In this analysis, we investigated two research questions on the effects of COVID on health systems in Uganda and Bangladesh. We aimed to understand how COVID affected reporting in routine information systems and how COVID affected health service utilization.
For the first research question, we found that COVID affected reporting for some services at the health facility level. COVID negatively affected reporting of immunization services in Bangladesh, for which data was reported at the upazila health complex level by the EPI reporting system. This general finding, that COVID had impacts on facility reporting of services, is reflected in other research. Shapira and colleagues also found drops in facility-level reporting for four of eight countries included in the analysis during COVID-19.7 When data was aggregated at higher levels, such as the upazila level for Bangladesh MIS3 services or district services in Uganda, no COVID effects were detected. However, this does not mean that there were no COVID effects on reporting at lower levels. COVID effects could have occurred at the facility level but been masked by the higher-level aggregation if any one facility reported providing any given service.
For the second research question, we found that health service utilization was affected by COVID in both countries. For many of the services, the most severe impacts were experienced within the first three months of the pandemic, corresponding with the same time during which most governments implemented shutdown policies to prevent spread. Overall outpatient attendance as well as malaria and pneumonia visits among young children were the services most persistently negatively affected by COVID nationally during the pandemic period and lacked stable recovery to expected levels. These results reflect existing studies that found negative impacts on overall utilization,4–8 as well as disease-specific service utilization.4,6
COVID effects on maternal and infant health services were not consistent across countries. In Bangladesh, there were clear negative effects on ANC1, ANC4, and facility deliveries. There were some initial negative COVID effects for all three services in Uganda, but after May 2020, there were multiple months exceeding levels predicted by our models. These inconsistent findings across countries were found by other studies. Research from Rwanda found significant decreases in ANC visits, facility deliveries, and live births nationally.9 A study from Ethiopia found month-to-month comparisons of ANC1 and ANC4 levels before and during COVID to be similar.4
During the COVID-19 period, supply-based services, which include family planning and vaccinations, were more volatile than what models had predicted. This volatility is particularly prominent in the graphs depicting vaccinations in Uganda and Bangladesh. There were initial negative COVID effects for the first several months in Uganda and Bangladesh, which is consistent with several other studies.7,9,11 Both countries either periodically achieved recovery or exceeded expected levels for multiple months in the remaining COVID period, during months that had not been included in those other studies.
It is also important to note the drastic negative COVID effects in Bangladesh on immunization services that occurred at the end of 2020. During this time, there was a health worker strike, particularly among the health workers whose responsibility it is to administer and implement national immunization programs. This drop in measles vaccinations is particularly prominent because this strike immediately preceded a national measles vaccination campaign.12 Due to a nationwide campaign, measles vaccinations that occurred in January 2021 were reported to a separate reporting system that was not integrated with the routine EPI information system.
When comparing national COVID effect patterns in Uganda to those in Kampala the COVID effects on Kampala were more severe. This pattern was consistent across almost every service investigated in this study. Relative to what occurred nationally, the increased severity in terms of COVID effects presented as both larger gaps between observed and predicted values toward the beginning of the pandemic as well as delayed recovery later into the pandemic period in Kampala. This finding is not surprising as urban areas had COVID outbreaks first and were often targeted for more stringent lockdown measures. Thus, the effects of the disruptions and lockdowns are likely to be more exacerbated in capital cities such as Kampala.
This study has several strengths and further contributes to the expanding body of literature on the impact of COVID-19 on health service utilization from low and middle-income countries. The inclusion of data from two different countries allows for conclusions to be drawn across borders, while also illustrating the variety of ways COVID could impact both reporting and service utilization. This research expands upon existing work that investigates similar questions. For almost all services in these countries, we predicted COVID effects for at least one year after the start of the pandemic. Additionally, we use rigorous, but simple modeling and prediction estimation strategies, that consider time trends, seasonal patterns, and unit-level fixed effects. These models are easily replicable as they do not depend on information found outside the RHIS and can be implemented by widely available statistical software. Instead of very sophisticated, novel statistical modeling, we applied a general estimation strategy with the aim for wider use of such analytical approaches on routinely available data. We, however, found that the service delivery trajectories varied considerably by country and services, which requires analysts to be careful adapting the estimation model specification to the peculiarities of each specific service.
There are several limitations to our analysis pertaining to the models that we use to estimated COVID effects. As time increases, the time trend estimated from pre-COVID data becomes less relevant, and thus, the potential for error in the counterfactual prediction increases. Such inaccuracies could stem from not just coefficient inaccuracy, but also changes in model functional form. An analyst should be careful about the uncertainly of medium- and long-term predictions.
Bangladesh’s routine information system for maternal and child health services delivery expanded during the pre-COVID period (in January 2019). Such changes were accounted for using the yearly control variables. In Uganda, the DHIS2 also changed in 2019, which had a new reporting mechanism and service definitions. However, we found consistency in the time series across years. Finally, we want to mention that adjusting for COVID effects on reporting in total models influenced the conclusions drawn from this analysis. Had we not considered the potential impact of COVID on facility reporting for EPI services in Bangladesh, we may have over-estimated COVID effects for some services and under-estimated others.