SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. METHODS
  4. RESULTS
  5. DISCUSSION
  6. LIMITATIONS
  7. CONCLUSIONS
  8. Acknowledgments
  9. References
  10. Supporting Information:
  11. Supporting Information

ACADEMIC EMERGENCY MEDICINE 2012; 19:180–188 © 2012 by the Society for Academic Emergency Medicine

Abstract

Objectives:  The main objective was to explore the relationship between socioeconomic status and the spatial distribution of ambulance calls, as modeled in the island nation of Singapore, at the Development Guide Plan (DGP) level (equivalent to census tracts in the United States).

Methods:  Ambulance call data came from a nationwide registry from January to May 2006. We used a conditional autoregressive (CAR) model to create smoothed maps of ambulance calls at the DGP level, as well as spatial regression models to evaluate the relationship between the risk of calls with regional measures of socioeconomic status, such as household type and both personal and household income.

Results:  There was geographical correlation in the ambulance calls, as well as a socioeconomic gradient in the relationship with ambulance calls of medical-related (but not trauma-related) reasons. For instance, the relative risk (RR) of medical ambulance calls decreased by a factor of 0.66 (95% credible interval [CrI] = 0.56 to 0.79) for every 10% increase in the proportion of those with monthly household income S$5000 and above. The top three DGPs with the highest risk of medical-related ambulance calls were Changi (RR = 29, 95% CrI = 24 to 35), downtown core (RR = 8, 95% CrI = 6 to 9), and Orchard (RR = 5, 95% CrI = 4 to 6).

Conclusions:  This study demonstrates the utility of geospatial analysis to relate population socioeconomic factors with ambulance call volumes. This can serve as a model for analysis of other public health systems.

There has been a growing understanding that emergency medical services (EMS) ambulance calls are not random events, but occur in patterns and trends that can be observed historically.1 This is related to movement patterns of people according to time of day, as well as geographical epidemiology and characteristics of the population. Spatial analytic tools can potentially help to advise public health and EMS policy.

The literature on spatial applications to ambulance calls data is rather restricted. Bassil and colleagues2 mapped the rates of heat-related illnesses as a proportion of all ambulance calls in Toronto from 2002 to 2005. The authors provided maps at the neighborhood level, but inherently failed to account for any spatial correlation in the data. The authors postulated a relationship between the spatial distribution of ambulance calls and socioeconomic status, but did not prove this in their study.

One of the earliest papers to suggest the use of geographical information systems (GIS) to plan and assess ambulance response times was from Peters and Hall.3 They found that sociospatial differentiation (e.g., location factors, demographic characteristics, and organizational characteristics of the delivery system) were important predictors of the spatial–temporal patterns of realized response times. There have been other attempts to use GIS methods to study out-of-hospital cardiac arrests,4–7 determine appropriate means of trauma patient transportation,8 and investigate new helipad locations and improve timely response,9 but to the best of our knowledge, none have examined the association between risk of ambulance calls and regional measures of socioeconomic status.

The main aim of this project was to study the relationship between the risk of medical- and trauma-related ambulance calls originating from residential areas in Singapore, with sociodemographic variables measured at the regional level. The secondary objective was to provide smoothed relative risk (RR) maps of medical- and trauma-related ambulance calls from all areas in Singapore. The motivation for this research was twofold. First, we intended to bridge the gap in literature about the relationship between socioeconomic status and ambulance calls. Second, given the projection that the volume of ambulance calls in Singapore was going to increase substantially in the future, mapping the risk of ambulance calls geographically would provide policy-makers with tools to identify areas at greater risk, as well as to prioritize and plan for health services.

METHODS

  1. Top of page
  2. Abstract
  3. METHODS
  4. RESULTS
  5. DISCUSSION
  6. LIMITATIONS
  7. CONCLUSIONS
  8. Acknowledgments
  9. References
  10. Supporting Information:
  11. Supporting Information

Study Design

This was an observational ecologic study design that used data at an aggregate level. Ethics approval was obtained from the Singapore General Hospital institutional review board. The National Medical Research Council, as well as the Duke–National University Singapore Graduate Medical School, helped fund this study, but had no role in the conduct and analysis of the study.

Study Setting and Population

Data were from two different sources. First, our ambulance calls data came from the Cardiac Arrest and Resuscitation Epidemiology (CARE) study.1 The Singapore CARE study has been ongoing since 2001. It is a national cardiac arrest database that integrates data from dispatch, the national ambulance service, and hospital outcomes. Data entry is from manual forms completed by ambulance and hospital staff, with a verification and quality process before electronic entry. In the CARE phase 3 study (2006), it was expanded to include all ambulance calls to the “9-9-5” national emergency number, not only cardiac arrests.10 Our study included all ambulance calls made from January 2006 to May 2006. Exclusion criteria were those calls that did not require an ambulance to be dispatched or “false” calls for which no ambulance was dispatched. The CARE study involved collaboration between the six major public hospitals in Singapore; the Singapore Civil Defence Force (SCDF); the Health Sciences Authority; and the Clinical Trials and Epidemiology Research Unit, Ministry of Health, Singapore.

Second, the Singapore Population Census 2000 provided us with data on Development Guide Plan (DGP) level measures such as age distribution, highest educational level, high personal and household monthly income (more than Singapore [S] $5,000), professional occupation, and average household size. A list of the variables and their descriptions are provided online in Data Supplement S1 (available as supporting information in the online version of this paper). The census is conducted every 10 years in Singapore. The census involves a population count of all Singapore residents, including citizens and permanent residents, but excludes foreign workers without residency, foreign students, and transient students. Detailed information on households is solicited from 20% of all households. Further details on the methodology used in the census can be found at the Singapore Department of Statistics Web site.11

Singapore is a city-state with a land area of 704 square kilometers and a population of 4.4 million.12 The country’s EMS system is run by the SCDF, which currently operates 46 ambulances based in 15 fire stations and 16 satellite stations. They use a fixed deployment system operating out of these geographically static locations. It is primarily a single-tier system, able to provide basic life support and defibrillation with automated external defibrillators. Singapore EMS is activated by a universal, centralized, enhanced “9-9-5” dispatching system run by the SCDF and using computer-aided dispatch, medical dispatch protocols, global positioning satellite automatic vehicle locating systems, and road traffic monitoring systems. Emergency ambulance call volumes have been increasing every year for the past 5 years, with 118,912 emergency calls received in 2009.13 It can be expected that with an aging population,14 emergency call volumes will continue to increase in the foreseeable future.

Study Protocol

Based on the address from where the call was made, each case was assigned to a DGP based on the Urban Redevelopment Authority Master Plan 2003.15 Addresses that had a missing postal code were looked up using the online yellow pages directory. Those with just street names were assigned the postal code in the middle of the street. Unfortunately, the geo-codes from the automatic vehicle locator system were not available for this study. For the main analysis involving association with risk factors, we included only calls made from known residential addresses, as this will allow us to map the calls with the subject’s socioeconomic status, as measured from their residential information. For the secondary analysis involving mapping the risks at the DGP level, we used all ambulance calls, including those from residential as well as nonresidential locations. Bayesian spatial models were used to analyze the data.

Measures

The main outcome measure in our study was the RR of ambulance calls at each DGP. Each DGP covers a planning area with a population of around 150,000 served by a town center. The planning areas are further divided into subzones, each served by a commercial center. Trauma cases were defined as those caused by blunt or penetrating injury, drowning, electrical shock, burns, poisoning, etc. Medical cases were those with “nonaccidental” causes.

We calculated the total observed (Oi) ambulance calls in each DGP by summing up the cases. The expected counts of ambulance calls were calculated as Ei = (DGP_Popi/Totalpop) × Total_Calls, where DGP_Popi refers to total population in the ith DGP and Totalpop and Total_Calls refer to the overall number of population and ambulance calls, respectively, during the study period. We performed separate analyses for trauma- and medical-related calls. In our study, we had 34 DGPs available for analysis. DGP-specific RR estimates were calculated as the ratio of the observed and expected counts for each area.

Data Analysis

Within the Bayesian framework of analysis, we used the conditional autoregressive (CAR) model to analyze the data. The CAR model is commonly used to study the associations between diseases and risk factors/covariates at a geographic level.16 This model has also been used previously to study out-of-hospital cardiac arrests in Singapore.17 When data are not independent, ignoring such correlation can lead to biased and inefficient inferences.18 Originally suggested by Besag et al.,19 in the context of image analysis, the CAR model (also referred to as the Besag, York, and Mollie [BYM] model) allows for the smoothing of RR estimates in each region toward the mean risk in the neighboring areas, and this gives a more precise and reliable estimate of both mean and variance compared to the crude rate.20

The formulation of the CAR model used in our analyses is

  • image(m1)

where Oi and Ei are the observed and expected ambulance calls for the ith DGP, ui is a spatially structured random effect that is assigned a CAR prior distribution, vi is a spatially unstructured random effect, and β is the coefficient for the corresponding covariate xi that we included in the analysis (e.g., proportion aged 65 and above). As for the secondary analysis of estimating the RR of ambulance calls for all areas, we used a modified version of Equation (1), this time excluding the covariate xi.

Estimation of the risk in any area is conditional on risks in neighboring areas. The weights for the adjacency are given as 1 for each pair of adjacent DGPs and 0 otherwise. In other words, all neighbors provide equal weights. For the purposes of our analysis, we used the Queen method of adjacency assignment, although other methods are also available.20 In other words, DGPs that shared a common boundary or vertex were considered as neighbors in our analysis. Our previous paper has shown the use of adjacency-based methods of neighborhood assignment to be adequate, when the aim of the study is to examine the relationship of risk estimates with other covariates.20 Out of the 55 DGPs in Singapore, we had 34 DGPs with available census data (the rest being either water catchment areas or islands without people living in them) included in our analysis. The mean number of neighbors for each DGP was four, with a range of one to seven.

The priors for the means were set to a normal distribution, with standard deviation (SD) set to cover a wide range of values, whereas the priors for the SDs of the precision estimates were set to a uniform distribution,21 with a wide yet plausible interval (i.e., range from 0.01 to 20). This range was selected from initial exploratory analysis of the data.

For the Bayesian analysis, we discarded the first 20,000 samples as burn-in (i.e., samples excluded as they were not stable estimates) and ran a further 50,000 iterations, from which we used every other observation (thinning), as we noticed from the initial analysis that the coefficients were highly correlated. We ran two different chains, starting from diverse initial values, and convergence was assessed using the Gelman-Rubin convergence statistic, as modified by Brooks and Gelman.22 We used the deviance information criterion (DIC) developed by Spiegelhalter et al.23 to assess the complexity and fit of the models. Generally, smaller values of DIC are preferred. Data analysis was performed using WinBUGS (version 1.4.1, Imperial College and Medical Research Council, London, UK), and maps were created in ArcMap version 9.0 (Esri, Redlands, CA).

RESULTS

  1. Top of page
  2. Abstract
  3. METHODS
  4. RESULTS
  5. DISCUSSION
  6. LIMITATIONS
  7. CONCLUSIONS
  8. Acknowledgments
  9. References
  10. Supporting Information:
  11. Supporting Information

Of the original 31,896 calls in our registry, 10 calls were excluded as they were false calls that were not of an emergency nature or calls that were transferred. We also had to exclude a further 1,556 calls, as the event occurred in an area (Sentosa Island, Tuas, Seletar, etc.) for which there was no census population data available, resulting in a final sample size of 30,330. Seventy-one percent of these calls were medical in nature, while the remaining 29% were trauma-related.

Table 1 shows the summary of demographics by DGP. There are clear geographic variations in the demographic profile at the DGP level. The highest proportion of persons aged 65 years and above can be found in Outram (19%), while the DGP with the lowest percentage of highly educated people is Woodlands (7.7%). The DGP with the highest proportion of household size five and above was Bukit Timah (44.8%).

Table 1.    Descriptive Summary of Sociodemographics by DGP
DGPAged ≥ 65 yrHousehold size ≥ 5High EducationLiving in ≥ 5 RoomsHigh Monthly Personal IncomeHigh Monthly Household IncomeWorkingSenior Official or Professional
  1. All values are percentages.

  2. DGP = Development Guide Plan.

Ang Mo Kio8.025.515.524.211.529.557.721.2
Bedok8.13113.845.217.336.557.227.6
Bishan6.832.712.352.823.351.761.334.1
Bukit Batok4.929.89.933.214.035.461.625.5
Bukit Merah12.819.816.920.31028.657.119.7
Bukit Panjang5.035.19.345.511.635.260.921.4
Bukit Timah8.244.811.892.953.372.860.166
Changi7.9317.876.410.427.362.322.1
Choa Chu Kang4.034.8946.410.537.963.220
Clementi7.624.117.526.416.436.659.127.9
Downtown core15.320.621.518.18.624.45620.7
Geylang10.626.312.430.610.731.155.722.4
Hougang6.534.7124212.335.159.422.8
Jurong East5.832.213.542.410.835.359.621.8
Jurong West4.433.210.736.67.231.661.516.3
Kallang12.821.913.625.410.426.556.321.9
Marine Parade11.427.914.261.130.244.355.843.0
Newton10.634.310.898.754.967.856.970.2
Novena11.429.614.658.627.847.059.141.3
Orchard10.634.310.898.754.967.856.970.2
Outram1914.620.82.86.818.154.816.7
Pasir Ris437.67.965.818.349.764.928.8
Punggol530.210.358.38.239.262.618.9
Queenstown1217.814.42413.629.756.424.9
River Valley1031.91510049.559.659.667.2
Rochor16.823.518.926.38.426.356.721.8
Sembawang4.927.711.355.69.331.661.920.9
Sengkang530.210.358.38.233.662.618.9
Serangoon7.135.711.955.723.148.360.733
Tampines5.134.79.540.611.936.46020.4
Tanglin9.637.29.997.960.871.659.873.1
Toa Payoh11.121.11429.111.228.556.321.9
Woodlands4.232.27.743.76.927.462.217
Yishun4.831.710.2268.329.159.419.7

Table 1 also describes the geographical variation in the socioeconomic indicators at the DGP level. The highest proportion living in five-room flats and larger was in River Valley (100%), whereas the largest proportion of people earning a monthly personal income of S$5,000 and above was in Tanglin (60.8%). On the other hand, Outram had the lowest proportion of people working (54.8%), and Jurong West had the lowest proportion of people working as senior officials and professionals (16.3%).

The smoothed RR of medical-related ambulance calls is shown in Table 2. The top three DGPs with the highest risk of medical-related ambulance calls were Changi (RR = 29.0, 95% credible interval [CrI] = 24.0 to 34.6), downtown core (RR = 7.8, 95% CrI = 6.5 to 9.2), and Orchard (RR = 4.7, 95% CrI = 3.9 to 5.6). At the other end of the scale, Punggol (RR = 0.3, 95% CrI = 0.27 to 0.41), Pasir Ris (RR = 0.6, 95% CrI = 0.5 to 0.7), and Bukit Timah (RR = 0.6, 95% CrI = 0.5 to 0.7) had the lowest risk of medical-related ambulance calls. Hence there is geographical correlation in the ambulance calls, with calls more likely to originate from the southern region of Singapore, as well as the eastern region of Changi.

Table 2.    Smoothed RR of Medical- and Trauma-related Ambulance Calls by DGP
DGPMedical-related Ambulance CallsTrauma-related Ambulance Calls
RR95% CrIRR95% CrI
  1. CrI = credible intervals; DGP = Development Guide Plan; RR = relative risk.

Ang Mo Kio1.100.96–1.250.880.73–1.04
Bedok0.810.71–0.920.800.67–0.94
Bishan0.840.72–0.970.890.72–1.08
Bukit Batok0.760.66–0.880.600.49–0.73
Bukit Merah1.581.38–1.791.201.01–1.43
Bukit Panjang0.690.59–0.800.720.58–0.88
Bukit Timah0.600.50–0.711.110.90–1.36
Changi28.9724.03–34.6438.0629.50–47.92
Choa Chu Kang0.690.59–0.790.550.45–0.68
Clementi0.930.80–1.080.920.75–1.11
Downtown core7.776.49–9.2118.3414.93–22.21
Geylang1.471.28–1.671.651.38–1.95
Hougang0.970.85–1.100.860.72–1.02
Jurong East0.910.79–1.060.820.67–1.00
Jurong West0.860.75–0.980.850.71–1.00
Kallang1.481.29–1.701.411.17–1.69
Marine Parade1.020.87–1.201.871.53–2.27
Newton0.730.48–1.032.121.43–2.98
Novena0.970.81–1.141.451.16–1.78
Orchard4.713.90–5.6414.5311.85–17.60
Outram2.412.05–2.812.401.90–2.97
Pasir Ris0.590.51–0.680.710.57–0.86
Punggol0.330.27–0.410.320.23–0.42
Queenstown1.291.12–1.481.120.92–1.35
River Valley0.920.68–1.201.961.39–2.64
Rochor3.112.65–3.635.674.65–6.83
Sembawang1.271.06–1.491.791.43–2.21
Sengkang1.421.22–1.630.920.73–1.13
Serangoon0.680.59–0.790.460.37–0.57
Tampines0.660.57–0.750.670.56–0.80
Tanglin0.660.49–0.861.871.38–2.43
Toa Payoh1.221.07–1.400.930.77–1.12
Woodlands1.060.93–1.210.890.74–1.05
Yishun1.090.95–1.240.890.74–1.06

Table 2 also indicates the smoothed RR of trauma-related ambulance calls in Singapore. Similar to medical-related calls, the risk was highest in Changi, followed by downtown core and Orchard. The risks were lower in the following areas: Punggol (RR = 0.3, 95% CrI = 0.2 to 0.4), Serangoon (RR = 0.5, 95% CrI = 0.4 to 0.6), and Choa Chu Kang (RR = 0.6, 95% CrI = 0.5 to 0.7).

Table 3 identifies sociodemographic factors associated with the risk of medical-related ambulance calls. Generally, we noticed that the risk decreased with an increase in the socioeconomic status of the region. For instance, the risk of ambulance calls decreased by 31% (95% CrI = 6% to 53%) for every absolute 10% increase in the proportion of senior officers and professionals in the DGP. The risk also decreased by 34% (95% CrI = 21% to 44%) for every 10% increase in the proportion with household income of S$5,000 and above. Travel by car alone and high personal income were the other significant variables identified.

Table 3.    Factors Associated With Risk of Medical- and Trauma-related Ambulance Calls
Factors*RR95% CrIDIC
  1. Note: RRs reported for a 10% point increase.

  2. Analysis involved calls made from residential locations only.

  3. CrI = credible interval; DIC = deviance information criterion; RR = relative risk.

  4. *Definitions to these can be found in Data Supplement S1, available as supporting information in the online version of this paper.

  5. †Statistically significant.

Medical-related ambulance calls
 Aged 65 yr and above0.550.12–4.22341.60
 High education1.680.29–5.47338.19
 Living in five-room or more flat 0.850.66–1.04334.39
 Working0.840.55–1.04335.66
 Senior officials and professionals†0.690.47–0.94342.64
 Traveling by car alone†0.710.57–0.98331.33
 Household income Singapore $5,000 and above†0.660.56–0.79330.03
 Personal income Singapore $5,000 and above†0.670.24–0.95344.02
Trauma-related ambulance calls
 Aged 65 yr and above2.510.88–19.89286.98
 High education1.880.13–10.28288.86
 Living in five-room or more flat0.900.79–1.06287.23
 Working0.630.35–2.72287.08
 Senior officials and professionals0.800.50–1.07290.30
 Household size five people and more†0.630.33–0.90285.98
 Traveling by car alone0.770.59–1.02286.92
 Household income Singapore $5,000 and above†0.760.57–0.91285.35
 Personal income Singapore $5,000 and above0.790.45–1.03290.00

The factors associated with risk of trauma-related ambulance calls are also listed in Table 3. The RR of trauma ambulance calls decreased by a factor of 0.63 (95% CrI = 0.33 to 0.90) for every 10% increase in the proportion of household size five and above. There was also a significant inverse relationship between risk of trauma ambulance calls and household income.

Figure 1 shows the smoothed RR of ambulance calls for medical cases. It appears that there is spatial correlation in calls, originating from the southern and eastern regions of Singapore. A similar pattern was also observed for trauma-related ambulance calls (Figure 2).

image

Figure 1.  Smoothed RR of ambulance calls for medical cases. Demarcations of maps are based on the Urban Redevelopment Authority of Singapore’s Master Plan 2003. Areas that are not colored indicate waterways or estates with no census data available, as well as uninhabited islands. RR = relative risk.

Download figure to PowerPoint

image

Figure 2.  Smoothed RR of ambulance calls for trauma cases. Demarcations of maps are based on the Urban Redevelopment Authority of Singapore’s Master Plan 2003. Areas that are not colored indicate waterways or estates with no census data available, as well as uninhabited islands. RR = relative risk.

Download figure to PowerPoint

In terms of diagnostics for our Bayesian models, convergence was generally seen after about 10,000 iterations for the posterior estimates of the regression coefficients. The samples also provided us with a reasonable Monte Carlo standard error (less than 0.1% of the SD) for the estimates.

DISCUSSION

  1. Top of page
  2. Abstract
  3. METHODS
  4. RESULTS
  5. DISCUSSION
  6. LIMITATIONS
  7. CONCLUSIONS
  8. Acknowledgments
  9. References
  10. Supporting Information:
  11. Supporting Information

This study demonstrates the utility of geospatial analysis to relate population socioeconomic factors with ambulance call volumes. We found regions in Singapore that demonstrate an increased or decreased risk of medical- or trauma-related emergency ambulance calls. This research can serve as a model for analysis of other public health systems.

For medical-related ambulance calls, we observed a number of socioeconomic variables that were associated with the RR of ambulance calls, such as occupation, travel by car alone, and both household and personal income. For trauma-related ambulance calls, the relationship with sociodemographic variables was less obvious. Only household size and household income were found to be significantly associated. Medical-related calls have a stronger association with socioeconomic status, and we believe that this is because most calls are due to chronic diseases such as heart disease and stroke that are known to have a socioeconomic gradient, while the risk factors for trauma calls (including falls and accidents) could be quite different (e.g., demographic and environmental). We postulate that the inverse relationship between household size and trauma calls could be related to the effect of caregivers and family members in the presence of vulnerable elders and young, but more research is needed to study this finding.

This study has also identified regions in Singapore that demonstrate an increased or decreased risk of medical- or trauma-related emergency ambulance calls. Changi was one of the areas shown in our study to have an elevated risk of ambulance calls. With closer examination of the location of calls, we found that a number of these cases actually occurred in Singapore’s international airport, which is located in Changi. Thus the increased cases were attributed to the transient population at the airport. The apparent increase in risk that we observed in the downtown core can also be explained by the large number of offices in that area, in particular Raffles Place. In fact, when we restricted the analysis to residential calls, the risks for Changi and the downtown core diminished to 3.51 and 0.84 for medical-related calls and 3.03 and 1.65 for trauma-related calls, respectively. We have presented the smoothed RR data for the subgroup analysis involving calls from residential addresses in Data Supplement S2. Orchard Road is the other area shown to have a higher risk for both medical- and trauma-related calls, and this can be explained by the high human density, as this area is the key shopping belt in Singapore with many shopping centers.

We hypothesize that socioeconomic status is related to health care help-seeking behavior and utilization of primary health care. In particular, those who are living in socioeconomically deprived areas have been shown to have higher risk of ischemic stroke after myocardial infarction,24 as well as higher risk of admissions for acute coronary syndrome.25 This may be attributed to their diet, lifestyle, limited access to and utilization of primary health care, and other social factors that are more difficult to measure. Although we used a number of socioeconomic variables measured at the DGP level, it is possible to create an overall index of socioeconomic status (e.g., Socio-Economic Indexes for Areas [SEIFA] used in Australia), but currently we do not have such an index in Singapore. Our finding that age, household income, and travel by car were positively associated with ambulance calls is corroborated by a Japanese study26 that looked at the same socioeconomic factors, but which were measured at an individual level.

The literature on relationship between socioeconomic status and ambulance calls has been limited. High-socioeconomic-status neighborhoods have been shown to be associated with shorter out-of-hospital transport intervals for patients with chest pain in one Canadian study,27 while a Japanese survey study has shown several individual-level socioeconomic factors to be associated with a person’s decision to call for an ambulance during nonemergency situations.26 Another study of out-of-hospital cardiac arrest victims found that socioeconomic status influences bystander CPR and survival rates.28 One other paper has looked at the determinants of emergency department (ED) visits by older adults29 and found that need is usually the primary determinant of ED visits in older people, while socioeconomic factors act as predisposing and enabling factors.29 Siler30 used linear regression models to predict demand for public emergency medical vehicles in Los Angeles, but the study failed to account for spatial correlation in the data. A recent paper looking at sudden cardiac arrests across seven North American sites also found that the incidence was lower for poorer neighborhoods, but that study used Poisson models, also with no adjustment for spatial correlation.31 Our study appears to be the first to use spatial models to demonstrate the relationship between geographic variations in ambulance utilization and underlying population socioeconomic factors.

Our findings suggest the need for geographically targeted programs, to educate the public on appropriate use of emergency services as well as to improve outcomes for specific conditions in high-risk communities. One example of a possible intervention was a randomized controlled community trial conducted in 20 pair-matched communities in the United States aiming to improve outcomes for acute myocardial ischemia.32 One community from each pair received an 18-month, multicomponent community education program on the need to activate EMS early for suspected myocardial ischemia. The study demonstrated a significant effect on the use of EMS among patients admitted to the hospital for suspected acute myocardial ischemia.

Geospatial analysis of ambulance demand by communities can also help with ambulance deployment planning.1 The investigators had previously used a simpler analysis to improve ambulance response times.33 We aim to follow up with the results from this study and see if further improvement can be made to our ambulance deployment plan. The results can be combined with other more useful information, such as traffic data, to provide more specific information on where to locate ambulances.

LIMITATIONS

  1. Top of page
  2. Abstract
  3. METHODS
  4. RESULTS
  5. DISCUSSION
  6. LIMITATIONS
  7. CONCLUSIONS
  8. Acknowledgments
  9. References
  10. Supporting Information:
  11. Supporting Information

One major drawback of our study is the ecologic nature of the study design. Associations found at the regional level cannot be extrapolated to the individual level, because of ecologic bias. In our study, the choice of using DGPs was because of practical data availability reasons. While we concede that it may be useful for policy-makers to have the results at a finer level (e.g., postal code level), a major setback was the lack of availability of population and other important covariate data at that level.

Our analysis made use of the “9-9-5” ambulance calls data in Singapore. Private ambulances do not respond to emergency “9-9-5” calls. However, anecdotally, we know the public infrequently calls for private ambulances. We acknowledge that our data may not represent all medical emergencies, as some would use their own transport, as well as private ambulances, to travel to a health care facility. We believe that the number is small and will not affect the generalizability of our results in Singapore. It is also possible that some of the ambulance calls originating from residential locations may have been made when the patient was in someone else’s home, and the socioeconomic information could be different. However, we believe that this may involve only a small number of patients, and we have no reason to believe that there is substantial differential misclassification geographically to affect our results.

CONCLUSIONS

  1. Top of page
  2. Abstract
  3. METHODS
  4. RESULTS
  5. DISCUSSION
  6. LIMITATIONS
  7. CONCLUSIONS
  8. Acknowledgments
  9. References
  10. Supporting Information:
  11. Supporting Information

Ambulance calls in Singapore demonstrate a clear spatial gradient. Our spatial regression models showed that the risk of making such calls decreases for areas with an increased socioeconomic status, measured at the regional level. Our results can help policy-makers target specific populations at risk with focused campaigns as well as more effective ambulance deployment. This study can serve as a model for analysis of other public health systems.

Acknowledgments

  1. Top of page
  2. Abstract
  3. METHODS
  4. RESULTS
  5. DISCUSSION
  6. LIMITATIONS
  7. CONCLUSIONS
  8. Acknowledgments
  9. References
  10. Supporting Information:
  11. Supporting Information

The authors acknowledge Susan Yap and Pek Pin Pin from the Department of Emergency Medicine, Singapore General Hospital.

References

  1. Top of page
  2. Abstract
  3. METHODS
  4. RESULTS
  5. DISCUSSION
  6. LIMITATIONS
  7. CONCLUSIONS
  8. Acknowledgments
  9. References
  10. Supporting Information:
  11. Supporting Information

Supporting Information:

  1. Top of page
  2. Abstract
  3. METHODS
  4. RESULTS
  5. DISCUSSION
  6. LIMITATIONS
  7. CONCLUSIONS
  8. Acknowledgments
  9. References
  10. Supporting Information:
  11. Supporting Information

The following supporting information is available in the online version of this paper:

Data Supplement S1. List of variables and their descriptions (Singapore Census 2000).

Data Supplement S2. Smoothed relative risk of medical- and trauma-related ambulance calls by DGP (calls from residential addresses only).

The document is in PDF format.

Please note: Wiley Periodicals Inc. is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.

Supporting Information

  1. Top of page
  2. Abstract
  3. METHODS
  4. RESULTS
  5. DISCUSSION
  6. LIMITATIONS
  7. CONCLUSIONS
  8. Acknowledgments
  9. References
  10. Supporting Information:
  11. Supporting Information

Data Supplement S1. List of variables and their descriptions (Singapore Census 2000).

Data Supplement S2. Smoothed relative risk of medical- and trauma-related ambulance calls by DGP (calls from residential addresses only).

Please note: Wiley Periodicals Inc. are not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.

FilenameFormatSizeDescription
ACEM_1280_sm_DataSupplementS1.pdf138KSupporting info item
ACEM_1280_sm_DataSupplementS2.pdf31KSupporting info item

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.