Clustering and risk factor analysis of pulmonary tuberculosis in a district in Ethiopia: a population-based cohort study

Abstract

Objective

A ‘cluster’ is an area with a higher occurrence of tuberculosis (TB) than would be expected in an average random distribution of that area. Tuberculosis clustering is commonly reported in Ethiopia, but most studies rely on registered data, which may miss patients who do not visit health facilities or those who attend but are not identified as having TB. This makes the detection of actual clusters challenging. This study analysed the clustering of pulmonary TB and associated risk factors using symptom-based population screening in Dale, Ethiopia.

Design

A prospective population-based cohort study.

Setting

All households in 383 enumeration areas were visited three times over a 1-year period, at 4-month intervals.

Participants

Individuals with pulmonary TB aged ≥15 years with demographic, socioeconomic, clinical and geographical data residing in 383 enumeration areas (ie, the lowest unit/village in the kebele, each with approximately 600 residents).

Outcome measures

Pulmonary TB (ie, bacteriologically confirmed by sputum microscopy, GeneXpert or culture plus clinically diagnosed pulmonary TB) and pulmonary TB clustering.

Results

We identified pulmonary TB clustering in 45 out of the 383 enumeration areas. During the first round of screening, 39 enumeration areas showed pulmonary TB clustering, compared with only 3 enumeration areas in the second and third rounds. Our multilevel analysis found that enumeration areas with clusters were located farther from the health centres than other enumeration areas. No other determinants examined were associated with clustering.

Conclusions

The distribution of pulmonary TB was clustered in enumeration areas distant from the health centres. Routine systematic community screening may be costly, but using existing health infrastructure with health extension workers through targeted screening, they can identify and refer persons with TB symptoms more quickly for diagnosis and treatment, thereby decreasing the duration of disease transmission and contributing to the reduction of TB burden.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• The study employed a three-round, total household symptom screening approach to identify both diagnosed and undiagnosed tuberculosis (TB) cases, enabling the identification of true disease clusters.

• Including visits to all households in all rounds using the existing health workforce for screening led to high coverage and low loss to follow-up, ensuring a more accurate population presentation.

• A delay in collecting residential addresses led to 17 missed pulmonary TB cases, potentially affecting the accuracy of cluster identification.

• The study may have missed asymptomatic TB cases, as not all individuals with TB exhibit symptoms.

• Smear microscopy will, due to its relatively low sensitivity, always miss some cases of TB.

Background

Although tuberculosis (TB) is preventable and curable, globally, an estimated 1.25 million people died from TB in 2023. Modelling shows that 2.7 million TB cases were not detected or reported globally, which sustains disease transmission and suffering in the community. According to the WHO, Ethiopia has a high burden of TB, with estimated incidence rates of 146 per 100 000 population.¹ The TB burden is particularly acute in certain areas, creating a ‘TB belt’ in the southern part of the country. Higher notification rates in the study area than the national average disproportionately affect these communities.

Efforts to improve services to the most vulnerable and disadvantaged populations using existing health systems are among strategic priorities.^{2 3} Clustering of TB in an area means there are more cases than expected from random distribution; this represents an increased risk of transmission³ but can also provide valuable information for the national TB programme. Clusters can be measured using surveillance data, genotypic data and statistical models.⁴ Many studies, including some Ethiopian, have reported TB clustering using surveillance data.4,7 These studies relied on notified data from passive case-finding, which may underestimate the actual TB rate because of undetected cases.^{8 9} Population-based studies may help better substantiate the existence of clusters, but they are relatively few.^{4 10}

A suggested prevention and care strategy is to target the population in clusters.¹¹ Repeated population screening may help identify these clusters, facilitating the design of focused, less expensive and easily accessible interventions. In our study of pulmonary TB (PTB) prevalence and incidence, we conducted three rounds of household screening in the Dale district in Ethiopia. We discovered that two-thirds of symptomatic PTB cases were undetected at the time of the household visit.¹² This study is based on active case-finding data, complementing information from TB registries, resulting in a more comprehensive dataset on PTB. The current study included people aged 15 years and older. Children under 15 years produce fewer bacilli loads, have limited capacity to expectorate and need more diagnostics, which include chest X-rays, the introduction of gastric aspirates and stool tests using GeneXpert, and clinicians capable of making clinical decisions.¹³ These are of limited presence in the study area. Some diagnostics are inaccessible. As the study used mainly sputum for examination, enrolling children was difficult. In addition, in line with most prevalence surveys, we preferred opting for the age category of 15 years and older. Our objectives in the Dale district of Ethiopia were to (1) identify symptomatic PTB clustering through population screening and (2) explore the association between PTB clustering and demographic, socioeconomic, clinical and environmental risk factors.

Methods

Study design and setting

This study is based on data from a cohort study with repeated household screening between October 2016 and September 2017, combined with a follow-up survey to collect geographical data in August and September 2018. The target population included everyone residing in the Dale district located in the Sidama region of southern Ethiopia. According to the 2017 census, the estimated adult population aged 15 years and older in Dale was 136 181. It is worth noting that most of the population in this region resides in rural areas. Dale has 36 kebeles, which are the lowest administrative units. A kebele is divided into 383 smaller villages called enumeration areas (EAs). EA has 600 people on average. TB healthcare services are delivered based on the End TB strategy across 10 health centres, 2 clinics and 36 health posts. However, neither GeneXpert nor X-ray services are available within Dale. Access to these services requires a referral to the hospital in neighbouring Yirgalem town. In Ethiopia, public local community health providers, health extension workers (HEWs), offer cost-effective health services, which include patient referrals and regular household visits. Approximately 33% of rural residents turn to HEWs as their initial source of care when they become ill.¹⁴ We followed the Strengthening the Reporting of Observational Studies in Epidemiology cohort reporting guidelines (online supplemental file 1).¹⁵

Study population

In the three rounds of household visits between 2016 and 2017, the initial visits occurred from October 2016 to January 2017, followed by visits from February to May 2017, and then again from June to September 2017. The visits occurred every 4-month interval and lasted 1–2 months each. During these visits, trained HEWs went door-to-door, asking all family members if they had any symptoms compatible with TB. They also asked about symptoms concerning household members who were not at home during the screening and revisited households that were missing. The participation rate averaged 97%. The study calculated the coverage by dividing the number of households visited by all registered households. A total of 45 384 households were screened, sampling 136 181 adults. Among them, 3746 (2.7%) were identified as having presumptive TB. None was reported as lost to follow-up. Healthcare workers collected sputum samples at home for those who could not come to the health facility for sputum collection.

We used the term ‘presumptive TB’ for individuals having a persistent cough for 14 days or more with or without haemoptysis, weight loss, fever, night sweats, chest pain or difficulty breathing. During these visits, demographic, socioeconomic, clinical and environmental data to assess associated risk factors were collected. Those with symptoms were closely followed and asked to come to the health post to provide sputum. Sputum samples were collected and transported to the health facility for smear microscopy. GeneXpert testing was performed on smear-positive samples to validate the results. This allowed us to link the previously collected data on risk factors to the geographical data. The addresses of 17 PTB patients from 2 kebeles (Hida Kaliti and Kaliti Simita) were not found due to industrial zone construction. Further details on household screening and follow-up can be found elsewhere.^{12 16}

Operational definitions

Bacteriologically confirmed PTB was defined as any TB diagnosed based on sputum smear-microscopy, culture or GeneXpert laboratory results. Clinically diagnosed PTB was defined by a clinical decision to start TB treatment in those with a persistent cough, usually supported by radiological findings and evidence from other tests according to the national TB and leprosy guidelines.¹⁶ An individual diagnosed with bacteriologically confirmed or clinically diagnosed PTB was considered a PTB case. The most likely and secondary clusters were reported by scan statistics.¹⁷ Scan statistics SaTScan methods compare the number of observed cases to the expected under the null hypothesis. To identify areas with the most likely and secondary PTB clusters, we used a method that scans circles of different sizes, with a maximum spatial cluster size of the population at risk of 50% as an upper limit.^{18 19} We calculated the likelihood ratio to measure a relative risk (RR), and the most likely and secondary clusters were detected and reported when a p value was less than 0.05. The location with the highest likelihood ratio, showing strong evidence of clustering, is called the ‘most likely clusters’. Significant but lower likelihood differences than the most likely clusters are called ‘secondary clusters’.^{19 20} Depending on the risk in the cluster, secondary clusters are further divided into secondary clusters 1 and 2. The most likely cluster areas are the key places where the disease is transmitted most. Secondary clusters are other risk areas that might need special attention.¹⁹ To estimate distance, we calculated the straight line from the health centres to the patient’s household.

Patient and public involvement

No patients were involved in setting the research question or the outcomes, interpretation, writing or dissemination of the results.

Data collection

The geographical information was collected using a handheld global positioning system (GPS). We used GPS to measure the location and altitude, computed population density from census data of the EA and linked it to the geographic data (GPS coordinates) corresponding to EAs. We also used GPS points in the EAs to calculate the straight-line distance between each household and the nearby health centre. Locations of patients were collected by trained personnel. HEWs who were familiar with the villages provided support to the data collectors. We used principal component analysis to calculate a wealth index (online supplemental file 2). Socioeconomic covariates obtained directly from the study population were assigned wealth-index scores, which were derived from the original variables measuring household socioeconomic status. We used the wealth index to indicate household income inequality. We determined a body mass index (BMI) cut-off point using national TB guidelines (<16 kg/m² refers to severe acute malnutrition; 16–17 kg/m² refers to moderate acute malnutrition; 17–18.5 kg/m² refers to mild malnutrition and ≥18.5 kg/m² refers to normal), while the Guinea-Bissau study for the mid-upper arm circumference (MUAC) <22 cm refers to malnutrition; MUAC ≥22 cm refers to normal.^{21 22} We calculated the median values of the dataset. We used the median value when we divided our dataset into two groups for distance, population, population density and altitude. EA served as the spatial unit of analysis. We used Scan statistics to identify clusters. The cluster was the outcome of interest. We classified EAs with PTB clusters as ‘yes’ while those without as ‘no’ to facilitate further analysis using the multilevel technique.

Statistical analysis

Spatial statistics are commonly used to detect risk factors, hotspots (areas with increased risk transmission or burden) and transmission patterns.¹⁹ We performed purely spatial analysis using a Poisson model to identify the most likely and secondary PTB clusters using Kulldroff’s scan statistics with the freely available software (SaTScan V.10.1).¹⁷ The scan statistics detect and evaluate the statistical significance of temporal and/or geographical disease clusters.²³ It performs disease cluster analysis by determining the size and location, calculating RR and providing a p value. We analysed data after exploration to avoid systematic errors that could occur during data collection and entry that impact our study’s results. For spatial analysis, we used corrected shape file data linked with EAs. We used the population size for each EA projected from the 2007 census and calculated the expected incidence rate of PTB for Ethiopia. We used ArcGIS V.10.4 (ESRI 2016) to calculate centroids for the polygons and construct output maps and maps created using the UTM Zone 37 North projected coordinate system. We used case, population and coordinate as input for the purely spatial analysis. This analysis distributes the number of events in an area according to the at-risk population. Spatial clustering analysis provides the data with a multilevel structure in which individuals are nested in cluster locations, which allows for an exact estimate of the effect. A between-class coefficient of zero means no clustering effects within the data; that is, all people within the same group have identical responses to the outcome variable.^{24 25} When dealing with multilevel analyses, multiple groups were involved.²⁶ The analysis included 425 individuals with PTB in two-level regression (patient and EA levels). We evaluated random effects at the EA level by using independent covariance and examined independent factors. We used three models: in model 1, we evaluated a null model with no independent covariates. In model 2, we evaluated individual-level covariates. In model 3, we added EA-level covariates to model 2 and allowed observation days to vary across both levels. We used Jamovi (V.2.3) software for multilevel analysis.²⁷

Results

Patient characteristics and distribution

From October 2016 to September 2017, the three-round household screening identified a total of 442 people with PTB, including 90 clinically diagnosed and 352 confirmed PTB cases. In 2018, we conducted a follow-up survey in the same population, collecting geographical information on the identified patients. Among them, 425 patients were included in this report, while 17 (3.8%) PTB patients were excluded due to missing residential addresses (figure 1). Out of 201 individuals with prevalent PTB, 79 (39%) PTB were on treatment (detected by routine services), while 122 (61%) PTB were identified by screening during the first round of the survey. The median age of the patients was 28 years, with an age range of 15–90 years. The patients’ mean number of observation days was 34 days with an SD of 2.6 days. Of the patients, 52% were male, 72% were ever married and only 15% had BCG scars. Of 383 EAs in the district, 197 reported at least one PTB patient, and the geographical distribution of PTB patients exhibited variations (figure 2).

Figure 1. Study flow chart for a household investigation for tuberculosis in Dale district, Ethiopia, 2018. PTB, pulmonary tuberculosis; TB, tuberculosis.

Figure 2. Rate per 100 000 population of pulmonary tuberculosis (PTB) by enumeration area in the three rounds of household visits from October 2016 to September 2017, Dale Ethiopia.

Table 1 shows the characteristics of patients inside and outside clusters. Most variables, such as age, sex, marital status, BMI, MUAC, wealth index and BCG scar, were comparable. Differences appear in schooling (74% vs 59%), distance from health facilities (84% vs 43% in more than 3 km) and population density (73% vs 53% in less dense areas <665/km²).

Table 1. Characteristics of symptomatic pulmonary tuberculosis patients inside and outside clusters, Dale, Ethiopia 2018.

Covariate		Cluster (n, 425)
		Yes, n (%)	No, n (%)
All patients		81	344
Sex
	Men	39 (48)	180 (52)
	Women	42 (52)	164 (48)
Age in years
	15–34	60 (74)	224 (65)
	≥35	21 (26)	125 (35)
Marital status
	Single	25 (31)	93 (27)
	Ever married	56 (69)	251 (73)
Schooling
	Yes	60 (74)	204 (59)
	No	21 (26)	140 (41)
BMI kg/m2
	<16	22 (27)	99 (29)
	16–17	12 (15)	79 (23)
	17–18.5	18 (35)	80 (23)
	≥18.5	19 (23)	86 (25)
MUAC
	<22 cm	65 (80)	267 (78)
	≥22 cm	16 (20)	77 (22)
Wealth index
	Poor	35 (43)	145 (42)
	Middle	19 (24)	106 (31)
	Rich	27 (33)	93 (27)
BCG scar
	Yes	13 (16)	51 (15)
	No	68 (84)	293 (85)
Distance from the health centre*
	<3 km	13 (16)	197 (57)
	≥3 km	68 (84)	147 (43)
Population density per square km*
	<665	59 (73)	183 (53)
	≥665	22 (27)	161 (47)
Population*
	<644	57 (70)	196 (57)
	≥644	24 (30)	148 (43
Altitude (m)*
	<1793	57 (70)	178 (52)
	≥1793	24 (30)	166 (48)

^*These variables used a cut-off for groups using the median values of the data set.

BMI, body mass index; MUAC, mid-upper arm circumference.

Clustering of symptomatic PTB

In purely spatial analysis, we found that 45 out of the 383 EAs visited in each round had PTB clustering during the study period, as shown in table 2. One EA exhibits the most likely clustering (RR 11.9, p<0.0001), while 41 EAs had secondary cluster 1 (RR 2.09, p<0.001) and 3 EAs had secondary cluster 2 (RR 3.9, p<0.044). PTB clusters were common in the southeast, northwest and south-central parts of the district, as shown in figure 3A–C. These figures show PTB clustering during (a) the entire study period, (b) the first round of household screening and (c) the second and third rounds of household screening. In the first round, 39 EAs (10%) had PTB clusters, but only 3 PTB clusters (1%) were detected in two kebeles during the second and third visits between February and September 2017.

Table 2. Results of cluster analysis of patients with pulmonary tuberculosis in Dale, Ethiopia, by scan statistics, in Dale, Ethiopia, 2018.

Spatial clusters	Location	Observed cases (n)	Expected cases (n)	RR	Likelihood ratio	P value
Most likely cluster	1	17	1.48	11.9	26.3	<0.001
Secondary clusters 1	41	50	23.7	2.09	15.4	<0.001
Secondary clusters 2	3	14	3.7	3.92	8.6	0.044

n, number; RR, relative risk.

Figure 3. (A) Pulmonary tuberculosis clustering in the three rounds of household visits from October 2016 to September 2017; (B) Pulmonary tuberculosis clustering In the first round of household visits (prevalent cases) from October 2016 to January 2017; (C) Pulmonary tuberculosis clustering in the second and third rounds (incident cases) from February 2017 to September 2017.

Risk factors for PTB clustering

Table 3 shows the results of a multilevel analysis of the patients and EAs. EAs in PTB clusters were more likely far from health facilities than EAs not in PTB clusters. In our analysis, the PTB cluster was not associated with other factors examined in our regression model.

Table 3. Results of multilevel analysis: association between pulmonary tuberculosis clusters and risk factors in Dale, Ethiopia 2018.

Covariates		Effect estimates		SE	Z	P value
		Coefficient	95% CI
(Intercept)		−25.5	(−38.2, 12.6)	6.5	−3.9	<0.001
Women		−2.8	(−8.0, 2.5)	2.7	-1	0.303
Age ≥35 years		−0.5	(−5.76, 4.82)	2.7	−0.2	0.862
Ever married		−1.1	(−5.8, 3.6)	2.4	−0.5	0.649
Schooling (No)		−4.9	(−10.47, 0.76)	2.9	−1.7	0.09
BMI kg/m2		1.6	(−0.57, 3.7)	1.1	1.4	0.152
	<16	1.0
	16–17	−1	(−3.59, 1.6)	1.3	−0.8	0.453
	17–18.5	1.2	(−0.73, 3.09)	1	1.2	0.225
	≥18.5	−0.2	(−2.76, 2.34)	1.3	−0.2	0.871
MUAC in cm (>22)		−5.1	(−13.36, 3.14)	4.2	−1.2	0.225
Wealth index		−0.6	(−1.6, 0.45)	0.5	−1.1	0.275
	Poor	1
	Middle	−0.7	(−5.92, 4.54)	2.7	−0.3	0.795
	Rich	−8.4	(−16.96, 0.19)	4.4	−1.9	0.055
BCG scar (no)		7.2	(−5.75, 20.1)	6.6	1.1	0.276
Number population ≥644*		−3.8	(−9.36, 1.74)	2.8	−1.4	0.178
Distance in km from health centre ≥3*		16.2	(4.9, 27.5)	5.8	2.8	0.005
Population density per square km ≥655*		0.7	(−4.15, 5.63)	2.5	0.3	0.767
Altitude in metres ≥1793*		−3.8	(−9.69, 2.18)	3	−1.2	0.215
Number of observation days†		0.1	(−0.03, 0.05)	0	0.7	0.506

^*These variables used a cut-off above the median values.

^†Continuous variable.

BMI, body mass index; MUAC, mid-upper arm circumference.

Discussion

Based on active case-finding data, this study revealed clusters of symptomatic PTB cases in the Dale district in Ethiopia. In the southeast, northwest and south-central parts of the district, there was more PTB clustering, and the EAs with PTB clusters were farther from health centres. These areas may play an important role in transmission and suggest the need for targeted screening for early detection and treatment in areas with limited TB care. Repeated active case-finding made TB care more accessible to everyone in the district. The strategy involves several rounds of screening, which may identify missing and subclinical TB cases in rural locations and improve our ability to pinpoint clusters since this technique basically removed most clusters in subsequent visits.

The current study was initiated because of over a 14% annual decline in notifications of smear-positive TB in Dale between 2011 and 2015.²⁸ The trend raised concerns. Was the decline due to low incidents, or was it a sign of poor programme performance? This uncertainty led to a population-based study that involved a three-round household visit. The timing of the intervention was particularly noteworthy, occurring during a phase marked by a decline in notifications within the district. This unique set of data offers a valuable opportunity to measure the actual incidence of PTB as well as to identify clusters at the village level—an approach that is not applied in other contexts. Such insights are helpful for the local programme, guiding their strategic efforts to design interventions based on real-time geographical clustering crucial for implementing targeted interventions, which include case-finding and prevention. The study will add to the knowledge and shape of TB interventions using patient mapping and associated clustering.

In programme settings, the locations of cases are often linked to the diagnostic centres and not the patients’ households. Since many patients must travel to health centres for diagnosis, the notification data can, therefore, not always identify ‘real’ geographical clusters, and units with clusters may include many visitors from other areas. Hence, the study findings from previous clustering studies using notification data are often reported in urban or semi-urban areas, where access to diagnosing and treating TB is less challenging.^{29 30} Populated urban areas have both more TB services and more TB contacts and transmission than rural areas, even though infection is often within close relatives in rural areas. Therefore, clusters of PTB cases may be more prevalent in urban areas where access to care is greater while being less reported in rural areas due to limited access to care. On the contrary, TB incidence has shown an inverse association with cities with developed, accessible health services and improved living standards.²⁹

Identifying TB patients and clusters based only on passive case-finding misses undiagnosed and unnotified patients. Contrary to passive case-finding, active screening can be more costly but indicates clusters are more likely to occur in areas farther from health centres. In this study, two-thirds of PTB cases were undetected by routine programme services identified during the household screening survey. It is unclear why people with TB symptoms do not seek care, given the high proportion of patients identified by the prevalence survey. The study highlights the benefits of more active TB case findings in these areas, as many patients had not sought health services despite having symptoms.^{12 30} Such screening can help decentralise TB services, which may offer an effective means of improving access to early diagnosis, treatment and preventive therapy that helps long-term efforts to reduce TB clusters and infectious sources in the community.^{30 31} Our case identification strategy did not depend on patient care-seeking behaviour. Unlike previous studies that relied solely on passive case-finding without accounting for undetected cases, we used trained HEWs who conducted regular screening visits on various health issues in their target population. These HEWs actively involve the entire population in the process, benefiting from their trust and familiarity with community screening. Expanding on this infrastructure, our study incorporated PTB symptom screening in all households within the district. The proportion of undetected symptomatic PTB cases is expected to be low compared with the notification data from passive case-finding. While there may be a latency period between TB infection and symptom onset, repeated screening of the entire population over an extended period could help identify new cases previously asymptomatic who later become symptomatic as the disease progresses.^{12 16} We used TB unit registrations to validate PTB cases and ensured that the identified patients received care and treatment.

In the multilevel analysis, we found that EAs in cluster areas were more distant from health facilities than other EAs; low TB notification and distance from the health facility were positively correlated.⁶ Distance from TB services affects access to TB services since it is interrelated with cost and time.³² Still, in rural settings, symptomatic people receive care late and with less sensitive diagnostics.¹² TB prevention and care efforts should address patient delays and stigma affecting healthcare-seeking.³³ TB screening targeted at remote areas could detect clusters more reliably and reduce community transmission. There has been no previous research linking socioeconomic position and population clustering. In our analysis at the EA level, we sought not only to identify clustering but also risk factors for clustering that may require attention from public health practices. However, our findings did not show associations between clustering and wealth index, and a multisite study would better identify socioeconomic-based differences in PTB clustering.

Our study has some limitations. We focused on patients with TB symptoms, missing the asymptomatic. While this may introduce bias for most individuals, this is usually a temporary condition, as symptoms in most will develop later.^{31 34} Moreover, our study observed the populations for a year with repeated screening of symptoms, so the number of undetected cases and asymptomatic individuals should be low. It is important to note that misdiagnosis can occur despite the implementation of quality assurance measures, primarily due to the low sensitivity of smear microscopy. We did not include 17 PTBs with missing location information in the study, which may influence identifying real clusters.

Generally, in the TB programme, many areas may perform well, but some areas with many cases and less-performing systems may need more attention and focused strategies. Active TB screening can enhance TB programme outcomes in these areas but is costly and usually regarded as almost impossible as a routine TB prevention and care strategy in limited-resource settings, particularly with X-ray and molecular diagnostic tests, which are very expensive. However, a cheap regular HEW symptom-based screening with routine monitoring of presumptive TB patients identified and confirmed TB among referred could help identify catchment facilities with lower-than-expected presumptive TB rates, as low cluster areas may be equally important as high cluster areas.³⁵ In this regard, although GIS studies based on notifications identifying clusters cannot distinguish between real geographical cluster differences and the role of health services, it has the potential to be helpful in close contact and neighbourhood TB screening in areas where TB cases are concentrated.²⁹ It can also indicate areas with low notification despite having many undiagnosed TB cases.

Conclusions

The distribution of PTB was not uniform but concentrated in particular regions distant from health centres, indicating the need to strengthen decentralised TB services in remote settings. Routine systematic community screening may be costly, but using existing health infrastructure with HEWs through targeted screening, they can identify and refer persons with TB symptoms more quickly for diagnosis and treatment, thereby decreasing the transmission and contributing to the reduction of the TB burden.

Funders have no role in the study design, data collection, analysis, decision to publish, or preparation of the manuscript.

Data availability statement

Data are available on reasonable request.