Community health worker interventions are key to optimal infant immunization coverage, evidence from a pretest-posttest experiment in Mwingi, Kenya

Introduction Immunization is a powerful and cost-effective health intervention which averts an estimated 2 to 3 million deaths every year. Kenya has a high infant and under five mortality and morbidity rates. Increasing routine child immunization coverage is one way of reducing child morbidity and mortality rates in Kenya. Community Health Workers (CHWs) have emerged as critical human resources for health in developing countries. The Community Strategy (CS) is one of the CHW led interventions promoting Maternal and Child Health (MCH) in Kenya. This study sought to establish the effect of CS on infant vaccination Coverage (IVC) in Mwingi west sub-county; Kenya. Methods This was a pretest - posttest experimental study design with 1 pretest and 2 post-test surveys conducted in intervention and control sites. Mwingi west and Mwingi north sub-counties where intervention and control sites respectively. Sample size in each survey was 422 households. Women with a child aged 9-12 months were main respondents. Results Intervention site end-term evaluation indicated that; the CS increased IVC by 10.1% (Z =6.0241, P <0.0001), from a suboptimal level of 88.7% at baseline survey to optimal level of 98.8% at end term survey. Infants in intervention site were 2.5 times more likely to receive all recommended immunizations within their first year of life [(crude OR= 2.475, P<0.0001; 95%CI: 1.794-3.414) (adj. OR=2.516, P<0.0001; 95%CI: 1.796-3.5240)]. Conclusion CS increased IVC in intervention site to optimal level (98.8%). To improve child health outcomes through immunization coverage, Kenya needs to fast-track nationwide implementation of the CS intervention.


Introduction
Immunization is a powerful and cost-effective health interventions which averts an estimated 2 to 3 million deaths every year [1,2].
World Health Organization (WHO) considers a child to have received all basic vaccinations if he or she has received: BCG vaccination against tuberculosis; three doses of DPT vaccine to prevent diphtheria, pertussis, and tetanus (or three doses of pentavalent, which includes DPT and vaccinations against both hepatitis B and haemophilus influenza type B); at least three doses of polio vaccine; and one dose of measles vaccine [2]. These vaccinations should be received during the first year of life [2]. In Kenya, an infant is considered to be fully vaccinated if the infant has received all WHO basic vaccinations and three doses of pneumococcal vaccine [3].
Global vaccination coverage -the proportion of the world's children who receive recommended vaccines -has remained steady for the past few years [1] . Since 2000, Global Alliance for Vaccines and Immunization (GAVI) and the Vaccine Alliance has supported vaccination of 500 million children in the world's poorest countries, saving an estimated 7 million lives [1]. Global routine vaccination coverage by 2014 was 86% -up from 20% in 1980 [1]. The world is closer to realizing a world free from polio. In 1988 there were  [3]. This coverage is low compared to 2015 global routine child immunization coverage of 86% [1]. This could be the cause of high infant mortality rate reported as 39 deaths per 1,000 live births and high under five mortality rates (52 deaths per 1,000 births) by 2014 KDHS [3]. In Kitui county where Mwingi west sub-county is located, Routine child immunization coverage is 52.7%; even lower than the national coverage [3]. These statistics point out to one thing; the need to design and implement innovative interventions to help increase child routine immunization coverage in Kenya. This in turn could help in reducing the high infant and under five mortality rates in the country.
Community health workers (CHWs) have emerged as critical human resources able to extend health systems and basic services directly to communities and households [5]. In resource poor countries, CHWs are increasingly recognized as a critical link in improving access to health services [5]

The study area
This was an experimental study with intervention and control site.  [16].

The intervention
CS is a CHW led intervention with the following key elements;

The research design
This was a non-randomized prospective experimental study in which 1 pre-test and 2 post-test time series household surveys were conducted in both intervention and control sites. Data was collected at 3 time points; a baseline survey was used to collect baseline data in both intervention site and control sites. First post intervention survey data was collected 9 months after implementation of the CS in intervention site and control site. This survey was defined as mid-Page number not for citation purposes 4 term evaluation. Second post intervention survey data was collected in both intervention and control sites 18 months after implementation of the CS. This is defined as end-term evaluation survey. Data was collected at household level with women of reproductive age with a child aged 9-12 months being the main respondents. Based on nature of phenomena to be examined, data was collected from different participants in all the three surveys. For example, it was not possible to guarantee that a woman who was sampled at baseline survey and data collected on her quality of ANC services provided, place of delivery etc., will be expectant again after 9 months or even 18 months to enable investigators to measure the same parameters again. This informed the choice of having different participants at baseline, midterm and end-term evaluation surveys.

Sample size determination
Reference [17] provides the Fisher's formula for calculating a representative sample size of a population with more than 10,000 participants. After employing this formula, a representative sample size of 384 households was established. Thirty-eight households (10 percent of 384 households) were added into this sample in order to carter for non-response. A total sample size of 422 households was determined.

Sampling procedure
Purposive and simple random sampling methods were employed.
Purposive sampling was used to identify intervention and control sites. Mwingi west Sub County was purposively selected as intervention site based on the fact that the CS program was to be implemented in the sub county. Mwingi north sub county was also purposively sampled as the control site based on the following; CS was not under implementation in the sub county, the sub county borders Mwingi West and both sub-counties have many similarities which include similar ecological and climatic characteristics [15].
Simple random sampling was applied in all the pre-and postintervention surveys in the study and control sites. The first step was to develop a sampling frame for each of the three surveys conducted in the study site and the control site respectively.
Sampling frames in Mwingi west Sub County was 1243 households (in Waita CU) at baseline and 927 households (in Kyethani CU) and 1107 households (Wikithuki CU) at midterm and end term surveys respectively. The sampling frame was developed using household registers which were developed during creation of CUs. In the control site, the researchers together with village elders and local chiefs conducted a series of community meetings to help in identification of households with a child or children aged between 9-12 months. This was done in Kyuso, Ngomeni and Mumoni wards. A sampling frame of 971 households, 1032 households and 1208 households was developed in Kyuso, Ngomeni and Mumoni wards respectively. Using SPSS a sample size of 422 households was drawn from each sampling frame.

Data collection process
The first step in data collection was to conduct a pre-intervention survey to collect baseline data in both intervention and control sites.
The aim was to obtain pretest measurements on both intervention and control groups to allow assessment of initial comparability of the two groups. In the intervention site, baseline data was collected

Variables in the study
The independent variable is the study was CS intervention while the influenza type B)) (given at 6wk, 10wk and 14wk), Pneumococcal vaccine-PCV 10 (administered in 6wk, 10wk and 14wk), and Measles (first dose administered at 9 months). The outcome variable in this study was change in IVC. It was measured in two ways: one; change in the proportion of infants in Mwingi west sub-county who were fully vaccinated by the age of 1 year and two; change in probability that an Infant would be fully vaccinated before and after the intervention.

Study validity and reliability
A pilot study was conducted in Nzeluni in Mwingi west sub-county before the main study. The objective of the pilot was to test the reliability of data collection tool. Data was collected in a randomly selected sample of 45 households (slightly above 10 per cent of the sample size) in three villages in Nzeluni sub location. Upon testing the data on reliability, the coefficient of internal consistency (Cronbach's alpa) was 0.864. This value was within the recommended range of 0.70-0.95 [18] and therefore we were assured that the data collection tool (questionnaire) was reliable.
Internal validity of the study was ensured by applying a sound methodology while external validity was ensured by use of a representative sample size.

Data analysis and presentation
Frequencies and percentages were used to provide descriptive statistics in this study. Z score tests were used to determine if proportions of IVC before and after the intervention were significantly different. Binary logistic regression was used to control for potential confounders (socio-demographic characteristics) and to establish the odds of infants who had received all recommended vaccines within 1 year before and after the intervention. Data was presented using tables.

Study limitations
The study had several important limitations; the most important of these was selection of intervention and control sites. Since implementation of the CS was a MoPHS and AMREF-Kenya project which was designed to be implemented in Mwingi West sub county as a whole, it was not feasible to randomly assign the CS intervention to community members in Mwingi west sub county. This is the reason why a non-randomized pre-test and post-test experimental study design was deemed appropriate. Though this method has been employed in other similar studies [8,[19][20][21][22] the design is weaker compared to a randomized controlled trial.
Secondly, researchers were also not able to account for possibility of other programs that could influence MCH outcomes of interest in the intervention site. However, there was an attempt to reduce the effect of confounding factors through, treating socio-demographic characteristics of both intervention and control sties as potential confounders and having them controlled in the binary logistic regression model used in data analysis, and by matching the control to the intervention sites by geographical location, and infrastructural characteristics.
Part of data collection involved collecting data from a Mother and Child Health (MCH) booklet at the household level. In the event that this booklet was not available, respondents were requested to remember the MCH events that happened in a span of 12 months.
Though this method has been successfully used in other studies including Demographic and Health Surveys (DHS) [3], the method introduced a retrospective data collection aspect that required respondents to recall past events. Though this was limited only to respondents who could not produce their mother and child booklets, it was a potential source of recall bias error.

Ethical considerations
Ethical clearance for this study was provided by the National Council of Science and Technology (NCST) of the Government of Kenya (GoK).

Socio-demographic characteristics of respondents
Data on socio-demographic characteristics of respondents of this study is summarized in Table 1.  Table 2.

Infant vaccination Coverage (IVC) in intervention site and control site
IVC: defined as the proportion of children aged 1 year and below who have received the basic WHO recommended vaccines plus three doses of pneumococcal vaccination [2] was as follows; 87.7%, 92.5%, and 98.8%, at baseline, midterm and end-term surveys in intervention site respectively and 84.4%, 83.3%, and 86% at baseline, midterm and end term surveys in control site respectively. These results are summarized in Table 3.

Effect of CS on IVC
Effect of CS intervention on IVC was estimated in two ways: one; by measuring if there is a significant difference between IVC before and after CS intervention in both intervention and control group, and two; by comparing the odds of infants who had received all recommended vaccines within 1 year before and after CS intervention in both intervention and control group.  Table 3. respectively. These results are summarized in Table 4.

Hypothesis testing
The null hypothesis in this study was; 'In the intervention arm, there was no difference in the odds of infants who received all GoK recommended vaccines within 9 to 12 months of life at baseline survey compared to end-term survey'.  Table 4.
Based on this test, the null hypothesis was rejected and the alternative hypothesis (In the intervention arm, there was a significant difference in the odds of infants who received all GoK recommended vaccines within 9-12 months of life at baseline survey compared to end-term survey) was adopted.

Effect of CS on immunization coverage of specific vaccines in the RCIP
This data reveals that coverage of vaccines recommended in RCIP improved in intervention site compared to control. As shown in Table 2 found to shun MCH service utilization including immunization services [23]. As shown in Table 2, the slight increment in immunization coverage of specific immunizations at end-term survey compared to midterm survey in control site was within range with baseline immunization coverages in the same site. Though no tests were done to show if proportions in specific immunization coverages were significantly different in intervention and control site, data indicates that in intervention site, end term survey immunization coverages were all at optimal levels compared to control site. This could only be attributed to the effect of CS intervention. Table 3, IVC in intervention site increased from baseline survey to midterm survey and from baseline survey to endterm survey by 3.8% and by 10.1% respectively. Z score tests revealed no significant difference in baseline IVC and midterm IVC proportions and a significant difference between baseline IVC and end-term IVC proportions (Z =6.0241; P <0.0001). In control site, Page number not for citation purposes 8 midterm survey IVC decreased by 1.1% compared to baseline survey IVC. In the same site end-term IVC increased by 1.6% compared to baseline survey IVC. As shown in Table 3, Z score tests revealed no significant difference between midterm IVC and baseline IVC as well as between end-term IVC and baseline IVC.  [24]. A WHO report on global experience of CHWs in delivery of health-related Millennium Development Goals (MDGs) indicates that CHWs played a critical role in improving MCH outcomes in the last decade through improving child immunization coverage in resource poor countries. CHWs role according to WHO was tracking child immunization defaulters and referring them to health facilities [25]. A review of 17 studies conducted in 10 developing countries reported that CHWs were effective in promoting child immunization coverage [26]. In Kenya a study conducted to evaluate effectiveness of the CS in promoting positive health outcomes also reported that CS was effective in improving measles immunization coverage [18]. These studies support that the considerable increase in IVC in the intervention site was the effect of CHWs working in the CS intervention and not by chance.

Conclusion
In Mwingi west sub county (intervention site), IVC improved from suboptimal level (88.7%) to optimal level (98.8%). Though the study deign limitations reduces the strength of evidence in this study, it is highly probable that the observed increase in IVC in intervention site was due to effect of CS in the site. To improve child health outcomes through immunization coverage, Kenya needs to fast-track nationwide implementation of the CS intervention.
What is known about this topic  Immunization is one of the most powerful and costeffective of all health interventions and averts an estimated 2 to 3 million deaths every year;  CHW led interventions have been associated with improved MCH outcomes.

What this study adds
 This study provides evidence proofing that a CHW led Primary Health Care (PHC) intervention in Kenya increased infant vaccination coverage to optimal levels.
Authors would like to thank all respondents for their willingness and consent to participate in the study. We would also like to acknowledge and thank all research assistants for their tireless efforts without which this research work would not have come into existence.