Am. J. Trop. Med. Hyg., 93(4), 2015, pp.

733–738
doi:10.4269/ajtmh.14-0754
Copyright © 2015 by The American Society of Tropical Medicine and Hygiene

Investigating Barriers to Tuberculosis Evaluation in Uganda Using Geographic

Information Systems
Jennifer M. Ross,* Adithya Cattamanchi, Cecily R. Miller, Andrew J. Tatem, Achilles Katamba, Priscilla Haguma,
Margaret A. Handley, and J. Lucian Davis
Division of Infectious Diseases, University of Washington, Seattle, Washington; Division of Pulmonary and Critical Care Medicine,
Curry International Tuberculosis Center, San Francisco, California; Center for Vulnerable Populations, Division of General Internal Medicine,
Department of Medicine, San Francisco General Hospital, San Francisco, California; Department of Epidemiology and Biostatistics,
University of California, San Francisco, California; MU-UCSF Research Collaboration, Clinical Epidemiology Unit, Department of Medicine,
Mulago Hospital, Makerere University, Kampala, Uganda; Department of Geography and Environment, University of Southampton,
Highfield, Southampton, United Kingdom; Fogarty International Center, National Institutes of Health, Bethesda, Maryland;
Department of Epidemiology of Microbial Diseases, School of Public Health, and Pulmonary, Critical Care, and
Sleep Medicine Section, School of Medicine, Yale University, New Haven, Connecticut

Abstract. Reducing geographic barriers to tuberculosis (TB) care is a priority in high-burden countries where patients
frequently initiate, but do not complete, the multi-day TB evaluation process. Using routine cross-sectional study
from six primary-health clinics in rural Uganda from 2009 to 2012, we explored whether geographic barriers affect com-
pletion of TB evaluation among adults with unexplained chronic cough. We measured distance from home parish to
health center and calculated individual travel time using a geographic information systems technique incorporating roads,
land cover, and slope, and measured its association with completion of TB evaluation. In 264,511 patient encounters,
4,640 adults (1.8%) had sputum smear microscopy ordered; 2,783 (60%) completed TB evaluation. Median travel time
was 68 minutes for patients with TB examination ordered compared with 60 minutes without (P < 0.010). Travel time
differed between those who did and did not complete TB evaluation at only one of six clinics, whereas distance to care
did not differ at any of them. Neither distance nor travel time predicted completion of TB evaluation in rural Uganda,
although limited detail in road and village maps restricted full implementation of these mapping techniques. Better data
are needed on geographic barriers to access clinics offering TB services to improve TB diagnosis.

INTRODUCTION In addition, newer techniques can be used to estimate travel

time, which may be preferred for analyses of access to care
The Global Plan to Stop TB calls for tuberculosis (TB) because travel time varies among regions based on geography
diagnostic services to be “easily accessible, with no or mini- and mode of transportation.12–17 Therefore, in this study, we
mal financial and geographic barriers to care” to achieve the used GIS to examine the relationship between the distance to
targets of halving TB mortality and TB prevalence set for care and patient travel time and the likelihood of complet-
the 2015 Millennium Development Goals (MDG).1 Unfortu- ing TB evaluation in a multicenter study in rural Uganda.
nately, many barriers remain. TB diagnostic algorithms requir- We hypothesized that geographic barriers, including distance
ing two or more sputum examinations over 2 or more days, and travel time, would be associated with failure to complete
with additional visits to receive results and to initiate treat- TB evaluation.
ment, remain the standard of care in sub-Saharan Africa and
other high-burden regions. Furthermore, diagnostic centers in
METHODS
rural areas where the majority of the population resides are
widely dispersed. Many patients initiate evaluation but do Study design, setting, and population. We carried out a
not complete it,2,3 and as a result experience adverse out- cross-sectional study of patients undergoing TB evaluation at
comes.4,5 Although the World Health Organization (WHO) government health facilities in Uganda. We examined rou-
has recently recommended that TB diagnostic algorithms be tinely collected individual patient data on demographic char-
shortened to one day,6,7 this change has not yet been widely acteristics and diagnostic services at six primary-health centers
adopted or promoted in countries where it could have the in six rural districts for the period January 2009 to January
greatest impact, and additional data are needed on factors 2012. These six health centers were selected because they
affecting initiation of therapy to inform public health policy. are located in a variety of different geographic areas of
A number of qualitative studies have highlighted the con- the country. The Ugandan government provides routine pri-
tribution of geography and extended travel times to failures mary care, obstetric, and basic surgical services at these
to complete TB evaluation and initiate treatment,8 but there facilities without charge to patients. Each health center has
is little quantitative information in this area. Geographic infor- an on-site laboratory that performs sputum smear micros-
mation systems (GIS) methods allow investigators to create copy for acid-fast bacilli (AFB), and participates in exter-
digital representations of the landscape to enable estimation nal quality-assurance activities overseen by the National TB
of the distance that patients travel, a technique that is com- Reference Laboratory.
plementary to information obtained from patient reports The health centers are part of an infectious disease sur-
and more scalable for data collection on large populations.9–13 veillance network in which clinic staff record clinical and
demographic data, presenting symptoms, results of labora-
tory testing, and prescriptions for treatment of every clini-
*Address correspondence to Jennifer M. Ross, Division of Infec- cal encounter on a standardized, one-page data collection
tious Diseases, University of Washington, 1959 NE Pacific Street, form that is routinely entered into an electronic database,
Box 356423, Seattle, WA 98195-6423. E-mail: [email protected] as previously described.18,19 Demographic data include age,
733
734 ROSS AND OTHERS

gender, and information about residence in one of 5,238 regis- six sites with a convenience sample of patients who had been
tered parishes,20 the smallest administrative units in Uganda evaluated by a clinician and referred for sputum evaluation.
with an average population of 4,625 and a mean area of The purpose of these interviews was to facilitate a sensitivity
38 km2. In this analysis, we sampled all patients who pre- analysis exploring the validity of our GIS measures of travel
sented to health centers with cough of more than 2 weeks’ time. We interviewed participants while they were waiting
duration and underwent sputum examination as part of for sputum results, using a standardized script (see Online
a TB diagnostic evaluation. We excluded those receiv- Supplement). Specifically, we asked each participant to pro-
ing follow-up monitoring and care for a previous diagnosis vide the total time spent traveling and the mode of transpor-
of TB. tation used to travel to clinic that day.
Outcome variable. The primary outcome for the cross- We first compared GIS-estimated and patient-reported travel
sectional analysis was the completion of TB evaluation, times using Kendall’s tau test. We then described travel time
which was treated as a dichotomous variable. In accordance differences between the two techniques by plotting the mean
with the International Standards for TB Care,21 we defined of the two measures against the percentage difference on the
TB evaluation as complete if a patient provided two or more y axis, and calculating the mean differences and their 95%
sputum samples which were read and reported as negative confidence limits using the Bland–Altman method.28,29
for AFB, or if a patient provided one or more sputum sam- Finally, as an additional way of comparing the distance
ples showing AFB and initiated TB therapy. and travel time metrics with respect to the likelihood of
Explanatory variables. We extracted data on participant patients attending health centers, we constructed a density
gender, age, and date of clinic visit, and categorized the visits function for patient visits per parish. We estimated the popu-
as occurring during the rainy season (March through May lation per parish using the population map for Uganda pro-
and September through November) or during the dry season duced by WorldPop.30 We assessed the relationship between
(December through February and June through August). the total number of visits per parish with the population of
We assigned distance and travel time by matching the names each parish using negative binomial regression models. We
of the patients’ home parishes to those in a parish bound- compared a model that included distance as a covariate with
aries dataset for Uganda.22 We used the Euclidean distance one that included travel time using the Akaike information
function in ArcMap 10 (Environmental Systems Research criterion (AIC).31
Institute [ESRI], Redlands, CA) to calculate the distance in Human subjects’ protection. The Makerere University
a straight line from the centroid of the patient’s home parish School of Medicine Research Ethics Committee, the Uganda
to the nearest clinic. We calculated travel time in minutes National Council for Science and Technology, and the Uni-
from the centroid of each patient’s home parish to the nearest versity of California San Francisco Committee on Human
clinic by assembling a friction surface in ArcMap. We included Research approved the protocol. The committees waived the
data layers for elevation,23 land cover,24 roads,25 rivers,26 and requirement for informed consent for the cross-sectional study
bodies of water26 at a 90 m by 90 m cell size. We calculated on grounds that the study posed minimal risk. Participants
slope from the elevation data using the Spatial Analyst toolset interviewed at the clinics provided written informed consent
in ArcMap. We assigned a time to cross each type of land with assistance from an interpreter. We previously presented
cover, assuming mechanized transport use on roads and walk- these data in abstract form.32
ing elsewhere, ranging from 1 minute/km over road surfaces;
to 2 minutes/km in urban areas; to 24 minutes/km for areas RESULTS
with sparse shrub cover; to 36 minutes/km for croplands,
forest and jungle; to 60 minutes/km over bodies of water, Distance and travel time to clinic. There were 264,511 adult
as previously described (Supplemental Table 1).10,27 We multi- visits to the six clinics over the 3-year surveillance period.
plied ambulatory travel rates by a slope factor27 to slow the Geographic information was missing for 15,245 visits, includ-
travel speed as a function of the steepness of the terrain. We ing 12,374 for which the patient’s home parish was not
then assembled geographic raster layers into a single mosaic recorded and 2,871 for which we were unable to locate the
layer using ArcMap. We assigned the rate of travel time for reported parish on a map. This left 249,266 visits with com-
each cell according to the fastest mode of travel available for plete geographic information. Figure 1 shows a map illus-
that cell. We then calculated the path requiring the shortest trating the travel times calculated using GIS for attendees of
travel time to the nearest clinic from the center of each parish the Walukuba clinic as an example of the maps generated
using the cost-distance function in ArcMap. around each site. The median distance from home parish
Primary analyses. We compared median differences in par- to clinic for all sites was 4.2 km (Interquartile range [IQR]:
ticipant age using the Wilcoxon rank sum test, differences 1.4–9.2) for patients with a TB examination ordered (N = 4,731),
in gender and season using the χ2 test, and differences in compared with 3.2km (IQR: 1.4 – 7.2) for patients who did not
median travel time using the Wilcoxon rank sum test. We have one ordered (N = 244,535, P < 0.001). Likewise, the median
evaluated bivariate associations between candidate covariates, calculated travel time from home to clinic for all sites was
including distance and travel time and the outcome variable, 68 minutes (IQR: 29–125) for patients who had a TB examination
completing TB evaluation, and incorporated a random effect ordered, compared with 60 minutes (IQR: 24–101) for patients
to account for clustering of data within health centers. We who did not have one ordered (P < 0.001). Of the patients with a
included candidate covariates with P value < 0.1 in the multi- TB examination ordered, 3,512 (74%) lived within an estimated
variate logistic regression model. We used Stata version 11 2-hour travel radius of the clinic. Median calculated travel time
(StataCorp, College Station, TX) for all statistical analyses. from home parish varied substantially among the six clinics,
Sensitivity analyses. In addition to the review of patient from 28 minutes in a periurban area (Walukuba); to 61 minutes
encounter data, we also carried out patient interviews at all (Kihihi) and 77 minutes (Kamwezi) in mountainous southwest
 GIS AND TB EVALUATION IN UGANDA 735

TABLE 1
Bivariate associations between clinical and demographic characteris-
tics and completion of TB evaluation status
Completed TB
Characteristic evaluation Bivariate odds ratio P value

N = 4,640 N = 2,783 (95% CI) –

Gender* – 0.59 (0.53–0.67) < 0.001
Women (N = 2,414) 1,305 (54.1%) – –
Men (N = 2,225) 1,477 (66.4%) – –
Age (years) – – < 0.001
15–35 (N = 2,226) 1,198 (53.8%) Referent –
36–49 (N = 1,198) 783 (64.0%) 1.52 (1.32–1.76) –
> 50 (N = 1,190) 802 (67.4%) 1.77 (1.53–2.05) –
Season – – < 0.001
Rainy (N = 2,500) 1,412 (56.5%) Referent –
Dry (N = 2,140) 1,371 (64.1%) 1.37 (1.22–1.55) –
Clinic – – < 0.001
Kasambya (N = 618) 535 (86.6%) 4.31 (3.39–5.62) –
Kihihi (N = 670) 523 (78.1%) 2.38 (1.90–2.98) –
Aduku (N = 646) 485 (75.1%) 2.01 (1.61–2.51) –
Nagongera (N = 914) 548 (60.0%) Referent –
Walukuba (N = 954) 383 (40.1%) 0.45 (0.37–0.54) –
FIGURE 1. Map of estimated individual travel times to Walukuba
Kamwezi (N = 838) 309 (36.9%) 0.39 (0.32–0.47) –
health center. This map displays travel times for patients from the
centers of parishes in Jinja District to Walukuba Level IV Health CI = confidence interval; TB = tuberculosis.
*One value missing.
Centre as estimated using the geographic information system (GIS)
modeling techniques.

wet season (OR: 1.37, 95% CI: 1.22–1.55), this effect was
Uganda; to 86 minutes adjacent to the Nile River (Aduku); to largely driven by the remote Kamwezi clinic (OR: 7.05
64 minutes (Nagongera) and 126 minutes (Kasambya) in two 95% CI: 5.05–9.8). Patients were equally likely to com-
less populated areas (P < 0.001). plete evaluation in the dry and the rainy seasons at Kasambya,
Demographics and TB diagnoses. Clinicians ordered sputum Nagongera, and Walukuba and modestly less likely to com-
examination in 4,731 patients. Ninety-one previously diag- plete evaluation in the dry season at Aduku (OR: 0.63 95%
nosed TB patients returning for medication refills were CI: 0.44–0.92) and Kihihi (OR: 0.45 95% CI: 0.31–0.66).
excluded, leaving 4,640 patients included in our analysis Human immunodeficiency virus (HIV) status was docu-
(Figure 2). Of these 4,640, 2,783 (60.0%) completed TB eval- mented for 1,766 of 4,640 participants (38.1%). Among
uation (Table 1). As previously shown,18 women were less these, 605 of 1,766 (34.3%) participants were HIV positive.
likely to complete evaluation than men (odds ratio [OR]: HIV-positive participants were less likely to complete their
0.60, 95% confidence interval [CI]: 0.53–0.67, P < 0.001). TB analysis than HIV-negative participants, but this effect did
Older patients were more likely to complete evaluation not remain significant when accounting for clustering by clinic
(OR: 1.20 for each additional decade of life, 95% CI: 1.15– site (OR: 0.75, 95% CI: 0.44–1.29).
1.24, P < 0.001). Although overall patients were more likely Euclidean or “straight line” distance from home parish
to complete their evaluation in the dry season than the to clinic was not associated with likelihood of completing TB

FIGURE 2. Flow diagram showing patient enrollment.

736 ROSS AND OTHERS

TABLE 2 difference between patient-reported and GIS-calculated travel

Site-specific models for adjusted odds of completing TB evaluation times was proportional to the mean, such that precision
as a function of each additional hour of GIS-predicted travel time decreased with increasing travel time from the clinic. The
Adjusted 95% Confidence mean absolute difference between the measures was 30.3 min-
Clinic District odds ratio* interval P value
utes (IQR: 16.6–51.2) for travel times less than 1 hour (N = 17),
Aduku Apac 0.99 0.85–1.15 0.9
Kamwezi Kabale 1.15 0.98–1.35 0.090
59.8 minutes (IQR: 23.8–88.2) for travel times between
Kasambya Mubende 1.28 1.05–1.56 0.016 1 and 2 hours (N = 30), and 129.1 minutes (IQR: 36.8–214.7)
Kihihi Kanungu 0.91 0.77–1.07 0.2 for travel times greater than 2 hours (N = 10). Limiting our
Nagongera Tororo 1.06 0.92–1.21 0.4 analysis to those patients who traveled by mechanized trans-
Walukuba Jinja 0.91 0.71–1.17 0.5 port (N = 26) did not improve agreement between the two
GIS = geographic information system; TB = tuberculosis.
*Adjusted for age and sex.
measures (95% limits of agreement −195 to +197 minutes).
Correlation of distance and travel time with density of
patient visits. The model including distance was more pre-
evaluation. Similarly, at five of the six clinic sites, travel time dictive of visit density than was the model with estimated
from home parish to health center did not consistently differ travel time (AIC 4957 for model including distance versus
between those who completed TB evaluation and those who AIC 5275 for model including time). This indicates that the
did not (Table 2), even among those traveling for the longest travel time estimate did not improve upon physical distance
time. At the sixth clinic, Kasambya, longer travel times were as a measure of accessibility.
actually associated with a higher likelihood of completing
evaluation (OR 1.3 for each additional hour of travel, 95%
CI: 1.0–1.6, P = 0.016). Median travel time also did not differ DISCUSSION
between those who were TB positive (N = 381, 78.4 minutes) Rigorously studying how geography acts as a barrier to TB
and those who completed TB evaluation with two negative evaluation is a critical but underutilized strategy for informing
samples (N = 2,402, 78.4 minutes, P = 0.57). efforts to increase access to TB care in high-burden countries.
Comparison of patient-reported and GIS-estimated travel Accessibility maps developed using GIS represent a novel and
times. We approached 67 patients for interview at the clinic potentially useful tool to estimate the amount of time required
sites. Sixty-five (97%) agreed to the interview and 57 (85%) to reach a diagnostic center. In rural Uganda, we found that
reported travel time and were included in further analysis patients who had a TB examination ordered travel further for
(Supplemental Table 2). Twenty-six (46%) traveled to their care on average than patients who did not have a TB exami-
chosen clinics using a motorized form of transport, including nation ordered. This likely reflects the restriction of TB diag-
24 (42%) by moped and two (4%) by shared taxi (matatu); nostic services to higher level clinics like these, whereas
36 (63%) came by non-motorized transport, including 20 (35%) primary care services are available in facilities of any level.
who walked and 13 (23%) who rode a bicycle. Three (5%) However, once patients reached a clinic to initiate TB eval-
used multiple modes of transport and are included in the uation, longer distance or travel times did not change the
counts above, and one declined to disclose the method of likelihood of completing the multiday process of evaluation
travel. Although patient-reported and GIS-calculated travel and treatment initiation.
times were associated by Kendall’s tau test (P = 0.045), neither Our application of GIS in this study was limited by the
method provided consistently higher travel times. When we lack of detail in publicly available, digitized maps of Uganda.
plotted the percent difference versus the mean using a modi- Available “complete” digital road maps of Uganda omit many
fication of the Bland–Altman28,29 method, the mean differ- smaller roads and tracks, providing low resolution for model-
ence was 22.9% (95% limits of agreement −156 to +202%) ing travel even when compared with maps of neighboring
(Figure 3). In a time-stratified analysis, we found that the Kenya. In addition, we were only able to locate our patients
to the center of their home parish because the majority of
their villages could not be found on digitized maps or gazet-
teers with geolocations.33 This meant that travel times were
calculated from the center of parishes, which often may not
correspond to an area on the road network, and this may
have led to unrealistically high travel times being calcu-
lated in many cases. Improving the detail of digital spatial
datasets of this region will increase the precision of the
technique used here. In the absence of detailed maps, an
alternative method would extend the use of mobile phone
technology to allow patients to transmit the coordinates of
their homes and/or their paths of travel rather than relying
on maps of villages. Mobile phones have already been used
in TB programs as platforms for data collection and to com-
municate with patients.34–36
Although neither travel time nor distance was associated
with likelihood of completing TB evaluation at these primary
FIGURE 3. Bland–Altman plot comparing the differences in geo- health clinics in Uganda, other studies have found that
graphic information systems (GIS)-calculated and patient-reported increased travel time as measured by self-report is associated
travel times as a proportion of the mean. with delays both in seeking and delivering TB care.8 In the
 GIS AND TB EVALUATION IN UGANDA 737

absence of more detailed individual covariates in our study, 40% of patients initiating TB evaluation at government
we propose several possible hypotheses for testing in future health centers in rural Uganda did not complete their evalu-
studies. One hypothesis is that patients traveling for a longer ation. In addition, large-scale application of the mapping
time period to access care may be wealthier or more moti- technique used in our study is currently limited by lack of
vated to complete evaluation than those traveling for a detail in publicly available road and village spatial datasets.
shorter period. Another hypothesis is that patients who have
to travel for longer to reach clinic may delay seeking care, Received November 26, 2014. Accepted for publication May 23, 2015.
but may be more likely to complete evaluation once they Published online July 27, 2015.
reach the clinic because traveling back is relatively more
Note: Supplemental survey and tables appear at www.ajtmh.org.
costly and time consuming.
This study makes several new contributions to understand- Acknowledgments: We thank the patients and staff at the facilities
participating in the Uganda TB Surveillance Project as well as the
ing the geographic accessibility of TB evaluation in Uganda. staff and administration of the Uganda Malaria Surveillance Project
Although GIS is increasingly used to assess the accessibility and the MU-UCSF Research Collaboration. We also acknowledge
of health services on the provincial10,37,38 or national12,13,39,40 Kevin Koy with the UC Berkeley Geospatial Innovations Facility (GIF).
level in sub-Saharan Africa, this is among the first analyses Funding supports: Jennifer M. Ross acknowledges funding support
in this region to use GIS to investigate travel time for TB from NIH/NIAID (T32AI007140). Andrew J. Tate acknowledges fund-
evaluation at a local level. This study also used a gridded ing support from the RAPIDD program of the Science and Technology
travel-time surface that permitted travel off of the road net- Directorate, Department of Homeland Security, and the Fogarty Inter-
national Center, National Institutes of Health, and is also supported by
work, in contrast to many accessibility studies done in devel- grants from NIH/NIAID (U19AI089674) and the Bill and Melinda
oped countries that calculate travel over the road network.41 Gates Foundation (#49446 and #1032350). J. Lucian Davis acknowl-
This technique could be well suited to Uganda given our edges support from NIH/NIAID (K23AI080147). Adithya Cattamanchi
finding that half of patients traveled to clinic without using acknowledges support from NIH/NHLBI (K23HL094141).
a mechanized form of transport, but would be improved with Authors’ addresses: Jennifer M. Ross, Division of Infectious Diseases,
more precise detail about the patient locations. In our study, University of Washington, Seattle, WA, E-mail: [email protected]. Adithya
where the median distance from home parish to clinic was Cattamanchi and Cecily R. Miller, Division of Pulmonary and Critical
Care Medicine, University of California, San Francisco, CA, E-mails:
less than 4 km, our inability to locate patients more precisely [email protected] and cecily.miller@ucsf. J. Lucian Davis,
than the parish center likely reduced our power to confirm Department of Epidemiology of Microbial Diseases, School of Public
or refute any association. Health, and Pulmonary, Critical Care, and Sleep Medicine Section,
A limitation of our study is that GIS-predicted travel times School of Medicine, Yale University, New Haven, CT, E-mail: lucian
did not correlate with those reported by patients, which may [email protected]. Andrew J. Tatem, Department of Geography and Envi-
ronment, University of Southampton, Southampton, United Kingdom,
be due to the limitations in the travel time estimate discussed E-mail: [email protected]. Achilles Katamba, College of Health
above. Socioeconomic status may be an unmeasured con- Sciences, Makerere University, Kampala, Uganda, E-mail: akatamba@
founder because patients from different regions may have yahoo.com. Priscilla Haguma, MU-UCSF Research Collaboration,
different levels of wealth, which may affect their access to Kampala, Uganda, E-mail: [email protected]. Margaret A.
Handley, Department of Epidemiology and Biostatistics, UCSF School of
transportation or their ability to take the time to seek care.
Medicine, San Francisco, CA, E-mail: [email protected].
An alternative explanation could be that patients traveled
to clinic by multiple methods that entailed different transfer
REFERENCES
times not accounted for in our analysis. Alternatively, patients
may not report travel times accurately, as shown in another 1. WHO, 2011. The Global Plan to Stop TB 2011–2015: Trans-
study from a rural region of Uganda where travel time forming the Fight Towards Elimination of Tuberculosis. Geneva,
reported by patients attending an HIV clinic differed from the Switzerland: WHO.
2. Kemp J, Squire S, Nyirenda I, Salaniponi F, 1996. Is tuberculosis
measured time of travel.42 It is unknown whether these dif- diagnosis a barrier to care? Trans R Soc Trop Med Hyg 90: 472.
ferences would persist if they were evaluated with more pre- 3. Botha E, den Boon S, Lawrence KA, Reuter H, Verver S,
cise residential localization. Although we assigned travel Lombard CJ, Dye C, Enarson DA, Beyers N, 2008. From
speeds to various land surfaces based on values used in other suspect to patient: tuberculosis diagnosis and treatment initi-
studies,10,27 they may not capture the actual travel times expe- ation in health facilities in South Africa. Int J Tuberc Lung
Dis 12: 936–941.
rienced by our patients. Attempts such as these to validate 4. Squire SB, Belaye AK, Kashoti A, Salaniponi FM, Mundy CJ,
GIS estimates of travel time are important because of the Theobald S, Kemp J, 2005. ‘Lost’ smear-positive pulmonary
goals set by health ministries to locate services within speci- tuberculosis cases: where are they and why did we lose them?
fied limits of travel time for their populations.10,13 Int J Tuberc Lung Dis 9: 25–31.
5. Botha E, Den Boon S, Verver S, Dunbar R, Lawrence KA,
A major focus of recent WHO TB diagnostic policy changes Bosman M, Enarson DA, Toms I, Beyers N, 2008. Initial
is on reducing diagnostic delays.6,43,44 As these and future default from tuberculosis treatment: how often does it happen
novel diagnostic strategies are implemented, it is important to and what are the reasons? Int J Tuberc Lung Dis 12: 820–823.
assess their impact on multiple levels, including equity ana- 6. World Health Organization, 2011. Same-day Diagnosis of Tuber-
lyses of sub-populations by gender, socioeconomic status, and, culosis by Microscopy. Geneva, Switzerland: WHO.
7. Davis JL, Cattamanchi A, Cuevas LE, Hopewell PC, Steingart
as we argue here, geography.45,46 Thus, future diagnostic imple- KR, 2013. Diagnostic accuracy of same-day microscopy versus
mentation studies might consider using GIS estimations of standard microscopy for pulmonary tuberculosis: a systematic
travel time to assess access to new diagnostics. Improved review and meta-analysis. Lancet Infect Dis 13: 147–154.
availability of detailed spatial datasets from governments and 8. Finnie RK, Khoza LB, van den Borne B, Mabunda T, Abotchie
P, Mullen PD, 2011. Factors associated with patient and health
researchers would greatly accelerate progress in this field. care system delay in diagnosis and treatment for TB in sub-
In conclusion, we found that distance to care and patient Saharan African countries with high burdens of TB and HIV.
travel time, as estimated by GIS, did not explain why nearly Trop Med Int Health 16: 394–411.
738 ROSS AND OTHERS

9. Houben RM, Van Boeckel TP, Mwinuka V, Mzumara P, Branson 29. Dewitte K, Fierens C, Stockl D, Thienpont LM, 2002. Applica-
K, Linard C, Chimbwandira F, French N, Glynn JR, Crampin tion of the Bland–Altman plot for interpretation of method-
AC, 2012. Monitoring the impact of decentralised chronic care comparison studies: a critical investigation of its practice. Clin
services on patient travel time in rural Africa–methods and Chem 48: 799–801.
results in northern Malawi. Int J Health Geogr 11: 49. 30. WorldPop, 2013. Uganda Population Map - 2010. Available at:
10. Huerta Munoz U, Kallestal C, 2012. Geographical accessibility http://www.worldpop.org. Accessed March 11, 2015.
and spatial coverage modeling of the primary health care net- 31. Akaike H, 1974. A new look at the statistical model identifica-
work in the Western Province of Rwanda. Int J Health Geogr tion. IEEE Trans Automat Contr 19: 716–723.
11: 40. 32. Ross J, Cattamanchi A, Miller C, Katamba A, Davis J, 2012.
11. Sudhof L, Amoroso C, Barebwanuwe P, Munyaneza F, Using geographic analysis to investigate barriers to TB evalu-
Karamaga A, Zambotti G, Drobac P, Hirschhorn LR, 2013. ation in Uganda. American Thoracic Society International
Local use of Geographic Information Systems to improve data Conference. New York.
utilisation and health services: mapping caesarean section 33. National Geospatial-Intelligence Agency. NGA GEOnet Names
coverage in rural Rwanda. Trop Med Int Health 18: 18–26. Server. Available at: http://geonames.nga.mil/gns/html/namefiles
12. Alegana VA, Wright JA, Pentrina U, Noor AM, Snow RW, .html. Accessed December 2011.
Atkinson PM, 2012. Spatial modelling of healthcare utili- 34. Khan AJ, Khowaja S, Khan FS, Qazi F, Lotia I, Habib A,
sation for treatment of fever in Namibia. Int J Health Geogr Mohammed S, Khan U, Amanullah F, Hussain H, Becerra
11: 6. MC, Creswell J, Keshavjee S, 2012. Engaging the private sector
13. Noor AM, Amin AA, Gething PW, Atkinson PM, Hay SI, Snow to increase tuberculosis case detection: an impact evaluation
RW, 2006. Modelling distances travelled to government health study. Lancet Infect Dis 12: 608–616.
services in Kenya. Trop Med Int Health 11: 188–196. 35. Kunawararak P, Pongpanich S, Chantawong S, Pokaew P,
14. Gabrysch S, Cousens S, Cox J, Campbell OM, 2011. The influ- Traisathit P, Srithanaviboonchai K, Plipat T, 2011. Tubercu-
ence of distance and level of care on delivery place in rural losis treatment with mobile-phone medication reminders in
Zambia: a study of linked national data in a geographic northern Thailand. Southeast Asian J Trop Med Public Health
information system. PLoS Med 8: e1000394. 42: 1444–1451.
15. Guagliardo MF, 2004. Spatial accessibility of primary care: con- 36. Fraser HS, Habib A, Goodrich M, Thomas D, Blaya JA, Fils-
cepts, methods and challenges. Int J Health Geogr 3: 3. Aime JR, Jazayeri D, Seaton M, Khan AJ, Choi SS, Kerrison
16. Blanford JI, Kumar S, Luo W, MacEachren AM, 2012. It’s a F, Falzon D, Becerra M, 2013. E-health systems for manage-
long, long walk: accessibility to hospitals, maternity and ment of MDR-TB in resource-poor environments: a decade of
integrated health centers in Niger. Int J Health Geogr 11: 24. experience and recommendations for future work. Stud Health
17. World Health Organization, 2001. Background Paper for the Tech- Technol Inform 192: 627–631.
nical Consultation on Effective Coverage of Health Systems. 37. Houben RM, Van Boeckel TP, Mwinuka V, Mzumara P, Branson
Geneva, Switzerland: WHO. K, Linard C, Chimbwandira F, French N, Glynn JR, Crampin
18. Miller CR, Davis JL, Katamba A, Sserwanga A, Kakeeto S, AC, 2012. Monitoring the impact of decentralised chronic
Kizito F, Cattamanchi A, 2013. Sex disparities in tuberculosis care services on patient travel time in rural Africa–methods
suspect evaluation: a cross-sectional analysis in rural Uganda. and results in Northern Malawi. Int J Health Geogr 11: 49.
Int J Tuberc Lung Dis 17: 480–485. 38. Feikin DR, Nguyen LM, Adazu K, Ombok M, Audi A, Slutsker
19. Davis J, Katamba A, Vasquez J, Crawford E, Sserwanga A, L, Lindblade KA, 2009. The impact of distance of residence
Kakeeto S, Kizito F, Dorsey G, den Boon S, Vittinghoff E, from a peripheral health facility on pediatric health utilisation
Huang L, Adatu F, Kamya MR, Hopewell PC, Cattamanchi in rural western Kenya. Trop Med Int Health 14: 54–61.
A, 2011. Evaluating tuberculosis case detection via real-time 39. Heard NJ, Larsen U, Hozumi D, 2004. Investigating access to
monitoring of tuberculosis diagnostic services. Am J Respir reproductive health services using GIS: proximity to services
Crit Care Med 184: 362–367. and the use of modern contraceptives in Malawi. Afr J Reprod
20. Uganda Bureau of Statistics, 2006. The 2002 Uganda Popula- Health 8: 164–179.
tion and Housing Census, Population Size and Distribution. 40. Barnighausen T, Wallrauch C, Welte A, McWalter TA, Mbizana
Kampala, Uganda: Uganda Bureau of Statistics. N, Viljoen J, Graham N, Tanser F, Puren A, Newell ML,
21. Hopewell PC, Pai M, Maher D, Uplekar M, Raviglione MC, 2008. HIV incidence in rural South Africa: comparison of
2006. International standards for tuberculosis care. Lancet Infect estimates from longitudinal surveillance and cross-sectional
Dis 6: 710–725. cBED assay testing. PLoS One 3: e3640.
22. Global Administrative Areas, 2012. GADM Database of Global 41. Pedigo AS, Odoi A, 2010. Investigation of disparities in geo-
Administrative Areas. Available at: www.gadm.org. Accessed graphic accessibility to emergency stroke and myocardial
December 12, 2011. infarction care in east tennessee using geographic information
23. Shuttle Radar Topography Mission (SRTM), 2011. U.S. Releases systems and network analysis. Ann Epidemiol 20: 924–930.
Enchanced Shuttle Land Elevation Data. Available at: http:// 42. Siedner MJ, Lankowski A, Tsai AC, Muzoora C, Martin JN,
www2.jpl.nasa.gov/srtm/. Accessed December 12, 2011. Hunt PW, Haberer JE, Bangsberg DR, 2013. GPS-measured
24. Di Gregorio A. Multipurpose Landcover Database of Uganda - distance to clinic, but not self-reported transportation factors,
AFRICOVER. United Nations Food and Agriculture Organiza- are associated with missed HIV clinic visits in rural Uganda.
tion, 2003. Available at: http://www.glcn.org/index_en.jsp. Accessed AIDS 27: 1503–1508.
December 12, 2011. 43. World Health Organization, 2011. Automated Real-Time Nucleic
25. Center for International Earth Science Information Network - Acid Amplification Technology for Rapid and Simultaneous
CIESIN - Columbia University, and Information Technology Detection of Tuberculosis and Rifampicin Resistance: Xpert
Outreach Services - ITOS - University of Georgia, 2013. Global MTB/RIF System. Geneva, Switzerland: WHO.
Roads Open Access Data Set, Version 1 (gROADSv1). Palisades, 44. World Health Organization, 2006. Improving the Diagnosis and
NY: NASA Socioeconomic Data and Applications Center Treatment of Smear-Negative Pulmonary and Extrapulonary
(SEDAC). Available at: http://dx.doi.org/10.7927/H4VD6WCT. Tuberculosis among Adults and Adolescents. Recommendations
Accessed December 12, 2011. for HIV-Prevalent and Resource-Constrained Settings. Geneva,
26. National Imagery and Mapping Agency (NIMA), 2000. Vector Switzerland: WHO.
Map Level 0 (VMAP0). Available at: http://earth-info.nga.mil/ 45. Mann G, Squire SB, Bissell K, Eliseev P, Du Toit E, Hesseling
publications/vmap0.html. Accessed December 12, 2011. A, Nicol M, Detjen A, Kritski A, 2010. Beyond accuracy: cre-
27. Nelson A, 2008. Estimated Travel Time to the Nearest City of ating a comprehensive evidence base for TB diagnostic tools.
50,000 or More People in Year 2000. Global Environment Moni- Int J Tuberc Lung Dis 14: 1518–1524.
toring Unit - Joint Research Centre of the European Commis- 46. Squire SB, Ramsay AR, van den Hof S, Millington KA, Langley I,
sion. Available at: http://forobs.jrc.ec.europa.eu/products/gam/. Bello G, Kritski A, Detjen A, Thomson R, Cobelens F, Mann
28. Bland JM, Altman DG, 1996. Measurement error proportional GH, 2011. Making innovations accessible to the poor through
to the mean. BMJ 313: 106. implementation research. Int J Tuberc Lung Dis 15: 862–870.