Study Design and Setting
The data for this study are drawn from the Lown scholars survey on healthcare gaps in informal-settlements conducted between June 2018 and July 2018 in Viwandani slums (Nairobi, Kenya). A total of 300 randomly selected households from the Nairobi Urban Health Demographic Surveillance System (NUHDSS) were surveyed in the Lown scholars study. The respondents comprised of adults aged 18 years and over who were either household heads or permanent household members. The design of the Lown scholars study has been previously published (20).
Data Collection
A structured interviewer-administered questionnaire was used to collect data from the respondents. The interview questions consisted of healthcare utilization patterns; health insurance coverage, timeliness of care, distance to the nearest primary care center, availability of essential healthcare services, affordability, acceptability, quality of care and treatment procedure. Other questions included; perceived health status, out-of-pocket healthcare expenditure and average monthly expenditure on health insurance. The responses were electronically recorded in a tablet. Secondary data on wealth tertile of the selected households were obtained from the latest NUHDSS data, which is run by the African Population and Health Research Center in Viwandani and Korogocho slum settlements.
Research model
Our study is underpinned by Panchansky and Thomas’ theory of access to primary healthcare (19). Previous studies in Nigeria (21) and Istanbul (22) have also adopted this model. Penchansky and Thomas’ theory proposes a taxonomic definition of "access." This theory summarizes a set of specific metrics that describe the fit between the healthcare system and the general population. These metrics are; availability, accessibility, accommodation, affordability, and acceptability of healthcare services. In particular, Penchansky and Thomas’ metrics of access form a formidable chain of access to primary care that is no stronger than its weakest link (19).
Using Panchansky and Thomas’ theory, we conceptualized seven independent and interconnected dimensions of access. These dimensions are health insurance coverage, timeliness of care, distance to the nearest primary care center, and availability of essential healthcare services, affordability, and acceptability, quality of care and treatment procedure. Health insurance coverage measures the extent of financial protection of patients from unexpected or high cost of healthcare services, distance to the nearest primary care center measures geographic accessibility of healthcare services. Timeliness of care measures the level of responsiveness of the health facility to the needs of the patient. Affordability measures the relationship between the costs of healthcare services versus the willingness and the ability of the patient to pay for the services. Availability determines the presence of requisite healthcare resources, such as infrastructure, personnel, technology and essential supplies needed to meet the healthcare needs of the patients. Quality of care and treatment procedure reflects the operational organization of the provider in a manner that meets the preferences of the patients. Acceptability measures the extent of comfortability of the patients with immutable characteristics of the healthcare service provider such as sex, age, ethnicity, and social class.
Measurements
Dependent variable
A proxy index for access to primary care was created based on healthcare utilization variables including timeliness of care, distance to the nearest primary care center, availability of essential healthcare services, affordability, acceptability, quality of care and treatment procedure. Access index was finally computed using principal component analysis (PCA) and varimax rotation method. Principal components are weighted averages of the variables used to construct them. The computed index was finally used to classify the sampled households into three categories (tertiles): poorest, middle, and highest. The first eigenvalue of the PCA was 1.72 and the proportion of variance explained by the first three components was 58%.
Independent variables
We conceptualized two categories of predictor variables (individual and household-level factors). Individual factors comprised of age, sex, level of education, employment status and perceived health status. Household-level factors comprised of the sex of the household head, household size, wealth tertile, the primary source of care and quarterly out-of-pocket (OOP) expenditure on healthcare. The quarterly OOP comprised of the total expenditures on consultation, diagnostics and laboratory tests, medication, emergency, and/or specialized care (such as dental care) in the three months preceding this survey. It was grouped into four categories: those who spent ≤ $5, between $5 and $9.9, between $10 and $29.9 USD and ≥$30. A wealth index was generated using PCA from socio-economic variables including type of dwelling, ownership of the dwelling, construction materials of the dwelling, source of cooking fuel, the source of lighting fuel, household possessions/goods, the source of water for household consumption and type of sanitation facility. The households were grouped into tertiles based on the generated wealth index (lowest, middle, and highest).
Data analysis
The outcome variable for this study is access to primary healthcare services. Descriptive statistics were used to summarize the background characteristics of the respondents and the frequency distribution of access to primary care with individual, household and community level factors. The outcome variable assumes an increasing order. Hence a multivariate ordered logistic regression was conducted. We fitted a proportional odds model; however, following the Brant test, we found that the critical assumption of parallel slopes[1] was violated in some of the covariates (age group, education level and primary source of care). Consequently, we implemented a partial proportional odds model, which is less restrictive, and relaxes the proportional odds assumption, allowing the effect of the explanatory covariates to vary (23). More information on the partial proportional odds model is available in numerous sources (24, 25). All explanatory variables in the unadjusted partial proportional odds model that were associated with the outcome were added to the adjusted partial proportional odds model. Results of the multivariate logistic regression compare a continuum of households ranging from those that have low access, moderate access to those that have adequate access to healthcare. We used Stata version 15.1 and statistical significance defined as a p -value less than 0.05 (2-sided). The gologit2 (26) Stata command was used to fit the partial proportional odds model.
Ethical considerations
The protocol for this study was reviewed and approved by the African Medical and Research Foundation based in Nairobi, Kenya (P482/2018). Written informed consent was sought from all respondents prior to participation.
[1] One assumption underlying the ordered logistic is the parallel lines assumption/proportional odds which posits that the relationship between each pair of the outcome groups is the same.