Variability in distribution and use of tuberculosis diagnostic tests in Kenya: a cross-sectional survey

Kenya is administratively divided into 47 counties, that are now largely responsible for health care since devolution in 2013 [29]. Health services are provided by public, private, non-profit non-governmental organisations (NGO) and faith based organisations (FBO). The healthcare system is structured in a hierarchical manner, starting with primary healthcare in the community and complicated cases referred upwards to secondary and tertiary levels of healthcare [29]. According to the Kenyan Master Facility List, there are approximately 10,000 health facilities in the country and just about half are TB treatment sites [30‐32]. For this analysis, we aggregated TB health facilities into two: lower level (dispensaries, health centres, and maternity/nursing homes) for primary care; and higher level (county hospitals and national referral facilities) that provide secondary and tertiary referral services.

We included patients of all ages who were notified to the Kenya National TB programme and started TB treatment in 2015.

We wished to explore the possible influence on use of Xpert MTB/RIF® of variables in three hierarchical levels: county, health facility and individual. Individual level co-variates included: age; gender; HIV status; nutritional status; and type of TB patient i.e. new/transfer-in or retreatment cases (relapse/defaults/treatment failures). An age cut-off of 15yrs for children is used locally and internationally for TB programming [14], therefore “child” here refers to 0-14yr olds. For nutrition status, weight-for-age Z (WAZ) scores were computed for those aged 0-23yr and body mass index (BMI) truncated at -5 to +5SD for those ≥10yr, and patients classified as underweight according to the scores. Those 10-23yr who met criteria of underweight by either criterion were also classified as underweight. Health facility level co-variates included: sector of care (public, private, or FBOs), level of care provided (higher vs lower level) and whether they were an Xpert® site or not. County level co-variates included: poverty; maternal education levels; travel time to nearest health facilities; and availability of Xpert® facilities per 100,000. All co-variates were determined for each of the 47 counties for 2015. These factors were decided upon a priori following review of literature on drivers of use of TB and health care services in general and data availability [21, 22].

The primary outcome of interest was evidence of Xpert® being done in patients who had been started on TB treatment. According to the 2015 Kenyan guidelines, all presumptive paediatric, HIV-infected smear negative, drug resistant (DR) cases, or retreatment cases i.e. relapse (recurrence)/defaults/treatment failures should have had at least one Xpert® assay as part of their diagnostic work up (Additional file 1) [15, 16].

We considered the best data sources for the three levels of variables of interest. For the individual level, de-identified data from the Kenya National TB Programme patients’ treatment register for 2015 were used (patients who were notified and/or started treatment in 2015). TB case definitions were as per Guidance for National TB programmes (Additional file 1) [33]. Health facility level data were from the Kenya Master Health Facility list [30] and Kenya Services Availability and Readiness Assessment Mapping Report (SARAM) 2014 [34]. Health facilities in the national TB register were geocoded using KEMRI-Wellcome Trust’s Kenya Health Facilities Database, which was last updated in 2016 using online digital place-name gazetteers and Global Positioning System (GPS) sources. County level data were from each county governments’ integrated development plans for 2015 [35]. The projected 2015 Kenya gridded population distribution surface at 100m spatial resolution was obtained from the WorldPop project [36, 37].

Stata version 15MP (StataCorp.2017, College Station, TX, USA) and ArcGIS 10.5 (ESRI, Redlands, CA, USA) were used for statistical analysis, and mapping and spatial analysis respectively. We described the proportion of adults and children reported to the TB programme, their socio-demographic characteristics, use of TB diagnostic tests and outcomes. We used the 2015 Kenya National TB programme data to construct maps for each TB case notification rate (CNR) in different counties and linked this with their capacity to perform diagnostic tests (chest x-rays, smear microscopy, Xpert®, culture and line probe assay) from the SARAM report [34].

For adherence to guidelines, we only had data for those who had tests done, and used these patients to describe patterns of use of TB diagnostic tests. Co-variates of theoretical and/or statistical significance were used to build hierarchical logistic regression models to establish determinants of use of Xpert® in adults and children. Possible collinearity was assessed using the variance inflation factor (VIF). Variables with VIF less than 10 were considered for analysis. The models converged at five integration points for complete case analysis, with county and health facility as random effects [38]. Models were built for the 0-14yrs and ≥15yrs separately and a model for the total population, with likelihood ratio tests, exploration for interactions in pre-specified covariates (HIV and nutrition status) and quantile-quantile plots of residuals used to determine best fit as seen in Additional file 2 [38, 39].

Figure 1 illustrates diagnostic practices amongst those patients who were started on anti-TB treatment. More than a third of children 0-14yr had no diagnostic test done, and were started on treatment based on clinical diagnosis only. Chest X-ray was the commonly used test in this age group (37.1%); while microscopy was the commonest amongst the ≥15yrs. Table 2 illustrates the patterns of use of bacteriological tests amongst those who had these tests done, relating them to the 2015 Kenyan guidelines (Additional file 1). Children 0-14yr represented <5% of those who had either Xpert® or Microscopy or Culture/LPA done. Those with extra-pulmonary TB also rarely got any of the tests done. Retreatment cases got a culture/LPA done more than any other test. Forty one percent of the patients got Xpert® as per guidelines, with better guideline adherence for culture/LPA (61.2%) and microscopy (97.8%), respectively.

Figures 2 and 3 and Additional files 3 and 4 show case notification rates (CNR) by age group and county, as well as variation in county use of Xpert® and microscopy. Median age was 32yr (IQR 24yr, 43yr). The highest CNR was amongst those 35-44yrs (455/100,000), and the lowest amongst the 5-9yrs (27/100,000) Fig. 2. The median CNR/100,000 amongst children 0-14yr old was 37.2 (IQR 20.6, 41.0) and 267.4 (IQR 202.6, 338.1) for ≥15yr olds respectively. Median use of Xpert® among 0-14yr olds was 1.45/100,000 (IQR 0.7, 2.1) and in ≥15yr olds was 29.9/100,000 (IQR 21.0, 42.8). Median use of microscopy among 0-14yr olds was 8.1/100,000 (IQR 6.6, 12.8) and 222/100,000 (IQR 169.8, 270.4) in ≥15yr olds. Xpert® use did not correlate with increase in county TB CNR/100,000 (Fig. 3 and Additional file 4) and in this was also observed in the univariate analysis where counties with higher Xpert® density did not have significantly higher CNRs (Additional file 2). Microscopy use in the ≥15yr tended to be higher in counties with highest CNRs (Fig. 3 and Additional file 4).

Figure 4 shows a panel of maps illustrating spatial distribution of TB patients in Kenya. The counties with two large cities had some of the highest CNRs, as did two counties serving the pastoralist communities of the northern and eastern frontiers (panel A). Panel B and C show the CNRs in the ≥15yr olds (“Adults”) and in children 0-14yr old respectively, while D shows the ratio of adult to child TB cases. These maps highlight some of the adult hot spots with low child CNRs and ensuing differences in adult: child ratios highlighting the gaps in case detection for children.

Figure 5 and Additional file 5 show the distribution of facilities with TB diagnostic facilities (Microscopy-Map A; Xpert MTB/RIF®-Map B; Culture and Line Probe Assay- Map C; and chest X Ray-Map D) overlaid with each county’s total population density. In 2015, there were 129 facilities providing Xpert® services (mean Xpert® density of 0.28/100,000); 304 chest X ray facilities (mean density 0.67/100,000); and 2,072 facilities with microscopy services (mean density 4.54/100,000). There were disparities noted in the distribution of Xpert® (panel B) and chest X-ray (panel D) facilities with some northern and eastern counties of Kenya having low facility densities/100,000 yet they carry a high burden of TB cases (Fig. 4).

Additional file 2 shows the univariate analysis of the factors that had been considered for the model building, and Table 3 shows the final adjusted hierarchical models of determinants of use of Xpert® in adults and children, with county and health facility as random effects. Amongst individual level factors, the retreatment cases (i.e. failures, relapses/recurrences, defaulters) had the highest odds of getting an Xpert® (total population AOR 7.81, 95% CI 7.33 to 8.33). From the model of the total population, children had reduced odds of getting an Xpert® test (AOR 0.41, CI 0.36 to 0.47), while being male was associated with increased odds in all age groups. Overall, the HIV positive individuals had nearly twice the odds of getting an Xpert® compared to the HIV negative. Nutrition status had a marginal effect, except in the 0-14yr olds where the underweight group had nearly twice the odds of getting an Xpert® compared with the well-nourished cases.

Examining factors at the facility and county levels, patients in private sector and from higher level facilities had a tendency towards lower odds of getting tested compared with those in public sector or lower level facilities, but this difference was not statistically significant for the 0-14yrs old. We also noted that counties with higher Xpert® density did not report significantly higher CNRs. From the intra-class correlation coefficients (ICC), county as a level explained approximately 4 & 7% and health facility 24 & 26% of the observed variability in each adjusted model.

These data are from notification data which may be incomplete because an unknown number of cases may be treated but not reported. We additionally had no information on those in whom a test was done and not documented, or in whom a test was attempted but failed. A large proportion of relevant cases probably had no record of a test being done and no indication of why a test was not done.

This work, however, still provides much needed insights about the utilisation of diagnostic tests amongst those accessing health care who have been diagnosed and started on TB treatment. It also contrasts with the prevalence survey, where up to 40% of TB cases were missed and, thus, got no TB diagnostic tests. The guidelines recommend that all patients should have a bacteriological test prior to treatment. We evaluated, with the information available, how well the guidelines were being followed.

Our complete case analysis dropped approximately 3% of the cases due to missing data on facility, but this did not reduce our ability to make meaningful inferences. We also considered all the 0-14yrs old as a single group but there may be differences in the sub categories. Because the numbers of children were low, subgroup analysis could have been misleading. This analysis together with the prevalence survey still provide deep insights into the known TB cases and guideline adherence, which can aid in planning services to improve TB case detection, especially in children.