Paramedic diagnostic processes: An experimental study of dual process theory

Method: Experiment 1
Participants
Participants were qualified paramedics employed by an Australian government ambulance service and paramedic students studying an accredited undergraduate paramedicine degree at an Australian university. Paramedics were invited to participate in the study during a regularly scheduled employer in-service program. Students were recruited during a face-to-face block of teaching on campus. No inducements or payments were offered for participation.
Sixty-four paramedics and forty-four students completed Experiment 1 (see Table 1). Paramedics were predominantly male (64%), while students had greater gender balance (41% male). Paramedics had a median of 13 years’ experience in the role; the least experienced in this group had 3 years, while the most experienced had practiced for 41 years.

Design
Experiment 1 used a repeated-measures design that manipulated four levels of likelihood and measured self-reports of typicality and confidence on five-point ordinal scales and impression as a forced choice from four pre-defined options. Answer fluency was measured as the response time for the impression; timing commenced from when the options were presented and stopped with the participant’s first recorded impression. The experiment aimed to induce an intuitive decision by placing participants under time pressure and by undertaking an unrelated concurrent task (De Neys, 2006; Thompson et al., 2011).

Ethical review
The study protocol was considered and approved by the Australian National University Human Research Ethics Committee (2017/141) and the Australian Catholic University Human Research Ethics Committee (2017-163R).

Materials
One area of paramedic practice where there is good knowledge of objective likelihood is Acute Coronary Syndrome (ACS). ACS is an umbrella term for several life-threatening conditions including acute angina pectoris, ST-elevation myocardial infarction , and non-ST-elevation myocardial infarction. Patients with ACS commonly, but not always, present with chest pain. However, chest pain is associated with a range of possible diagnoses ranging from life threatening through to the relatively benign. Paramedics need to accurately identify the cause of chest pain as different diagnoses are associated with different (sometimes contradictory) treatment pathways.
The signs and symptoms of ACS have been studied in several retrospective case-series (Body et al., 2010; Colbeck, 2016; Fanaroff, Rymer, Goldstein, Simel, & Newby, 2015; Sharif & Upadhye, 2016). These studies have generated likelihood ratios (LR+) for a range of signs, symptoms and risk factors. From these, a list of ‘cues’ was generated with an associated LR+ for ACS. For experiment 1, four written vignettes were designed to show a progression from most likely to least likely to be associated with ACS (Table 2). The information provided was deliberately restricted to a few key points similar to what paramedics would receive in reality to form an impression en route to an emergency incident.
The experimental materials were presented on Qualtrics survey software (Qualtrics, Provo, UT). Paramedics accessed the software using a 10-inch touch-screen tablet in the classroom, while students used a desktop computer located in a university computer lab. Participants interacted with the software using a touch screen (with their finger or a stylus) or a mouse. Paramedics completed the experiment individually in a classroom environment ranging in size from two to 12. The paramedicine students participated individually in class groups of 20-30.

Procedure
Typically, a paramedic’s first exposure to a patient occurs before they meet. In many countries, members of the public requesting paramedic attendance in an emergency do so via a centralised call centre (e.g. 000 in Australia, 911 in the USA). Call centre staff seek basic information about the location and nature of the request, including limited patient information. This information is passed to the attending paramedic verbally or electronically before traveling to the patient, or en route to the patient. This allows the paramedic to form an “impression” of what might be wrong with the patient before ever seeing them, often while concurrently performing other tasks, such as navigating and driving to the location. When the paramedic arrives at the patient’s location, they can begin a more comprehensive and structured assessment leading to a diagnosis.
The experimental design followed this structure to assess the role of this limited “pre-arrival” clinical information. After providing consent, participants answered a series of demographic questions using the software. Participants were then provided with on-screen instructions and prompted to press the “next” button when ready. This initiated the first of six practice tasks. The first three practice tasks consisted of an image of a street map with two markers on it. The map was taken from a location expected to be unfamiliar to Australian participants (the outer suburbs of Des Moines, Iowa, USA). Participants were instructed to mark a route between the markers. To ensure consistent task difficulty in both practice and experimental tasks, the markers were approximately the same linear distance apart and required a similar number of turns in the optimal route. The map appeared on the screen for only 12 seconds before automatically advancing, regardless of whether the full route had been marked or not.
The first three practice tasks consisted of the navigation exercise only; the second three practice tasks combined the navigation exercise with clinical information. For these tasks, at the top of the map was brief written clinical information similar to that seen by paramedics when traveling to a patient. The information simulated information gathered by an emergency dispatcher from a caller during an emergency request for service (Figure 1). The word length of each vignette was similar, as were the number of cues. Again, the participant had only 12 seconds to complete the task and read the information provided.
Following the navigation task, participants were asked the following questions (possible responses in square brackets):
1. Thinking just about the pre-arrival information, what do you think is most likely to be wrong with the patient? (Choose one option) [Presented in random order: Acute Coronary Syndrome; Respiratory Tract Infection; Musculo-skeletal injury; Pulmonary embolism]
2. For the previous question, you chose [response to previous question]. Based on the pre-arrival information you saw, how confident are you about your answer? [Not at all confident; Low confidence; Confident; Very confident]
3. Based just on the pre-arrival information you saw, how representative or typical was that patient compared to others with [response to first question] you know about or have experienced? [Very atypical; Atypical; Neither typical nor atypical; Typical; Very Typical]
Responses to question one are referred to below as ‘impression’; responses to questions two and three are referred to as ‘confidence’ and ‘typicality’, respectively. Following the six practice tasks, participants were presented with the four experimental tasks in random order. The experimental tasks appeared to the participant as identical to the last three practice tasks. Figure 2 outlines the experimental procedure.

Method: Experiment 2
Participants
Participants were recruited using the same processes as experiment 1. No individual participated in both experiments 1 and 2. Sixty-five paramedics and eighty-three students completed experiment 2 (Table 7). Paramedics were predominantly male (69%), while students had the opposite gender balance (25% male). Paramedics had a median 14 years’ experience; the least experienced in this group were nine graduate interns with less than 12 months experience, while the most experienced had practiced for 32 years. The paramedic group was divided according to the median experience: Less experienced (= 14 years).

Materials
Participants accessed the experimental materials as described in experiment 1. The tasks leading to an impression were identical to the “Most Likely” and “Least Likely” conditions in experiment 1 (see Table 3).
Experiment 1 presented participants with limited information to make a decision, simulating the sort of information available to paramedics prior to meeting a patient, known as pre-arrival or dispatch information. Experiment 2 added additional information such as that available to a paramedic after they meet their patient; post-arrival information. Post-arrival information used previously identified cues to form one scenario likely to be ACS and another that was unlikely; for simplicity these will be referred to as “Likely” and “Unlikely” (see Table 8). Based on previous research, the likely scenario had an estimated positive likelihood ratio for ACS of 2.1 to 6.4, and the unlikely scenario had an estimated positive likelihood ratio of 0.1 to 1.0 (Body et al., 2010; Colbeck, 2016; Fanaroff et al., 2015; Sharif & Upadhye, 2016). In addition to the diagnostic cues identified by previous research, the vignettes contained additional information to provide participants with information typically gathered by paramedics in their patient assessment. The additional information was non-diagnostic; that is, they were in the normal range or were innocuous. Both vignettes had similar word counts and presented the same number of cues. Such vignettes are used commonly in paramedic initial and ongoing training, so the style and format were familiar to participants.
The pre-arrival and post-arrival information was combined to form four vignettes each with two parts: Unlikely-Likely (UL); Likely-Unlikely (LU); Unlikely-Unlikely (UU); Likely-Likely (LL). The LL and UU vignettes were considered congruent; and the LU and UL vignettes were considered incongruent.

Procedure
The procedure was identical to experiment 1. After completing the practice tasks, participants were randomly assigned by the software to one of two groups: congruent or incongruent. Two vignettes were each presented in two parts. The first part was presented identically to that in experiment 1 and consisted of brief pre-arrival clinical information accompanied by a navigation task. As in experiment 1, the participant had 12 seconds to complete the task and read the information provided. They were then asked the same questions as in experiment 1 and the responses were recorded.
Following this, they were presented with the second part of the vignette; the post-arrival information. There was no time limit applied to this information and the participant could move onto the next questions when they were ready. When they did so, they were asked to consider the information they had just read and were asked to respond to the same three questions presented earlier, slightly reworded to draw participants’ attention to the post-arrival information they had just read (Figure 4).

方法：实验1
一、被试
本研究的被试为澳大利亚政府救护车服务机构聘用的合格急救医护人员，以及在澳大利亚高校修读经认证的本科急救医学专业的学生。急救医护人员通过雇主定期安排的在职培训项目受邀参与本研究，学生则在校园面授教学时段招募。本次研究未为被试提供任何参与报酬或激励。
共有64名急救医护人员与44名学生完成了实验1（详见表1）。急救医护人员以男性为主（占比64%），而学生群体的性别分布更为均衡（男性占比41%）。该群体中急救医护人员的从业年限中位数为13年，从业时长最短者为3年，最长者达41年。

二、实验设计
实验1采用重复测量设计，操纵四种可能性水平，以5点顺序量表测量被试的典型性与自信心自评结果，并通过4个预定义选项的迫选范式收集被试的印象判断。作答流畅性以印象判断的反应时为指标：计时始于选项呈现时刻，止于被试首次记录的印象选择时刻。本实验通过给被试施加时间压力，并要求其完成一项无关的并行任务（De Neys, 2006; Thompson et al., 2011），旨在诱发直觉性决策。

三、伦理审查
本研究方案已通过澳大利亚国立大学人类研究伦理委员会（审批号：2017/141）与澳大利亚天主教大学人类研究伦理委员会（审批号：2017-163R）的审查与批准。

四、实验材料
急救医护实践中，对急性冠状动脉综合征（Acute Coronary Syndrome, ACS）的客观可能性已有较为充分的研究。ACS是一类涵盖多种致命病症的总称，包括急性心绞痛、ST段抬高型心肌梗死与非ST段抬高型心肌梗死。ACS患者通常（但非必然）表现为胸痛，但胸痛可对应多种诊断结果，从致命性疾病到相对良性的病症均有涉及。由于不同诊断对应的治疗路径存在差异（有时甚至相互矛盾），急救医护人员需准确识别胸痛的病因。
此前已有多项回顾性病例系列研究探讨了ACS的体征与症状（Body et al., 2010; Colbeck, 2016; Fanaroff, Rymer, Goldstein, Simel, & Newby, 2015; Sharif & Upadhye, 2016），这些研究为一系列体征、症状与风险因素生成了阳性似然比（likelihood ratios, LR+），据此筛选出与ACS相关的“线索”并附上对应LR+值。针对实验1，研究人员设计了4份情境材料（vignette），呈现了ACS关联可能性从高到低的递进过程（详见表2）。实验材料刻意限定为少量关键信息，模拟急救医护人员在前往急救现场途中所能获取的真实信息，以帮助其形成初步印象。
实验材料通过Qualtrics调查软件（Qualtrics, 普罗沃，犹他州，美国）呈现。急救医护人员在教室中使用10英寸触控平板访问该软件，学生则通过高校计算机实验室的台式电脑完成实验。被试可通过触控屏（手指或触控笔）或鼠标与软件交互。急救医护人员在2至12人规模的教室中单独完成实验，急救医学专业学生则以20至30人的班级为单位，在课堂中单独参与实验。

五、实验流程
通常而言，急救医护人员在见到患者之前便已首次接触患者相关信息。在诸多国家，公众拨打急救电话求助时，会通过集中化呼叫中心（如澳大利亚的000急救热线、美国的911急救热线）进行报备。呼叫中心工作人员会询问求助地点与事件基本信息，包括有限的患者相关信息。这些信息会在急救医护人员前往现场途中或抵达现场前，以口头或电子形式传递给负责出诊的医护人员，使其在见到患者前便形成对患者病情的“初步印象”，且此过程通常伴随其他并行任务，例如导航与驾车前往现场。当急救医护人员抵达患者所在地后，便可开展更全面、结构化的评估以作出诊断。
本实验设计模拟了该流程，以评估有限的“院前”临床信息的作用。被试在签署知情同意书后，需通过软件填写一系列人口学问卷。随后，被试将收到屏幕提示的实验说明，并在准备就绪后点击“下一步”按钮，启动6项练习任务中的第一项。前三项练习任务仅包含导航练习：屏幕上呈现一张街道地图，其上设有两个标记点，该地图取自澳大利亚被试大概率陌生的区域（美国艾奥瓦州得梅因市郊区）。被试需在两点之间规划最优路径。为确保练习与实验任务的难度一致，两个标记点的直线距离相近，且最优路径所需转弯次数相同。无论被试是否完成路径规划，地图仅会在屏幕上显示12秒后自动跳转。
后三项练习任务则将导航练习与临床信息相结合：在地图顶部会显示简短的书面临床信息，模拟急救调度员在急救求助请求中从来电者处收集的信息（见图1）。每份情境材料的字数与线索数量均相近。与此前一致，被试仅有12秒的时间完成任务并阅读所提供的信息。
完成导航任务后，被试需回答以下问题（可选答案以方括号标注）：
1. 仅基于院前信息，您认为患者最可能罹患何种疾病？（请选择一项）[随机呈现选项：急性冠状动脉综合征；呼吸道感染；肌肉骨骼损伤；肺栓塞]
2. 您在上一题选择了[上一题的作答结果]。基于您所查看的院前信息，您对该答案的自信心如何？[完全不自信；低自信心；自信；非常自信]
3. 仅基于您所查看的院前信息，相较于您所了解或经历过的[第一题所选疾病]患者，该患者的典型性如何？[非常不典型；不典型；既不典型也不典型；典型；非常典型]
下文中，第一题的作答结果被称为“印象判断”，第二题与第三题的作答结果分别被称为“自信心”与“典型性”。完成6项练习任务后，被试将随机获得4项实验任务。实验任务的呈现形式与后三项练习任务完全一致。图2概述了实验流程。

方法：实验2
一、被试
实验2的被试招募流程与实验1一致，且无被试同时参与两项实验。共有65名急救医护人员与83名学生完成了实验2（详见表7）。急救医护人员仍以男性为主（占比69%），而学生群体的性别分布则与之相反（男性占比25%）。该群体中急救医护人员的从业年限中位数为14年，从业时长最短者为9名毕业实习医护人员，从业时间不足12个月，最长者达32年。急救医护人员群体按从业年限中位数分为两组：低经验组（≤14年）。

二、实验材料
被试可通过与实验1相同的方式访问实验材料。用于形成初步印象的任务与实验1中的“高可能性”与“低可能性”条件完全一致（详见表3）。
实验1向被试呈现有限的决策信息，模拟急救医护人员在见到患者前所能获取的信息，即院前信息或调度信息。实验2则新增了急救医护人员见到患者后所能获取的额外信息，即抵达后信息。抵达后信息通过此前已识别的线索，分别构建了ACS高可能性与低可能性两种情境，为简化表述，二者分别被称为“高可能性”与“低可能性”情境（详见表8）。基于此前研究，高可能性情境的ACS阳性似然比估计值为2.1至6.4，低可能性情境的阳性似然比估计值为0.1至1.0（Body et al., 2010; Colbeck, 2016; Fanaroff et al., 2015; Sharif & Upadhye, 2016）。除了此前研究已识别的诊断线索外，情境材料还包含急救医护人员在患者评估中通常会收集的额外信息，这些额外信息无诊断价值，即处于正常范围或为无害信息。两份情境材料的字数与线索数量均相近。此类情境材料常被用于急救医护人员的初始与持续培训中，因此其风格与格式为被试所熟悉。
研究人员将院前信息与抵达后信息相结合，构建了4份各包含两个部分的情境材料：低可能性-高可能性（UL）、高可能性-低可能性（LU）、低可能性-低可能性（UU）与高可能性-高可能性（LL）。其中LL与UU情境被视为一致性情境，LU与UL情境则被视为不一致性情境。

三、实验流程
实验2的流程与实验1完全一致。完成练习任务后，软件会将被试随机分配至一致性组或不一致性组。实验包含两份情境材料，每份均分为两个部分。第一部分的呈现形式与实验1完全一致，包含简短的院前临床信息与配套的导航任务。与实验1相同，被试仅有12秒的时间完成任务并阅读所提供的信息，随后需回答与实验1完全一致的问题，作答结果将被记录。
完成第一部分后，被试将看到情境材料的第二部分，即抵达后信息。该部分无时间限制，被试可在准备就绪后进入下一环节。此时，被试需基于刚刚阅读的信息，回答与此前相同的三个问题（问题表述略有调整，以引导被试关注刚查看的抵达后信息，见图4）。