal child immunization, and all-cause mortality among under 5 years old 428 population), therefore enhancing credibility of our findings. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint 

. CC-BY-NC-ND 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted November 26, 2021. ; https://doi.org/10.1101/2021.11.24.21266809 doi: medRxiv preprint COVID-19-related in-hospital mortality in a high HIV and tuberculosis prevalence 540 setting in South Africa: a cohort study. Lancet HIV. 2021;8(9):e554-67. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted November 26, 2021. ; https://doi.org/10.1101/2021.11.24.21266809 doi: medRxiv preprint Table 1 : Individual, community, health care characteristics and outcomes of COVID-19 cases in DKI Jakarta, March 2, 2020 to August 31, 2021 values were categorised into quarters (Q), i.e., below 25 th percentile (Lowest), 25 th -50 th 580 percentile (Q2), 50 th -75 th percentile (Q3) and above 75 th percentile (Highest) for each sub-581 district level variable. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted November 26, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted November 26, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 Supplementary data is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted November 26, 2021. ; Table S2 presents the final multi-level multivariable logistic regression with multiple imputations for missing comorbidities. Sub-district was treated as the random effect variable. First wave: March 2020 to April 2021, Second wave: May 2021 to August 2021. Numeric values were categorised into quartiles (Q), i.e., below 25 th percentile (Lowest), 25 th -50 th percentile (Q2), 50 th -75 th percentile (Q3) and above 75 th percentile (Highest) for each subdistrict level variable. Table S4 presents the effect of poverty on mortality based on population density, after controlling for sex, comorbidities, time period, vaccine coverage, and correlation at subdistrict level (random effect variable). OR=odds ratio. First wave=March 2020 to April 2021.

. CC-BY-NC-ND 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted November 26, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 Second wave=May 2021 to August 2021. First wave: March 2020 to April 2021, Second wave: May 2021 to August 2021. Table S5 : Association between population density and other covariates. Numeric values were categorised into quartiles (Q), i.e., below 25 th percentile (Lowest), 25 th -50 th percentile (Q2), 50 th -75 th percentile (Q3) and above 75 th percentile (Highest) for each sub-district level variable. Table S6 : Association between poverty and other covariates. Numeric values were categorised into quartiles (Q), i.e., below 25 th percentile (Lowest), 25 th -50 th percentile (Q2), 50 th -75 th percentile (Q3) and above 75 th percentile (Highest) for each sub-district level variable.

. CC-BY-NC-ND 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted November 26, 2021. ; https://doi.org/10.1101/2021.11.24.21266809 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted November 26, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 

Association of Social and Demographic Factors 506 with COVID-19 Incidence and Death Rates in the US

Q2 (1·5-2·1)

24·6%) 2,729 (25·3%) 170,793 (24·6%) Health care capacity Doctor-population ratio, doctor per 10,000 population 0·158

Q3 (7·9-9·9)

24·0%) 2,627 (24·4%) 166,539 (24·0%) Nurse-population ratio, nurse per 10,000 population Q1 (2·9-11·2)

Q2 (11·2-22·8)

Q4 (83·9-416·1)

COVID-19 vaccination coverage, % <0·0001

Q4 (40·5-50·0)

Q1 (94·5-96·2)

Q2 (96·2-98·0)

Q3 (98·0-98·9)

25·1%) 2,656 (24·6%) 173,999 (25·1%) Health related characteristics Prevalence of Hypertension

Q2 (0·2-0·3)

Q3 (0·3-0·5)

Q1 (0·8-2·1)

Q4 (4·0-6·3)