Menu
Menu
Menu
Menu
Background The Coronavirus Disease 2019 (COVID-19) pandemic has had wide-reaching direct and indirect impacts on population health. In low- and middle-income countries, these impacts can halt progress toward reducing maternal and child mortality. This study estimates changes in health services utilization during the pandemic and the associated consequences for maternal, neonatal, and child mortality. Methods and findings Data on service utilization from January 2018 to June 2021 were extracted from health management information systems of 18 low- and lower-middle-income countries (Afghanistan, Bangladesh, Cameroon, Democratic Republic of the Congo (DRC), Ethiopia, Ghana, Guinea, Haiti, Kenya, Liberia, Madagascar, Malawi, Mali, Nigeria, Senegal, Sierra Leone, Somalia, and Uganda). An interrupted time-series design was used to estimate the percent change in the volumes of outpatient consultations and maternal and child health services delivered during the pandemic compared to projected volumes based on prepandemic trends. The Lives Saved Tool mathematical model was used to project the impact of the service utilization disruptions on child and maternal mortality. In addition, the estimated monthly disruptions were also correlated to the monthly number of COVID-19 deaths officially reported, time since the start of the pandemic, and relative severity of mobility restrictions. Across the 18 countries, we estimate an average decline in OPD volume of 13.1% and average declines of 2.6% to 4.6% for maternal and child services. We projected that decreases in essential health service utilization between March 2020 and June 2021 were associated with 113,962 excess deaths (110,686 children under 5, and 3,276 mothers), representing 3.6% and 1.5% increases in child and maternal mortality, respectively. This excess mortality is associated with the decline in utilization of the essential health services included in the analysis, but the utilization shortfalls vary substantially between countries, health services, and over time. The largest disruptions, associated with 27.5% of the excess deaths, occurred during the second quarter of 2020, regardless of whether countries reported the highest rate of COVID-19-related mortality during the same months. There is a significant relationship between the magnitude of service disruptions and the stringency of mobility restrictions. The study is limited by the extent to which administrative data, which varies in quality across countries, can accurately capture the changes in service coverage in the population. Conclusions Declines in healthcare utilization during the COVID-19 pandemic amplified the pandemic’s harmful impacts on health outcomes and threaten to reverse gains in reducing maternal and child mortality. As efforts and resource allocation toward prevention and treatment of COVID-19 continue, essential health services must be maintained, particularly in low- and middle-income countries.
We used an interrupted time-series design to estimate the percent change in the volumes of essential health services delivered during the pandemic. These estimates of lost services were translated into relative changes in coverage of interventions delivered during those periods to project the number of lives lost. The estimated monthly disruptions were also correlated to officially reported COVID-19 mortality rates, time since the start of the pandemic, and relative severity of mobility restrictions to determine which drivers are associated with changes in measured disruptions over time. The analysis was modeled on a previous study described elsewhere [27]. No changes to the analysis plan were made due to comments from reviewers or observations in the data. Data sharing agreements were established with all governments. Analysis of these secondary data did not constitute human subjects research and was considered public health practice. Thus, institutional research board approval was not required nor sought. Monthly administrative data on the volume of key essential health services between January 2018 and June 2021 were collated from 18 countries participating in a monitoring activity supported by the Global Financing Facility for Women, Children, and Adolescents (GFF). Eleven countries are classified as low-income by the World Bank: Afghanistan, the Democratic Republic of Congo (DRC), Ethiopia, Guinea, Liberia, Madagascar, Malawi, Mali, Sierra Leone, Somalia, and Uganda. The other 7 countries, Bangladesh, Cameroon, Ghana, Haiti, Kenya, Nigeria, and Senegal, are classified as lower-middle-income. Seven services were selected to represent the continuum of reproductive, maternal, and child health: family planning, antenatal care initiation (ANC1), antenatal care completion (ANC4), delivery, postnatal care initiation (PNC1), bacillus Calmette–Guérin (BCG) vaccine administration, and completion of pentavalent schedule (Penta3). These services were selected because they are high completeness across countries and serve as proxies for other services and interventions delivered at the same point. In addition, outpatient consultations (OPDs) were used as a proxy for the general use of health services. Outcome measures were not included since rare outcomes (e.g., maternal death, stillbirths) are difficult to accurately capture in facility data, or the data completeness was too poor for this analysis. Seven countries’ administrative data systems were missing 1 indicator, and Uganda was missing 2 indicators. Family planning volume was the most frequently missing indicator and was not reported in 5 countries (see Table A in S1 Text). The analysis used available-case analysis, where facilities with partial facility-month observations were included in the analysis. Differences in indicator definitions were observed across countries, particularly in OPD (total attendance versus total outpatient consultations), delivery (institutional deliveries versus institutional deliveries with a skilled birth attendant), and PNC1 (first postnatal visit versus time-bound PNC visits). In countries with both versions of indicators, a sensitivity check was conducted to demonstrate that both reporting methods yielded similar results (see Tables B, C, and D in S1 Text). HMIS data validity is often assessed in the context of measuring service coverage levels and can reflect challenges due to factors such as poor representativeness and the accuracy of population denominators [28]. Despite finding shortcomings in measuring service coverage, previous authors have called for the greater use of HMIS data, specifically the absolute number of services provided each month, in research and policy decisions. In this study, we do not attempt to estimate population service coverage but rather assume that the change in service-specific utilization reported by facilities in the HMIS represents the percentage change in population service coverage. We believe this use of HMIS, not as an estimate of coverage but as an estimate of coverage change, is less subject to various potential biases. The ability to rigorously estimate changes in service volume despite limitations of facility data has been previously demonstrated [29]. With the exception of Bangladesh and Nigeria, there should be high representativeness of facilities that report to HMIS since the public sector delivers the majority of care. The primary concern is the possible differential change in utilization between reporting and nonreporting facilities. Findings from household surveys and interviews with key health system stakeholders during the pandemic confirm that private facilities and community programs did not compensate for the disruptions in the public sector, and there were substantial levels of foregone care in the population [17,20]. HMIS data were downloaded on 22 August 2021, and were prepared for analysis by removing outlier values and restricting data for indicators with low completeness. These preparation steps are detailed in the Supporting information (see Text A in S1 Text), and the advantages and disadvantages of HMIS data are discussed in previous work [14]. To further assess the quality and reliability of the data, we present a range of sensitivity tests; we describe data reporting completeness (see Fig A in S1 Text) and include a sensitivity check showing that changes in reporting patterns did not drive the results. We also specify for each country and indicator the dates dropped due to poor completeness, or data availability that may reduce the prepandemic follow-up (see Table L in S1 Text). The final dataset included 137,192 health facilities ranging from 478 facilities in Guinea to 34,701 in Nigeria. The reports cover 42 months and 8 services, for 21,421,125 nonmissing facility-month-service observations in the 18 countries. We obtained data from 2 additional sources to assess whether service disruptions correlate with officially reported COVID-related death rates or with mobility restrictions. Data on reported COVID-19 deaths were obtained from the Center for Systems Science and Engineering (CSSE) at Johns Hopkins University, which compiles data from official government COVID-19 surveillance reports. Official accounts are likely to underreport actual mortality in settings with limited testing capacity, particularly at the beginning of the pandemic [30]. However, even if the official reports are inaccurate, there are many mechanisms through which these reports may affect service utilization, such as changing perceptions by health providers and the population on the state of the outbreak. Therefore, one way to interpret these data is as a proxy for the perceived risk of infection. Information on the policy measures affecting population mobility was obtained from the Oxford COVID-19 Government Policy Tracker, which systematically tracks implementation dates and scores the stringency of policy interventions. We selected a subset of policies that may affect population access to health facilities: public transport closures, stay-at-home requirements, movement limitations, school closures, and workplace closures. The dataset includes ordinal severity scores for each policy to capture the stringency of restriction, ranging from no restrictions to recommendations to requirements with minimal exceptions. We constructed an index representing the daily severity of mobility restrictions using the first component of a principal component analysis of these selected indicators. There is a correlation of 0.92 between the index we construct and the Oxford response stringency index, composed of a wider set of interventions. The Oxford COVID-19 Policy Tracker captures as-written strictness of policies but not levels of enforcement. As we are unaware of a reliable source on levels of enforcement of restrictions, differences in levels of enforcement between countries were not taken into account. We used an interrupted time-series approach to predict the volume of services that would have been delivered had the pandemic not occurred. The interruption period starts with the WHO pandemic declaration in March 2020, coinciding with the start of community transmission and mobility restrictions in most countries. Service and countries were modeled separately using a linear regression equation with the following form: where Ytf is the service volume reported by facility f in month t. T represents the time in months since January 2018 to account for a linear secular trend (β1), Month represents calendar months to account for seasonality (β2..12), and αf represents the facility-level fixed effect accounting for time-invariant facility characteristics. Fixed effects were replaced with facility characteristic covariates (province and facility type) in Uganda, where an update to the administrative system did not allow for consistent identification of facilities over time. PandemicMonth denotes a series of dummy variables for each of the months between March 2020 and June 2021. That is, β13..29 contain estimated disruption for each month since the pandemic. To calculate the percentage change in service utilization during the pandemic months, we first used the estimation results to calculate the expected volume in the absence of the pandemic (counterfactual). Then, we divided the reported volumes by these expectations. The cumulative shortfall was estimated using the same model with a single pandemic period. A 2-year prepandemic time horizon was chosen to minimize confounding from changes in data collection practices, policy changes, or other health shocks while still allowing separation of seasonality effects from secular trends. We further examined whether the estimated monthly changes in service volumes showed statistical association with the time elapsed since the start of the pandemic, the monthly number of official reported COVID deaths, and the stringency of mobility restrictions. These relationships were assessed by running the following linear regression separately for each service: where Dtc is the estimated monthly change in service volume for month t in country c. CovidMortalitytc represents the officially reported monthly number of COVID-19-related deaths per 100,000 people, and RestrictionStringencytc represents the average monthly stringency of the mobility restrictions. αc represents a country fixed effect, and εtc is a normally distributed error term. We estimated the impact of the service utilization disruptions on the absolute number of child, neonatal, and maternal deaths using the Lives Saved Tool (LiST). LiST is a mathematical model that forecasts mortality estimates from the coverage of 70+ RMNCH+N health interventions, considering the specific demographic and epidemiological context of a country [31]. We assumed that the relative change in the coverage of the interventions included in the LiST model was the same as the estimated relative changes in service utilization. Each intervention was linked to the service during which the intervention is typically delivered or proxied by the service assumed to have a similar utilization pattern. For interventions without a reasonable proxy, such as child nutrition services, the conservative default assumed no change in the intervention coverage. This linking of service indicators to LiST interventions is described in Table E of S1 Text. As multiple RMNCH interventions were linked to a small set of indicators, small variations in the few service indicators significantly affect the overall mortality results. To address this, we ran a sensitivity analysis using different linking combinations to understand how these linking decisions alter the results and limit their potential effect. For each LiST intervention and country, we obtained coverage values from the most recent household survey for the country (typically a DHS or MICS), which we took as the coverage value that we would have expected in the absence of the pandemic (i.e., as a “counterfactual”). To estimate the coverage value during the pandemic, we multiplied the counterfactual coverage value by the estimated disruption of the service (proxy) indicator. This approach assumes that, during the pandemic, the change in population-level coverage was proportional to the change in reported facility-level utilization. In this way, we obtained an estimated coverage value for each intervention, country, and period. We used 3-month periods (quarters), aggregating the service disruption for the relevant proxy indicator for each quarter and calculating disrupted coverage values for each quarter of the pandemic for each country and intervention. We ran 2 LiST analyses for each country and quarter: first, a “without pandemic” scenario, using only the counterfactual coverage values, to obtain the expected deaths in the absence of the pandemic; and second, a “with pandemic” scenario, to obtain the expected deaths during the pandemic. LiST only takes yearly input values, so we entered quarterly values as yearly values (for 2020 or 2021, as appropriate), and divided the resulting expected deaths by 4, to obtain the expected deaths for the quarter. We took the difference in expected deaths between the “with pandemic” and “without pandemic” scenarios to represent the additional deaths due to the pandemic. For each country and age group, we report the number of deaths that we would have expected in the absence of the pandemic for the period March 2020 to June 2021, the estimated number of additional maternal and child (including neonatal) deaths due to the change in service utilization during this same period, and the relative increase in mortality because of service utilization declines during the pandemic. The RECORD checklist is included in the Supporting information (S1 RECORD Checklist). Study limitations include the assumption that the changes in service-specific utilization reported in the HMIS represents the percentage of service coverage change in the population, and the inability to account for differential changes in health-seeking behaviors across severity of need.
