You are here

  • Home
  • Archive
  • Volume 31, Issue 9
  • A comparison of actual versus predicted emergency ambulance journey times using generic Geographic Information System software

Article Text

Article menu

Download PDFPDF

Prehospital care
A comparison of actual versus predicted emergency ambulance journey times using generic Geographic Information System software
  1. Peter McMeekin1,
  2. Jo Gray1,
  3. Gary A Ford2,
  4. Jay Duckett3,
  5. Christopher I Price2
  1. 1Institute of Health and Society, Newcastle University, Newcastle Upon Tyne, UK
  2. 2Institute for Ageing and Health (Stroke Research Group), Newcastle University, Newcastle Upon Tyne, UK
  3. 3North East Ambulance Service NHS Foundation Trust, Newcastle Upon Tyne, UK
  1. Correspondence to Dr Peter McMeekin, Institute of Health and Society, Newcastle University, Newcastle Upon Tyne NE2 4AX, UK; peter.mcmeekin{at}newcastle.ac.uk, peter.mcmeekin{at}ncl.ac.uk

Abstract

Study objective The planning of regional emergency medical services is aided by accurate prediction of urgent ambulance journey times, but it is unclear whether it is appropriate to use Geographical Information System (GIS) products designed for general traffic. We examined the accuracy of a commercially available generic GIS package when predicting emergency ambulance journey times under different population and temporal conditions.

Methods We undertook a retrospective cohort study of emergency ambulance admissions to three emergency departments (ED) serving differing population distributions in northeast England (urban/suburban/rural). The transport time from scene to ED for all the highest priority dispatches between 1 October 2009 and 30 September 2010 was compared with predictions made by generic GIS software.

Results For 10 156 emergency ambulance journeys, the mean prediction discrepancy between actual and predicted journey times across all EDs was an underprediction of 1.6 min (SD 4.9). Underprediction was statistically significant at all population densities, but unlikely to be of clinical significance. Ambulances in urban areas were able to exceed general traffic speed, whereas, the opposite effect was seen in suburban and rural road networks. There were minor effects due to travel outside the busiest traffic times (mean overprediction 0.8 min) and during winter months (mean underprediction 0.4 min).

Conclusions It is reasonable to estimate emergency ambulance journey times using generic GIS software, but in order to avoid insufficient regional ambulance provision it would be necessary to make small adjustments because of the tendency towards systematic underprediction.

  • emergency ambulance systems
  • planning

https://doi.org/10.1136/emermed-2012-202246

Statistics from Altmetric.com

    Request Permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

    Introduction

    Background

    During the planning of Emergency Medical Services (EMS), it is very important to consider the time taken during transfer of patients to a regional centre, rather than a local hospital, for urgent services, such as cardiac, stroke, trauma and critical care.1–5 It is inevitable that some patients will travel a longer distance to receive treatment, but it is anticipated that the expertise and efficiency of the care provided at the regional site will offset delays and be more cost effective at a population level.6–10 The development of new clinical care pathways involving prehospital redirection should include accurate prediction of the impact of EMS centralisation on ambulance transportation times and resources. Geographic Information Systems (GIS) are software tools that enable researchers to manipulate and visualise locality-specific data to support organisational decision making based upon regional characteristics.11 ,12 GIS has been used extensively to describe emergency ambulance activity13 and access to existing health services.14 During the planning of a large-scale healthcare centralisation, activity estimates should reflect the local population distribution and influences, such as time of day and season, but such refinements are poorly developed.15–17

    Importance

    As high-priority calls contribute more than a quarter of ambulance service activity, the impact of additional journey time to specialist centres could be significant at a regional level.18 ,19 The availability of user-friendly commercial GIS software increases the likelihood that the transport implications of different EMS configurations can be more easily modelled by healthcare providers, but it is unclear whether these generic systems can accurately describe emergency ambulance journeys instead of general traffic. Given these uncertainties, it is necessary to validate whether the estimates produced by a generic GIS package are sufficiently robust to inform regional healthcare reconfiguration decisions which may affect a large number of patients.

    Goals of investigation

    We compared predictions by non-specialised GIS software with the actual times of emergency ambulance journeys between incident location, and three admitting emergency departments (ED) in northeast England, and considered the effects of population density and temporal activity on GIS accuracy.

    Methods

    Study design and setting

    We undertook a retrospective cohort study of emergency ambulance calls in northeast England. The North East Ambulance Service (NEAS) National Health Service (NHS) Foundation Trust is the sole provider of paramedic and emergency transport services for patients in the region. From the 10 ED sites within this area, we selected three neighbouring sites (A, B and C) with differing population distributions providing healthcare for a total population of 550 000 (table 1).20

    View this table:
    Table 1

    Site populations served 201020

    At any time of day, unselected patients are admitted to each medically staffed ED without redirection between sites. The destination site chosen by paramedics reflects the nearest ED according to their local knowledge at the time of each incident.

    Selection of participants

    Patient transport details of all highest priority (‘Category A’) emergency calls between 1 October 2009 and 30 September 2010 were extracted from the NEAS Computer Aided Dispatch (CAD) database for incidents occurring within the usual catchment area of the three sites.

    Methods and measurements

    For each emergency ambulance journey, we retrieved the date and time, destination ED and the journey time from scene to ED. Commercially available generic GIS software was used to estimate journey time over the ‘quickest’ rather than ‘shortest’ route based upon the commonly used default settings for the average speed of travel on a particular road type (motorways 70 mph/113 kmh, dual carriageways 62 mph/100 kmh, A roads 40 mph/65 kmh, restricted A roads 34 mph/55 kmh and urban road 19 mph/30 kmh). A journey was excluded from the analysis if the difference between the actual and predicted time was more than 30 min, or if the predicted journey time was 10 times greater or smaller than the observed journey time, with the assumption that there was an error in the recording of the journey details, or extraordinary conditions which would not normally be considered during EMS planning.

    Outcomes

    The outcome of interest was ‘prediction error’ that is, the difference between predicted and actual observed journey times. ‘Prediction’ was defined as the GIS estimate; ‘underprediction’ as when actual exceeded predicted; and ‘overprediction’ as when actual was less than predicted.

    Analysis

    To explore overprediction and underprediction, multiple linear regression was used to model the relationship between the prediction error of journey times using GIS software and the following independent variables: distance travelled from incident to ED site (in order to capture any systematic underprediction of journey speed) and binary variables representing population density, the destination site, whether the journey occurred outside of office hours (OH) during a period of reduced traffic volume (20:00–8:00, Monday to Friday, or any time at weekends), and whether a journey occurred during the first quarter of the calendar year (Q1) to reflect the poorest seasonal road conditions. Interaction terms were also included in the model to determine whether prediction errors were associated with particular sites at particular times of day with differing traffic volumes, season or distance travelled. Intercept coefficients for site B and C destination variables represented the degree of prediction error for each site relative to site A. Slope coefficients for sites B and C destination variables showed by how much the rate of change in prediction error as a function of distance travelled differed for each site relative to site A. The base case model that all other categories and associated coefficients were referenced against was the mean prediction error from travelling to site A between April and December (Q2–Q4, ie, not Q1) during hours with greater traffic volume (Monday–Friday; 8:00–20:00, ie, not OH).

    A backward stepwise regression analysis was performed to identify a subset of independent variables related to prediction error using the statistical package R.21 We would reject the hypothesis that predictions made by generic GIS software can be used to accurately estimate ambulance journey times if there was a statistically significant estimate (p<0.05) attached to the coefficients describing site type.

    Results

    Characteristics of the emergency patient transport journeys

    Ten thousand three hundred relevant emergency calls occurred during the 12-month period; 144 (1.4%) were excluded due to excessive discrepancies between predicted and actual journey times. Table 2 shows the characteristics of the remaining 10 156 emergency ambulance journeys included in the analysis: 51.1% were to site B, 40.5% to site A and 8.4% to site C, reflecting the size of each catchment population. Overall, 25.4% of journeys were in Q1, and 34.1% took place outside the busiest traffic hours (ie, at weekends and after 20:00).

    View this table:
    Table 2

    Characteristics of emergency ambulance journeys October 2009 to September 2010

    Main results

    After calculation of journey times between incident and destination site postcodes by GIS software, the mean prediction discrepancy between the actual and predicted journey times across all sites was an underprediction of 1.6 min (SD 4.9). Table 3 shows the results of the best model fit for journey time in order to illustrate local influences. Log transformation did not affect the model output. As the dependent variable is the difference between predicted and actual journey times, a positive sign on a coefficient indicates overprediction, and a negative sign indicates underprediction. All variables were statistically significant at the 5% level. Independent variables which dropped out from the model due to a lack of predictive effect were the interaction terms between OH and both sites B and C, and the interaction between Q1 and site C.

    View this table:
    Table 3

    Results of the generalised linear model regression for Geographical Information System prediction error

    For a hypothetical zero mile journey to an urban hospital (site A) made during busier traffic times outside of Q1, GIS software made an absolute underprediction of journey time by 7.0 min (ie, the intercept value was −7.0). Relative to site A, the software showed much lower degrees of journey time underprediction for site B (−7.0+6.5=−0.5 min) and site C (−7.0+6.5=−0.5 min).

    Holding all other factors constant, journey durations made outside of busier traffic hours were overpredicted for all sites by a mean of 0.8 min. Journeys made during Q1 were associated with an underprediction of journey time (0.4 min), including an additional impact on admissions to site B (0.5 min). These effects were small (<7% of mean regional journey time) and no other site-specific associations were found with the time of day or season when journeys took place.

    The effect of distance travelled on the prediction of journey duration varied between sites. As the distance from incident location to site A increased, the absolute overprediction of journey duration relative to observed grew by 0.9 min per mile, suggesting that ambulances have an advantage over general traffic when negotiating a path through urban road networks. For sites B and C, the opposite effect was found, and longer journeys had increasing underprediction, compared with those observed, that is, on suburban and rural roads, ambulances cannot travel as fast as general traffic. Figure 1 shows the effect of these underpredictions and overpredictions for the median and interquartile ranges (middle 50%) of journey distances to each site.

    Figure 1
    Figure 1

    Prediction of ambulance journey times in the interquartile journey range for each site.

    Using the results from table 3, it is possible to calculate that an unadjusted GIS software model for a journey of 1 mile to site A between 8:00 and 20:00 on a weekday during Q2–Q4, would result in an underprediction of −7.0+0.9=−6.1 min. However, most ambulance journeys are longer than one mile. Table 4 shows the predicted discrepancies and CIs across median journeys in the sample. The distribution of journey distances shows that an unadjusted GIS approach more commonly results in underprediction than overprediction, irrespective of population density.

    View this table:
    Table 4

    Underpredictions of typical emergency ambulance journeys per site

    For all the journeys made to each site during the 12-month period, the gross error of unadjusted GIS estimates was: A: 16 823; B: 16 423 and C: 3525 min (280; 274; 59 h). However, due to the tendency towards underprediction, the net effect from the sum of underprediction and overprediction was an overall underestimation of journey times by: A: −4003; B: −9385 and C: −2816 min (−67; −156; −47 h).

    Limitations

    Our study has some limitations. We only considered the transport times for patients triggering an urgent ambulance dispatch in one region. Some patients may not have needed emergency transportation, but the dispatch decision was according to standard NHS criteria, and would have been independent of the destination hospital. The CAD data reflects paramedic-initiated electronic recording of arrival and departure, and although this is a routine function, inaccurate data capture might explain why generic GIS software calculated different journey times. We did not consider the call to scene, or time on scene by paramedics which would be necessary to examine the impact on service efficiency, and is likely to reduce the overall effect of error due to GIS prediction variability. There may be other patient influences which explain prediction errors, such as the nature and severity of patient illness, although a whole year of ambulance activity was used to reduce random effects. The estimates were predicted by a popular commercially available GIS software package which applied commonly used traffic speeds, but we did not test the accuracy of multiple packages which may vary through the use of different algorithms even if the same road speeds are employed.

    Discussion

    During prehospital redirection, patients may bypass closer hospitals in order to access additional resources further away, creating a delay which should be offset by health and/or economic benefits. Our findings suggest that if healthcare planners use generic GIS software to predict emergency ambulance travel times to new destinations, then ideally, this should be calibrated against local data which reflects population density. However, the impact of prediction error for any typical journey was small and may not have direct clinical consequence apart from at the perimeter of large regional services. If no local data were available, it would be acceptable to use a generic package, but the tendency to underpredict journey times should be recognised due to service planning to avoid slight underprovision of resources, especially during longer rural journeys, very short urban journeys, winter months and peak traffic hours.

    GIS has been extensively used for examining the spatial patterns of health services and access to new facilities.22 It has previously been acknowledged that standard proprietary packages might have limitations in examining different aspects of accessibility if a specific mode of transport is being considered (eg, emergency ambulance), or if wider influences need to be taken into account, such as temporal effects.14 There have been efforts to incorporate a temporal component when considering individual service accessibility based on personal travel diary data,23 ,24 but not in the context of emergency ambulance transportation. Our study is the first to test the accuracy of a non-specialised GIS package when predicting emergency ambulance journey times by comparison with data for individual patients. By estimating the size of these effects, we have provided a method for increasing the accuracy of predicted journey times through adjustment of generic GIS software output.

    The effect of population density as depicted by hospital site location should be considered in the EMS planning process, as it reflects local carriageway sizes, speed limits, the frequency of road junctions and accessibility to incident locations. For very short urban journeys, it would seem that the status of being an ambulance conveys no overall advantage compared with general traffic, and the process of leaving the scene and/or arriving at hospital in urban areas creates more of a delay than in semiurban and rural locations. As urban journey length increases, underprediction quickly reduces, whereas in semiurban and semirural settings, the underprediction effect increases with longer journeys. One interpretation of this observation is that in urban settings, after spending a longer time leaving the scene due to poorer access to properties, the status of being an ambulance confers more benefit in finding a path through traffic than in suburban and rural locations. In the latter, topography and road terrain may be more difficult for ambulances to negotiate at higher than average speeds while retaining patient safety and comfort.25 ,26 Due to a lack of information, we were unable to investigate further how much of the gain during urban travel was specifically due to the use of blue lights and sirens after leaving the scene, but it is likely these were often employed following a high-priority dispatch. Temporal activity factors may also need consideration for strategic use of resources. Journeys outside of normal OH took less time than predicted, probably as a result of lower traffic volumes. However, journeys were longer than predicted in the first quarter of the year, particularly in semiurban areas, which may also result from greater seasonal traffic volumes with the additional effect of poor weather conditions.

    Although the discrepancies between predicted and actual travel time were small, the impact of underprediction may have service provision consequences for patients needing highly time-dependent treatments, such as thrombolysis for ischaemic stroke27 and respiratory support for asthma.28 For any particular condition, the meaning of prediction error would have to be weighed up against the other factors in the specific patient care pathway which could be more significant and/or amenable to intervention, such as delays in assessment after arrival at hospital. A systematic underprediction error might directly impact on regional ambulance availability and, hence, the response time to other emergency calls. Within the region examined, there were 120 000 dispatches expected to arrive on scene in less than 8 min (‘Category A’) during 2010, and more than two million nationally.18 Accumulated underestimation of the time spent with patients during transportation to hospital by generic GIS packages during healthcare planning might lead to an overestimation of the capacity of a service to respond effectively with an available vehicle, especially during peak times and winter months.

    In summary, during the planning of emergency healthcare provision, it is reasonable to estimate emergency ambulance journey times using generic GIS software, but to avoid an overall tendency towards systematic underprediction in large cohorts it would be necessary to make small adjustments reflecting local population density and time of travel.

    References

    1. ↵
      1. Antman EM,
      2. Anbe DT,
      3. Armstrong PW,
      4. et al
      . ACC/AHA guidelines for the management of patients with ST-elevation myocardial infarction; a report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines. J Am Coll Cardiol 2004;44:e1–211.
      OpenUrlCrossRefPubMed
    2. ↵
      1. Nichol G,
      2. Aufderheide TP,
      3. Eigel B,
      4. et al
      . Regional systems of care for out-of-hospital cardiac arrest. A policy statement from the American Heart Association. Circulation 2010;121:709–29.
      OpenUrlAbstract/FREE Full Text
    3. ↵
      1. Price CI,
      2. Clement F,
      3. Gray J,
      4. et al
      . Systematic review of stroke thrombolysis service configuration. Exp Rev Neurotherap 2009;9:211–33.
      OpenUrlCrossRef
    4. ↵
      1. Celso B,
      2. Tepas J,
      3. Langland-Orban B,
      4. et al
      . A systematic review and meta-analysis comparing outcome of severely injured patients treated in trauma centers following the establishment of trauma systems. J Trauma 2006;60:371–8.
      OpenUrlCrossRefPubMedWeb of Science
    5. ↵
      1. Meadows CIS,
      2. Rattenberry W,
      3. Waldman C
      . Centralisation of specialist critical care services. JICS 2011;12:87–9.
      OpenUrl
    6. ↵
      1. Spaite DW,
      2. Bobrow BJ,
      3. Vadeboncoeur TF,
      4. et al
      . The impact of prehospital transport interval on survival in out-of-hospital cardiac arrest: implications for regionalization of post-resuscitation care. Resuscitation 2008;79:61–6.
      OpenUrlCrossRefPubMedWeb of Science
    7. ↵
      1. Pearson G,
      2. Barry P,
      3. Timmins C,
      4. et al
      . Changes in the profile of paediatric intensive care associated with centralisation. Intensive Care Med 2001;27:1670–73.
      OpenUrlCrossRefPubMedWeb of Science
    8. ↵
      1. Nathens AB,
      2. Jurkovich GJ,
      3. Maier RV,
      4. et al
      . Relationship between trauma center volume and outcomes. JAMA 2001;285:1164–71.
      OpenUrlCrossRefPubMedWeb of Science
    9. ↵
      1. Moy R,
      2. Han K,
      3. Smith G Dean,
      4. et al
      . Trauma systems: the potential impact of a trauma divert policy on a regional ambulance service. Emerg Med J 2012;29:404–8.
      OpenUrlAbstract/FREE Full Text
    10. ↵
      Nuffield Institute for Health & NHS Centre for Reviews and Dissemination. Hospital volume and health care outcomes, costs and patient access. University of York: Effective Health Care 1996;2.
    11. ↵
      1. Carlson T,
      2. York S,
      3. Primomo J
      . The utilization of geographic information systems to create a site selection strategy to disseminate an older adult fall prevention program. Soc Sci J 2011;48:159–74.
      OpenUrlCrossRef
    12. ↵
      1. O'Sullivan D,
      2. Unwin DJ
      . Geographic information analysis. Hoboken, NJ: JohnWiley and Sons, 2003.
    13. ↵
      1. Peleg K,
      2. Pliskin JS
      . A geographic information system simulation model of EMS: reducing ambulance response time. Am J Emerg Med 2004;22:164–70.
      OpenUrlCrossRefPubMedWeb of Science
    14. ↵
      1. Higgs G
      . A literature review of the use of GIS-based measures of access to health care services. Health Serv Outcomes Res Methodol 2004;5:119–39.
      OpenUrlCrossRef
    15. ↵
      1. Blanchard IE,
      2. Doig CJ,
      3. Hagel BE,
      4. et al
      . Emergency medical services response time and mortality in an urban setting. Prehosp Emerg Care 2012;16:142–51.
      OpenUrlCrossRefPubMedWeb of Science
    16. ↵
      1. Miller HJ,
      2. Wu YH
      . GIS software for measuring space-time accessibility in transportation planning and analysis. GeoInformatica 2000;4:141–59.
      OpenUrlCrossRef
    17. ↵
      1. McLafferty SL
      . GIS and health care. Annual Review of Public Health 2003;24:25–42.
      OpenUrlCrossRefPubMedWeb of Science
    18. ↵
      Ambulance Services in England. The NHS Information Centre, Workforce and Facilities. June 2010. http://www.ic.nhs.uk/ (accessed 22 Nov 2012).
    19. ↵
      1. Pitts SR,
      2. Niska RW,
      3. Xu J,
      4. et al
      . National Hospital Ambulatory Medical Care Survey: 2006 emergency department summary. National health statistics reports; no 7. Hyattsville, MD: National Center for Health Statistics, 2008.
    20. ↵
      England Local Authorities: Population, Land Area & Density. http://www.demographia.com/db-engla.htm (accessed 22 Nov 2012).
    21. ↵
      R Core Team (2012). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, http://www.R-project.org/
    22. ↵
      1. Gatrell A,
      2. Senior M
      . Health and health care applications. In: Longley P, Maguire D, Rhind D, eds. Geographical information systems: management issues and applications. 2nd edn. New York: Wiley, 1999:925–38.
    23. ↵
      1. Weber J,
      2. Kwan MP
      . Bringing time back in: a study on the influence of travel time variations and facility opening hours on individual accessibility. Prof Geographer 2002;54:226–40.
      OpenUrlCrossRefWeb of Science
    24. ↵
      1. Weber J,
      2. Kwan MP
      . Evaluating the effects of geographic contexts on individual accessibility: a multilevel approach. Urban Geography 2003;24:647–71.
      OpenUrlCrossRefWeb of Science
    25. ↵
      1. Lin G,
      2. Allan DE,
      3. Penning MJ
      . Examining distance effects on hospitalization using GIS: a study of three health regions in British Columbia, Canada. Environ Plann 2003;34:2037–53.
      OpenUrl
    26. ↵
      1. Schuurman N
      . Defining rational hospital catchments for non-urban areas based on travel time. Int J Health Geogr 2006;5:43–7.
      OpenUrlCrossRefPubMed
    27. ↵
      1. Wardlaw JM,
      2. Murray V,
      3. Berge E,
      4. et al
      . Thrombolysis for acute ischaemic stroke. Cochrane Database Syst Rev 2009;(4):CD000213.
    28. ↵
      1. Jones AP,
      2. Bentham G
      . Health service accessibility and deaths from asthma in 401 local authority districts in England and Wales, 1988–92. Thorax 1997; 52:218–22.
      OpenUrlAbstract

    Footnotes

    • Contributors PM, CP and JD conceived the research; PM and JG were responsible for the analyses; PM, CP, JG and GF interpreted the data prepared the draft and contributed to the final revision.

    • Funding This research was funded by an NIHR Programme Grant for Applied Research (RP-PG-0606-1241). The views and opinions expressed here are those of the authors and do not necessarily reflect those of the Department of Health.

    • Competing interests None.

    • Provenance and peer review Not commissioned; externally peer reviewed.

    • Data sharing statement Any agreement to further use the data would be with North East Ambulance Service, although the researchers only obtained the data reported in the study.