The COVID-19 pandemic and disruptions to essential health services in Kenya: a retrospective time-series analysis

Background: Public health emergencies can disrupt the provision of and access to essential health-care services, exacerbating health crises. We aimed to assess the effect of the COVID-19 pandemic on essential health-care services in Kenya. Methods: Using county-level data routinely collected from the health information system from health facilities across the country, we used a robust mixed-effect model to examine changes in 17 indicators of essential health services across four periods: the pre-pandemic period (from January, 2018 to February, 2020), two pandemic periods (from March to November 2020, and February to October, 2021), and the period during the COVID-19-associated health-care workers’ strike (from December, 2020 to January, 2021). Findings: In the pre-pandemic period, we observed a positive trend for multiple indicators. The onset of the pandemic was associated with statistically significant decreases in multiple indicators, including outpatient visits (28·7%; 95% CI 16·0–43·5%), cervical cancer screening (49·8%; 20·6–57·9%), number of HIV tests conducted (45·3%; 23·9–63·0%), patients tested for malaria (31·9%; 16·7–46·7%), number of notified tuberculosis cases (26·6%; 14·7–45·1%), hypertension cases (10·4%; 6·0–39·4%), vitamin A supplements (8·7%; 7·9–10·5%), and three doses of the diphtheria, tetanus toxoid, and pertussis vaccine administered (0·9%; 0·5–1·3%). Pneumonia cases reduced by 50·6% (31·3–67·3%), diarrhoea by 39·7% (24·8–62·7%), and children attending welfare clinics by 39·6% (23·5–47·1%). Cases of sexual violence increased by 8·0% (4·3–25·0%). Skilled deliveries, antenatal care, people with HIV infection newly started on antiretroviral therapy, confirmed cases of malaria, and diabetes cases detected were not significantly affected negatively. Although most of the health indicators began to recover during the pandemic, the health-care workers’ strike resulted in nearly all indicators falling to numbers lower than those observed at the onset or during the pre-strike pandemic period. Interpretation: The COVID-19 pandemic and the associated health-care workers’ strike in Kenya have been associated with a substantial disruption of essential health services, with the use of outpatient visits, screening and diagnostic services, and child immunisation adversely affected. Efforts to maintain the provision of these essential health services during a health-care crisis should target the susceptible services to prevent the exacerbation of associated disease burdens during such health crises. Funding: Bill & Melinda Gates Foundation. To examine the effects of the COVID-19 pandemic on the use of essential health services, we did a retrospective time-series analysis, examining data at the county level for Kenya. Because this study used existing routine aggregated health information that does not qualify as human patient research, written informed consent was not required. The use of these data was approved by Kenya’s Ministry of Health. Kenya has implemented a health information system that captures health data from the lowest level of health facilities across the country, which are then reported monthly to a central national database.22 By means of District Health Information Software (DHIS2)—the main national data aggregation platform deployed in most health facilities in Kenya and used by all public health facilities—the data are reported as aggregate numbers for each subcounty, county, and at the national level. The monthly reporting rate is calculated on the basis of the number of registered facilities, and the number of monthly reports submitted per month. We used the reporting rate to calculate an adjusted estimate of the monthly aggregate for each county. Using this dataset, we abstracted aggregated county-level data on indicators of the use of primary health-care services; reproductive, maternal, newborn, child, and adolescent health; sexual violence; communicable and non-communicable diseases; and the reporting rates for each indicator (appendix p 1). The indicators used were: primary health-care use, skilled deliveries, antenatal care, children presenting with pneumonia, vitamin A supplements, number of third doses of the diphtheria, tetanus toxoid, and pertussis vaccine (DTP3) administered, children attending a child welfare clinic who are underweight, children treated for diarrhoea, sexual violence, HIV tests conducted, people with an HIV infection newly started on antiretroviral therapy, number of notified tuberculosis cases, patients tested for malaria, confirmed cases of malaria, cervical cancer screening, hypertension cases, and diabetes cases. The data were obtained as monthly aggregates for the period from January, 2018 to October, 2021. During the pandemic period, a nationwide health-care workers’ strike, primarily involving clinical officers and nurses, occurred in the months of December, 2020 and January, 2021. The health-care workers were advocating for the adequate provision of personal protective equipment and insurance to protect themselves while responding to the pandemic. An exploratory data analysis on the trends of the essential health services showed four unique periods: the pre-pandemic period (from January, 2018 to February, 2020), the two pandemic periods before and after the health-care workers’ strike (from March to November 2020 and February to October 2021), and the within-pandemic period when there was a national health-care workers’ strike (from December, 2020 to January, 2021). The first case of SARS-CoV-2 in Kenya was reported on March 13, 2020. We obtained data on human movement in Kenya during the pandemic period from Google and Facebook.23, 24 The Google mobility anonymised data were used to estimate the within-county human mobility. This estimate was done by comparing visits to specific categories of locations (eg, retail shops, parks, workplaces, residential areas, and public transport areas) during the pre-pandemic period (baseline) and the pandemic period. The baseline values were the median values for each day of the week over a 5-week period from Jan 3 to Feb 6, 2020. Data from Facebook were used to estimate the between-county mobility data by comparing the number of individuals moving between defined administrative regions before and during the pandemic period in Kenya. A monthly average of the between-county and within-county movement data was used as a measure of adherence to movement restrictions for each county during the pandemic period. Data on the number of people tested and the daily cases of COVID-19 confirmed in Kenya were obtained from Kenya’s Ministry of Health. A monthly attack rate (number of new cases divided by the total population per 100 000) was computed for each county and incorporated in the model. We maintained a record of the type of COVID-19 restrictions that were instituted, and the dates when these restrictions came into effect and when they were lifted (appendix p 2). For the night curfew that was implemented as a partial lockdown measure, we used data on the total number of curfew hours each month in each county and incorporated these data into the models to account for the stringency of movement restrictions. Our analysis aimed to answer the two following questions: firstly, did the COVID-19 pandemic lead to statistically significant changes in the chosen indicators of essential health services in Kenya? Secondly, what was the direction and magnitude of the change in these indicators? To answer these questions, we planned to: firstly, establish the monthly incidence of each indicator per county; secondly, conduct a robust mixed-effect regression model for each indicator comparing the pre-pandemic, pandemic, and health-care workers’ strike periods; and thirdly, compare the estimates of the slopes for each indicator at the end of each study period and the intercept estimate of the indicators at the start of the next period to establish the magnitude of change in the indicators associated with the pandemic and the health-care workers’ strike. We estimated the population sizes and population density of each county per year using the 2019 Kenya Census Data and World Bank population growth rate estimates, which were 2·31 for 2018 and 2·27 for 2019. We assumed an estimated growth rate of 2·23 (because we assumed a 0·04 reduction in growth rate from 2·27 to 2·23, as observed between 2018 and 2019) for the years 2020 and 2021 because these data were unavailable. We calculated the incidence of each indicator per month, expressed as the number of cases per 100 000 people. For indicators that concerned women of childbearing age, we used the estimated population of women aged between 15 and 49 years to calculate the incidence of these indicators. To estimate the incidence of cervical cancer screening, we used the female population aged between 25 and 49 years, in accordance with the Ministry of Health Kenya National Cancer Screening Guidelines.25 To establish the effect of the COVID-19 pandemic on the use of and access to essential health services, we used a robust mixed-effect model with random intercepts and random slopes for each county over time on each of the 17 indicators of essential health services across the pre-pandemic period, the pandemic period, and the period during the health-care workers’ strike.6 The robust mixed-effect model was selected to account for the correlation between observations from the same county, and to minimise the influence of outliers or other contamination on model estimates.26 Before running the robust mixed-effect models, we analysed the time-series data for each indicator to test for seasonality using the Friedman rank test implemented in the R package seastest.27 For indicators with a seasonal effect, we included the month as a fixed effect in the model. The model equations used are as shown here: Where Y t represents a study indicator of essential health services, β0 represents the estimated incidence for every indicator at the beginning of the pre-pandemic period, β1 represents the average monthly change in the incidence over the pre-pandemic period, T t represents the time since the start of the study period, β2 is the change in incidence immediately after the COVID-19 period, which is represented by X t, β3 represents the average difference in trend in incidence between the pandemic and pre-pandemic period, β4 is the change in incidence before and immediately after the health-care workers’ strike, which is represented by S t, β5 represents the estimated difference between the pandemic period before the health-care workers’ strike and the strike period, βi represents other independent variables, W t, which comprise the hours of the nationwide dusk-to-dawn curfew, mobility, attack rate, population density, and seasonality. β6 is the change in incidence immediately after the end of the health-care workers’ strike, which was still a pandemic period represented by Z t. β7 is the average difference in trend in incidence between the pandemic period during and after the strike period. The monthly curfew hours, mobility (both within and between countries), and attack rate were averaged for every county during the pandemic period, with a value of 0 given for the months before the pandemic. The random effects (county) were represented by γ and the error terms were represented by ɛt. The linear trend during the pandemic (β1 + β3) and during the health-care workers’ strike period (β1 + β5) were calculated. All the analysis and data visualisation was done using R statistical software (version 4.0.2). The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report.