Menu
Menu
Menu
Menu
Background: Health workers’ strikes are a global occurrence. Kenya has had several strikes by health workers in recent years but their effect on mortality is unknown. We assessed the effect on mortality of six strikes by health workers that occurred from 2010 to 2016 in Kilifi, Kenya. Methods: Using daily mortality data obtained from the Kilifi Health and Demographic Surveillance System, we fitted a negative binomial regression model to estimate the change in mortality during strike periods and in the 2 weeks immediately after strikes. We did subgroup analyses by age, cause of death, and strike week. Findings: Between Jan 1, 2010, and Nov 30, 2016, we recorded 1 829 929 person-years of observation, 6396 deaths, and 128 strike days (median duration of strikes, 18·5 days [range 9–42]). In the primary analysis, no change in all-cause mortality was noted during strike periods (adjusted rate ratio [RR] 0·93, 95% CI 0·81–1·08; p=0·34). Weak evidence was recorded of variation in mortality rates by age group, with an apparent decrease among infants aged 1–11 months (adjusted RR 0·58, 95% CI 0·33–1·03; p=0·064) and an increase among children aged 12–59 months (1·75, 1·11–2·76; p=0·016). No change was noted in mortality rates in post-strike periods and for any category of cause of death. Interpretation: The brief strikes by health workers during the period 2010–16 were not associated with obvious changes in overall mortality in Kilifi. The combined effects of private (and some public) health care during strike periods, a high proportion of out-of-hospital deaths, and a low number of events might have led us to underestimate the effect. Funding: Wellcome Trust and MRC Tropical Epidemiology Group.
We obtained data for our study from the KHDSS. This health database covers an area of 891 km2 comprising both rural and semiurban regions in Kilifi county in coastal Kenya, which has a population of approximately 300 000 people.22 Information on pregnancies, births, deaths, and migrations within the KHDSS is updated every 4 months, and cause of death data are obtained using verbal autopsies.23 The crude death rate within the KHDSS for the period 2006–10 was 5·85 deaths per 1000 person-years of observation,22 and the prevalence of HIV in Kilifi county in 2015 was 4·4%.24 The main referral facility within the KHDSS is the Kilifi County Hospital (KCH), which is a level 4 government-run facility25 that provides both inpatient and outpatient services. Additional facilities within the KHDSS comprise three government-run health centres, 16 dispensaries, and approximately 42 private health facilities, of which only about 10% offer inpatient services. The proportion of babies born at home within the KHDSS decreased from 53% in 2012 to 30% in 2016. The proportion of babies in the KHDSS born at a health facility increased from 44% in 2012 to 68% in 2016. At KCH, inpatient services are provided for adults in the adult and maternity wards and for children in the general paediatric ward and in the high dependency unit (HDU). The HDU is run by the KEMRI-Wellcome Trust Research Programme. On average, 20% of paediatric patients are admitted to the HDU and 80% to the general paediatric ward. The general paediatric ward has a 70-bed capacity and is staffed on average with two nurses, five clinical officer interns, two medical officer interns, and one consultant paediatrician. The HDU has six beds, six cots, and four incubators and is staffed on average with three nurses, three clinical officer interns, one medical officer intern, and two consultant paediatricians. Approximately 4% of all deaths recorded in the KHDSS and 38% of all paediatric inpatient deaths occur in the HDU. During strikes by health workers in the period 2010–16, service delivery was disrupted by the striking primary staff cadre (eg, doctors), which led to hampered service delivery by other non-striking staff cadres (eg, nurses). All non-striking staff were expected to be present at their respective facilities. During strike periods, the HDU at KCH remained operational but limited paediatric inpatient services were provided. Admissions to this unit were restricted to the most critical cases because of the limited capacity, although there was a provision for multiple patients to share a bed. Staff in the HDU provided services only within this unit and were not redistributed to other hospital departments to offer services such as high-risk deliveries or caesarean sections. No arrangements were made for hiring replacement staff at KCH during strike periods. Mortality data within the KHDSS were obtained with approval from the Kenya Medical Research Institute Scientific Ethics Review Unit (SSC 1348). We ascertained the dates of strikes by doctors, nurses, or both by searching through digital archives of Kenyan newspapers. We confirmed these dates by checking admission data from KCH. A strike was defined by a national announcement of cessation of service provision by doctors, nurses, or both in government hospitals across the country and a concomitant decline in admissions at KCH. We obtained dates of death from an existing electronic database of admissions to KCH (maintained by KEMRI-Wellcome Trust Research Programme), from people living in the same or a neighbouring homestead who knew the deceased, or by both these ways. In the primary analysis, we excluded deaths for which the exact date on which the death occurred was not known. The routine verbal autopsy and cause of death allocation process in the KHDSS has been previously described.23 Briefly, people with information on the deceased were interviewed using standard WHO verbal autopsy questionnaires. From their responses, causes of death were assigned using the InterVA-4 computer-based probabilistic model.23 We categorised causes of death assigned by verbal autopsy into six broad groups—maternal, medical, surgical, medical or surgical, trauma-associated, and other causes. We defined childbearing age as 15–49 years old. The period of analysis was from Jan 1, 2010, to Nov 30, 2016. We did not include December, 2016, in our analysis because another strike by health workers started at the beginning of this month and continued into 2017, covering a period for which KHDSS data were not available at the time of analysis. The mid-month population in the KHDSS was used to estimate person-days of observation. We combined all strike periods to form a strike days category (referred to as the strike period) and all non-strike periods to form a non-strike days category (referred to as the non-strike period). We compared mortality during the strike period with mortality during the non-strike period using negative binomial regression to account for overdispersion. We used Newey West SEs to adjust for autocorrelation, allowing a lag of up to 7 days.26 We calculated the maximum non-zero lag using the formula 4(n/100)2/9, for which n refers to the length of the time series.26 We adjusted for categorical variables that we had identified a priori as possible confounders—namely, year and month (adjusting for long-term mortality trends and seasonality, respectively), day of the week, and public holiday. Prespecified subgroup analyses included comparisons of mortality during the strike and non-strike periods by age group, cause of death, and strike week. We investigated possible delayed effects of strikes on mortality by comparing mortality in the first and second weeks immediately after a strike with mortality in the non-strike period. We checked whether inclusion of interaction terms for age group and strike week improved the regression model using a multiparameter Wald test. The reference category in the strike week analysis was the non-strike period. Finally, we did two sensitivity analyses to examine the effect of excluding deaths with missing dates. In the first sensitivity analysis, we included all these deaths but assigned the date of death as the 15th of the month in which the death was reported to have occurred, and we included an indicator variable for day 15 in the model. In the second analysis, we ran the regression analysis on 100 imputed datasets, in which the date of death was randomly assigned to any day within the month in which it occurred. We did statistical analyses with Stata, version 15.1. The funder had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all data in the study and had final responsibility for the decision to submit for publication.
