Prehospital Emergency Care
ISSN: 1090-3127 (Print) 1545-0066 (Online) Journal homepage: http://www.tandfonline.com/loi/ipec20
Predictors of Nontransport of Older Fallers Who Receive Ambulance Care Paul M. Simpson MScM (ClinEpi), BEd, BHSc, Jason C. Bendall MM (ClinEpi), MBBS, PhD, Barbara Toson BStatEc, Anne Tiedemann BSc, PhD, Stephen R. Lord PhD & Jacqueline C. T. Close MD To cite this article: Paul M. Simpson MScM (ClinEpi), BEd, BHSc, Jason C. Bendall MM (ClinEpi), MBBS, PhD, Barbara Toson BStatEc, Anne Tiedemann BSc, PhD, Stephen R. Lord PhD & Jacqueline C. T. Close MD (2014) Predictors of Nontransport of Older Fallers Who Receive Ambulance Care, Prehospital Emergency Care, 18:3, 342-349 To link to this article: http://dx.doi.org/10.3109/10903127.2013.864355
Published online: 24 Jan 2014.
Submit your article to this journal
Article views: 234
View related articles
View Crossmark data
Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=ipec20 Download by: [Chinese University of Hong Kong]
Date: 05 November 2015, At: 18:51
PREDICTORS OF NONTRANSPORT OF OLDER FALLERS WHO RECEIVE AMBULANCE CARE Paul M. Simpson, MScM (ClinEpi), BEd, BHSc, Jason C. Bendall, MM (ClinEpi), MBBS, PhD, Barbara Toson, BStatEc, Anne Tiedemann, BSc, PhD, Stephen R. Lord, PhD, Jacqueline C. T. Close, MD
Downloaded by [Chinese University of Hong Kong] at 18:51 05 November 2015
ABSTRACT
included response nonurgent response priority, younger age, and the presence of a personal alarm. The AUC was 0.68 (95% CI 0.64–0.71; p < 0.0001) (imputed data AUC 0.69 (95% CI 0.66–0.72)). Conclusion. In this population of confirmed older fallers attended to by paramedics, determination of the prehospital transport outcome is greatly influenced by on-scene findings resulting from paramedic assessment. The presence of new pain, abnormal physiology, and altered function post-fall were strongly associated with increased odds of transport. Conversely the presence of a personal alarm and allocation of a nonurgent dispatch priority increased the odds of nontransport. Accurate discrimination between older fallers who were and were not transported using dispatch data only was not possible. Key words: ambulance; emergencies; aged; accidental falls
Objectives. To identify patient, clinical, and operational factors associated with nontransport of older people who have fallen and received ambulance care; and to develop a nontransport prediction tool that could be utilized during the dispatch process to rationalize allocation of emergency ambulance resources. Methods. The study was a planned subanalysis using data collected during a prospective observational cohort study of nonconsecutive emergency responses to older people aged 65 years or more who had fallen between October 1, 2010 and June 30, 2011. The data consisted of routinely collected ambulance dispatch and clinical records, combined with prospectively collected fall-specific information. Missing data were managed using multiple imputation. Multivariate logistic regression modeling was undertaken to identify predictors of nontransport. Results are described for original and imputated data sets, presented as odds ratios (OR) with 95%CI (confidence interval). Receiver operating curve (ROC) statistics were generated, with model discrimination determined by the area under the curve (AUC). Results. There were 1,484 cases eligible for this subanalysis of which 419 (28.2%) were recorded as nontransport. Multivariate regression including dispatch and clinical variables identified a 6-item final model. Younger age group, nonurgent response priority, and presence of a personal alarm were predictors of nontransport, along with clinical variables, including normal vital signs, absence of injury, and unchanged functional status post-fall. The AUC was 0.88 (95% CI 0.86–0.90; p < 0.0001) (imputed data AUC 0.86 (95% CI 0.84–0.88)). Multivariate modeling of dispatch variables only identified a 3-item final model, which
PREHOSPITAL EMERGENCY CARE 2014;18:342–349
INTRODUCTION Provision of emergency responses to older people who have fallen is a core function of ambulance services. While the majority of older people are transported to an emergency department (ED) following assessment by paramedics, between one-quarter and onethird are not transported after ambulance care.1 Evidence from the United Kingdom (UK), United States (US), and Australia suggests that nontransport of this older population may result in an increased risk of adverse patient outcomes and high rates of ambulance reattendance.2–4 Equally, there is insufficient evidence to indicate whether these people would do better if transported to an ED or if they received alternate models of care.5,6 While the transport decision-making process of paramedics has been previously described,7 external factors associated with nontransport of older fallers following paramedic attendance have not been explored. Identification of patient, clinical, or operational factors associated with nontransport could reveal aspects of ambulance management of older fallers that will elucidate this process that impacts significantly on patient outcomes and ambulance resource utilisation. A better understanding of factors associated with nontransport also provides an opportunity to explore alternate models of care and possibly the development of a nontransport dispatch prediction tool that could be used to rationalize ambulance resource allocation and optimize patient safety. Dispatch prediction rules
Received July 11, 2013 from the University of Western Sydney, Sydney, Australia (PMS), Ambulance Service of New South Wales, Sydney, Australia (PMS), Department of Anaesthesia, Gosford Hospital, Gosford, Australia (JCB), Neuroscience Research Australia (BT, SRT, JCTC), School of Public Health and Community Medicine, University of New South Wales, Sydney, Australia (PMS, SRL), The George Institute for Global Health, Sydney, Australia (AT), Sydney Medical School, The University of Sydney, Sydney, Australia (AT), Prince of Wales Clinical School, University of New South Wales, Sydney, Australia (JCTC). Revision received September 9, 2013; accepted for publication October 7, 2013. The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper. Address correspondence to Mr. Paul Simpson, University of Western Sydney, Locked Bag 1797, Penrith NSW 2751, Australia. e-mail: [email protected] doi: 10.3109/10903127.2013.864355
342
Downloaded by [Chinese University of Hong Kong] at 18:51 05 November 2015
P. M. Simpson et al.
343
Falls Nontransport Prediction
have been previously explored in other contexts. For example, Isenberg et al. developed a simple dispatch prediction tool to reduce unnecessary urgent responses to minor car accidents.8 The ability to predict nontransportation of older fallers could offer several benefits for ambulance services. First, dispatching an appropriate resource to a case likely to result in nontransport might streamline patient care and minimize the clinical risk. Second, with most ambulance services facing unprecedented emergency call demands, a predictive tool with good discrimination could enable dispatch of, for example, a single appropriately skilled paramedic responder, leaving a double-crewed transport ambulance available to respond to cases where transport is more likely. With this context in mind, the two key aims of this study were to 1) identify patient, clinical, and operational factors associated with nontransport of older people who have fallen and received ambulance care; and 2) develop a nontransport prediction tool that could be utilized during the dispatch process to rationalize allocation of emergency ambulance resources.
METHODS
ple of which 1.02 million (14%) are aged 65 years or more. NSW is serviced by a single, state-based government ambulance service that has approximately 3,500 paramedics within three clinical levels: paramedic interns (first to third year of training); qualified paramedics (have completed 3 years of core training); and paramedic specialists (either intensive care paramedic or extended care paramedic). Training is provided internally using a vocational training model, but with an increasing number of university-trained paramedics. Paramedic management of cases involving falls experienced by older people is guided by a “falls in the elderly” protocol. Paramedics can pursue mutually agreed upon nontransport “clinical pathways” where strictly defined criteria are met following detailed history taking and patient assessment, although any patient requesting transport following discussion must be transported to ED. Dispatch of ambulance resources is managed by a computer-aided dispatch (CAD) system, in which incoming calls are categorized and prioritized using the Medical Priority Dispatch System (MPDS) and ProQA software (Priority Dispatch Corporation, Salt Lake City, Utah, US).
Study Design
Data Sources
This study was a planned subanalysis of a cohort of patients extracted from a larger dataset that had been collected during a prospective observational study investigating prehospital management of older people who had fallen and received an emergency response.9 The larger epidemiological study used a prospective, convenience-sampling (nonconsecutive enrollment) design to collect fall-specific data over a 9-month period from 1 October, 2010 through 30 June, 2011. During that period, paramedics who had volunteered to participate in the study prospectively enrolled people aged 65 years and older who were confirmed after arriving on scene to have had a fall, and collected fallspecific data, which was subsequently linked to routinely collected clinical record and dispatch data.
The data used in this study were drawn from three sources. Data relating specifically to falls and patient demography were collected prospectively using a study-specific data sheet (Figure 1) by paramedics at the point of care. The other two data sources were clinical and dispatch data, retrieved from routinely collected administrative databases. The clinical data consists of the patient health-care clinical record that is completed by paramedics for every incident attended. The dispatch data, sourced from the CAD database, includes dispatch priority details and operational times for each phase of ambulance response and care. Selection of the variables and their relevance to the outcome of nontransport was based on existing literature in this area and the consensus of an advisory group working in the fields of prehospital care, falls and balance, and aged care (Tables 1 and 2).
Population The study sample comprised people aged 65 years or older who had received an emergency ambulance response and were confirmed by paramedics after arriving at the scene to have had a fall using a standard definition.10
Study Setting The study was conducted in New South Wales (NSW), Australia, an area of 800,000 square kilometers containing metropolitan and rural/remote regions. NSW has a population of approximately 7.13 million peo-
Statistical Analysis Descriptive statistical analysis and univariate logistic analysis were undertaken for the variables of interest. In order to maximize the number of cases available for inclusion, missing values found not to be missing completely at random were managed using multiple imputation (MI) methods incorporated within the SPSS missing data analysis module (IBM SPSS Statistics for Windows, Version 21.0, Armonk, NY). The MI process involved 10 imputations of data so as to maximize relative efficiency of the results arising from the logistic
PREHOSPITAL EMERGENCY CARE
JULY/SEPTEMBER 2014
VOLUME 18 / NUMBER 3
Downloaded by [Chinese University of Hong Kong] at 18:51 05 November 2015
344
FIGURE 1. Fall information data collection tool.
regression modelling process.11 The decision to use MI was undertaken in accordance with established conventions regarding handling of missing data.12–14 Two independent multivariate logistic regression analyses were then undertaken. The dependent variable for both was prehospital transport outcome, a dichotomous variable consisting of “transport” or “no transport” as recorded by the computerized emergency dispatch system, which tracks each phase of an emergency case. Only predictors that satisfied the inclusion criteria of p ≤ 0.25 in the univariate analysis
were included in the multivariate logistic regression models. The first “all variables” analysis was an exploratory analysis conducted to identify demographic, clinical, and operational factors predictive of nontransport of older people following a fall. Included in this were the 20 variables derived from information gathered during the initial emergency phone call and the ambulance dispatch process and by paramedics on scene after conducting an assessment of the patient (Table 1). The resulting “all variables model” (AVM) was also
P. M. Simpson et al.
TABLE 1. Predictor variables subjected to univariate logistic regression for “all variables” and “dispatch variables” analyses “All variables” regression analysis
Downloaded by [Chinese University of Hong Kong] at 18:51 05 November 2015
345
Falls Nontransport Prediction
Patient gender Patient age group Response priority Time of day Time of week Personal alarm present Residential status Fall location Operational region Presence of carer Non-English-speaking background (NESB) On floor at ambulance arrival New injury/pain Initial presenting physiology Fall in past 12 months Patient of anticoagulants Number of prescription medications Use of mobility/walking aid Change in usual function post fall Highest paramedic clinical level on scene
“Dispatch variables” regression analysis
Patient gender Patient age group Response priority Time of day Time of week Personal alarm present Residential status Fall location Operational region Presence of carer
deemed to constitute an optimal predictive model for nontransport of older people following a fall, against which a second regression model would be compared. The second “dispatch variables” analysis was conducted to develop a predictive “dispatch variables model” (DVM) from which a prediction rule might be derived that could potentially be applied during the dispatch process to guide emergency resource allocation. As such, this analysis consisted only of those variables that could be derived from information routinely gathered during the existing dispatch process, or from information that could potentially be reliably and accurately gathered by adding additional questions during this process (Table 1). A stepwise backward elimination process was undertaken in both regression models. At each step the least significant variable was removed until only variables significant at p < 0.05 remained. Interactions between remaining variables were tested, and retained in the model if significant at p < 0.01. The final multivariate model was assessed for goodness-of-fit using the Hosmer and Lemeshow method with p ≥ 0.05 considered an acceptable fit.15 Multivariate regression analysis results are presented for both the complete and imputed datasets, expressed as odds ratios (OR) with 95% confidence intervals (CI). Receiver operating curve (ROC) statistics were generated to determine the area under the curve (AUC) for both regression models using the complete data set, in order to compare the potential performance of the “dispatch variables” model with the “all variables” optimal model. An AUC of 0.7 for the DVM was selected
as the minimum value for model discrimination that would allow creation of a dispatch prediction rule. The chi-squared test was used to determine statistical difference in AUC for the two ROC curves, with significance established at p = 0.05.
Ethical Approval Ethical approval was granted by the Sydney Local Area District (SLHD) Human Research Ethics Committee (HREC), Royal Prince Alfred Zone (Protocol No X10-0152 & HREC/10/RPAH/282).
RESULTS There were 1,610 cases identified prospectively by paramedics as having fallen during the study period. Of these, 55 were excluded due to not being emergency response cases (i.e., booked medical responses, routine ambulance transports), and another 71 were excluded due to not having a transport outcome (i.e., transport or nontransport) recorded. There were 1,484 eligible cases forming the complete data set for analysis of which 419 (28.2%) were recorded as nontransport. The missing data analysis identified several variables of interest with between 5 and 15% missing data, resulting in the list-wise loss of approximately one-third of cases in the regression analysis. Table 2 shows the results of the univariate “all variables” analysis, while Table 3 shows the results of the multivariate logistic regression, which led to a 6item final predictive model. Younger age, lower response priority, the presence of an alarm device, the absence of new injury or pain or change in usual level of function, and normal physiology were all associated with increased odds of a nontransport outcome. The goodness-of-fit test indicated good model fit (8 DF, χ 2 = 7.43, p = 0.49). For the “dispatch variables”, 10 variables underwent univariate analysis (Table 1). The multivariate logistic regression analysis identified a 3-item final model, which included response priority, age, and the presence of a personal alarm (Table 4). The goodness-offit test indicated good model fit (7DF, χ 2 = 10.01, p = 0.19). The AUC for the optimal “all variables” prediction model was 0.88 (95% CI 0.86–0.90; p < 0.0001) (imputed data AUC 0.86 (95% CI 0.84–0.88)), demonstrating excellent model discrimination (Figure 2). Using a probability cutoff of 50%, the sensitivity and specificity of this model were 59 and 91%, respectively, with a false-positive rate of 31% and a false-negative rate of 14%. However, for the “dispatch variables” model, the AUC of 0.68 (95% CI 0.64–0.71; p < 0.0001) (imputed data AUC 0.69 (95% CI 0.66–0.72)) demonstrated at best only moderate discrimination (Figure 2). The sensitivity and specificity of this model were 27.2 and
346
PREHOSPITAL EMERGENCY CARE
JULY/SEPTEMBER 2014
VOLUME 18 / NUMBER 3
TABLE 2. Univariate logistic regression analysis of all variables modeling unadjusted odds of nontransport for people aged ≥65 who have fallen and received an emergency ambulance response (complete data set n = 1,484)
Downloaded by [Chinese University of Hong Kong] at 18:51 05 November 2015
Variable
NESB Yes (referent) No Patient gender Male (referent) Female Patient age 65–74 years (referent) 75–84 years 85+ years Response priority Urgent (referent) Nonurgent (within 60 min) Nonurgent (within 90 min) On floor at ambulance arrival Yes No Time of day Dayshift (0700–1859) (referent) Nightshift (1900–0659) Time of week Weekday (referent) Weekend Initial patient physiology Normal (referent) Abnormal New injury/pain Yes (referent) No Personal alarm present Yes (referent) No Fall in past 12 months Yes (referent) No Patient on anticoagulant medication Yes (referent) No Number of prescription medications 0–3 (referent) 4–7 8+ Residential status of patient Community dwelling (referent) RACF Use of a mobility/walking aid No (referent) Yes Change in usual function post fall No (referent) Yes Carer status Lives with carer (referent) Lives alone Location of fall Place of residence (referent) Public area Geographic area Metropolitan Regional/rural Highest paramedic clinical level on scene Paramedic intern (referent) Qualified paramedic Paramedic specialist
Transport N (%) (n = 1,065)
Nontransport N (%) (n = 419)
Univariate OR (95% CI)
132 (78.1) 833 (70.2)
37 (21.9) 353 (29.8)
1 1.51 (1.03–2.22)
0.03
381 (69.7) 678 (73.1)
166 (30.4) 249 (26.9)
1 0.84 (0.67–1.06)
0.15
155 (66.8) 372 (69.3) 484 (75.6)
77 (33.2) 165 (30.7) 156 (24.4)
1 0.89 (0.64–1.24) 0.65 (0.47–0.90)
0.01
688 (80.3) 354 (67.2) 23 (23.0)
169 (19.7) 173 (32.8) 77 (77.0)
1 2.0 (1.56–2.55) 13.63 (8.31–22.36)
< 0.0001
508 (64.5) 517 (80.2)
280 (35.5) 128 (19.8)
1 0.45 (0.35–0.57)
< 0.0001
826 (73.2) 239 (67.3)
303 (26.9) 116 (32.7)
1 1.32 (1.02–1.71)
0.03
774 (70.7) 291 (74.8)
321 (29.3) 98 (25.2)
1 0.81 (0.62–1.06)
0.12
527 (65.8) 471 (81.8)
274 (34.2) 105 (18.2)
1 0.43 (0.33–0.56)
< 0.0001
898 (83.9) 126 (34.2)
172 (16.1) 243 (65.9)
1 10.07 (7.69–13.19)
< 0.0001
185 (57.1) 830 (75.6)
132 (42.9) 268 (24.4)
1 0.43 (0.33–0.56)
< 0.0001
494 (68) 474 (75.2)
232 (32) 156 (24.8)
1 0.70 (0.55–0.89)
0.04
349 (71.7) 622 (71.4)
128 (28.3) 249 (28.6)
1 1.01 (0.79–1.30)
0.92
292 (70.7) 433 (72.7) 224 (70.2)
121 (29.3) 163 (27.4) 95 (29.8)
1 0.9 (0.69–1.20) 1.02 (0.74–1.40)
0.68
878 (70) 172 (81.5)
377 (30) 39 (18.5)
1 0.53 (0.37–0.74)
0.0007
440 (79.9) 546 (65.6)
111 (20.2) 287 (34.5)
1 2.08 (1.62–2.68)
< 0.0001
356 (53.2) 547 (96.6)
313 (46.8) 19 (3.4)
1 0.04 (0.02–0.06)
< 0.0001
373 (72.9) 554 (69)
139 (27.2) 249 (31)
1 1.21 (0.94–1.54)
0.14
817 (69.6) 197 (78.5)
357 (30.4) 54 (21.5)
1 0.63 (0.45–0.87)
0.01
642 (72.5) 423 (70.6)
243 (27.5) 176 (29.4)
1 1.1 (0.87–1.38)
0.42
195 (77.4) 250 (71.6) 226 (77.4)
57 (22.6) 99 (28.4) 66 (22.6)
1 1.36 (0.93–1.97) 0.99 (0.67–1.50)
0.15
χ 2 LR test (p)
NESB, Non-English-speaking background; RACF, residential aged care facility; OR, odds ratio; CI, confidence interval; LR, likelihood ratio; p, probability.
P. M. Simpson et al.
347
Falls Nontransport Prediction
TABLE 3. “All variables” multivariate logistic regression analysis modeling the adjusted odds of nontransport for people aged ≥65 who have fallen and received an emergency ambulance response Adjusted OR (95% CI) (complete data n = 1,031)
Downloaded by [Chinese University of Hong Kong] at 18:51 05 November 2015
Variable
Patient age 65–74 years (referent) 75–84 years 85+ years Response priority Urgent (referent) Nonurgent (within 60 min) Nonurgent (within 90 min) Personal alarm present No (referent) Yes Initial patient physiology Normal (referent) Abnormal New injury/pain Yes (referent) No Change in usual function post fall No (referent) Yes
Adjusted OR (95% CI) (imputation data n = 1,484)
1 0.61 (0.37–1.03) 0.40 (0.23–0.67)
1 0.64 (0.41–0.99) 0.50 (0.32–0.78)
1 2.50 (1.53–3.30) 4.99 (2.24–11.10)
1 2.04 (1.5–2.78) 4.84 (2.63–8.91)
1 1.81 (1.17–2.79)
1 2.0 (1.41–2.81)
1 0.44 (0.30–0.64)
1 0.5 (0.37–0.69)
1 3.54 (2.37–5.27)
1 3.67 (2.66–5.07)
1 0.06 (0.03–0.10)
1 0.12 (0.08–0.19)
OR, odds ratio; CI, confidence interval.
96.7%, respectively, with a false-positive rate of 23.4% and a false-negative rate of 23.2%. The AUCs for the “all variable” and “dispatch variable” models differed significantly (χ 2 = 80.5; p < 0.0001) (Figure 2).
DISCUSSION Using data from a prospective cohort study, our analyses identified operational and clinical variables that are predictive of a nontransport outcome for older people who have fallen and been attended to by paramedics. The “all variables” model, which demonstrated excellent discrimination, provides important information on predictors of nontransport following a fall. Patient age group was a powerful predictor that continued to influence transport disposition even after adjusting for other potential confounders. Older people were more likely to be transported, a finding consistent with an earlier retrospective study conducted within the
FIGURE 2. ROC curve comparison between “all variables” model and “dispatch variables” model (complete data: n = 1,031 for both models). AVM, all variables model; DVM, dispatch variables model.
TABLE 4. “Dispatch variables” multivariate logistic regression analysis modeling the adjusted odds of nontransport for people aged ≥65 who have fallen and received an emergency ambulance response Variable
Patient age 65–74 years (referent) 75–84 years 85+ years Response priority Urgent (referent) Nonurgent (within 60 min) Nonurgent (within 90 min) Personal alarm present No (referent) Yes OR, odds ratio; CI, confidence interval.
Adjusted OR (95% CI) (complete data n = 1,031)
Adjusted OR (95% CI) (imputation data n = 1,484)
1 0.80 (0.53–1.22) 0.50 (0.33–0.77)
1 0.78 (0.55–1.11) 0.54 (0.38–0.79)
1 1.94 (1.43–2.63) 11.93 (6.35–22.40)
1 1.92 (1.49–2.47) 12.64 (7.64–20.92)
1 2.24 (1.57–3.19)
1 2.51 (1.88–3.35)
Downloaded by [Chinese University of Hong Kong] at 18:51 05 November 2015
348 same ambulance service.1 The “oldest” old are also likely to be increasingly frail and present with more comorbidities, which together may serve to flag a higher level of clinical risk, leading to the default outcome of transport to ED being preferred by paramedics. Also, this group had a higher proportion of patients who were residents of aged-care facilities, where the highest rate of transport was recorded (almost twice that of community dwelling fallers). It is not clear whether this is clinically driven or a reflection of operational policies and practice of residential aged-care facilities. The presence of a personal alarm device reduced the odds of transportation by almost 50%, but the reasons for this association are unclear. It could be a function of a sense of safety or reassurance on behalf of the patient or paramedic that should the patient fall again, they will have a mechanism by which they can seek help if required. Interestingly, older people have been found to be reluctant to use personal alarm devices when they fall, even if they cannot get up on their own.16,17 From an operational perspective, Johnston et al. reported that personal alarms do not impact on ambulance response times or on the proportion of older fallers experiencing “long lies.”18 Also noted was a high rate of accidental alarm activation, generating additional emergency call volume and subsequent operational demand. Allocation of a nonurgent response priority remained a strong predictor of nontransport after adjusting for potential confounders, with the odds of nontransport increasing markedly as response priority decreased. This finding is consistent with what one would expect of the dispatch system, as it stands to reason that falls cases with high odds of nontransport are less serious and therefore triage to a nonurgent response priority constitutes an appropriate result. While the “dispatch variables” predictive model demonstrated a relatively low level of discrimination, there are important operational implications arising from these findings that could assist in the rationalization of emergency dispatch to older fallers and optimization of dispatch efficiency. The “dispatch variables” model demonstrated poor sensitivity (27%) but excellent specificity (97%); that is, the model appears to be poor at predicting nontransport cases but very effective at predicting those who will be transported. Based on these data, dispatching a resource other than a transport ambulance to falls involving the “oldest” old who have been allocated a nonurgent response following a fall could represent an inefficient use of resources, as a transport ambulance would almost certainly also be required soon after. Avoiding unnecessary dual responses to a single incident is desirable considering the high demand volume confronted by modern ambulance services.19 Based on the almost 50% increase in sensitivity seen in the “all variables” model after adding clinical variables, predicting nontransport cases appears to require
PREHOSPITAL EMERGENCY CARE
JULY/SEPTEMBER 2014
VOLUME 18 / NUMBER 3
a paramedic on scene to assess the patient. Models of care involving diversion of “low-acuity” emergency calls away from immediate dispatch toward secondary telephone triage (“hear and treat”) are increasingly common, aiming to reduce the number of unnecessary emergency responses.17,20 There is a risk that older fallers could be seen as a “nonserious” population for whom such secondary telephone triage might be appropriate21 ; however, our findings suggest that “hands-on” assessment of this population is necessary in identifying cases likely to result in nontransport. This in turn has important implications for training and education of paramedics, as there is evidently a need to ensure they are equipped to undertake a systematic assessment and clinical examination so as to accurately gather clinical information relating to physiology, injury, and post-fall function. Paramedics with additional training in patient assessment and examination beyond that of a standard paramedic have been shown to have high levels of safety and good outcomes in relation to older fallers.22–24 With patients accessing ambulance services being increasingly older, and falls accounting for a large proportion of responses in that group, it could be argued that the “enhanced” assessment and examination skills should represent the standard for all paramedics instead of being reserved for specialist groups of paramedics. While this study provides information regarding quantitative factors and their association with transport outcomes for older people who have fallen, it is important to recognize that other less quantifiable factors are likely to contribute also. The transaction between paramedic and patient, and the processes through which a transport decision is reached, appear to be complex, and exactly how a nontransport outcome for an older faller evolves remains unclear.7 Further qualitative research investigating attitudes, perceptions, and beliefs of paramedics, as well as organizational and cultural enablers and barriers to nontransport, may provide valuable insight that would complement these quantitative findings.
LIMITATIONS This study has several limitations. First, the nonconsecutive nature of the prospect cohort study from which this study population was extracted could introduce a risk of selection bias which may affect the generalizability of the results. Second, it is possible that being part of an observational study could have caused paramedics to alter their usual clinical practice in relation to the transportation decision. However, paramedics were na¨ıve to the specific research questions being addressed until after the data collection period had been completed. There was also no difference in nontransport rates during the study period when compared to that reported in a previous populationbased retrospective study conducted within the same
Downloaded by [Chinese University of Hong Kong] at 18:51 05 November 2015
P. M. Simpson et al.
349
Falls Nontransport Prediction
ambulance service.1 Third, missing values were a concern across several variables of interest, but the use of multiple imputation for managing data mitigates the statistical uncertainty that the presence of missing values may introduce. The findings are presented for the original dataset and the multiple imputation dataset so as to enable comparison of the two approaches and the impact of using a multiple imputation analysis. Fourth, it should be recognized that the predictive models presented in these findings constitute “unvalidated” prediction models. Validation was considered to be unnecessary for the “all variables” model as it was not designed or intended to be used on patients in clinical or operational practice. Similarly, a validation process using an external sample was also not undertaken for the “dispatch variables” predictive model as its discriminative ability was insufficient to be clinically or operationally useful. Finally, clinical practice by paramedics relating to management of fallers is not homogenous between ambulance services, particularly in relation to nontransport situations. Therefore, the results of this study may not be generalizable to other ambulance services.
6.
7.
8.
9.
10.
11.
12.
13.
CONCLUSION In this population of confirmed older fallers attended to by paramedics, determination of the prehospital transport outcome is greatly influenced by on-scene findings resulting from paramedic assessment. The presence of new pain, abnormal physiology, and altered function post fall were strongly associated with increased odds of transport. Conversely, the presence of a personal alarm and allocation of a nonurgent dispatch priority increased the odds of nontransport. Accurate discrimination between older fallers who were and were not transported using dispatch data only was not possible.
14.
15. 16.
17.
18.
19.
References 1.
Simpson PM, Bendall JC, Tiedamann A, Close JCT, Middleton PM. Epidemiology of ambulance responses to older fallers in New South Wales, Australia. Australas J Ageing. 2012;32(3):171–176. 2. Snooks HA, Halter M, Close JC, Cheung WY, Moore F, Roberts SE. Emergency care of older people who fall: a missed opportunity. Qual Saf Health Care. 2006;15(6):390–2. 3. Cone D, Ahern J, Lee C, Baker D, Murphy T, Bogucki S. A Descriptive study of the “lift-assist” call. Prehosp Emerg Care. 2013;17(1):51–56. 4. Tiedemann A, Mikolaizak S, Sherrington C, Segin K, Lord SR, Close JCT. Older fallers attended to by an ambulance but not transported to hospital:a vulnerable population at high risk of future falls. Aust NZ J Public Health. 2013;37:179–85. 5. Salter AE, Khan KM, Donaldson MG, Davis JC, Buchanan J, Abu-Laban RB, Cook WL, Lord SR, McKay HA. Communitydwelling seniors who present to the emergency department with a fall do not receive Guideline care and their fall risk pro-
20.
21.
22.
23.
24.
file worsens significantly: a 6-month prospective study. Osteoporos Int. 2006;17(5):672–83. Logan PA, Coupland CA, Gladman JR, Sahota O, Stoner-Hobbs V, Robertson K, Tomlinson V, Ward M, Sach T, Avery AJ. Community falls prevention for people who call an emergency ambulance after a fall: randomised controlled trial. BMJ. 2010;340:c2102. Halter M, Vernon S, Snooks H, Porter A, Close J, Moore F, Porsz S. Complexity of the decision-making process of ambulance staff for assessment and referral of older people who have fallen: a qualitative study. Emerg Med J. 2010;28(1):44–50. Isenberg D, Cone DC, Stiell IG. A simple three-step dispatch rule may reduce lights and sirens responses to motor vehicle crashes. Emerg Med J. 2012;29(7):592–5. Simpson PM, Bendall JC, Tiedemann A, Lord SR, Close JCT. Epidemiology of emergency medical service responses to older people who have fallen: a prospective cohort study. Prehosp Emerg Care. 2013;in press (accepted for publication September 10, 2013). The prevention of falls in later life. A report of the Kellogg International Work Group on the Prevention of Falls by the Elderly. Danish Med Bull. 1987;34 (Suppl 4):1–24. Hertel BR. Minimizing error variance introduced by missing data routines in survey analysis. Sociol Methods Res. 1976;4(4):459–74. Raymond MR, Roberts DM. A comparison of methods for treating incomplete data in selection research. Educ Psychol Measurement. 1987;47(1):13–26. Hair, Joseph F, et al. Multivariate Data Analysis: A Global Perspective. 6th ed. Upper Saddle River, NJ: Prentice Hall, 2006. Malhotra NK. Analyzing marketing research data with incomplete information on the dependent variable. J Marketing Res. 1987;24(1):74–84. Hosmer D, Lemeshow S. Applied Logistic Regression, 2nd ed. New York: Wiley-Interscience; 2000. Fleming J, Brayne C, Cambridge City over-75s Cohort study c. Inability to get up after falling, subsequent time on floor, and summoning help: prospective cohort study in people over 90. BMJ. 2008;337:a2227. Johnston K, Grimmer-Somers K, Sutherland M. Perspectives on use of personal alarms by older fallers. Int J Gen Med. 2010;30(3):231–7. Johnston K, Worley A, Grimmer-Somers K, Sutherland M, Amos L. Personal alarm use to call the ambulance after a fall in older people: characteristics of clients and falls. J Emerg Primary Healthcare. 2010;8(4):1–9. Lowthian JA, Cameron PA, Stoelwinder JU, Curtis A, Currell A, Cooke MW, McNeil JJ. Increasing utilisation of emergency ambulances. Aust Health Rev. 2011;35(1):63–9. Dale J, Williams S, Foster T, Higgins J, Snooks H, Crouch R, Hartley-Sharpe C, Glucksman E, George S. Safety of telephone consultation for “non-serious” emergency ambulance service patients. Qual Saf Health Care. 2004;13(5):363–73. Snooks H, Cheung WY, Gwini SM, Humphreys I, Sanchez A, Siriwardena N. Can older people who fall be identified in the ambulance call centre to enable alternative responses or care pathways? Emerg Med J. 2011;28(3):e1. Mason S, Knowles E, Freeman J, Snooks H. Safety of paramedics with extended skills. Acad Emerg Med. 2008;15(7): 607–12. Mason S, Wardrope J, Perrin J. Developing a community paramedic practitioner intermediate care support scheme for older people with minor conditions. Emerg Med J. 2003;20(2): 196–8. Hoyle S, Swain AH, Fake P, Larsen PD. Introduction of an extended care paramedic model in New Zealand. Emerg Med Australas. 2012;24(6):652–656.
ISSN: 1090-3127 (Print) 1545-0066 (Online) Journal homepage: http://www.tandfonline.com/loi/ipec20
Predictors of Nontransport of Older Fallers Who Receive Ambulance Care Paul M. Simpson MScM (ClinEpi), BEd, BHSc, Jason C. Bendall MM (ClinEpi), MBBS, PhD, Barbara Toson BStatEc, Anne Tiedemann BSc, PhD, Stephen R. Lord PhD & Jacqueline C. T. Close MD To cite this article: Paul M. Simpson MScM (ClinEpi), BEd, BHSc, Jason C. Bendall MM (ClinEpi), MBBS, PhD, Barbara Toson BStatEc, Anne Tiedemann BSc, PhD, Stephen R. Lord PhD & Jacqueline C. T. Close MD (2014) Predictors of Nontransport of Older Fallers Who Receive Ambulance Care, Prehospital Emergency Care, 18:3, 342-349 To link to this article: http://dx.doi.org/10.3109/10903127.2013.864355
Published online: 24 Jan 2014.
Submit your article to this journal
Article views: 234
View related articles
View Crossmark data
Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=ipec20 Download by: [Chinese University of Hong Kong]
Date: 05 November 2015, At: 18:51
PREDICTORS OF NONTRANSPORT OF OLDER FALLERS WHO RECEIVE AMBULANCE CARE Paul M. Simpson, MScM (ClinEpi), BEd, BHSc, Jason C. Bendall, MM (ClinEpi), MBBS, PhD, Barbara Toson, BStatEc, Anne Tiedemann, BSc, PhD, Stephen R. Lord, PhD, Jacqueline C. T. Close, MD
Downloaded by [Chinese University of Hong Kong] at 18:51 05 November 2015
ABSTRACT
included response nonurgent response priority, younger age, and the presence of a personal alarm. The AUC was 0.68 (95% CI 0.64–0.71; p < 0.0001) (imputed data AUC 0.69 (95% CI 0.66–0.72)). Conclusion. In this population of confirmed older fallers attended to by paramedics, determination of the prehospital transport outcome is greatly influenced by on-scene findings resulting from paramedic assessment. The presence of new pain, abnormal physiology, and altered function post-fall were strongly associated with increased odds of transport. Conversely the presence of a personal alarm and allocation of a nonurgent dispatch priority increased the odds of nontransport. Accurate discrimination between older fallers who were and were not transported using dispatch data only was not possible. Key words: ambulance; emergencies; aged; accidental falls
Objectives. To identify patient, clinical, and operational factors associated with nontransport of older people who have fallen and received ambulance care; and to develop a nontransport prediction tool that could be utilized during the dispatch process to rationalize allocation of emergency ambulance resources. Methods. The study was a planned subanalysis using data collected during a prospective observational cohort study of nonconsecutive emergency responses to older people aged 65 years or more who had fallen between October 1, 2010 and June 30, 2011. The data consisted of routinely collected ambulance dispatch and clinical records, combined with prospectively collected fall-specific information. Missing data were managed using multiple imputation. Multivariate logistic regression modeling was undertaken to identify predictors of nontransport. Results are described for original and imputated data sets, presented as odds ratios (OR) with 95%CI (confidence interval). Receiver operating curve (ROC) statistics were generated, with model discrimination determined by the area under the curve (AUC). Results. There were 1,484 cases eligible for this subanalysis of which 419 (28.2%) were recorded as nontransport. Multivariate regression including dispatch and clinical variables identified a 6-item final model. Younger age group, nonurgent response priority, and presence of a personal alarm were predictors of nontransport, along with clinical variables, including normal vital signs, absence of injury, and unchanged functional status post-fall. The AUC was 0.88 (95% CI 0.86–0.90; p < 0.0001) (imputed data AUC 0.86 (95% CI 0.84–0.88)). Multivariate modeling of dispatch variables only identified a 3-item final model, which
PREHOSPITAL EMERGENCY CARE 2014;18:342–349
INTRODUCTION Provision of emergency responses to older people who have fallen is a core function of ambulance services. While the majority of older people are transported to an emergency department (ED) following assessment by paramedics, between one-quarter and onethird are not transported after ambulance care.1 Evidence from the United Kingdom (UK), United States (US), and Australia suggests that nontransport of this older population may result in an increased risk of adverse patient outcomes and high rates of ambulance reattendance.2–4 Equally, there is insufficient evidence to indicate whether these people would do better if transported to an ED or if they received alternate models of care.5,6 While the transport decision-making process of paramedics has been previously described,7 external factors associated with nontransport of older fallers following paramedic attendance have not been explored. Identification of patient, clinical, or operational factors associated with nontransport could reveal aspects of ambulance management of older fallers that will elucidate this process that impacts significantly on patient outcomes and ambulance resource utilisation. A better understanding of factors associated with nontransport also provides an opportunity to explore alternate models of care and possibly the development of a nontransport dispatch prediction tool that could be used to rationalize ambulance resource allocation and optimize patient safety. Dispatch prediction rules
Received July 11, 2013 from the University of Western Sydney, Sydney, Australia (PMS), Ambulance Service of New South Wales, Sydney, Australia (PMS), Department of Anaesthesia, Gosford Hospital, Gosford, Australia (JCB), Neuroscience Research Australia (BT, SRT, JCTC), School of Public Health and Community Medicine, University of New South Wales, Sydney, Australia (PMS, SRL), The George Institute for Global Health, Sydney, Australia (AT), Sydney Medical School, The University of Sydney, Sydney, Australia (AT), Prince of Wales Clinical School, University of New South Wales, Sydney, Australia (JCTC). Revision received September 9, 2013; accepted for publication October 7, 2013. The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper. Address correspondence to Mr. Paul Simpson, University of Western Sydney, Locked Bag 1797, Penrith NSW 2751, Australia. e-mail: [email protected] doi: 10.3109/10903127.2013.864355
342
Downloaded by [Chinese University of Hong Kong] at 18:51 05 November 2015
P. M. Simpson et al.
343
Falls Nontransport Prediction
have been previously explored in other contexts. For example, Isenberg et al. developed a simple dispatch prediction tool to reduce unnecessary urgent responses to minor car accidents.8 The ability to predict nontransportation of older fallers could offer several benefits for ambulance services. First, dispatching an appropriate resource to a case likely to result in nontransport might streamline patient care and minimize the clinical risk. Second, with most ambulance services facing unprecedented emergency call demands, a predictive tool with good discrimination could enable dispatch of, for example, a single appropriately skilled paramedic responder, leaving a double-crewed transport ambulance available to respond to cases where transport is more likely. With this context in mind, the two key aims of this study were to 1) identify patient, clinical, and operational factors associated with nontransport of older people who have fallen and received ambulance care; and 2) develop a nontransport prediction tool that could be utilized during the dispatch process to rationalize allocation of emergency ambulance resources.
METHODS
ple of which 1.02 million (14%) are aged 65 years or more. NSW is serviced by a single, state-based government ambulance service that has approximately 3,500 paramedics within three clinical levels: paramedic interns (first to third year of training); qualified paramedics (have completed 3 years of core training); and paramedic specialists (either intensive care paramedic or extended care paramedic). Training is provided internally using a vocational training model, but with an increasing number of university-trained paramedics. Paramedic management of cases involving falls experienced by older people is guided by a “falls in the elderly” protocol. Paramedics can pursue mutually agreed upon nontransport “clinical pathways” where strictly defined criteria are met following detailed history taking and patient assessment, although any patient requesting transport following discussion must be transported to ED. Dispatch of ambulance resources is managed by a computer-aided dispatch (CAD) system, in which incoming calls are categorized and prioritized using the Medical Priority Dispatch System (MPDS) and ProQA software (Priority Dispatch Corporation, Salt Lake City, Utah, US).
Study Design
Data Sources
This study was a planned subanalysis of a cohort of patients extracted from a larger dataset that had been collected during a prospective observational study investigating prehospital management of older people who had fallen and received an emergency response.9 The larger epidemiological study used a prospective, convenience-sampling (nonconsecutive enrollment) design to collect fall-specific data over a 9-month period from 1 October, 2010 through 30 June, 2011. During that period, paramedics who had volunteered to participate in the study prospectively enrolled people aged 65 years and older who were confirmed after arriving on scene to have had a fall, and collected fallspecific data, which was subsequently linked to routinely collected clinical record and dispatch data.
The data used in this study were drawn from three sources. Data relating specifically to falls and patient demography were collected prospectively using a study-specific data sheet (Figure 1) by paramedics at the point of care. The other two data sources were clinical and dispatch data, retrieved from routinely collected administrative databases. The clinical data consists of the patient health-care clinical record that is completed by paramedics for every incident attended. The dispatch data, sourced from the CAD database, includes dispatch priority details and operational times for each phase of ambulance response and care. Selection of the variables and their relevance to the outcome of nontransport was based on existing literature in this area and the consensus of an advisory group working in the fields of prehospital care, falls and balance, and aged care (Tables 1 and 2).
Population The study sample comprised people aged 65 years or older who had received an emergency ambulance response and were confirmed by paramedics after arriving at the scene to have had a fall using a standard definition.10
Study Setting The study was conducted in New South Wales (NSW), Australia, an area of 800,000 square kilometers containing metropolitan and rural/remote regions. NSW has a population of approximately 7.13 million peo-
Statistical Analysis Descriptive statistical analysis and univariate logistic analysis were undertaken for the variables of interest. In order to maximize the number of cases available for inclusion, missing values found not to be missing completely at random were managed using multiple imputation (MI) methods incorporated within the SPSS missing data analysis module (IBM SPSS Statistics for Windows, Version 21.0, Armonk, NY). The MI process involved 10 imputations of data so as to maximize relative efficiency of the results arising from the logistic
PREHOSPITAL EMERGENCY CARE
JULY/SEPTEMBER 2014
VOLUME 18 / NUMBER 3
Downloaded by [Chinese University of Hong Kong] at 18:51 05 November 2015
344
FIGURE 1. Fall information data collection tool.
regression modelling process.11 The decision to use MI was undertaken in accordance with established conventions regarding handling of missing data.12–14 Two independent multivariate logistic regression analyses were then undertaken. The dependent variable for both was prehospital transport outcome, a dichotomous variable consisting of “transport” or “no transport” as recorded by the computerized emergency dispatch system, which tracks each phase of an emergency case. Only predictors that satisfied the inclusion criteria of p ≤ 0.25 in the univariate analysis
were included in the multivariate logistic regression models. The first “all variables” analysis was an exploratory analysis conducted to identify demographic, clinical, and operational factors predictive of nontransport of older people following a fall. Included in this were the 20 variables derived from information gathered during the initial emergency phone call and the ambulance dispatch process and by paramedics on scene after conducting an assessment of the patient (Table 1). The resulting “all variables model” (AVM) was also
P. M. Simpson et al.
TABLE 1. Predictor variables subjected to univariate logistic regression for “all variables” and “dispatch variables” analyses “All variables” regression analysis
Downloaded by [Chinese University of Hong Kong] at 18:51 05 November 2015
345
Falls Nontransport Prediction
Patient gender Patient age group Response priority Time of day Time of week Personal alarm present Residential status Fall location Operational region Presence of carer Non-English-speaking background (NESB) On floor at ambulance arrival New injury/pain Initial presenting physiology Fall in past 12 months Patient of anticoagulants Number of prescription medications Use of mobility/walking aid Change in usual function post fall Highest paramedic clinical level on scene
“Dispatch variables” regression analysis
Patient gender Patient age group Response priority Time of day Time of week Personal alarm present Residential status Fall location Operational region Presence of carer
deemed to constitute an optimal predictive model for nontransport of older people following a fall, against which a second regression model would be compared. The second “dispatch variables” analysis was conducted to develop a predictive “dispatch variables model” (DVM) from which a prediction rule might be derived that could potentially be applied during the dispatch process to guide emergency resource allocation. As such, this analysis consisted only of those variables that could be derived from information routinely gathered during the existing dispatch process, or from information that could potentially be reliably and accurately gathered by adding additional questions during this process (Table 1). A stepwise backward elimination process was undertaken in both regression models. At each step the least significant variable was removed until only variables significant at p < 0.05 remained. Interactions between remaining variables were tested, and retained in the model if significant at p < 0.01. The final multivariate model was assessed for goodness-of-fit using the Hosmer and Lemeshow method with p ≥ 0.05 considered an acceptable fit.15 Multivariate regression analysis results are presented for both the complete and imputed datasets, expressed as odds ratios (OR) with 95% confidence intervals (CI). Receiver operating curve (ROC) statistics were generated to determine the area under the curve (AUC) for both regression models using the complete data set, in order to compare the potential performance of the “dispatch variables” model with the “all variables” optimal model. An AUC of 0.7 for the DVM was selected
as the minimum value for model discrimination that would allow creation of a dispatch prediction rule. The chi-squared test was used to determine statistical difference in AUC for the two ROC curves, with significance established at p = 0.05.
Ethical Approval Ethical approval was granted by the Sydney Local Area District (SLHD) Human Research Ethics Committee (HREC), Royal Prince Alfred Zone (Protocol No X10-0152 & HREC/10/RPAH/282).
RESULTS There were 1,610 cases identified prospectively by paramedics as having fallen during the study period. Of these, 55 were excluded due to not being emergency response cases (i.e., booked medical responses, routine ambulance transports), and another 71 were excluded due to not having a transport outcome (i.e., transport or nontransport) recorded. There were 1,484 eligible cases forming the complete data set for analysis of which 419 (28.2%) were recorded as nontransport. The missing data analysis identified several variables of interest with between 5 and 15% missing data, resulting in the list-wise loss of approximately one-third of cases in the regression analysis. Table 2 shows the results of the univariate “all variables” analysis, while Table 3 shows the results of the multivariate logistic regression, which led to a 6item final predictive model. Younger age, lower response priority, the presence of an alarm device, the absence of new injury or pain or change in usual level of function, and normal physiology were all associated with increased odds of a nontransport outcome. The goodness-of-fit test indicated good model fit (8 DF, χ 2 = 7.43, p = 0.49). For the “dispatch variables”, 10 variables underwent univariate analysis (Table 1). The multivariate logistic regression analysis identified a 3-item final model, which included response priority, age, and the presence of a personal alarm (Table 4). The goodness-offit test indicated good model fit (7DF, χ 2 = 10.01, p = 0.19). The AUC for the optimal “all variables” prediction model was 0.88 (95% CI 0.86–0.90; p < 0.0001) (imputed data AUC 0.86 (95% CI 0.84–0.88)), demonstrating excellent model discrimination (Figure 2). Using a probability cutoff of 50%, the sensitivity and specificity of this model were 59 and 91%, respectively, with a false-positive rate of 31% and a false-negative rate of 14%. However, for the “dispatch variables” model, the AUC of 0.68 (95% CI 0.64–0.71; p < 0.0001) (imputed data AUC 0.69 (95% CI 0.66–0.72)) demonstrated at best only moderate discrimination (Figure 2). The sensitivity and specificity of this model were 27.2 and
346
PREHOSPITAL EMERGENCY CARE
JULY/SEPTEMBER 2014
VOLUME 18 / NUMBER 3
TABLE 2. Univariate logistic regression analysis of all variables modeling unadjusted odds of nontransport for people aged ≥65 who have fallen and received an emergency ambulance response (complete data set n = 1,484)
Downloaded by [Chinese University of Hong Kong] at 18:51 05 November 2015
Variable
NESB Yes (referent) No Patient gender Male (referent) Female Patient age 65–74 years (referent) 75–84 years 85+ years Response priority Urgent (referent) Nonurgent (within 60 min) Nonurgent (within 90 min) On floor at ambulance arrival Yes No Time of day Dayshift (0700–1859) (referent) Nightshift (1900–0659) Time of week Weekday (referent) Weekend Initial patient physiology Normal (referent) Abnormal New injury/pain Yes (referent) No Personal alarm present Yes (referent) No Fall in past 12 months Yes (referent) No Patient on anticoagulant medication Yes (referent) No Number of prescription medications 0–3 (referent) 4–7 8+ Residential status of patient Community dwelling (referent) RACF Use of a mobility/walking aid No (referent) Yes Change in usual function post fall No (referent) Yes Carer status Lives with carer (referent) Lives alone Location of fall Place of residence (referent) Public area Geographic area Metropolitan Regional/rural Highest paramedic clinical level on scene Paramedic intern (referent) Qualified paramedic Paramedic specialist
Transport N (%) (n = 1,065)
Nontransport N (%) (n = 419)
Univariate OR (95% CI)
132 (78.1) 833 (70.2)
37 (21.9) 353 (29.8)
1 1.51 (1.03–2.22)
0.03
381 (69.7) 678 (73.1)
166 (30.4) 249 (26.9)
1 0.84 (0.67–1.06)
0.15
155 (66.8) 372 (69.3) 484 (75.6)
77 (33.2) 165 (30.7) 156 (24.4)
1 0.89 (0.64–1.24) 0.65 (0.47–0.90)
0.01
688 (80.3) 354 (67.2) 23 (23.0)
169 (19.7) 173 (32.8) 77 (77.0)
1 2.0 (1.56–2.55) 13.63 (8.31–22.36)
< 0.0001
508 (64.5) 517 (80.2)
280 (35.5) 128 (19.8)
1 0.45 (0.35–0.57)
< 0.0001
826 (73.2) 239 (67.3)
303 (26.9) 116 (32.7)
1 1.32 (1.02–1.71)
0.03
774 (70.7) 291 (74.8)
321 (29.3) 98 (25.2)
1 0.81 (0.62–1.06)
0.12
527 (65.8) 471 (81.8)
274 (34.2) 105 (18.2)
1 0.43 (0.33–0.56)
< 0.0001
898 (83.9) 126 (34.2)
172 (16.1) 243 (65.9)
1 10.07 (7.69–13.19)
< 0.0001
185 (57.1) 830 (75.6)
132 (42.9) 268 (24.4)
1 0.43 (0.33–0.56)
< 0.0001
494 (68) 474 (75.2)
232 (32) 156 (24.8)
1 0.70 (0.55–0.89)
0.04
349 (71.7) 622 (71.4)
128 (28.3) 249 (28.6)
1 1.01 (0.79–1.30)
0.92
292 (70.7) 433 (72.7) 224 (70.2)
121 (29.3) 163 (27.4) 95 (29.8)
1 0.9 (0.69–1.20) 1.02 (0.74–1.40)
0.68
878 (70) 172 (81.5)
377 (30) 39 (18.5)
1 0.53 (0.37–0.74)
0.0007
440 (79.9) 546 (65.6)
111 (20.2) 287 (34.5)
1 2.08 (1.62–2.68)
< 0.0001
356 (53.2) 547 (96.6)
313 (46.8) 19 (3.4)
1 0.04 (0.02–0.06)
< 0.0001
373 (72.9) 554 (69)
139 (27.2) 249 (31)
1 1.21 (0.94–1.54)
0.14
817 (69.6) 197 (78.5)
357 (30.4) 54 (21.5)
1 0.63 (0.45–0.87)
0.01
642 (72.5) 423 (70.6)
243 (27.5) 176 (29.4)
1 1.1 (0.87–1.38)
0.42
195 (77.4) 250 (71.6) 226 (77.4)
57 (22.6) 99 (28.4) 66 (22.6)
1 1.36 (0.93–1.97) 0.99 (0.67–1.50)
0.15
χ 2 LR test (p)
NESB, Non-English-speaking background; RACF, residential aged care facility; OR, odds ratio; CI, confidence interval; LR, likelihood ratio; p, probability.
P. M. Simpson et al.
347
Falls Nontransport Prediction
TABLE 3. “All variables” multivariate logistic regression analysis modeling the adjusted odds of nontransport for people aged ≥65 who have fallen and received an emergency ambulance response Adjusted OR (95% CI) (complete data n = 1,031)
Downloaded by [Chinese University of Hong Kong] at 18:51 05 November 2015
Variable
Patient age 65–74 years (referent) 75–84 years 85+ years Response priority Urgent (referent) Nonurgent (within 60 min) Nonurgent (within 90 min) Personal alarm present No (referent) Yes Initial patient physiology Normal (referent) Abnormal New injury/pain Yes (referent) No Change in usual function post fall No (referent) Yes
Adjusted OR (95% CI) (imputation data n = 1,484)
1 0.61 (0.37–1.03) 0.40 (0.23–0.67)
1 0.64 (0.41–0.99) 0.50 (0.32–0.78)
1 2.50 (1.53–3.30) 4.99 (2.24–11.10)
1 2.04 (1.5–2.78) 4.84 (2.63–8.91)
1 1.81 (1.17–2.79)
1 2.0 (1.41–2.81)
1 0.44 (0.30–0.64)
1 0.5 (0.37–0.69)
1 3.54 (2.37–5.27)
1 3.67 (2.66–5.07)
1 0.06 (0.03–0.10)
1 0.12 (0.08–0.19)
OR, odds ratio; CI, confidence interval.
96.7%, respectively, with a false-positive rate of 23.4% and a false-negative rate of 23.2%. The AUCs for the “all variable” and “dispatch variable” models differed significantly (χ 2 = 80.5; p < 0.0001) (Figure 2).
DISCUSSION Using data from a prospective cohort study, our analyses identified operational and clinical variables that are predictive of a nontransport outcome for older people who have fallen and been attended to by paramedics. The “all variables” model, which demonstrated excellent discrimination, provides important information on predictors of nontransport following a fall. Patient age group was a powerful predictor that continued to influence transport disposition even after adjusting for other potential confounders. Older people were more likely to be transported, a finding consistent with an earlier retrospective study conducted within the
FIGURE 2. ROC curve comparison between “all variables” model and “dispatch variables” model (complete data: n = 1,031 for both models). AVM, all variables model; DVM, dispatch variables model.
TABLE 4. “Dispatch variables” multivariate logistic regression analysis modeling the adjusted odds of nontransport for people aged ≥65 who have fallen and received an emergency ambulance response Variable
Patient age 65–74 years (referent) 75–84 years 85+ years Response priority Urgent (referent) Nonurgent (within 60 min) Nonurgent (within 90 min) Personal alarm present No (referent) Yes OR, odds ratio; CI, confidence interval.
Adjusted OR (95% CI) (complete data n = 1,031)
Adjusted OR (95% CI) (imputation data n = 1,484)
1 0.80 (0.53–1.22) 0.50 (0.33–0.77)
1 0.78 (0.55–1.11) 0.54 (0.38–0.79)
1 1.94 (1.43–2.63) 11.93 (6.35–22.40)
1 1.92 (1.49–2.47) 12.64 (7.64–20.92)
1 2.24 (1.57–3.19)
1 2.51 (1.88–3.35)
Downloaded by [Chinese University of Hong Kong] at 18:51 05 November 2015
348 same ambulance service.1 The “oldest” old are also likely to be increasingly frail and present with more comorbidities, which together may serve to flag a higher level of clinical risk, leading to the default outcome of transport to ED being preferred by paramedics. Also, this group had a higher proportion of patients who were residents of aged-care facilities, where the highest rate of transport was recorded (almost twice that of community dwelling fallers). It is not clear whether this is clinically driven or a reflection of operational policies and practice of residential aged-care facilities. The presence of a personal alarm device reduced the odds of transportation by almost 50%, but the reasons for this association are unclear. It could be a function of a sense of safety or reassurance on behalf of the patient or paramedic that should the patient fall again, they will have a mechanism by which they can seek help if required. Interestingly, older people have been found to be reluctant to use personal alarm devices when they fall, even if they cannot get up on their own.16,17 From an operational perspective, Johnston et al. reported that personal alarms do not impact on ambulance response times or on the proportion of older fallers experiencing “long lies.”18 Also noted was a high rate of accidental alarm activation, generating additional emergency call volume and subsequent operational demand. Allocation of a nonurgent response priority remained a strong predictor of nontransport after adjusting for potential confounders, with the odds of nontransport increasing markedly as response priority decreased. This finding is consistent with what one would expect of the dispatch system, as it stands to reason that falls cases with high odds of nontransport are less serious and therefore triage to a nonurgent response priority constitutes an appropriate result. While the “dispatch variables” predictive model demonstrated a relatively low level of discrimination, there are important operational implications arising from these findings that could assist in the rationalization of emergency dispatch to older fallers and optimization of dispatch efficiency. The “dispatch variables” model demonstrated poor sensitivity (27%) but excellent specificity (97%); that is, the model appears to be poor at predicting nontransport cases but very effective at predicting those who will be transported. Based on these data, dispatching a resource other than a transport ambulance to falls involving the “oldest” old who have been allocated a nonurgent response following a fall could represent an inefficient use of resources, as a transport ambulance would almost certainly also be required soon after. Avoiding unnecessary dual responses to a single incident is desirable considering the high demand volume confronted by modern ambulance services.19 Based on the almost 50% increase in sensitivity seen in the “all variables” model after adding clinical variables, predicting nontransport cases appears to require
PREHOSPITAL EMERGENCY CARE
JULY/SEPTEMBER 2014
VOLUME 18 / NUMBER 3
a paramedic on scene to assess the patient. Models of care involving diversion of “low-acuity” emergency calls away from immediate dispatch toward secondary telephone triage (“hear and treat”) are increasingly common, aiming to reduce the number of unnecessary emergency responses.17,20 There is a risk that older fallers could be seen as a “nonserious” population for whom such secondary telephone triage might be appropriate21 ; however, our findings suggest that “hands-on” assessment of this population is necessary in identifying cases likely to result in nontransport. This in turn has important implications for training and education of paramedics, as there is evidently a need to ensure they are equipped to undertake a systematic assessment and clinical examination so as to accurately gather clinical information relating to physiology, injury, and post-fall function. Paramedics with additional training in patient assessment and examination beyond that of a standard paramedic have been shown to have high levels of safety and good outcomes in relation to older fallers.22–24 With patients accessing ambulance services being increasingly older, and falls accounting for a large proportion of responses in that group, it could be argued that the “enhanced” assessment and examination skills should represent the standard for all paramedics instead of being reserved for specialist groups of paramedics. While this study provides information regarding quantitative factors and their association with transport outcomes for older people who have fallen, it is important to recognize that other less quantifiable factors are likely to contribute also. The transaction between paramedic and patient, and the processes through which a transport decision is reached, appear to be complex, and exactly how a nontransport outcome for an older faller evolves remains unclear.7 Further qualitative research investigating attitudes, perceptions, and beliefs of paramedics, as well as organizational and cultural enablers and barriers to nontransport, may provide valuable insight that would complement these quantitative findings.
LIMITATIONS This study has several limitations. First, the nonconsecutive nature of the prospect cohort study from which this study population was extracted could introduce a risk of selection bias which may affect the generalizability of the results. Second, it is possible that being part of an observational study could have caused paramedics to alter their usual clinical practice in relation to the transportation decision. However, paramedics were na¨ıve to the specific research questions being addressed until after the data collection period had been completed. There was also no difference in nontransport rates during the study period when compared to that reported in a previous populationbased retrospective study conducted within the same
Downloaded by [Chinese University of Hong Kong] at 18:51 05 November 2015
P. M. Simpson et al.
349
Falls Nontransport Prediction
ambulance service.1 Third, missing values were a concern across several variables of interest, but the use of multiple imputation for managing data mitigates the statistical uncertainty that the presence of missing values may introduce. The findings are presented for the original dataset and the multiple imputation dataset so as to enable comparison of the two approaches and the impact of using a multiple imputation analysis. Fourth, it should be recognized that the predictive models presented in these findings constitute “unvalidated” prediction models. Validation was considered to be unnecessary for the “all variables” model as it was not designed or intended to be used on patients in clinical or operational practice. Similarly, a validation process using an external sample was also not undertaken for the “dispatch variables” predictive model as its discriminative ability was insufficient to be clinically or operationally useful. Finally, clinical practice by paramedics relating to management of fallers is not homogenous between ambulance services, particularly in relation to nontransport situations. Therefore, the results of this study may not be generalizable to other ambulance services.
6.
7.
8.
9.
10.
11.
12.
13.
CONCLUSION In this population of confirmed older fallers attended to by paramedics, determination of the prehospital transport outcome is greatly influenced by on-scene findings resulting from paramedic assessment. The presence of new pain, abnormal physiology, and altered function post fall were strongly associated with increased odds of transport. Conversely, the presence of a personal alarm and allocation of a nonurgent dispatch priority increased the odds of nontransport. Accurate discrimination between older fallers who were and were not transported using dispatch data only was not possible.
14.
15. 16.
17.
18.
19.
References 1.
Simpson PM, Bendall JC, Tiedamann A, Close JCT, Middleton PM. Epidemiology of ambulance responses to older fallers in New South Wales, Australia. Australas J Ageing. 2012;32(3):171–176. 2. Snooks HA, Halter M, Close JC, Cheung WY, Moore F, Roberts SE. Emergency care of older people who fall: a missed opportunity. Qual Saf Health Care. 2006;15(6):390–2. 3. Cone D, Ahern J, Lee C, Baker D, Murphy T, Bogucki S. A Descriptive study of the “lift-assist” call. Prehosp Emerg Care. 2013;17(1):51–56. 4. Tiedemann A, Mikolaizak S, Sherrington C, Segin K, Lord SR, Close JCT. Older fallers attended to by an ambulance but not transported to hospital:a vulnerable population at high risk of future falls. Aust NZ J Public Health. 2013;37:179–85. 5. Salter AE, Khan KM, Donaldson MG, Davis JC, Buchanan J, Abu-Laban RB, Cook WL, Lord SR, McKay HA. Communitydwelling seniors who present to the emergency department with a fall do not receive Guideline care and their fall risk pro-
20.
21.
22.
23.
24.
file worsens significantly: a 6-month prospective study. Osteoporos Int. 2006;17(5):672–83. Logan PA, Coupland CA, Gladman JR, Sahota O, Stoner-Hobbs V, Robertson K, Tomlinson V, Ward M, Sach T, Avery AJ. Community falls prevention for people who call an emergency ambulance after a fall: randomised controlled trial. BMJ. 2010;340:c2102. Halter M, Vernon S, Snooks H, Porter A, Close J, Moore F, Porsz S. Complexity of the decision-making process of ambulance staff for assessment and referral of older people who have fallen: a qualitative study. Emerg Med J. 2010;28(1):44–50. Isenberg D, Cone DC, Stiell IG. A simple three-step dispatch rule may reduce lights and sirens responses to motor vehicle crashes. Emerg Med J. 2012;29(7):592–5. Simpson PM, Bendall JC, Tiedemann A, Lord SR, Close JCT. Epidemiology of emergency medical service responses to older people who have fallen: a prospective cohort study. Prehosp Emerg Care. 2013;in press (accepted for publication September 10, 2013). The prevention of falls in later life. A report of the Kellogg International Work Group on the Prevention of Falls by the Elderly. Danish Med Bull. 1987;34 (Suppl 4):1–24. Hertel BR. Minimizing error variance introduced by missing data routines in survey analysis. Sociol Methods Res. 1976;4(4):459–74. Raymond MR, Roberts DM. A comparison of methods for treating incomplete data in selection research. Educ Psychol Measurement. 1987;47(1):13–26. Hair, Joseph F, et al. Multivariate Data Analysis: A Global Perspective. 6th ed. Upper Saddle River, NJ: Prentice Hall, 2006. Malhotra NK. Analyzing marketing research data with incomplete information on the dependent variable. J Marketing Res. 1987;24(1):74–84. Hosmer D, Lemeshow S. Applied Logistic Regression, 2nd ed. New York: Wiley-Interscience; 2000. Fleming J, Brayne C, Cambridge City over-75s Cohort study c. Inability to get up after falling, subsequent time on floor, and summoning help: prospective cohort study in people over 90. BMJ. 2008;337:a2227. Johnston K, Grimmer-Somers K, Sutherland M. Perspectives on use of personal alarms by older fallers. Int J Gen Med. 2010;30(3):231–7. Johnston K, Worley A, Grimmer-Somers K, Sutherland M, Amos L. Personal alarm use to call the ambulance after a fall in older people: characteristics of clients and falls. J Emerg Primary Healthcare. 2010;8(4):1–9. Lowthian JA, Cameron PA, Stoelwinder JU, Curtis A, Currell A, Cooke MW, McNeil JJ. Increasing utilisation of emergency ambulances. Aust Health Rev. 2011;35(1):63–9. Dale J, Williams S, Foster T, Higgins J, Snooks H, Crouch R, Hartley-Sharpe C, Glucksman E, George S. Safety of telephone consultation for “non-serious” emergency ambulance service patients. Qual Saf Health Care. 2004;13(5):363–73. Snooks H, Cheung WY, Gwini SM, Humphreys I, Sanchez A, Siriwardena N. Can older people who fall be identified in the ambulance call centre to enable alternative responses or care pathways? Emerg Med J. 2011;28(3):e1. Mason S, Knowles E, Freeman J, Snooks H. Safety of paramedics with extended skills. Acad Emerg Med. 2008;15(7): 607–12. Mason S, Wardrope J, Perrin J. Developing a community paramedic practitioner intermediate care support scheme for older people with minor conditions. Emerg Med J. 2003;20(2): 196–8. Hoyle S, Swain AH, Fake P, Larsen PD. Introduction of an extended care paramedic model in New Zealand. Emerg Med Australas. 2012;24(6):652–656.
