potential versus revealed access to care during a dengue
TRANSCRIPT
1
Potential versus revealed access to care during a dengue fever outbreak 1
2
Irene Casas, PhD12 3
Eric Delmelle, PhD3 4
Elizabeth C. Delmelle, PhD4 5
6
Working paper. Manuscript accepted in July 2016 for Transport and Health 7
8
Abstract 9 Dengue fever is a vector-borne disease which spreads quickly under suitable conditions and puts 10
certain segments of the population at higher risk, especially in developing countries. Prompt 11
diagnosis of the disease is critical to (1) substantially reduce risks of morbidity and mortality and 12
(2) prevent further expansion of an existing outbreak. Suitable geographic access to medical care 13
and effective prevention campaigns can help achieve these two objectives. This paper examines 14
patterns of health care use based on travel time during an outbreak of dengue fever in the city of 15
Cali, Colombia. We map patterns of diagnosis by facility and identify associated travel 16
disparities. Results indicate that travel times exhibited significant spatial autocorrelation, 17
showing that patients living in the periphery of the city experienced substantially greater travel 18
times compared with patients residing in the north-south corridor of the city, owing to the 19
proximity of health care centers. Revealed travel times were nearly six times longer than 20
estimated travel to the closest hospital, suggesting that the healthcare center where care was 21
received rarely coincided with the closest option (only in 8% of cases). A multilevel model, 22
accounting for both individual and neighborhood characteristics, was developed to explain 23
variation in travel estimates. Results revealed that males and patients from affluent 24
neighborhoods were more likely to travel longer and beyond their closest facility to receive care. 25
These findings are important for policy related decisions such as (1) timely allocation and 26
training of health personnel and (2) organizing informative campaigns aiming at increasing the 27
awareness about the disease, especially for the population that is at high risk and/or most 28
vulnerable. 29
30
Keywords: dengue fever, revealed and potential accessibility, travel times, multilevel modeling. 31
1
School of History and Social Sciences, Louisiana Tech University GTM 137, Railroad Av. Ruston, LA, 71270, USA
Corresponding author 3
Department of Geography and Earth Sciences and Center for Applied Geographic Information Science,
University of North Carolina at Charlotte 9201 University City Blvd, Charlotte, NC, 28223, USA
Department of Geography and Earth Sciences, University of North Carolina at Charlotte 9201 University City Blvd, Charlotte, NC, 28223, USA [email protected]
2
1 Introduction 32
Dengue fever is an arboviral disease with a presence in more than one hundred tropical and 33
subtropical countries (Kittayapong et al. 2008). The disease has gradually re-emerged across the 34
global South, particularly affecting urban areas of the tropics and sub-tropics. In 2015, 2.35 35
million cases were reported in the Americas alone where severe cases resulted in 1181 deaths. In 36
Colombia, the fatality rate for the year 2010 was 2.28% which was above the 2% threshold 37
imposed for the virus (Rojas 2011). In that year, there were 146,354 dengue cases, of which 5420 38
were categorized as severe, resulting in a total of 174 deaths (Salud 2016). 39
The dengue virus (DEN) comprises of four serotypes (DENV-1, DENV-2, DENV-3 and 40
DENV-4) and is transmitted by the Aedes Aegypti mosquito. The virus passes from one 41
individual to another through the bites of the female Aedes mosquito and is acquired by the 42
mosquito while feeding on an infected person (WHO 2016). After approximately 10 days, the 43
virus spreads to the salivary glands of the mosquito and the virus can be transmitted to other 44
humans. Once a human is infected, and after an incubation period of 4 to10 days, symptoms -45
which include fever, joint and back pain, severe headache, and nausea - will last for 2 to7 days. 46
Once infected, humans become the main carriers and multipliers of the virus, serving as a source 47
for uninfected mosquitoes. 48
The female Aedes Aegypti typically spend their lifetime in or around the houses where they 49
emerge as adults. The transmission of the virus is thus focal (Kuno 1995, Morrison et al. 1998, 50
Getis et al. 2003). Female Aedes Aegypti usually fly no more than 400 meters. This means that 51
people, rather than mosquitoes, rapidly move the virus within and between communities and 52
places. The virus has the potential to develop into a life-threatening version identified as severe 53
dengue fever. There is no vaccine for dengue, but early detection and proper access to health care 54
lowers fatality rates below 1% (WHO 2016). Treatment includes rehydration and may require 55
blood transfusions for severe cases (Kolivras 2006). Timely diagnosis prevents existing 56
outbreaks from expanding. 57
Several studies have used Geographic Information Systems (GIS) to document the temporal 58
and spatial variation of the Aedes mosquito (Morrison et al. 2004, Koenraadt et al. 2008) and to 59
identify clusters of dengue fever outbreaks (Mammen et al. 2008, Siqueira-Junior et al. 2008, 60
Vazquez-Prokopec et al. 2010, Aldstadt et al. 2012, Hsueh et al. 2012, Delmelle et al. 2013a). 61
However, no study has examined how individuals access health services during such outbreaks. 62
Travel impedance may impact the timely receipt of health services, which is particularly critical 63
since delays in diagnosis are likely to impact the outcome of the disease (Noor et al. 2003). 64
In this paper, we examine revealed travel patterns of individuals diagnosed with dengue fever 65
during the 2010 outbreak in Cali, a rapidly expanding urban environment of Colombia, Latin 66
America. For each diagnosis, we estimate travel time from the geocoded residential address of 67
each patient to the healthcare facility where care was received. We compare actual travel time to 68
the time necessary to reach the closest healthcare facility where care could have been received, 69
or potential accessibility. Second, we develop a multilevel model, accounting for both individual 70
and neighborhood characteristics to explain variation in travel estimates. The study therefore 71
3
identifies potential disparities in access to healtcare, but also highlights the increased choice of 72
facilities available to those with greater resources to travel. 73
Our approach is descriptive and helps to (1) determine the adequacy and equity of health 74
services, (2) identify areas characterized by an imbalance between revealed and potential 75
accessibility and (3) determine whether patients who died from complications were at a 76
disadvantage in accessing a health facility. Answers to these issues will inform health authorities 77
on strategies to better plan preventive and control actions against dengue fever. 78
2 Background 79
Access is a fundamental measure of health care quality. This notion has been categorized into 80
five different dimensions including: availability, affordability, acceptability, appropriateness, and 81
accessibility (Penchansky and Thomas 1981). The focus of this study is on spatial accessibility, 82
or the ability to reach a particular service from a given location (Luo and Whippo 2012, 83
Levesque et al. 2013). Factors that may impact spatial accessibility include the supply of health 84
care and the distribution of patients, as well as geographic impedance which is linked to the 85
transport system (Delamater et al. 2012). 86
Various metrics have been developed to measure spatial accessibility ranging from the 87
simpler distance or travel time to the nearest service (Rosero-Bixby 2004, Haynes et al. 2006, 88
Tanser et al. 2006, Delamater et al. 2012, McGrail 2012, Leese et al. 2013), to metrics based on 89
the notion of gravity, or the idea that larger and closer facilities are more attractive (Delmelle and 90
Casas 2012, McGrail 2012, Hu et al. 2013, Wang and Tang 2013), and population-to-provider 91
ratios (Schonfeld et al. 1972, Connor et al. 1995, Phillips and Lindbloom 2000). Improvements 92
to these methods have also been proposed including inclusion of distance-decay functions and 93
the use of variable catchment sizes (McGrail and Humphreys 2009, Dony et al. 2015). As 94
expected, a large number of studies in access to health have been conducted in the developed 95
world (Arcury et al. 2005, Langford and Higgs 2006, Delamater et al. 2012, Wong et al. 2012). 96
However, recently there have been more conducted in the developing world (Rosero-Bixby 97
2004, Tanser et al. 2006, Friedman et al. 2013, Hu et al. 2013). 98
Spatial accessibility can be distinguished between potential and revealed accessibility; the 99
former is based on the possibility for health care use given the services offered, while the latter is 100
based on the actual utilization of the services and the act of receiving health care (Guagliardo 101
2004). The presence of geographical barriers affects access to service and may further result in 102
unbalanced –or inaccessible- supply (Joseph and Phillip 1984, Haynes 1986, Jones and Bentham 103
1997, Perry and Gesler 2000). The overwhelming majority of studies in the literature examine 104
potential accessibility (Luo and Whippo 2012, McGrail 2012, Hu et al. 2013, Kaljee et al. 2013, 105
Wang and Tang 2013); while those on revealed accessibility less common given that they require 106
revealed preference or utilization data sets which are expensive and time consuming to acquire 107
(Parker and Campbell 1998, Perry and Gesler 2000, Brabyn and Skelly 2002, Buor 2003, Haynes 108
et al. 2003, Rosero-Bixby 2004, Haynes et al. 2006, Tanser et al. 2006, Delamater et al. 2012). 109
This literature has generally recognized that the use of health services varies according to the 110
4
patient’s age, gender and income as well as with the separating distance between the patient and 111
the place where care is received. An individual may elect to travel further than the closest facility 112
when specific medical procedures are required (Buchmueller et al. 2006). 113
In developing countries, access to health care is uneven (Fosu 1989, Delmelle and Casas 114
2012) as the planning of the distribution of health services often does not consider socio-medical 115
priorities or how services are used (Fosu 1989). This makes measuring access to health care in 116
the developing world difficult (Bahr and Wehrhahn 1993), not only due to the populations’ 117
characteristics (remoteness and lack of infrastructure to reach rural communities), but also the 118
costs involved in the process. However, countries in Latin America, for example, have made 119
substantial and consistent investments in research for health (Council on Healh Research for 120
Development 2006). 121
In building upon work previously conducted by Casas et al. (2010), which illustrated how a 122
particular hospital in the city of Cali served a population well beyond its intended service area, 123
this paper presents results of an empirical study on access to health care by patients diagnosed 124
with dengue fever. The focus is on both potential and revealed access to healthcare facilities. 125
Factors influencing the choice of healthcare facility are examined at the individual and 126
neighborhood level using multilevel analysis. This paper contributes to the literature by 127
examining a disease of an endemic nature that constitutes a health threat in most of the countries 128
where it is present. It also contributes by presenting a revealed accessibility travel scenario that 129
helps in understanding individual choice and identifying estimation differences in travel times 130
when compared with potential destinations. 131
3 Data and Study Area 132
3.1 Study Area 133
Cali is the third largest city in Colombia with approximately 2.3 million people according to 134
the 2013 population census (Cali 2014). The city is characterized by a tropical climate with two 135
rainy seasons, its yearly precipitation totals 100mm while an average temperature of 26°C 136
provides a thriving environment for the reproduction of the Aedes Aegypti mosquito. The city of 137
Cali is administratively divided into twenty-two Communes (see Figure 1a), representing 138
groupings of neighborhoods exhibiting homogeneous demographic and socioeconomic 139
characteristics. Neighborhoods are classified into six classes based on socioeconomic strata, six 140
being the highest and one the lowest (see Figure 1b). The stratification process is based on two 141
aspects: (1) the external physical characteristics of the housing stock and (2) its immediate 142
surroundings and urban context. Variables are observed under each aspect including, but not 143
limited to things such as: front yard, garage, door material, façade material, access road, 144
sidewalks, poverty, urban deterioration, and industry (DANE 2016). The stratification reflects 145
the urbanization process of the city from the old colonial core centered in communes three and 146
nine, suggesting an expansion pattern along the spinal corridor to the north and south and 147
subsequently on a radial pattern towards the periphery. Peripheral areas are characterized by 148
5
unplanned urbanization (Restrepo 2006), resulting in various squatter settlements with poor 149
infrastructure. 150
151
152
Figure 1: (a) Administrative units in the city of Cali and (b) socio-economic strata at the neighborhood level. 153 154
3.2 Dengue Fever Data 155
The city of Cali has been an endemic region for dengue fever since the reemergence of the 156
virus in the 1970s. Outbreaks were reported in 1995, 2002, 2005, 2010 (Cali 2010b) and more 157
recently in 2013, with 2010 being the second worst in the last 25 years. Positively diagnosed 158
cases of dengue fever are reported to the health municipality and recorded in the “Sistema de 159
Vigilancia en Salud Pública” (SIVIGILA, English: Public health surveillance system). Individual 160
information includes: patient’s data (sex, age, race and patient’s occupation), date of diagnosis, 161
epidemiological week, day symptoms started, whether the patient was hospitalized, their final 162
condition (dead or alive), and apparent symptoms. The database also includes the reporting 163
institution (i.e., the health facility where the patient was diagnosed). Data used here corresponds 164
to the 2010 outbreak and from SIVIGILA patient data (sex and age) and the reporting institution 165
are included in the models. After standardizing the data to a format that would allow address 166
6
matching, residential addresses of patients were geocoded at the street intersection level. This is 167
done to mask the data and maintain the privacy of the patients (Kwan et al. 2004). A total of 168
8,668 cases are included in the data set (see Figure 2a) for 2010 with clusters of patients in 169
various neighborhoods throughout the city (see Figure 2b). 170
171
172
Figure 2: (a) Spatial distribution of dengue fever patients reported in 2010, (b) kernel density estimation, and (c) 173 locations of hospitals utilized at least once by patients. 174
3.3 Healthcare Centers 175
In the city of Cali, medical services are provided by private and public (state owned) 176
facilities (Cali 2010a). Healthcare facilities are further categorized based on their level of service 177
(I: basic services with no beds, II: specialized services with no beds, level III: teaching university 178
hospital with beds; level IV has recently been added to this list to refer to a particular private 179
health facility with state of the art technology and infrastructure, the Fundación Clinica Valle del 180
Lili in Figure 2c). Health posts are the simplest of all the healthcare facilities, followed by 181
healthcare centers, hospital centers, and hospitals (or clinics) and other private and cooperative 182
institutions providing more complex services. Cooperatives are a special kind of facility where a 183
group of people come together to provide themselves associatively with services, therefore 184
individuals are owners and members of the facility (Table 1). 185
186
187
188
7
Institution Number of Units Institution Number of Units
Health Posts 66 Comfandi (cooperative) 13 Healthcare Centers 34 Comfenalco (cooperative) 5 Hospital Centers 5 Police Clinics 1 Hospitals – Level II 5 Private Clinics 22 Hospitals- Level III 2 Private Health Promoters 82 Hospitals- Level IV 1
Table 1: Types of healthcare facilities in the city of Cali, 2010 189 190
The healthcare facilities where a patient was diagnosed are geocoded to the intersection level. 191
In 2010, the patients who were diagnosed with dengue fever visited 167 different healthcare 192
facilities (Figure 2c). However, due to incomplete data filed at the municipality, healthcare 193
facilities reported by some patients could not be linked to the health facility database resulting in 194
the removal of those cases (938 cases removed from 9606). 195
4 Methods 196
Potential and revealed accessibility are calculated for each geocoded patient. Potential 197
accessibility is estimated based on the one-way travel time by private car from each patient’s 198
residential address to their closest health facility, while revealed accessibility is estimated using 199
the travel time to the health facility where the patient was diagnosed with dengue virus. Private 200
car ownership for the year 2010 was reported based on the number of cars per 100000 people as 201
15418.9 cars (2244536 individuals/ 346082.76 cars). Motorcycle ownership was 4410 202
motorcycles per 100000 people (total of 98984 motorcycles for the city). The number of trips by 203
public transport was not reported for 2010 because there was a complete change in the public 204
transit system with a move to a BRT system. However, in 2008 the number of trips per 100000 205
people by public transport was 1862.4. Travel time is used as a proxy for travel cost (Martin et 206
al. 2002, Apparicio et al. 2008, Shahid et al. 2009). Similar to Delmelle et al. (2013b), travel 207
time is estimated in a commercial GIS environment, using the Dijkstra shortest-path algorithm. 208
A road network to estimate travel impedance for the city of Cali is used, assuming a 40km/h 209
travel speed. Excess travel is estimated as the difference in travel time between the closest 210
hospital (potential) and the hospital where diagnosis was confirmed (revealed). Higher values of 211
excess travel are indicative of patients incurring longer travel than if they had received care at the 212
closest facility. 213
To explain variations in travel time among patients and to determine what factors might 214
influence travel time, a multi-level modeling framework is utilized. This approach considers the 215
nested structure of the data: patients are clustered within urban neighborhoods. Such clustering 216
would violate the assumption of data independence in an ordinary least squares regression 217
model. In this mixed model, two levels of independent variables are selected: individual and 218
neighborhood. Unfortunately, patient data limitations restrict the availability of many individual 219
level factors, however, age and gender of the patient is accounted for. Given the potential non-220
linear relationship between age and travel time (i.e., both older and younger adults potentially 221
8
have different travel behaviors than middle-aged patients), age is categorized into the following 222
groups: (1) Less than 10 years old, (2), 10-20 years old, (3) 20-50 years old, and (4) greater than 223
50. At the neighborhood level, the socioeconomic strata of the neighborhood is included as well 224
as the average travel time from a neighborhood’s centroid to all city hospitals. This latter 225
variable serves as a proxy for urban centrality while the former serves as a proxy for income. 226
For instance, for individuals residing in neighborhoods along the periphery who choose not to 227
visit their closest hospital, the next closest will be much further away than for those located 228
toward the urban core that have a greater number of hospitals to choose from within a closer 229
proximity. 230
5 Results and Discussion 231
The sample is composed of 8,668 patients, with a similar number of males and females, 232
however nearly 50% of the patients are younger than 20 years old. Studies have previously 233
shown that children are at a higher risk of contracting dengue fever than other age cohorts (WHO 234
2009, Martín et al. 2010). From the sample, 8,049 had the classic manifestation of dengue fever, 235
while 609 exhibited severe symptoms. A total of ten patients died as a result of contracting the 236
virus. 237
5.1 Demand Side 238
The demand side analysis presents accessibility results from the patient’s perspective, using 239
travel times from the patient’s home to the healthcare facility where they were diagnosed. On 240
average, the revealed travel time is nearly 6 minutes longer than the time needed to reach the 241
closest health facility suggesting that individuals traveled beyond their closest option (see Table 242
2). The average potential travel time to healthcare facilities for patients with dengue fever is 243
estimated to be approximately 1 minute (median: 54 seconds; range: 0 – 6.79 minutes). Nearly 244
60% of patients had to travel 1 minute or less to their closest health facility, but in reality the 245
revealed pattern departed significantly from the potential, pointing to an important 246
underestimation in travel time (6.82 minutes, median: 6.15 minutes; range: 0 – 24.94 minutes). 247
Separating data by age groups results in similar patterns. The difference is largest in children 248
younger than 10 years old. When analyzing excess travel grouped by age groups (less than 10, 249
10-20, 20-50, and greater than 50), a slightly greater mean in travel time difference is identified 250
for the youngest and oldest patients as compared with the two middle groups. The distribution 251
according to revealed travel time alone is identical. 252
Stratifying patients by type of dengue virus suggests significant differences in revealed travel 253
times (see Table 3). Individuals diagnosed with classic dengue fever travel on average 6.6 254
minutes to a health facility, while individuals diagnosed with the severe type travel almost 9 255
minutes. Travel times increase even more to 10.3 minutes when the person died as a result of the 256
virus. Some of this variation can be explained by the protocol used to treat dengue fever. If an 257
individual suspects he/she presents symptoms aligned with the severe dengue type or they have 258
been diagnosed at a healthcare facility and symptoms suggest a severe type, the patient needs to 259
9
be hospitalized. Not all healthcare facilities are equipped to care for this type of complication, 260
therefore an individual may have to travel a longer distance to receive appropriate services which 261
may require a referral (this is also the case for patients who died of complications). There is a 262
significant difference in revealed travel times for dengue and severe dengue (t= 10.68, 263
p=0.0001). 264
265
Mean
(min) Median
(min) St. Deviation
(min)
Travel Time – Revealed 6.82 6.15 4.53
Travel Time – Potential 1.03 0.89 0.68
Excess Travel Time 5.78 5.09 4.46
Total Patients 8668
Table 2: One-way travel time for patients with dengue fever. 266 267
268
Dengue Severe Dengue Deaths
Mean
(min) St.
Deviation Mean (min)
St.
Deviation Mean (min)
St.
Deviation Travel Time –
Revealed 6.67 4.49 8.69 4.60 10.35 4.29
Travel Time –
Potential 1.03 0.68 1 0.65 1.1 0.80
Excess Travel Time 5.64 4.38 7.6 4.4 9.18 4.44
Total 8049 609 10
Table 3: One-way travel time, aggregated by type of dengue virus. 269 270
A significant difference in revealed travel times is observed among private and public 271
healthcare facilities (t= 16.29, p<0.01). Table 4 shows that almost twice as many individuals 272
were diagnosed at a private health facility and that those individuals were willing to travel 273
longer. Private healthcare facilities tend to have more resources. Therefore, people demonstrate a 274
preference for these facilities under the assumption that they will receive better care (e.g., more 275
doctors, shorter wait time, and better quality of service). 276
277
278
279
280
10
Type of Hospital Cooperative Private Public
Mean (min)
St.
Deviation Mean
(min) St.
Deviation Mean (min)
St.
Deviation Travel Time – Revealed 7.67 3.672 7.34 4.39 5.61 4.71
Travel Time – Potential 1.00 0.68 1.02 0.67 1.05 0.71
Total Patients 613 5314 2741
Table 4: One-way travel time, aggregated by type of health facility. 281 282
The analysis of travel time from the perspective of patients seeking a diagnosis suggests that 283
in spite of distance being a barrier, individuals find a way to overcome it to obtain better care. As 284
the disease becomes more life threatening (i.e., severe dengue) the desire or need to do so 285
becomes even greater. That is also the case when children and the elderly are affected. It is 286
important to note that dengue fever is similar to other endemic hemorrhagic fevers (e.g., malaria 287
and leptospirosis) and misdiagnosis could occur, leading patients to search for more qualified 288
doctors. It is clear that patients have a preference for privately owned healthcare facilities and are 289
willing to travel further to reach those. The results further suggest that it is important to consider 290
individual’s preferences in regards to health care when an endemic disease is the subject of 291
study. Personal knowledge of the disease and how it affects them might influence their 292
willingness to overcome distance barriers to obtain proper treatment. 293
5.2 Supply Side 294
The supply side analysis presents accessibility results from the healthcare facility perspective 295
where the patient was diagnosed with dengue fever virus. Average one-way travel time for each 296
health facility is estimated by aggregating all the travel times of that facility. The pattern in 297
Figure 3a clearly indicates that longer travel times are incurred at those facilities located in the 298
center of town and along the north-south spinal corridor. Assignment of patients to the facility 299
that minimizes travel time ('closest assignment') is often a desirable outcome of public facility 300
location decisions. Intuitively, hospitals with the longest travel time are also the ones with the 301
least number of closest assignments (correlation = -0.68, p<.01, Figure 3b). Comparing Figures 302
3a and 3b, it is clear that if patients were to use their closest health facility, there would be a 303
more even distribution of services within the city. However, this would be ideal if all facilities 304
provided hospitalization comparable services, had similar carrying capacities, and were equally 305
equipped, which is not the case (Table 1). 306
Facilities which are more evenly distributed throughout the city, health posts and healthcare 307
centers, are located and designed to serve neighborhoods. They correspond to level I service 308
which results in a general doctor’s visit. At this level, dengue fever can be diagnosed but no 309
treatment can be provided if the patient is diagnosed with the severe type. In 2004, the Isaias 310
Duarte Cancino Hospital (HIDC) was built in the eastern side of the city to help serve the 311
demand for health services (Casas et al. 2010, see Figure 2c). However, the results show that 312
individuals have a preference for hospitals that are further away. 313
11
314
315
Figure 3: (a) Average one-way travel time, aggregated at the facility level, (b) percentage of closest assignments: A 316 larger dot denotes a healthcare center that receives patients for whom that center is their closest choice. 317 318
5.3 Neighborhood Level 319
Travel times are aggregated at the neighborhood level (Figure 4), as this scale of aggregation 320
allows us to account for socio-economic information when modeling travel time. In the three 321
figures, a darker blue color indicates regions where a health facility could not be accessed within 322
a 15 minute driving budget. Lighter blue colors are reflective of a shorter travel time, pointing to 323
a higher level of geographic accessibility. Significant geographic differences in potential travel 324
time are observed across different neighborhoods of the city. Patients residing along the north-325
south corridor experience better geographic access to healthcare facilities than patients living in 326
the periphery, owing to the geographical proximity of healthcare centers. This is the result of the 327
city’s urbanization process, where many services concentrated in the core and along the spine. 328
Following a health reform in the early 1990s (Echeverri 1996), existing newer and smaller 329
healthcare facilities were transformed to comply with the new directives. The new directives 330
proposed a health system model based on free regulated competition with demand subsidies to 331
12
protect the poor population. This also resulted in the construction of new facilities and the 332
delineation of service areas. 333
Excess travel results are shown in Figure 4c. Small values are indicative of patients treated at 334
their closest healthcare centers, while larger values (dark blue) indicate longer diagnostic travel 335
times. The differences are smaller in the center of the city along the North-South corridor and 336
greater along the Eastern edge of the city. An outlier to the southwest of the city corresponds to 337
the Fundación Valle del Lili, the only level IV facility containing state-of-the art technology and 338
facilities; it is the only one of its kind in the city. 339
340
341
Figure 4: (a) and (b) Potential and revealed one-way travel time aggregated at the neighborhood level, respectively, 342 (c) excess travel time (revealed minus potential) by car for patients. Neighborhoods forming a significant local 343 cluster of higher (lower) travel time are outlined in white (orange) (LISA statistic). 344
345
The local Moran’s I statistic (Anselin 1995) is used to test whether neighborhoods in close 346
spatial proximity to one another have similar travel impedance values. A cluster of high excess 347
travel is found in the neighborhoods in the Eastern edge of the city, while another cluster of low 348
excess values is found in the old center of Cali. The former suggests that people do not utilize the 349
newly created public hospital in the area. This might be for lack of knowledge of what kind of 350
services are available, for safety reasons (the hospital is in an area characterized by the presence 351
of gangs), or the belief that private institutions provide a better service. Collectively, the results 352
in Figure 4 demonstrate that while the potential access to a healthcare facility appears quite even 353
throughout the city, the revealed access paints a very different picture. 354
13
5.4 Multi-Level Model Results 355
Results of the regression analysis are displayed in Tables 4 and 5. Three models are run in 356
sequential order: a null model with no explanatory variables, a second, model which incorporates 357
individual-level independent variables, and a third model which adds neighborhood-level 358
explanatory factors. 359
For the first dependent variable, travel time to the visited hospital for each patient, the null 360
model indicates that the variance component associated with the random intercept is 5.95. Given 361
the variance component of the residual, the intraclass correlation coefficient is computed 362
revealing that 28 percent of the variation in travel time is attributed to neighborhood-level 363
factors. The intercept of the fixed effect indicates the average travel time for all individuals to 364
their chosen hospital across neighborhoods is 6.86 minutes. In the second model, individual-365
level variables are incorporated to explain variations in travel time. These include three age 366
categories and gender. The age category of 20-50 is used as the reference category. Results 367
suggest that compared with this latter group, those aged 11-20 have significantly lower travel 368
times, while those in the oldest age group have significantly higher travel times to reach care. 369
Compared with females, travel times for males are statistically significantly higher. 370
Finally, in the third model, two neighborhood-level variables are incorporated to help 371
account for some of the neighborhood variation: the neighborhood’s socioeconomic strata and 372
the average travel time from a neighborhood’s centroid to all city healthcare facilities, which 373
serves as a measure of urban centrality. Both variables are statistically significant at p< 0.05. 374
Those living in neighborhoods of a higher socioeconomic strata have longer travel times to reach 375
care while those with longer average travel time to all healthcare facilities (less centrally located) 376
similarly experience longer travel. The coefficient and significance of socioeconomic strata 377
indicates that it is the more important factor in explaining longer travel times. The results support 378
those found by others which suggest that access to healthcare facilities varies by age, sex, and 379
income (Ensor and San 1996, Parker and Campbell 1998, Arcury et al. 2005). In terms of model 380
fit, the AIC and BIC is lowest in the third model indicating the best fit for the data. 381
In the second set of models, excess travel is used as the dependent variable. The null model 382
produces nearly identical results to the prior set of models with an interclass coefficient of 26 383
percent. The second model with individual-level factors suggests that compared with patients in 384
the middle age category (ages 20-50), younger patients (ages 11-20) have significantly less 385
excess travel to health care and older patients (aged 50 and above) have significantly longer 386
excess travel times. Consistent with the previous set of models, males have significantly more 387
excess travel as compared with females. Finally, in the third model, when neighborhood factors 388
are accounted for, the socio-economic strata of a neighborhood is positively associated with 389
excess travel, but the variable representing the average travel time to all healthcare facilities is 390
not significant. This latter result stands in contrast to the previous models explaining travel time 391
alone where this variable was positive and significant. In both sets of models, neighborhood 392
socioeconomic strata serves as a proxy for the socioeconomic status of individuals, assuming that 393
residents sort themselves in neighborhoods according to education, income leading to relatively 394
14
homogeneous groupings of individuals in each neighborhood. This variable is significant and 395
positive in both models suggesting that residents of a higher socioeconomic status travel farther 396
and are more likely to travel to a hospital father from their closest facility. This is plausibly the 397
result of an increased choice set amongst the more well-off. Given that traveling further 398
distances incurs as a greater cost (both monetarily as well as in respect to travel time), the 399
decision to travel a farther distance to utilize a particular hospital. Lower-income residents may 400
also be restricted by transportation options such as the availability of private vehicles to get to 401
particular destinations. 402
403
Dependent Variable = Travel Time (by car)
Model 1
(Null) Model 2:
Individual level
variables
Model 3: Individual and
Neighborhood
Fixed
Effects
Intercept 6.86* (0.15) 6.75* (0.17) 1.49* (0.31) Aged 10 and
Younger 0.12 (0.10) 0.19 (0.10)
Aged 11-20 -0.30* (0.11) -0.29* (0.11)
Aged 20-50 (Ref.) ------ Aged >50 0.37* (0.21) 0.19 (0.13)
Male 0.21* (0.08) 0.21* (0.08)
SES Strata 1.17* (0.04)
Average distance 0.002* (0.0004)
Random
Effects
Intercept
(Neighborhood) 5.95 (2.44) 5.99 (2.44) 6.13 (2.47)
Residual 15.11 (3.89) 15.05 (3.88) 13.49 (3.67)
Model Fit
Statistics
AIC 48834.23 48814.48 47899.10 BIC 488553.43 48863.96 47962.71 LogLik -24414.11 -24400.24 -23940.55 Deviance 48828.23 48800.48 47881.10 Number of Obs. 8668 8668 8668 Number of Groups 327 327 327
*Significant at p<0.05; Standard errors shown in parentheses
Table 5: Multi-level model results: dependent variable travel time to health facility by car. 404 405 406 407 408 409 410 411 412 413 414 415 416 417
15
Dependent Variable = Excess Travel
Model 1
(Null) Model 2:
Individual level
variables
Model 3: Individual and Neighborhood
Fixed
Effects
Intercept 5.82* (0.14) 5.69* (0.16) 1.28* (0.30) Aged 10 and Younger 0.12 (0.11) 0.20* (0.10)
Aged 11-20 -0.29* (0.11) -0.28* (0.11)
Aged 20-50 (Ref.) ------ ------
Aged >50 0.38* (0.14) 0.20 (0.13)
Male 0.21* (0.08) 0.22* (0.08)
SES Strata 1.17* (0.04)
Average distance -0.0001 (0.0004)
Random
Effects Intercept
(Neighborhood) 5.41 (2.44) 5.44 (2.33) 5.88 (2.43)
Residual 15.15 (3.89) 15.09 (3.88) 13.52 (3.68)
Model Fit
Statistics
AIC 48834.23 48810.36 47915.31 BIC 488553.43 48859.93 47978.91 LogLik -24414.11 -24398.18 -23948.65 Deviance 48828.23 48796.36 47897.31 Number of Observations 8668 8668 8668 Number of Groups 327 327 327
*Significant at p<0.05; Standard errors shown in parentheses
Table 6: Multi-level model results: dependent variable excess travel time to health facility. 418 419 420
Taken together, these results suggest that those located on the outer periphery of the city 421
travel longer, in general, to reach health care, but their spatial location does not dictate whether 422
or not they are willing to travel beyond their closest hospital to reach care. This choice is a 423
factor of the socio-economic conditions of that individual’s neighborhood and of other individual 424
characteristics. Regardless of their proximity to the overall offering of urban healthcare 425
facilities, those with greater resources seem to be willing to overcome spatial barriers to access 426
better care. In this final model, the variable representing the youngest patients is positive and 427
statistically significant. Given the vulnerability of these patients, it makes sense that they would 428
be taken for care at a health facility with more critical care resources that may not be their closest 429
facility. Model fit criteria point to the third model as the best fit with the data. 430
431
6 Conclusions 432
Improving access to primary health care can have a direct impact in improving health 433
outcomes (Perry and Gesler 2000). Although travel time for patients diagnosed with dengue 434
fever in the city of Cali may appear relatively small (average = 6 minutes), the departure from 435
the potential travel time to the closest facility is significant, revealing an underestimation of 436
potential travel time. The map illustrating the distribution of potential access to the closest 437
healthcare facility indicates an equitable distribution of potential travel times, however, the 438
16
revealed travel times pointed to significant clusters of longer travel times throughout the urban 439
area. Given that the overwhelming majority of accessibility studies are potential in nature – often 440
used for evaluating the equity of the distribution of facilities or a transportation network – this 441
study emphasizes that actual travel times may depart from these estimations and highlight 442
inequalities that are not apparent in potential access scenarios. 443
In this study, we found that both individual and neighborhood characteristics helped explain 444
travel times and excess travel times, or the difference between the travel time to the closest 445
facility and the chosen facility. Individuals in neighborhoods of a higher socioeconomic status 446
chose to travel to facilities located further away than their closest facility. Travel times also 447
increase for patients diagnosed with severe dengue fever and those whose diagnosis resulted in 448
death. There are other endemic hemorrhagic fevers (e.g., malaria and leptospirosis) similar to 449
dengue fever, that can lead patients to search for more qualified doctors to obtain an accurate 450
diagnosis. In the year 2010, there were 608 cases of malaria (Municipal 2012) and 344 of 451
leptospiroris (Pública 2011). Even though these numbers are not comparable to those of dengue 452
for the year 2010, there is potential for misdiagnosing diseases. 453
It is important to note that besides comparisons in travel times, the results suggest that 454
patients favor healthcare facilities located in the city’s central core. This area has a concentration 455
of services with facilities that (1) are better equipped to treat the virus if necessary, (2) are mostly 456
privately owned and (3) have been in operation for at least 30 years or more. In the rest of the 457
city, a more uniform distribution of public healthcare facilities can be observed. A personal 458
choice to receive private or public health care may thus motivate individuals to travel further. 459
The multi-level models show that travel time is influenced mainly by income and distance to the 460
facilities. This corroborates findings by Ensor and San (1996), Parker and Campbell (1998), and 461
Arcury et al. (2005) and again may point to a wider choice set of hospitals for wealthier 462
residents. Those living in neighborhoods of a higher socioeconomic status were more likely to 463
travel farther and go beyond their closest facility to receive care, while residents in lower income 464
neighborhoods were more likely to use the closest facility. 465
6.1 Assumptions and Limitations 466
This study relied on several assumptions which may affect the validity of our results. First, 467
we estimated travel impedance as the distance between the patient’s residential address to the 468
hospital where the diagnosis was conducted. The residential address may not have been the place 469
where the patient was infected. Second, travel was assumed to be done by private automobile 470
and not by public transportation. Third, our measure of travel impedance does not account for 471
road construction/congestion and as such our distances are likely to be underestimates of the true 472
travel. Fourth, geocoding was conducted at the nearest intersection from the place of residence; 473
this choice is motivated by a lack of accurate address ranges and an effort to preserve patient’s 474
privacy. Fifth, we did not incorporate information related to the capacity of each healthcare 475
center (such as the number of doctors and nurses). This may influence the decision to seek health 476
care service elsewhere. 477
17
6.2 Strengths 478
Despite these limitations, our paper demonstrated several strengths. To the authors’ 479
knowledge, this is the first attempt to estimate spatial accessibility to health care center during a 480
dengue fever epidemic. Second, the paper estimates accessibility using healthcare center 481
discharge information. The quasi majority of studies modeling spatial accessibility have used 482
potential accessibility metrics, which are based on the estimated travel impedance to the closest 483
healthcare center. As demonstrated in this paper, the closest healthcare center is rarely the 484
patient’s choice. Third, our accessibility is estimated at the individual level unlike most studies 485
which use aggregated information. 486
6.3 Implications 487
Results from this work are very valuable to health officials to design and plan protocols for 488
prevention and control of the dengue virus. For example, campaigns geared towards children 489
showing where they can be most efficiently treated can be designed as to capitalize on the 490
facilities which can provide this particular service (i.e., children’s hospital). Second, awareness 491
can be raised as to the effects of the virus on this vulnerable group. Knowing which facilities are 492
preferred by patients for treatment will help focus resources at these locations to educate people 493
on how to prevent such disease. This is all the more important given that facilities offer different 494
services and it is important not to waste resources if they are not needed. Not all public facilities 495
can be equipped the same way, which is why the different levels exist. Being aware of patients’ 496
preferences regarding their choice of facilities can help officials in resource allocation. It is also 497
possible to identify focal points of the virus by identifying locations that are overwhelmed with 498
dengue patients. Knowledge on the distribution of the Aedes Aegypti inside the household and 499
the existence of the larvae and pupae can further aid in this process (Getis et al. 2003). 500
6.4 Future Research 501
Although data availability has constrained the ability to identify other potential variables that 502
could influence patients’ choice for a healthcare facility, personal interviews geared towards the 503
decision making process will help in the collection of such data. For instance, if symptoms were 504
discovered shortly after the incubation period, an individual may have more time to choose 505
among different health care facilities. Alternatively, a late discovery may require a more urgent 506
diagnosis and an individual may decide to visit the closest facility. Second, future research will 507
investigate how public transit affects travel estimates and its impact on the patient’s choice of 508
healthcare facility. Third, we will compare our travel estimates to the ones from online providers, 509
such as Google Maps (see for instance Delmelle et al. 2013a). Finally, it would be valuable to 510
contrast the results of this study for individuals who are diagnosed with severe hemorrhagic 511
dengue fever symptoms. 512
18
7 Acknowledgments 513
The authors would like to thank MetroCali for providing the city’s street network and the 514
Public Health Secretariat of the city of Cali for the dengue fever surveillance data. 515
516
19
8 References 517
Aldstadt, J., I. K. Yoon, R. G. Tannitisupawong, S. J. Thomas, R. V. Gibbons, A. Uppapong, S. 518
Iamsirithaworn, A. L. Rothman & T. W. Scott (2012) Space-time analysis of hospitalised 519
dengue patients in rural Thailand reveals important temporal intervals in the pattern of 520
dengue virus transmission. Tropical Medicine & International Health, 17, 1076-1085. 521
Anselin, L. (1995) Local indicators of spatial association - LISA. Geographical Analysis, 27, 93-522
115. 523
Apparicio, P., M. Abdelmajid, M. Riva & R. Shearmur (2008) Comparing alternative approaches 524
to measuring the geographical accessibility of urban health services: Distance types and 525
aggregation-error issues. International Journal of Health Geographics, 7, 7. 526
Arcury, T. A., W. M. Gesler, J. S. Preisser, J. Sherman, J. Spencer & J. Perin (2005) The Effects 527
of Geography and Spatial Behavior on Health Care Utilization among the Residents of a 528
Rural Region. Health Services Research, 40, 135-156. 529
Bahr, J. & R. Wehrhahn (1993) Life expectancy and infant mortality in Latin America. Social 530
Science and Medicine, 36, 1373-1382. 531
Brabyn, L. & C. Skelly (2002) Modeling population access to New Zealand public hospitals. 532
International Journal of Health Geographics, 1, 1-9. 533
Buchmueller, T. C., M. Jacobson & C. Wold (2006) How far to the hospital? The effect of 534
hospital closures on access to care. Journal of Health Economics, 25, 740-761. 535
Buor, D. (2003) Analysing the primacy of distance in the utilization of health services in the 536
Ahafo-Ano south district, Ghana. International Journal of Health Planning and 537
Management, 18, 293-311. 538
Cali, A. d. S. d. 2010a. Cali en cifras. Cali: Alcaldia de Cali. 539
20
---. 2014. Cali en cifras. Cali: Alcaldia de Cali. 540
Cali, S. d. S. P. M. d. 2010b. Historia del dengue en Cali. Endemia o una continua epidemia. 541
Cali: Secretaria de Salud Publica Municipal de Cali. 542
Casas, I., E. Delmelle & A. Varela (2010) A space-time approach to diffusion of health service 543
provision information. International Regional Science Review, 33, 134-156. 544
Connor, R. A., S. D. Hillson & J. E. Krawelski (1995) Competition, professional synergism, and 545
the geographic distribution of rural physicians. Med Care, 33, 1067-1078. 546
Council on Healh Research for Development, C. 2006. Supporting health research systems 547
developmentin Latin America. 1-26. 548
DANE. 2016. Metodología de estratificación urbana tipo 1. DANE Website: DANE. 549
Delamater, P. L., J. P. Messina, A. M. Shortridge & S. C. Grady (2012) Measuring geographic 550
access to health care: raster and network-based methods. International Journal of Health 551
Geographics, 11. 552
Delmelle, E., I. Casas, J. Rojas & A. Varela (2013a) An exploratory analysis of spatio-temporal 553
patterns of sengue fever in Cali, Colombia. International Journal of Applied Geospatial 554
Research, 4. 555
Delmelle, E. C. & I. Casas (2012) Evaluating the spatial equity of bus rapid transit-based 556
accessibility patterns in a developing country: The case of Cali, Colombia. Transport 557
Policy, 20, 36-46. 558
Delmelle, E. M., C. H. Cassell, C. Dony, E. Radcliff, J. P. Tanner, C. Siffel & R. S. Kirby 559
(2013b) Modeling travel impedance to medical care for children with birth defects using 560
Geographic Information Systems. Birth Defects Research Part A: Clinical and Molecular 561
Teratology, 97, 673-684. 562
21
Dony, C. C., E. M. Delmelle & E. C. Delmelle. (2015) Re-conceptualizing accessibility to parks 563
in multi-modal cities: A Variable-width Floating Catchment Area (VFCA) method. 564
Landscape and Urban Planning, 143, 90-99. 565
Echeverri, O. (1996) Ley 100: Quo Vadis? Colombia Medica, 27, 39-41. 566
Ensor, T. & P. B. San (1996) Access and payment for health care: the poor of Northern Vietnam. 567
The International Journal of Health Planning and Management, 11, 69-83. 568
Fosu, G. B. (1989) Access to health care in urban areas of developiong societies. Journal of 569
Geographical Systems, 30, 398-411. 570
Friedman, J. M., L. Hagander, C. D. Hughes, K. A. Nash, A. F. Linden, J. Blossom & J. G. 571
Meara (2013) Distance to hospital and utilization of surgical services in Haiti: do children 572
, delivering mothers , and patients with emergent surgical conditions experience greater 573
geographical barriers to surgical care ? The International Journal of Health Planning and 574
Management, 28, 248-256. 575
Getis, A., A. C. Morrison, K. Gray & T. W. Scott (2003) Characteristics of the spatial pattern of 576
the dengue vector, Aedes aegypti, in Iquitos, Peru. The American Journal of Tropical 577
Medicine and Hygiene, 69, 494-505. 578
Guagliardo, M. F. (2004) Spatial accessibility of primary care: concepts, methods and 579
challenges. International Journal of Health Geographics, 3, 1-13. 580
Haynes, R., A. Lovett & G. Sünnenberg (2003) Potential accessibility, travel time, and consumer 581
choice: geographical variations in general medical practice registrations in Eastern 582
England. Environment and Planning A, 35, 1733-1750. 583
Haynes, R., A. P. Jones, V. Sauerzapf & H. Zhao (2006) Validation of travel times to hospital 584
estimated by GIS. International Journal of Health Geographics, 5, 1-8. 585
22
Haynes, R. M. 1986. The geography of health services in Britain. London: Croom Helm. 586
Hsueh, Y.-H., J. Lee & L. Beltz (2012) Spatio-temporal patterns of dengue fever cases in 587
Kaoshiung City, Taiwan, 2003–2008. Applied Geography, 34, 587-594. 588
Hu, R., S. Dong, Y. Zhao, H. Hu & Z. Li (2013) Assessing potential spatial accessibility of 589
health services in rural China: a case study of Donghai County. International Journal for 590
Equity in Health, 12. 591
Jones, A. P. & G. Bentham (1997) Health service accessibility and deaths from asthma in 401 592
local authority districts in England and Wales, 1988-92. Thorax, 52, 218-222. 593
Joseph, A. E. & D. R. Phillip. 1984. Accessibility & utilization: Geographical perspectives on 594
health care delivery. New York: Harper & Row. 595
Kaljee, L. M., A. Pach, K. Thriemer, B. Ley, S. M. Ali, M. Jiddawi, M. Puri, L. von Seidlein, J. 596
Deen, L. Ochiai, T. Wierzba & J. Clemens (2013) Utilization and accessibility of 597
healthcare on Pemba Island, Tanzania: implications for health outcomes and disease 598
surveillance for typhoid fever. The American Journal of Tropical Medicine and Hygiene, 599
88, 144-152. 600
Kittayapong, P., S. Yoksan, U. Chansang, C. Chansang & A. Bhumiratana (2008) Suppression of 601
dengue transmission by application of integrated vector control strategies at sero-positive 602
GIS-based foci. The American Journal of Tropical Medicine and Hygiene, 78, 70-76. 603
Koenraadt, C. J., J. Aldstadt, U. Kijchalao, R. Sithiprasasna, A. Getis, J. W. Jones & T. W. Scott 604
(2008) Spatial and temporal patterns in pupal and adult production of the dengue vector 605
Aedes aegypti in Kamphaeng Phet, Thailand. The American Journal of Tropical 606
Medicine and Hygiene, 79, 230-238. 607
23
Kolivras, K. (2006) Mosquito Habitat and Dengue Risk Potential in Hawaii: A Conceptual 608
Framework and GIS Application. The Professional Geographer, 58, 139-154. 609
Kuno, G. (1995) Review of the factors modulating dengue transmission. Epidemiological 610
Review, 17, 321-335. 611
Kwan, M.-P., I. Casas & B. C. Schmitz (2004) Protection of geo-privacy and accuracy of spatial 612
information: how effective are geographical masks? Cartographica, 39, 15-38. 613
Langford, M. & G. Higgs (2006) Measuring Potential Access to Primary Healthcare Services: 614
The Influence of Alternative Spatial Representations of Population. The Professional 615
Geographer, 58, 294-306. 616
Leese, G. P., Z. Feng, R. M. Leese, C. Dibben & A. Emslie-Smith (2013) Impact of health-care 617
accessibility and social deprivation on diabetes related foot disease. Diabetic medicine: A 618
Journal of the British Diabetic Association, 30, 484-490. 619
Levesque, J.-F., M. F. Harris & G. Russell (2013) Patient-centred access to health care: 620
conceptualising access at the interface of health systems and populations. International 621
Journal for Equity in Health, 12. 622
Luo, W. & T. Whippo (2012) Variable catchment sizes for the two-step floating catchment area 623
(2SFCA) method. Health & Place, 18, 789-795. 624
Mammen, M., C. Pimgate, C. Koenraadt, A. Rothman, J. Aldstadt, A. Nisalak, R. Jarman, J. 625
Jones, A. Srikiatkhachorn, C. Ypil-Butac, A. Getis, S. Thammapalo, A. Morrison, D. 626
Libraty, S. Green & T. Scott (2008) Spatial and temporal clustering of dengue virus 627
transmission in Thai villages. PLoS Medicine, 5, e205. 628
Martin, D., H. Wrigley, S. Barnett & P. Roderick (2002) Increasing the sophistication of access 629
measurement in a rural healthcare study. Health & Place, 8, 3-13. 630
24
Martín, J. L. S., O. Brathwaite, B. Zambrano, J. O. Solórzano, A. Bouckenooghe, G. H. Dayan & 631
M. G. Guzmán (2010) The Epidemiology of Dengue in the Americas Over the Last Three 632
Decades: A Worrisome Reality. American Journal of Tropical Medicine and Hygiene, 633
82, 128-135. 634
McGrail, M. R. & J. S. Humphreys (2009) Measuring spatial accessibility to primary care in 635
rural areas: Improving the effectiveness of the two-step floating catchment area method. 636
Applied Geography, 29, 533-541. 637
McGrail, M. R. (2012) Spatial accessibility of primary health care utilising the two step floating 638
catchment area method: an assessment of recent improvements. International Journal of 639
Health Geographics, 11. 640
Morrison, A. C., A. Getis, M. Santiago, J. G. Rigau-Perez & P. Reiter (1998) Exploratory space-641
time analysis of reported dengue cases during an outbreak in Florida, Puerto Rico, 1991-642
1992. The American Journal of Tropical Medicine and Hygiene, 58, 287-298. 643
Morrison, A. C., K. Gray, A. Getis, H. Astete, M. Sihuincha, D. Focks, D. Watts, J. D. Stancil, J. 644
G. Olson, P. Blair & T. W. Scott (2004) Temporal and geographic patterns of Aedes 645
aegypti (Diptera: Culicidae) production in Iquitos, Peru. Journal of Medical Entomology, 646
41, 1123-1142. 647
Municipal, S. d. S. P. 2012. Boletím epidemiológico 2011 y 2012. Cali, Colombia: Secretaría de 648
Salud Pública Municipal. 649
Noor, A. M., D. Zurovac, S. Hay, S. Ochola & R. W. Snow (2003) Defining equity in physical 650
access to clinical services using geographical information systems as part of malaria 651
planning and monitoring in Kenya. Tropical Medicine & International Health, 8, 917-652
926. 653
25
Parker, E. B. & J. L. Campbell (1998) Measuring access to primary medical care: some examples 654
of the use of geographical information systems. Health & Place, 4, 183-193. 655
Penchansky, R. & J. W. Thomas (1981) The concept of access. Med Care, 19, 110-140. 656
Perry, B. & W. Gesler (2000) Physical access to primary health care in Andean Bolivia. Social 657
Science & Medicine, 50, 1177-1188. 658
Phillips, R. L. & E. J. Lindbloom (2000) Using Geographic Information Systems to Understand 659
Health Care Access. Archives of Family medicine, 9, 971-978. 660
Pública, S. d. S. 2011. Salud en cifras 2011. Cali, Colombia: Secretaría de Salud Pública. 661
Restrepo, L. D. E. (2006) El plan piloto de cali 1950. Revista Bitacora Urbano Territorial, 1, 662
222-223. 663
Rojas, J. 2011. Dengue y dengue grave. ed. S. d. S. M. Cali, 18. Secretaría de Salud Municipal 664
Cali. 665
Rosero-Bixby, L. (2004) Spatial access to health care in Costa Rica and its equity: a GIS-based 666
study. Social Science & Medicine, 58, 1271-1284. 667
Salud, I. N. d. 2016. Vigilancia rutinaria. 668
Schonfeld, H. K., J. F. Heston & I. S. Falk (1972) Numbers of physicians required for primary 669
medical care. N. Engl J. Med, 286, 571-576. 670
Shahid, R., S. Bertazzon, M. L. Knudtson & W. A. Ghali (2009) Comparison of distance 671
measures in spatial analytical modeling for health service planning. BMC health services 672
research, 9, 200. 673
26
Siqueira-Junior, J. B., I. J. Maciel, C. Barcellos, W. V. Souza, M. S. Carvalho, N. E. Nascimento, 674
R. M. Oliveira, O. Morais-Neto & C. M. Martelli (2008) Spatial point analysis based on 675
dengue surveys at household level in central Brazil. BMC Public Health, 8, 361. 676
Tanser, F., B. Gijsbertsen & K. Herbst (2006) Modelling and understanding primary health care 677
accessibility and utilization in rural South Africa: An exploration using a geographical 678
information system. Social Science & Medicine, 63, 691-705. 679
Vazquez-Prokopec, G. M., U. Kitron, B. Montgomery, P. Horne & S. A. Ritchie (2010) 680
Quantifying the spatial dimension of dengue virus epidemic spread within a tropical 681
urban environment. PLoS Neglected Tropical Diseases, 4, e920. 682
Wang, F. & Q. Tang (2013) Planning toward equal accessibility to services: a quadratic 683
programming approach. Environment and Planning B: Planning and Design, 40, 195-684
212. 685
WHO. 2009. Dengue and dengue haemorrhagic fever - Fact sheet no. 117. WHO. 686
---. 2016. Dengue and dengue haemorrhagic fever - Fact sheet no. 117. WHO. 687
Wong, L. Y., B. H. Heng, J. Tiang, S. Cheah & C. Tan (2012) Using spatial accessibility to 688
identify polyclinic service gaps and volume of under-served population in Singapore 689
using Geographic Information System. The International Journal of Health Planning and 690
Management, 27, 173-185. 691
692