Original Contribution

Does sex influence the allocation of life support level by dispatchers in acute chest pain?^?

Martin Gellerstedt MSc, PhD ^a, Angela Bang RN, PhD ^b, Emma Andreasson MSc ^a,

Anna Johansson MSc ^a, Johan Herlitz MD, PhD ^b,c,?

^aUniversity of Trollhattan/Uddevalla, Box 795, SE-451 26 Uddevalla, Sweden

^bPrehospital Research Centre of Western Sweden, University College of Boras, SE- 501 90 Boras, Sweden

^cThe Prehospital Research Centre of Western Sweden, Institute of Medicine, Department of Molecular and Clinical Medicine, Sahlgrenska University Hospital, SE- 413 45 Goteborg, Sweden

Received 5 May 2009; accepted 12 May 2009

Abstract

Aim: The aim of this study was to evaluate (a) the differences between men and women in Symptom profile, allocated life support level (LSL), and presence of acute myocardial infarction (AMI), life- threatening condition (LTC), or death and (b) whether a computer-based decision support system could improve the allocation of LSL.

Patients: All patients in Goteborg, Sweden, who called the dispatch center because of chest pain during 3 months (n = 503) were included in this study.

Methods: Age, sex, and symptom profile were background variables. Based on these, we studied allocation of LSL by the dispatchers and its relationship to AMI, LTC, and death. All evaluations were made from a sex perspective. Finally, we studied the potential benefit of using a statistical model for allocating LSL. Results: The advanced life support level (ALSL) was used equally frequently for men and women. There was no difference in age or symptom profile between men and women in relation to allocation. However, the allocation of ALSL was predictive of AMI and LTC only in men. The sensitivity was far lower for women than for men. When a statistical model was used for allocation, the ALSL was predictive for both men and women. Using a separate model for men and women respectively, sensitivity increased, especially for women, and specificity was kept at the same level.

Conclusion: This exploratory study indicates that women would benefit most from the allocation of LSL using a statistical model and computer-based decision support among patients who call for an ambulance because of acute chest pain. This needs further evaluation.

(C) 2010

^? This study was supported by grants from The Laerdal Foundation in Norway.

* Corresponding author. The Prehospital Research Centre of Western

Sweden, Institute of Medicine, Department of Molecular and Clinical Medicine, Sahlgrenska University Hospital, SE-413 45 Goteborg, Sweden. Tel.: +46 31 342 1000.

E-mail address: [email protected] (J. Herlitz).

Introduction

For patients with acute chest pain, very early handling has a crucial impact on the outcome. Around half of all patients with a confirmed acute myocardial infarction (AMI) and

0735-6757/$ - see front matter (C) 2010 doi:10.1016/j.ajem.2009.05.009

roughly one third of all patients with acute chest pain call a dispatch center for prehospital emergency care [1].

At the time of this study, Goteborg had 2 different types of ambulance, standard ambulances manned with personnel with 7 weeks of medical education and special ambulances staffed by paramedics and a nurse.

The dispatchers had the task of allocating the basic life support level (BLSL) and the standard ambulance in less alarming cases, whereas the special ambulance and the advanced life support level (ALSL) were to be used if the case history indicated an AMI or other life-threatening diseases. Disappointing statistics reveal that a large percentage of patients with AMI were allocated to the BLSL [2].

Assessing the relatively limited information given by the patient is a complex task, and symptoms such as chest pain could occur for a number of reasons apart from ischemic heart disease. To evaluate the dispatcher’s subjective assessment, a study comprising 503 consecutive patients calling the emergency number because of chest pain was conducted. In a previous study, we showed that there is a relationship between the dispatcher’s initial suspicion of an AMI and the final diagnosis [3]. However, there was no relationship between the suspicion of an AMI and the early mortality rate [3]. It is therefore important to evaluate potential ways of improving the allocation of life support level (LSL).

In a previous study, we showed that a computer-based Decision support system, including a prevalence function, could be a valuable tool for allocating the level of life support

[4] The computer-based decision support system could improve the sensitivity, that is, the percentage of patients with AMI allocated to the ALSL, and could even slightly improve the specificity. Furthermore, the mortality rate among patients with AMI allocated to the BLSL was lower using the computer-based decision support system than the dispatcher’s original allocation [4]. Similar studies with computer-based decision support systems for patients with AMI have been reported [5,6].

Over the last decade, differences in symptom profiles for men and women with AMI have been the subject of discussion [7-10]. The aim of the present study was to further investigate the data from a sex perspective. The primary aim was to evaluate whether men and women experiencing AMI and calling the emergency medical services dispatch center differ in symptom profile and the assessment made by the emergency medical services dispatchers. The secondary aim was to evaluate whether men and women benefit equally from a computer-based decision support system.

Patients and methods

A more detailed description of patients, methods, and organization is given in our previous studies [3,4,11].

Study population

All the patients in the Municipality of Goteborg, Sweden (500 000 inhabitants), who called the emergency number and were assessed by the dispatchers as having acute chest pain during a 3-month period were included in the study.

In all, 503 patients fulfilled the criteria for inclusion and were therefore included in the survey.

Variables and definitions

For each enrolled patient, a specific case record form was used; it included a standardized set of questions relating to duration of pain, intensity of pain (severe/vague), and the possible presence of the following symptoms: dyspnea, cold sweat, nausea, vertigo, and syncope. At the end of the telephone interview, the dispatcher assessed the suspicion of an AMI on a 5-grade scale (1, no suspicion; 2, vague suspicion; 3, moderate suspicion; 4, strong suspicion; and 5, confirmed). According to the established routines, there were 3 levels of dispatch based on the suspected severity: (1) highest priority and call for a special ambulance, (2) highest priority but do not call for a special ambulance, and (3) not highest priority and do not call for a special ambulance.

Each patient who was enrolled was then the subject of a follow-up of hospital records. The content of variables in hospital records included medical investigations, complica- tions, and final diagnosis.

For a confirmed AMI, at least 2 of the following had to be fulfilled: chest pain with a duration of at least 15 minutes, increased serum enzyme activity of the MB form of creatine kinase at least 2 times above the upper normal limit, or the development of Q waves in at least 2 leads in a 12-lead standard electrocardiogram.

Retrospectively, each call was classified as life-threaten- ing condition (LTC) or not. To be judged as LTC, one of the following must be fulfilled: death before discharge; a final diagnosis of AMI or myocardial ischemia (including minor elevation of biochemical markers and/or electrocardiogram changes such as ST elevation, ST depression, or T-wave inversion), pulmonary embolism, aortic aneurysm, or pneumothorax; or any of the following either before hospital admission or during hospitalization: ventricular fibrillation, ventricular tachycardia, asystole, or pulseless electrical activity.

Organization

All the ambulances were dispatched by 1 center. For each call assessed as priority 1 with a need for a special ambulance, an ALSL unit, if available, and the nearest standard ambulance were dispatched simultaneously. A nurse was on board the ALSL unit from 8 AM to 5 PM, and in addition, 2 paramedics, with 39 weeks of medical training, were always on board. The personnel in the BLSL unit had 7 weeks of

medical training. The dispatchers had all received 2 weeks of medical training (repeated for 3 days every year), focusing on identifying relevant symptoms via telephone interviews.

The ALSL unit crew was authorized to intubate, and part of the crew was authorized to give intravenous medication. thrombolytic agents were not given in the prehospital phase, and there were no facilities for external pacing. All BLSL units were equipped with automated external defibrillators.

Statistical methods

The differences between men and women in categorical variables, for example, presence of symptoms, allocated ALSL, AMI, and LTC, were analyzed using a standard ?2 test. The difference in average age was analyzed using Student t test for independent samples. The primary variables in this study were a final diagnosis of AMI, if a patient was retrospectively classified as having an LTC, and whether the patient died during 1 year of follow-up. Bivariate analyses of sex and symptoms in relationship to AMI, LTC, and death were analyzed by a ?2 test. Possible differences in age between patients with or without AMI, LTC, or death were analyzed by a standard t test. The relationship between the target variables of AMI, LTC, and death explained by the factors of allocated chosen LSL (BLSL or ALSL) and sex was analyzed using a logistic regression model including an interaction between LSL and sex.

To analyze the relationship between patient characteristics (according to the case record form used during the interview) and the dichotomous response variables (AMI, yes or no; and LTC, yes or no), logistic regression was used. The logistic model was regarded as a prevalence function, which estimates the individual probability of AMI or LTC for each patient. The model has been described in more detail in a previous study [4]. To take account of possible differences in symptom profile between men and women with AMI or LTC, we also used 2 separate models, one for men and one for women. These probabilities were used retrospectively for a fictive allocation of LSL, that is, ALSL or BLSL. This model allocation was then compared with the true allocation made by the dispatchers. The allocation based on the model uses the prevalence function and, for each patient, the allocation of the ALSL follows the following rule:

• • • If the probability of AMI is greater than threshold value, allocate the ALSL; otherwise, allocate the BLSL.

The same rule is also applied in the second model with LTC as a target variable. The choice of threshold value affects the sensitivity and specificity of the allocation; that is, sensitivity represents the probability that a patient with AMI or LTC is allocated to the ALSL, whereas specificity is the probability that a patient without AMI or LTC will be allocated to the BLSL. The choice of a threshold value is arbitrary, but to make a fair comparison, we identified the threshold value that gave 381 ALSL allocations and 122

BLSL allocations, which is identical to the true allocation made by the dispatchers. Given the same distribution between the ALSL and BLSL, we therefore have an opportunity to evaluate possible benefits, such as reduced medical risk, defined as a patient with AMI being allocated to the BLSL. Moreover, by developing sex-specific models, we have an opportunity to evaluate whether both sexes benefit equally. Obviously, this study contains a large amount of statistical testing in several different models, and all the P values should therefore be interpreted with care, that is, regarded as interesting findings rather than conclusive evidence. The study is an exploratory prestudy and must be followed by more confirmatory studies.

Results

Symptom profile by sex

Even though there were some tendencies, we found no significant differences in the presence of symptoms between men and women. However, there was a significant difference in age; women were around 8 years older on average, see Table 1.

Allocation of LSL, AMI, LTC, and death by sex

Men and women were allocated to the ALSL equally frequently, around 3 of 4 times (men, 77% ALSL; women, 74% ALSL). The percentage of patients allocated to the ALSL was not related to age but to all symptoms apart from syncope. The relationships between the allocation to the ALSL and the age and symptom profile looked more or less the same for men and women (see Table 2). The allocation to the ALSL therefore seemed to be based on the same criteria, regardless of sex.

When it came to the relationship between age and symptom profile and AMI, LTC, and death, it also seemed to

Table 1 Age and symptom profile, by sex

Sex

No.

%

Age (y)

Total

503

100

Woman

398

79.1

Man

105

20.9

P

.008

Mean 70.8 75.0			66.9
SD 14.2 12.6			14.5
Strong pain (% yes)	68.6	67.5	69.6	N.20
Dyspnea (% yes)	34.4	37.0	31.9	N.20
Cold sweat (% yes)	17.3	14.4	20.0	.10
Nausea (% yes)	20.5	20.6	20.4	N.20
Vertigo (% yes)	7.6	8.2	6.9	N.20
Syncope (% yes)	4.6	2.9	6.2	.08

	Woman					Man
	% ALSL	% AMI	% LTC	% Death		% ALSL	% AMI	% LTC	% Death
Total	74	17	42	10		77	24	46	8
Above median age
Yes	76	22	51	13		76	33	55	16
No	71	10	25	6		78	20	41	4
P	N.20	.018	b.001	.056		N.20	.022	.032	.001
Strong pain
Yes	89	20	43	10	90		29	49	8
No	43	11	38	12	48		13	39	9
P	b.001	.092	N.20	N.20	b.001		.004	.163	N.20
Dyspnea
Yes	90	13	37	8	93		31	54	12
No	65	20	44	12	70		21	42	6
P	b.001	N.20	N.20	N.20	b.001		.068	.061	.113

be similar for men and women, apart from dyspnea. The presence of dyspnea, even though it was not significant, indicated AMI, LTC, or death for men, whereas the tendency for women was the reverse (see Table 2).

Table 2 The percentage of ALSL, AMI, LTC, and death, given age, and symptom profile, by sex

Is the accuracy of the allocation of LSL the same for men and women?

There was a relationship between LSL and a final diagnosis of AMI, but only for men (men, P = .001; women, P = .731; see Table 3). For men, the sensitivity was 92% and the percentage of false-positives was 73%. For women, the sensitivity and the percentage of false-positives were almost the same: 76% and 74%, respectively. This was also confirmed in a logistic regression model showing that the interaction effect between sex and LSL was significant (P =

.034). The use of the ALSL therefore had a predictive value for AMI for men but not for women.

The same relationships were found between LSL and LTC, but they were not as strong. For men, the sensitivity was 84% and the false-positive rate was 72% (P = .017), whereas the corresponding figures for women were 77% and 72% (P N .20). There were no relationships between LSL and death, which may be due to the low number of cases of death.

Table 3 The relationship between final diagnosis of AMI (yes/no) and LSL (BLSL/ALSL), by sex

AMI

LSL

BLSL

P

ALSL

Man (n = 260)	Yes, n = 63	8%	92%	.001
	No, n = 197	27%	73%
Woman (n = 243)	Yes, n = 42	24%	76%	N.20
	No, n = 201	26%	74%

Does a statistical model for allocating LSL benefit both men and women?

In a previous study, we have shown that the sensitivity for AMI and LTC may be improved by using a computer-based decision support system and a statistical model [4]. We showed that the sensitivity for AMI could be increased from 86% to 92% and that the sensitivity for LTC could be increased from 81% to 86% while reducing the percentage of false-positives slightly at the same time [4]. In terms of medical risk, there were 15 patients with AMI who were originally allocated to the BLSL by dispatchers. Using the model, only 8 patients with AMI were allocated to the BLSL. The corresponding figures for LTC were 42 allocated to the BLSL by dispatchers and 30 allocated to the BLSL by the model. There was also a difference in terms of death. Among the 15 cases with AMI allocated to the BLSL by the dispatchers, 9 died, whereas only 1 of the 8 patients with AMI allocated to the BLSL by the model died [4].

In this study, we took a more detailed look at the 15 patients with AMI who were allocated to the BLSL and divided these patients by sex. As can be seen in Table 4, two thirds of these patients were actually women (among patients with AMI, 60% were men). For men, the model increased the sensitivity from the originally somewhat high value of 92% to 97%. For women, the original sensitivity was much lower, that is, 76%, and the model increased it to 86%. This was still lower than for men, but the model unified the sensitivity slightly. In that sense, the model produced a more equal allocation between sexes. The relationship between allocated LSL according to the model and a final diagnosis of AMI was now significant for both men and women (P = .002 and P = .004, respectively).

As can be seen in Table 4, the percentage of false- positives was reduced by the model, but only for women. For men, the percentage of false-positives actually increased.

Table 4 Improvement in allocation when using the model compared with the dispatchers’ original allocation, by sex
	Patients with AMI allocated to BLSL by dispatchers (no. of deaths)	Dispatcher sensitivity (%)	Dispatcher false-positives (%)	Patients with AMI allocated to BLSL by model (no. of deaths)	Model sensitivity (%)	Model false-positives (%)
Men	5 (4)	92	73	2	97	80
Women	10 (5)	76	74	6 (1)	86	63
Total	15 (9)	86	73	8 (1)	92	71

However, when we developed 2 separate Statistical models, that is, 1 model for women and 1 model for men, it was found that these models produced the same improvement in sensitivity but kept the percentage of false-positives almost unchanged for men (75%) and reduced it for women (68%). In overall terms, the percentage of false-positives was reduced no matter whether a single statistical model or 2 sex-Separated models were used.

Discussion

Similarities and differences between men and women in AMI are the subject of discussion. There are some indications of possible differences in symptom profile [7-10]. There may also be differences in the way different symptoms are subjectively experienced and objectively interpreted. Another potential difference is the way different question- and-answer alternatives are interpreted. For instance, a description of pain as weak or strong may depend on the person’s reference framework; for example, a women who has delivered a baby may use that as a reference. There may also be a difference in the way the dispatchers recognize anxiety and other qualitative information, depending on sex, age, and possibly other demographic variables [12-14].

In this study, we found that the ALSL was used equally frequently regardless of sex, and we found no marked differences in symptom profile between men and women in relation to the allocation of the ALSL. Nevertheless, the accuracy of ALSL allocation was shown to be higher for men than for women. As a matter of fact, ALSL allocation was only predictive of AMI in the case of men. For women, there was no relationship between allocated LSL and AMI. At this stage, we can only speculate about possible explanations. Perhaps, there are subtle differences in symptom profile or slightly different interactions between different symptoms that are important and differ between men and women. In our exploratory analyses, we noticed that it may be a good idea to use one statistical model for men and another for women. This is equivalent to using a statistical model with interactions between sexes and symptoms. This kind of modeling enables the predictive value of a symptom to differ between men and women. From our previous study, we know that a computer-based decision support system may be a valuable tool for improving the allocation of LSL. As pointed out, there was a relationship

between dispatchers’ allocated LSL and AMI, but only for men. With allocations based on the model, there was a relationship between allocated LSL and AMI for both men and women. The results of this study therefore imply that both sexes will benefit from a computer-based decision support system, but there may possibly be an even more marked improvement for women.

Limitations

The criteria for AMI are not the same today as they were at the time of the survey. It is not unlikely that a few of the patients who were judged as having unstable angina pectoris at the time of the survey would be regarded today as having AMI. We do not know how that would have influenced the results. These facts underscore the importance of interpreting our findings as hypothesis generating.

Conclusion

This exploratory study indicates that women would benefit most from the allocation of LSL using a statistical model and computer-based decision support among patients who call for an ambulance because of acute chest pain. We recommend further studies to enable the development of sex- specific models for improving allocation and for making allocation more sex equal.

References

1. Herlitz J, Karlson BW, Bang A, et al. Characteristics and outcome for patients with acute chest pain in relation to whether or not they were transported by ambulance. Eur J Emerg Med 2000;7:195-200.
2. Herlitz J, Karlsson BW, Richter A, et al. Early identification of acute myocardial and prognosis in relation to Mode of transport. Am J Emerg Med 1992;10(5):406-12.
3. Herlitz J, Bang A, Isaksson L, et al. Outcome for patients who called for an ambulance for chest pain in relation to the dispatcher’s initial suspicion of acute myocardial infarction. Eur J Emerg Med 1995;2: 75-82.
4. Gellersted M, Bang A, Herlitz J. Could a computer based system including a prevalence function support emergency medical system and improve allocation of life support level? Eur J Emerg Med 2006;13 (5):290-4.
5. Baxt WG. Use of an artificial neural network for the diagnosis of myocardial infarction. Ann Intern Med 1991;115:843-8.
6. Eggers KM, Ellenius J, Dellborg M, et al. Artificial neural network algorithms for early diagnosis of acute myocardial infarction and prediction of infarct size in chest pain patients. Int J Cardiol 2007;114: 366-74.
7. Goldberg RJ, O’Donnell C, Yarzebski J, et al. sex differences in symptom presentation associated with acute myocardial infarction: a population-based perspective. Am Heart J 1998;136:189-95.
8. Pengue S, Halm M, Smith M, et al, From United Hospital and St Paul Heart Clinic, St Paul, Minn. Women and coronary disease: relationship between descriptors of signs and symptoms and diagnostic and treatment course. Am J Crit Care 1998;7:175-82.
9. Culic V, Eterovic D, Miric D, et al. Symptom presentation of acute myocardial infarction: influence of sex, age and risk factors. Am Heart J 2002;144:1012-7.
10. Thuresson M, Berglin Jarlov M, Lindahl B, et al. Symptoms and type of symptom onset in acute coronary syndrome in relation to ST-

elevation, sex, age, and a history of diabetes. Am Heart J 2005;150: 234-42.

1. Herlitz J, Bang A, Isaksson L, et al. Ambulance despatchers’ estimation of intensity of pain and presence of associated symptoms in relation to outcome in patients who call for an ambulance because of acute chest pain. Eur Heart J 1995;16:1789-94.
2. Dale J, Higgins J, Williams S, et al. Computer assisted assessment and advice for “non-serious” 999 ambulance service callers: the potential impact on ambulance dispatch. Emerg Med J 2003;20: 178-83.
3. Wilson S, Cooke M, Morrell R, et al. A systematic review of the evidence supporting the use of priority dispatch of emergency ambulances. Prehosp Emerg Care 2002;6(1):42-9.
4. Ekberg R, Nillson J, Hermansson M. (in Swedish) Better decisions with data mining? Available at url: http://www.bibliotek.htu.se/ exarb/2000sv3-17.pdf 2000.